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> | |
9ebefb18 | 35 | #include <linux/bit_spinlock.h> |
5103e947 | 36 | #include <linux/xattr.h> |
33268eaf | 37 | #include <linux/posix_acl.h> |
d899e052 | 38 | #include <linux/falloc.h> |
4b4e25f2 | 39 | #include "compat.h" |
39279cc3 CM |
40 | #include "ctree.h" |
41 | #include "disk-io.h" | |
42 | #include "transaction.h" | |
43 | #include "btrfs_inode.h" | |
44 | #include "ioctl.h" | |
45 | #include "print-tree.h" | |
0b86a832 | 46 | #include "volumes.h" |
e6dcd2dc | 47 | #include "ordered-data.h" |
95819c05 | 48 | #include "xattr.h" |
e02119d5 | 49 | #include "tree-log.h" |
c8b97818 | 50 | #include "compression.h" |
b4ce94de | 51 | #include "locking.h" |
39279cc3 CM |
52 | |
53 | struct btrfs_iget_args { | |
54 | u64 ino; | |
55 | struct btrfs_root *root; | |
56 | }; | |
57 | ||
6e1d5dcc AD |
58 | static const struct inode_operations btrfs_dir_inode_operations; |
59 | static const struct inode_operations btrfs_symlink_inode_operations; | |
60 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
61 | static const struct inode_operations btrfs_special_inode_operations; | |
62 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
63 | static const struct address_space_operations btrfs_aops; |
64 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 65 | static const struct file_operations btrfs_dir_file_operations; |
d1310b2e | 66 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
67 | |
68 | static struct kmem_cache *btrfs_inode_cachep; | |
69 | struct kmem_cache *btrfs_trans_handle_cachep; | |
70 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 CM |
71 | struct kmem_cache *btrfs_path_cachep; |
72 | ||
73 | #define S_SHIFT 12 | |
74 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
75 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
76 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
77 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
78 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
79 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
80 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
81 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
82 | }; | |
83 | ||
7b128766 | 84 | static void btrfs_truncate(struct inode *inode); |
c8b97818 | 85 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end); |
771ed689 CM |
86 | static noinline int cow_file_range(struct inode *inode, |
87 | struct page *locked_page, | |
88 | u64 start, u64 end, int *page_started, | |
89 | unsigned long *nr_written, int unlock); | |
7b128766 | 90 | |
f34f57a3 YZ |
91 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
92 | struct inode *inode, struct inode *dir) | |
0279b4cd JO |
93 | { |
94 | int err; | |
95 | ||
f34f57a3 | 96 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 97 | if (!err) |
f34f57a3 | 98 | err = btrfs_xattr_security_init(trans, inode, dir); |
0279b4cd JO |
99 | return err; |
100 | } | |
101 | ||
c8b97818 CM |
102 | /* |
103 | * this does all the hard work for inserting an inline extent into | |
104 | * the btree. The caller should have done a btrfs_drop_extents so that | |
105 | * no overlapping inline items exist in the btree | |
106 | */ | |
d397712b | 107 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
108 | struct btrfs_root *root, struct inode *inode, |
109 | u64 start, size_t size, size_t compressed_size, | |
110 | struct page **compressed_pages) | |
111 | { | |
112 | struct btrfs_key key; | |
113 | struct btrfs_path *path; | |
114 | struct extent_buffer *leaf; | |
115 | struct page *page = NULL; | |
116 | char *kaddr; | |
117 | unsigned long ptr; | |
118 | struct btrfs_file_extent_item *ei; | |
119 | int err = 0; | |
