Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
8f18cf13 | 19 | #include <linux/kernel.h> |
065631f6 | 20 | #include <linux/bio.h> |
39279cc3 | 21 | #include <linux/buffer_head.h> |
f2eb0a24 | 22 | #include <linux/file.h> |
39279cc3 CM |
23 | #include <linux/fs.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/init.h> | |
28 | #include <linux/string.h> | |
39279cc3 CM |
29 | #include <linux/backing-dev.h> |
30 | #include <linux/mpage.h> | |
31 | #include <linux/swap.h> | |
32 | #include <linux/writeback.h> | |
33 | #include <linux/statfs.h> | |
34 | #include <linux/compat.h> | |
a27bb332 | 35 | #include <linux/aio.h> |
9ebefb18 | 36 | #include <linux/bit_spinlock.h> |
5103e947 | 37 | #include <linux/xattr.h> |
33268eaf | 38 | #include <linux/posix_acl.h> |
d899e052 | 39 | #include <linux/falloc.h> |
5a0e3ad6 | 40 | #include <linux/slab.h> |
7a36ddec | 41 | #include <linux/ratelimit.h> |
22c44fe6 | 42 | #include <linux/mount.h> |
55e301fd | 43 | #include <linux/btrfs.h> |
53b381b3 | 44 | #include <linux/blkdev.h> |
f23b5a59 | 45 | #include <linux/posix_acl_xattr.h> |
4b4e25f2 | 46 | #include "compat.h" |
39279cc3 CM |
47 | #include "ctree.h" |
48 | #include "disk-io.h" | |
49 | #include "transaction.h" | |
50 | #include "btrfs_inode.h" | |
39279cc3 | 51 | #include "print-tree.h" |
e6dcd2dc | 52 | #include "ordered-data.h" |
95819c05 | 53 | #include "xattr.h" |
e02119d5 | 54 | #include "tree-log.h" |
4a54c8c1 | 55 | #include "volumes.h" |
c8b97818 | 56 | #include "compression.h" |
b4ce94de | 57 | #include "locking.h" |
dc89e982 | 58 | #include "free-space-cache.h" |
581bb050 | 59 | #include "inode-map.h" |
38c227d8 | 60 | #include "backref.h" |
f23b5a59 | 61 | #include "hash.h" |
39279cc3 CM |
62 | |
63 | struct btrfs_iget_args { | |
64 | u64 ino; | |
65 | struct btrfs_root *root; | |
66 | }; | |
67 | ||
6e1d5dcc AD |
68 | static const struct inode_operations btrfs_dir_inode_operations; |
69 | static const struct inode_operations btrfs_symlink_inode_operations; | |
70 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
71 | static const struct inode_operations btrfs_special_inode_operations; | |
72 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
73 | static const struct address_space_operations btrfs_aops; |
74 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 75 | static const struct file_operations btrfs_dir_file_operations; |
d1310b2e | 76 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
77 | |
78 | static struct kmem_cache *btrfs_inode_cachep; | |
8ccf6f19 | 79 | static struct kmem_cache *btrfs_delalloc_work_cachep; |
39279cc3 CM |
80 | struct kmem_cache *btrfs_trans_handle_cachep; |
81 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 82 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 83 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
84 | |
85 | #define S_SHIFT 12 | |
86 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
87 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
88 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
89 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
90 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
91 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
92 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
93 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
94 | }; | |
95 | ||
3972f260 | 96 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
a41ad394 | 97 | static int btrfs_truncate(struct inode *inode); |
5fd02043 | 98 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
99 | static noinline int cow_file_range(struct inode *inode, |
100 | struct page *locked_page, | |
101 | u64 start, u64 end, int *page_started, | |
102 | unsigned long *nr_written, int unlock); | |
70c8a91c JB |
103 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
104 | u64 len, u64 orig_start, | |
105 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
106 | u64 orig_block_len, u64 ram_bytes, |
107 | int type); | |
7b128766 | 108 | |
48a3b636 | 109 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 110 | |
f34f57a3 | 111 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
112 | struct inode *inode, struct inode *dir, |
113 | const struct qstr *qstr) | |
0279b4cd JO |
114 | { |
115 | int err; | |
116 | ||
f34f57a3 | 117 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 118 | if (!err) |
2a7dba39 | 119 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
120 | return err; |
121 | } | |
122 | ||
c8b97818 CM |
123 | /* |
124 | * this does all the hard work for inserting an inline extent into | |
125 | * the btree. The caller should have done a btrfs_drop_extents so that | |
126 | * no overlapping inline items exist in the btree | |
127 | */ | |
d397712b | 128 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
129 | struct btrfs_root *root, struct inode *inode, |
130 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 131 | int compress_type, |
c8b97818 CM |
132 | struct page **compressed_pages) |
133 | { | |
134 | struct btrfs_key key; | |
135 | struct btrfs_path *path; | |
136 | struct extent_buffer *leaf; | |
137 | struct page *page = NULL; | |
138 | char *kaddr; | |
139 | unsigned long ptr; | |
140 | struct btrfs_file_extent_item *ei; | |
141 | int err = 0; | |
142 | int ret; | |
143 | size_t cur_size = size; | |
144 | size_t datasize; | |
145 | unsigned long offset; | |
c8b97818 | 146 | |
fe3f566c | 147 | if (compressed_size && compressed_pages) |
c8b97818 | 148 | cur_size = compressed_size; |
c8b97818 | 149 | |
d397712b CM |
150 | path = btrfs_alloc_path(); |
151 | if (!path) | |
c8b97818 CM |
152 | return -ENOMEM; |
153 | ||
b9473439 | 154 | path->leave_spinning = 1; |
c8b97818 | 155 | |
33345d01 | 156 | key.objectid = btrfs_ino(inode); |
c8b97818 CM |
157 | key.offset = start; |
158 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
159 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
160 | ||
161 | inode_add_bytes(inode, size); | |
162 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
163 | datasize); | |
c8b97818 CM |
164 | if (ret) { |
165 | err = ret; | |
c8b97818 CM |
166 | goto fail; |
167 | } | |
168 | leaf = path->nodes[0]; | |
169 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
170 | struct btrfs_file_extent_item); | |
171 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
172 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
173 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
174 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
175 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
176 | ptr = btrfs_file_extent_inline_start(ei); | |
177 | ||
261507a0 | 178 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
179 | struct page *cpage; |
180 | int i = 0; | |
d397712b | 181 | while (compressed_size > 0) { |
c8b97818 | 182 | cpage = compressed_pages[i]; |
5b050f04 | 183 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
184 | PAGE_CACHE_SIZE); |
185 | ||
7ac687d9 | 186 | kaddr = kmap_atomic(cpage); |
c8b97818 | 187 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 188 | kunmap_atomic(kaddr); |
c8b97818 CM |
189 | |
190 | i++; | |
191 | ptr += cur_size; | |
192 | compressed_size -= cur_size; | |
193 | } | |
194 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 195 | compress_type); |
c8b97818 CM |
196 | } else { |
197 | page = find_get_page(inode->i_mapping, | |
198 | start >> PAGE_CACHE_SHIFT); | |
199 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7ac687d9 | 200 | kaddr = kmap_atomic(page); |
c8b97818 CM |
201 | offset = start & (PAGE_CACHE_SIZE - 1); |
202 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
7ac687d9 | 203 | kunmap_atomic(kaddr); |
c8b97818 CM |
204 | page_cache_release(page); |
205 | } | |
206 | btrfs_mark_buffer_dirty(leaf); | |
207 | btrfs_free_path(path); | |
208 | ||
c2167754 YZ |
209 | /* |
210 | * we're an inline extent, so nobody can | |
211 | * extend the file past i_size without locking | |
212 | * a page we already have locked. | |
213 | * | |
214 | * We must do any isize and inode updates | |
215 | * before we unlock the pages. Otherwise we | |
216 | * could end up racing with unlink. | |
217 | */ | |
c8b97818 | 218 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 219 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 220 | |
79787eaa | 221 | return ret; |
c8b97818 CM |
222 | fail: |
223 | btrfs_free_path(path); | |
224 | return err; | |
225 | } | |
226 | ||
227 | ||
228 | /* | |
229 | * conditionally insert an inline extent into the file. This | |
230 | * does the checks required to make sure the data is small enough | |
231 | * to fit as an inline extent. | |
232 | */ | |
7f366cfe | 233 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
234 | struct btrfs_root *root, |
235 | struct inode *inode, u64 start, u64 end, | |
fe3f566c | 236 | size_t compressed_size, int compress_type, |
c8b97818 CM |
237 | struct page **compressed_pages) |
238 | { | |
239 | u64 isize = i_size_read(inode); | |
240 | u64 actual_end = min(end + 1, isize); | |
241 | u64 inline_len = actual_end - start; | |
fda2832f | 242 | u64 aligned_end = ALIGN(end, root->sectorsize); |
c8b97818 CM |
243 | u64 data_len = inline_len; |
244 | int ret; | |
245 | ||
246 | if (compressed_size) | |
247 | data_len = compressed_size; | |
248 | ||
249 | if (start > 0 || | |
70b99e69 | 250 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
251 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
252 | (!compressed_size && | |
253 | (actual_end & (root->sectorsize - 1)) == 0) || | |
254 | end + 1 < isize || | |
255 | data_len > root->fs_info->max_inline) { | |
256 | return 1; | |
257 | } | |
258 | ||
2671485d | 259 | ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1); |
79787eaa JM |
260 | if (ret) |
261 | return ret; | |
c8b97818 CM |
262 | |
263 | if (isize > actual_end) | |
264 | inline_len = min_t(u64, isize, actual_end); | |
265 | ret = insert_inline_extent(trans, root, inode, start, | |
266 | inline_len, compressed_size, | |
fe3f566c | 267 | compress_type, compressed_pages); |
2adcac1a | 268 | if (ret && ret != -ENOSPC) { |
79787eaa JM |
269 | btrfs_abort_transaction(trans, root, ret); |
270 | return ret; | |
2adcac1a JB |
271 | } else if (ret == -ENOSPC) { |
272 | return 1; | |
79787eaa | 273 | } |
2adcac1a | 274 | |
bdc20e67 | 275 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
0ca1f7ce | 276 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 277 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
278 | return 0; |
279 | } | |
280 | ||
771ed689 CM |
281 | struct async_extent { |
282 | u64 start; | |
283 | u64 ram_size; | |
284 | u64 compressed_size; | |
285 | struct page **pages; | |
286 | unsigned long nr_pages; | |
261507a0 | 287 | int compress_type; |
771ed689 CM |
288 | struct list_head list; |
289 | }; | |
290 | ||
291 | struct async_cow { | |
292 | struct inode *inode; | |
293 | struct btrfs_root *root; | |
294 | struct page *locked_page; | |
295 | u64 start; | |
296 | u64 end; | |
297 | struct list_head extents; | |
298 | struct btrfs_work work; | |
299 | }; | |
300 | ||
301 | static noinline int add_async_extent(struct async_cow *cow, | |
302 | u64 start, u64 ram_size, | |
303 | u64 compressed_size, | |
304 | struct page **pages, | |
261507a0 LZ |
305 | unsigned long nr_pages, |
306 | int compress_type) | |
771ed689 CM |
307 | { |
308 | struct async_extent *async_extent; | |
309 | ||
310 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 311 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
312 | async_extent->start = start; |
313 | async_extent->ram_size = ram_size; | |
314 | async_extent->compressed_size = compressed_size; | |
315 | async_extent->pages = pages; | |
316 | async_extent->nr_pages = nr_pages; | |
261507a0 | 317 | async_extent->compress_type = compress_type; |
771ed689 CM |
318 | list_add_tail(&async_extent->list, &cow->extents); |
319 | return 0; | |
320 | } | |
321 | ||
d352ac68 | 322 | /* |
771ed689 CM |
323 | * we create compressed extents in two phases. The first |
324 | * phase compresses a range of pages that have already been | |
325 | * locked (both pages and state bits are locked). | |
c8b97818 | 326 | * |
771ed689 CM |
327 | * This is done inside an ordered work queue, and the compression |
328 | * is spread across many cpus. The actual IO submission is step | |
329 | * two, and the ordered work queue takes care of making sure that | |
330 | * happens in the same order things were put onto the queue by | |
331 | * writepages and friends. | |
c8b97818 | 332 | * |
771ed689 CM |
333 | * If this code finds it can't get good compression, it puts an |
334 | * entry onto the work queue to write the uncompressed bytes. This | |
335 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
336 | * are written in the same order that the flusher thread sent them |
337 | * down. | |
d352ac68 | 338 | */ |
771ed689 CM |
339 | static noinline int compress_file_range(struct inode *inode, |
340 | struct page *locked_page, | |
341 | u64 start, u64 end, | |
342 | struct async_cow *async_cow, | |
343 | int *num_added) | |
b888db2b CM |
344 | { |
345 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
346 | struct btrfs_trans_handle *trans; | |
db94535d | 347 | u64 num_bytes; |
db94535d | 348 | u64 blocksize = root->sectorsize; |
c8b97818 | 349 | u64 actual_end; |
42dc7bab | 350 | u64 isize = i_size_read(inode); |
e6dcd2dc | 351 | int ret = 0; |
c8b97818 CM |
352 | struct page **pages = NULL; |
353 | unsigned long nr_pages; | |
354 | unsigned long nr_pages_ret = 0; | |
355 | unsigned long total_compressed = 0; | |
356 | unsigned long total_in = 0; | |
357 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 358 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
359 | int i; |
360 | int will_compress; | |
261507a0 | 361 | int compress_type = root->fs_info->compress_type; |
4adaa611 | 362 | int redirty = 0; |
b888db2b | 363 | |
4cb13e5d LB |
364 | /* if this is a small write inside eof, kick off a defrag */ |
365 | if ((end - start + 1) < 16 * 1024 && | |
366 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
367 | btrfs_add_inode_defrag(NULL, inode); |
368 | ||
42dc7bab | 369 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
370 | again: |
371 | will_compress = 0; | |
372 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
373 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 374 | |
f03d9301 CM |
375 | /* |
376 | * we don't want to send crud past the end of i_size through | |
377 | * compression, that's just a waste of CPU time. So, if the | |
378 | * end of the file is before the start of our current | |
379 | * requested range of bytes, we bail out to the uncompressed | |
380 | * cleanup code that can deal with all of this. | |
381 | * | |
382 | * It isn't really the fastest way to fix things, but this is a | |
383 | * very uncommon corner. | |
384 | */ | |
385 | if (actual_end <= start) | |
386 | goto cleanup_and_bail_uncompressed; | |
387 | ||
c8b97818 CM |
388 | total_compressed = actual_end - start; |
389 | ||
390 | /* we want to make sure that amount of ram required to uncompress | |
391 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
392 | * of a compressed extent to 128k. This is a crucial number |
393 | * because it also controls how easily we can spread reads across | |
394 | * cpus for decompression. | |
395 | * | |
396 | * We also want to make sure the amount of IO required to do | |
397 | * a random read is reasonably small, so we limit the size of | |
398 | * a compressed extent to 128k. | |
c8b97818 CM |
399 | */ |
400 | total_compressed = min(total_compressed, max_uncompressed); | |
fda2832f | 401 | num_bytes = ALIGN(end - start + 1, blocksize); |
be20aa9d | 402 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
403 | total_in = 0; |
404 | ret = 0; | |
db94535d | 405 | |
771ed689 CM |
406 | /* |
407 | * we do compression for mount -o compress and when the | |
408 | * inode has not been flagged as nocompress. This flag can | |
409 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 410 | */ |
6cbff00f | 411 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 | 412 | (btrfs_test_opt(root, COMPRESS) || |
75e7cb7f LB |
413 | (BTRFS_I(inode)->force_compress) || |
414 | (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) { | |
c8b97818 | 415 | WARN_ON(pages); |
cfbc246e | 416 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
417 | if (!pages) { |
418 | /* just bail out to the uncompressed code */ | |
419 | goto cont; | |
420 | } | |
c8b97818 | 421 | |
261507a0 LZ |
422 | if (BTRFS_I(inode)->force_compress) |
423 | compress_type = BTRFS_I(inode)->force_compress; | |
424 | ||
4adaa611 CM |
425 | /* |
426 | * we need to call clear_page_dirty_for_io on each | |
427 | * page in the range. Otherwise applications with the file | |
428 | * mmap'd can wander in and change the page contents while | |
429 | * we are compressing them. | |
430 | * | |
431 | * If the compression fails for any reason, we set the pages | |
432 | * dirty again later on. | |
433 | */ | |
434 | extent_range_clear_dirty_for_io(inode, start, end); | |
435 | redirty = 1; | |
261507a0 LZ |
436 | ret = btrfs_compress_pages(compress_type, |
437 | inode->i_mapping, start, | |
438 | total_compressed, pages, | |
439 | nr_pages, &nr_pages_ret, | |
440 | &total_in, | |
441 | &total_compressed, | |
442 | max_compressed); | |
c8b97818 CM |
443 | |
444 | if (!ret) { | |
445 | unsigned long offset = total_compressed & | |
446 | (PAGE_CACHE_SIZE - 1); | |
447 | struct page *page = pages[nr_pages_ret - 1]; | |
448 | char *kaddr; | |
449 | ||
450 | /* zero the tail end of the last page, we might be | |
451 | * sending it down to disk | |
452 | */ | |
453 | if (offset) { | |
7ac687d9 | 454 | kaddr = kmap_atomic(page); |
c8b97818 CM |
455 | memset(kaddr + offset, 0, |
456 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 457 | kunmap_atomic(kaddr); |
c8b97818 CM |
458 | } |
459 | will_compress = 1; | |
460 | } | |
461 | } | |
560f7d75 | 462 | cont: |
c8b97818 | 463 | if (start == 0) { |
7a7eaa40 | 464 | trans = btrfs_join_transaction(root); |
79787eaa JM |
465 | if (IS_ERR(trans)) { |
466 | ret = PTR_ERR(trans); | |
467 | trans = NULL; | |
468 | goto cleanup_and_out; | |
469 | } | |
0ca1f7ce | 470 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 471 | |
c8b97818 | 472 | /* lets try to make an inline extent */ |
771ed689 | 473 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 474 | /* we didn't compress the entire range, try |
771ed689 | 475 | * to make an uncompressed inline extent. |
c8b97818 CM |
476 | */ |
477 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 478 | start, end, 0, 0, NULL); |
c8b97818 | 479 | } else { |
771ed689 | 480 | /* try making a compressed inline extent */ |
c8b97818 CM |
481 | ret = cow_file_range_inline(trans, root, inode, |
482 | start, end, | |
fe3f566c LZ |
483 | total_compressed, |
484 | compress_type, pages); | |
c8b97818 | 485 | } |
79787eaa | 486 | if (ret <= 0) { |
151a41bc JB |
487 | unsigned long clear_flags = EXTENT_DELALLOC | |
488 | EXTENT_DEFRAG; | |
489 | clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0; | |
490 | ||
771ed689 | 491 | /* |
79787eaa JM |
492 | * inline extent creation worked or returned error, |
493 | * we don't need to create any more async work items. | |
494 | * Unlock and free up our temp pages. | |
771ed689 | 495 | */ |
c2790a2e | 496 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
151a41bc | 497 | clear_flags, PAGE_UNLOCK | |
c2790a2e JB |
498 | PAGE_CLEAR_DIRTY | |
499 | PAGE_SET_WRITEBACK | | |
500 | PAGE_END_WRITEBACK); | |
c2167754 | 501 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
502 | goto free_pages_out; |
503 | } | |
c2167754 | 504 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
505 | } |
506 | ||
507 | if (will_compress) { | |
508 | /* | |
509 | * we aren't doing an inline extent round the compressed size | |
510 | * up to a block size boundary so the allocator does sane | |
511 | * things | |
512 | */ | |
fda2832f | 513 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
514 | |
515 | /* | |
516 | * one last check to make sure the compression is really a | |
517 | * win, compare the page count read with the blocks on disk | |
518 | */ | |
fda2832f | 519 | total_in = ALIGN(total_in, PAGE_CACHE_SIZE); |
c8b97818 CM |
520 | if (total_compressed >= total_in) { |
521 | will_compress = 0; | |
522 | } else { | |
c8b97818 CM |
523 | num_bytes = total_in; |
524 | } | |
525 | } | |
526 | if (!will_compress && pages) { | |
527 | /* | |
528 | * the compression code ran but failed to make things smaller, | |
529 | * free any pages it allocated and our page pointer array | |
530 | */ | |
531 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 532 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
533 | page_cache_release(pages[i]); |
534 | } | |
535 | kfree(pages); | |
536 | pages = NULL; | |
537 | total_compressed = 0; | |
538 | nr_pages_ret = 0; | |
539 | ||
540 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
541 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
542 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 543 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 544 | } |
c8b97818 | 545 | } |
771ed689 CM |
546 | if (will_compress) { |
547 | *num_added += 1; | |
c8b97818 | 548 | |
771ed689 CM |
549 | /* the async work queues will take care of doing actual |
550 | * allocation on disk for these compressed pages, | |
551 | * and will submit them to the elevator. | |
552 | */ | |
553 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
554 | total_compressed, pages, nr_pages_ret, |
555 | compress_type); | |
179e29e4 | 556 | |
24ae6365 | 557 | if (start + num_bytes < end) { |
771ed689 CM |
558 | start += num_bytes; |
559 | pages = NULL; | |
560 | cond_resched(); | |
561 | goto again; | |
562 | } | |
563 | } else { | |
f03d9301 | 564 | cleanup_and_bail_uncompressed: |
771ed689 CM |
565 | /* |
566 | * No compression, but we still need to write the pages in | |
567 | * the file we've been given so far. redirty the locked | |
568 | * page if it corresponds to our extent and set things up | |
569 | * for the async work queue to run cow_file_range to do | |
570 | * the normal delalloc dance | |
571 | */ | |
572 | if (page_offset(locked_page) >= start && | |
573 | page_offset(locked_page) <= end) { | |
574 | __set_page_dirty_nobuffers(locked_page); | |
575 | /* unlocked later on in the async handlers */ | |
576 | } | |
4adaa611 CM |
577 | if (redirty) |
578 | extent_range_redirty_for_io(inode, start, end); | |
261507a0 LZ |
579 | add_async_extent(async_cow, start, end - start + 1, |
580 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
581 | *num_added += 1; |
582 | } | |
3b951516 | 583 | |
771ed689 | 584 | out: |
79787eaa | 585 | return ret; |
771ed689 CM |
586 | |
587 | free_pages_out: | |
588 | for (i = 0; i < nr_pages_ret; i++) { | |
589 | WARN_ON(pages[i]->mapping); | |
590 | page_cache_release(pages[i]); | |
591 | } | |
d397712b | 592 | kfree(pages); |
771ed689 CM |
593 | |
594 | goto out; | |
79787eaa JM |
595 | |
596 | cleanup_and_out: | |
c2790a2e | 597 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
151a41bc JB |
598 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | |
599 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
600 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | | |
601 | PAGE_END_WRITEBACK); | |
79787eaa JM |
602 | if (!trans || IS_ERR(trans)) |
603 | btrfs_error(root->fs_info, ret, "Failed to join transaction"); | |
604 | else | |
605 | btrfs_abort_transaction(trans, root, ret); | |
606 | goto free_pages_out; | |
771ed689 CM |
607 | } |
608 | ||
609 | /* | |
610 | * phase two of compressed writeback. This is the ordered portion | |
611 | * of the code, which only gets called in the order the work was | |
612 | * queued. We walk all the async extents created by compress_file_range | |
613 | * and send them down to the disk. | |
614 | */ | |
615 | static noinline int submit_compressed_extents(struct inode *inode, | |
616 | struct async_cow *async_cow) | |
617 | { | |
618 | struct async_extent *async_extent; | |
619 | u64 alloc_hint = 0; | |
620 | struct btrfs_trans_handle *trans; | |
621 | struct btrfs_key ins; | |
622 | struct extent_map *em; | |
623 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
624 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
625 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 626 | int ret = 0; |
771ed689 CM |
627 | |
628 | if (list_empty(&async_cow->extents)) | |
629 | return 0; | |
630 | ||
3e04e7f1 | 631 | again: |
d397712b | 632 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
633 | async_extent = list_entry(async_cow->extents.next, |
634 | struct async_extent, list); | |
635 | list_del(&async_extent->list); | |
c8b97818 | 636 | |
771ed689 CM |
637 | io_tree = &BTRFS_I(inode)->io_tree; |
638 | ||
f5a84ee3 | 639 | retry: |
771ed689 CM |
640 | /* did the compression code fall back to uncompressed IO? */ |
641 | if (!async_extent->pages) { | |
642 | int page_started = 0; | |
643 | unsigned long nr_written = 0; | |
644 | ||
645 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 646 | async_extent->start + |
d0082371 | 647 | async_extent->ram_size - 1); |
771ed689 CM |
648 | |
649 | /* allocate blocks */ | |
f5a84ee3 JB |
650 | ret = cow_file_range(inode, async_cow->locked_page, |
651 | async_extent->start, | |
652 | async_extent->start + | |
653 | async_extent->ram_size - 1, | |
654 | &page_started, &nr_written, 0); | |
771ed689 | 655 | |
79787eaa JM |
656 | /* JDM XXX */ |
657 | ||
771ed689 CM |
658 | /* |
659 | * if page_started, cow_file_range inserted an | |
660 | * inline extent and took care of all the unlocking | |
661 | * and IO for us. Otherwise, we need to submit | |
662 | * all those pages down to the drive. | |
663 | */ | |
f5a84ee3 | 664 | if (!page_started && !ret) |
771ed689 CM |
665 | extent_write_locked_range(io_tree, |
666 | inode, async_extent->start, | |
d397712b | 667 | async_extent->start + |
771ed689 CM |
668 | async_extent->ram_size - 1, |
669 | btrfs_get_extent, | |
670 | WB_SYNC_ALL); | |
3e04e7f1 JB |
671 | else if (ret) |
672 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
673 | kfree(async_extent); |
674 | cond_resched(); | |
675 | continue; | |
676 | } | |
677 | ||
678 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 679 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 680 | |
7a7eaa40 | 681 | trans = btrfs_join_transaction(root); |
79787eaa JM |
682 | if (IS_ERR(trans)) { |
683 | ret = PTR_ERR(trans); | |
684 | } else { | |
685 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
686 | ret = btrfs_reserve_extent(trans, root, | |
771ed689 CM |
687 | async_extent->compressed_size, |
688 | async_extent->compressed_size, | |
81c9ad23 | 689 | 0, alloc_hint, &ins, 1); |
962197ba | 690 | if (ret && ret != -ENOSPC) |
79787eaa JM |
691 | btrfs_abort_transaction(trans, root, ret); |
692 | btrfs_end_transaction(trans, root); | |
693 | } | |
c2167754 | 694 | |
f5a84ee3 JB |
695 | if (ret) { |
696 | int i; | |
3e04e7f1 | 697 | |
f5a84ee3 JB |
698 | for (i = 0; i < async_extent->nr_pages; i++) { |
699 | WARN_ON(async_extent->pages[i]->mapping); | |
700 | page_cache_release(async_extent->pages[i]); | |
701 | } | |
702 | kfree(async_extent->pages); | |
703 | async_extent->nr_pages = 0; | |
704 | async_extent->pages = NULL; | |
3e04e7f1 | 705 | |
fdf8e2ea JB |
706 | if (ret == -ENOSPC) { |
707 | unlock_extent(io_tree, async_extent->start, | |
708 | async_extent->start + | |
709 | async_extent->ram_size - 1); | |
79787eaa | 710 | goto retry; |
fdf8e2ea | 711 | } |
3e04e7f1 | 712 | goto out_free; |
f5a84ee3 JB |
713 | } |
714 | ||
c2167754 YZ |
715 | /* |
716 | * here we're doing allocation and writeback of the | |
717 | * compressed pages | |
718 | */ | |
719 | btrfs_drop_extent_cache(inode, async_extent->start, | |
720 | async_extent->start + | |
721 | async_extent->ram_size - 1, 0); | |
722 | ||
172ddd60 | 723 | em = alloc_extent_map(); |
b9aa55be LB |
724 | if (!em) { |
725 | ret = -ENOMEM; | |
3e04e7f1 | 726 | goto out_free_reserve; |
b9aa55be | 727 | } |
771ed689 CM |
728 | em->start = async_extent->start; |
729 | em->len = async_extent->ram_size; | |
445a6944 | 730 | em->orig_start = em->start; |
2ab28f32 JB |
731 | em->mod_start = em->start; |
732 | em->mod_len = em->len; | |
c8b97818 | 733 | |
771ed689 CM |
734 | em->block_start = ins.objectid; |
735 | em->block_len = ins.offset; | |
b4939680 | 736 | em->orig_block_len = ins.offset; |
cc95bef6 | 737 | em->ram_bytes = async_extent->ram_size; |
771ed689 | 738 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
261507a0 | 739 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
740 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
741 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
70c8a91c | 742 | em->generation = -1; |
771ed689 | 743 | |
d397712b | 744 | while (1) { |
890871be | 745 | write_lock(&em_tree->lock); |
09a2a8f9 | 746 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 747 | write_unlock(&em_tree->lock); |
771ed689 CM |
748 | if (ret != -EEXIST) { |
749 | free_extent_map(em); | |
750 | break; | |
751 | } | |
752 | btrfs_drop_extent_cache(inode, async_extent->start, | |
753 | async_extent->start + | |
754 | async_extent->ram_size - 1, 0); | |
755 | } | |
756 | ||
3e04e7f1 JB |
757 | if (ret) |
758 | goto out_free_reserve; | |
759 | ||
261507a0 LZ |
760 | ret = btrfs_add_ordered_extent_compress(inode, |
761 | async_extent->start, | |
762 | ins.objectid, | |
763 | async_extent->ram_size, | |
764 | ins.offset, | |
765 | BTRFS_ORDERED_COMPRESSED, | |
766 | async_extent->compress_type); | |
3e04e7f1 JB |
767 | if (ret) |
768 | goto out_free_reserve; | |
771ed689 | 769 | |
771ed689 CM |
770 | /* |
771 | * clear dirty, set writeback and unlock the pages. | |
772 | */ | |
c2790a2e | 773 | extent_clear_unlock_delalloc(inode, async_extent->start, |
a791e35e CM |
774 | async_extent->start + |
775 | async_extent->ram_size - 1, | |
151a41bc JB |
776 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
777 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 778 | PAGE_SET_WRITEBACK); |
771ed689 | 779 | ret = btrfs_submit_compressed_write(inode, |
d397712b CM |
780 | async_extent->start, |
781 | async_extent->ram_size, | |
782 | ins.objectid, | |
783 | ins.offset, async_extent->pages, | |
784 | async_extent->nr_pages); | |
771ed689 CM |
785 | alloc_hint = ins.objectid + ins.offset; |
786 | kfree(async_extent); | |
3e04e7f1 JB |
787 | if (ret) |
788 | goto out; | |
771ed689 CM |
789 | cond_resched(); |
790 | } | |
79787eaa JM |
791 | ret = 0; |
792 | out: | |
793 | return ret; | |
3e04e7f1 JB |
794 | out_free_reserve: |
795 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
79787eaa | 796 | out_free: |
c2790a2e | 797 | extent_clear_unlock_delalloc(inode, async_extent->start, |
3e04e7f1 JB |
798 | async_extent->start + |
799 | async_extent->ram_size - 1, | |
c2790a2e | 800 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
801 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
802 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
803 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 804 | kfree(async_extent); |
3e04e7f1 | 805 | goto again; |
771ed689 CM |
806 | } |
807 | ||
4b46fce2 JB |
808 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
809 | u64 num_bytes) | |
810 | { | |
811 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
812 | struct extent_map *em; | |
813 | u64 alloc_hint = 0; | |
814 | ||
815 | read_lock(&em_tree->lock); | |
816 | em = search_extent_mapping(em_tree, start, num_bytes); | |
817 | if (em) { | |
818 | /* | |
819 | * if block start isn't an actual block number then find the | |
820 | * first block in this inode and use that as a hint. If that | |
821 | * block is also bogus then just don't worry about it. | |
822 | */ | |
823 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
824 | free_extent_map(em); | |
825 | em = search_extent_mapping(em_tree, 0, 0); | |
826 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
827 | alloc_hint = em->block_start; | |
828 | if (em) | |
829 | free_extent_map(em); | |
830 | } else { | |
831 | alloc_hint = em->block_start; | |
832 | free_extent_map(em); | |
833 | } | |
834 | } | |
835 | read_unlock(&em_tree->lock); | |
836 | ||
837 | return alloc_hint; | |
838 | } | |
839 | ||
771ed689 CM |
840 | /* |
841 | * when extent_io.c finds a delayed allocation range in the file, | |
842 | * the call backs end up in this code. The basic idea is to | |
843 | * allocate extents on disk for the range, and create ordered data structs | |
844 | * in ram to track those extents. | |
845 | * | |
846 | * locked_page is the page that writepage had locked already. We use | |
847 | * it to make sure we don't do extra locks or unlocks. | |
848 | * | |
849 | * *page_started is set to one if we unlock locked_page and do everything | |
850 | * required to start IO on it. It may be clean and already done with | |
851 | * IO when we return. | |
852 | */ | |
b7d5b0a8 MX |
853 | static noinline int __cow_file_range(struct btrfs_trans_handle *trans, |
854 | struct inode *inode, | |
855 | struct btrfs_root *root, | |
856 | struct page *locked_page, | |
857 | u64 start, u64 end, int *page_started, | |
858 | unsigned long *nr_written, | |
859 | int unlock) | |
771ed689 | 860 | { |
771ed689 CM |
861 | u64 alloc_hint = 0; |
862 | u64 num_bytes; | |
863 | unsigned long ram_size; | |
864 | u64 disk_num_bytes; | |
865 | u64 cur_alloc_size; | |
866 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
867 | struct btrfs_key ins; |
868 | struct extent_map *em; | |
869 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
870 | int ret = 0; | |
871 | ||
83eea1f1 | 872 | BUG_ON(btrfs_is_free_space_inode(inode)); |
771ed689 | 873 | |
fda2832f | 874 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
875 | num_bytes = max(blocksize, num_bytes); |
876 | disk_num_bytes = num_bytes; | |
771ed689 | 877 | |
4cb5300b | 878 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
879 | if (num_bytes < 64 * 1024 && |
880 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
881 | btrfs_add_inode_defrag(trans, inode); |
882 | ||
771ed689 CM |
883 | if (start == 0) { |
884 | /* lets try to make an inline extent */ | |
885 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 886 | start, end, 0, 0, NULL); |
771ed689 | 887 | if (ret == 0) { |
c2790a2e JB |
888 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
889 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc | 890 | EXTENT_DEFRAG, PAGE_UNLOCK | |
c2790a2e JB |
891 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
892 | PAGE_END_WRITEBACK); | |
c2167754 | 893 | |
771ed689 CM |
894 | *nr_written = *nr_written + |
895 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
896 | *page_started = 1; | |
771ed689 | 897 | goto out; |
79787eaa JM |
898 | } else if (ret < 0) { |
899 | btrfs_abort_transaction(trans, root, ret); | |
900 | goto out_unlock; | |
771ed689 CM |
901 | } |
902 | } | |
903 | ||
904 | BUG_ON(disk_num_bytes > | |
6c41761f | 905 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 906 | |
4b46fce2 | 907 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
908 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
909 | ||
d397712b | 910 | while (disk_num_bytes > 0) { |
a791e35e CM |
911 | unsigned long op; |
912 | ||
287a0ab9 | 913 | cur_alloc_size = disk_num_bytes; |
e6dcd2dc | 914 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 915 | root->sectorsize, 0, alloc_hint, |
81c9ad23 | 916 | &ins, 1); |
79787eaa JM |
917 | if (ret < 0) { |
918 | btrfs_abort_transaction(trans, root, ret); | |
919 | goto out_unlock; | |
920 | } | |
d397712b | 921 | |
172ddd60 | 922 | em = alloc_extent_map(); |
b9aa55be LB |
923 | if (!em) { |
924 | ret = -ENOMEM; | |
ace68bac | 925 | goto out_reserve; |
b9aa55be | 926 | } |
e6dcd2dc | 927 | em->start = start; |
445a6944 | 928 | em->orig_start = em->start; |
771ed689 CM |
929 | ram_size = ins.offset; |
930 | em->len = ins.offset; | |
2ab28f32 JB |
931 | em->mod_start = em->start; |
932 | em->mod_len = em->len; | |
c8b97818 | 933 | |
e6dcd2dc | 934 | em->block_start = ins.objectid; |
c8b97818 | 935 | em->block_len = ins.offset; |
b4939680 | 936 | em->orig_block_len = ins.offset; |
cc95bef6 | 937 | em->ram_bytes = ram_size; |
e6dcd2dc | 938 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 939 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
70c8a91c | 940 | em->generation = -1; |
c8b97818 | 941 | |
d397712b | 942 | while (1) { |
890871be | 943 | write_lock(&em_tree->lock); |
09a2a8f9 | 944 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 945 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
946 | if (ret != -EEXIST) { |
947 | free_extent_map(em); | |
948 | break; | |
949 | } | |
950 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 951 | start + ram_size - 1, 0); |
e6dcd2dc | 952 | } |
ace68bac LB |
953 | if (ret) |
954 | goto out_reserve; | |
e6dcd2dc | 955 | |
98d20f67 | 956 | cur_alloc_size = ins.offset; |
e6dcd2dc | 957 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 958 | ram_size, cur_alloc_size, 0); |
ace68bac LB |
959 | if (ret) |
960 | goto out_reserve; | |
c8b97818 | 961 | |
17d217fe YZ |
962 | if (root->root_key.objectid == |
963 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
964 | ret = btrfs_reloc_clone_csums(inode, start, | |
965 | cur_alloc_size); | |
79787eaa JM |
966 | if (ret) { |
967 | btrfs_abort_transaction(trans, root, ret); | |
ace68bac | 968 | goto out_reserve; |
79787eaa | 969 | } |
17d217fe YZ |
970 | } |
971 | ||
d397712b | 972 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 973 | break; |
d397712b | 974 | |
c8b97818 CM |
975 | /* we're not doing compressed IO, don't unlock the first |
976 | * page (which the caller expects to stay locked), don't | |
977 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
978 | * |
979 | * Do set the Private2 bit so we know this page was properly | |
980 | * setup for writepage | |
c8b97818 | 981 | */ |
c2790a2e JB |
982 | op = unlock ? PAGE_UNLOCK : 0; |
983 | op |= PAGE_SET_PRIVATE2; | |
a791e35e | 984 | |
c2790a2e JB |
985 | extent_clear_unlock_delalloc(inode, start, |
986 | start + ram_size - 1, locked_page, | |
987 | EXTENT_LOCKED | EXTENT_DELALLOC, | |
988 | op); | |
c8b97818 | 989 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
990 | num_bytes -= cur_alloc_size; |
991 | alloc_hint = ins.objectid + ins.offset; | |
992 | start += cur_alloc_size; | |
b888db2b | 993 | } |
79787eaa | 994 | out: |
be20aa9d | 995 | return ret; |
b7d5b0a8 | 996 | |
ace68bac LB |
997 | out_reserve: |
998 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
79787eaa | 999 | out_unlock: |
c2790a2e | 1000 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
151a41bc JB |
1001 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
1002 | EXTENT_DELALLOC | EXTENT_DEFRAG, | |
1003 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
1004 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 1005 | goto out; |
771ed689 | 1006 | } |
c8b97818 | 1007 | |
b7d5b0a8 MX |
1008 | static noinline int cow_file_range(struct inode *inode, |
1009 | struct page *locked_page, | |
1010 | u64 start, u64 end, int *page_started, | |
1011 | unsigned long *nr_written, | |
1012 | int unlock) | |
1013 | { | |
1014 | struct btrfs_trans_handle *trans; | |
1015 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1016 | int ret; | |
1017 | ||
1018 | trans = btrfs_join_transaction(root); | |
1019 | if (IS_ERR(trans)) { | |
c2790a2e JB |
1020 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1021 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1022 | EXTENT_DO_ACCOUNTING | |
1023 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1024 | PAGE_CLEAR_DIRTY | |
1025 | PAGE_SET_WRITEBACK | | |
1026 | PAGE_END_WRITEBACK); | |
b7d5b0a8 MX |
1027 | return PTR_ERR(trans); |
1028 | } | |
1029 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
1030 | ||
1031 | ret = __cow_file_range(trans, inode, root, locked_page, start, end, | |
1032 | page_started, nr_written, unlock); | |
1033 | ||
1034 | btrfs_end_transaction(trans, root); | |
1035 | ||
1036 | return ret; | |
1037 | } | |
1038 | ||
771ed689 CM |
1039 | /* |
1040 | * work queue call back to started compression on a file and pages | |
1041 | */ | |
1042 | static noinline void async_cow_start(struct btrfs_work *work) | |
1043 | { | |
1044 | struct async_cow *async_cow; | |
1045 | int num_added = 0; | |
1046 | async_cow = container_of(work, struct async_cow, work); | |
1047 | ||
1048 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1049 | async_cow->start, async_cow->end, async_cow, | |
1050 | &num_added); | |
8180ef88 | 1051 | if (num_added == 0) { |
cb77fcd8 | 1052 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1053 | async_cow->inode = NULL; |
8180ef88 | 1054 | } |
771ed689 CM |
1055 | } |
1056 | ||
1057 | /* | |
1058 | * work queue call back to submit previously compressed pages | |
1059 | */ | |
1060 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1061 | { | |
1062 | struct async_cow *async_cow; | |
1063 | struct btrfs_root *root; | |
1064 | unsigned long nr_pages; | |
1065 | ||
1066 | async_cow = container_of(work, struct async_cow, work); | |
1067 | ||
1068 | root = async_cow->root; | |
1069 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1070 | PAGE_CACHE_SHIFT; | |
1071 | ||
66657b31 | 1072 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
287082b0 | 1073 | 5 * 1024 * 1024 && |
771ed689 CM |
1074 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1075 | wake_up(&root->fs_info->async_submit_wait); | |
1076 | ||
d397712b | 1077 | if (async_cow->inode) |
771ed689 | 1078 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1079 | } |