120 | int ret; | |
121 | size_t cur_size = size; | |
122 | size_t datasize; | |
123 | unsigned long offset; | |
124 | int use_compress = 0; | |
125 | ||
126 | if (compressed_size && compressed_pages) { | |
127 | use_compress = 1; | |
128 | cur_size = compressed_size; | |
129 | } | |
130 | ||
d397712b CM |
131 | path = btrfs_alloc_path(); |
132 | if (!path) | |
c8b97818 CM |
133 | return -ENOMEM; |
134 | ||
b9473439 | 135 | path->leave_spinning = 1; |
c8b97818 CM |
136 | btrfs_set_trans_block_group(trans, inode); |
137 | ||
138 | key.objectid = inode->i_ino; | |
139 | key.offset = start; | |
140 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
141 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
142 | ||
143 | inode_add_bytes(inode, size); | |
144 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
145 | datasize); | |
146 | BUG_ON(ret); | |
147 | if (ret) { | |
148 | err = ret; | |
c8b97818 CM |
149 | goto fail; |
150 | } | |
151 | leaf = path->nodes[0]; | |
152 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
153 | struct btrfs_file_extent_item); | |
154 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
155 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
156 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
157 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
158 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
159 | ptr = btrfs_file_extent_inline_start(ei); | |
160 | ||
161 | if (use_compress) { | |
162 | struct page *cpage; | |
163 | int i = 0; | |
d397712b | 164 | while (compressed_size > 0) { |
c8b97818 | 165 | cpage = compressed_pages[i]; |
5b050f04 | 166 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
167 | PAGE_CACHE_SIZE); |
168 | ||
b9473439 | 169 | kaddr = kmap_atomic(cpage, KM_USER0); |
c8b97818 | 170 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
b9473439 | 171 | kunmap_atomic(kaddr, KM_USER0); |
c8b97818 CM |
172 | |
173 | i++; | |
174 | ptr += cur_size; | |
175 | compressed_size -= cur_size; | |
176 | } | |
177 | btrfs_set_file_extent_compression(leaf, ei, | |
178 | BTRFS_COMPRESS_ZLIB); | |
179 | } else { | |
180 | page = find_get_page(inode->i_mapping, | |
181 | start >> PAGE_CACHE_SHIFT); | |
182 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
183 | kaddr = kmap_atomic(page, KM_USER0); | |
184 | offset = start & (PAGE_CACHE_SIZE - 1); | |
185 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
186 | kunmap_atomic(kaddr, KM_USER0); | |
187 | page_cache_release(page); | |
188 | } | |
189 | btrfs_mark_buffer_dirty(leaf); | |
190 | btrfs_free_path(path); | |
191 | ||
c2167754 YZ |
192 | /* |
193 | * we're an inline extent, so nobody can | |
194 | * extend the file past i_size without locking | |
195 | * a page we already have locked. | |
196 | * | |
197 | * We must do any isize and inode updates | |
198 | * before we unlock the pages. Otherwise we | |
199 | * could end up racing with unlink. | |
200 | */ | |
c8b97818 CM |
201 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
202 | btrfs_update_inode(trans, root, inode); | |
c2167754 | 203 | |
c8b97818 CM |
204 | return 0; |
205 | fail: | |
206 | btrfs_free_path(path); | |
207 | return err; | |
208 | } | |
209 | ||
210 | ||
211 | /* | |
212 | * conditionally insert an inline extent into the file. This | |
213 | * does the checks required to make sure the data is small enough | |
214 | * to fit as an inline extent. | |
215 | */ | |
7f366cfe | 216 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
217 | struct btrfs_root *root, |
218 | struct inode *inode, u64 start, u64 end, | |
219 | size_t compressed_size, | |
220 | struct page **compressed_pages) | |
221 | { | |