c8b97818 | 1080 | |
771ed689 CM |
1081 | static noinline void async_cow_free(struct btrfs_work *work) |
1082 | { | |
1083 | struct async_cow *async_cow; | |
1084 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1085 | if (async_cow->inode) |
cb77fcd8 | 1086 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1087 | kfree(async_cow); |
1088 | } | |
1089 | ||
1090 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1091 | u64 start, u64 end, int *page_started, | |
1092 | unsigned long *nr_written) | |
1093 | { | |
1094 | struct async_cow *async_cow; | |
1095 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1096 | unsigned long nr_pages; | |
1097 | u64 cur_end; | |
287082b0 | 1098 | int limit = 10 * 1024 * 1024; |
771ed689 | 1099 | |
a3429ab7 CM |
1100 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1101 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1102 | while (start < end) { |
771ed689 | 1103 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1104 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1105 | async_cow->inode = igrab(inode); |
771ed689 CM |
1106 | async_cow->root = root; |
1107 | async_cow->locked_page = locked_page; | |
1108 | async_cow->start = start; | |
1109 | ||
6cbff00f | 1110 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
1111 | cur_end = end; |
1112 | else | |
1113 | cur_end = min(end, start + 512 * 1024 - 1); | |
1114 | ||
1115 | async_cow->end = cur_end; | |
1116 | INIT_LIST_HEAD(&async_cow->extents); | |
1117 | ||
1118 | async_cow->work.func = async_cow_start; | |
1119 | async_cow->work.ordered_func = async_cow_submit; | |
1120 | async_cow->work.ordered_free = async_cow_free; | |
1121 | async_cow->work.flags = 0; | |
1122 | ||
771ed689 CM |
1123 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1124 | PAGE_CACHE_SHIFT; | |
1125 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1126 | ||
1127 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
1128 | &async_cow->work); | |
1129 | ||
1130 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1131 | wait_event(root->fs_info->async_submit_wait, | |
1132 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1133 | limit)); | |
1134 | } | |
1135 | ||
d397712b | 1136 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1137 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1138 | wait_event(root->fs_info->async_submit_wait, | |
1139 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1140 | 0)); | |
1141 | } | |
1142 | ||
1143 | *nr_written += nr_pages; | |
1144 | start = cur_end + 1; | |
1145 | } | |
1146 | *page_started = 1; | |
1147 | return 0; | |
be20aa9d CM |
1148 | } |
1149 | ||
d397712b | 1150 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1151 | u64 bytenr, u64 num_bytes) |
1152 | { | |
1153 | int ret; | |
1154 | struct btrfs_ordered_sum *sums; | |
1155 | LIST_HEAD(list); | |
1156 | ||
07d400a6 | 1157 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1158 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1159 | if (ret == 0 && list_empty(&list)) |
1160 | return 0; | |
1161 | ||
1162 | while (!list_empty(&list)) { | |
1163 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1164 | list_del(&sums->list); | |
1165 | kfree(sums); | |
1166 | } | |
1167 | return 1; | |
1168 | } | |
1169 | ||
d352ac68 CM |
1170 | /* |
1171 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1172 | * of the extents that exist in the file, and COWs the file as required. | |
1173 | * | |
1174 | * If no cow copies or snapshots exist, we write directly to the existing | |
1175 | * blocks on disk | |
1176 | */ | |
7f366cfe CM |
1177 | static noinline int run_delalloc_nocow(struct inode *inode, |
1178 | struct page *locked_page, | |
771ed689 CM |
1179 | u64 start, u64 end, int *page_started, int force, |
1180 | unsigned long *nr_written) | |
be20aa9d | 1181 | { |
be20aa9d | 1182 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1183 | struct btrfs_trans_handle *trans; |
be20aa9d | 1184 | struct extent_buffer *leaf; |
be20aa9d | 1185 | struct btrfs_path *path; |
80ff3856 | 1186 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1187 | struct btrfs_key found_key; |
80ff3856 YZ |
1188 | u64 cow_start; |
1189 | u64 cur_offset; | |
1190 | u64 extent_end; | |
5d4f98a2 | 1191 | u64 extent_offset; |
80ff3856 YZ |
1192 | u64 disk_bytenr; |
1193 | u64 num_bytes; | |
b4939680 | 1194 | u64 disk_num_bytes; |
cc95bef6 | 1195 | u64 ram_bytes; |
80ff3856 | 1196 | int extent_type; |
79787eaa | 1197 | int ret, err; |
d899e052 | 1198 | int type; |
80ff3856 YZ |
1199 | int nocow; |
1200 | int check_prev = 1; | |
82d5902d | 1201 | bool nolock; |
33345d01 | 1202 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1203 | |
1204 | path = btrfs_alloc_path(); | |
17ca04af | 1205 | if (!path) { |
c2790a2e JB |
1206 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1207 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1208 | EXTENT_DO_ACCOUNTING | |
1209 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1210 | PAGE_CLEAR_DIRTY | |
1211 | PAGE_SET_WRITEBACK | | |
1212 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1213 | return -ENOMEM; |
17ca04af | 1214 | } |
82d5902d | 1215 | |
83eea1f1 | 1216 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1217 | |
1218 | if (nolock) | |
7a7eaa40 | 1219 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1220 | else |
7a7eaa40 | 1221 | trans = btrfs_join_transaction(root); |
ff5714cc | 1222 | |
79787eaa | 1223 | if (IS_ERR(trans)) { |
c2790a2e JB |
1224 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1225 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1226 | EXTENT_DO_ACCOUNTING | |
1227 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1228 | PAGE_CLEAR_DIRTY | |
1229 | PAGE_SET_WRITEBACK | | |
1230 | PAGE_END_WRITEBACK); | |
79787eaa JM |
1231 | btrfs_free_path(path); |
1232 | return PTR_ERR(trans); | |
1233 | } | |
1234 | ||
74b21075 | 1235 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1236 | |
80ff3856 YZ |
1237 | cow_start = (u64)-1; |
1238 | cur_offset = start; | |
1239 | while (1) { | |
33345d01 | 1240 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1241 | cur_offset, 0); |
79787eaa JM |
1242 | if (ret < 0) { |
1243 | btrfs_abort_transaction(trans, root, ret); | |
1244 | goto error; | |
1245 | } | |
80ff3856 YZ |
1246 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1247 | leaf = path->nodes[0]; | |
1248 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1249 | path->slots[0] - 1); | |
33345d01 | 1250 | if (found_key.objectid == ino && |
80ff3856 YZ |
1251 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1252 | path->slots[0]--; | |
1253 | } | |
1254 | check_prev = 0; | |
1255 | next_slot: | |
1256 | leaf = path->nodes[0]; | |
1257 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1258 | ret = btrfs_next_leaf(root, path); | |
79787eaa JM |
1259 | if (ret < 0) { |
1260 | btrfs_abort_transaction(trans, root, ret); | |
1261 | goto error; | |
1262 | } | |
80ff3856 YZ |
1263 | if (ret > 0) |
1264 | break; | |
1265 | leaf = path->nodes[0]; | |
1266 | } | |
be20aa9d | 1267 | |
80ff3856 YZ |
1268 | nocow = 0; |
1269 | disk_bytenr = 0; | |
17d217fe | 1270 | num_bytes = 0; |
80ff3856 YZ |
1271 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1272 | ||
33345d01 | 1273 | if (found_key.objectid > ino || |
80ff3856 YZ |
1274 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1275 | found_key.offset > end) | |
1276 | break; | |
1277 | ||
1278 | if (found_key.offset > cur_offset) { | |
1279 | extent_end = found_key.offset; | |
e9061e21 | 1280 | extent_type = 0; |
80ff3856 YZ |
1281 | goto out_check; |
1282 | } | |
1283 | ||
1284 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1285 | struct btrfs_file_extent_item); | |
1286 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1287 | ||
cc95bef6 | 1288 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1289 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1290 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1291 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1292 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1293 | extent_end = found_key.offset + |
1294 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1295 | disk_num_bytes = |
1296 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1297 | if (extent_end <= start) { |
1298 | path->slots[0]++; | |
1299 | goto next_slot; | |
1300 | } | |
17d217fe YZ |
1301 | if (disk_bytenr == 0) |
1302 | goto out_check; | |
80ff3856 YZ |
1303 | if (btrfs_file_extent_compression(leaf, fi) || |
1304 | btrfs_file_extent_encryption(leaf, fi) || | |
1305 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1306 | goto out_check; | |
d899e052 YZ |
1307 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1308 | goto out_check; | |
d2fb3437 | 1309 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1310 | goto out_check; |
33345d01 | 1311 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1312 | found_key.offset - |
1313 | extent_offset, disk_bytenr)) | |
17d217fe | 1314 | goto out_check; |
5d4f98a2 | 1315 | disk_bytenr += extent_offset; |
17d217fe YZ |
1316 | disk_bytenr += cur_offset - found_key.offset; |
1317 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1318 | /* | |
1319 | * force cow if csum exists in the range. | |
1320 | * this ensure that csum for a given extent are | |
1321 | * either valid or do not exist. | |
1322 | */ | |
1323 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1324 | goto out_check; | |
80ff3856 YZ |
1325 | nocow = 1; |
1326 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1327 | extent_end = found_key.offset + | |
1328 | btrfs_file_extent_inline_len(leaf, fi); | |
1329 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1330 | } else { | |
1331 | BUG_ON(1); | |
1332 | } | |
1333 | out_check: | |
1334 | if (extent_end <= start) { | |
1335 | path->slots[0]++; | |
1336 | goto next_slot; | |
1337 | } | |
1338 | if (!nocow) { | |
1339 | if (cow_start == (u64)-1) | |
1340 | cow_start = cur_offset; | |
1341 | cur_offset = extent_end; | |
1342 | if (cur_offset > end) | |
1343 | break; | |
1344 | path->slots[0]++; | |
1345 | goto next_slot; | |
7ea394f1 YZ |
1346 | } |
1347 | ||
b3b4aa74 | 1348 | btrfs_release_path(path); |
80ff3856 | 1349 | if (cow_start != (u64)-1) { |
b7d5b0a8 MX |
1350 | ret = __cow_file_range(trans, inode, root, locked_page, |
1351 | cow_start, found_key.offset - 1, | |
1352 | page_started, nr_written, 1); | |
79787eaa JM |
1353 | if (ret) { |
1354 | btrfs_abort_transaction(trans, root, ret); | |
1355 | goto error; | |
1356 | } | |
80ff3856 | 1357 | cow_start = (u64)-1; |
7ea394f1 | 1358 | } |
80ff3856 | 1359 | |
d899e052 YZ |
1360 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1361 | struct extent_map *em; | |
1362 | struct extent_map_tree *em_tree; | |
1363 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1364 | em = alloc_extent_map(); |
79787eaa | 1365 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1366 | em->start = cur_offset; |
70c8a91c | 1367 | em->orig_start = found_key.offset - extent_offset; |
d899e052 YZ |
1368 | em->len = num_bytes; |
1369 | em->block_len = num_bytes; | |
1370 | em->block_start = disk_bytenr; | |
b4939680 | 1371 | em->orig_block_len = disk_num_bytes; |
cc95bef6 | 1372 | em->ram_bytes = ram_bytes; |
d899e052 | 1373 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
2ab28f32 JB |
1374 | em->mod_start = em->start; |
1375 | em->mod_len = em->len; | |
d899e052 | 1376 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
b11e234d | 1377 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
70c8a91c | 1378 | em->generation = -1; |
d899e052 | 1379 | while (1) { |
890871be | 1380 | write_lock(&em_tree->lock); |
09a2a8f9 | 1381 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1382 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1383 | if (ret != -EEXIST) { |
1384 | free_extent_map(em); | |
1385 | break; | |
1386 | } | |
1387 | btrfs_drop_extent_cache(inode, em->start, | |
1388 | em->start + em->len - 1, 0); | |
1389 | } | |
1390 | type = BTRFS_ORDERED_PREALLOC; | |
1391 | } else { | |
1392 | type = BTRFS_ORDERED_NOCOW; | |
1393 | } | |
80ff3856 YZ |
1394 | |
1395 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1396 | num_bytes, num_bytes, type); |
79787eaa | 1397 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1398 | |
efa56464 YZ |
1399 | if (root->root_key.objectid == |
1400 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1401 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1402 | num_bytes); | |
79787eaa JM |
1403 | if (ret) { |
1404 | btrfs_abort_transaction(trans, root, ret); | |
1405 | goto error; | |
1406 | } | |
efa56464 YZ |
1407 | } |
1408 | ||
c2790a2e JB |
1409 | extent_clear_unlock_delalloc(inode, cur_offset, |
1410 | cur_offset + num_bytes - 1, | |
1411 | locked_page, EXTENT_LOCKED | | |
1412 | EXTENT_DELALLOC, PAGE_UNLOCK | | |
1413 | PAGE_SET_PRIVATE2); | |
80ff3856 YZ |
1414 | cur_offset = extent_end; |
1415 | if (cur_offset > end) | |
1416 | break; | |
be20aa9d | 1417 | } |
b3b4aa74 | 1418 | btrfs_release_path(path); |
80ff3856 | 1419 | |
17ca04af | 1420 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1421 | cow_start = cur_offset; |
17ca04af JB |
1422 | cur_offset = end; |
1423 | } | |
1424 | ||
80ff3856 | 1425 | if (cow_start != (u64)-1) { |
b7d5b0a8 MX |
1426 | ret = __cow_file_range(trans, inode, root, locked_page, |
1427 | cow_start, end, | |
1428 | page_started, nr_written, 1); | |
79787eaa JM |
1429 | if (ret) { |
1430 | btrfs_abort_transaction(trans, root, ret); | |
1431 | goto error; | |
1432 | } | |
80ff3856 YZ |
1433 | } |
1434 | ||
79787eaa | 1435 | error: |
a698d075 | 1436 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1437 | if (!ret) |
1438 | ret = err; | |
1439 | ||
17ca04af | 1440 | if (ret && cur_offset < end) |
c2790a2e JB |
1441 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
1442 | locked_page, EXTENT_LOCKED | | |
151a41bc JB |
1443 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1444 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1445 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1446 | PAGE_SET_WRITEBACK | |
1447 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1448 | btrfs_free_path(path); |
79787eaa | 1449 | return ret; |
be20aa9d CM |
1450 | } |
1451 | ||
d352ac68 CM |
1452 | /* |
1453 | * extent_io.c call back to do delayed allocation processing | |
1454 | */ | |
c8b97818 | 1455 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1456 | u64 start, u64 end, int *page_started, |
1457 | unsigned long *nr_written) | |
be20aa9d | 1458 | { |
be20aa9d | 1459 | int ret; |
7f366cfe | 1460 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1461 | |
7ddf5a42 | 1462 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) { |
c8b97818 | 1463 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1464 | page_started, 1, nr_written); |
7ddf5a42 | 1465 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) { |
d899e052 | 1466 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1467 | page_started, 0, nr_written); |
7ddf5a42 JB |
1468 | } else if (!btrfs_test_opt(root, COMPRESS) && |
1469 | !(BTRFS_I(inode)->force_compress) && | |
1470 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) { | |
7f366cfe CM |
1471 | ret = cow_file_range(inode, locked_page, start, end, |
1472 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1473 | } else { |
1474 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1475 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1476 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1477 | page_started, nr_written); |
7ddf5a42 | 1478 | } |
b888db2b CM |
1479 | return ret; |
1480 | } | |
1481 | ||
1bf85046 JM |
1482 | static void btrfs_split_extent_hook(struct inode *inode, |
1483 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1484 | { |
0ca1f7ce | 1485 | /* not delalloc, ignore it */ |
9ed74f2d | 1486 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1487 | return; |
9ed74f2d | 1488 | |
9e0baf60 JB |
1489 | spin_lock(&BTRFS_I(inode)->lock); |
1490 | BTRFS_I(inode)->outstanding_extents++; | |
1491 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1492 | } |
1493 | ||
1494 | /* | |
1495 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1496 | * extents so we can keep track of new extents that are just merged onto old | |
1497 | * extents, such as when we are doing sequential writes, so we can properly | |
1498 | * account for the metadata space we'll need. | |
1499 | */ | |
1bf85046 JM |
1500 | static void btrfs_merge_extent_hook(struct inode *inode, |
1501 | struct extent_state *new, | |
1502 | struct extent_state *other) | |
9ed74f2d | 1503 | { |
9ed74f2d JB |
1504 | /* not delalloc, ignore it */ |
1505 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1506 | return; |
9ed74f2d | 1507 | |
9e0baf60 JB |
1508 | spin_lock(&BTRFS_I(inode)->lock); |
1509 | BTRFS_I(inode)->outstanding_extents--; | |
1510 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1511 | } |
1512 | ||
eb73c1b7 MX |
1513 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1514 | struct inode *inode) | |
1515 | { | |
1516 | spin_lock(&root->delalloc_lock); | |
1517 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1518 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1519 | &root->delalloc_inodes); | |
1520 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1521 | &BTRFS_I(inode)->runtime_flags); | |
1522 | root->nr_delalloc_inodes++; | |
1523 | if (root->nr_delalloc_inodes == 1) { | |
1524 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1525 | BUG_ON(!list_empty(&root->delalloc_root)); | |
1526 | list_add_tail(&root->delalloc_root, | |
1527 | &root->fs_info->delalloc_roots); | |
1528 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1529 | } | |
1530 | } | |
1531 | spin_unlock(&root->delalloc_lock); | |
1532 | } | |
1533 | ||
1534 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1535 | struct inode *inode) | |
1536 | { | |
1537 | spin_lock(&root->delalloc_lock); | |
1538 | if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1539 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1540 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1541 | &BTRFS_I(inode)->runtime_flags); | |
1542 | root->nr_delalloc_inodes--; | |
1543 | if (!root->nr_delalloc_inodes) { | |
1544 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1545 | BUG_ON(list_empty(&root->delalloc_root)); | |
1546 | list_del_init(&root->delalloc_root); | |
1547 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1548 | } | |
1549 | } | |
1550 | spin_unlock(&root->delalloc_lock); | |
1551 | } | |
1552 | ||
d352ac68 CM |
1553 | /* |
1554 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1555 | * bytes in this file, and to maintain the list of inodes that | |
1556 | * have pending delalloc work to be done. | |
1557 | */ | |
1bf85046 | 1558 | static void btrfs_set_bit_hook(struct inode *inode, |
41074888 | 1559 | struct extent_state *state, unsigned long *bits) |
291d673e | 1560 | { |
9ed74f2d | 1561 | |
75eff68e CM |
1562 | /* |
1563 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1564 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1565 | * bit, which is only set or cleared with irqs on |
1566 | */ | |
0ca1f7ce | 1567 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1568 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1569 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1570 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1571 | |
9e0baf60 | 1572 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1573 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1574 | } else { |
1575 | spin_lock(&BTRFS_I(inode)->lock); | |
1576 | BTRFS_I(inode)->outstanding_extents++; | |
1577 | spin_unlock(&BTRFS_I(inode)->lock); | |
1578 | } | |
287a0ab9 | 1579 | |
963d678b MX |
1580 | __percpu_counter_add(&root->fs_info->delalloc_bytes, len, |
1581 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1582 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1583 | BTRFS_I(inode)->delalloc_bytes += len; |
df0af1a5 | 1584 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1585 | &BTRFS_I(inode)->runtime_flags)) |
1586 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1587 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1588 | } |
291d673e CM |
1589 | } |
1590 | ||
d352ac68 CM |
1591 | /* |
1592 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1593 | */ | |
1bf85046 | 1594 | static void btrfs_clear_bit_hook(struct inode *inode, |
41074888 DS |
1595 | struct extent_state *state, |
1596 | unsigned long *bits) | |
291d673e | 1597 | { |
75eff68e CM |
1598 | /* |
1599 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1600 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1601 | * bit, which is only set or cleared with irqs on |
1602 | */ | |
0ca1f7ce | 1603 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1604 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1605 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1606 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1607 | |
9e0baf60 | 1608 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1609 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1610 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1611 | spin_lock(&BTRFS_I(inode)->lock); | |
1612 | BTRFS_I(inode)->outstanding_extents--; | |
1613 | spin_unlock(&BTRFS_I(inode)->lock); | |
1614 | } | |
0ca1f7ce YZ |
1615 | |
1616 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1617 | btrfs_delalloc_release_metadata(inode, len); | |
1618 | ||
0cb59c99 | 1619 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
7ee9e440 | 1620 | && do_list && !(state->state & EXTENT_NORESERVE)) |
0ca1f7ce | 1621 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1622 | |
963d678b MX |
1623 | __percpu_counter_add(&root->fs_info->delalloc_bytes, -len, |
1624 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1625 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1626 | BTRFS_I(inode)->delalloc_bytes -= len; |
0cb59c99 | 1627 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
df0af1a5 | 1628 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1629 | &BTRFS_I(inode)->runtime_flags)) |
1630 | btrfs_del_delalloc_inode(root, inode); | |
df0af1a5 | 1631 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1632 | } |
291d673e CM |
1633 | } |
1634 | ||
d352ac68 CM |
1635 | /* |
1636 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1637 | * we don't create bios that span stripes or chunks | |
1638 | */ | |
64a16701 | 1639 | int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset, |
c8b97818 CM |
1640 | size_t size, struct bio *bio, |
1641 | unsigned long bio_flags) | |
239b14b3 CM |
1642 | { |
1643 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
a62b9401 | 1644 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1645 | u64 length = 0; |
1646 | u64 map_length; | |
239b14b3 CM |
1647 | int ret; |
1648 | ||
771ed689 CM |
1649 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1650 | return 0; | |
1651 | ||
f2d8d74d | 1652 | length = bio->bi_size; |
239b14b3 | 1653 | map_length = length; |
64a16701 | 1654 | ret = btrfs_map_block(root->fs_info, rw, logical, |
f188591e | 1655 | &map_length, NULL, 0); |
3ec706c8 | 1656 | /* Will always return 0 with map_multi == NULL */ |
3444a972 | 1657 | BUG_ON(ret < 0); |
d397712b | 1658 | if (map_length < length + size) |
239b14b3 | 1659 | return 1; |
3444a972 | 1660 | return 0; |
239b14b3 CM |
1661 | } |
1662 | ||
d352ac68 CM |
1663 | /* |
1664 | * in order to insert checksums into the metadata in large chunks, | |
1665 | * we wait until bio submission time. All the pages in the bio are | |
1666 | * checksummed and sums are attached onto the ordered extent record. | |
1667 | * | |
1668 | * At IO completion time the cums attached on the ordered extent record | |
1669 | * are inserted into the btree | |
1670 | */ | |
d397712b CM |
1671 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1672 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1673 | unsigned long bio_flags, |
1674 | u64 bio_offset) | |
065631f6 | 1675 | { |
065631f6 | 1676 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1677 | int ret = 0; |
e015640f | 1678 | |
d20f7043 | 1679 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1680 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1681 | return 0; |
1682 | } | |
e015640f | 1683 | |
4a69a410 CM |
1684 | /* |
1685 | * in order to insert checksums into the metadata in large chunks, | |
1686 | * we wait until bio submission time. All the pages in the bio are | |
1687 | * checksummed and sums are attached onto the ordered extent record. | |
1688 | * | |
1689 | * At IO completion time the cums attached on the ordered extent record | |
1690 | * are inserted into the btree | |
1691 | */ | |
b2950863 | 1692 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1693 | int mirror_num, unsigned long bio_flags, |
1694 | u64 bio_offset) | |
4a69a410 CM |
1695 | { |
1696 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
61891923 SB |
1697 | int ret; |
1698 | ||
1699 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 1); | |
1700 | if (ret) | |
1701 | bio_endio(bio, ret); | |
1702 | return ret; | |
44b8bd7e CM |
1703 | } |
1704 | ||
d352ac68 | 1705 | /* |
cad321ad CM |
1706 | * extent_io.c submission hook. This does the right thing for csum calculation |
1707 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1708 | */ |
b2950863 | 1709 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1710 | int mirror_num, unsigned long bio_flags, |
1711 | u64 bio_offset) | |
44b8bd7e CM |
1712 | { |
1713 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1714 | int ret = 0; | |
19b9bdb0 | 1715 | int skip_sum; |
0417341e | 1716 | int metadata = 0; |
b812ce28 | 1717 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1718 | |
6cbff00f | 1719 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1720 | |
83eea1f1 | 1721 | if (btrfs_is_free_space_inode(inode)) |
0417341e JM |
1722 | metadata = 2; |
1723 | ||
7b6d91da | 1724 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1725 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1726 | if (ret) | |
61891923 | 1727 | goto out; |
5fd02043 | 1728 | |
d20f7043 | 1729 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1730 | ret = btrfs_submit_compressed_read(inode, bio, |
1731 | mirror_num, | |
1732 | bio_flags); | |
1733 | goto out; | |
c2db1073 TI |
1734 | } else if (!skip_sum) { |
1735 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1736 | if (ret) | |
61891923 | 1737 | goto out; |
c2db1073 | 1738 | } |
4d1b5fb4 | 1739 | goto mapit; |
b812ce28 | 1740 | } else if (async && !skip_sum) { |
17d217fe YZ |
1741 | /* csum items have already been cloned */ |
1742 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1743 | goto mapit; | |
19b9bdb0 | 1744 | /* we're doing a write, do the async checksumming */ |
61891923 | 1745 | ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, |
44b8bd7e | 1746 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1747 | bio_flags, bio_offset, |
1748 | __btrfs_submit_bio_start, | |
4a69a410 | 1749 | __btrfs_submit_bio_done); |
61891923 | 1750 | goto out; |
b812ce28 JB |
1751 | } else if (!skip_sum) { |
1752 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); | |
1753 | if (ret) | |
1754 | goto out; | |
19b9bdb0 CM |
1755 | } |
1756 | ||
0b86a832 | 1757 | mapit: |
61891923 SB |
1758 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); |
1759 | ||
1760 | out: | |
1761 | if (ret < 0) | |
1762 | bio_endio(bio, ret); | |
1763 | return ret; | |
065631f6 | 1764 | } |
6885f308 | 1765 | |
d352ac68 CM |
1766 | /* |
1767 | * given a list of ordered sums record them in the inode. This happens | |
1768 | * at IO completion time based on sums calculated at bio submission time. | |
1769 | */ | |
ba1da2f4 | 1770 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1771 | struct inode *inode, u64 file_offset, |
1772 | struct list_head *list) | |
1773 | { | |
e6dcd2dc CM |
1774 | struct btrfs_ordered_sum *sum; |
1775 | ||
c6e30871 | 1776 | list_for_each_entry(sum, list, list) { |
39847c4d | 1777 | trans->adding_csums = 1; |
d20f7043 CM |
1778 | btrfs_csum_file_blocks(trans, |
1779 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 1780 | trans->adding_csums = 0; |
e6dcd2dc CM |
1781 | } |
1782 | return 0; | |
1783 | } | |
1784 | ||
2ac55d41 JB |
1785 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1786 | struct extent_state **cached_state) | |
ea8c2819 | 1787 | { |
6c1500f2 | 1788 | WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0); |
ea8c2819 | 1789 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1790 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1791 | } |
1792 | ||
d352ac68 | 1793 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1794 | struct btrfs_writepage_fixup { |
1795 | struct page *page; | |
1796 | struct btrfs_work work; | |
1797 | }; | |
1798 | ||
b2950863 | 1799 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1800 | { |
1801 | struct btrfs_writepage_fixup *fixup; | |
1802 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1803 | struct extent_state *cached_state = NULL; |
247e743c CM |
1804 | struct page *page; |
1805 | struct inode *inode; | |
1806 | u64 page_start; | |
1807 | u64 page_end; | |
87826df0 | 1808 | int ret; |
247e743c CM |
1809 | |
1810 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1811 | page = fixup->page; | |
4a096752 | 1812 | again: |
247e743c CM |
1813 | lock_page(page); |
1814 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1815 | ClearPageChecked(page); | |
1816 | goto out_page; | |
1817 | } | |
1818 | ||
1819 | inode = page->mapping->host; | |
1820 | page_start = page_offset(page); | |
1821 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1822 | ||
2ac55d41 | 1823 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1824 | &cached_state); |
4a096752 CM |
1825 | |
1826 | /* already ordered? We're done */ | |
8b62b72b | 1827 | if (PagePrivate2(page)) |
247e743c | 1828 | goto out; |
4a096752 CM |
1829 | |
1830 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1831 | if (ordered) { | |
2ac55d41 JB |
1832 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1833 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1834 | unlock_page(page); |
1835 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1836 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1837 | goto again; |
1838 | } | |
247e743c | 1839 | |
87826df0 JM |
1840 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1841 | if (ret) { | |
1842 | mapping_set_error(page->mapping, ret); | |
1843 | end_extent_writepage(page, ret, page_start, page_end); | |
1844 | ClearPageChecked(page); | |
1845 | goto out; | |
1846 | } | |
1847 | ||
2ac55d41 | 1848 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1849 | ClearPageChecked(page); |
87826df0 | 1850 | set_page_dirty(page); |
247e743c | 1851 | out: |
2ac55d41 JB |
1852 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1853 | &cached_state, GFP_NOFS); | |
247e743c CM |
1854 | out_page: |
1855 | unlock_page(page); | |
1856 | page_cache_release(page); | |
b897abec | 1857 | kfree(fixup); |
247e743c CM |
1858 | } |
1859 | ||
1860 | /* | |
1861 | * There are a few paths in the higher layers of the kernel that directly | |
1862 | * set the page dirty bit without asking the filesystem if it is a | |
1863 | * good idea. This causes problems because we want to make sure COW | |
1864 | * properly happens and the data=ordered rules are followed. | |
1865 | * | |
c8b97818 | 1866 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1867 | * hasn't been properly setup for IO. We kick off an async process |
1868 | * to fix it up. The async helper will wait for ordered extents, set | |
1869 | * the delalloc bit and make it safe to write the page. | |
1870 | */ | |
b2950863 | 1871 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1872 | { |
1873 | struct inode *inode = page->mapping->host; | |
1874 | struct btrfs_writepage_fixup *fixup; | |
1875 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1876 | |
8b62b72b CM |
1877 | /* this page is properly in the ordered list */ |
1878 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1879 | return 0; |
1880 | ||
1881 | if (PageChecked(page)) | |
1882 | return -EAGAIN; | |
1883 | ||
1884 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1885 | if (!fixup) | |
1886 | return -EAGAIN; | |
f421950f | 1887 | |
247e743c CM |
1888 | SetPageChecked(page); |
1889 | page_cache_get(page); | |
1890 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1891 | fixup->page = page; | |
1892 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
87826df0 | 1893 | return -EBUSY; |
247e743c CM |
1894 | } |
1895 | ||
d899e052 YZ |
1896 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1897 | struct inode *inode, u64 file_pos, | |
1898 | u64 disk_bytenr, u64 disk_num_bytes, | |
1899 | u64 num_bytes, u64 ram_bytes, | |
1900 | u8 compression, u8 encryption, | |
1901 | u16 other_encoding, int extent_type) | |
1902 | { | |
1903 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1904 | struct btrfs_file_extent_item *fi; | |
1905 | struct btrfs_path *path; | |
1906 | struct extent_buffer *leaf; | |
1907 | struct btrfs_key ins; | |
d899e052 YZ |
1908 | int ret; |
1909 | ||
1910 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1911 | if (!path) |
1912 | return -ENOMEM; | |
d899e052 | 1913 | |
b9473439 | 1914 | path->leave_spinning = 1; |
a1ed835e CM |
1915 | |
1916 | /* | |
1917 | * we may be replacing one extent in the tree with another. | |
1918 | * The new extent is pinned in the extent map, and we don't want | |
1919 | * to drop it from the cache until it is completely in the btree. | |
1920 | * | |
1921 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1922 | * the caller is expected to unpin it and allow it to be merged | |
1923 | * with the others. | |
1924 | */ | |
5dc562c5 | 1925 | ret = btrfs_drop_extents(trans, root, inode, file_pos, |
2671485d | 1926 | file_pos + num_bytes, 0); |
79787eaa JM |
1927 | if (ret) |
1928 | goto out; | |
d899e052 | 1929 | |
33345d01 | 1930 | ins.objectid = btrfs_ino(inode); |
d899e052 YZ |
1931 | ins.offset = file_pos; |
1932 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1933 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
79787eaa JM |
1934 | if (ret) |
1935 | goto out; | |
d899e052 YZ |
1936 | leaf = path->nodes[0]; |
1937 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1938 | struct btrfs_file_extent_item); | |
1939 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1940 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1941 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1942 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1943 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1944 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1945 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1946 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1947 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1948 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 1949 | |
d899e052 | 1950 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 1951 | btrfs_release_path(path); |
d899e052 YZ |
1952 | |
1953 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1954 | |
1955 | ins.objectid = disk_bytenr; | |
1956 | ins.offset = disk_num_bytes; | |
1957 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1958 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1959 | root->root_key.objectid, | |
33345d01 | 1960 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 1961 | out: |
d899e052 | 1962 | btrfs_free_path(path); |
b9473439 | 1963 | |
79787eaa | 1964 | return ret; |
d899e052 YZ |
1965 | } |
1966 | ||
38c227d8 LB |
1967 | /* snapshot-aware defrag */ |
1968 | struct sa_defrag_extent_backref { | |
1969 | struct rb_node node; | |
1970 | struct old_sa_defrag_extent *old; | |
1971 | u64 root_id; | |
1972 | u64 inum; | |
1973 | u64 file_pos; | |
1974 | u64 extent_offset; | |
1975 | u64 num_bytes; | |
1976 | u64 generation; | |
1977 | }; | |
1978 | ||
1979 | struct old_sa_defrag_extent { | |
1980 | struct list_head list; | |
1981 | struct new_sa_defrag_extent *new; | |
1982 | ||
1983 | u64 extent_offset; | |
1984 | u64 bytenr; | |
1985 | u64 offset; | |
1986 | u64 len; | |
1987 | int count; | |
1988 | }; | |
1989 | ||
1990 | struct new_sa_defrag_extent { | |
1991 | struct rb_root root; | |
1992 | struct list_head head; | |
1993 | struct btrfs_path *path; | |
1994 | struct inode *inode; | |
1995 | u64 file_pos; | |
1996 | u64 len; | |
1997 | u64 bytenr; | |
1998 | u64 disk_len; | |
1999 | u8 compress_type; | |
2000 | }; | |
2001 | ||
2002 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2003 | struct sa_defrag_extent_backref *b2) | |
2004 | { | |
2005 | if (b1->root_id < b2->root_id) | |
2006 | return -1; | |
2007 | else if (b1->root_id > b2->root_id) | |
2008 | return 1; | |
2009 | ||
2010 | if (b1->inum < b2->inum) | |
2011 | return -1; | |
2012 | else if (b1->inum > b2->inum) | |
2013 | return 1; | |
2014 | ||
2015 | if (b1->file_pos < b2->file_pos) | |
2016 | return -1; | |
2017 | else if (b1->file_pos > b2->file_pos) | |
2018 | return 1; | |
2019 | ||
2020 | /* | |
2021 | * [------------------------------] ===> (a range of space) | |
2022 | * |<--->| |<---->| =============> (fs/file tree A) | |
2023 | * |<---------------------------->| ===> (fs/file tree B) | |
2024 | * | |
2025 | * A range of space can refer to two file extents in one tree while | |
2026 | * refer to only one file extent in another tree. | |
2027 | * | |
2028 | * So we may process a disk offset more than one time(two extents in A) | |
2029 | * and locate at the same extent(one extent in B), then insert two same | |
2030 | * backrefs(both refer to the extent in B). | |
2031 | */ | |
2032 | return 0; | |
2033 | } | |
2034 | ||
2035 | static void backref_insert(struct rb_root *root, | |
2036 | struct sa_defrag_extent_backref *backref) | |
2037 | { | |
2038 | struct rb_node **p = &root->rb_node; | |
2039 | struct rb_node *parent = NULL; | |
2040 | struct sa_defrag_extent_backref *entry; | |
2041 | int ret; | |
2042 | ||
2043 | while (*p) { | |
2044 | parent = *p; | |
2045 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2046 | ||
2047 | ret = backref_comp(backref, entry); | |
2048 | if (ret < 0) | |
2049 | p = &(*p)->rb_left; | |
2050 | else | |
2051 | p = &(*p)->rb_right; | |
2052 | } | |
2053 | ||
2054 | rb_link_node(&backref->node, parent, p); | |
2055 | rb_insert_color(&backref->node, root); | |
2056 | } | |
2057 | ||
2058 | /* | |
2059 | * Note the backref might has changed, and in this case we just return 0. | |
2060 | */ | |
2061 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2062 | void *ctx) | |
2063 | { | |
2064 | struct btrfs_file_extent_item *extent; | |
2065 | struct btrfs_fs_info *fs_info; | |
2066 | struct old_sa_defrag_extent *old = ctx; | |
2067 | struct new_sa_defrag_extent *new = old->new; | |
2068 | struct btrfs_path *path = new->path; | |
2069 | struct btrfs_key key; | |
2070 | struct btrfs_root *root; | |
2071 | struct sa_defrag_extent_backref *backref; | |
2072 | struct extent_buffer *leaf; | |
2073 | struct inode *inode = new->inode; | |
2074 | int slot; | |
2075 | int ret; | |
2076 | u64 extent_offset; | |
2077 | u64 num_bytes; | |
2078 | ||
2079 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
2080 | inum == btrfs_ino(inode)) | |
2081 | return 0; | |
2082 | ||
2083 | key.objectid = root_id; | |
2084 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2085 | key.offset = (u64)-1; | |
2086 | ||
2087 | fs_info = BTRFS_I(inode)->root->fs_info; | |
2088 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2089 | if (IS_ERR(root)) { | |
2090 | if (PTR_ERR(root) == -ENOENT) | |
2091 | return 0; | |
2092 | WARN_ON(1); | |
2093 | pr_debug("inum=%llu, offset=%llu, root_id=%llu\n", | |
2094 | inum, offset, root_id); | |
2095 | return PTR_ERR(root); | |
2096 | } | |
2097 | ||
2098 | key.objectid = inum; | |
2099 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2100 | if (offset > (u64)-1 << 32) | |
2101 | key.offset = 0; | |
2102 | else | |
2103 | key.offset = offset; | |
2104 | ||
2105 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2106 | if (ret < 0) { | |
2107 | WARN_ON(1); | |
2108 | return ret; | |
2109 | } | |
50f1319c | 2110 | ret = 0; |
38c227d8 LB |
2111 | |
2112 | while (1) { | |
2113 | cond_resched(); | |
2114 | ||
2115 | leaf = path->nodes[0]; | |
2116 | slot = path->slots[0]; | |
2117 | ||
2118 | if (slot >= btrfs_header_nritems(leaf)) { | |
2119 | ret = btrfs_next_leaf(root, path); | |
2120 | if (ret < 0) { | |
2121 | goto out; | |
2122 | } else if (ret > 0) { | |
2123 | ret = 0; | |
2124 | goto out; | |
2125 | } | |
2126 | continue; | |
2127 | } | |
2128 | ||
2129 | path->slots[0]++; | |
2130 | ||
2131 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2132 | ||
2133 | if (key.objectid > inum) | |
2134 | goto out; | |
2135 | ||
2136 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2137 | continue; | |
2138 | ||
2139 | extent = btrfs_item_ptr(leaf, slot, | |
2140 | struct btrfs_file_extent_item); | |
2141 | ||
2142 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2143 | continue; | |
2144 | ||
e68afa49 LB |
2145 | /* |
2146 | * 'offset' refers to the exact key.offset, | |
2147 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2148 | * (key.offset - extent_offset). | |
2149 | */ | |
2150 | if (key.offset != offset) | |
38c227d8 LB |
2151 | continue; |
2152 | ||
e68afa49 | 2153 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2154 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2155 | |
38c227d8 LB |