222 | u64 isize = i_size_read(inode); | |
223 | u64 actual_end = min(end + 1, isize); | |
224 | u64 inline_len = actual_end - start; | |
225 | u64 aligned_end = (end + root->sectorsize - 1) & | |
226 | ~((u64)root->sectorsize - 1); | |
227 | u64 hint_byte; | |
228 | u64 data_len = inline_len; | |
229 | int ret; | |
230 | ||
231 | if (compressed_size) | |
232 | data_len = compressed_size; | |
233 | ||
234 | if (start > 0 || | |
70b99e69 | 235 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
236 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
237 | (!compressed_size && | |
238 | (actual_end & (root->sectorsize - 1)) == 0) || | |
239 | end + 1 < isize || | |
240 | data_len > root->fs_info->max_inline) { | |
241 | return 1; | |
242 | } | |
243 | ||
920bbbfb | 244 | ret = btrfs_drop_extents(trans, inode, start, aligned_end, |
a1ed835e | 245 | &hint_byte, 1); |
c8b97818 CM |
246 | BUG_ON(ret); |
247 | ||
248 | if (isize > actual_end) | |
249 | inline_len = min_t(u64, isize, actual_end); | |
250 | ret = insert_inline_extent(trans, root, inode, start, | |
251 | inline_len, compressed_size, | |
252 | compressed_pages); | |
253 | BUG_ON(ret); | |
a1ed835e | 254 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
255 | return 0; |
256 | } | |
257 | ||
771ed689 CM |
258 | struct async_extent { |
259 | u64 start; | |
260 | u64 ram_size; | |
261 | u64 compressed_size; | |
262 | struct page **pages; | |
263 | unsigned long nr_pages; | |
264 | struct list_head list; | |
265 | }; | |
266 | ||
267 | struct async_cow { | |
268 | struct inode *inode; | |
269 | struct btrfs_root *root; | |
270 | struct page *locked_page; | |
271 | u64 start; | |
272 | u64 end; | |
273 | struct list_head extents; | |
274 | struct btrfs_work work; | |
275 | }; | |
276 | ||
277 | static noinline int add_async_extent(struct async_cow *cow, | |
278 | u64 start, u64 ram_size, | |
279 | u64 compressed_size, | |
280 | struct page **pages, | |
281 | unsigned long nr_pages) | |
282 | { | |
283 | struct async_extent *async_extent; | |
284 | ||
285 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
286 | async_extent->start = start; | |
287 | async_extent->ram_size = ram_size; | |
288 | async_extent->compressed_size = compressed_size; | |
289 | async_extent->pages = pages; | |
290 | async_extent->nr_pages = nr_pages; | |
291 | list_add_tail(&async_extent->list, &cow->extents); | |
292 | return 0; | |
293 | } | |
294 | ||
d352ac68 | 295 | /* |
771ed689 CM |
296 | * we create compressed extents in two phases. The first |
297 | * phase compresses a range of pages that have already been | |
298 | * locked (both pages and state bits are locked). | |
c8b97818 | 299 | * |
771ed689 CM |
300 | * This is done inside an ordered work queue, and the compression |
301 | * is spread across many cpus. The actual IO submission is step | |
302 | * two, and the ordered work queue takes care of making sure that | |
303 | * happens in the same order things were put onto the queue by | |
304 | * writepages and friends. | |
c8b97818 | 305 | * |
771ed689 CM |
306 | * If this code finds it can't get good compression, it puts an |
307 | * entry onto the work queue to write the uncompressed bytes. This | |
308 | * makes sure that both compressed inodes and uncompressed inodes | |
309 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 310 | */ |
771ed689 CM |
311 | static noinline int compress_file_range(struct inode *inode, |
312 | struct page *locked_page, | |
313 | u64 start, u64 end, | |
314 | struct async_cow *async_cow, | |
315 | int *num_added) | |
b888db2b CM |
316 | { |
317 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
318 | struct btrfs_trans_handle *trans; | |