2156 | if (extent_offset >= old->extent_offset + old->offset + |
2157 | old->len || extent_offset + num_bytes <= | |
2158 | old->extent_offset + old->offset) | |
2159 | continue; | |
38c227d8 LB |
2160 | break; |
2161 | } | |
2162 | ||
2163 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2164 | if (!backref) { | |
2165 | ret = -ENOENT; | |
2166 | goto out; | |
2167 | } | |
2168 | ||
2169 | backref->root_id = root_id; | |
2170 | backref->inum = inum; | |
e68afa49 | 2171 | backref->file_pos = offset; |
38c227d8 LB |
2172 | backref->num_bytes = num_bytes; |
2173 | backref->extent_offset = extent_offset; | |
2174 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2175 | backref->old = old; | |
2176 | backref_insert(&new->root, backref); | |
2177 | old->count++; | |
2178 | out: | |
2179 | btrfs_release_path(path); | |
2180 | WARN_ON(ret); | |
2181 | return ret; | |
2182 | } | |
2183 | ||
2184 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2185 | struct new_sa_defrag_extent *new) | |
2186 | { | |
2187 | struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info; | |
2188 | struct old_sa_defrag_extent *old, *tmp; | |
2189 | int ret; | |
2190 | ||
2191 | new->path = path; | |
2192 | ||
2193 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2194 | ret = iterate_inodes_from_logical(old->bytenr + |
2195 | old->extent_offset, fs_info, | |
38c227d8 LB |
2196 | path, record_one_backref, |
2197 | old); | |
2198 | BUG_ON(ret < 0 && ret != -ENOENT); | |
2199 | ||
2200 | /* no backref to be processed for this extent */ | |
2201 | if (!old->count) { | |
2202 | list_del(&old->list); | |
2203 | kfree(old); | |
2204 | } | |
2205 | } | |
2206 | ||
2207 | if (list_empty(&new->head)) | |
2208 | return false; | |
2209 | ||
2210 | return true; | |
2211 | } | |
2212 | ||
2213 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2214 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2215 | struct new_sa_defrag_extent *new) |
38c227d8 | 2216 | { |
116e0024 | 2217 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2218 | return 0; |
2219 | ||
2220 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2221 | return 0; | |
2222 | ||
116e0024 LB |
2223 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2224 | return 0; | |
2225 | ||
2226 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2227 | btrfs_file_extent_other_encoding(leaf, fi)) |
2228 | return 0; | |
2229 | ||
2230 | return 1; | |
2231 | } | |
2232 | ||
2233 | /* | |
2234 | * Note the backref might has changed, and in this case we just return 0. | |
2235 | */ | |
2236 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2237 | struct sa_defrag_extent_backref *prev, | |
2238 | struct sa_defrag_extent_backref *backref) | |
2239 | { | |
2240 | struct btrfs_file_extent_item *extent; | |
2241 | struct btrfs_file_extent_item *item; | |
2242 | struct btrfs_ordered_extent *ordered; | |
2243 | struct btrfs_trans_handle *trans; | |
2244 | struct btrfs_fs_info *fs_info; | |
2245 | struct btrfs_root *root; | |
2246 | struct btrfs_key key; | |
2247 | struct extent_buffer *leaf; | |
2248 | struct old_sa_defrag_extent *old = backref->old; | |
2249 | struct new_sa_defrag_extent *new = old->new; | |
2250 | struct inode *src_inode = new->inode; | |
2251 | struct inode *inode; | |
2252 | struct extent_state *cached = NULL; | |
2253 | int ret = 0; | |
2254 | u64 start; | |
2255 | u64 len; | |
2256 | u64 lock_start; | |
2257 | u64 lock_end; | |
2258 | bool merge = false; | |
2259 | int index; | |
2260 | ||
2261 | if (prev && prev->root_id == backref->root_id && | |
2262 | prev->inum == backref->inum && | |
2263 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2264 | merge = true; | |
2265 | ||
2266 | /* step 1: get root */ | |
2267 | key.objectid = backref->root_id; | |
2268 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2269 | key.offset = (u64)-1; | |
2270 | ||
2271 | fs_info = BTRFS_I(src_inode)->root->fs_info; | |
2272 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
2273 | ||
2274 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2275 | if (IS_ERR(root)) { | |
2276 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2277 | if (PTR_ERR(root) == -ENOENT) | |
2278 | return 0; | |
2279 | return PTR_ERR(root); | |
2280 | } | |
38c227d8 LB |
2281 | |
2282 | /* step 2: get inode */ | |
2283 | key.objectid = backref->inum; | |
2284 | key.type = BTRFS_INODE_ITEM_KEY; | |
2285 | key.offset = 0; | |
2286 | ||
2287 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2288 | if (IS_ERR(inode)) { | |
2289 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2290 | return 0; | |
2291 | } | |
2292 | ||
2293 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2294 | ||
2295 | /* step 3: relink backref */ | |
2296 | lock_start = backref->file_pos; | |
2297 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2298 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2299 | 0, &cached); | |
2300 | ||
2301 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2302 | if (ordered) { | |
2303 | btrfs_put_ordered_extent(ordered); | |
2304 | goto out_unlock; | |
2305 | } | |
2306 | ||
2307 | trans = btrfs_join_transaction(root); | |
2308 | if (IS_ERR(trans)) { | |
2309 | ret = PTR_ERR(trans); | |
2310 | goto out_unlock; | |
2311 | } | |
2312 | ||
2313 | key.objectid = backref->inum; | |
2314 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2315 | key.offset = backref->file_pos; | |
2316 | ||
2317 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2318 | if (ret < 0) { | |
2319 | goto out_free_path; | |
2320 | } else if (ret > 0) { | |
2321 | ret = 0; | |
2322 | goto out_free_path; | |
2323 | } | |
2324 | ||
2325 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2326 | struct btrfs_file_extent_item); | |
2327 | ||
2328 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2329 | backref->generation) | |
2330 | goto out_free_path; | |
2331 | ||
2332 | btrfs_release_path(path); | |
2333 | ||
2334 | start = backref->file_pos; | |
2335 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2336 | start += old->extent_offset + old->offset - | |
2337 | backref->extent_offset; | |
2338 | ||
2339 | len = min(backref->extent_offset + backref->num_bytes, | |
2340 | old->extent_offset + old->offset + old->len); | |
2341 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2342 | ||
2343 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2344 | start + len, 1); | |
2345 | if (ret) | |
2346 | goto out_free_path; | |
2347 | again: | |
2348 | key.objectid = btrfs_ino(inode); | |
2349 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2350 | key.offset = start; | |
2351 | ||
a09a0a70 | 2352 | path->leave_spinning = 1; |
38c227d8 LB |
2353 | if (merge) { |
2354 | struct btrfs_file_extent_item *fi; | |
2355 | u64 extent_len; | |
2356 | struct btrfs_key found_key; | |
2357 | ||
2358 | ret = btrfs_search_slot(trans, root, &key, path, 1, 1); | |
2359 | if (ret < 0) | |
2360 | goto out_free_path; | |
2361 | ||
2362 | path->slots[0]--; | |
2363 | leaf = path->nodes[0]; | |
2364 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2365 | ||
2366 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2367 | struct btrfs_file_extent_item); | |
2368 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2369 | ||
116e0024 LB |
2370 | if (extent_len + found_key.offset == start && |
2371 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2372 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2373 | extent_len + len); | |
2374 | btrfs_mark_buffer_dirty(leaf); | |
2375 | inode_add_bytes(inode, len); | |
2376 | ||
2377 | ret = 1; | |
2378 | goto out_free_path; | |
2379 | } else { | |
2380 | merge = false; | |
2381 | btrfs_release_path(path); | |
2382 | goto again; | |
2383 | } | |
2384 | } | |
2385 | ||
2386 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2387 | sizeof(*extent)); | |
2388 | if (ret) { | |
2389 | btrfs_abort_transaction(trans, root, ret); | |
2390 | goto out_free_path; | |
2391 | } | |
2392 | ||
2393 | leaf = path->nodes[0]; | |
2394 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2395 | struct btrfs_file_extent_item); | |
2396 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2397 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2398 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2399 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2400 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2401 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2402 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2403 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2404 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2405 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2406 | ||
2407 | btrfs_mark_buffer_dirty(leaf); | |
2408 | inode_add_bytes(inode, len); | |
a09a0a70 | 2409 | btrfs_release_path(path); |
38c227d8 LB |
2410 | |
2411 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, | |
2412 | new->disk_len, 0, | |
2413 | backref->root_id, backref->inum, | |
2414 | new->file_pos, 0); /* start - extent_offset */ | |
2415 | if (ret) { | |
2416 | btrfs_abort_transaction(trans, root, ret); | |
2417 | goto out_free_path; | |
2418 | } | |
2419 | ||
2420 | ret = 1; | |
2421 | out_free_path: | |
2422 | btrfs_release_path(path); | |
a09a0a70 | 2423 | path->leave_spinning = 0; |
38c227d8 LB |
2424 | btrfs_end_transaction(trans, root); |
2425 | out_unlock: | |
2426 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2427 | &cached, GFP_NOFS); | |
2428 | iput(inode); | |
2429 | return ret; | |
2430 | } | |
2431 | ||
2432 | static void relink_file_extents(struct new_sa_defrag_extent *new) | |
2433 | { | |
2434 | struct btrfs_path *path; | |
2435 | struct old_sa_defrag_extent *old, *tmp; | |
2436 | struct sa_defrag_extent_backref *backref; | |
2437 | struct sa_defrag_extent_backref *prev = NULL; | |
2438 | struct inode *inode; | |
2439 | struct btrfs_root *root; | |
2440 | struct rb_node *node; | |
2441 | int ret; | |
2442 | ||
2443 | inode = new->inode; | |
2444 | root = BTRFS_I(inode)->root; | |
2445 | ||
2446 | path = btrfs_alloc_path(); | |
2447 | if (!path) | |
2448 | return; | |
2449 | ||
2450 | if (!record_extent_backrefs(path, new)) { | |
2451 | btrfs_free_path(path); | |
2452 | goto out; | |
2453 | } | |
2454 | btrfs_release_path(path); | |
2455 | ||
2456 | while (1) { | |
2457 | node = rb_first(&new->root); | |
2458 | if (!node) | |
2459 | break; | |
2460 | rb_erase(node, &new->root); | |
2461 | ||
2462 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2463 | ||
2464 | ret = relink_extent_backref(path, prev, backref); | |
2465 | WARN_ON(ret < 0); | |
2466 | ||
2467 | kfree(prev); | |
2468 | ||
2469 | if (ret == 1) | |
2470 | prev = backref; | |
2471 | else | |
2472 | prev = NULL; | |
2473 | cond_resched(); | |
2474 | } | |
2475 | kfree(prev); | |
2476 | ||
2477 | btrfs_free_path(path); | |
2478 | ||
2479 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
2480 | list_del(&old->list); | |
2481 | kfree(old); | |
2482 | } | |
2483 | out: | |
2484 | atomic_dec(&root->fs_info->defrag_running); | |
2485 | wake_up(&root->fs_info->transaction_wait); | |
2486 | ||
2487 | kfree(new); | |
2488 | } | |
2489 | ||
2490 | static struct new_sa_defrag_extent * | |
2491 | record_old_file_extents(struct inode *inode, | |
2492 | struct btrfs_ordered_extent *ordered) | |
2493 | { | |
2494 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2495 | struct btrfs_path *path; | |
2496 | struct btrfs_key key; | |
2497 | struct old_sa_defrag_extent *old, *tmp; | |
2498 | struct new_sa_defrag_extent *new; | |
2499 | int ret; | |
2500 | ||
2501 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2502 | if (!new) | |
2503 | return NULL; | |
2504 | ||
2505 | new->inode = inode; | |
2506 | new->file_pos = ordered->file_offset; | |
2507 | new->len = ordered->len; | |
2508 | new->bytenr = ordered->start; | |
2509 | new->disk_len = ordered->disk_len; | |
2510 | new->compress_type = ordered->compress_type; | |
2511 | new->root = RB_ROOT; | |
2512 | INIT_LIST_HEAD(&new->head); | |
2513 | ||
2514 | path = btrfs_alloc_path(); | |
2515 | if (!path) | |
2516 | goto out_kfree; | |
2517 | ||
2518 | key.objectid = btrfs_ino(inode); | |
2519 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2520 | key.offset = new->file_pos; | |
2521 | ||
2522 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2523 | if (ret < 0) | |
2524 | goto out_free_path; | |
2525 | if (ret > 0 && path->slots[0] > 0) | |
2526 | path->slots[0]--; | |
2527 | ||
2528 | /* find out all the old extents for the file range */ | |
2529 | while (1) { | |
2530 | struct btrfs_file_extent_item *extent; | |
2531 | struct extent_buffer *l; | |
2532 | int slot; | |
2533 | u64 num_bytes; | |
2534 | u64 offset; | |
2535 | u64 end; | |
2536 | u64 disk_bytenr; | |
2537 | u64 extent_offset; | |
2538 | ||
2539 | l = path->nodes[0]; | |
2540 | slot = path->slots[0]; | |
2541 | ||
2542 | if (slot >= btrfs_header_nritems(l)) { | |
2543 | ret = btrfs_next_leaf(root, path); | |
2544 | if (ret < 0) | |
2545 | goto out_free_list; | |
2546 | else if (ret > 0) | |
2547 | break; | |
2548 | continue; | |
2549 | } | |
2550 | ||
2551 | btrfs_item_key_to_cpu(l, &key, slot); | |
2552 | ||
2553 | if (key.objectid != btrfs_ino(inode)) | |
2554 | break; | |
2555 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2556 | break; | |
2557 | if (key.offset >= new->file_pos + new->len) | |
2558 | break; | |
2559 | ||
2560 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2561 | ||
2562 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2563 | if (key.offset + num_bytes < new->file_pos) | |
2564 | goto next; | |
2565 | ||
2566 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2567 | if (!disk_bytenr) | |
2568 | goto next; | |
2569 | ||
2570 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2571 | ||
2572 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2573 | if (!old) | |
2574 | goto out_free_list; | |
2575 | ||
2576 | offset = max(new->file_pos, key.offset); | |
2577 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2578 | ||
2579 | old->bytenr = disk_bytenr; | |
2580 | old->extent_offset = extent_offset; | |
2581 | old->offset = offset - key.offset; | |
2582 | old->len = end - offset; | |
2583 | old->new = new; | |
2584 | old->count = 0; | |
2585 | list_add_tail(&old->list, &new->head); | |
2586 | next: | |
2587 | path->slots[0]++; | |
2588 | cond_resched(); | |
2589 | } | |
2590 | ||
2591 | btrfs_free_path(path); | |
2592 | atomic_inc(&root->fs_info->defrag_running); | |
2593 | ||
2594 | return new; | |
2595 | ||
2596 | out_free_list: | |
2597 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
2598 | list_del(&old->list); | |
2599 | kfree(old); | |
2600 | } | |
2601 | out_free_path: | |
2602 | btrfs_free_path(path); | |
2603 | out_kfree: | |
2604 | kfree(new); | |
2605 | return NULL; | |
2606 | } | |
2607 | ||
5d13a98f CM |
2608 | /* |
2609 | * helper function for btrfs_finish_ordered_io, this | |
2610 | * just reads in some of the csum leaves to prime them into ram | |
2611 | * before we start the transaction. It limits the amount of btree | |
2612 | * reads required while inside the transaction. | |
2613 | */ | |
d352ac68 CM |
2614 | /* as ordered data IO finishes, this gets called so we can finish |
2615 | * an ordered extent if the range of bytes in the file it covers are | |
2616 | * fully written. | |
2617 | */ | |
5fd02043 | 2618 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2619 | { |
5fd02043 | 2620 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 2621 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2622 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2623 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2624 | struct extent_state *cached_state = NULL; |
38c227d8 | 2625 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2626 | int compress_type = 0; |
e6dcd2dc | 2627 | int ret; |
82d5902d | 2628 | bool nolock; |
e6dcd2dc | 2629 | |
83eea1f1 | 2630 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 2631 | |
5fd02043 JB |
2632 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2633 | ret = -EIO; | |
2634 | goto out; | |
2635 | } | |
2636 | ||
c2167754 | 2637 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2638 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
6c760c07 JB |
2639 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2640 | if (nolock) | |
2641 | trans = btrfs_join_transaction_nolock(root); | |
2642 | else | |
2643 | trans = btrfs_join_transaction(root); | |
2644 | if (IS_ERR(trans)) { | |
2645 | ret = PTR_ERR(trans); | |
2646 | trans = NULL; | |
2647 | goto out; | |
c2167754 | 2648 | } |
6c760c07 JB |
2649 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2650 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2651 | if (ret) /* -ENOMEM or corruption */ | |
2652 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
2653 | goto out; |
2654 | } | |
e6dcd2dc | 2655 | |
2ac55d41 JB |
2656 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2657 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 2658 | 0, &cached_state); |
e6dcd2dc | 2659 | |
38c227d8 LB |
2660 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2661 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2662 | EXTENT_DEFRAG, 1, cached_state); | |
2663 | if (ret) { | |
2664 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
2665 | if (last_snapshot >= BTRFS_I(inode)->generation) | |
2666 | /* the inode is shared */ | |
2667 | new = record_old_file_extents(inode, ordered_extent); | |
2668 | ||
2669 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2670 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2671 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
2672 | } | |
2673 | ||
0cb59c99 | 2674 | if (nolock) |
7a7eaa40 | 2675 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 2676 | else |
7a7eaa40 | 2677 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2678 | if (IS_ERR(trans)) { |
2679 | ret = PTR_ERR(trans); | |
2680 | trans = NULL; | |
2681 | goto out_unlock; | |
2682 | } | |
0ca1f7ce | 2683 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 2684 | |
c8b97818 | 2685 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2686 | compress_type = ordered_extent->compress_type; |
d899e052 | 2687 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2688 | BUG_ON(compress_type); |
920bbbfb | 2689 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
2690 | ordered_extent->file_offset, |
2691 | ordered_extent->file_offset + | |
2692 | ordered_extent->len); | |
d899e052 | 2693 | } else { |
0af3d00b | 2694 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
2695 | ret = insert_reserved_file_extent(trans, inode, |
2696 | ordered_extent->file_offset, | |
2697 | ordered_extent->start, | |
2698 | ordered_extent->disk_len, | |
2699 | ordered_extent->len, | |
2700 | ordered_extent->len, | |
261507a0 | 2701 | compress_type, 0, 0, |
d899e052 | 2702 | BTRFS_FILE_EXTENT_REG); |
d899e052 | 2703 | } |
5dc562c5 JB |
2704 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
2705 | ordered_extent->file_offset, ordered_extent->len, | |
2706 | trans->transid); | |
79787eaa JM |
2707 | if (ret < 0) { |
2708 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 2709 | goto out_unlock; |
79787eaa | 2710 | } |
2ac55d41 | 2711 | |
e6dcd2dc CM |
2712 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
2713 | &ordered_extent->list); | |
2714 | ||
6c760c07 JB |
2715 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2716 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2717 | if (ret) { /* -ENOMEM or corruption */ | |
2718 | btrfs_abort_transaction(trans, root, ret); | |
2719 | goto out_unlock; | |
1ef30be1 JB |
2720 | } |
2721 | ret = 0; | |
5fd02043 JB |
2722 | out_unlock: |
2723 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2724 | ordered_extent->file_offset + | |
2725 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2726 | out: |
5b0e95bf | 2727 | if (root != root->fs_info->tree_root) |
0cb59c99 | 2728 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
2729 | if (trans) |
2730 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 2731 | |
0bec9ef5 | 2732 | if (ret) { |
5fd02043 JB |
2733 | clear_extent_uptodate(io_tree, ordered_extent->file_offset, |
2734 | ordered_extent->file_offset + | |
2735 | ordered_extent->len - 1, NULL, GFP_NOFS); | |
2736 | ||
0bec9ef5 JB |
2737 | /* |
2738 | * If the ordered extent had an IOERR or something else went | |
2739 | * wrong we need to return the space for this ordered extent | |
2740 | * back to the allocator. | |
2741 | */ | |
2742 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
2743 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) | |
2744 | btrfs_free_reserved_extent(root, ordered_extent->start, | |
2745 | ordered_extent->disk_len); | |
2746 | } | |
2747 | ||
2748 | ||
5fd02043 | 2749 | /* |
8bad3c02 LB |
2750 | * This needs to be done to make sure anybody waiting knows we are done |
2751 | * updating everything for this ordered extent. | |
5fd02043 JB |
2752 | */ |
2753 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
2754 | ||
38c227d8 LB |
2755 | /* for snapshot-aware defrag */ |
2756 | if (new) | |
2757 | relink_file_extents(new); | |
2758 | ||
e6dcd2dc CM |
2759 | /* once for us */ |
2760 | btrfs_put_ordered_extent(ordered_extent); | |
2761 | /* once for the tree */ | |
2762 | btrfs_put_ordered_extent(ordered_extent); | |
2763 | ||
5fd02043 JB |
2764 | return ret; |
2765 | } | |
2766 | ||
2767 | static void finish_ordered_fn(struct btrfs_work *work) | |
2768 | { | |
2769 | struct btrfs_ordered_extent *ordered_extent; | |
2770 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
2771 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
2772 | } |
2773 | ||
b2950863 | 2774 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
2775 | struct extent_state *state, int uptodate) |
2776 | { | |
5fd02043 JB |
2777 | struct inode *inode = page->mapping->host; |
2778 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2779 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
2780 | struct btrfs_workers *workers; | |
2781 | ||
1abe9b8a | 2782 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2783 | ||
8b62b72b | 2784 | ClearPagePrivate2(page); |
5fd02043 JB |
2785 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2786 | end - start + 1, uptodate)) | |
2787 | return 0; | |
2788 | ||
2789 | ordered_extent->work.func = finish_ordered_fn; | |
2790 | ordered_extent->work.flags = 0; | |
2791 | ||
83eea1f1 | 2792 | if (btrfs_is_free_space_inode(inode)) |
5fd02043 JB |
2793 | workers = &root->fs_info->endio_freespace_worker; |
2794 | else | |
2795 | workers = &root->fs_info->endio_write_workers; | |
2796 | btrfs_queue_worker(workers, &ordered_extent->work); | |
2797 | ||
2798 | return 0; | |
211f90e6 CM |
2799 | } |
2800 | ||
d352ac68 CM |
2801 | /* |
2802 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
2803 | * if there's a match, we allow the bio to finish. If not, the code in |
2804 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 2805 | */ |
facc8a22 MX |
2806 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
2807 | u64 phy_offset, struct page *page, | |
2808 | u64 start, u64 end, int mirror) | |
07157aac | 2809 | { |
4eee4fa4 | 2810 | size_t offset = start - page_offset(page); |
07157aac | 2811 | struct inode *inode = page->mapping->host; |
d1310b2e | 2812 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 2813 | char *kaddr; |
ff79f819 | 2814 | struct btrfs_root *root = BTRFS_I(inode)->root; |
facc8a22 | 2815 | u32 csum_expected; |
ff79f819 | 2816 | u32 csum = ~(u32)0; |
c2cf52eb SK |
2817 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, |
2818 | DEFAULT_RATELIMIT_BURST); | |
d1310b2e | 2819 | |
d20f7043 CM |
2820 | if (PageChecked(page)) { |
2821 | ClearPageChecked(page); | |
2822 | goto good; | |
2823 | } | |
6cbff00f CH |
2824 | |
2825 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 2826 | goto good; |
17d217fe YZ |
2827 | |
2828 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 2829 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
2830 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
2831 | GFP_NOFS); | |
b6cda9bc | 2832 | return 0; |
17d217fe | 2833 | } |
d20f7043 | 2834 | |
facc8a22 MX |
2835 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
2836 | csum_expected = *(((u32 *)io_bio->csum) + phy_offset); | |
d397712b | 2837 | |
facc8a22 | 2838 | kaddr = kmap_atomic(page); |
b0496686 | 2839 | csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1); |
ff79f819 | 2840 | btrfs_csum_final(csum, (char *)&csum); |
facc8a22 | 2841 | if (csum != csum_expected) |
07157aac | 2842 | goto zeroit; |
d397712b | 2843 | |
7ac687d9 | 2844 | kunmap_atomic(kaddr); |
d20f7043 | 2845 | good: |
07157aac CM |
2846 | return 0; |
2847 | ||
2848 | zeroit: | |
c2cf52eb | 2849 | if (__ratelimit(&_rs)) |
facc8a22 | 2850 | btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u", |
c2cf52eb | 2851 | (unsigned long long)btrfs_ino(page->mapping->host), |
facc8a22 | 2852 | (unsigned long long)start, csum, csum_expected); |
db94535d CM |
2853 | memset(kaddr + offset, 1, end - start + 1); |
2854 | flush_dcache_page(page); | |
7ac687d9 | 2855 | kunmap_atomic(kaddr); |
facc8a22 | 2856 | if (csum_expected == 0) |
3b951516 | 2857 | return 0; |
7e38326f | 2858 | return -EIO; |
07157aac | 2859 | } |
b888db2b | 2860 | |
24bbcf04 YZ |
2861 | struct delayed_iput { |
2862 | struct list_head list; | |
2863 | struct inode *inode; | |
2864 | }; | |
2865 | ||
79787eaa JM |
2866 | /* JDM: If this is fs-wide, why can't we add a pointer to |
2867 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
2868 | void btrfs_add_delayed_iput(struct inode *inode) |
2869 | { | |
2870 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2871 | struct delayed_iput *delayed; | |
2872 | ||
2873 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2874 | return; | |
2875 | ||
2876 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2877 | delayed->inode = inode; | |
2878 | ||
2879 | spin_lock(&fs_info->delayed_iput_lock); | |
2880 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2881 | spin_unlock(&fs_info->delayed_iput_lock); | |
2882 | } | |
2883 | ||
2884 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2885 | { | |
2886 | LIST_HEAD(list); | |
2887 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2888 | struct delayed_iput *delayed; | |
2889 | int empty; | |
2890 | ||
2891 | spin_lock(&fs_info->delayed_iput_lock); | |
2892 | empty = list_empty(&fs_info->delayed_iputs); | |
2893 | spin_unlock(&fs_info->delayed_iput_lock); | |
2894 | if (empty) | |
2895 | return; | |
2896 | ||
24bbcf04 YZ |
2897 | spin_lock(&fs_info->delayed_iput_lock); |
2898 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2899 | spin_unlock(&fs_info->delayed_iput_lock); | |
2900 | ||
2901 | while (!list_empty(&list)) { | |
2902 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2903 | list_del(&delayed->list); | |
2904 | iput(delayed->inode); | |
2905 | kfree(delayed); | |
2906 | } | |
24bbcf04 YZ |
2907 | } |
2908 | ||
d68fc57b | 2909 | /* |
42b2aa86 | 2910 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
2911 | * files in the subvolume, it removes orphan item and frees block_rsv |
2912 | * structure. | |
2913 | */ | |
2914 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2915 | struct btrfs_root *root) | |
2916 | { | |
90290e19 | 2917 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
2918 | int ret; |
2919 | ||
8a35d95f | 2920 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
2921 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
2922 | return; | |
2923 | ||
90290e19 | 2924 | spin_lock(&root->orphan_lock); |
8a35d95f | 2925 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
2926 | spin_unlock(&root->orphan_lock); |
2927 | return; | |
2928 | } | |
2929 | ||
2930 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
2931 | spin_unlock(&root->orphan_lock); | |
2932 | return; | |
2933 | } | |
2934 | ||
2935 | block_rsv = root->orphan_block_rsv; | |
2936 | root->orphan_block_rsv = NULL; | |
2937 | spin_unlock(&root->orphan_lock); | |
2938 | ||
d68fc57b YZ |
2939 | if (root->orphan_item_inserted && |
2940 | btrfs_root_refs(&root->root_item) > 0) { | |
2941 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2942 | root->root_key.objectid); | |
4ef31a45 JB |
2943 | if (ret) |
2944 | btrfs_abort_transaction(trans, root, ret); | |
2945 | else | |
2946 | root->orphan_item_inserted = 0; | |
d68fc57b YZ |
2947 | } |
2948 | ||
90290e19 JB |
2949 | if (block_rsv) { |
2950 | WARN_ON(block_rsv->size > 0); | |
2951 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
2952 | } |
2953 | } | |
2954 | ||
7b128766 JB |
2955 | /* |
2956 | * This creates an orphan entry for the given inode in case something goes | |
2957 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2958 | * |
2959 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2960 | * this function. | |
7b128766 JB |
2961 | */ |
2962 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2963 | { | |
2964 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2965 | struct btrfs_block_rsv *block_rsv = NULL; |
2966 | int reserve = 0; | |
2967 | int insert = 0; | |
2968 | int ret; | |
7b128766 | 2969 | |
d68fc57b | 2970 | if (!root->orphan_block_rsv) { |
66d8f3dd | 2971 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
2972 | if (!block_rsv) |
2973 | return -ENOMEM; | |
d68fc57b | 2974 | } |
7b128766 | 2975 | |
d68fc57b YZ |
2976 | spin_lock(&root->orphan_lock); |
2977 | if (!root->orphan_block_rsv) { | |
2978 | root->orphan_block_rsv = block_rsv; | |
2979 | } else if (block_rsv) { | |
2980 | btrfs_free_block_rsv(root, block_rsv); | |
2981 | block_rsv = NULL; | |
7b128766 | 2982 | } |
7b128766 | 2983 | |
8a35d95f JB |
2984 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2985 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
2986 | #if 0 |
2987 | /* | |
2988 | * For proper ENOSPC handling, we should do orphan | |
2989 | * cleanup when mounting. But this introduces backward | |
2990 | * compatibility issue. | |
2991 | */ | |
2992 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2993 | insert = 2; | |
2994 | else | |
2995 | insert = 1; | |
2996 | #endif | |
2997 | insert = 1; | |
321f0e70 | 2998 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
2999 | } |
3000 | ||
72ac3c0d JB |
3001 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3002 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3003 | reserve = 1; |
d68fc57b | 3004 | spin_unlock(&root->orphan_lock); |
7b128766 | 3005 | |
d68fc57b YZ |
3006 | /* grab metadata reservation from transaction handle */ |
3007 | if (reserve) { | |
3008 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 3009 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 3010 | } |
7b128766 | 3011 | |
d68fc57b YZ |
3012 | /* insert an orphan item to track this unlinked/truncated file */ |
3013 | if (insert >= 1) { | |
33345d01 | 3014 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 3015 | if (ret) { |
8a35d95f JB |
3016 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3017 | &BTRFS_I(inode)->runtime_flags); | |
4ef31a45 JB |
3018 | if (reserve) { |
3019 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
3020 | &BTRFS_I(inode)->runtime_flags); | |
3021 | btrfs_orphan_release_metadata(inode); | |
3022 | } | |
3023 | if (ret != -EEXIST) { | |
3024 | btrfs_abort_transaction(trans, root, ret); | |
3025 | return ret; | |
3026 | } | |
79787eaa JM |
3027 | } |
3028 | ret = 0; | |
d68fc57b YZ |
3029 | } |
3030 | ||
3031 | /* insert an orphan item to track subvolume contains orphan files */ | |
3032 | if (insert >= 2) { | |
3033 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
3034 | root->root_key.objectid); | |
79787eaa JM |
3035 | if (ret && ret != -EEXIST) { |
3036 | btrfs_abort_transaction(trans, root, ret); | |
3037 | return ret; | |
3038 | } | |
d68fc57b YZ |
3039 | } |
3040 | return 0; | |
7b128766 JB |
3041 | } |
3042 | ||
3043 | /* | |
3044 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3045 | * item for this particular inode. | |
3046 | */ | |
48a3b636 ES |
3047 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3048 | struct inode *inode) | |
7b128766 JB |
3049 | { |
3050 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3051 | int delete_item = 0; |
3052 | int release_rsv = 0; | |
7b128766 JB |
3053 | int ret = 0; |
3054 | ||
d68fc57b | 3055 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
3056 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3057 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3058 | delete_item = 1; |
7b128766 | 3059 | |
72ac3c0d JB |
3060 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3061 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3062 | release_rsv = 1; |
d68fc57b | 3063 | spin_unlock(&root->orphan_lock); |
7b128766 | 3064 | |
4ef31a45 | 3065 | if (trans && delete_item) |
33345d01 | 3066 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); |
7b128766 | 3067 | |
8a35d95f | 3068 | if (release_rsv) { |
d68fc57b | 3069 | btrfs_orphan_release_metadata(inode); |
8a35d95f JB |
3070 | atomic_dec(&root->orphan_inodes); |
3071 | } | |
7b128766 | 3072 | |
4ef31a45 | 3073 | return ret; |
7b128766 JB |
3074 | } |
3075 | ||
3076 | /* | |
3077 | * this cleans up any orphans that may be left on the list from the last use | |
3078 | * of this root. | |
3079 | */ | |
66b4ffd1 | 3080 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
3081 | { |
3082 | struct btrfs_path *path; | |
3083 | struct extent_buffer *leaf; | |
7b128766 JB |
3084 | struct btrfs_key key, found_key; |
3085 | struct btrfs_trans_handle *trans; | |
3086 | struct inode *inode; | |
8f6d7f4f | 3087 | u64 last_objectid = 0; |
7b128766 JB |
3088 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3089 | ||
d68fc57b | 3090 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3091 | return 0; |
c71bf099 YZ |
3092 | |
3093 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3094 | if (!path) { |
3095 | ret = -ENOMEM; | |
3096 | goto out; | |
3097 | } | |
7b128766 JB |
3098 | path->reada = -1; |
3099 | ||
3100 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
3101 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
3102 | key.offset = (u64)-1; | |
3103 | ||
7b128766 JB |
3104 | while (1) { |
3105 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3106 | if (ret < 0) |
3107 | goto out; | |
7b128766 JB |
3108 | |
3109 | /* | |
3110 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3111 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3112 | * find the key and see if we have stuff that matches |
3113 | */ | |
3114 | if (ret > 0) { | |
66b4ffd1 | 3115 | ret = 0; |
7b128766 JB |
3116 | if (path->slots[0] == 0) |
3117 | break; | |
3118 | path->slots[0]--; | |
3119 | } | |
3120 | ||
3121 | /* pull out the item */ | |
3122 | leaf = path->nodes[0]; | |
7b128766 JB |
3123 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3124 | ||
3125 | /* make sure the item matches what we want */ | |
3126 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3127 | break; | |
3128 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
3129 | break; | |
3130 | ||
3131 | /* release the path since we're done with it */ | |
b3b4aa74 | 3132 | btrfs_release_path(path); |
7b128766 JB |
3133 | |
3134 | /* | |
3135 | * this is where we are basically btrfs_lookup, without the | |
3136 | * crossing root thing. we store the inode number in the | |
3137 | * offset of the orphan item. | |
3138 | */ | |
8f6d7f4f JB |
3139 | |
3140 | if (found_key.offset == last_objectid) { | |
c2cf52eb SK |
3141 | btrfs_err(root->fs_info, |
3142 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3143 | ret = -EINVAL; |
3144 | goto out; | |
3145 | } | |
3146 | ||
3147 | last_objectid = found_key.offset; | |
3148 | ||
5d4f98a2 YZ |
3149 | found_key.objectid = found_key.offset; |
3150 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3151 | found_key.offset = 0; | |
73f73415 | 3152 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
a8c9e576 JB |
3153 | ret = PTR_RET(inode); |
3154 | if (ret && ret != -ESTALE) | |
66b4ffd1 | 3155 | goto out; |
7b128766 | 3156 | |
f8e9e0b0 AJ |
3157 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
3158 | struct btrfs_root *dead_root; | |
3159 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3160 | int is_dead_root = 0; | |
3161 | ||
3162 | /* | |
3163 | * this is an orphan in the tree root. Currently these | |
3164 | * could come from 2 sources: | |
3165 | * a) a snapshot deletion in progress | |
3166 | * b) a free space cache inode | |
3167 | * We need to distinguish those two, as the snapshot | |
3168 | * orphan must not get deleted. | |
3169 | * find_dead_roots already ran before us, so if this | |
3170 | * is a snapshot deletion, we should find the root | |
3171 | * in the dead_roots list | |
3172 | */ | |
3173 | spin_lock(&fs_info->trans_lock); | |
3174 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3175 | root_list) { | |
3176 | if (dead_root->root_key.objectid == | |
3177 | found_key.objectid) { | |
3178 | is_dead_root = 1; | |
3179 | break; | |
3180 | } | |
3181 | } | |
3182 | spin_unlock(&fs_info->trans_lock); | |
3183 | if (is_dead_root) { | |
3184 | /* prevent this orphan from being found again */ | |
3185 | key.offset = found_key.objectid - 1; | |
3186 | continue; | |
3187 | } | |
3188 | } | |
7b128766 | 3189 | /* |
a8c9e576 JB |
3190 | * Inode is already gone but the orphan item is still there, |
3191 | * kill the orphan item. | |
7b128766 | 3192 | */ |
a8c9e576 JB |
3193 | if (ret == -ESTALE) { |
3194 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
3195 | if (IS_ERR(trans)) { |
3196 | ret = PTR_ERR(trans); | |
3197 | goto out; | |
3198 | } | |
c2cf52eb SK |
3199 | btrfs_debug(root->fs_info, "auto deleting %Lu", |
3200 | found_key.objectid); | |
a8c9e576 JB |
3201 | ret = btrfs_del_orphan_item(trans, root, |
3202 | found_key.objectid); | |
5b21f2ed | 3203 | btrfs_end_transaction(trans, root); |
4ef31a45 JB |
3204 | if (ret) |
3205 | goto out; | |
7b128766 JB |
3206 | continue; |
3207 | } | |
3208 | ||
a8c9e576 JB |
3209 | /* |
3210 | * add this inode to the orphan list so btrfs_orphan_del does | |
3211 | * the proper thing when we hit it | |
3212 | */ | |
8a35d95f JB |
3213 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3214 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3215 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3216 | |
7b128766 JB |
3217 | /* if we have links, this was a truncate, lets do that */ |
3218 | if (inode->i_nlink) { | |
a41ad394 JB |
3219 | if (!S_ISREG(inode->i_mode)) { |
3220 | WARN_ON(1); | |
3221 | iput(inode); | |
3222 | continue; | |
3223 | } | |
7b128766 | 3224 | nr_truncate++; |
f3fe820c JB |
3225 | |
3226 | /* 1 for the orphan item deletion. */ | |
3227 | trans = btrfs_start_transaction(root, 1); | |
3228 | if (IS_ERR(trans)) { | |
c69b26b0 | 3229 | iput(inode); |
f3fe820c JB |
3230 | ret = PTR_ERR(trans); |
3231 | goto out; | |
3232 | } | |
3233 | ret = btrfs_orphan_add(trans, inode); | |
3234 | btrfs_end_transaction(trans, root); | |
c69b26b0 JB |
3235 | if (ret) { |
3236 | iput(inode); | |
f3fe820c | 3237 | goto out; |
c69b26b0 | 3238 | } |
f3fe820c | 3239 | |
66b4ffd1 | 3240 | ret = btrfs_truncate(inode); |
4a7d0f68 JB |
3241 | if (ret) |
3242 | btrfs_orphan_del(NULL, inode); | |
7b128766 JB |
3243 | } else { |
3244 | nr_unlink++; | |
3245 | } | |
3246 | ||
3247 | /* this will do delete_inode and everything for us */ | |
3248 | iput(inode); | |
66b4ffd1 JB |
3249 | if (ret) |
3250 | goto out; | |
7b128766 | 3251 | } |
3254c876 MX |
3252 | /* release the path since we're done with it */ |
3253 | btrfs_release_path(path); | |
3254 | ||
d68fc57b YZ |
3255 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3256 | ||
3257 | if (root->orphan_block_rsv) | |
3258 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
3259 | (u64)-1); | |
3260 | ||
3261 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
7a7eaa40 | 3262 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
3263 | if (!IS_ERR(trans)) |
3264 | btrfs_end_transaction(trans, root); | |
d68fc57b | 3265 | } |
7b128766 JB |
3266 | |
3267 | if (nr_unlink) | |
4884b476 | 3268 | btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3269 | if (nr_truncate) |
4884b476 | 3270 | btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3271 | |
3272 | out: | |
3273 | if (ret) | |
c2cf52eb SK |
3274 | btrfs_crit(root->fs_info, |
3275 | "could not do orphan cleanup %d", ret); | |
66b4ffd1 JB |
3276 | btrfs_free_path(path); |
3277 | return ret; | |
7b128766 JB |
3278 | } |
3279 | ||
46a53cca CM |
3280 | /* |
3281 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3282 | * don't find any xattrs, we know there can't be any acls. | |
3283 | * | |
3284 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3285 | */ | |
3286 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
3287 | int slot, u64 objectid) | |
3288 | { | |
3289 | u32 nritems = btrfs_header_nritems(leaf); | |
3290 | struct btrfs_key found_key; | |
f23b5a59 JB |
3291 | static u64 xattr_access = 0; |
3292 | static u64 xattr_default = 0; | |
46a53cca CM |
3293 | int scanned = 0; |
3294 | ||
f23b5a59 JB |
3295 | if (!xattr_access) { |
3296 | xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS, | |
3297 | strlen(POSIX_ACL_XATTR_ACCESS)); | |
3298 | xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT, | |
3299 | strlen(POSIX_ACL_XATTR_DEFAULT)); | |
3300 | } | |
3301 | ||
46a53cca CM |
3302 | slot++; |
3303 | while (slot < nritems) { | |
3304 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3305 | ||
3306 | /* we found a different objectid, there must not be acls */ | |
3307 | if (found_key.objectid != objectid) | |
3308 | return 0; | |
3309 | ||
3310 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 JB |
3311 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
3312 | if (found_key.offset == xattr_access || | |
3313 | found_key.offset == xattr_default) | |
3314 | return 1; | |
3315 | } | |
46a53cca CM |
3316 | |
3317 | /* | |
3318 | * we found a key greater than an xattr key, there can't | |
3319 | * be any acls later on | |
3320 | */ | |
3321 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3322 | return 0; | |
3323 | ||
3324 | slot++; | |
3325 | scanned++; | |
3326 | ||
3327 | /* | |
3328 | * it goes inode, inode backrefs, xattrs, extents, | |
3329 | * so if there are a ton of hard links to an inode there can | |
3330 | * be a lot of backrefs. Don't waste time searching too hard, | |
3331 | * this is just an optimization | |
3332 | */ | |
3333 | if (scanned >= 8) | |
3334 | break; | |
3335 | } | |
3336 | /* we hit the end of the leaf before we found an xattr or | |
3337 | * something larger than an xattr. We have to assume the inode | |
3338 | * has acls | |
3339 | */ | |
3340 | return 1; | |
3341 | } | |
3342 | ||
d352ac68 CM |
3343 | /* |
3344 | * read an inode from the btree into the in-memory inode | |
3345 | */ | |
5d4f98a2 | 3346 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
3347 | { |
3348 | struct btrfs_path *path; | |
5f39d397 | 3349 | struct extent_buffer *leaf; |
39279cc3 | 3350 | struct btrfs_inode_item *inode_item; |
0b86a832 | 3351 | struct btrfs_timespec *tspec; |
39279cc3 CM |
3352 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3353 | struct btrfs_key location; | |
46a53cca | 3354 | int maybe_acls; |
618e21d5 | 3355 | u32 rdev; |
39279cc3 | 3356 | int ret; |
2f7e33d4 MX |
3357 | bool filled = false; |
3358 | ||
3359 | ret = btrfs_fill_inode(inode, &rdev); | |
3360 | if (!ret) | |
3361 | filled = true; | |
39279cc3 CM |
3362 | |
3363 | path = btrfs_alloc_path(); | |
1748f843 MF |
3364 | if (!path) |
3365 | goto make_bad; | |
3366 | ||
d90c7321 | 3367 | path->leave_spinning = 1; |
39279cc3 | 3368 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3369 | |
39279cc3 | 3370 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 3371 | if (ret) |
39279cc3 | 3372 | goto make_bad; |
39279cc3 | 3373 | |
5f39d397 | 3374 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3375 | |
3376 | if (filled) | |
3377 | goto cache_acl; | |
3378 | ||
5f39d397 CM |
3379 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3380 | struct btrfs_inode_item); | |
5f39d397 | 3381 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3382 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3383 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3384 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 3385 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
3386 | |
3387 | tspec = btrfs_inode_atime(inode_item); | |
3388 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3389 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3390 | ||
3391 | tspec = btrfs_inode_mtime(inode_item); | |
3392 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3393 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3394 | ||
3395 | tspec = btrfs_inode_ctime(inode_item); | |
3396 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3397 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3398 | ||
a76a3cd4 | 3399 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3400 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3401 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3402 | ||
3403 | /* | |
3404 | * If we were modified in the current generation and evicted from memory | |
3405 | * and then re-read we need to do a full sync since we don't have any | |
3406 | * idea about which extents were modified before we were evicted from | |
3407 | * cache. | |
3408 | */ | |
3409 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
3410 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3411 | &BTRFS_I(inode)->runtime_flags); | |
3412 | ||
0c4d2d95 | 3413 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 3414 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 3415 | inode->i_rdev = 0; |
5f39d397 CM |
3416 | rdev = btrfs_inode_rdev(leaf, inode_item); |
3417 | ||
aec7477b | 3418 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 3419 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
2f7e33d4 | 3420 | cache_acl: |
46a53cca CM |
3421 | /* |
3422 | * try to precache a NULL acl entry for files that don't have | |
3423 | * any xattrs or acls | |
3424 | */ | |
33345d01 LZ |
3425 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
3426 | btrfs_ino(inode)); | |
72c04902 AV |
3427 | if (!maybe_acls) |
3428 | cache_no_acl(inode); | |
46a53cca | 3429 | |
39279cc3 | 3430 | btrfs_free_path(path); |
39279cc3 | 3431 | |
39279cc3 | 3432 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3433 | case S_IFREG: |
3434 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 3435 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 3436 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3437 | inode->i_fop = &btrfs_file_operations; |
3438 | inode->i_op = &btrfs_file_inode_operations; | |
3439 | break; | |
3440 | case S_IFDIR: | |
3441 | inode->i_fop = &btrfs_dir_file_operations; | |
3442 | if (root == root->fs_info->tree_root) | |
3443 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
3444 | else | |
3445 | inode->i_op = &btrfs_dir_inode_operations; | |
3446 | break; | |
3447 | case S_IFLNK: | |
3448 | inode->i_op = &btrfs_symlink_inode_operations; | |
3449 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 3450 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 3451 | break; |
618e21d5 | 3452 | default: |
0279b4cd | 3453 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3454 | init_special_inode(inode, inode->i_mode, rdev); |
3455 | break; | |
39279cc3 | 3456 | } |
6cbff00f CH |
3457 | |
3458 | btrfs_update_iflags(inode); | |
39279cc3 CM |
3459 | return; |
3460 | ||
3461 | make_bad: | |
39279cc3 | 3462 | btrfs_free_path(path); |
39279cc3 CM |
3463 | make_bad_inode(inode); |
3464 | } | |
3465 | ||
d352ac68 CM |
3466 | /* |
3467 | * given a leaf and an inode, copy the inode fields into the leaf | |
3468 | */ | |
e02119d5 CM |
3469 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3470 | struct extent_buffer *leaf, | |
5f39d397 | 3471 | struct btrfs_inode_item *item, |
39279cc3 CM |
3472 | struct inode *inode) |
3473 | { | |
51fab693 LB |
3474 | struct btrfs_map_token token; |
3475 | ||
3476 | btrfs_init_map_token(&token); | |
5f39d397 | 3477 | |
51fab693 LB |
3478 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3479 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3480 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3481 | &token); | |
3482 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3483 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3484 | |
51fab693 LB |
3485 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item), |
3486 | inode->i_atime.tv_sec, &token); | |
3487 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item), | |
3488 | inode->i_atime.tv_nsec, &token); | |
5f39d397 | 3489 | |
51fab693 LB |
3490 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item), |
3491 | inode->i_mtime.tv_sec, &token); | |
3492 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
3493 | inode->i_mtime.tv_nsec, &token); | |
5f39d397 | 3494 | |
51fab693 LB |
3495 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item), |
3496 | inode->i_ctime.tv_sec, &token); | |
3497 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
3498 | inode->i_ctime.tv_nsec, &token); | |
5f39d397 | 3499 | |
51fab693 LB |
3500 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3501 | &token); | |
3502 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3503 | &token); | |
3504 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3505 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3506 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3507 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3508 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3509 | } |
3510 | ||
d352ac68 CM |
3511 | /* |
3512 | * copy everything in the in-memory inode into the btree. | |
3513 | */ | |
2115133f | 3514 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3515 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3516 | { |
3517 | struct btrfs_inode_item *inode_item; | |
3518 | struct btrfs_path *path; | |
5f39d397 | 3519 | struct extent_buffer *leaf; |
39279cc3 CM |
3520 | int ret; |
3521 | ||
3522 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3523 | if (!path) |
3524 | return -ENOMEM; | |
3525 | ||
b9473439 | 3526 | path->leave_spinning = 1; |
16cdcec7 MX |
3527 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3528 | 1); | |
39279cc3 CM |
3529 | if (ret) { |
3530 | if (ret > 0) | |
3531 | ret = -ENOENT; | |
3532 | goto failed; | |
3533 | } | |
3534 | ||
b4ce94de | 3535 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
3536 | leaf = path->nodes[0]; |
3537 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3538 | struct btrfs_inode_item); |
39279cc3 | 3539 | |
e02119d5 | 3540 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3541 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3542 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3543 | ret = 0; |
3544 | failed: | |
39279cc3 CM |
3545 | btrfs_free_path(path); |
3546 | return ret; | |
3547 | } | |
3548 | ||
2115133f CM |
3549 | /* |
3550 | * copy everything in the in-memory inode into the btree. | |
3551 | */ | |
3552 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3553 | struct btrfs_root *root, struct inode *inode) | |
3554 | { | |
3555 | int ret; | |
3556 | ||
3557 | /* | |
3558 | * If the inode is a free space inode, we can deadlock during commit | |
3559 | * if we put it into the delayed code. | |
3560 | * | |
3561 | * The data relocation inode should also be directly updated | |
3562 | * without delay | |
3563 | */ | |
83eea1f1 | 3564 | if (!btrfs_is_free_space_inode(inode) |
2115133f | 3565 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { |
8ea05e3a AB |
3566 | btrfs_update_root_times(trans, root); |
3567 | ||
2115133f CM |
3568 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3569 | if (!ret) | |
3570 | btrfs_set_inode_last_trans(trans, inode); | |
3571 | return ret; | |
3572 | } | |
3573 | ||
3574 | return btrfs_update_inode_item(trans, root, inode); | |
3575 | } | |
3576 | ||
be6aef60 JB |
3577 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3578 | struct btrfs_root *root, | |
3579 | struct inode *inode) | |
2115133f CM |
3580 | { |
3581 | int ret; | |
3582 | ||
3583 | ret = btrfs_update_inode(trans, root, inode); | |
3584 | if (ret == -ENOSPC) | |
3585 | return btrfs_update_inode_item(trans, root, inode); | |
3586 | return ret; | |
3587 | } | |
3588 | ||
d352ac68 CM |
3589 | /* |
3590 | * unlink helper that gets used here in inode.c and in the tree logging | |
3591 | * recovery code. It remove a link in a directory with a given name, and | |
3592 | * also drops the back refs in the inode to the directory | |
3593 | */ | |
92986796 AV |
3594 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3595 | struct btrfs_root *root, | |
3596 | struct inode *dir, struct inode *inode, | |
3597 | const char *name, int name_len) | |
39279cc3 CM |
3598 | { |
3599 | struct btrfs_path *path; | |
39279cc3 | 3600 | int ret = 0; |
5f39d397 | 3601 | struct extent_buffer *leaf; |
39279cc3 | 3602 | struct btrfs_dir_item *di; |
5f39d397 | 3603 | struct btrfs_key key; |
aec7477b | 3604 | u64 index; |
33345d01 LZ |
3605 | u64 ino = btrfs_ino(inode); |
3606 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3607 | |
3608 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3609 | if (!path) { |
3610 | ret = -ENOMEM; | |
554233a6 | 3611 | goto out; |
54aa1f4d CM |
3612 | } |
3613 | ||
b9473439 | 3614 | path->leave_spinning = 1; |
33345d01 | 3615 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
3616 | name, name_len, -1); |
3617 | if (IS_ERR(di)) { | |
3618 | ret = PTR_ERR(di); | |
3619 | goto err; | |
3620 | } | |
3621 | if (!di) { | |
3622 | ret = -ENOENT; | |
3623 | goto err; | |
3624 | } | |
5f39d397 CM |
3625 | leaf = path->nodes[0]; |
3626 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 3627 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3628 | if (ret) |
3629 | goto err; | |
b3b4aa74 | 3630 | btrfs_release_path(path); |
39279cc3 | 3631 | |
33345d01 LZ |
3632 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3633 | dir_ino, &index); | |
aec7477b | 3634 | if (ret) { |
c2cf52eb SK |
3635 | btrfs_info(root->fs_info, |
3636 | "failed to delete reference to %.*s, inode %llu parent %llu", | |
3637 | name_len, name, | |
3638 | (unsigned long long)ino, (unsigned long long)dir_ino); | |
79787eaa | 3639 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
3640 | goto err; |
3641 | } | |
3642 | ||
16cdcec7 | 3643 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3644 | if (ret) { |
3645 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 3646 | goto err; |
79787eaa | 3647 | } |
39279cc3 | 3648 | |
e02119d5 | 3649 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 3650 | inode, dir_ino); |
79787eaa JM |
3651 | if (ret != 0 && ret != -ENOENT) { |
3652 | btrfs_abort_transaction(trans, root, ret); | |
3653 | goto err; | |
3654 | } | |
e02119d5 CM |
3655 | |
3656 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
3657 | dir, index); | |
6418c961 CM |
3658 | if (ret == -ENOENT) |
3659 | ret = 0; | |
d4e3991b ZB |
3660 | else if (ret) |
3661 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 CM |
3662 | err: |
3663 | btrfs_free_path(path); | |
e02119d5 CM |
3664 | if (ret) |
3665 | goto out; | |
3666 | ||
3667 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
3668 | inode_inc_iversion(inode); |
3669 | inode_inc_iversion(dir); | |
e02119d5 | 3670 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
b9959295 | 3671 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 3672 | out: |
39279cc3 CM |
3673 | return ret; |
3674 | } | |
3675 | ||
92986796 AV |
3676 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3677 | struct btrfs_root *root, | |
3678 | struct inode *dir, struct inode *inode, | |
3679 | const char *name, int name_len) | |
3680 | { | |
3681 | int ret; | |
3682 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
3683 | if (!ret) { | |
3684 | btrfs_drop_nlink(inode); | |
3685 | ret = btrfs_update_inode(trans, root, inode); | |
3686 | } | |
3687 | return ret; | |
3688 | } | |
39279cc3 | 3689 | |
a22285a6 YZ |
3690 | /* |
3691 | * helper to start transaction for unlink and rmdir. | |
3692 | * | |
d52be818 JB |
3693 | * unlink and rmdir are special in btrfs, they do not always free space, so |
3694 | * if we cannot make our reservations the normal way try and see if there is | |
3695 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
3696 | * allow the unlink to occur. | |
a22285a6 | 3697 | */ |
d52be818 | 3698 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 3699 | { |
39279cc3 | 3700 | struct btrfs_trans_handle *trans; |
a22285a6 | 3701 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d YZ |
3702 | int ret; |
3703 | ||
e70bea5f JB |
3704 | /* |
3705 | * 1 for the possible orphan item | |
3706 | * 1 for the dir item | |
3707 | * 1 for the dir index | |
3708 | * 1 for the inode ref | |
e70bea5f JB |
3709 | * 1 for the inode |
3710 | */ | |
6e137ed3 | 3711 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
3712 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
3713 | return trans; | |
4df27c4d | 3714 | |
d52be818 JB |
3715 | if (PTR_ERR(trans) == -ENOSPC) { |
3716 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); | |
4df27c4d | 3717 | |
d52be818 JB |
3718 | trans = btrfs_start_transaction(root, 0); |
3719 | if (IS_ERR(trans)) | |
3720 | return trans; | |
3721 | ret = btrfs_cond_migrate_bytes(root->fs_info, | |
3722 | &root->fs_info->trans_block_rsv, | |
3723 | num_bytes, 5); | |
3724 | if (ret) { | |
3725 | btrfs_end_transaction(trans, root); | |
3726 | return ERR_PTR(ret); | |
a22285a6 | 3727 | } |
5a77d76c | 3728 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
d52be818 | 3729 | trans->bytes_reserved = num_bytes; |
a22285a6 | 3730 | } |
d52be818 | 3731 | return trans; |
a22285a6 YZ |
3732 | } |
3733 | ||
3734 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3735 | { | |
3736 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3737 | struct btrfs_trans_handle *trans; | |
3738 | struct inode *inode = dentry->d_inode; | |
3739 | int ret; | |
a22285a6 | 3740 | |
d52be818 | 3741 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
3742 | if (IS_ERR(trans)) |
3743 | return PTR_ERR(trans); | |
5f39d397 | 3744 | |
12fcfd22 CM |
3745 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
3746 | ||
e02119d5 CM |
3747 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3748 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
3749 | if (ret) |
3750 | goto out; | |
7b128766 | 3751 | |
a22285a6 | 3752 | if (inode->i_nlink == 0) { |
7b128766 | 3753 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
3754 | if (ret) |
3755 | goto out; | |
a22285a6 | 3756 | } |
7b128766 | 3757 | |
b532402e | 3758 | out: |
d52be818 | 3759 | btrfs_end_transaction(trans, root); |
b53d3f5d | 3760 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
3761 | return ret; |
3762 | } | |
3763 | ||
4df27c4d YZ |
3764 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
3765 | struct btrfs_root *root, | |
3766 | struct inode *dir, u64 objectid, | |
3767 | const char *name, int name_len) | |
3768 | { | |
3769 | struct btrfs_path *path; | |
3770 | struct extent_buffer *leaf; | |
3771 | struct btrfs_dir_item *di; | |
3772 | struct btrfs_key key; | |
3773 | u64 index; | |
3774 | int ret; | |
33345d01 | 3775 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
3776 | |
3777 | path = btrfs_alloc_path(); | |
3778 | if (!path) | |
3779 | return -ENOMEM; | |
3780 | ||
33345d01 | 3781 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3782 | name, name_len, -1); |
79787eaa JM |
3783 | if (IS_ERR_OR_NULL(di)) { |
3784 | if (!di) | |
3785 | ret = -ENOENT; | |
3786 | else | |
3787 | ret = PTR_ERR(di); | |
3788 | goto out; | |
3789 | } | |
4df27c4d YZ |
3790 | |
3791 | leaf = path->nodes[0]; | |
3792 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3793 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3794 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
3795 | if (ret) { |
3796 | btrfs_abort_transaction(trans, root, ret); | |
3797 | goto out; | |
3798 | } | |
b3b4aa74 | 3799 | btrfs_release_path(path); |
4df27c4d YZ |
3800 | |
3801 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3802 | objectid, root->root_key.objectid, | |
33345d01 | 3803 | dir_ino, &index, name, name_len); |
4df27c4d | 3804 | if (ret < 0) { |
79787eaa JM |
3805 | if (ret != -ENOENT) { |
3806 | btrfs_abort_transaction(trans, root, ret); | |
3807 | goto out; | |
3808 | } | |
33345d01 | 3809 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3810 | name, name_len); |
79787eaa JM |
3811 | if (IS_ERR_OR_NULL(di)) { |
3812 | if (!di) | |
3813 | ret = -ENOENT; | |
3814 | else | |
3815 | ret = PTR_ERR(di); | |
3816 | btrfs_abort_transaction(trans, root, ret); | |
3817 | goto out; | |
3818 | } | |
4df27c4d YZ |
3819 | |
3820 | leaf = path->nodes[0]; | |
3821 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 3822 | btrfs_release_path(path); |
4df27c4d YZ |
3823 | index = key.offset; |
3824 | } | |
945d8962 | 3825 | btrfs_release_path(path); |
4df27c4d | 3826 | |
16cdcec7 | 3827 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3828 | if (ret) { |
3829 | btrfs_abort_transaction(trans, root, ret); | |
3830 | goto out; | |
3831 | } | |
4df27c4d YZ |
3832 | |
3833 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 3834 | inode_inc_iversion(dir); |
4df27c4d | 3835 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
5a24e84c | 3836 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
3837 | if (ret) |
3838 | btrfs_abort_transaction(trans, root, ret); | |
3839 | out: | |
71d7aed0 | 3840 | btrfs_free_path(path); |
79787eaa | 3841 | return ret; |
4df27c4d YZ |
3842 | } |
3843 | ||
39279cc3 CM |
3844 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3845 | { | |
3846 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3847 | int err = 0; |
39279cc3 | 3848 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3849 | struct btrfs_trans_handle *trans; |
39279cc3 | 3850 | |
b3ae244e | 3851 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 3852 | return -ENOTEMPTY; |
b3ae244e DS |
3853 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
3854 | return -EPERM; | |
134d4512 | 3855 | |
d52be818 | 3856 | trans = __unlink_start_trans(dir); |
a22285a6 | 3857 | if (IS_ERR(trans)) |
5df6a9f6 | 3858 | return PTR_ERR(trans); |
5df6a9f6 | 3859 | |
33345d01 | 3860 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
3861 | err = btrfs_unlink_subvol(trans, root, dir, |
3862 | BTRFS_I(inode)->location.objectid, | |
3863 | dentry->d_name.name, | |
3864 | dentry->d_name.len); | |
3865 | goto out; | |
3866 | } | |
3867 | ||
7b128766 JB |
3868 | err = btrfs_orphan_add(trans, inode); |
3869 | if (err) | |
4df27c4d | 3870 | goto out; |
7b128766 | 3871 | |
39279cc3 | 3872 | /* now the directory is empty */ |
e02119d5 CM |
3873 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3874 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3875 | if (!err) |
dbe674a9 | 3876 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3877 | out: |
d52be818 | 3878 | btrfs_end_transaction(trans, root); |
b53d3f5d | 3879 | btrfs_btree_balance_dirty(root); |
3954401f | 3880 | |
39279cc3 CM |
3881 | return err; |
3882 | } | |
3883 | ||
39279cc3 CM |
3884 | /* |
3885 | * this can truncate away extent items, csum items and directory items. | |
3886 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3887 | * any higher than new_size |
39279cc3 CM |
3888 | * |
3889 | * csum items that cross the new i_size are truncated to the new size | |
3890 | * as well. | |
7b128766 JB |
3891 | * |
3892 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3893 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3894 | */ |
8082510e YZ |
3895 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3896 | struct btrfs_root *root, | |
3897 | struct inode *inode, | |
3898 | u64 new_size, u32 min_type) | |
39279cc3 | 3899 | { |
39279cc3 | 3900 | struct btrfs_path *path; |
5f39d397 | 3901 | struct extent_buffer *leaf; |
39279cc3 | 3902 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3903 | struct btrfs_key key; |
3904 | struct btrfs_key found_key; | |
39279cc3 | 3905 | u64 extent_start = 0; |
db94535d | 3906 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3907 | u64 extent_offset = 0; |
39279cc3 | 3908 | u64 item_end = 0; |
8082510e | 3909 | u32 found_type = (u8)-1; |
39279cc3 CM |
3910 | int found_extent; |
3911 | int del_item; | |
85e21bac CM |
3912 | int pending_del_nr = 0; |
3913 | int pending_del_slot = 0; | |
179e29e4 | 3914 | int extent_type = -1; |
8082510e YZ |
3915 | int ret; |
3916 | int err = 0; | |
33345d01 | 3917 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
3918 | |
3919 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3920 | |
0eb0e19c MF |
3921 | path = btrfs_alloc_path(); |
3922 | if (!path) | |
3923 | return -ENOMEM; | |
3924 | path->reada = -1; | |
3925 | ||
5dc562c5 JB |
3926 | /* |
3927 | * We want to drop from the next block forward in case this new size is | |
3928 | * not block aligned since we will be keeping the last block of the | |
3929 | * extent just the way it is. | |
3930 | */ | |
0af3d00b | 3931 | if (root->ref_cows || root == root->fs_info->tree_root) |
fda2832f QW |
3932 | btrfs_drop_extent_cache(inode, ALIGN(new_size, |
3933 | root->sectorsize), (u64)-1, 0); | |
8082510e | 3934 | |
16cdcec7 MX |
3935 | /* |
3936 | * This function is also used to drop the items in the log tree before | |
3937 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
3938 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
3939 | * items. | |
3940 | */ | |
3941 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
3942 | btrfs_kill_delayed_inode_items(inode); | |
3943 | ||
33345d01 | 3944 | key.objectid = ino; |
39279cc3 | 3945 | key.offset = (u64)-1; |
5f39d397 CM |
3946 | key.type = (u8)-1; |
3947 | ||
85e21bac | 3948 | search_again: |
b9473439 | 3949 | path->leave_spinning = 1; |
85e21bac | 3950 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3951 | if (ret < 0) { |
3952 | err = ret; | |
3953 | goto out; | |
3954 | } | |
d397712b | 3955 | |
85e21bac | 3956 | if (ret > 0) { |
e02119d5 CM |
3957 | /* there are no items in the tree for us to truncate, we're |
3958 | * done | |
3959 | */ | |
8082510e YZ |
3960 | if (path->slots[0] == 0) |
3961 | goto out; | |
85e21bac CM |
3962 | path->slots[0]--; |
3963 | } | |
3964 | ||
d397712b | 3965 | while (1) { |
39279cc3 | 3966 | fi = NULL; |
5f39d397 CM |
3967 | leaf = path->nodes[0]; |
3968 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3969 | found_type = btrfs_key_type(&found_key); | |
39279cc3 | 3970 | |
33345d01 | 3971 | if (found_key.objectid != ino) |
39279cc3 | 3972 | break; |
5f39d397 | 3973 | |
85e21bac | 3974 | if (found_type < min_type) |
39279cc3 CM |
3975 | break; |
3976 | ||
5f39d397 | 3977 | item_end = found_key.offset; |
39279cc3 | 3978 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3979 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3980 | struct btrfs_file_extent_item); |
179e29e4 CM |
3981 | extent_type = btrfs_file_extent_type(leaf, fi); |
3982 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 3983 | item_end += |
db94535d | 3984 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3985 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3986 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3987 | fi); |
39279cc3 | 3988 | } |
008630c1 | 3989 | item_end--; |
39279cc3 | 3990 | } |
8082510e YZ |
3991 | if (found_type > min_type) { |
3992 | del_item = 1; | |
3993 | } else { | |
3994 | if (item_end < new_size) | |
b888db2b | 3995 | break; |
8082510e YZ |
3996 | if (found_key.offset >= new_size) |
3997 | del_item = 1; | |
3998 | else | |
3999 | del_item = 0; | |
39279cc3 | 4000 | } |
39279cc3 | 4001 | found_extent = 0; |
39279cc3 | 4002 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4003 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4004 | goto delete; | |
4005 | ||
4006 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 4007 | u64 num_dec; |
db94535d | 4008 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4009 | if (!del_item) { |
db94535d CM |
4010 | u64 orig_num_bytes = |
4011 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4012 | extent_num_bytes = ALIGN(new_size - |
4013 | found_key.offset, | |
4014 | root->sectorsize); | |
db94535d CM |
4015 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4016 | extent_num_bytes); | |
4017 | num_dec = (orig_num_bytes - | |
9069218d | 4018 | extent_num_bytes); |
e02119d5 | 4019 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 4020 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4021 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4022 | } else { |
db94535d CM |
4023 | extent_num_bytes = |
4024 | btrfs_file_extent_disk_num_bytes(leaf, | |
4025 | fi); | |
5d4f98a2 YZ |
4026 | extent_offset = found_key.offset - |
4027 | btrfs_file_extent_offset(leaf, fi); | |
4028 | ||
39279cc3 | 4029 | /* FIXME blocksize != 4096 */ |
9069218d | 4030 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4031 | if (extent_start != 0) { |
4032 | found_extent = 1; | |
e02119d5 | 4033 | if (root->ref_cows) |
a76a3cd4 | 4034 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4035 | } |
39279cc3 | 4036 | } |
9069218d | 4037 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4038 | /* |
4039 | * we can't truncate inline items that have had | |
4040 | * special encodings | |
4041 | */ | |
4042 | if (!del_item && | |
4043 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
4044 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
4045 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
4046 | u32 size = new_size - found_key.offset; |
4047 | ||
4048 | if (root->ref_cows) { | |
a76a3cd4 YZ |
4049 | inode_sub_bytes(inode, item_end + 1 - |
4050 | new_size); | |
e02119d5 CM |
4051 | } |
4052 | size = | |
4053 | btrfs_file_extent_calc_inline_size(size); | |
afe5fea7 | 4054 | btrfs_truncate_item(root, path, size, 1); |
e02119d5 | 4055 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
4056 | inode_sub_bytes(inode, item_end + 1 - |
4057 | found_key.offset); | |
9069218d | 4058 | } |
39279cc3 | 4059 | } |
179e29e4 | 4060 | delete: |
39279cc3 | 4061 | if (del_item) { |
85e21bac CM |
4062 | if (!pending_del_nr) { |
4063 | /* no pending yet, add ourselves */ | |
4064 | pending_del_slot = path->slots[0]; | |
4065 | pending_del_nr = 1; | |
4066 | } else if (pending_del_nr && | |
4067 | path->slots[0] + 1 == pending_del_slot) { | |
4068 | /* hop on the pending chunk */ | |
4069 | pending_del_nr++; | |
4070 | pending_del_slot = path->slots[0]; | |
4071 | } else { | |
d397712b | 4072 | BUG(); |
85e21bac | 4073 | } |
39279cc3 CM |
4074 | } else { |
4075 | break; | |
4076 | } | |
0af3d00b JB |
4077 | if (found_extent && (root->ref_cows || |
4078 | root == root->fs_info->tree_root)) { | |
b9473439 | 4079 | btrfs_set_path_blocking(path); |
39279cc3 | 4080 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4081 | extent_num_bytes, 0, |
4082 | btrfs_header_owner(leaf), | |
66d7e7f0 | 4083 | ino, extent_offset, 0); |
39279cc3 CM |
4084 | BUG_ON(ret); |
4085 | } | |
85e21bac | 4086 | |
8082510e YZ |
4087 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4088 | break; | |
4089 | ||
4090 | if (path->slots[0] == 0 || | |
4091 | path->slots[0] != pending_del_slot) { | |
8082510e YZ |
4092 | if (pending_del_nr) { |
4093 | ret = btrfs_del_items(trans, root, path, | |
4094 | pending_del_slot, | |
4095 | pending_del_nr); | |
79787eaa JM |
4096 | if (ret) { |
4097 | btrfs_abort_transaction(trans, | |
4098 | root, ret); | |
4099 | goto error; | |
4100 | } | |
8082510e YZ |
4101 | pending_del_nr = 0; |
4102 | } | |
b3b4aa74 | 4103 | btrfs_release_path(path); |
85e21bac | 4104 | goto search_again; |
8082510e YZ |
4105 | } else { |
4106 | path->slots[0]--; | |
85e21bac | 4107 | } |
39279cc3 | 4108 | } |
8082510e | 4109 | out: |
85e21bac CM |
4110 | if (pending_del_nr) { |
4111 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4112 | pending_del_nr); | |
79787eaa JM |
4113 | if (ret) |
4114 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 4115 | } |
79787eaa | 4116 | error: |
39279cc3 | 4117 | btrfs_free_path(path); |
8082510e | 4118 | return err; |
39279cc3 CM |
4119 | } |
4120 | ||
4121 | /* | |
2aaa6655 JB |
4122 | * btrfs_truncate_page - read, zero a chunk and write a page |
4123 | * @inode - inode that we're zeroing | |
4124 | * @from - the offset to start zeroing | |
4125 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4126 | * offset | |
4127 | * @front - zero up to the offset instead of from the offset on | |
4128 | * | |
4129 | * This will find the page for the "from" offset and cow the page and zero the | |
4130 | * part we want to zero. This is used with truncate and hole punching. | |
39279cc3 | 4131 | */ |
2aaa6655 JB |
4132 | int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len, |
4133 | int front) | |
39279cc3 | 4134 | { |
2aaa6655 | 4135 | struct address_space *mapping = inode->i_mapping; |
db94535d | 4136 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4137 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4138 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4139 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 4140 | char *kaddr; |
db94535d | 4141 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
4142 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
4143 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
4144 | struct page *page; | |
3b16a4e3 | 4145 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4146 | int ret = 0; |
a52d9a80 | 4147 | u64 page_start; |
e6dcd2dc | 4148 | u64 page_end; |
39279cc3 | 4149 | |
2aaa6655 JB |
4150 | if ((offset & (blocksize - 1)) == 0 && |
4151 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4152 | goto out; |
0ca1f7ce | 4153 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
4154 | if (ret) |
4155 | goto out; | |
39279cc3 | 4156 | |
211c17f5 | 4157 | again: |
3b16a4e3 | 4158 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4159 | if (!page) { |
0ca1f7ce | 4160 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
ac6a2b36 | 4161 | ret = -ENOMEM; |
39279cc3 | 4162 | goto out; |
5d5e103a | 4163 | } |
e6dcd2dc CM |
4164 | |
4165 | page_start = page_offset(page); | |
4166 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
4167 | ||
39279cc3 | 4168 | if (!PageUptodate(page)) { |
9ebefb18 | 4169 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4170 | lock_page(page); |
211c17f5 CM |
4171 | if (page->mapping != mapping) { |
4172 | unlock_page(page); | |
4173 | page_cache_release(page); | |
4174 | goto again; | |
4175 | } | |
39279cc3 CM |
4176 | if (!PageUptodate(page)) { |
4177 | ret = -EIO; | |
89642229 | 4178 | goto out_unlock; |
39279cc3 CM |
4179 | } |
4180 | } | |
211c17f5 | 4181 | wait_on_page_writeback(page); |
e6dcd2dc | 4182 | |
d0082371 | 4183 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
4184 | set_page_extent_mapped(page); |
4185 | ||
4186 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
4187 | if (ordered) { | |
2ac55d41 JB |
4188 | unlock_extent_cached(io_tree, page_start, page_end, |
4189 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
4190 | unlock_page(page); |
4191 | page_cache_release(page); | |
eb84ae03 | 4192 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4193 | btrfs_put_ordered_extent(ordered); |
4194 | goto again; | |
4195 | } | |
4196 | ||
2ac55d41 | 4197 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
4198 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4199 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4200 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4201 | |
2ac55d41 JB |
4202 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
4203 | &cached_state); | |
9ed74f2d | 4204 | if (ret) { |
2ac55d41 JB |
4205 | unlock_extent_cached(io_tree, page_start, page_end, |
4206 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
4207 | goto out_unlock; |
4208 | } | |
4209 | ||
e6dcd2dc | 4210 | if (offset != PAGE_CACHE_SIZE) { |
2aaa6655 JB |
4211 | if (!len) |
4212 | len = PAGE_CACHE_SIZE - offset; | |
e6dcd2dc | 4213 | kaddr = kmap(page); |
2aaa6655 JB |
4214 | if (front) |
4215 | memset(kaddr, 0, offset); | |
4216 | else | |
4217 | memset(kaddr + offset, 0, len); | |
e6dcd2dc CM |
4218 | flush_dcache_page(page); |
4219 | kunmap(page); | |
4220 | } | |
247e743c | 4221 | ClearPageChecked(page); |
e6dcd2dc | 4222 | set_page_dirty(page); |
2ac55d41 JB |
4223 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
4224 | GFP_NOFS); | |
39279cc3 | 4225 | |
89642229 | 4226 | out_unlock: |
5d5e103a | 4227 | if (ret) |
0ca1f7ce | 4228 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
4229 | unlock_page(page); |
4230 | page_cache_release(page); | |
4231 | out: | |
4232 | return ret; | |
4233 | } | |
4234 | ||
695a0d0d JB |
4235 | /* |
4236 | * This function puts in dummy file extents for the area we're creating a hole | |
4237 | * for. So if we are truncating this file to a larger size we need to insert | |
4238 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4239 | * the range between oldsize and size | |
4240 | */ | |
a41ad394 | 4241 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4242 | { |
9036c102 YZ |
4243 | struct btrfs_trans_handle *trans; |
4244 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4245 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4246 | struct extent_map *em = NULL; |
2ac55d41 | 4247 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4248 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
fda2832f QW |
4249 | u64 hole_start = ALIGN(oldsize, root->sectorsize); |
4250 | u64 block_end = ALIGN(size, root->sectorsize); | |
9036c102 YZ |
4251 | u64 last_byte; |
4252 | u64 cur_offset; | |
4253 | u64 hole_size; | |
9ed74f2d | 4254 | int err = 0; |
39279cc3 | 4255 | |
a71754fc JB |
4256 | /* |
4257 | * If our size started in the middle of a page we need to zero out the | |
4258 | * rest of the page before we expand the i_size, otherwise we could | |
4259 | * expose stale data. | |
4260 | */ | |
4261 | err = btrfs_truncate_page(inode, oldsize, 0, 0); | |
4262 | if (err) | |
4263 | return err; | |
4264 | ||
9036c102 YZ |
4265 | if (size <= hole_start) |
4266 | return 0; | |
4267 | ||
9036c102 YZ |
4268 | while (1) { |
4269 | struct btrfs_ordered_extent *ordered; | |
4270 | btrfs_wait_ordered_range(inode, hole_start, | |
4271 | block_end - hole_start); | |
2ac55d41 | 4272 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 4273 | &cached_state); |
9036c102 YZ |
4274 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
4275 | if (!ordered) | |
4276 | break; | |
2ac55d41 JB |
4277 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4278 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
4279 | btrfs_put_ordered_extent(ordered); |
4280 | } | |
39279cc3 | 4281 | |
9036c102 YZ |
4282 | cur_offset = hole_start; |
4283 | while (1) { | |
4284 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4285 | block_end - cur_offset, 0); | |
79787eaa JM |
4286 | if (IS_ERR(em)) { |
4287 | err = PTR_ERR(em); | |
f2767956 | 4288 | em = NULL; |
79787eaa JM |
4289 | break; |
4290 | } | |
9036c102 | 4291 | last_byte = min(extent_map_end(em), block_end); |
fda2832f | 4292 | last_byte = ALIGN(last_byte , root->sectorsize); |
8082510e | 4293 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4294 | struct extent_map *hole_em; |
9036c102 | 4295 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4296 | |
3642320e | 4297 | trans = btrfs_start_transaction(root, 3); |
a22285a6 YZ |
4298 | if (IS_ERR(trans)) { |
4299 | err = PTR_ERR(trans); | |
9ed74f2d | 4300 | break; |
a22285a6 | 4301 | } |
8082510e | 4302 | |
5dc562c5 JB |
4303 | err = btrfs_drop_extents(trans, root, inode, |
4304 | cur_offset, | |
2671485d | 4305 | cur_offset + hole_size, 1); |
5b397377 | 4306 | if (err) { |
79787eaa | 4307 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 4308 | btrfs_end_transaction(trans, root); |
3893e33b | 4309 | break; |
5b397377 | 4310 | } |
8082510e | 4311 | |
9036c102 | 4312 | err = btrfs_insert_file_extent(trans, root, |
33345d01 | 4313 | btrfs_ino(inode), cur_offset, 0, |
9036c102 YZ |
4314 | 0, hole_size, 0, hole_size, |
4315 | 0, 0, 0); | |
5b397377 | 4316 | if (err) { |
79787eaa | 4317 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 4318 | btrfs_end_transaction(trans, root); |
3893e33b | 4319 | break; |
5b397377 | 4320 | } |
8082510e | 4321 | |
5dc562c5 JB |
4322 | btrfs_drop_extent_cache(inode, cur_offset, |
4323 | cur_offset + hole_size - 1, 0); | |
4324 | hole_em = alloc_extent_map(); | |
4325 | if (!hole_em) { | |
4326 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4327 | &BTRFS_I(inode)->runtime_flags); | |
4328 | goto next; | |
4329 | } | |
4330 | hole_em->start = cur_offset; | |
4331 | hole_em->len = hole_size; | |
4332 | hole_em->orig_start = cur_offset; | |
8082510e | 4333 | |
5dc562c5 JB |
4334 | hole_em->block_start = EXTENT_MAP_HOLE; |
4335 | hole_em->block_len = 0; | |
b4939680 | 4336 | hole_em->orig_block_len = 0; |
cc95bef6 | 4337 | hole_em->ram_bytes = hole_size; |
5dc562c5 JB |
4338 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
4339 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
4340 | hole_em->generation = trans->transid; | |
8082510e | 4341 | |
5dc562c5 JB |
4342 | while (1) { |
4343 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4344 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4345 | write_unlock(&em_tree->lock); |
4346 | if (err != -EEXIST) | |
4347 | break; | |
4348 | btrfs_drop_extent_cache(inode, cur_offset, | |
4349 | cur_offset + | |
4350 | hole_size - 1, 0); | |
4351 | } | |
4352 | free_extent_map(hole_em); | |
4353 | next: | |
3642320e | 4354 | btrfs_update_inode(trans, root, inode); |
8082510e | 4355 | btrfs_end_transaction(trans, root); |
9036c102 YZ |
4356 | } |
4357 | free_extent_map(em); | |
a22285a6 | 4358 | em = NULL; |
9036c102 | 4359 | cur_offset = last_byte; |
8082510e | 4360 | if (cur_offset >= block_end) |
9036c102 YZ |
4361 | break; |
4362 | } | |
1832a6d5 | 4363 | |
a22285a6 | 4364 | free_extent_map(em); |
2ac55d41 JB |
4365 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
4366 | GFP_NOFS); | |
9036c102 YZ |
4367 | return err; |
4368 | } | |
39279cc3 | 4369 | |
3972f260 | 4370 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4371 | { |
f4a2f4c5 MX |
4372 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4373 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4374 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4375 | loff_t newsize = attr->ia_size; |
4376 | int mask = attr->ia_valid; | |
8082510e YZ |
4377 | int ret; |
4378 | ||
3972f260 ES |
4379 | /* |
4380 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4381 | * special case where we need to update the times despite not having | |
4382 | * these flags set. For all other operations the VFS set these flags | |
4383 | * explicitly if it wants a timestamp update. | |
4384 | */ | |
4385 | if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME)))) | |
4386 | inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb); | |
4387 | ||
a41ad394 | 4388 | if (newsize > oldsize) { |