db94535d | 319 | u64 num_bytes; |
c8b97818 CM |
320 | u64 orig_start; |
321 | u64 disk_num_bytes; | |
db94535d | 322 | u64 blocksize = root->sectorsize; |
c8b97818 | 323 | u64 actual_end; |
42dc7bab | 324 | u64 isize = i_size_read(inode); |
e6dcd2dc | 325 | int ret = 0; |
c8b97818 CM |
326 | struct page **pages = NULL; |
327 | unsigned long nr_pages; | |
328 | unsigned long nr_pages_ret = 0; | |
329 | unsigned long total_compressed = 0; | |
330 | unsigned long total_in = 0; | |
331 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 332 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
333 | int i; |
334 | int will_compress; | |
b888db2b | 335 | |
c8b97818 CM |
336 | orig_start = start; |
337 | ||
42dc7bab | 338 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
339 | again: |
340 | will_compress = 0; | |
341 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
342 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 343 | |
f03d9301 CM |
344 | /* |
345 | * we don't want to send crud past the end of i_size through | |
346 | * compression, that's just a waste of CPU time. So, if the | |
347 | * end of the file is before the start of our current | |
348 | * requested range of bytes, we bail out to the uncompressed | |
349 | * cleanup code that can deal with all of this. | |
350 | * | |
351 | * It isn't really the fastest way to fix things, but this is a | |
352 | * very uncommon corner. | |
353 | */ | |
354 | if (actual_end <= start) | |
355 | goto cleanup_and_bail_uncompressed; | |
356 | ||
c8b97818 CM |
357 | total_compressed = actual_end - start; |
358 | ||
359 | /* we want to make sure that amount of ram required to uncompress | |
360 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
361 | * of a compressed extent to 128k. This is a crucial number |
362 | * because it also controls how easily we can spread reads across | |
363 | * cpus for decompression. | |
364 | * | |
365 | * We also want to make sure the amount of IO required to do | |
366 | * a random read is reasonably small, so we limit the size of | |
367 | * a compressed extent to 128k. | |
c8b97818 CM |
368 | */ |
369 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 370 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 371 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
372 | disk_num_bytes = num_bytes; |
373 | total_in = 0; | |
374 | ret = 0; | |
db94535d | 375 | |
771ed689 CM |
376 | /* |
377 | * we do compression for mount -o compress and when the | |
378 | * inode has not been flagged as nocompress. This flag can | |
379 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 380 | */ |
6cbff00f | 381 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
c8b97818 CM |
382 | btrfs_test_opt(root, COMPRESS)) { |
383 | WARN_ON(pages); | |
cfbc246e | 384 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
c8b97818 | 385 | |
c8b97818 CM |
386 | ret = btrfs_zlib_compress_pages(inode->i_mapping, start, |
387 | total_compressed, pages, | |
388 | nr_pages, &nr_pages_ret, | |
389 | &total_in, | |
390 | &total_compressed, | |
391 | max_compressed); | |
392 | ||
393 | if (!ret) { | |
394 | unsigned long offset = total_compressed & | |
395 | (PAGE_CACHE_SIZE - 1); | |
396 | struct page *page = pages[nr_pages_ret - 1]; | |
397 | char *kaddr; | |
398 | ||
399 | /* zero the tail end of the last page, we might be | |
400 | * sending it down to disk | |
401 | */ | |
402 | if (offset) { | |
403 | kaddr = kmap_atomic(page, KM_USER0); | |
404 | memset(kaddr + offset, 0, | |
405 | PAGE_CACHE_SIZE - offset); | |
406 | kunmap_atomic(kaddr, KM_USER0); | |
407 | } | |
408 | will_compress = 1; | |
409 | } | |
410 | } | |
411 | if (start == 0) { | |
771ed689 CM |