a41ad394 JB |
4389 | truncate_pagecache(inode, oldsize, newsize); |
4390 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
f4a2f4c5 | 4391 | if (ret) |
8082510e | 4392 | return ret; |
8082510e | 4393 | |
f4a2f4c5 MX |
4394 | trans = btrfs_start_transaction(root, 1); |
4395 | if (IS_ERR(trans)) | |
4396 | return PTR_ERR(trans); | |
4397 | ||
4398 | i_size_write(inode, newsize); | |
4399 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
4400 | ret = btrfs_update_inode(trans, root, inode); | |
7ad85bb7 | 4401 | btrfs_end_transaction(trans, root); |
a41ad394 | 4402 | } else { |
8082510e | 4403 | |
a41ad394 JB |
4404 | /* |
4405 | * We're truncating a file that used to have good data down to | |
4406 | * zero. Make sure it gets into the ordered flush list so that | |
4407 | * any new writes get down to disk quickly. | |
4408 | */ | |
4409 | if (newsize == 0) | |
72ac3c0d JB |
4410 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
4411 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 4412 | |
f3fe820c JB |
4413 | /* |
4414 | * 1 for the orphan item we're going to add | |
4415 | * 1 for the orphan item deletion. | |
4416 | */ | |
4417 | trans = btrfs_start_transaction(root, 2); | |
4418 | if (IS_ERR(trans)) | |
4419 | return PTR_ERR(trans); | |
4420 | ||
4421 | /* | |
4422 | * We need to do this in case we fail at _any_ point during the | |
4423 | * actual truncate. Once we do the truncate_setsize we could | |
4424 | * invalidate pages which forces any outstanding ordered io to | |
4425 | * be instantly completed which will give us extents that need | |
4426 | * to be truncated. If we fail to get an orphan inode down we | |
4427 | * could have left over extents that were never meant to live, | |
4428 | * so we need to garuntee from this point on that everything | |
4429 | * will be consistent. | |
4430 | */ | |
4431 | ret = btrfs_orphan_add(trans, inode); | |
4432 | btrfs_end_transaction(trans, root); | |
4433 | if (ret) | |
4434 | return ret; | |
4435 | ||
a41ad394 JB |
4436 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
4437 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
4438 | |
4439 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
4440 | btrfs_inode_block_unlocked_dio(inode); | |
4441 | inode_dio_wait(inode); | |
4442 | btrfs_inode_resume_unlocked_dio(inode); | |
4443 | ||
a41ad394 | 4444 | ret = btrfs_truncate(inode); |
f3fe820c JB |
4445 | if (ret && inode->i_nlink) |
4446 | btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
4447 | } |
4448 | ||
a41ad394 | 4449 | return ret; |
8082510e YZ |
4450 | } |
4451 | ||
9036c102 YZ |
4452 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
4453 | { | |
4454 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 4455 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 4456 | int err; |
39279cc3 | 4457 | |
b83cc969 LZ |
4458 | if (btrfs_root_readonly(root)) |
4459 | return -EROFS; | |
4460 | ||
9036c102 YZ |
4461 | err = inode_change_ok(inode, attr); |
4462 | if (err) | |
4463 | return err; | |
2bf5a725 | 4464 | |
5a3f23d5 | 4465 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 4466 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
4467 | if (err) |
4468 | return err; | |
39279cc3 | 4469 | } |
9036c102 | 4470 | |
1025774c CH |
4471 | if (attr->ia_valid) { |
4472 | setattr_copy(inode, attr); | |
0c4d2d95 | 4473 | inode_inc_iversion(inode); |
22c44fe6 | 4474 | err = btrfs_dirty_inode(inode); |
1025774c | 4475 | |
22c44fe6 | 4476 | if (!err && attr->ia_valid & ATTR_MODE) |
1025774c CH |
4477 | err = btrfs_acl_chmod(inode); |
4478 | } | |
33268eaf | 4479 | |
39279cc3 CM |
4480 | return err; |
4481 | } | |
61295eb8 | 4482 | |
bd555975 | 4483 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
4484 | { |
4485 | struct btrfs_trans_handle *trans; | |
4486 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 4487 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 4488 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 CM |
4489 | int ret; |
4490 | ||
1abe9b8a | 4491 | trace_btrfs_inode_evict(inode); |
4492 | ||
39279cc3 | 4493 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b | 4494 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
83eea1f1 | 4495 | btrfs_is_free_space_inode(inode))) |
bd555975 AV |
4496 | goto no_delete; |
4497 | ||
39279cc3 | 4498 | if (is_bad_inode(inode)) { |
7b128766 | 4499 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
4500 | goto no_delete; |
4501 | } | |
bd555975 | 4502 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 4503 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 4504 | |
c71bf099 | 4505 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 4506 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 4507 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
4508 | goto no_delete; |
4509 | } | |
4510 | ||
76dda93c YZ |
4511 | if (inode->i_nlink > 0) { |
4512 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
4513 | goto no_delete; | |
4514 | } | |
4515 | ||
0e8c36a9 MX |
4516 | ret = btrfs_commit_inode_delayed_inode(inode); |
4517 | if (ret) { | |
4518 | btrfs_orphan_del(NULL, inode); | |
4519 | goto no_delete; | |
4520 | } | |
4521 | ||
66d8f3dd | 4522 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
4523 | if (!rsv) { |
4524 | btrfs_orphan_del(NULL, inode); | |
4525 | goto no_delete; | |
4526 | } | |
4a338542 | 4527 | rsv->size = min_size; |
ca7e70f5 | 4528 | rsv->failfast = 1; |
726c35fa | 4529 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 4530 | |
dbe674a9 | 4531 | btrfs_i_size_write(inode, 0); |
5f39d397 | 4532 | |
4289a667 | 4533 | /* |
8407aa46 MX |
4534 | * This is a bit simpler than btrfs_truncate since we've already |
4535 | * reserved our space for our orphan item in the unlink, so we just | |
4536 | * need to reserve some slack space in case we add bytes and update | |
4537 | * inode item when doing the truncate. | |
4289a667 | 4538 | */ |
8082510e | 4539 | while (1) { |
08e007d2 MX |
4540 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
4541 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
4542 | |
4543 | /* | |
4544 | * Try and steal from the global reserve since we will | |
4545 | * likely not use this space anyway, we want to try as | |
4546 | * hard as possible to get this to work. | |
4547 | */ | |
4548 | if (ret) | |
4549 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 4550 | |
d68fc57b | 4551 | if (ret) { |
c2cf52eb SK |
4552 | btrfs_warn(root->fs_info, |
4553 | "Could not get space for a delete, will truncate on mount %d", | |
4554 | ret); | |
4289a667 JB |
4555 | btrfs_orphan_del(NULL, inode); |
4556 | btrfs_free_block_rsv(root, rsv); | |
4557 | goto no_delete; | |
d68fc57b | 4558 | } |
7b128766 | 4559 | |
0e8c36a9 | 4560 | trans = btrfs_join_transaction(root); |
4289a667 JB |
4561 | if (IS_ERR(trans)) { |
4562 | btrfs_orphan_del(NULL, inode); | |
4563 | btrfs_free_block_rsv(root, rsv); | |
4564 | goto no_delete; | |
d68fc57b | 4565 | } |
7b128766 | 4566 | |
4289a667 JB |
4567 | trans->block_rsv = rsv; |
4568 | ||
d68fc57b | 4569 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
ca7e70f5 | 4570 | if (ret != -ENOSPC) |
8082510e | 4571 | break; |
85e21bac | 4572 | |
8407aa46 | 4573 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e YZ |
4574 | btrfs_end_transaction(trans, root); |
4575 | trans = NULL; | |
b53d3f5d | 4576 | btrfs_btree_balance_dirty(root); |
8082510e | 4577 | } |
5f39d397 | 4578 | |
4289a667 JB |
4579 | btrfs_free_block_rsv(root, rsv); |
4580 | ||
4ef31a45 JB |
4581 | /* |
4582 | * Errors here aren't a big deal, it just means we leave orphan items | |
4583 | * in the tree. They will be cleaned up on the next mount. | |
4584 | */ | |
8082510e | 4585 | if (ret == 0) { |
4289a667 | 4586 | trans->block_rsv = root->orphan_block_rsv; |
4ef31a45 JB |
4587 | btrfs_orphan_del(trans, inode); |
4588 | } else { | |
4589 | btrfs_orphan_del(NULL, inode); | |
8082510e | 4590 | } |
54aa1f4d | 4591 | |
4289a667 | 4592 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
4593 | if (!(root == root->fs_info->tree_root || |
4594 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 4595 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 4596 | |
54aa1f4d | 4597 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4598 | btrfs_btree_balance_dirty(root); |
39279cc3 | 4599 | no_delete: |
89042e5a | 4600 | btrfs_remove_delayed_node(inode); |
dbd5768f | 4601 | clear_inode(inode); |
8082510e | 4602 | return; |
39279cc3 CM |
4603 | } |
4604 | ||
4605 | /* | |
4606 | * this returns the key found in the dir entry in the location pointer. | |
4607 | * If no dir entries were found, location->objectid is 0. | |
4608 | */ | |
4609 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
4610 | struct btrfs_key *location) | |
4611 | { | |
4612 | const char *name = dentry->d_name.name; | |
4613 | int namelen = dentry->d_name.len; | |
4614 | struct btrfs_dir_item *di; | |
4615 | struct btrfs_path *path; | |
4616 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 4617 | int ret = 0; |
39279cc3 CM |
4618 | |
4619 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
4620 | if (!path) |
4621 | return -ENOMEM; | |
3954401f | 4622 | |
33345d01 | 4623 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 4624 | namelen, 0); |
0d9f7f3e Y |
4625 | if (IS_ERR(di)) |
4626 | ret = PTR_ERR(di); | |
d397712b | 4627 | |
c704005d | 4628 | if (IS_ERR_OR_NULL(di)) |
3954401f | 4629 | goto out_err; |
d397712b | 4630 | |
5f39d397 | 4631 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 4632 | out: |
39279cc3 CM |
4633 | btrfs_free_path(path); |
4634 | return ret; | |
3954401f CM |
4635 | out_err: |
4636 | location->objectid = 0; | |
4637 | goto out; | |
39279cc3 CM |
4638 | } |
4639 | ||
4640 | /* | |
4641 | * when we hit a tree root in a directory, the btrfs part of the inode | |
4642 | * needs to be changed to reflect the root directory of the tree root. This | |
4643 | * is kind of like crossing a mount point. | |
4644 | */ | |
4645 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
4646 | struct inode *dir, |
4647 | struct dentry *dentry, | |
4648 | struct btrfs_key *location, | |
4649 | struct btrfs_root **sub_root) | |
39279cc3 | 4650 | { |
4df27c4d YZ |
4651 | struct btrfs_path *path; |
4652 | struct btrfs_root *new_root; | |
4653 | struct btrfs_root_ref *ref; | |
4654 | struct extent_buffer *leaf; | |
4655 | int ret; | |
4656 | int err = 0; | |
39279cc3 | 4657 | |
4df27c4d YZ |
4658 | path = btrfs_alloc_path(); |
4659 | if (!path) { | |
4660 | err = -ENOMEM; | |
4661 | goto out; | |
4662 | } | |
39279cc3 | 4663 | |
4df27c4d YZ |
4664 | err = -ENOENT; |
4665 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
4666 | BTRFS_I(dir)->root->root_key.objectid, | |
4667 | location->objectid); | |
4668 | if (ret) { | |
4669 | if (ret < 0) | |
4670 | err = ret; | |
4671 | goto out; | |
4672 | } | |
39279cc3 | 4673 | |
4df27c4d YZ |
4674 | leaf = path->nodes[0]; |
4675 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 4676 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
4677 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
4678 | goto out; | |
39279cc3 | 4679 | |
4df27c4d YZ |
4680 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
4681 | (unsigned long)(ref + 1), | |
4682 | dentry->d_name.len); | |
4683 | if (ret) | |
4684 | goto out; | |
4685 | ||
b3b4aa74 | 4686 | btrfs_release_path(path); |
4df27c4d YZ |
4687 | |
4688 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
4689 | if (IS_ERR(new_root)) { | |
4690 | err = PTR_ERR(new_root); | |
4691 | goto out; | |
4692 | } | |
4693 | ||
4df27c4d YZ |
4694 | *sub_root = new_root; |
4695 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
4696 | location->type = BTRFS_INODE_ITEM_KEY; | |
4697 | location->offset = 0; | |
4698 | err = 0; | |
4699 | out: | |
4700 | btrfs_free_path(path); | |
4701 | return err; | |
39279cc3 CM |
4702 | } |
4703 | ||
5d4f98a2 YZ |
4704 | static void inode_tree_add(struct inode *inode) |
4705 | { | |
4706 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4707 | struct btrfs_inode *entry; | |
03e860bd FNP |
4708 | struct rb_node **p; |
4709 | struct rb_node *parent; | |
33345d01 | 4710 | u64 ino = btrfs_ino(inode); |
5d4f98a2 | 4711 | |
1d3382cb | 4712 | if (inode_unhashed(inode)) |
76dda93c | 4713 | return; |
e1409cef MX |
4714 | again: |
4715 | parent = NULL; | |
5d4f98a2 | 4716 | spin_lock(&root->inode_lock); |
e1409cef | 4717 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
4718 | while (*p) { |
4719 | parent = *p; | |
4720 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
4721 | ||
33345d01 | 4722 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4723 | p = &parent->rb_left; |
33345d01 | 4724 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4725 | p = &parent->rb_right; |
5d4f98a2 YZ |
4726 | else { |
4727 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 4728 | (I_WILL_FREE | I_FREEING))); |
03e860bd FNP |
4729 | rb_erase(parent, &root->inode_tree); |
4730 | RB_CLEAR_NODE(parent); | |
4731 | spin_unlock(&root->inode_lock); | |
4732 | goto again; | |
5d4f98a2 YZ |
4733 | } |
4734 | } | |
4735 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
4736 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
4737 | spin_unlock(&root->inode_lock); | |
4738 | } | |
4739 | ||
4740 | static void inode_tree_del(struct inode *inode) | |
4741 | { | |
4742 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 4743 | int empty = 0; |
5d4f98a2 | 4744 | |
03e860bd | 4745 | spin_lock(&root->inode_lock); |
5d4f98a2 | 4746 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 4747 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 4748 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 4749 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 4750 | } |
03e860bd | 4751 | spin_unlock(&root->inode_lock); |
76dda93c | 4752 | |
0af3d00b JB |
4753 | /* |
4754 | * Free space cache has inodes in the tree root, but the tree root has a | |
4755 | * root_refs of 0, so this could end up dropping the tree root as a | |
4756 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
4757 | * make sure we don't drop it. | |
4758 | */ | |
4759 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
4760 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
4761 | synchronize_srcu(&root->fs_info->subvol_srcu); |
4762 | spin_lock(&root->inode_lock); | |
4763 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
4764 | spin_unlock(&root->inode_lock); | |
4765 | if (empty) | |
4766 | btrfs_add_dead_root(root); | |
4767 | } | |
4768 | } | |
4769 | ||
143bede5 | 4770 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
4771 | { |
4772 | struct rb_node *node; | |
4773 | struct rb_node *prev; | |
4774 | struct btrfs_inode *entry; | |
4775 | struct inode *inode; | |
4776 | u64 objectid = 0; | |
4777 | ||
4778 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4779 | ||
4780 | spin_lock(&root->inode_lock); | |
4781 | again: | |
4782 | node = root->inode_tree.rb_node; | |
4783 | prev = NULL; | |
4784 | while (node) { | |
4785 | prev = node; | |
4786 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4787 | ||
33345d01 | 4788 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 4789 | node = node->rb_left; |
33345d01 | 4790 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
4791 | node = node->rb_right; |
4792 | else | |
4793 | break; | |
4794 | } | |
4795 | if (!node) { | |
4796 | while (prev) { | |
4797 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 4798 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
4799 | node = prev; |
4800 | break; | |
4801 | } | |
4802 | prev = rb_next(prev); | |
4803 | } | |
4804 | } | |
4805 | while (node) { | |
4806 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 4807 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
4808 | inode = igrab(&entry->vfs_inode); |
4809 | if (inode) { | |
4810 | spin_unlock(&root->inode_lock); | |
4811 | if (atomic_read(&inode->i_count) > 1) | |
4812 | d_prune_aliases(inode); | |
4813 | /* | |
45321ac5 | 4814 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
4815 | * the inode cache when its usage count |
4816 | * hits zero. | |
4817 | */ | |
4818 | iput(inode); | |
4819 | cond_resched(); | |
4820 | spin_lock(&root->inode_lock); | |
4821 | goto again; | |
4822 | } | |
4823 | ||
4824 | if (cond_resched_lock(&root->inode_lock)) | |
4825 | goto again; | |
4826 | ||
4827 | node = rb_next(node); | |
4828 | } | |
4829 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
4830 | } |
4831 | ||
e02119d5 CM |
4832 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
4833 | { | |
4834 | struct btrfs_iget_args *args = p; | |
4835 | inode->i_ino = args->ino; | |
e02119d5 | 4836 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
4837 | return 0; |
4838 | } | |
4839 | ||
4840 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
4841 | { | |
4842 | struct btrfs_iget_args *args = opaque; | |
33345d01 | 4843 | return args->ino == btrfs_ino(inode) && |
d397712b | 4844 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
4845 | } |
4846 | ||
5d4f98a2 YZ |
4847 | static struct inode *btrfs_iget_locked(struct super_block *s, |
4848 | u64 objectid, | |
4849 | struct btrfs_root *root) | |
39279cc3 CM |
4850 | { |
4851 | struct inode *inode; | |
4852 | struct btrfs_iget_args args; | |
4853 | args.ino = objectid; | |
4854 | args.root = root; | |
4855 | ||
4856 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
4857 | btrfs_init_locked_inode, | |
4858 | (void *)&args); | |
4859 | return inode; | |
4860 | } | |
4861 | ||
1a54ef8c BR |
4862 | /* Get an inode object given its location and corresponding root. |
4863 | * Returns in *is_new if the inode was read from disk | |
4864 | */ | |
4865 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 4866 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
4867 | { |
4868 | struct inode *inode; | |
4869 | ||
4870 | inode = btrfs_iget_locked(s, location->objectid, root); | |
4871 | if (!inode) | |
5d4f98a2 | 4872 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
4873 | |
4874 | if (inode->i_state & I_NEW) { | |
4875 | BTRFS_I(inode)->root = root; | |
4876 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
4877 | btrfs_read_locked_inode(inode); | |
1748f843 MF |
4878 | if (!is_bad_inode(inode)) { |
4879 | inode_tree_add(inode); | |
4880 | unlock_new_inode(inode); | |
4881 | if (new) | |
4882 | *new = 1; | |
4883 | } else { | |
e0b6d65b ST |
4884 | unlock_new_inode(inode); |
4885 | iput(inode); | |
4886 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
4887 | } |
4888 | } | |
4889 | ||
1a54ef8c BR |
4890 | return inode; |
4891 | } | |
4892 | ||
4df27c4d YZ |
4893 | static struct inode *new_simple_dir(struct super_block *s, |
4894 | struct btrfs_key *key, | |
4895 | struct btrfs_root *root) | |
4896 | { | |
4897 | struct inode *inode = new_inode(s); | |
4898 | ||
4899 | if (!inode) | |
4900 | return ERR_PTR(-ENOMEM); | |
4901 | ||
4df27c4d YZ |
4902 | BTRFS_I(inode)->root = root; |
4903 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 4904 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
4905 | |
4906 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 4907 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
4908 | inode->i_fop = &simple_dir_operations; |
4909 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
4910 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
4911 | ||
4912 | return inode; | |
4913 | } | |
4914 | ||
3de4586c | 4915 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4916 | { |
d397712b | 4917 | struct inode *inode; |
4df27c4d | 4918 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4919 | struct btrfs_root *sub_root = root; |
4920 | struct btrfs_key location; | |
76dda93c | 4921 | int index; |
b4aff1f8 | 4922 | int ret = 0; |
39279cc3 CM |
4923 | |
4924 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
4925 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4926 | |
39e3c955 | 4927 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
4928 | if (ret < 0) |
4929 | return ERR_PTR(ret); | |
5f39d397 | 4930 | |
4df27c4d YZ |
4931 | if (location.objectid == 0) |
4932 | return NULL; | |
4933 | ||
4934 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4935 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4936 | return inode; |
4937 | } | |
4938 | ||
4939 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4940 | ||
76dda93c | 4941 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4942 | ret = fixup_tree_root_location(root, dir, dentry, |
4943 | &location, &sub_root); | |
4944 | if (ret < 0) { | |
4945 | if (ret != -ENOENT) | |
4946 | inode = ERR_PTR(ret); | |
4947 | else | |
4948 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4949 | } else { | |
73f73415 | 4950 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4951 | } |
76dda93c YZ |
4952 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4953 | ||
34d19bad | 4954 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
4955 | down_read(&root->fs_info->cleanup_work_sem); |
4956 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 4957 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 4958 | up_read(&root->fs_info->cleanup_work_sem); |
01cd3367 JB |
4959 | if (ret) { |
4960 | iput(inode); | |
66b4ffd1 | 4961 | inode = ERR_PTR(ret); |
01cd3367 | 4962 | } |
c71bf099 YZ |
4963 | } |
4964 | ||
3de4586c CM |
4965 | return inode; |
4966 | } | |
4967 | ||
fe15ce44 | 4968 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
4969 | { |
4970 | struct btrfs_root *root; | |
848cce0d | 4971 | struct inode *inode = dentry->d_inode; |
76dda93c | 4972 | |
848cce0d LZ |
4973 | if (!inode && !IS_ROOT(dentry)) |
4974 | inode = dentry->d_parent->d_inode; | |
76dda93c | 4975 | |
848cce0d LZ |
4976 | if (inode) { |
4977 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
4978 | if (btrfs_root_refs(&root->root_item) == 0) |
4979 | return 1; | |
848cce0d LZ |
4980 | |
4981 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
4982 | return 1; | |
efefb143 | 4983 | } |
76dda93c YZ |
4984 | return 0; |
4985 | } | |
4986 | ||
b4aff1f8 JB |
4987 | static void btrfs_dentry_release(struct dentry *dentry) |
4988 | { | |
4989 | if (dentry->d_fsdata) | |
4990 | kfree(dentry->d_fsdata); | |
4991 | } | |
4992 | ||
3de4586c | 4993 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 4994 | unsigned int flags) |
3de4586c | 4995 | { |
a66e7cc6 JB |
4996 | struct dentry *ret; |
4997 | ||
4998 | ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry); | |
a66e7cc6 | 4999 | return ret; |
39279cc3 CM |
5000 | } |
5001 | ||
16cdcec7 | 5002 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5003 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5004 | }; | |
5005 | ||
9cdda8d3 | 5006 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5007 | { |
9cdda8d3 | 5008 | struct inode *inode = file_inode(file); |
39279cc3 CM |
5009 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5010 | struct btrfs_item *item; | |
5011 | struct btrfs_dir_item *di; | |
5012 | struct btrfs_key key; | |
5f39d397 | 5013 | struct btrfs_key found_key; |
39279cc3 | 5014 | struct btrfs_path *path; |
16cdcec7 MX |
5015 | struct list_head ins_list; |
5016 | struct list_head del_list; | |
39279cc3 | 5017 | int ret; |
5f39d397 | 5018 | struct extent_buffer *leaf; |
39279cc3 | 5019 | int slot; |
39279cc3 CM |
5020 | unsigned char d_type; |
5021 | int over = 0; | |
5022 | u32 di_cur; | |
5023 | u32 di_total; | |
5024 | u32 di_len; | |
5025 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
5026 | char tmp_name[32]; |
5027 | char *name_ptr; | |
5028 | int name_len; | |
9cdda8d3 | 5029 | int is_curr = 0; /* ctx->pos points to the current index? */ |
39279cc3 CM |
5030 | |
5031 | /* FIXME, use a real flag for deciding about the key type */ | |
5032 | if (root->fs_info->tree_root == root) | |
5033 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 5034 | |
9cdda8d3 AV |
5035 | if (!dir_emit_dots(file, ctx)) |
5036 | return 0; | |
5037 | ||
49593bfa | 5038 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5039 | if (!path) |
5040 | return -ENOMEM; | |
ff5714cc | 5041 | |
026fd317 | 5042 | path->reada = 1; |
49593bfa | 5043 | |
16cdcec7 MX |
5044 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5045 | INIT_LIST_HEAD(&ins_list); | |
5046 | INIT_LIST_HEAD(&del_list); | |
5047 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
5048 | } | |
5049 | ||
39279cc3 | 5050 | btrfs_set_key_type(&key, key_type); |
9cdda8d3 | 5051 | key.offset = ctx->pos; |
33345d01 | 5052 | key.objectid = btrfs_ino(inode); |
5f39d397 | 5053 | |
39279cc3 CM |
5054 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5055 | if (ret < 0) | |
5056 | goto err; | |
49593bfa DW |
5057 | |
5058 | while (1) { | |
5f39d397 | 5059 | leaf = path->nodes[0]; |
39279cc3 | 5060 | slot = path->slots[0]; |
b9e03af0 LZ |
5061 | if (slot >= btrfs_header_nritems(leaf)) { |
5062 | ret = btrfs_next_leaf(root, path); | |
5063 | if (ret < 0) | |
5064 | goto err; | |
5065 | else if (ret > 0) | |
5066 | break; | |
5067 | continue; | |
39279cc3 | 5068 | } |
3de4586c | 5069 | |
5f39d397 CM |
5070 | item = btrfs_item_nr(leaf, slot); |
5071 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
5072 | ||
5073 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5074 | break; |
5f39d397 | 5075 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 5076 | break; |
9cdda8d3 | 5077 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5078 | goto next; |
16cdcec7 MX |
5079 | if (key_type == BTRFS_DIR_INDEX_KEY && |
5080 | btrfs_should_delete_dir_index(&del_list, | |
5081 | found_key.offset)) | |
5082 | goto next; | |
5f39d397 | 5083 | |
9cdda8d3 | 5084 | ctx->pos = found_key.offset; |
16cdcec7 | 5085 | is_curr = 1; |
49593bfa | 5086 | |
39279cc3 CM |
5087 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
5088 | di_cur = 0; | |
5f39d397 | 5089 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
5090 | |
5091 | while (di_cur < di_total) { | |
5f39d397 CM |
5092 | struct btrfs_key location; |
5093 | ||
22a94d44 JB |
5094 | if (verify_dir_item(root, leaf, di)) |
5095 | break; | |
5096 | ||
5f39d397 | 5097 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 5098 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
5099 | name_ptr = tmp_name; |
5100 | } else { | |
5101 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
5102 | if (!name_ptr) { |
5103 | ret = -ENOMEM; | |
5104 | goto err; | |
5105 | } | |
5f39d397 CM |
5106 | } |
5107 | read_extent_buffer(leaf, name_ptr, | |
5108 | (unsigned long)(di + 1), name_len); | |
5109 | ||
5110 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
5111 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 5112 | |
fede766f | 5113 | |
3de4586c | 5114 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
5115 | * skip it. |
5116 | * | |
5117 | * In contrast to old kernels, we insert the snapshot's | |
5118 | * dir item and dir index after it has been created, so | |
5119 | * we won't find a reference to our own snapshot. We | |
5120 | * still keep the following code for backward | |
5121 | * compatibility. | |
3de4586c CM |
5122 | */ |
5123 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
5124 | location.objectid == root->root_key.objectid) { | |
5125 | over = 0; | |
5126 | goto skip; | |
5127 | } | |
9cdda8d3 AV |
5128 | over = !dir_emit(ctx, name_ptr, name_len, |
5129 | location.objectid, d_type); | |
5f39d397 | 5130 | |
3de4586c | 5131 | skip: |
5f39d397 CM |
5132 | if (name_ptr != tmp_name) |
5133 | kfree(name_ptr); | |
5134 | ||
39279cc3 CM |
5135 | if (over) |
5136 | goto nopos; | |
5103e947 | 5137 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 5138 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
5139 | di_cur += di_len; |
5140 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
5141 | } | |
b9e03af0 LZ |
5142 | next: |
5143 | path->slots[0]++; | |
39279cc3 | 5144 | } |
49593bfa | 5145 | |
16cdcec7 MX |
5146 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5147 | if (is_curr) | |
9cdda8d3 AV |
5148 | ctx->pos++; |
5149 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); | |
16cdcec7 MX |
5150 | if (ret) |
5151 | goto nopos; | |
5152 | } | |
5153 | ||
49593bfa | 5154 | /* Reached end of directory/root. Bump pos past the last item. */ |
db62efbb ZB |
5155 | ctx->pos++; |
5156 | ||
5157 | /* | |
5158 | * Stop new entries from being returned after we return the last | |
5159 | * entry. | |
5160 | * | |
5161 | * New directory entries are assigned a strictly increasing | |
5162 | * offset. This means that new entries created during readdir | |
5163 | * are *guaranteed* to be seen in the future by that readdir. | |
5164 | * This has broken buggy programs which operate on names as | |
5165 | * they're returned by readdir. Until we re-use freed offsets | |
5166 | * we have this hack to stop new entries from being returned | |
5167 | * under the assumption that they'll never reach this huge | |
5168 | * offset. | |
5169 | * | |
5170 | * This is being careful not to overflow 32bit loff_t unless the | |
5171 | * last entry requires it because doing so has broken 32bit apps | |
5172 | * in the past. | |
5173 | */ | |
5174 | if (key_type == BTRFS_DIR_INDEX_KEY) { | |
5175 | if (ctx->pos >= INT_MAX) | |
5176 | ctx->pos = LLONG_MAX; | |
5177 | else | |
5178 | ctx->pos = INT_MAX; | |
5179 | } | |
39279cc3 CM |
5180 | nopos: |
5181 | ret = 0; | |
5182 | err: | |
16cdcec7 MX |
5183 | if (key_type == BTRFS_DIR_INDEX_KEY) |
5184 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 5185 | btrfs_free_path(path); |
39279cc3 CM |
5186 | return ret; |
5187 | } | |
5188 | ||
a9185b41 | 5189 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
5190 | { |
5191 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5192 | struct btrfs_trans_handle *trans; | |
5193 | int ret = 0; | |
0af3d00b | 5194 | bool nolock = false; |
39279cc3 | 5195 | |
72ac3c0d | 5196 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
5197 | return 0; |
5198 | ||
83eea1f1 | 5199 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 5200 | nolock = true; |
0af3d00b | 5201 | |
a9185b41 | 5202 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 5203 | if (nolock) |
7a7eaa40 | 5204 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 5205 | else |
7a7eaa40 | 5206 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5207 | if (IS_ERR(trans)) |
5208 | return PTR_ERR(trans); | |
a698d075 | 5209 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
5210 | } |
5211 | return ret; | |
5212 | } | |
5213 | ||
5214 | /* | |
54aa1f4d | 5215 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5216 | * inode changes. But, it is most likely to find the inode in cache. |
5217 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5218 | * to keep or drop this code. | |
5219 | */ | |
48a3b636 | 5220 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
5221 | { |
5222 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5223 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5224 | int ret; |
5225 | ||
72ac3c0d | 5226 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5227 | return 0; |
39279cc3 | 5228 | |
7a7eaa40 | 5229 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5230 | if (IS_ERR(trans)) |
5231 | return PTR_ERR(trans); | |
8929ecfa YZ |
5232 | |
5233 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
5234 | if (ret && ret == -ENOSPC) { |
5235 | /* whoops, lets try again with the full transaction */ | |
5236 | btrfs_end_transaction(trans, root); | |
5237 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
5238 | if (IS_ERR(trans)) |
5239 | return PTR_ERR(trans); | |
8929ecfa | 5240 | |
94b60442 | 5241 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 5242 | } |
39279cc3 | 5243 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
5244 | if (BTRFS_I(inode)->delayed_node) |
5245 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
5246 | |
5247 | return ret; | |
5248 | } | |
5249 | ||
5250 | /* | |
5251 | * This is a copy of file_update_time. We need this so we can return error on | |
5252 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5253 | */ | |
e41f941a JB |
5254 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
5255 | int flags) | |
22c44fe6 | 5256 | { |
2bc55652 AB |
5257 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5258 | ||
5259 | if (btrfs_root_readonly(root)) | |
5260 | return -EROFS; | |
5261 | ||
e41f941a | 5262 | if (flags & S_VERSION) |
22c44fe6 | 5263 | inode_inc_iversion(inode); |
e41f941a JB |
5264 | if (flags & S_CTIME) |
5265 | inode->i_ctime = *now; | |
5266 | if (flags & S_MTIME) | |
5267 | inode->i_mtime = *now; | |
5268 | if (flags & S_ATIME) | |
5269 | inode->i_atime = *now; | |
5270 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
5271 | } |
5272 | ||
d352ac68 CM |
5273 | /* |
5274 | * find the highest existing sequence number in a directory | |
5275 | * and then set the in-memory index_cnt variable to reflect | |
5276 | * free sequence numbers | |
5277 | */ | |
aec7477b JB |
5278 | static int btrfs_set_inode_index_count(struct inode *inode) |
5279 | { | |
5280 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5281 | struct btrfs_key key, found_key; | |
5282 | struct btrfs_path *path; | |
5283 | struct extent_buffer *leaf; | |
5284 | int ret; | |
5285 | ||
33345d01 | 5286 | key.objectid = btrfs_ino(inode); |
aec7477b JB |
5287 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); |
5288 | key.offset = (u64)-1; | |
5289 | ||
5290 | path = btrfs_alloc_path(); | |
5291 | if (!path) | |
5292 | return -ENOMEM; | |
5293 | ||
5294 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5295 | if (ret < 0) | |
5296 | goto out; | |
5297 | /* FIXME: we should be able to handle this */ | |
5298 | if (ret == 0) | |
5299 | goto out; | |
5300 | ret = 0; | |
5301 | ||
5302 | /* | |
5303 | * MAGIC NUMBER EXPLANATION: | |
5304 | * since we search a directory based on f_pos we have to start at 2 | |
5305 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
5306 | * else has to start at 2 | |
5307 | */ | |
5308 | if (path->slots[0] == 0) { | |
5309 | BTRFS_I(inode)->index_cnt = 2; | |
5310 | goto out; | |
5311 | } | |
5312 | ||
5313 | path->slots[0]--; | |
5314 | ||
5315 | leaf = path->nodes[0]; | |
5316 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5317 | ||
33345d01 | 5318 | if (found_key.objectid != btrfs_ino(inode) || |
aec7477b JB |
5319 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { |
5320 | BTRFS_I(inode)->index_cnt = 2; | |
5321 | goto out; | |
5322 | } | |
5323 | ||
5324 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
5325 | out: | |
5326 | btrfs_free_path(path); | |
5327 | return ret; | |
5328 | } | |
5329 | ||
d352ac68 CM |
5330 | /* |
5331 | * helper to find a free sequence number in a given directory. This current | |
5332 | * code is very simple, later versions will do smarter things in the btree | |
5333 | */ | |
3de4586c | 5334 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
5335 | { |
5336 | int ret = 0; | |
5337 | ||
5338 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
5339 | ret = btrfs_inode_delayed_dir_index_count(dir); |
5340 | if (ret) { | |
5341 | ret = btrfs_set_inode_index_count(dir); | |
5342 | if (ret) | |
5343 | return ret; | |
5344 | } | |
aec7477b JB |
5345 | } |
5346 | ||
00e4e6b3 | 5347 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
5348 | BTRFS_I(dir)->index_cnt++; |
5349 | ||
5350 | return ret; | |
5351 | } | |
5352 | ||
39279cc3 CM |
5353 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
5354 | struct btrfs_root *root, | |
aec7477b | 5355 | struct inode *dir, |
9c58309d | 5356 | const char *name, int name_len, |
175a4eb7 AV |
5357 | u64 ref_objectid, u64 objectid, |
5358 | umode_t mode, u64 *index) | |
39279cc3 CM |
5359 | { |
5360 | struct inode *inode; | |
5f39d397 | 5361 | struct btrfs_inode_item *inode_item; |
39279cc3 | 5362 | struct btrfs_key *location; |
5f39d397 | 5363 | struct btrfs_path *path; |
9c58309d CM |
5364 | struct btrfs_inode_ref *ref; |
5365 | struct btrfs_key key[2]; | |
5366 | u32 sizes[2]; | |
5367 | unsigned long ptr; | |
39279cc3 CM |
5368 | int ret; |
5369 | int owner; | |
5370 | ||
5f39d397 | 5371 | path = btrfs_alloc_path(); |
d8926bb3 MF |
5372 | if (!path) |
5373 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 5374 | |
39279cc3 | 5375 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
5376 | if (!inode) { |
5377 | btrfs_free_path(path); | |
39279cc3 | 5378 | return ERR_PTR(-ENOMEM); |
8fb27640 | 5379 | } |
39279cc3 | 5380 | |
581bb050 LZ |
5381 | /* |
5382 | * we have to initialize this early, so we can reclaim the inode | |
5383 | * number if we fail afterwards in this function. | |
5384 | */ | |
5385 | inode->i_ino = objectid; | |
5386 | ||
aec7477b | 5387 | if (dir) { |
1abe9b8a | 5388 | trace_btrfs_inode_request(dir); |
5389 | ||
3de4586c | 5390 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 5391 | if (ret) { |
8fb27640 | 5392 | btrfs_free_path(path); |
09771430 | 5393 | iput(inode); |
aec7477b | 5394 | return ERR_PTR(ret); |
09771430 | 5395 | } |
aec7477b JB |
5396 | } |
5397 | /* | |
5398 | * index_cnt is ignored for everything but a dir, | |
5399 | * btrfs_get_inode_index_count has an explanation for the magic | |
5400 | * number | |
5401 | */ | |
5402 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 5403 | BTRFS_I(inode)->root = root; |
e02119d5 | 5404 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 5405 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 5406 | |
5dc562c5 JB |
5407 | /* |
5408 | * We could have gotten an inode number from somebody who was fsynced | |
5409 | * and then removed in this same transaction, so let's just set full | |
5410 | * sync since it will be a full sync anyway and this will blow away the | |
5411 | * old info in the log. | |
5412 | */ | |
5413 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
5414 | ||
569254b0 | 5415 | if (S_ISDIR(mode)) |
39279cc3 CM |
5416 | owner = 0; |
5417 | else | |
5418 | owner = 1; | |
9c58309d CM |
5419 | |
5420 | key[0].objectid = objectid; | |
5421 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
5422 | key[0].offset = 0; | |
5423 | ||
f186373f MF |
5424 | /* |
5425 | * Start new inodes with an inode_ref. This is slightly more | |
5426 | * efficient for small numbers of hard links since they will | |
5427 | * be packed into one item. Extended refs will kick in if we | |
5428 | * add more hard links than can fit in the ref item. | |
5429 | */ | |
9c58309d CM |
5430 | key[1].objectid = objectid; |
5431 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
5432 | key[1].offset = ref_objectid; | |
5433 | ||
5434 | sizes[0] = sizeof(struct btrfs_inode_item); | |
5435 | sizes[1] = name_len + sizeof(*ref); | |
5436 | ||
b9473439 | 5437 | path->leave_spinning = 1; |
9c58309d CM |
5438 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
5439 | if (ret != 0) | |
5f39d397 CM |
5440 | goto fail; |
5441 | ||
ecc11fab | 5442 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 5443 | inode_set_bytes(inode, 0); |
39279cc3 | 5444 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
5445 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
5446 | struct btrfs_inode_item); | |
293f7e07 LZ |
5447 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
5448 | sizeof(*inode_item)); | |
e02119d5 | 5449 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
5450 | |
5451 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
5452 | struct btrfs_inode_ref); | |
5453 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 5454 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
5455 | ptr = (unsigned long)(ref + 1); |
5456 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
5457 | ||
5f39d397 CM |
5458 | btrfs_mark_buffer_dirty(path->nodes[0]); |
5459 | btrfs_free_path(path); | |
5460 | ||
39279cc3 CM |
5461 | location = &BTRFS_I(inode)->location; |
5462 | location->objectid = objectid; | |
39279cc3 CM |
5463 | location->offset = 0; |
5464 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
5465 | ||
6cbff00f CH |
5466 | btrfs_inherit_iflags(inode, dir); |
5467 | ||
569254b0 | 5468 | if (S_ISREG(mode)) { |
94272164 CM |
5469 | if (btrfs_test_opt(root, NODATASUM)) |
5470 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
213490b3 | 5471 | if (btrfs_test_opt(root, NODATACOW)) |
f2bdf9a8 JB |
5472 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
5473 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
5474 | } |
5475 | ||
39279cc3 | 5476 | insert_inode_hash(inode); |
5d4f98a2 | 5477 | inode_tree_add(inode); |
1abe9b8a | 5478 | |
5479 | trace_btrfs_inode_new(inode); | |
1973f0fa | 5480 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 5481 | |
8ea05e3a AB |
5482 | btrfs_update_root_times(trans, root); |
5483 | ||
39279cc3 | 5484 | return inode; |
5f39d397 | 5485 | fail: |
aec7477b JB |
5486 | if (dir) |
5487 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 5488 | btrfs_free_path(path); |
09771430 | 5489 | iput(inode); |
5f39d397 | 5490 | return ERR_PTR(ret); |
39279cc3 CM |
5491 | } |
5492 | ||
5493 | static inline u8 btrfs_inode_type(struct inode *inode) | |
5494 | { | |
5495 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
5496 | } | |
5497 | ||
d352ac68 CM |
5498 | /* |
5499 | * utility function to add 'inode' into 'parent_inode' with | |
5500 | * a give name and a given sequence number. | |
5501 | * if 'add_backref' is true, also insert a backref from the | |
5502 | * inode to the parent directory. | |
5503 | */ | |
e02119d5 CM |
5504 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
5505 | struct inode *parent_inode, struct inode *inode, | |
5506 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 5507 | { |
4df27c4d | 5508 | int ret = 0; |