412 | trans = btrfs_join_transaction(root, 1); |
413 | BUG_ON(!trans); | |
414 | btrfs_set_trans_block_group(trans, inode); | |
415 | ||
c8b97818 | 416 | /* lets try to make an inline extent */ |
771ed689 | 417 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 418 | /* we didn't compress the entire range, try |
771ed689 | 419 | * to make an uncompressed inline extent. |
c8b97818 CM |
420 | */ |
421 | ret = cow_file_range_inline(trans, root, inode, | |
422 | start, end, 0, NULL); | |
423 | } else { | |
771ed689 | 424 | /* try making a compressed inline extent */ |
c8b97818 CM |
425 | ret = cow_file_range_inline(trans, root, inode, |
426 | start, end, | |
427 | total_compressed, pages); | |
428 | } | |
429 | if (ret == 0) { | |
771ed689 CM |
430 | /* |
431 | * inline extent creation worked, we don't need | |
432 | * to create any more async work items. Unlock | |
433 | * and free up our temp pages. | |
434 | */ | |
c8b97818 | 435 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
436 | &BTRFS_I(inode)->io_tree, |
437 | start, end, NULL, | |
438 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 439 | EXTENT_CLEAR_DELALLOC | |
32c00aff | 440 | EXTENT_CLEAR_ACCOUNTING | |
a791e35e | 441 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
442 | |
443 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
444 | goto free_pages_out; |
445 | } | |
c2167754 | 446 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
447 | } |
448 | ||
449 | if (will_compress) { | |
450 | /* | |
451 | * we aren't doing an inline extent round the compressed size | |
452 | * up to a block size boundary so the allocator does sane | |
453 | * things | |
454 | */ | |
455 | total_compressed = (total_compressed + blocksize - 1) & | |
456 | ~(blocksize - 1); | |
457 | ||
458 | /* | |
459 | * one last check to make sure the compression is really a | |
460 | * win, compare the page count read with the blocks on disk | |
461 | */ | |
462 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
463 | ~(PAGE_CACHE_SIZE - 1); | |
464 | if (total_compressed >= total_in) { | |
465 | will_compress = 0; | |
466 | } else { | |
467 | disk_num_bytes = total_compressed; | |
468 | num_bytes = total_in; | |
469 | } | |
470 | } | |
471 | if (!will_compress && pages) { | |
472 | /* | |
473 | * the compression code ran but failed to make things smaller, | |
474 | * free any pages it allocated and our page pointer array | |
475 | */ | |
476 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 477 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
478 | page_cache_release(pages[i]); |
479 | } | |
480 | kfree(pages); | |
481 | pages = NULL; | |
482 | total_compressed = 0; | |
483 | nr_pages_ret = 0; | |
484 | ||
485 | /* flag the file so we don't compress in the future */ | |
a555f810 CM |
486 | if (!btrfs_test_opt(root, FORCE_COMPRESS)) |
487 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; | |
c8b97818 | 488 | } |
771ed689 CM |
489 | if (will_compress) { |
490 | *num_added += 1; | |
c8b97818 | 491 | |
771ed689 CM |
492 | /* the async work queues will take care of doing actual |
493 | * allocation on disk for these compressed pages, | |
494 | * and will submit them to the elevator. | |
495 | */ | |
496 | add_async_extent(async_cow, start, num_bytes, | |
497 | total_compressed, pages, nr_pages_ret); | |
179e29e4 | 498 | |
42dc7bab | 499 | if (start + num_bytes < end && start + num_bytes < actual_end) { |
771ed689 CM |
500 | start += num_bytes; |
501 | pages = NULL; | |
502 | cond_resched(); | |
503 | goto again; | |
504 | } | |
505 | } else { | |
f03d9301 | 506 | cleanup_and_bail_uncompressed: |
771ed689 CM |
507 | /* |
508 | * No compression, but we still need to write the pages in | |