39279cc3 | 5509 | struct btrfs_key key; |
e02119d5 | 5510 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
5511 | u64 ino = btrfs_ino(inode); |
5512 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 5513 | |
33345d01 | 5514 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5515 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
5516 | } else { | |
33345d01 | 5517 | key.objectid = ino; |
4df27c4d YZ |
5518 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
5519 | key.offset = 0; | |
5520 | } | |
5521 | ||
33345d01 | 5522 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5523 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
5524 | key.objectid, root->root_key.objectid, | |
33345d01 | 5525 | parent_ino, index, name, name_len); |
4df27c4d | 5526 | } else if (add_backref) { |
33345d01 LZ |
5527 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
5528 | parent_ino, index); | |
4df27c4d | 5529 | } |
39279cc3 | 5530 | |
79787eaa JM |
5531 | /* Nothing to clean up yet */ |
5532 | if (ret) | |
5533 | return ret; | |
4df27c4d | 5534 | |
79787eaa JM |
5535 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
5536 | parent_inode, &key, | |
5537 | btrfs_inode_type(inode), index); | |
9c52057c | 5538 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
5539 | goto fail_dir_item; |
5540 | else if (ret) { | |
5541 | btrfs_abort_transaction(trans, root, ret); | |
5542 | return ret; | |
39279cc3 | 5543 | } |
79787eaa JM |
5544 | |
5545 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
5546 | name_len * 2); | |
0c4d2d95 | 5547 | inode_inc_iversion(parent_inode); |
79787eaa JM |
5548 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
5549 | ret = btrfs_update_inode(trans, root, parent_inode); | |
5550 | if (ret) | |
5551 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 5552 | return ret; |
fe66a05a CM |
5553 | |
5554 | fail_dir_item: | |
5555 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
5556 | u64 local_index; | |
5557 | int err; | |
5558 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
5559 | key.objectid, root->root_key.objectid, | |
5560 | parent_ino, &local_index, name, name_len); | |
5561 | ||
5562 | } else if (add_backref) { | |
5563 | u64 local_index; | |
5564 | int err; | |
5565 | ||
5566 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
5567 | ino, parent_ino, &local_index); | |
5568 | } | |
5569 | return ret; | |
39279cc3 CM |
5570 | } |
5571 | ||
5572 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
5573 | struct inode *dir, struct dentry *dentry, |
5574 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 5575 | { |
a1b075d2 JB |
5576 | int err = btrfs_add_link(trans, dir, inode, |
5577 | dentry->d_name.name, dentry->d_name.len, | |
5578 | backref, index); | |
39279cc3 CM |
5579 | if (err > 0) |
5580 | err = -EEXIST; | |
5581 | return err; | |
5582 | } | |
5583 | ||
618e21d5 | 5584 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 5585 | umode_t mode, dev_t rdev) |
618e21d5 JB |
5586 | { |
5587 | struct btrfs_trans_handle *trans; | |
5588 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 5589 | struct inode *inode = NULL; |
618e21d5 JB |
5590 | int err; |
5591 | int drop_inode = 0; | |
5592 | u64 objectid; | |
00e4e6b3 | 5593 | u64 index = 0; |
618e21d5 JB |
5594 | |
5595 | if (!new_valid_dev(rdev)) | |
5596 | return -EINVAL; | |
5597 | ||
9ed74f2d JB |
5598 | /* |
5599 | * 2 for inode item and ref | |
5600 | * 2 for dir items | |
5601 | * 1 for xattr if selinux is on | |
5602 | */ | |
a22285a6 YZ |
5603 | trans = btrfs_start_transaction(root, 5); |
5604 | if (IS_ERR(trans)) | |
5605 | return PTR_ERR(trans); | |
1832a6d5 | 5606 | |
581bb050 LZ |
5607 | err = btrfs_find_free_ino(root, &objectid); |
5608 | if (err) | |
5609 | goto out_unlock; | |
5610 | ||
aec7477b | 5611 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5612 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5613 | mode, &index); |
7cf96da3 TI |
5614 | if (IS_ERR(inode)) { |
5615 | err = PTR_ERR(inode); | |
618e21d5 | 5616 | goto out_unlock; |
7cf96da3 | 5617 | } |
618e21d5 | 5618 | |
2a7dba39 | 5619 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5620 | if (err) { |
5621 | drop_inode = 1; | |
5622 | goto out_unlock; | |
5623 | } | |
5624 | ||
ad19db71 CS |
5625 | /* |
5626 | * If the active LSM wants to access the inode during | |
5627 | * d_instantiate it needs these. Smack checks to see | |
5628 | * if the filesystem supports xattrs by looking at the | |
5629 | * ops vector. | |
5630 | */ | |
5631 | ||
5632 | inode->i_op = &btrfs_special_inode_operations; | |
a1b075d2 | 5633 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
5634 | if (err) |
5635 | drop_inode = 1; | |
5636 | else { | |
618e21d5 | 5637 | init_special_inode(inode, inode->i_mode, rdev); |
1b4ab1bb | 5638 | btrfs_update_inode(trans, root, inode); |
08c422c2 | 5639 | d_instantiate(dentry, inode); |
618e21d5 | 5640 | } |
618e21d5 | 5641 | out_unlock: |
7ad85bb7 | 5642 | btrfs_end_transaction(trans, root); |
b53d3f5d | 5643 | btrfs_btree_balance_dirty(root); |
618e21d5 JB |
5644 | if (drop_inode) { |
5645 | inode_dec_link_count(inode); | |
5646 | iput(inode); | |
5647 | } | |
618e21d5 JB |
5648 | return err; |
5649 | } | |
5650 | ||
39279cc3 | 5651 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 5652 | umode_t mode, bool excl) |
39279cc3 CM |
5653 | { |
5654 | struct btrfs_trans_handle *trans; | |
5655 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 5656 | struct inode *inode = NULL; |
43baa579 | 5657 | int drop_inode_on_err = 0; |
a22285a6 | 5658 | int err; |
39279cc3 | 5659 | u64 objectid; |
00e4e6b3 | 5660 | u64 index = 0; |
39279cc3 | 5661 | |
9ed74f2d JB |
5662 | /* |
5663 | * 2 for inode item and ref | |
5664 | * 2 for dir items | |
5665 | * 1 for xattr if selinux is on | |
5666 | */ | |
a22285a6 YZ |
5667 | trans = btrfs_start_transaction(root, 5); |
5668 | if (IS_ERR(trans)) | |
5669 | return PTR_ERR(trans); | |
9ed74f2d | 5670 | |
581bb050 LZ |
5671 | err = btrfs_find_free_ino(root, &objectid); |
5672 | if (err) | |
5673 | goto out_unlock; | |
5674 | ||
aec7477b | 5675 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5676 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5677 | mode, &index); |
7cf96da3 TI |
5678 | if (IS_ERR(inode)) { |
5679 | err = PTR_ERR(inode); | |
39279cc3 | 5680 | goto out_unlock; |
7cf96da3 | 5681 | } |
43baa579 | 5682 | drop_inode_on_err = 1; |
39279cc3 | 5683 | |
2a7dba39 | 5684 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
43baa579 | 5685 | if (err) |
33268eaf | 5686 | goto out_unlock; |
33268eaf | 5687 | |
9185aa58 FB |
5688 | err = btrfs_update_inode(trans, root, inode); |
5689 | if (err) | |
5690 | goto out_unlock; | |
5691 | ||
ad19db71 CS |
5692 | /* |
5693 | * If the active LSM wants to access the inode during | |
5694 | * d_instantiate it needs these. Smack checks to see | |
5695 | * if the filesystem supports xattrs by looking at the | |
5696 | * ops vector. | |
5697 | */ | |
5698 | inode->i_fop = &btrfs_file_operations; | |
5699 | inode->i_op = &btrfs_file_inode_operations; | |
5700 | ||
a1b075d2 | 5701 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 5702 | if (err) |
43baa579 FB |
5703 | goto out_unlock; |
5704 | ||
5705 | inode->i_mapping->a_ops = &btrfs_aops; | |
5706 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
5707 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
5708 | d_instantiate(dentry, inode); | |
5709 | ||
39279cc3 | 5710 | out_unlock: |
7ad85bb7 | 5711 | btrfs_end_transaction(trans, root); |
43baa579 | 5712 | if (err && drop_inode_on_err) { |
39279cc3 CM |
5713 | inode_dec_link_count(inode); |
5714 | iput(inode); | |
5715 | } | |
b53d3f5d | 5716 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5717 | return err; |
5718 | } | |
5719 | ||
5720 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
5721 | struct dentry *dentry) | |
5722 | { | |
5723 | struct btrfs_trans_handle *trans; | |
5724 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5725 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 5726 | u64 index; |
39279cc3 CM |
5727 | int err; |
5728 | int drop_inode = 0; | |
5729 | ||
4a8be425 TH |
5730 | /* do not allow sys_link's with other subvols of the same device */ |
5731 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 5732 | return -EXDEV; |
4a8be425 | 5733 | |
f186373f | 5734 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 5735 | return -EMLINK; |
4a8be425 | 5736 | |
3de4586c | 5737 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
5738 | if (err) |
5739 | goto fail; | |
5740 | ||
a22285a6 | 5741 | /* |
7e6b6465 | 5742 | * 2 items for inode and inode ref |
a22285a6 | 5743 | * 2 items for dir items |
7e6b6465 | 5744 | * 1 item for parent inode |
a22285a6 | 5745 | */ |
7e6b6465 | 5746 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
5747 | if (IS_ERR(trans)) { |
5748 | err = PTR_ERR(trans); | |
5749 | goto fail; | |
5750 | } | |
5f39d397 | 5751 | |
3153495d | 5752 | btrfs_inc_nlink(inode); |
0c4d2d95 | 5753 | inode_inc_iversion(inode); |
3153495d | 5754 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 5755 | ihold(inode); |
e9976151 | 5756 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 5757 | |
a1b075d2 | 5758 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 5759 | |
a5719521 | 5760 | if (err) { |
54aa1f4d | 5761 | drop_inode = 1; |
a5719521 | 5762 | } else { |
10d9f309 | 5763 | struct dentry *parent = dentry->d_parent; |
a5719521 | 5764 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
5765 | if (err) |
5766 | goto fail; | |
08c422c2 | 5767 | d_instantiate(dentry, inode); |
6a912213 | 5768 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 5769 | } |
39279cc3 | 5770 | |
7ad85bb7 | 5771 | btrfs_end_transaction(trans, root); |
1832a6d5 | 5772 | fail: |
39279cc3 CM |
5773 | if (drop_inode) { |
5774 | inode_dec_link_count(inode); | |
5775 | iput(inode); | |
5776 | } | |
b53d3f5d | 5777 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5778 | return err; |
5779 | } | |
5780 | ||
18bb1db3 | 5781 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 5782 | { |
b9d86667 | 5783 | struct inode *inode = NULL; |
39279cc3 CM |
5784 | struct btrfs_trans_handle *trans; |
5785 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5786 | int err = 0; | |
5787 | int drop_on_err = 0; | |
b9d86667 | 5788 | u64 objectid = 0; |
00e4e6b3 | 5789 | u64 index = 0; |
39279cc3 | 5790 | |
9ed74f2d JB |
5791 | /* |
5792 | * 2 items for inode and ref | |
5793 | * 2 items for dir items | |
5794 | * 1 for xattr if selinux is on | |
5795 | */ | |
a22285a6 YZ |
5796 | trans = btrfs_start_transaction(root, 5); |
5797 | if (IS_ERR(trans)) | |
5798 | return PTR_ERR(trans); | |
39279cc3 | 5799 | |
581bb050 LZ |
5800 | err = btrfs_find_free_ino(root, &objectid); |
5801 | if (err) | |
5802 | goto out_fail; | |
5803 | ||
aec7477b | 5804 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5805 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5806 | S_IFDIR | mode, &index); |
39279cc3 CM |
5807 | if (IS_ERR(inode)) { |
5808 | err = PTR_ERR(inode); | |
5809 | goto out_fail; | |
5810 | } | |
5f39d397 | 5811 | |
39279cc3 | 5812 | drop_on_err = 1; |
33268eaf | 5813 | |
2a7dba39 | 5814 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5815 | if (err) |
5816 | goto out_fail; | |
5817 | ||
39279cc3 CM |
5818 | inode->i_op = &btrfs_dir_inode_operations; |
5819 | inode->i_fop = &btrfs_dir_file_operations; | |
39279cc3 | 5820 | |
dbe674a9 | 5821 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
5822 | err = btrfs_update_inode(trans, root, inode); |
5823 | if (err) | |
5824 | goto out_fail; | |
5f39d397 | 5825 | |
a1b075d2 JB |
5826 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
5827 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
5828 | if (err) |
5829 | goto out_fail; | |
5f39d397 | 5830 | |
39279cc3 CM |
5831 | d_instantiate(dentry, inode); |
5832 | drop_on_err = 0; | |
39279cc3 CM |
5833 | |
5834 | out_fail: | |
7ad85bb7 | 5835 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
5836 | if (drop_on_err) |
5837 | iput(inode); | |
b53d3f5d | 5838 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5839 | return err; |
5840 | } | |
5841 | ||
d352ac68 CM |
5842 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
5843 | * and an extent that you want to insert, deal with overlap and insert | |
5844 | * the new extent into the tree. | |
5845 | */ | |
3b951516 CM |
5846 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
5847 | struct extent_map *existing, | |
e6dcd2dc CM |
5848 | struct extent_map *em, |
5849 | u64 map_start, u64 map_len) | |
3b951516 CM |
5850 | { |
5851 | u64 start_diff; | |
3b951516 | 5852 | |
e6dcd2dc CM |
5853 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
5854 | start_diff = map_start - em->start; | |
5855 | em->start = map_start; | |
5856 | em->len = map_len; | |
c8b97818 CM |
5857 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
5858 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 5859 | em->block_start += start_diff; |
c8b97818 CM |
5860 | em->block_len -= start_diff; |
5861 | } | |
09a2a8f9 | 5862 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
5863 | } |
5864 | ||
c8b97818 CM |
5865 | static noinline int uncompress_inline(struct btrfs_path *path, |
5866 | struct inode *inode, struct page *page, | |
5867 | size_t pg_offset, u64 extent_offset, | |
5868 | struct btrfs_file_extent_item *item) | |
5869 | { | |
5870 | int ret; | |
5871 | struct extent_buffer *leaf = path->nodes[0]; | |
5872 | char *tmp; | |
5873 | size_t max_size; | |
5874 | unsigned long inline_size; | |
5875 | unsigned long ptr; | |
261507a0 | 5876 | int compress_type; |
c8b97818 CM |
5877 | |
5878 | WARN_ON(pg_offset != 0); | |
261507a0 | 5879 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
5880 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
5881 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
5882 | btrfs_item_nr(leaf, path->slots[0])); | |
5883 | tmp = kmalloc(inline_size, GFP_NOFS); | |
8d413713 TI |
5884 | if (!tmp) |
5885 | return -ENOMEM; | |
c8b97818 CM |
5886 | ptr = btrfs_file_extent_inline_start(item); |
5887 | ||
5888 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
5889 | ||
5b050f04 | 5890 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
5891 | ret = btrfs_decompress(compress_type, tmp, page, |
5892 | extent_offset, inline_size, max_size); | |
c8b97818 | 5893 | if (ret) { |
7ac687d9 | 5894 | char *kaddr = kmap_atomic(page); |
c8b97818 CM |
5895 | unsigned long copy_size = min_t(u64, |
5896 | PAGE_CACHE_SIZE - pg_offset, | |
5897 | max_size - extent_offset); | |
5898 | memset(kaddr + pg_offset, 0, copy_size); | |
7ac687d9 | 5899 | kunmap_atomic(kaddr); |
c8b97818 CM |
5900 | } |
5901 | kfree(tmp); | |
5902 | return 0; | |
5903 | } | |
5904 | ||
d352ac68 CM |
5905 | /* |
5906 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
5907 | * the ugly parts come from merging extents from the disk with the in-ram |
5908 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
5909 | * where the in-ram extents might be locked pending data=ordered completion. |
5910 | * | |
5911 | * This also copies inline extents directly into the page. | |
5912 | */ | |
d397712b | 5913 | |
a52d9a80 | 5914 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 5915 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
5916 | int create) |
5917 | { | |
5918 | int ret; | |
5919 | int err = 0; | |
db94535d | 5920 | u64 bytenr; |
a52d9a80 CM |
5921 | u64 extent_start = 0; |
5922 | u64 extent_end = 0; | |
33345d01 | 5923 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 5924 | u32 found_type; |
f421950f | 5925 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
5926 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5927 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
5928 | struct extent_buffer *leaf; |
5929 | struct btrfs_key found_key; | |
a52d9a80 CM |
5930 | struct extent_map *em = NULL; |
5931 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 5932 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 5933 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 5934 | int compress_type; |
a52d9a80 | 5935 | |
a52d9a80 | 5936 | again: |
890871be | 5937 | read_lock(&em_tree->lock); |
d1310b2e | 5938 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
5939 | if (em) |
5940 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 5941 | read_unlock(&em_tree->lock); |
d1310b2e | 5942 | |
a52d9a80 | 5943 | if (em) { |
e1c4b745 CM |
5944 | if (em->start > start || em->start + em->len <= start) |
5945 | free_extent_map(em); | |
5946 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
5947 | free_extent_map(em); |
5948 | else | |
5949 | goto out; | |
a52d9a80 | 5950 | } |
172ddd60 | 5951 | em = alloc_extent_map(); |
a52d9a80 | 5952 | if (!em) { |
d1310b2e CM |
5953 | err = -ENOMEM; |
5954 | goto out; | |
a52d9a80 | 5955 | } |
e6dcd2dc | 5956 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 5957 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 5958 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 5959 | em->len = (u64)-1; |
c8b97818 | 5960 | em->block_len = (u64)-1; |
f421950f CM |
5961 | |
5962 | if (!path) { | |
5963 | path = btrfs_alloc_path(); | |
026fd317 JB |
5964 | if (!path) { |
5965 | err = -ENOMEM; | |
5966 | goto out; | |
5967 | } | |
5968 | /* | |
5969 | * Chances are we'll be called again, so go ahead and do | |
5970 | * readahead | |
5971 | */ | |
5972 | path->reada = 1; | |
f421950f CM |
5973 | } |
5974 | ||
179e29e4 CM |
5975 | ret = btrfs_lookup_file_extent(trans, root, path, |
5976 | objectid, start, trans != NULL); | |
a52d9a80 CM |
5977 | if (ret < 0) { |
5978 | err = ret; | |
5979 | goto out; | |
5980 | } | |
5981 | ||
5982 | if (ret != 0) { | |
5983 | if (path->slots[0] == 0) | |
5984 | goto not_found; | |
5985 | path->slots[0]--; | |
5986 | } | |
5987 | ||
5f39d397 CM |
5988 | leaf = path->nodes[0]; |
5989 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 5990 | struct btrfs_file_extent_item); |
a52d9a80 | 5991 | /* are we inside the extent that was found? */ |
5f39d397 CM |
5992 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5993 | found_type = btrfs_key_type(&found_key); | |
5994 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
5995 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5996 | goto not_found; | |
5997 | } | |
5998 | ||
5f39d397 CM |
5999 | found_type = btrfs_file_extent_type(leaf, item); |
6000 | extent_start = found_key.offset; | |
261507a0 | 6001 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
6002 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6003 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6004 | extent_end = extent_start + |
db94535d | 6005 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
6006 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
6007 | size_t size; | |
6008 | size = btrfs_file_extent_inline_len(leaf, item); | |
fda2832f | 6009 | extent_end = ALIGN(extent_start + size, root->sectorsize); |
9036c102 YZ |
6010 | } |
6011 | ||
6012 | if (start >= extent_end) { | |
6013 | path->slots[0]++; | |
6014 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6015 | ret = btrfs_next_leaf(root, path); | |
6016 | if (ret < 0) { | |
6017 | err = ret; | |
6018 | goto out; | |
a52d9a80 | 6019 | } |
9036c102 YZ |
6020 | if (ret > 0) |
6021 | goto not_found; | |
6022 | leaf = path->nodes[0]; | |
a52d9a80 | 6023 | } |
9036c102 YZ |
6024 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6025 | if (found_key.objectid != objectid || | |
6026 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6027 | goto not_found; | |
6028 | if (start + len <= found_key.offset) | |
6029 | goto not_found; | |
6030 | em->start = start; | |
70c8a91c | 6031 | em->orig_start = start; |
9036c102 YZ |
6032 | em->len = found_key.offset - start; |
6033 | goto not_found_em; | |
6034 | } | |
6035 | ||
cc95bef6 | 6036 | em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item); |
d899e052 YZ |
6037 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6038 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
6039 | em->start = extent_start; |
6040 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
6041 | em->orig_start = extent_start - |
6042 | btrfs_file_extent_offset(leaf, item); | |
b4939680 JB |
6043 | em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, |
6044 | item); | |
db94535d CM |
6045 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
6046 | if (bytenr == 0) { | |
5f39d397 | 6047 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
6048 | goto insert; |
6049 | } | |
261507a0 | 6050 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 6051 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 6052 | em->compress_type = compress_type; |
c8b97818 | 6053 | em->block_start = bytenr; |
b4939680 | 6054 | em->block_len = em->orig_block_len; |
c8b97818 CM |
6055 | } else { |
6056 | bytenr += btrfs_file_extent_offset(leaf, item); | |
6057 | em->block_start = bytenr; | |
6058 | em->block_len = em->len; | |
d899e052 YZ |
6059 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
6060 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 6061 | } |
a52d9a80 CM |
6062 | goto insert; |
6063 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 6064 | unsigned long ptr; |
a52d9a80 | 6065 | char *map; |
3326d1b0 CM |
6066 | size_t size; |
6067 | size_t extent_offset; | |
6068 | size_t copy_size; | |
a52d9a80 | 6069 | |
689f9346 | 6070 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 6071 | if (!page || create) { |
689f9346 | 6072 | em->start = extent_start; |
9036c102 | 6073 | em->len = extent_end - extent_start; |
689f9346 Y |
6074 | goto out; |
6075 | } | |
5f39d397 | 6076 | |
9036c102 YZ |
6077 | size = btrfs_file_extent_inline_len(leaf, item); |
6078 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 6079 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 6080 | size - extent_offset); |
3326d1b0 | 6081 | em->start = extent_start + extent_offset; |
fda2832f | 6082 | em->len = ALIGN(copy_size, root->sectorsize); |
b4939680 | 6083 | em->orig_block_len = em->len; |
70c8a91c | 6084 | em->orig_start = em->start; |
261507a0 | 6085 | if (compress_type) { |
c8b97818 | 6086 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
6087 | em->compress_type = compress_type; |
6088 | } | |
689f9346 | 6089 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 6090 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
6091 | if (btrfs_file_extent_compression(leaf, item) != |
6092 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
6093 | ret = uncompress_inline(path, inode, page, |
6094 | pg_offset, | |
6095 | extent_offset, item); | |
79787eaa | 6096 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
6097 | } else { |
6098 | map = kmap(page); | |
6099 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6100 | copy_size); | |
93c82d57 CM |
6101 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
6102 | memset(map + pg_offset + copy_size, 0, | |
6103 | PAGE_CACHE_SIZE - pg_offset - | |
6104 | copy_size); | |
6105 | } | |
c8b97818 CM |
6106 | kunmap(page); |
6107 | } | |
179e29e4 CM |
6108 | flush_dcache_page(page); |
6109 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 6110 | BUG(); |
179e29e4 CM |
6111 | if (!trans) { |
6112 | kunmap(page); | |
6113 | free_extent_map(em); | |
6114 | em = NULL; | |
ff5714cc | 6115 | |
b3b4aa74 | 6116 | btrfs_release_path(path); |
7a7eaa40 | 6117 | trans = btrfs_join_transaction(root); |
ff5714cc | 6118 | |
3612b495 TI |
6119 | if (IS_ERR(trans)) |
6120 | return ERR_CAST(trans); | |
179e29e4 CM |
6121 | goto again; |
6122 | } | |
c8b97818 | 6123 | map = kmap(page); |
70dec807 | 6124 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 6125 | copy_size); |
c8b97818 | 6126 | kunmap(page); |
179e29e4 | 6127 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 6128 | } |
d1310b2e | 6129 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6130 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
6131 | goto insert; |
6132 | } else { | |
31b1a2bd | 6133 | WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
6134 | } |
6135 | not_found: | |
6136 | em->start = start; | |
70c8a91c | 6137 | em->orig_start = start; |
d1310b2e | 6138 | em->len = len; |
a52d9a80 | 6139 | not_found_em: |
5f39d397 | 6140 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 6141 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 6142 | insert: |
b3b4aa74 | 6143 | btrfs_release_path(path); |
d1310b2e | 6144 | if (em->start > start || extent_map_end(em) <= start) { |
c2cf52eb SK |
6145 | btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]", |
6146 | (unsigned long long)em->start, | |
6147 | (unsigned long long)em->len, | |
6148 | (unsigned long long)start, | |
6149 | (unsigned long long)len); | |
a52d9a80 CM |
6150 | err = -EIO; |
6151 | goto out; | |
6152 | } | |
d1310b2e CM |
6153 | |
6154 | err = 0; | |
890871be | 6155 | write_lock(&em_tree->lock); |
09a2a8f9 | 6156 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6157 | /* it is possible that someone inserted the extent into the tree |
6158 | * while we had the lock dropped. It is also possible that | |
6159 | * an overlapping map exists in the tree | |
6160 | */ | |
a52d9a80 | 6161 | if (ret == -EEXIST) { |
3b951516 | 6162 | struct extent_map *existing; |
e6dcd2dc CM |
6163 | |
6164 | ret = 0; | |
6165 | ||
3b951516 | 6166 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
6167 | if (existing && (existing->start > start || |
6168 | existing->start + existing->len <= start)) { | |
6169 | free_extent_map(existing); | |
6170 | existing = NULL; | |
6171 | } | |
3b951516 CM |
6172 | if (!existing) { |
6173 | existing = lookup_extent_mapping(em_tree, em->start, | |
6174 | em->len); | |
6175 | if (existing) { | |
6176 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
6177 | em, start, |
6178 | root->sectorsize); | |
3b951516 CM |
6179 | free_extent_map(existing); |
6180 | if (err) { | |
6181 | free_extent_map(em); | |
6182 | em = NULL; | |
6183 | } | |
6184 | } else { | |
6185 | err = -EIO; | |
3b951516 CM |
6186 | free_extent_map(em); |
6187 | em = NULL; | |
6188 | } | |
6189 | } else { | |
6190 | free_extent_map(em); | |
6191 | em = existing; | |
e6dcd2dc | 6192 | err = 0; |
a52d9a80 | 6193 | } |
a52d9a80 | 6194 | } |
890871be | 6195 | write_unlock(&em_tree->lock); |
a52d9a80 | 6196 | out: |
1abe9b8a | 6197 | |
f0bd95ea TI |
6198 | if (em) |
6199 | trace_btrfs_get_extent(root, em); | |
1abe9b8a | 6200 | |
f421950f CM |
6201 | if (path) |
6202 | btrfs_free_path(path); | |
a52d9a80 CM |
6203 | if (trans) { |
6204 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 6205 | if (!err) |
a52d9a80 CM |
6206 | err = ret; |
6207 | } | |
a52d9a80 CM |
6208 | if (err) { |
6209 | free_extent_map(em); | |
a52d9a80 CM |
6210 | return ERR_PTR(err); |
6211 | } | |
79787eaa | 6212 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
6213 | return em; |
6214 | } | |
6215 | ||
ec29ed5b CM |
6216 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
6217 | size_t pg_offset, u64 start, u64 len, | |
6218 | int create) | |
6219 | { | |
6220 | struct extent_map *em; | |
6221 | struct extent_map *hole_em = NULL; | |
6222 | u64 range_start = start; | |
6223 | u64 end; | |
6224 | u64 found; | |
6225 | u64 found_end; | |
6226 | int err = 0; | |
6227 | ||
6228 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
6229 | if (IS_ERR(em)) | |
6230 | return em; | |
6231 | if (em) { | |
6232 | /* | |
f9e4fb53 LB |
6233 | * if our em maps to |
6234 | * - a hole or | |
6235 | * - a pre-alloc extent, | |
6236 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 6237 | */ |
f9e4fb53 LB |
6238 | if (em->block_start != EXTENT_MAP_HOLE && |
6239 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
6240 | return em; |
6241 | else | |
6242 | hole_em = em; | |
6243 | } | |
6244 | ||
6245 | /* check to see if we've wrapped (len == -1 or similar) */ | |
6246 | end = start + len; | |
6247 | if (end < start) | |
6248 | end = (u64)-1; | |
6249 | else | |
6250 | end -= 1; | |
6251 | ||
6252 | em = NULL; | |
6253 | ||
6254 | /* ok, we didn't find anything, lets look for delalloc */ | |
6255 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
6256 | end, len, EXTENT_DELALLOC, 1); | |
6257 | found_end = range_start + found; | |
6258 | if (found_end < range_start) | |
6259 | found_end = (u64)-1; | |
6260 | ||
6261 | /* | |
6262 | * we didn't find anything useful, return | |
6263 | * the original results from get_extent() | |
6264 | */ | |
6265 | if (range_start > end || found_end <= start) { | |
6266 | em = hole_em; | |
6267 | hole_em = NULL; | |
6268 | goto out; | |
6269 | } | |
6270 | ||
6271 | /* adjust the range_start to make sure it doesn't | |
6272 | * go backwards from the start they passed in | |
6273 | */ | |
6274 | range_start = max(start,range_start); | |
6275 | found = found_end - range_start; | |
6276 | ||
6277 | if (found > 0) { | |
6278 | u64 hole_start = start; | |
6279 | u64 hole_len = len; | |
6280 | ||
172ddd60 | 6281 | em = alloc_extent_map(); |
ec29ed5b CM |
6282 | if (!em) { |
6283 | err = -ENOMEM; | |
6284 | goto out; | |
6285 | } | |
6286 | /* | |
6287 | * when btrfs_get_extent can't find anything it | |
6288 | * returns one huge hole | |
6289 | * | |
6290 | * make sure what it found really fits our range, and | |
6291 | * adjust to make sure it is based on the start from | |
6292 | * the caller | |
6293 | */ | |
6294 | if (hole_em) { | |
6295 | u64 calc_end = extent_map_end(hole_em); | |
6296 | ||
6297 | if (calc_end <= start || (hole_em->start > end)) { | |
6298 | free_extent_map(hole_em); | |
6299 | hole_em = NULL; | |
6300 | } else { | |
6301 | hole_start = max(hole_em->start, start); | |
6302 | hole_len = calc_end - hole_start; | |
6303 | } | |
6304 | } | |
6305 | em->bdev = NULL; | |
6306 | if (hole_em && range_start > hole_start) { | |
6307 | /* our hole starts before our delalloc, so we | |
6308 | * have to return just the parts of the hole | |
6309 | * that go until the delalloc starts | |
6310 | */ | |
6311 | em->len = min(hole_len, | |
6312 | range_start - hole_start); | |
6313 | em->start = hole_start; | |
6314 | em->orig_start = hole_start; | |
6315 | /* | |
6316 | * don't adjust block start at all, | |
6317 | * it is fixed at EXTENT_MAP_HOLE | |
6318 | */ | |
6319 | em->block_start = hole_em->block_start; | |
6320 | em->block_len = hole_len; | |
f9e4fb53 LB |
6321 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
6322 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
6323 | } else { |
6324 | em->start = range_start; | |
6325 | em->len = found; | |
6326 | em->orig_start = range_start; | |
6327 | em->block_start = EXTENT_MAP_DELALLOC; | |
6328 | em->block_len = found; | |
6329 | } | |
6330 | } else if (hole_em) { | |
6331 | return hole_em; | |
6332 | } | |
6333 | out: | |
6334 | ||
6335 | free_extent_map(hole_em); | |
6336 | if (err) { | |
6337 | free_extent_map(em); | |
6338 | return ERR_PTR(err); | |
6339 | } | |
6340 | return em; | |
6341 | } | |
6342 | ||
4b46fce2 JB |
6343 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
6344 | u64 start, u64 len) | |
6345 | { | |
6346 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6347 | struct btrfs_trans_handle *trans; | |
70c8a91c | 6348 | struct extent_map *em; |
4b46fce2 JB |
6349 | struct btrfs_key ins; |
6350 | u64 alloc_hint; | |
6351 | int ret; | |
4b46fce2 | 6352 | |
7a7eaa40 | 6353 | trans = btrfs_join_transaction(root); |
3612b495 TI |
6354 | if (IS_ERR(trans)) |
6355 | return ERR_CAST(trans); | |
4b46fce2 JB |
6356 | |
6357 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
6358 | ||
6359 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
6360 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
81c9ad23 | 6361 | alloc_hint, &ins, 1); |
4b46fce2 JB |
6362 | if (ret) { |
6363 | em = ERR_PTR(ret); | |
6364 | goto out; | |
6365 | } | |
6366 | ||
70c8a91c | 6367 | em = create_pinned_em(inode, start, ins.offset, start, ins.objectid, |
cc95bef6 | 6368 | ins.offset, ins.offset, ins.offset, 0); |
70c8a91c JB |
6369 | if (IS_ERR(em)) |
6370 | goto out; | |
4b46fce2 JB |
6371 | |
6372 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
6373 | ins.offset, ins.offset, 0); | |
6374 | if (ret) { | |
6375 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
6376 | em = ERR_PTR(ret); | |
6377 | } | |
6378 | out: | |
6379 | btrfs_end_transaction(trans, root); | |
6380 | return em; | |
6381 | } | |
6382 | ||
46bfbb5c CM |
6383 | /* |
6384 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
6385 | * block must be cow'd | |
6386 | */ | |
7ee9e440 JB |
6387 | noinline int can_nocow_extent(struct btrfs_trans_handle *trans, |
6388 | struct inode *inode, u64 offset, u64 *len, | |
6389 | u64 *orig_start, u64 *orig_block_len, | |
6390 | u64 *ram_bytes) | |
46bfbb5c CM |
6391 | { |
6392 | struct btrfs_path *path; | |
6393 | int ret; | |
6394 | struct extent_buffer *leaf; | |
6395 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6396 | struct btrfs_file_extent_item *fi; | |
6397 | struct btrfs_key key; | |
6398 | u64 disk_bytenr; | |
6399 | u64 backref_offset; | |
6400 | u64 extent_end; | |
6401 | u64 num_bytes; | |
6402 | int slot; | |
6403 | int found_type; | |
7ee9e440 | 6404 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
46bfbb5c CM |
6405 | path = btrfs_alloc_path(); |
6406 | if (!path) | |
6407 | return -ENOMEM; | |
6408 | ||
33345d01 | 6409 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
46bfbb5c CM |
6410 | offset, 0); |
6411 | if (ret < 0) | |
6412 | goto out; | |
6413 | ||
6414 | slot = path->slots[0]; | |
6415 | if (ret == 1) { | |
6416 | if (slot == 0) { | |
6417 | /* can't find the item, must cow */ | |
6418 | ret = 0; | |
6419 | goto out; | |
6420 | } | |
6421 | slot--; | |
6422 | } | |
6423 | ret = 0; | |
6424 | leaf = path->nodes[0]; | |
6425 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 6426 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
6427 | key.type != BTRFS_EXTENT_DATA_KEY) { |
6428 | /* not our file or wrong item type, must cow */ | |
6429 | goto out; | |
6430 | } | |
6431 | ||
6432 | if (key.offset > offset) { | |
6433 | /* Wrong offset, must cow */ | |
6434 | goto out; | |
6435 | } | |
6436 | ||
6437 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
6438 | found_type = btrfs_file_extent_type(leaf, fi); | |
6439 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
6440 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
6441 | /* not a regular extent, must cow */ | |
6442 | goto out; | |
6443 | } | |
7ee9e440 JB |
6444 | |
6445 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
6446 | goto out; | |
6447 | ||
46bfbb5c | 6448 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
6449 | if (disk_bytenr == 0) |
6450 | goto out; | |
6451 | ||
6452 | if (btrfs_file_extent_compression(leaf, fi) || | |
6453 | btrfs_file_extent_encryption(leaf, fi) || | |
6454 | btrfs_file_extent_other_encoding(leaf, fi)) | |
6455 | goto out; | |
6456 | ||
46bfbb5c CM |
6457 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
6458 | ||
7ee9e440 JB |
6459 | if (orig_start) { |
6460 | *orig_start = key.offset - backref_offset; | |
6461 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
6462 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
6463 | } | |
eb384b55 | 6464 | |
46bfbb5c | 6465 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
46bfbb5c CM |
6466 | |
6467 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
6468 | goto out; | |
6469 | ||
6470 | /* | |
6471 | * look for other files referencing this extent, if we | |
6472 | * find any we must cow | |
6473 | */ | |
33345d01 | 6474 | if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), |
46bfbb5c CM |
6475 | key.offset - backref_offset, disk_bytenr)) |
6476 | goto out; | |
6477 | ||
6478 | /* | |
6479 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
6480 | * in this extent we are about to write. If there | |
6481 | * are any csums in that range we have to cow in order | |
6482 | * to keep the csums correct | |
6483 | */ | |
6484 | disk_bytenr += backref_offset; | |
6485 | disk_bytenr += offset - key.offset; | |
eb384b55 | 6486 | num_bytes = min(offset + *len, extent_end) - offset; |
46bfbb5c CM |
6487 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) |
6488 | goto out; | |
6489 | /* | |
6490 | * all of the above have passed, it is safe to overwrite this extent | |
6491 | * without cow | |
6492 | */ | |
eb384b55 | 6493 | *len = num_bytes; |
46bfbb5c CM |
6494 | ret = 1; |
6495 | out: | |
6496 | btrfs_free_path(path); | |
6497 | return ret; | |
6498 | } | |
6499 | ||
eb838e73 JB |
6500 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
6501 | struct extent_state **cached_state, int writing) | |
6502 | { | |
6503 | struct btrfs_ordered_extent *ordered; | |
6504 | int ret = 0; | |
6505 | ||
6506 | while (1) { | |
6507 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6508 | 0, cached_state); | |
6509 | /* | |
6510 | * We're concerned with the entire range that we're going to be | |
6511 | * doing DIO to, so we need to make sure theres no ordered | |
6512 | * extents in this range. | |
6513 | */ | |
6514 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
6515 | lockend - lockstart + 1); | |
6516 | ||
6517 | /* | |
6518 | * We need to make sure there are no buffered pages in this | |
6519 | * range either, we could have raced between the invalidate in | |
6520 | * generic_file_direct_write and locking the extent. The | |
6521 | * invalidate needs to happen so that reads after a write do not | |
6522 | * get stale data. | |
6523 | */ | |
6524 | if (!ordered && (!writing || | |
6525 | !test_range_bit(&BTRFS_I(inode)->io_tree, | |
6526 | lockstart, lockend, EXTENT_UPTODATE, 0, | |
6527 | *cached_state))) | |
6528 | break; | |
6529 | ||
6530 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6531 | cached_state, GFP_NOFS); | |
6532 | ||
6533 | if (ordered) { | |
6534 | btrfs_start_ordered_extent(inode, ordered, 1); | |
6535 | btrfs_put_ordered_extent(ordered); | |
6536 | } else { | |
6537 | /* Screw you mmap */ | |
6538 | ret = filemap_write_and_wait_range(inode->i_mapping, | |
6539 | lockstart, | |
6540 | lockend); | |
6541 | if (ret) | |
6542 | break; | |
6543 | ||
6544 | /* | |
6545 | * If we found a page that couldn't be invalidated just | |
6546 | * fall back to buffered. | |
6547 | */ | |
6548 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
6549 | lockstart >> PAGE_CACHE_SHIFT, | |
6550 | lockend >> PAGE_CACHE_SHIFT); | |
6551 | if (ret) | |
6552 | break; | |
6553 | } | |
6554 | ||
6555 | cond_resched(); | |
6556 | } | |
6557 | ||
6558 | return ret; | |
6559 | } | |
6560 | ||
69ffb543 JB |
6561 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
6562 | u64 len, u64 orig_start, | |
6563 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
6564 | u64 orig_block_len, u64 ram_bytes, |
6565 | int type) | |
69ffb543 JB |
6566 | { |
6567 | struct extent_map_tree *em_tree; | |
6568 | struct extent_map *em; | |
6569 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6570 | int ret; | |
6571 | ||
6572 | em_tree = &BTRFS_I(inode)->extent_tree; | |
6573 | em = alloc_extent_map(); | |
6574 | if (!em) | |
6575 | return ERR_PTR(-ENOMEM); | |
6576 | ||
6577 | em->start = start; | |
6578 | em->orig_start = orig_start; | |
2ab28f32 JB |
6579 | em->mod_start = start; |
6580 | em->mod_len = len; | |
69ffb543 JB |
6581 | em->len = len; |
6582 | em->block_len = block_len; | |
6583 | em->block_start = block_start; | |
6584 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 6585 | em->orig_block_len = orig_block_len; |
cc95bef6 | 6586 | em->ram_bytes = ram_bytes; |
70c8a91c | 6587 | em->generation = -1; |
69ffb543 JB |
6588 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
6589 | if (type == BTRFS_ORDERED_PREALLOC) | |
b11e234d | 6590 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
69ffb543 JB |
6591 | |
6592 | do { | |
6593 | btrfs_drop_extent_cache(inode, em->start, | |
6594 | em->start + em->len - 1, 0); | |
6595 | write_lock(&em_tree->lock); | |
09a2a8f9 | 6596 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 JB |
6597 | write_unlock(&em_tree->lock); |
6598 | } while (ret == -EEXIST); | |
6599 | ||
6600 | if (ret) { | |
6601 | free_extent_map(em); | |
6602 | return ERR_PTR(ret); | |
6603 | } | |
6604 | ||
6605 | return em; | |
6606 | } | |
6607 | ||
6608 | ||
4b46fce2 JB |
6609 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
6610 | struct buffer_head *bh_result, int create) | |
6611 | { | |
6612 | struct extent_map *em; | |
6613 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 6614 | struct extent_state *cached_state = NULL; |
4b46fce2 | 6615 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 6616 | u64 lockstart, lockend; |
4b46fce2 | 6617 | u64 len = bh_result->b_size; |
46bfbb5c | 6618 | struct btrfs_trans_handle *trans; |
eb838e73 | 6619 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 6620 | int ret = 0; |
eb838e73 | 6621 | |
172a5049 | 6622 | if (create) |
eb838e73 | 6623 | unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY; |
172a5049 | 6624 | else |
c329861d | 6625 | len = min_t(u64, len, root->sectorsize); |
eb838e73 | 6626 | |
c329861d JB |
6627 | lockstart = start; |
6628 | lockend = start + len - 1; | |
6629 | ||
eb838e73 JB |
6630 | /* |
6631 | * If this errors out it's because we couldn't invalidate pagecache for | |
6632 | * this range and we need to fallback to buffered. | |
6633 | */ | |
6634 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create)) | |
6635 | return -ENOTBLK; | |
6636 | ||
4b46fce2 | 6637 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
eb838e73 JB |
6638 | if (IS_ERR(em)) { |
6639 | ret = PTR_ERR(em); | |
6640 | goto unlock_err; | |
6641 | } | |
4b46fce2 JB |
6642 | |
6643 | /* | |
6644 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
6645 | * io. INLINE is special, and we could probably kludge it in here, but | |
6646 | * it's still buffered so for safety lets just fall back to the generic | |
6647 | * buffered path. | |
6648 | * | |
6649 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
6650 | * decompress it, so there will be buffering required no matter what we | |
6651 | * do, so go ahead and fallback to buffered. | |
6652 | * | |
6653 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
6654 | * to buffered IO. Don't blame me, this is the price we pay for using | |
6655 | * the generic code. | |
6656 | */ | |
6657 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
6658 | em->block_start == EXTENT_MAP_INLINE) { | |
6659 | free_extent_map(em); | |
eb838e73 JB |
6660 | ret = -ENOTBLK; |
6661 | goto unlock_err; | |
4b46fce2 JB |
6662 | } |
6663 | ||
6664 | /* Just a good old fashioned hole, return */ | |
6665 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
6666 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
6667 | free_extent_map(em); | |
eb838e73 | 6668 | goto unlock_err; |
4b46fce2 JB |
6669 | } |
6670 | ||
6671 | /* | |
6672 | * We don't allocate a new extent in the following cases | |
6673 | * | |
6674 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
6675 | * existing extent. | |
6676 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
6677 | * just use the extent. | |
6678 | * | |
6679 | */ | |
46bfbb5c | 6680 | if (!create) { |
eb838e73 JB |
6681 | len = min(len, em->len - (start - em->start)); |
6682 | lockstart = start + len; | |
6683 | goto unlock; | |
46bfbb5c | 6684 | } |
4b46fce2 JB |
6685 | |
6686 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
6687 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
6688 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
6689 | int type; |
6690 | int ret; | |
eb384b55 | 6691 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
6692 | |
6693 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6694 | type = BTRFS_ORDERED_PREALLOC; | |
6695 | else | |
6696 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 6697 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 6698 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
6699 | |
6700 | /* | |
6701 | * we're not going to log anything, but we do need | |
6702 | * to make sure the current transaction stays open | |
6703 | * while we look for nocow cross refs | |
6704 | */ | |
7a7eaa40 | 6705 | trans = btrfs_join_transaction(root); |
3612b495 | 6706 | if (IS_ERR(trans)) |
46bfbb5c CM |
6707 | goto must_cow; |
6708 | ||
7ee9e440 JB |
6709 | if (can_nocow_extent(trans, inode, start, &len, &orig_start, |
6710 | &orig_block_len, &ram_bytes) == 1) { | |
69ffb543 JB |
6711 | if (type == BTRFS_ORDERED_PREALLOC) { |
6712 | free_extent_map(em); | |
6713 | em = create_pinned_em(inode, start, len, | |
6714 | orig_start, | |
b4939680 | 6715 | block_start, len, |
cc95bef6 JB |
6716 | orig_block_len, |
6717 | ram_bytes, type); | |
69ffb543 JB |
6718 | if (IS_ERR(em)) { |
6719 | btrfs_end_transaction(trans, root); | |
6720 | goto unlock_err; | |
6721 | } | |
6722 | } | |
6723 | ||
46bfbb5c CM |
6724 | ret = btrfs_add_ordered_extent_dio(inode, start, |
6725 | block_start, len, len, type); | |
6726 | btrfs_end_transaction(trans, root); | |
6727 | if (ret) { | |
6728 | free_extent_map(em); | |
eb838e73 | 6729 | goto unlock_err; |
46bfbb5c CM |
6730 | } |
6731 | goto unlock; | |
4b46fce2 | 6732 | } |
46bfbb5c | 6733 | btrfs_end_transaction(trans, root); |
4b46fce2 | 6734 | } |
46bfbb5c CM |
6735 | must_cow: |
6736 | /* | |
6737 | * this will cow the extent, reset the len in case we changed | |
6738 | * it above | |
6739 | */ | |
6740 | len = bh_result->b_size; | |
70c8a91c JB |
6741 | free_extent_map(em); |
6742 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
6743 | if (IS_ERR(em)) { |
6744 | ret = PTR_ERR(em); | |
6745 | goto unlock_err; | |
6746 | } | |
46bfbb5c CM |
6747 | len = min(len, em->len - (start - em->start)); |
6748 | unlock: | |
4b46fce2 JB |
6749 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
6750 | inode->i_blkbits; | |
46bfbb5c | 6751 | bh_result->b_size = len; |
4b46fce2 JB |
6752 | bh_result->b_bdev = em->bdev; |
6753 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
6754 | if (create) { |
6755 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6756 | set_buffer_new(bh_result); | |
6757 | ||
6758 | /* | |
6759 | * Need to update the i_size under the extent lock so buffered | |
6760 | * readers will get the updated i_size when we unlock. | |
6761 | */ | |
6762 | if (start + len > i_size_read(inode)) | |
6763 | i_size_write(inode, start + len); | |
0934856d | 6764 | |
172a5049 MX |
6765 | spin_lock(&BTRFS_I(inode)->lock); |
6766 | BTRFS_I(inode)->outstanding_extents++; | |
6767 | spin_unlock(&BTRFS_I(inode)->lock); | |
6768 | ||
0934856d MX |
6769 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
6770 | lockstart + len - 1, EXTENT_DELALLOC, NULL, | |
6771 | &cached_state, GFP_NOFS); | |
6772 | BUG_ON(ret); | |
c3473e83 | 6773 | } |
4b46fce2 | 6774 | |
eb838e73 JB |
6775 | /* |
6776 | * In the case of write we need to clear and unlock the entire range, | |
6777 | * in the case of read we need to unlock only the end area that we | |
6778 | * aren't using if there is any left over space. | |
6779 | */ | |
24c03fa5 | 6780 | if (lockstart < lockend) { |
0934856d MX |
6781 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
6782 | lockend, unlock_bits, 1, 0, | |
6783 | &cached_state, GFP_NOFS); | |
24c03fa5 | 6784 | } else { |
eb838e73 | 6785 | free_extent_state(cached_state); |
24c03fa5 | 6786 | } |
eb838e73 | 6787 | |
4b46fce2 JB |
6788 | free_extent_map(em); |
6789 | ||
6790 | return 0; | |
eb838e73 JB |
6791 | |
6792 | unlock_err: | |
eb838e73 JB |
6793 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6794 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
6795 | return ret; | |
4b46fce2 JB |
6796 | } |
6797 | ||
4b46fce2 JB |
6798 | static void btrfs_endio_direct_read(struct bio *bio, int err) |
6799 | { | |
e65e1535 | 6800 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
6801 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
6802 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
6803 | struct inode *inode = dip->inode; |
6804 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9be3395b | 6805 | struct bio *dio_bio; |
facc8a22 MX |
6806 | u32 *csums = (u32 *)dip->csum; |
6807 | int index = 0; | |
4b46fce2 | 6808 | u64 start; |
4b46fce2 JB |
6809 | |
6810 | start = dip->logical_offset; | |
6811 | do { | |
6812 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
6813 | struct page *page = bvec->bv_page; | |
6814 | char *kaddr; | |
6815 | u32 csum = ~(u32)0; | |
6816 | unsigned long flags; | |
6817 | ||
6818 | local_irq_save(flags); | |
7ac687d9 | 6819 | kaddr = kmap_atomic(page); |
b0496686 | 6820 | csum = btrfs_csum_data(kaddr + bvec->bv_offset, |
4b46fce2 JB |
6821 | csum, bvec->bv_len); |
6822 | btrfs_csum_final(csum, (char *)&csum); | |
7ac687d9 | 6823 | kunmap_atomic(kaddr); |
4b46fce2 JB |
6824 | local_irq_restore(flags); |
6825 | ||
6826 | flush_dcache_page(bvec->bv_page); | |
facc8a22 MX |
6827 | if (csum != csums[index]) { |
6828 | btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u", | |
6829 | (unsigned long long)btrfs_ino(inode), | |
6830 | (unsigned long long)start, | |
6831 | csum, csums[index]); | |
4b46fce2 JB |
6832 | err = -EIO; |
6833 | } | |
6834 | } | |
6835 | ||
6836 | start += bvec->bv_len; | |
4b46fce2 | 6837 | bvec++; |
facc8a22 | 6838 | index++; |
4b46fce2 JB |
6839 | } while (bvec <= bvec_end); |
6840 | ||
6841 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
d0082371 | 6842 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 6843 | dio_bio = dip->dio_bio; |
4b46fce2 | 6844 | |
4b46fce2 | 6845 | kfree(dip); |
c0da7aa1 JB |
6846 | |
6847 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
6848 | if (err) | |
9be3395b CM |
6849 | clear_bit(BIO_UPTODATE, &dio_bio->bi_flags); |
6850 | dio_end_io(dio_bio, err); | |
6851 | bio_put(bio); | |
4b46fce2 JB |
6852 | } |
6853 | ||
6854 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
6855 | { | |
6856 | struct btrfs_dio_private *dip = bio->bi_private; | |
6857 | struct inode *inode = dip->inode; | |
6858 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 6859 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
6860 | u64 ordered_offset = dip->logical_offset; |
6861 | u64 ordered_bytes = dip->bytes; | |
9be3395b | 6862 | struct bio *dio_bio; |
4b46fce2 JB |
6863 | int ret; |
6864 | ||
6865 | if (err) | |
6866 | goto out_done; | |
163cf09c CM |
6867 | again: |
6868 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
6869 | &ordered_offset, | |
5fd02043 | 6870 | ordered_bytes, !err); |
4b46fce2 | 6871 | if (!ret) |
163cf09c | 6872 | goto out_test; |
4b46fce2 | 6873 | |
5fd02043 JB |
6874 | ordered->work.func = finish_ordered_fn; |
6875 | ordered->work.flags = 0; | |
6876 | btrfs_queue_worker(&root->fs_info->endio_write_workers, | |
6877 | &ordered->work); | |
163cf09c CM |
6878 | out_test: |
6879 | /* | |
6880 | * our bio might span multiple ordered extents. If we haven't | |
6881 | * completed the accounting for the whole dio, go back and try again | |
6882 | */ | |
6883 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
6884 | ordered_bytes = dip->logical_offset + dip->bytes - | |
6885 | ordered_offset; | |
5fd02043 | 6886 | ordered = NULL; |
163cf09c CM |
6887 | goto again; |
6888 | } | |
4b46fce2 | 6889 | out_done: |
9be3395b | 6890 | dio_bio = dip->dio_bio; |
4b46fce2 | 6891 | |
4b46fce2 | 6892 | kfree(dip); |
c0da7aa1 JB |
6893 | |
6894 | /* If we had an error make sure to clear the uptodate flag */ | |
6895 | if (err) | |
9be3395b CM |
6896 | clear_bit(BIO_UPTODATE, &dio_bio->bi_flags); |
6897 | dio_end_io(dio_bio, err); | |
6898 | bio_put(bio); | |
4b46fce2 JB |
6899 | } |
6900 | ||
eaf25d93 CM |
6901 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
6902 | struct bio *bio, int mirror_num, | |
6903 | unsigned long bio_flags, u64 offset) | |
6904 | { | |
6905 | int ret; | |
6906 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6907 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 6908 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
6909 | return 0; |
6910 | } | |
6911 | ||
e65e1535 MX |
6912 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
6913 | { | |
6914 | struct btrfs_dio_private *dip = bio->bi_private; | |
6915 | ||
6916 | if (err) { | |
33345d01 | 6917 | printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu " |
3dd1462e | 6918 | "sector %#Lx len %u err no %d\n", |
33345d01 | 6919 | (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw, |
3dd1462e | 6920 | (unsigned long long)bio->bi_sector, bio->bi_size, err); |
e65e1535 MX |
6921 | dip->errors = 1; |
6922 | ||
6923 | /* | |
6924 | * before atomic variable goto zero, we must make sure | |
6925 | * dip->errors is perceived to be set. | |
6926 | */ | |
6927 | smp_mb__before_atomic_dec(); | |
6928 | } | |
6929 | ||
6930 | /* if there are more bios still pending for this dio, just exit */ | |
6931 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
6932 | goto out; | |
6933 | ||
9be3395b | 6934 | if (dip->errors) { |
e65e1535 | 6935 | bio_io_error(dip->orig_bio); |
9be3395b CM |
6936 | } else { |
6937 | set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags); | |
e65e1535 MX |
6938 | bio_endio(dip->orig_bio, 0); |
6939 | } | |
6940 | out: | |
6941 | bio_put(bio); | |
6942 | } | |
6943 | ||
6944 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
6945 | u64 first_sector, gfp_t gfp_flags) | |
6946 | { | |
6947 | int nr_vecs = bio_get_nr_vecs(bdev); | |
6948 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
6949 | } | |
6950 | ||
6951 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
6952 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 6953 | int async_submit) |
e65e1535 | 6954 | { |
facc8a22 | 6955 | struct btrfs_dio_private *dip = bio->bi_private; |
e65e1535 MX |
6956 | int write = rw & REQ_WRITE; |
6957 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6958 | int ret; | |
6959 | ||
b812ce28 JB |
6960 | if (async_submit) |
6961 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
6962 | ||
e65e1535 | 6963 | bio_get(bio); |
5fd02043 JB |
6964 | |
6965 | if (!write) { | |
6966 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
6967 | if (ret) | |
6968 | goto err; | |
6969 | } | |
e65e1535 | 6970 | |
1ae39938 JB |
6971 | if (skip_sum) |
6972 | goto map; | |
6973 | ||
6974 | if (write && async_submit) { | |
e65e1535 MX |
6975 | ret = btrfs_wq_submit_bio(root->fs_info, |
6976 | inode, rw, bio, 0, 0, | |
6977 | file_offset, | |
6978 | __btrfs_submit_bio_start_direct_io, | |
6979 | __btrfs_submit_bio_done); | |
6980 | goto err; | |
1ae39938 JB |
6981 | } else if (write) { |
6982 | /* | |
6983 | * If we aren't doing async submit, calculate the csum of the | |
6984 | * bio now. | |
6985 | */ | |
6986 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
6987 | if (ret) | |
6988 | goto err; | |
c2db1073 | 6989 | } else if (!skip_sum) { |
facc8a22 MX |
6990 | ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio, |
6991 | file_offset); | |
c2db1073 TI |
6992 | if (ret) |
6993 | goto err; | |
6994 | } | |
e65e1535 | 6995 | |
1ae39938 JB |
6996 | map: |
6997 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
6998 | err: |
6999 | bio_put(bio); | |
7000 | return ret; | |
7001 | } | |
7002 | ||
7003 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
7004 | int skip_sum) | |
7005 | { | |
7006 | struct inode *inode = dip->inode; | |
7007 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
e65e1535 MX |
7008 | struct bio *bio; |
7009 | struct bio *orig_bio = dip->orig_bio; | |
7010 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
7011 | u64 start_sector = orig_bio->bi_sector; | |
7012 | u64 file_offset = dip->logical_offset; | |
7013 | u64 submit_len = 0; | |
7014 | u64 map_length; | |
7015 | int nr_pages = 0; | |
e65e1535 | 7016 | int ret = 0; |
1ae39938 | 7017 | int async_submit = 0; |
e65e1535 | 7018 | |
e65e1535 | 7019 | map_length = orig_bio->bi_size; |
53b381b3 | 7020 | ret = btrfs_map_block(root->fs_info, rw, start_sector << 9, |
e65e1535 MX |
7021 | &map_length, NULL, 0); |
7022 | if (ret) { | |
64728bbb | 7023 | bio_put(orig_bio); |
e65e1535 MX |
7024 | return -EIO; |
7025 | } | |
facc8a22 | 7026 | |
02f57c7a JB |
7027 | if (map_length >= orig_bio->bi_size) { |
7028 | bio = orig_bio; | |
7029 | goto submit; | |
7030 | } | |
7031 | ||
53b381b3 DW |
7032 | /* async crcs make it difficult to collect full stripe writes. */ |
7033 | if (btrfs_get_alloc_profile(root, 1) & | |
7034 | (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) | |
7035 | async_submit = 0; | |
7036 | else | |
7037 | async_submit = 1; | |
7038 | ||
02f57c7a JB |
7039 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
7040 | if (!bio) | |
7041 | return -ENOMEM; | |
7042 | bio->bi_private = dip; | |
7043 | bio->bi_end_io = btrfs_end_dio_bio; | |
7044 | atomic_inc(&dip->pending_bios); | |
7045 | ||
e65e1535 MX |
7046 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
7047 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
7048 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
7049 | bvec->bv_offset) < bvec->bv_len)) { | |
7050 | /* | |
7051 | * inc the count before we submit the bio so | |
7052 | * we know the end IO handler won't happen before | |
7053 | * we inc the count. Otherwise, the dip might get freed | |
7054 | * before we're done setting it up | |
7055 | */ | |
7056 | atomic_inc(&dip->pending_bios); | |
7057 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
7058 | file_offset, skip_sum, | |
c329861d | 7059 | async_submit); |
e65e1535 MX |
7060 | if (ret) { |
7061 | bio_put(bio); | |
7062 | atomic_dec(&dip->pending_bios); | |
7063 | goto out_err; | |
7064 | } | |
7065 | ||
e65e1535 MX |
7066 | start_sector += submit_len >> 9; |
7067 | file_offset += submit_len; | |
7068 | ||
7069 | submit_len = 0; | |
7070 | nr_pages = 0; | |
7071 | ||
7072 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
7073 | start_sector, GFP_NOFS); | |
7074 | if (!bio) | |
7075 | goto out_err; | |
7076 | bio->bi_private = dip; | |
7077 | bio->bi_end_io = btrfs_end_dio_bio; | |
7078 | ||
7079 | map_length = orig_bio->bi_size; | |
53b381b3 | 7080 | ret = btrfs_map_block(root->fs_info, rw, |
3ec706c8 | 7081 | start_sector << 9, |
e65e1535 MX |
7082 | &map_length, NULL, 0); |
7083 | if (ret) { | |
7084 | bio_put(bio); | |
7085 | goto out_err; | |
7086 | } | |
7087 | } else { | |
7088 | submit_len += bvec->bv_len; | |
7089 | nr_pages ++; | |
7090 | bvec++; | |
7091 | } | |
7092 | } | |
7093 | ||
02f57c7a | 7094 | submit: |
e65e1535 | 7095 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 7096 | async_submit); |
e65e1535 MX |
7097 | if (!ret) |
7098 | return 0; | |
7099 | ||
7100 | bio_put(bio); | |
7101 | out_err: | |
7102 | dip->errors = 1; | |
7103 | /* | |
7104 | * before atomic variable goto zero, we must | |
7105 | * make sure dip->errors is perceived to be set. | |
7106 | */ | |
7107 | smp_mb__before_atomic_dec(); | |
7108 | if (atomic_dec_and_test(&dip->pending_bios)) | |
7109 | bio_io_error(dip->orig_bio); | |
7110 | ||
7111 | /* bio_end_io() will handle error, so we needn't return it */ | |
7112 | return 0; | |
7113 | } | |
7114 | ||
9be3395b CM |
7115 | static void btrfs_submit_direct(int rw, struct bio *dio_bio, |
7116 | struct inode *inode, loff_t file_offset) | |
4b46fce2 JB |
7117 | { |
7118 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7119 | struct btrfs_dio_private *dip; | |
9be3395b | 7120 | struct bio *io_bio; |
4b46fce2 | 7121 | int skip_sum; |
facc8a22 | 7122 | int sum_len; |
7b6d91da | 7123 | int write = rw & REQ_WRITE; |
4b46fce2 | 7124 | int ret = 0; |
facc8a22 | 7125 | u16 csum_size; |
4b46fce2 JB |
7126 | |
7127 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
7128 | ||
9be3395b | 7129 | io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS); |
9be3395b CM |
7130 | if (!io_bio) { |
7131 | ret = -ENOMEM; | |
7132 | goto free_ordered; | |
7133 | } | |
7134 | ||
facc8a22 MX |
7135 | if (!skip_sum && !write) { |
7136 | csum_size = btrfs_super_csum_size(root->fs_info->super_copy); | |
7137 | sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits; | |
7138 | sum_len *= csum_size; | |
7139 | } else { | |
7140 | sum_len = 0; | |
7141 | } | |
7142 | ||
7143 | dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS); | |
4b46fce2 JB |
7144 | if (!dip) { |
7145 | ret = -ENOMEM; | |
9be3395b | 7146 | goto free_io_bio; |
4b46fce2 | 7147 | } |
4b46fce2 | 7148 | |
9be3395b | 7149 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
7150 | dip->inode = inode; |
7151 | dip->logical_offset = file_offset; | |
e6da5d2e | 7152 | dip->bytes = dio_bio->bi_size; |
9be3395b CM |
7153 | dip->disk_bytenr = (u64)dio_bio->bi_sector << 9; |
7154 | io_bio->bi_private = dip; | |
e65e1535 | 7155 | dip->errors = 0; |
9be3395b CM |
7156 | dip->orig_bio = io_bio; |
7157 | dip->dio_bio = dio_bio; | |
e65e1535 | 7158 | atomic_set(&dip->pending_bios, 0); |
4b46fce2 JB |
7159 | |
7160 | if (write) | |
9be3395b | 7161 | io_bio->bi_end_io = btrfs_endio_direct_write; |
4b46fce2 | 7162 | else |
9be3395b | 7163 | io_bio->bi_end_io = btrfs_endio_direct_read; |
4b46fce2 | 7164 | |
e65e1535 MX |
7165 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
7166 | if (!ret) | |
eaf25d93 | 7167 | return; |
9be3395b CM |
7168 | |
7169 | free_io_bio: | |
7170 | bio_put(io_bio); | |
7171 | ||
4b46fce2 JB |
7172 | free_ordered: |
7173 | /* | |
7174 | * If this is a write, we need to clean up the reserved space and kill | |
7175 | * the ordered extent. | |
7176 | */ | |
7177 | if (write) { | |
7178 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 7179 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
7180 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
7181 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
7182 | btrfs_free_reserved_extent(root, ordered->start, | |
7183 | ordered->disk_len); | |
7184 | btrfs_put_ordered_extent(ordered); | |
7185 | btrfs_put_ordered_extent(ordered); | |
7186 | } | |
9be3395b | 7187 | bio_endio(dio_bio, ret); |
4b46fce2 JB |
7188 | } |
7189 | ||
5a5f79b5 CM |
7190 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
7191 | const struct iovec *iov, loff_t offset, | |
7192 | unsigned long nr_segs) | |
7193 | { | |
7194 | int seg; | |
a1b75f7d | 7195 | int i; |
5a5f79b5 CM |
7196 | size_t size; |
7197 | unsigned long addr; | |
7198 | unsigned blocksize_mask = root->sectorsize - 1; | |
7199 | ssize_t retval = -EINVAL; | |
7200 | loff_t end = offset; | |
7201 | ||
7202 | if (offset & blocksize_mask) | |
7203 | goto out; | |
7204 | ||
7205 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
7206 | for (seg = 0; seg < nr_segs; seg++) { | |
7207 | addr = (unsigned long)iov[seg].iov_base; | |
7208 | size = iov[seg].iov_len; | |
7209 | end += size; | |
a1b75f7d | 7210 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 7211 | goto out; |
a1b75f7d JB |
7212 | |
7213 | /* If this is a write we don't need to check anymore */ | |
7214 | if (rw & WRITE) | |
7215 | continue; | |
7216 | ||
7217 | /* | |
7218 | * Check to make sure we don't have duplicate iov_base's in this | |
7219 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
7220 | * when reading back. | |
7221 | */ | |
7222 | for (i = seg + 1; i < nr_segs; i++) { | |
7223 | if (iov[seg].iov_base == iov[i].iov_base) | |
7224 | goto out; | |
7225 | } | |
5a5f79b5 CM |
7226 | } |
7227 | retval = 0; | |
7228 | out: | |
7229 | return retval; | |
7230 | } | |
eb838e73 | 7231 | |
16432985 CM |
7232 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
7233 | const struct iovec *iov, loff_t offset, | |
7234 | unsigned long nr_segs) | |
7235 | { | |
4b46fce2 JB |
7236 | struct file *file = iocb->ki_filp; |
7237 | struct inode *inode = file->f_mapping->host; | |
0934856d | 7238 | size_t count = 0; |
2e60a51e | 7239 | int flags = 0; |
38851cc1 MX |
7240 | bool wakeup = true; |
7241 | bool relock = false; | |
0934856d | 7242 | ssize_t ret; |
4b46fce2 | 7243 | |
5a5f79b5 | 7244 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
eb838e73 | 7245 | offset, nr_segs)) |
5a5f79b5 | 7246 | return 0; |
3f7c579c | 7247 | |
38851cc1 MX |
7248 | atomic_inc(&inode->i_dio_count); |
7249 | smp_mb__after_atomic_inc(); | |
7250 | ||
0e267c44 JB |
7251 | /* |
7252 | * The generic stuff only does filemap_write_and_wait_range, which isn't | |
7253 | * enough if we've written compressed pages to this area, so we need to | |
7254 | * call btrfs_wait_ordered_range to make absolutely sure that any | |
7255 | * outstanding dirty pages are on disk. | |
7256 | */ | |
7257 | count = iov_length(iov, nr_segs); | |
7258 | btrfs_wait_ordered_range(inode, offset, count); | |
7259 | ||
0934856d | 7260 | if (rw & WRITE) { |
38851cc1 MX |
7261 | /* |
7262 | * If the write DIO is beyond the EOF, we need update | |
7263 | * the isize, but it is protected by i_mutex. So we can | |
7264 | * not unlock the i_mutex at this case. | |
7265 | */ | |
7266 | if (offset + count <= inode->i_size) { | |
7267 | mutex_unlock(&inode->i_mutex); | |
7268 | relock = true; | |
7269 | } | |
0934856d MX |
7270 | ret = btrfs_delalloc_reserve_space(inode, count); |
7271 | if (ret) | |
38851cc1 MX |
7272 | goto out; |
7273 | } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK, | |
7274 | &BTRFS_I(inode)->runtime_flags))) { | |
7275 | inode_dio_done(inode); | |
7276 | flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
7277 | wakeup = false; | |
0934856d MX |
7278 | } |
7279 | ||
7280 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
7281 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
7282 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
2e60a51e | 7283 | btrfs_submit_direct, flags); |
0934856d MX |
7284 | if (rw & WRITE) { |
7285 | if (ret < 0 && ret != -EIOCBQUEUED) | |
7286 | btrfs_delalloc_release_space(inode, count); | |
172a5049 | 7287 | else if (ret >= 0 && (size_t)ret < count) |
0934856d MX |
7288 | btrfs_delalloc_release_space(inode, |
7289 | count - (size_t)ret); | |
172a5049 MX |
7290 | else |
7291 | btrfs_delalloc_release_metadata(inode, 0); | |
0934856d | 7292 | } |
38851cc1 | 7293 | out: |
2e60a51e MX |
7294 | if (wakeup) |
7295 | inode_dio_done(inode); | |
38851cc1 MX |
7296 | if (relock) |
7297 | mutex_lock(&inode->i_mutex); | |
0934856d MX |
7298 | |
7299 | return ret; | |
16432985 CM |
7300 | } |
7301 | ||
05dadc09 TI |
7302 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
7303 | ||
1506fcc8 YS |
7304 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
7305 | __u64 start, __u64 len) | |
7306 | { | |
05dadc09 TI |
7307 | int ret; |
7308 | ||
7309 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
7310 | if (ret) | |
7311 | return ret; | |
7312 | ||
ec29ed5b | 7313 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
7314 | } |
7315 | ||
a52d9a80 | 7316 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 7317 | { |
d1310b2e CM |
7318 | struct extent_io_tree *tree; |
7319 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 7320 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 7321 | } |
1832a6d5 | 7322 | |
a52d9a80 | 7323 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 7324 | { |
d1310b2e | 7325 | struct extent_io_tree *tree; |
b888db2b CM |
7326 | |
7327 | ||
7328 | if (current->flags & PF_MEMALLOC) { | |
7329 | redirty_page_for_writepage(wbc, page); | |
7330 | unlock_page(page); | |
7331 | return 0; | |
7332 | } | |
d1310b2e | 7333 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 7334 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
7335 | } |
7336 | ||
48a3b636 ES |
7337 | static int btrfs_writepages(struct address_space *mapping, |
7338 | struct writeback_control *wbc) | |
b293f02e | 7339 | { |
d1310b2e | 7340 | struct extent_io_tree *tree; |
771ed689 | 7341 | |
d1310b2e | 7342 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
7343 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
7344 | } | |
7345 | ||
3ab2fb5a CM |
7346 | static int |
7347 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
7348 | struct list_head *pages, unsigned nr_pages) | |
7349 | { | |
d1310b2e CM |
7350 | struct extent_io_tree *tree; |
7351 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
7352 | return extent_readpages(tree, mapping, pages, nr_pages, |
7353 | btrfs_get_extent); | |
7354 | } | |
e6dcd2dc | 7355 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 7356 | { |
d1310b2e CM |
7357 | struct extent_io_tree *tree; |
7358 | struct extent_map_tree *map; | |
a52d9a80 | 7359 | int ret; |
8c2383c3 | 7360 | |
d1310b2e CM |
7361 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
7362 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 7363 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
7364 | if (ret == 1) { |
7365 | ClearPagePrivate(page); | |
7366 | set_page_private(page, 0); | |
7367 | page_cache_release(page); | |
39279cc3 | 7368 | } |
a52d9a80 | 7369 | return ret; |
39279cc3 CM |
7370 | } |
7371 | ||
e6dcd2dc CM |
7372 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
7373 | { | |
98509cfc CM |
7374 | if (PageWriteback(page) || PageDirty(page)) |
7375 | return 0; | |
b335b003 | 7376 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
7377 | } |
7378 | ||
d47992f8 LC |
7379 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
7380 | unsigned int length) | |
39279cc3 | 7381 | { |
5fd02043 | 7382 | struct inode *inode = page->mapping->host; |
d1310b2e | 7383 | struct extent_io_tree *tree; |
e6dcd2dc | 7384 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 7385 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
7386 | u64 page_start = page_offset(page); |
7387 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 7388 | |
8b62b72b CM |
7389 | /* |
7390 | * we have the page locked, so new writeback can't start, | |
7391 | * and the dirty bit won't be cleared while we are here. | |
7392 | * | |
7393 | * Wait for IO on this page so that we can safely clear | |
7394 | * the PagePrivate2 bit and do ordered accounting | |
7395 | */ | |
e6dcd2dc | 7396 | wait_on_page_writeback(page); |
8b62b72b | 7397 | |
5fd02043 | 7398 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
7399 | if (offset) { |
7400 | btrfs_releasepage(page, GFP_NOFS); | |
7401 | return; | |
7402 | } | |
d0082371 | 7403 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
4eee4fa4 | 7404 | ordered = btrfs_lookup_ordered_extent(inode, page_offset(page)); |
e6dcd2dc | 7405 | if (ordered) { |
eb84ae03 CM |
7406 | /* |
7407 | * IO on this page will never be started, so we need | |
7408 | * to account for any ordered extents now | |
7409 | */ | |
e6dcd2dc CM |
7410 | clear_extent_bit(tree, page_start, page_end, |
7411 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
9e8a4a8b LB |
7412 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
7413 | EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS); | |
8b62b72b CM |
7414 | /* |
7415 | * whoever cleared the private bit is responsible | |
7416 | * for the finish_ordered_io | |
7417 | */ | |
5fd02043 JB |
7418 | if (TestClearPagePrivate2(page) && |
7419 | btrfs_dec_test_ordered_pending(inode, &ordered, page_start, | |
7420 | PAGE_CACHE_SIZE, 1)) { | |
7421 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 7422 | } |
e6dcd2dc | 7423 | btrfs_put_ordered_extent(ordered); |
2ac55d41 | 7424 | cached_state = NULL; |
d0082371 | 7425 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
7426 | } |
7427 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 7428 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
9e8a4a8b LB |
7429 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, |
7430 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
7431 | __btrfs_releasepage(page, GFP_NOFS); |
7432 | ||
4a096752 | 7433 | ClearPageChecked(page); |
9ad6b7bc | 7434 | if (PagePrivate(page)) { |
9ad6b7bc CM |
7435 | ClearPagePrivate(page); |
7436 | set_page_private(page, 0); | |
7437 | page_cache_release(page); | |
7438 | } | |
39279cc3 CM |
7439 | } |
7440 | ||
9ebefb18 CM |
7441 | /* |
7442 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
7443 | * called from a page fault handler when a page is first dirtied. Hence we must | |
7444 | * be careful to check for EOF conditions here. We set the page up correctly | |
7445 | * for a written page which means we get ENOSPC checking when writing into | |
7446 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
7447 | * support these features. | |
7448 | * | |
7449 | * We are not allowed to take the i_mutex here so we have to play games to | |
7450 | * protect against truncate races as the page could now be beyond EOF. Because | |
7451 | * vmtruncate() writes the inode size before removing pages, once we have the | |
7452 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
7453 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
7454 | * unlock the page. | |
7455 | */ | |
c2ec175c | 7456 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 7457 | { |
c2ec175c | 7458 | struct page *page = vmf->page; |
496ad9aa | 7459 | struct inode *inode = file_inode(vma->vm_file); |
1832a6d5 | 7460 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
7461 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
7462 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 7463 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
7464 | char *kaddr; |
7465 | unsigned long zero_start; | |
9ebefb18 | 7466 | loff_t size; |
1832a6d5 | 7467 | int ret; |
9998eb70 | 7468 | int reserved = 0; |
a52d9a80 | 7469 | u64 page_start; |
e6dcd2dc | 7470 | u64 page_end; |
9ebefb18 | 7471 | |
b2b5ef5c | 7472 | sb_start_pagefault(inode->i_sb); |
0ca1f7ce | 7473 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 7474 | if (!ret) { |
e41f941a | 7475 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
7476 | reserved = 1; |
7477 | } | |
56a76f82 NP |
7478 | if (ret) { |
7479 | if (ret == -ENOMEM) | |
7480 | ret = VM_FAULT_OOM; | |
7481 | else /* -ENOSPC, -EIO, etc */ | |
7482 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
7483 | if (reserved) |
7484 | goto out; | |
7485 | goto out_noreserve; | |
56a76f82 | 7486 | } |
1832a6d5 | 7487 | |
56a76f82 | 7488 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 7489 | again: |
9ebefb18 | 7490 | lock_page(page); |
9ebefb18 | 7491 | size = i_size_read(inode); |
e6dcd2dc CM |
7492 | page_start = page_offset(page); |
7493 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 7494 | |
9ebefb18 | 7495 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 7496 | (page_start >= size)) { |
9ebefb18 CM |
7497 | /* page got truncated out from underneath us */ |
7498 | goto out_unlock; | |
7499 | } | |
e6dcd2dc CM |
7500 | wait_on_page_writeback(page); |
7501 | ||
d0082371 | 7502 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
7503 | set_page_extent_mapped(page); |
7504 | ||
eb84ae03 CM |
7505 | /* |
7506 | * we can't set the delalloc bits if there are pending ordered | |
7507 | * extents. Drop our locks and wait for them to finish | |
7508 | */ | |
e6dcd2dc CM |
7509 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
7510 | if (ordered) { | |
2ac55d41 JB |
7511 | unlock_extent_cached(io_tree, page_start, page_end, |
7512 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 7513 | unlock_page(page); |
eb84ae03 | 7514 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
7515 | btrfs_put_ordered_extent(ordered); |
7516 | goto again; | |
7517 | } | |
7518 | ||
fbf19087 JB |
7519 | /* |
7520 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
7521 | * if it was already dirty, so for space accounting reasons we need to | |
7522 | * clear any delalloc bits for the range we are fixing to save. There | |
7523 | * is probably a better way to do this, but for now keep consistent with | |
7524 | * prepare_pages in the normal write path. | |
7525 | */ | |
2ac55d41 | 7526 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
7527 | EXTENT_DIRTY | EXTENT_DELALLOC | |
7528 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 7529 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 7530 | |
2ac55d41 JB |
7531 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
7532 | &cached_state); | |
9ed74f2d | 7533 | if (ret) { |
2ac55d41 JB |
7534 | unlock_extent_cached(io_tree, page_start, page_end, |
7535 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
7536 | ret = VM_FAULT_SIGBUS; |
7537 | goto out_unlock; | |
7538 | } | |
e6dcd2dc | 7539 | ret = 0; |
9ebefb18 CM |
7540 | |
7541 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 7542 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 7543 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 7544 | else |
e6dcd2dc | 7545 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 7546 | |
e6dcd2dc CM |
7547 | if (zero_start != PAGE_CACHE_SIZE) { |
7548 | kaddr = kmap(page); | |
7549 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
7550 | flush_dcache_page(page); | |
7551 | kunmap(page); | |
7552 | } | |
247e743c | 7553 | ClearPageChecked(page); |
e6dcd2dc | 7554 | set_page_dirty(page); |
50a9b214 | 7555 | SetPageUptodate(page); |
5a3f23d5 | 7556 | |
257c62e1 CM |
7557 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
7558 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 7559 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 7560 | |
2ac55d41 | 7561 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
7562 | |
7563 | out_unlock: | |
b2b5ef5c JK |
7564 | if (!ret) { |
7565 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 7566 | return VM_FAULT_LOCKED; |
b2b5ef5c | 7567 | } |
9ebefb18 | 7568 | unlock_page(page); |
1832a6d5 | 7569 | out: |
ec39e180 | 7570 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 7571 | out_noreserve: |
b2b5ef5c | 7572 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
7573 | return ret; |
7574 | } | |
7575 | ||
a41ad394 | 7576 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
7577 | { |
7578 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 7579 | struct btrfs_block_rsv *rsv; |
a71754fc | 7580 | int ret = 0; |
3893e33b | 7581 | int err = 0; |
39279cc3 | 7582 | struct btrfs_trans_handle *trans; |
dbe674a9 | 7583 | u64 mask = root->sectorsize - 1; |
07127184 | 7584 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 7585 | |
4a096752 | 7586 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 7587 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 7588 | |
fcb80c2a JB |
7589 | /* |
7590 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
7591 | * 3 things going on here | |
7592 | * | |
7593 | * 1) We need to reserve space for our orphan item and the space to | |
7594 | * delete our orphan item. Lord knows we don't want to have a dangling | |
7595 | * orphan item because we didn't reserve space to remove it. | |
7596 | * | |
7597 | * 2) We need to reserve space to update our inode. | |
7598 | * | |
7599 | * 3) We need to have something to cache all the space that is going to | |
7600 | * be free'd up by the truncate operation, but also have some slack | |
7601 | * space reserved in case it uses space during the truncate (thank you | |
7602 | * very much snapshotting). | |
7603 | * | |
7604 | * And we need these to all be seperate. The fact is we can use alot of | |
7605 | * space doing the truncate, and we have no earthly idea how much space | |
7606 | * we will use, so we need the truncate reservation to be seperate so it | |
7607 | * doesn't end up using space reserved for updating the inode or | |
7608 | * removing the orphan item. We also need to be able to stop the | |
7609 | * transaction and start a new one, which means we need to be able to | |
7610 | * update the inode several times, and we have no idea of knowing how | |
7611 | * many times that will be, so we can't just reserve 1 item for the | |
7612 | * entirety of the opration, so that has to be done seperately as well. | |
7613 | * Then there is the orphan item, which does indeed need to be held on | |
7614 | * to for the whole operation, and we need nobody to touch this reserved | |
7615 | * space except the orphan code. | |
7616 | * | |
7617 | * So that leaves us with | |
7618 | * | |
7619 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
7620 | * 2) rsv - for the truncate reservation, which we will steal from the | |
7621 | * transaction reservation. | |
7622 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
7623 | * updating the inode. | |
7624 | */ | |
66d8f3dd | 7625 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
7626 | if (!rsv) |
7627 | return -ENOMEM; | |
4a338542 | 7628 | rsv->size = min_size; |
ca7e70f5 | 7629 | rsv->failfast = 1; |
f0cd846e | 7630 | |
907cbceb | 7631 | /* |
07127184 | 7632 | * 1 for the truncate slack space |
907cbceb JB |
7633 | * 1 for updating the inode. |
7634 | */ | |
f3fe820c | 7635 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
7636 | if (IS_ERR(trans)) { |
7637 | err = PTR_ERR(trans); | |
7638 | goto out; | |
7639 | } | |
f0cd846e | 7640 | |
907cbceb JB |
7641 | /* Migrate the slack space for the truncate to our reserve */ |
7642 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
7643 | min_size); | |
fcb80c2a | 7644 | BUG_ON(ret); |
f0cd846e | 7645 | |
5a3f23d5 CM |
7646 | /* |
7647 | * setattr is responsible for setting the ordered_data_close flag, | |
7648 | * but that is only tested during the last file release. That | |
7649 | * could happen well after the next commit, leaving a great big | |
7650 | * window where new writes may get lost if someone chooses to write | |
7651 | * to this file after truncating to zero | |
7652 | * | |
7653 | * The inode doesn't have any dirty data here, and so if we commit | |
7654 | * this is a noop. If someone immediately starts writing to the inode | |
7655 | * it is very likely we'll catch some of their writes in this | |
7656 | * transaction, and the commit will find this file on the ordered | |
7657 | * data list with good things to send down. | |
7658 | * | |
7659 | * This is a best effort solution, there is still a window where | |
7660 | * using truncate to replace the contents of the file will | |
7661 | * end up with a zero length file after a crash. | |
7662 | */ | |
72ac3c0d JB |
7663 | if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
7664 | &BTRFS_I(inode)->runtime_flags)) | |
5a3f23d5 CM |
7665 | btrfs_add_ordered_operation(trans, root, inode); |
7666 | ||
5dc562c5 JB |
7667 | /* |
7668 | * So if we truncate and then write and fsync we normally would just | |
7669 | * write the extents that changed, which is a problem if we need to | |
7670 | * first truncate that entire inode. So set this flag so we write out | |
7671 | * all of the extents in the inode to the sync log so we're completely | |
7672 | * safe. | |
7673 | */ | |
7674 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 7675 | trans->block_rsv = rsv; |
907cbceb | 7676 | |
8082510e YZ |
7677 | while (1) { |
7678 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
7679 | inode->i_size, | |
7680 | BTRFS_EXTENT_DATA_KEY); | |
ca7e70f5 | 7681 | if (ret != -ENOSPC) { |
3893e33b | 7682 | err = ret; |
8082510e | 7683 | break; |
3893e33b | 7684 | } |
39279cc3 | 7685 | |
fcb80c2a | 7686 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 7687 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
7688 | if (ret) { |
7689 | err = ret; | |
7690 | break; | |
7691 | } | |
ca7e70f5 | 7692 | |
8082510e | 7693 | btrfs_end_transaction(trans, root); |
b53d3f5d | 7694 | btrfs_btree_balance_dirty(root); |
ca7e70f5 JB |
7695 | |