509 | * the file we've been given so far. redirty the locked | |
510 | * page if it corresponds to our extent and set things up | |
511 | * for the async work queue to run cow_file_range to do | |
512 | * the normal delalloc dance | |
513 | */ | |
514 | if (page_offset(locked_page) >= start && | |
515 | page_offset(locked_page) <= end) { | |
516 | __set_page_dirty_nobuffers(locked_page); | |
517 | /* unlocked later on in the async handlers */ | |
518 | } | |
519 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0); | |
520 | *num_added += 1; | |
521 | } | |
3b951516 | 522 | |
771ed689 CM |
523 | out: |
524 | return 0; | |
525 | ||
526 | free_pages_out: | |
527 | for (i = 0; i < nr_pages_ret; i++) { | |
528 | WARN_ON(pages[i]->mapping); | |
529 | page_cache_release(pages[i]); | |
530 | } | |
d397712b | 531 | kfree(pages); |
771ed689 CM |
532 | |
533 | goto out; | |
534 | } | |
535 | ||
536 | /* | |
537 | * phase two of compressed writeback. This is the ordered portion | |
538 | * of the code, which only gets called in the order the work was | |
539 | * queued. We walk all the async extents created by compress_file_range | |
540 | * and send them down to the disk. | |
541 | */ | |
542 | static noinline int submit_compressed_extents(struct inode *inode, | |
543 | struct async_cow *async_cow) | |
544 | { | |
545 | struct async_extent *async_extent; | |
546 | u64 alloc_hint = 0; | |
547 | struct btrfs_trans_handle *trans; | |
548 | struct btrfs_key ins; | |
549 | struct extent_map *em; | |
550 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
551 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
552 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 553 | int ret = 0; |
771ed689 CM |
554 | |
555 | if (list_empty(&async_cow->extents)) | |
556 | return 0; | |
557 | ||
771ed689 | 558 | |
d397712b | 559 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
560 | async_extent = list_entry(async_cow->extents.next, |
561 | struct async_extent, list); | |
562 | list_del(&async_extent->list); | |
c8b97818 | 563 | |
771ed689 CM |
564 | io_tree = &BTRFS_I(inode)->io_tree; |
565 | ||
f5a84ee3 | 566 | retry: |
771ed689 CM |
567 | /* did the compression code fall back to uncompressed IO? */ |
568 | if (!async_extent->pages) { | |
569 | int page_started = 0; | |
570 | unsigned long nr_written = 0; | |
571 | ||
572 | lock_extent(io_tree, async_extent->start, | |
d397712b CM |
573 | async_extent->start + |
574 | async_extent->ram_size - 1, GFP_NOFS); | |
771ed689 CM |
575 | |
576 | /* allocate blocks */ | |
f5a84ee3 JB |
577 | ret = cow_file_range(inode, async_cow->locked_page, |
578 | async_extent->start, | |
579 | async_extent->start + | |
580 | async_extent->ram_size - 1, | |
581 | &page_started, &nr_written, 0); | |
771ed689 CM |
582 | |
583 | /* | |
584 | * if page_started, cow_file_range inserted an | |
585 | * inline extent and took care of all the unlocking | |
586 | * and IO for us. Otherwise, we need to submit | |
587 | * all those pages down to the drive. | |
588 | */ | |
f5a84ee3 | 589 | if (!page_started && !ret) |
771ed689 CM |
590 | extent_write_locked_range(io_tree, |
591 | inode, async_extent->start, | |
d397712b | 592 | async_extent->start + |
771ed689 CM |
593 | async_extent->ram_size - 1, |
594 | btrfs_get_extent, | |
595 | WB_SYNC_ALL); | |
596 | kfree(async_extent); | |
597 | cond_resched(); | |
598 | continue; | |
599 | } | |
600 | ||
601 | lock_extent(io_tree, async_extent->start, | |
602 | async_extent->start + async_extent->ram_size - 1, | |
603 | GFP_NOFS); | |
771ed689 | 604 | |
c2167754 | 605 | trans = btrfs_join_transaction(root, 1); |
771ed689 CM |
606 | ret = btrfs_reserve_extent(trans, root, |