7696 | trans = btrfs_start_transaction(root, 2); | |
7697 | if (IS_ERR(trans)) { | |
7698 | ret = err = PTR_ERR(trans); | |
7699 | trans = NULL; | |
7700 | break; | |
7701 | } | |
7702 | ||
7703 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
7704 | rsv, min_size); | |
7705 | BUG_ON(ret); /* shouldn't happen */ | |
7706 | trans->block_rsv = rsv; | |
8082510e YZ |
7707 | } |
7708 | ||
7709 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 7710 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 7711 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
7712 | if (ret) |
7713 | err = ret; | |
8082510e YZ |
7714 | } |
7715 | ||
917c16b2 CM |
7716 | if (trans) { |
7717 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
7718 | ret = btrfs_update_inode(trans, root, inode); | |
7719 | if (ret && !err) | |
7720 | err = ret; | |
7b128766 | 7721 | |
7ad85bb7 | 7722 | ret = btrfs_end_transaction(trans, root); |
b53d3f5d | 7723 | btrfs_btree_balance_dirty(root); |
917c16b2 | 7724 | } |
fcb80c2a JB |
7725 | |
7726 | out: | |
7727 | btrfs_free_block_rsv(root, rsv); | |
7728 | ||
3893e33b JB |
7729 | if (ret && !err) |
7730 | err = ret; | |
a41ad394 | 7731 | |
3893e33b | 7732 | return err; |
39279cc3 CM |
7733 | } |
7734 | ||
d352ac68 CM |
7735 | /* |
7736 | * create a new subvolume directory/inode (helper for the ioctl). | |
7737 | */ | |
d2fb3437 | 7738 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
d82a6f1d | 7739 | struct btrfs_root *new_root, u64 new_dirid) |
39279cc3 | 7740 | { |
39279cc3 | 7741 | struct inode *inode; |
76dda93c | 7742 | int err; |
00e4e6b3 | 7743 | u64 index = 0; |
39279cc3 | 7744 | |
12fc9d09 FA |
7745 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
7746 | new_dirid, new_dirid, | |
7747 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
7748 | &index); | |
54aa1f4d | 7749 | if (IS_ERR(inode)) |
f46b5a66 | 7750 | return PTR_ERR(inode); |
39279cc3 CM |
7751 | inode->i_op = &btrfs_dir_inode_operations; |
7752 | inode->i_fop = &btrfs_dir_file_operations; | |
7753 | ||
bfe86848 | 7754 | set_nlink(inode, 1); |
dbe674a9 | 7755 | btrfs_i_size_write(inode, 0); |
3b96362c | 7756 | |
76dda93c | 7757 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 7758 | |
76dda93c | 7759 | iput(inode); |
ce598979 | 7760 | return err; |
39279cc3 CM |
7761 | } |
7762 | ||
39279cc3 CM |
7763 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
7764 | { | |
7765 | struct btrfs_inode *ei; | |
2ead6ae7 | 7766 | struct inode *inode; |
39279cc3 CM |
7767 | |
7768 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
7769 | if (!ei) | |
7770 | return NULL; | |
2ead6ae7 YZ |
7771 | |
7772 | ei->root = NULL; | |
2ead6ae7 | 7773 | ei->generation = 0; |
15ee9bc7 | 7774 | ei->last_trans = 0; |
257c62e1 | 7775 | ei->last_sub_trans = 0; |
e02119d5 | 7776 | ei->logged_trans = 0; |
2ead6ae7 | 7777 | ei->delalloc_bytes = 0; |
2ead6ae7 YZ |
7778 | ei->disk_i_size = 0; |
7779 | ei->flags = 0; | |
7709cde3 | 7780 | ei->csum_bytes = 0; |
2ead6ae7 YZ |
7781 | ei->index_cnt = (u64)-1; |
7782 | ei->last_unlink_trans = 0; | |
46d8bc34 | 7783 | ei->last_log_commit = 0; |
2ead6ae7 | 7784 | |
9e0baf60 JB |
7785 | spin_lock_init(&ei->lock); |
7786 | ei->outstanding_extents = 0; | |
7787 | ei->reserved_extents = 0; | |
2ead6ae7 | 7788 | |
72ac3c0d | 7789 | ei->runtime_flags = 0; |
261507a0 | 7790 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 7791 | |
16cdcec7 MX |
7792 | ei->delayed_node = NULL; |
7793 | ||
2ead6ae7 | 7794 | inode = &ei->vfs_inode; |
a8067e02 | 7795 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
7796 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
7797 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
7798 | ei->io_tree.track_uptodate = 1; |
7799 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 7800 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 7801 | mutex_init(&ei->log_mutex); |
f248679e | 7802 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 7803 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 7804 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 7805 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
7806 | RB_CLEAR_NODE(&ei->rb_node); |
7807 | ||
7808 | return inode; | |
39279cc3 CM |
7809 | } |
7810 | ||
fa0d7e3d NP |
7811 | static void btrfs_i_callback(struct rcu_head *head) |
7812 | { | |
7813 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
7814 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
7815 | } | |
7816 | ||
39279cc3 CM |
7817 | void btrfs_destroy_inode(struct inode *inode) |
7818 | { | |
e6dcd2dc | 7819 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
7820 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7821 | ||
b3d9b7a3 | 7822 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 7823 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
7824 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7825 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
7826 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
7827 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
39279cc3 | 7828 | |
a6dbd429 JB |
7829 | /* |
7830 | * This can happen where we create an inode, but somebody else also | |
7831 | * created the same inode and we need to destroy the one we already | |
7832 | * created. | |
7833 | */ | |
7834 | if (!root) | |
7835 | goto free; | |
7836 | ||
5a3f23d5 CM |
7837 | /* |
7838 | * Make sure we're properly removed from the ordered operation | |
7839 | * lists. | |
7840 | */ | |
7841 | smp_mb(); | |
7842 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
199c2a9c | 7843 | spin_lock(&root->fs_info->ordered_root_lock); |
5a3f23d5 | 7844 | list_del_init(&BTRFS_I(inode)->ordered_operations); |
199c2a9c | 7845 | spin_unlock(&root->fs_info->ordered_root_lock); |
5a3f23d5 CM |
7846 | } |
7847 | ||
8a35d95f JB |
7848 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
7849 | &BTRFS_I(inode)->runtime_flags)) { | |
c2cf52eb SK |
7850 | btrfs_info(root->fs_info, "inode %llu still on the orphan list", |
7851 | (unsigned long long)btrfs_ino(inode)); | |
8a35d95f | 7852 | atomic_dec(&root->orphan_inodes); |
7b128766 | 7853 | } |
7b128766 | 7854 | |
d397712b | 7855 | while (1) { |
e6dcd2dc CM |
7856 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
7857 | if (!ordered) | |
7858 | break; | |
7859 | else { | |
c2cf52eb SK |
7860 | btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup", |
7861 | (unsigned long long)ordered->file_offset, | |
7862 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
7863 | btrfs_remove_ordered_extent(inode, ordered); |
7864 | btrfs_put_ordered_extent(ordered); | |
7865 | btrfs_put_ordered_extent(ordered); | |
7866 | } | |
7867 | } | |
5d4f98a2 | 7868 | inode_tree_del(inode); |
5b21f2ed | 7869 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 7870 | free: |
fa0d7e3d | 7871 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
7872 | } |
7873 | ||
45321ac5 | 7874 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
7875 | { |
7876 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 7877 | |
6379ef9f NA |
7878 | if (root == NULL) |
7879 | return 1; | |
7880 | ||
fa6ac876 | 7881 | /* the snap/subvol tree is on deleting */ |
0af3d00b | 7882 | if (btrfs_root_refs(&root->root_item) == 0 && |
fa6ac876 | 7883 | root != root->fs_info->tree_root) |
45321ac5 | 7884 | return 1; |
76dda93c | 7885 | else |
45321ac5 | 7886 | return generic_drop_inode(inode); |
76dda93c YZ |
7887 | } |
7888 | ||
0ee0fda0 | 7889 | static void init_once(void *foo) |
39279cc3 CM |
7890 | { |
7891 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
7892 | ||
7893 | inode_init_once(&ei->vfs_inode); | |
7894 | } | |
7895 | ||
7896 | void btrfs_destroy_cachep(void) | |
7897 | { | |
8c0a8537 KS |
7898 | /* |
7899 | * Make sure all delayed rcu free inodes are flushed before we | |
7900 | * destroy cache. | |
7901 | */ | |
7902 | rcu_barrier(); | |
39279cc3 CM |
7903 | if (btrfs_inode_cachep) |
7904 | kmem_cache_destroy(btrfs_inode_cachep); | |
7905 | if (btrfs_trans_handle_cachep) | |
7906 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
7907 | if (btrfs_transaction_cachep) | |
7908 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
7909 | if (btrfs_path_cachep) |
7910 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
7911 | if (btrfs_free_space_cachep) |
7912 | kmem_cache_destroy(btrfs_free_space_cachep); | |
8ccf6f19 MX |
7913 | if (btrfs_delalloc_work_cachep) |
7914 | kmem_cache_destroy(btrfs_delalloc_work_cachep); | |
39279cc3 CM |
7915 | } |
7916 | ||
7917 | int btrfs_init_cachep(void) | |
7918 | { | |
837e1972 | 7919 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 CH |
7920 | sizeof(struct btrfs_inode), 0, |
7921 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
7922 | if (!btrfs_inode_cachep) |
7923 | goto fail; | |
9601e3f6 | 7924 | |
837e1972 | 7925 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
7926 | sizeof(struct btrfs_trans_handle), 0, |
7927 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7928 | if (!btrfs_trans_handle_cachep) |
7929 | goto fail; | |
9601e3f6 | 7930 | |
837e1972 | 7931 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
7932 | sizeof(struct btrfs_transaction), 0, |
7933 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7934 | if (!btrfs_transaction_cachep) |
7935 | goto fail; | |
9601e3f6 | 7936 | |
837e1972 | 7937 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
7938 | sizeof(struct btrfs_path), 0, |
7939 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7940 | if (!btrfs_path_cachep) |
7941 | goto fail; | |
9601e3f6 | 7942 | |
837e1972 | 7943 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
7944 | sizeof(struct btrfs_free_space), 0, |
7945 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
7946 | if (!btrfs_free_space_cachep) | |
7947 | goto fail; | |
7948 | ||
8ccf6f19 MX |
7949 | btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work", |
7950 | sizeof(struct btrfs_delalloc_work), 0, | |
7951 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
7952 | NULL); | |
7953 | if (!btrfs_delalloc_work_cachep) | |
7954 | goto fail; | |
7955 | ||
39279cc3 CM |
7956 | return 0; |
7957 | fail: | |
7958 | btrfs_destroy_cachep(); | |
7959 | return -ENOMEM; | |
7960 | } | |
7961 | ||
7962 | static int btrfs_getattr(struct vfsmount *mnt, | |
7963 | struct dentry *dentry, struct kstat *stat) | |
7964 | { | |
df0af1a5 | 7965 | u64 delalloc_bytes; |
39279cc3 | 7966 | struct inode *inode = dentry->d_inode; |
fadc0d8b DS |
7967 | u32 blocksize = inode->i_sb->s_blocksize; |
7968 | ||
39279cc3 | 7969 | generic_fillattr(inode, stat); |
0ee5dc67 | 7970 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 7971 | stat->blksize = PAGE_CACHE_SIZE; |
df0af1a5 MX |
7972 | |
7973 | spin_lock(&BTRFS_I(inode)->lock); | |
7974 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
7975 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 7976 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 7977 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
7978 | return 0; |
7979 | } | |
7980 | ||
d397712b CM |
7981 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
7982 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
7983 | { |
7984 | struct btrfs_trans_handle *trans; | |
7985 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 7986 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
7987 | struct inode *new_inode = new_dentry->d_inode; |
7988 | struct inode *old_inode = old_dentry->d_inode; | |
7989 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 7990 | u64 index = 0; |
4df27c4d | 7991 | u64 root_objectid; |
39279cc3 | 7992 | int ret; |
33345d01 | 7993 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 7994 | |
33345d01 | 7995 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
7996 | return -EPERM; |
7997 | ||
4df27c4d | 7998 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 7999 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
8000 | return -EXDEV; |
8001 | ||
33345d01 LZ |
8002 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
8003 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 8004 | return -ENOTEMPTY; |
5f39d397 | 8005 | |
4df27c4d YZ |
8006 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
8007 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
8008 | return -ENOTEMPTY; | |
9c52057c CM |
8009 | |
8010 | ||
8011 | /* check for collisions, even if the name isn't there */ | |
8012 | ret = btrfs_check_dir_item_collision(root, new_dir->i_ino, | |
8013 | new_dentry->d_name.name, | |
8014 | new_dentry->d_name.len); | |
8015 | ||
8016 | if (ret) { | |
8017 | if (ret == -EEXIST) { | |
8018 | /* we shouldn't get | |
8019 | * eexist without a new_inode */ | |
8020 | if (!new_inode) { | |
8021 | WARN_ON(1); | |
8022 | return ret; | |
8023 | } | |
8024 | } else { | |
8025 | /* maybe -EOVERFLOW */ | |
8026 | return ret; | |
8027 | } | |
8028 | } | |
8029 | ret = 0; | |
8030 | ||
5a3f23d5 CM |
8031 | /* |
8032 | * we're using rename to replace one file with another. | |
8033 | * and the replacement file is large. Start IO on it now so | |
8034 | * we don't add too much work to the end of the transaction | |
8035 | */ | |
4baf8c92 | 8036 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
8037 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
8038 | filemap_flush(old_inode->i_mapping); | |
8039 | ||
76dda93c | 8040 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 8041 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 8042 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
8043 | /* |
8044 | * We want to reserve the absolute worst case amount of items. So if | |
8045 | * both inodes are subvols and we need to unlink them then that would | |
8046 | * require 4 item modifications, but if they are both normal inodes it | |
8047 | * would require 5 item modifications, so we'll assume their normal | |
8048 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
8049 | * should cover the worst case number of items we'll modify. | |
8050 | */ | |
6e137ed3 | 8051 | trans = btrfs_start_transaction(root, 11); |
b44c59a8 JL |
8052 | if (IS_ERR(trans)) { |
8053 | ret = PTR_ERR(trans); | |
8054 | goto out_notrans; | |
8055 | } | |
76dda93c | 8056 | |
4df27c4d YZ |
8057 | if (dest != root) |
8058 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 8059 | |
a5719521 YZ |
8060 | ret = btrfs_set_inode_index(new_dir, &index); |
8061 | if (ret) | |
8062 | goto out_fail; | |
5a3f23d5 | 8063 | |
33345d01 | 8064 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
8065 | /* force full log commit if subvolume involved. */ |
8066 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
8067 | } else { | |
a5719521 YZ |
8068 | ret = btrfs_insert_inode_ref(trans, dest, |
8069 | new_dentry->d_name.name, | |
8070 | new_dentry->d_name.len, | |
33345d01 LZ |
8071 | old_ino, |
8072 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
8073 | if (ret) |
8074 | goto out_fail; | |
4df27c4d YZ |
8075 | /* |
8076 | * this is an ugly little race, but the rename is required | |
8077 | * to make sure that if we crash, the inode is either at the | |
8078 | * old name or the new one. pinning the log transaction lets | |
8079 | * us make sure we don't allow a log commit to come in after | |
8080 | * we unlink the name but before we add the new name back in. | |
8081 | */ | |
8082 | btrfs_pin_log_trans(root); | |
8083 | } | |
5a3f23d5 CM |
8084 | /* |
8085 | * make sure the inode gets flushed if it is replacing | |
8086 | * something. | |
8087 | */ | |
33345d01 | 8088 | if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode)) |
5a3f23d5 | 8089 | btrfs_add_ordered_operation(trans, root, old_inode); |
5a3f23d5 | 8090 | |
0c4d2d95 JB |
8091 | inode_inc_iversion(old_dir); |
8092 | inode_inc_iversion(new_dir); | |
8093 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
8094 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
8095 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
8096 | old_inode->i_ctime = ctime; | |
5f39d397 | 8097 | |
12fcfd22 CM |
8098 | if (old_dentry->d_parent != new_dentry->d_parent) |
8099 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
8100 | ||
33345d01 | 8101 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
8102 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
8103 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
8104 | old_dentry->d_name.name, | |
8105 | old_dentry->d_name.len); | |
8106 | } else { | |
92986796 AV |
8107 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
8108 | old_dentry->d_inode, | |
8109 | old_dentry->d_name.name, | |
8110 | old_dentry->d_name.len); | |
8111 | if (!ret) | |
8112 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 8113 | } |
79787eaa JM |
8114 | if (ret) { |
8115 | btrfs_abort_transaction(trans, root, ret); | |
8116 | goto out_fail; | |
8117 | } | |
39279cc3 CM |
8118 | |
8119 | if (new_inode) { | |
0c4d2d95 | 8120 | inode_inc_iversion(new_inode); |
39279cc3 | 8121 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 8122 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
8123 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
8124 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
8125 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
8126 | root_objectid, | |
8127 | new_dentry->d_name.name, | |
8128 | new_dentry->d_name.len); | |
8129 | BUG_ON(new_inode->i_nlink == 0); | |
8130 | } else { | |
8131 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
8132 | new_dentry->d_inode, | |
8133 | new_dentry->d_name.name, | |
8134 | new_dentry->d_name.len); | |
8135 | } | |
4ef31a45 | 8136 | if (!ret && new_inode->i_nlink == 0) |
e02119d5 | 8137 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
79787eaa JM |
8138 | if (ret) { |
8139 | btrfs_abort_transaction(trans, root, ret); | |
8140 | goto out_fail; | |
8141 | } | |
39279cc3 | 8142 | } |
aec7477b | 8143 | |
4df27c4d YZ |
8144 | ret = btrfs_add_link(trans, new_dir, old_inode, |
8145 | new_dentry->d_name.name, | |
a5719521 | 8146 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
8147 | if (ret) { |
8148 | btrfs_abort_transaction(trans, root, ret); | |
8149 | goto out_fail; | |
8150 | } | |
39279cc3 | 8151 | |
33345d01 | 8152 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 8153 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 8154 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
8155 | btrfs_end_log_trans(root); |
8156 | } | |
39279cc3 | 8157 | out_fail: |
7ad85bb7 | 8158 | btrfs_end_transaction(trans, root); |
b44c59a8 | 8159 | out_notrans: |
33345d01 | 8160 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 8161 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 8162 | |
39279cc3 CM |
8163 | return ret; |
8164 | } | |
8165 | ||
8ccf6f19 MX |
8166 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
8167 | { | |
8168 | struct btrfs_delalloc_work *delalloc_work; | |
8169 | ||
8170 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
8171 | work); | |
8172 | if (delalloc_work->wait) | |
8173 | btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1); | |
8174 | else | |
8175 | filemap_flush(delalloc_work->inode->i_mapping); | |
8176 | ||
8177 | if (delalloc_work->delay_iput) | |
8178 | btrfs_add_delayed_iput(delalloc_work->inode); | |
8179 | else | |
8180 | iput(delalloc_work->inode); | |
8181 | complete(&delalloc_work->completion); | |
8182 | } | |
8183 | ||
8184 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
8185 | int wait, int delay_iput) | |
8186 | { | |
8187 | struct btrfs_delalloc_work *work; | |
8188 | ||
8189 | work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS); | |
8190 | if (!work) | |
8191 | return NULL; | |
8192 | ||
8193 | init_completion(&work->completion); | |
8194 | INIT_LIST_HEAD(&work->list); | |
8195 | work->inode = inode; | |
8196 | work->wait = wait; | |
8197 | work->delay_iput = delay_iput; | |
8198 | work->work.func = btrfs_run_delalloc_work; | |
8199 | ||
8200 | return work; | |
8201 | } | |
8202 | ||
8203 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
8204 | { | |
8205 | wait_for_completion(&work->completion); | |
8206 | kmem_cache_free(btrfs_delalloc_work_cachep, work); | |
8207 | } | |
8208 | ||
d352ac68 CM |
8209 | /* |
8210 | * some fairly slow code that needs optimization. This walks the list | |
8211 | * of all the inodes with pending delalloc and forces them to disk. | |
8212 | */ | |
eb73c1b7 | 8213 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 | 8214 | { |
ea8c2819 | 8215 | struct btrfs_inode *binode; |
5b21f2ed | 8216 | struct inode *inode; |
8ccf6f19 MX |
8217 | struct btrfs_delalloc_work *work, *next; |
8218 | struct list_head works; | |
1eafa6c7 | 8219 | struct list_head splice; |
8ccf6f19 | 8220 | int ret = 0; |
ea8c2819 | 8221 | |
8ccf6f19 | 8222 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 8223 | INIT_LIST_HEAD(&splice); |
63607cc8 | 8224 | |
eb73c1b7 MX |
8225 | spin_lock(&root->delalloc_lock); |
8226 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
8227 | while (!list_empty(&splice)) { |
8228 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 8229 | delalloc_inodes); |
1eafa6c7 | 8230 | |
eb73c1b7 MX |
8231 | list_move_tail(&binode->delalloc_inodes, |
8232 | &root->delalloc_inodes); | |
5b21f2ed | 8233 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 8234 | if (!inode) { |
eb73c1b7 | 8235 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 8236 | continue; |
df0af1a5 | 8237 | } |
eb73c1b7 | 8238 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 MX |
8239 | |
8240 | work = btrfs_alloc_delalloc_work(inode, 0, delay_iput); | |
8241 | if (unlikely(!work)) { | |
8242 | ret = -ENOMEM; | |
8243 | goto out; | |
5b21f2ed | 8244 | } |
1eafa6c7 MX |
8245 | list_add_tail(&work->list, &works); |
8246 | btrfs_queue_worker(&root->fs_info->flush_workers, | |
8247 | &work->work); | |
8248 | ||
5b21f2ed | 8249 | cond_resched(); |
eb73c1b7 | 8250 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 8251 | } |
eb73c1b7 | 8252 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 8253 | |
1eafa6c7 MX |
8254 | list_for_each_entry_safe(work, next, &works, list) { |
8255 | list_del_init(&work->list); | |
8256 | btrfs_wait_and_free_delalloc_work(work); | |
8257 | } | |
eb73c1b7 MX |
8258 | return 0; |
8259 | out: | |
8260 | list_for_each_entry_safe(work, next, &works, list) { | |
8261 | list_del_init(&work->list); | |
8262 | btrfs_wait_and_free_delalloc_work(work); | |
8263 | } | |
8264 | ||
8265 | if (!list_empty_careful(&splice)) { | |
8266 | spin_lock(&root->delalloc_lock); | |
8267 | list_splice_tail(&splice, &root->delalloc_inodes); | |
8268 | spin_unlock(&root->delalloc_lock); | |
8269 | } | |
8270 | return ret; | |
8271 | } | |
1eafa6c7 | 8272 | |
eb73c1b7 MX |
8273 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
8274 | { | |
8275 | int ret; | |
1eafa6c7 | 8276 | |
eb73c1b7 MX |
8277 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
8278 | return -EROFS; | |
8279 | ||
8280 | ret = __start_delalloc_inodes(root, delay_iput); | |
8281 | /* | |
8282 | * the filemap_flush will queue IO into the worker threads, but | |
8c8bee1d CM |
8283 | * we have to make sure the IO is actually started and that |
8284 | * ordered extents get created before we return | |
8285 | */ | |
8286 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 8287 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 8288 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 8289 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
8290 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
8291 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
8292 | } |
8293 | atomic_dec(&root->fs_info->async_submit_draining); | |
eb73c1b7 MX |
8294 | return ret; |
8295 | } | |
8296 | ||
8297 | int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info *fs_info, | |
8298 | int delay_iput) | |
8299 | { | |
8300 | struct btrfs_root *root; | |
8301 | struct list_head splice; | |
8302 | int ret; | |
8303 | ||
8304 | if (fs_info->sb->s_flags & MS_RDONLY) | |
8305 | return -EROFS; | |
8306 | ||
8307 | INIT_LIST_HEAD(&splice); | |
8308 | ||
8309 | spin_lock(&fs_info->delalloc_root_lock); | |
8310 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
8311 | while (!list_empty(&splice)) { | |
8312 | root = list_first_entry(&splice, struct btrfs_root, | |
8313 | delalloc_root); | |
8314 | root = btrfs_grab_fs_root(root); | |
8315 | BUG_ON(!root); | |
8316 | list_move_tail(&root->delalloc_root, | |
8317 | &fs_info->delalloc_roots); | |
8318 | spin_unlock(&fs_info->delalloc_root_lock); | |
8319 | ||
8320 | ret = __start_delalloc_inodes(root, delay_iput); | |
8321 | btrfs_put_fs_root(root); | |
8322 | if (ret) | |
8323 | goto out; | |
8324 | ||
8325 | spin_lock(&fs_info->delalloc_root_lock); | |
8ccf6f19 | 8326 | } |
eb73c1b7 | 8327 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 8328 | |
eb73c1b7 MX |
8329 | atomic_inc(&fs_info->async_submit_draining); |
8330 | while (atomic_read(&fs_info->nr_async_submits) || | |
8331 | atomic_read(&fs_info->async_delalloc_pages)) { | |
8332 | wait_event(fs_info->async_submit_wait, | |
8333 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
8334 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
8335 | } | |
8336 | atomic_dec(&fs_info->async_submit_draining); | |
8337 | return 0; | |
8338 | out: | |
1eafa6c7 | 8339 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
8340 | spin_lock(&fs_info->delalloc_root_lock); |
8341 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
8342 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 8343 | } |
8ccf6f19 | 8344 | return ret; |
ea8c2819 CM |
8345 | } |
8346 | ||
39279cc3 CM |
8347 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
8348 | const char *symname) | |
8349 | { | |
8350 | struct btrfs_trans_handle *trans; | |
8351 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
8352 | struct btrfs_path *path; | |
8353 | struct btrfs_key key; | |
1832a6d5 | 8354 | struct inode *inode = NULL; |
39279cc3 CM |
8355 | int err; |
8356 | int drop_inode = 0; | |
8357 | u64 objectid; | |
00e4e6b3 | 8358 | u64 index = 0 ; |
39279cc3 CM |
8359 | int name_len; |
8360 | int datasize; | |
5f39d397 | 8361 | unsigned long ptr; |
39279cc3 | 8362 | struct btrfs_file_extent_item *ei; |
5f39d397 | 8363 | struct extent_buffer *leaf; |
39279cc3 CM |
8364 | |
8365 | name_len = strlen(symname) + 1; | |
8366 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
8367 | return -ENAMETOOLONG; | |
1832a6d5 | 8368 | |
9ed74f2d JB |
8369 | /* |
8370 | * 2 items for inode item and ref | |
8371 | * 2 items for dir items | |
8372 | * 1 item for xattr if selinux is on | |
8373 | */ | |
a22285a6 YZ |
8374 | trans = btrfs_start_transaction(root, 5); |
8375 | if (IS_ERR(trans)) | |
8376 | return PTR_ERR(trans); | |
1832a6d5 | 8377 | |
581bb050 LZ |
8378 | err = btrfs_find_free_ino(root, &objectid); |
8379 | if (err) | |
8380 | goto out_unlock; | |
8381 | ||
aec7477b | 8382 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 8383 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 8384 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
8385 | if (IS_ERR(inode)) { |
8386 | err = PTR_ERR(inode); | |
39279cc3 | 8387 | goto out_unlock; |
7cf96da3 | 8388 | } |
39279cc3 | 8389 | |
2a7dba39 | 8390 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
8391 | if (err) { |
8392 | drop_inode = 1; | |
8393 | goto out_unlock; | |
8394 | } | |
8395 | ||
ad19db71 CS |
8396 | /* |
8397 | * If the active LSM wants to access the inode during | |
8398 | * d_instantiate it needs these. Smack checks to see | |
8399 | * if the filesystem supports xattrs by looking at the | |
8400 | * ops vector. | |
8401 | */ | |
8402 | inode->i_fop = &btrfs_file_operations; | |
8403 | inode->i_op = &btrfs_file_inode_operations; | |
8404 | ||
a1b075d2 | 8405 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
8406 | if (err) |
8407 | drop_inode = 1; | |
8408 | else { | |
8409 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 8410 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 8411 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 8412 | } |
39279cc3 CM |
8413 | if (drop_inode) |
8414 | goto out_unlock; | |
8415 | ||
8416 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
8417 | if (!path) { |
8418 | err = -ENOMEM; | |
8419 | drop_inode = 1; | |
8420 | goto out_unlock; | |
8421 | } | |
33345d01 | 8422 | key.objectid = btrfs_ino(inode); |
39279cc3 | 8423 | key.offset = 0; |
39279cc3 CM |
8424 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
8425 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
8426 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
8427 | datasize); | |
54aa1f4d CM |
8428 | if (err) { |
8429 | drop_inode = 1; | |
b0839166 | 8430 | btrfs_free_path(path); |
54aa1f4d CM |
8431 | goto out_unlock; |
8432 | } | |
5f39d397 CM |
8433 | leaf = path->nodes[0]; |
8434 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
8435 | struct btrfs_file_extent_item); | |
8436 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
8437 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 8438 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
8439 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
8440 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
8441 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
8442 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
8443 | ||
39279cc3 | 8444 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
8445 | write_extent_buffer(leaf, symname, ptr, name_len); |
8446 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 8447 | btrfs_free_path(path); |
5f39d397 | 8448 | |
39279cc3 CM |
8449 | inode->i_op = &btrfs_symlink_inode_operations; |
8450 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 8451 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 8452 | inode_set_bytes(inode, name_len); |
dbe674a9 | 8453 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
8454 | err = btrfs_update_inode(trans, root, inode); |
8455 | if (err) | |
8456 | drop_inode = 1; | |
39279cc3 CM |
8457 | |
8458 | out_unlock: | |
08c422c2 AV |
8459 | if (!err) |
8460 | d_instantiate(dentry, inode); | |
7ad85bb7 | 8461 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
8462 | if (drop_inode) { |
8463 | inode_dec_link_count(inode); | |
8464 | iput(inode); | |
8465 | } | |
b53d3f5d | 8466 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
8467 | return err; |
8468 | } | |
16432985 | 8469 | |
0af3d00b JB |
8470 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
8471 | u64 start, u64 num_bytes, u64 min_size, | |
8472 | loff_t actual_len, u64 *alloc_hint, | |
8473 | struct btrfs_trans_handle *trans) | |
d899e052 | 8474 | { |
5dc562c5 JB |
8475 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
8476 | struct extent_map *em; | |
d899e052 YZ |
8477 | struct btrfs_root *root = BTRFS_I(inode)->root; |
8478 | struct btrfs_key ins; | |
d899e052 | 8479 | u64 cur_offset = start; |
55a61d1d | 8480 | u64 i_size; |
154ea289 | 8481 | u64 cur_bytes; |
d899e052 | 8482 | int ret = 0; |
0af3d00b | 8483 | bool own_trans = true; |
d899e052 | 8484 | |
0af3d00b JB |
8485 | if (trans) |
8486 | own_trans = false; | |
d899e052 | 8487 | while (num_bytes > 0) { |
0af3d00b JB |
8488 | if (own_trans) { |
8489 | trans = btrfs_start_transaction(root, 3); | |
8490 | if (IS_ERR(trans)) { | |
8491 | ret = PTR_ERR(trans); | |
8492 | break; | |
8493 | } | |
5a303d5d YZ |
8494 | } |
8495 | ||
154ea289 CM |
8496 | cur_bytes = min(num_bytes, 256ULL * 1024 * 1024); |
8497 | cur_bytes = max(cur_bytes, min_size); | |
8498 | ret = btrfs_reserve_extent(trans, root, cur_bytes, | |
24542bf7 | 8499 | min_size, 0, *alloc_hint, &ins, 1); |
5a303d5d | 8500 | if (ret) { |
0af3d00b JB |
8501 | if (own_trans) |
8502 | btrfs_end_transaction(trans, root); | |
a22285a6 | 8503 | break; |
d899e052 | 8504 | } |
5a303d5d | 8505 | |
d899e052 YZ |
8506 | ret = insert_reserved_file_extent(trans, inode, |
8507 | cur_offset, ins.objectid, | |
8508 | ins.offset, ins.offset, | |
920bbbfb | 8509 | ins.offset, 0, 0, 0, |
d899e052 | 8510 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa JM |
8511 | if (ret) { |
8512 | btrfs_abort_transaction(trans, root, ret); | |
8513 | if (own_trans) | |
8514 | btrfs_end_transaction(trans, root); | |
8515 | break; | |
8516 | } | |
a1ed835e CM |
8517 | btrfs_drop_extent_cache(inode, cur_offset, |
8518 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 8519 | |
5dc562c5 JB |
8520 | em = alloc_extent_map(); |
8521 | if (!em) { | |
8522 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
8523 | &BTRFS_I(inode)->runtime_flags); | |
8524 | goto next; | |
8525 | } | |
8526 | ||
8527 | em->start = cur_offset; | |
8528 | em->orig_start = cur_offset; | |
8529 | em->len = ins.offset; | |
8530 | em->block_start = ins.objectid; | |
8531 | em->block_len = ins.offset; | |
b4939680 | 8532 | em->orig_block_len = ins.offset; |
cc95bef6 | 8533 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
8534 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
8535 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
8536 | em->generation = trans->transid; | |
8537 | ||
8538 | while (1) { | |
8539 | write_lock(&em_tree->lock); | |
09a2a8f9 | 8540 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
8541 | write_unlock(&em_tree->lock); |
8542 | if (ret != -EEXIST) | |
8543 | break; | |
8544 | btrfs_drop_extent_cache(inode, cur_offset, | |
8545 | cur_offset + ins.offset - 1, | |
8546 | 0); | |
8547 | } | |
8548 | free_extent_map(em); | |
8549 | next: | |
d899e052 YZ |
8550 | num_bytes -= ins.offset; |
8551 | cur_offset += ins.offset; | |
efa56464 | 8552 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 8553 | |
0c4d2d95 | 8554 | inode_inc_iversion(inode); |
d899e052 | 8555 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 8556 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 8557 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
8558 | (actual_len > inode->i_size) && |
8559 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 8560 | if (cur_offset > actual_len) |
55a61d1d | 8561 | i_size = actual_len; |
d1ea6a61 | 8562 | else |
55a61d1d JB |
8563 | i_size = cur_offset; |
8564 | i_size_write(inode, i_size); | |
8565 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
8566 | } |
8567 | ||
d899e052 | 8568 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
8569 | |
8570 | if (ret) { | |
8571 | btrfs_abort_transaction(trans, root, ret); | |
8572 | if (own_trans) | |
8573 | btrfs_end_transaction(trans, root); | |
8574 | break; | |
8575 | } | |
d899e052 | 8576 | |
0af3d00b JB |
8577 | if (own_trans) |
8578 | btrfs_end_transaction(trans, root); | |
5a303d5d | 8579 | } |
d899e052 YZ |
8580 | return ret; |
8581 | } | |
8582 | ||
0af3d00b JB |
8583 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
8584 | u64 start, u64 num_bytes, u64 min_size, | |
8585 | loff_t actual_len, u64 *alloc_hint) | |
8586 | { | |
8587 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
8588 | min_size, actual_len, alloc_hint, | |
8589 | NULL); | |
8590 | } | |
8591 | ||
8592 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
8593 | struct btrfs_trans_handle *trans, int mode, | |
8594 | u64 start, u64 num_bytes, u64 min_size, | |
8595 | loff_t actual_len, u64 *alloc_hint) | |
8596 | { | |
8597 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
8598 | min_size, actual_len, alloc_hint, trans); | |
8599 | } | |
8600 | ||
e6dcd2dc CM |
8601 | static int btrfs_set_page_dirty(struct page *page) |
8602 | { | |
e6dcd2dc CM |
8603 | return __set_page_dirty_nobuffers(page); |
8604 | } | |
8605 | ||
10556cb2 | 8606 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 8607 | { |
b83cc969 | 8608 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 8609 | umode_t mode = inode->i_mode; |
b83cc969 | 8610 | |
cb6db4e5 JM |
8611 | if (mask & MAY_WRITE && |
8612 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
8613 | if (btrfs_root_readonly(root)) | |
8614 | return -EROFS; | |
8615 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
8616 | return -EACCES; | |
8617 | } | |
2830ba7f | 8618 | return generic_permission(inode, mask); |
fdebe2bd | 8619 | } |
39279cc3 | 8620 | |
6e1d5dcc | 8621 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 8622 | .getattr = btrfs_getattr, |
39279cc3 CM |
8623 | .lookup = btrfs_lookup, |
8624 | .create = btrfs_create, | |
8625 | .unlink = btrfs_unlink, | |
8626 | .link = btrfs_link, | |
8627 | .mkdir = btrfs_mkdir, | |
8628 | .rmdir = btrfs_rmdir, | |
8629 | .rename = btrfs_rename, | |
8630 | .symlink = btrfs_symlink, | |
8631 | .setattr = btrfs_setattr, | |
618e21d5 | 8632 | .mknod = btrfs_mknod, |
95819c05 CH |
8633 | .setxattr = btrfs_setxattr, |
8634 | .getxattr = btrfs_getxattr, | |
5103e947 | 8635 | .listxattr = btrfs_listxattr, |
95819c05 | 8636 | .removexattr = btrfs_removexattr, |
fdebe2bd | 8637 | .permission = btrfs_permission, |
4e34e719 | 8638 | .get_acl = btrfs_get_acl, |
39279cc3 | 8639 | }; |
6e1d5dcc | 8640 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 8641 | .lookup = btrfs_lookup, |
fdebe2bd | 8642 | .permission = btrfs_permission, |
4e34e719 | 8643 | .get_acl = btrfs_get_acl, |
39279cc3 | 8644 | }; |
76dda93c | 8645 | |
828c0950 | 8646 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
8647 | .llseek = generic_file_llseek, |
8648 | .read = generic_read_dir, | |
9cdda8d3 | 8649 | .iterate = btrfs_real_readdir, |
34287aa3 | 8650 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 8651 | #ifdef CONFIG_COMPAT |
34287aa3 | 8652 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 8653 | #endif |
6bf13c0c | 8654 | .release = btrfs_release_file, |
e02119d5 | 8655 | .fsync = btrfs_sync_file, |
39279cc3 CM |
8656 | }; |
8657 | ||
d1310b2e | 8658 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 8659 | .fill_delalloc = run_delalloc_range, |
065631f6 | 8660 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 8661 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 8662 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 8663 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 8664 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
8665 | .set_bit_hook = btrfs_set_bit_hook, |
8666 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
8667 | .merge_extent_hook = btrfs_merge_extent_hook, |
8668 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
8669 | }; |
8670 | ||
35054394 CM |
8671 | /* |
8672 | * btrfs doesn't support the bmap operation because swapfiles | |
8673 | * use bmap to make a mapping of extents in the file. They assume | |
8674 | * these extents won't change over the life of the file and they | |
8675 | * use the bmap result to do IO directly to the drive. | |
8676 | * | |
8677 | * the btrfs bmap call would return logical addresses that aren't | |
8678 | * suitable for IO and they also will change frequently as COW | |
8679 | * operations happen. So, swapfile + btrfs == corruption. | |
8680 | * | |
8681 | * For now we're avoiding this by dropping bmap. | |
8682 | */ | |
7f09410b | 8683 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
8684 | .readpage = btrfs_readpage, |
8685 | .writepage = btrfs_writepage, | |
b293f02e | 8686 | .writepages = btrfs_writepages, |
3ab2fb5a | 8687 | .readpages = btrfs_readpages, |
16432985 | 8688 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
8689 | .invalidatepage = btrfs_invalidatepage, |
8690 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 8691 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 8692 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
8693 | }; |
8694 | ||
7f09410b | 8695 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
8696 | .readpage = btrfs_readpage, |
8697 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
8698 | .invalidatepage = btrfs_invalidatepage, |
8699 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
8700 | }; |
8701 | ||
6e1d5dcc | 8702 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
8703 | .getattr = btrfs_getattr, |
8704 | .setattr = btrfs_setattr, | |
95819c05 CH |
8705 | .setxattr = btrfs_setxattr, |
8706 | .getxattr = btrfs_getxattr, | |
5103e947 | 8707 | .listxattr = btrfs_listxattr, |
95819c05 | 8708 | .removexattr = btrfs_removexattr, |
fdebe2bd | 8709 | .permission = btrfs_permission, |
1506fcc8 | 8710 | .fiemap = btrfs_fiemap, |
4e34e719 | 8711 | .get_acl = btrfs_get_acl, |
e41f941a | 8712 | .update_time = btrfs_update_time, |
39279cc3 | 8713 | }; |
6e1d5dcc | 8714 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
8715 | .getattr = btrfs_getattr, |
8716 | .setattr = btrfs_setattr, | |
fdebe2bd | 8717 | .permission = btrfs_permission, |
95819c05 CH |
8718 | .setxattr = btrfs_setxattr, |
8719 | .getxattr = btrfs_getxattr, | |
33268eaf | 8720 | .listxattr = btrfs_listxattr, |
95819c05 | 8721 | .removexattr = btrfs_removexattr, |
4e34e719 | 8722 | .get_acl = btrfs_get_acl, |
e41f941a | 8723 | .update_time = btrfs_update_time, |
618e21d5 | 8724 | }; |
6e1d5dcc | 8725 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
8726 | .readlink = generic_readlink, |
8727 | .follow_link = page_follow_link_light, | |
8728 | .put_link = page_put_link, | |
f209561a | 8729 | .getattr = btrfs_getattr, |
22c44fe6 | 8730 | .setattr = btrfs_setattr, |
fdebe2bd | 8731 | .permission = btrfs_permission, |
0279b4cd JO |
8732 | .setxattr = btrfs_setxattr, |
8733 | .getxattr = btrfs_getxattr, | |
8734 | .listxattr = btrfs_listxattr, | |
8735 | .removexattr = btrfs_removexattr, | |
4e34e719 | 8736 | .get_acl = btrfs_get_acl, |
e41f941a | 8737 | .update_time = btrfs_update_time, |
39279cc3 | 8738 | }; |
76dda93c | 8739 | |
82d339d9 | 8740 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 8741 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 8742 | .d_release = btrfs_dentry_release, |
76dda93c | 8743 | }; |