607 | async_extent->compressed_size, | |
608 | async_extent->compressed_size, | |
609 | 0, alloc_hint, | |
610 | (u64)-1, &ins, 1); | |
c2167754 YZ |
611 | btrfs_end_transaction(trans, root); |
612 | ||
f5a84ee3 JB |
613 | if (ret) { |
614 | int i; | |
615 | for (i = 0; i < async_extent->nr_pages; i++) { | |
616 | WARN_ON(async_extent->pages[i]->mapping); | |
617 | page_cache_release(async_extent->pages[i]); | |
618 | } | |
619 | kfree(async_extent->pages); | |
620 | async_extent->nr_pages = 0; | |
621 | async_extent->pages = NULL; | |
622 | unlock_extent(io_tree, async_extent->start, | |
623 | async_extent->start + | |
624 | async_extent->ram_size - 1, GFP_NOFS); | |
625 | goto retry; | |
626 | } | |
627 | ||
c2167754 YZ |
628 | /* |
629 | * here we're doing allocation and writeback of the | |
630 | * compressed pages | |
631 | */ | |
632 | btrfs_drop_extent_cache(inode, async_extent->start, | |
633 | async_extent->start + | |
634 | async_extent->ram_size - 1, 0); | |
635 | ||
771ed689 CM |
636 | em = alloc_extent_map(GFP_NOFS); |
637 | em->start = async_extent->start; | |
638 | em->len = async_extent->ram_size; | |
445a6944 | 639 | em->orig_start = em->start; |
c8b97818 | 640 | |
771ed689 CM |
641 | em->block_start = ins.objectid; |
642 | em->block_len = ins.offset; | |
643 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
644 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
645 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
646 | ||
d397712b | 647 | while (1) { |
890871be | 648 | write_lock(&em_tree->lock); |
771ed689 | 649 | ret = add_extent_mapping(em_tree, em); |
890871be | 650 | write_unlock(&em_tree->lock); |
771ed689 CM |
651 | if (ret != -EEXIST) { |
652 | free_extent_map(em); | |
653 | break; | |
654 | } | |
655 | btrfs_drop_extent_cache(inode, async_extent->start, | |
656 | async_extent->start + | |
657 | async_extent->ram_size - 1, 0); | |
658 | } | |
659 | ||
660 | ret = btrfs_add_ordered_extent(inode, async_extent->start, | |
661 | ins.objectid, | |
662 | async_extent->ram_size, | |
663 | ins.offset, | |
664 | BTRFS_ORDERED_COMPRESSED); | |
665 | BUG_ON(ret); | |
666 | ||
771ed689 CM |
667 | /* |
668 | * clear dirty, set writeback and unlock the pages. | |
669 | */ | |
670 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
671 | &BTRFS_I(inode)->io_tree, |
672 | async_extent->start, | |
673 | async_extent->start + | |
674 | async_extent->ram_size - 1, | |
675 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
676 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 677 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 678 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
679 | |
680 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
681 | async_extent->start, |
682 | async_extent->ram_size, | |
683 | ins.objectid, | |
684 | ins.offset, async_extent->pages, | |
685 | async_extent->nr_pages); | |
771ed689 CM |
686 | |
687 | BUG_ON(ret); | |
771ed689 CM |
688 | alloc_hint = ins.objectid + ins.offset; |
689 | kfree(async_extent); | |
690 | cond_resched(); | |
691 | } | |
692 | ||
771ed689 CM |
693 | return 0; |
694 | } | |
695 | ||
696 | /* | |
697 | * when extent_io.c finds a delayed allocation range in the file, | |
698 | * the call backs end up in this code. The basic idea is to | |
699 | * allocate extents on disk for the range, and create ordered data structs | |
700 | * in ram to track those extents. | |
701 | * | |
702 | * locked_page is the page that writepage had locked already. We use | |
703 | * it to make sure we don't do extra locks or unlocks. | |
704 | * | |
705 | * *page_started is set to one if we unlock locked_page and do everything | |
706 | * required to start IO on it. It may be clean and already done with | |
707 | * IO when we return. | |
708 | */ | |
709 | static noinline int cow_file_range(struct inode *inode, | |
710 | struct page *locked_page, | |
711 | u64 start, u64 end, int *page_started, | |
712 | unsigned long *nr_written, | |
713 | int unlock) | |
714 | { | |
715 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
716 | struct btrfs_trans_handle *trans; | |
717 | u64 alloc_hint = 0; | |
718 | u64 num_bytes; | |
719 | unsigned long ram_size; | |
720 | u64 disk_num_bytes; | |
721 | u64 cur_alloc_size; | |
722 | u64 blocksize = root->sectorsize; | |
723 | u64 actual_end; | |
42dc7bab | 724 | u64 isize = i_size_read(inode); |
771ed689 CM |
725 | struct btrfs_key ins; |
726 | struct extent_map *em; | |
727 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
728 | int ret = 0; | |
729 | ||
730 | trans = btrfs_join_transaction(root, 1); | |
731 | BUG_ON(!trans); | |
732 | btrfs_set_trans_block_group(trans, inode); | |
733 | ||
42dc7bab | 734 | actual_end = min_t(u64, isize, end + 1); |
771ed689 CM |
735 | |
736 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); | |
737 | num_bytes = max(blocksize, num_bytes); | |
738 | disk_num_bytes = num_bytes; | |
739 | ret = 0; | |
740 | ||
741 | if (start == 0) { | |
742 | /* lets try to make an inline extent */ | |
743 | ret = cow_file_range_inline(trans, root, inode, | |
744 | start, end, 0, NULL); | |
745 | if (ret == 0) { | |
746 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
747 | &BTRFS_I(inode)->io_tree, |
748 | start, end, NULL, | |
749 | EXTENT_CLEAR_UNLOCK_PAGE | | |
750 | EXTENT_CLEAR_UNLOCK | | |
751 | EXTENT_CLEAR_DELALLOC | | |
32c00aff | 752 | EXTENT_CLEAR_ACCOUNTING | |
a791e35e CM |
753 | EXTENT_CLEAR_DIRTY | |
754 | EXTENT_SET_WRITEBACK | | |
755 | EXTENT_END_WRITEBACK); | |
c2167754 | 756 | |
771ed689 CM |
757 | *nr_written = *nr_written + |
758 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
759 | *page_started = 1; | |
760 | ret = 0; | |
761 | goto out; | |
762 | } | |
763 | } | |
764 | ||
765 | BUG_ON(disk_num_bytes > | |
766 | btrfs_super_total_bytes(&root->fs_info->super_copy)); | |
767 | ||
b917b7c3 CM |
768 | |
769 | read_lock(&BTRFS_I(inode)->extent_tree.lock); | |
770 | em = search_extent_mapping(&BTRFS_I(inode)->extent_tree, | |
771 | start, num_bytes); | |
772 | if (em) { | |
6346c939 JB |
773 | /* |
774 | * if block start isn't an actual block number then find the | |
775 | * first block in this inode and use that as a hint. If that | |
776 | * block is also bogus then just don't worry about it. | |
777 | */ | |
778 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
779 | free_extent_map(em); | |
780 | em = search_extent_mapping(em_tree, 0, 0); | |
781 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
782 | alloc_hint = em->block_start; | |
783 | if (em) | |
784 | free_extent_map(em); | |
785 | } else { | |
786 | alloc_hint = em->block_start; | |
787 | free_extent_map(em); | |
788 | } | |
b917b7c3 CM |
789 | } |
790 | read_unlock(&BTRFS_I(inode)->extent_tree.lock); | |
771ed689 CM |
791 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
792 | ||
d397712b | 793 | while (disk_num_bytes > 0) { |
a791e35e CM |
794 | unsigned long op; |
795 | ||
c8b97818 | 796 | cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent); |
e6dcd2dc | 797 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 798 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 799 | (u64)-1, &ins, 1); |
d397712b CM |
800 | BUG_ON(ret); |
801 | ||
e6dcd2dc CM |
802 | em = alloc_extent_map(GFP_NOFS); |
803 | em->start = start; | |
445a6944 | 804 | em->orig_start = em->start; |
771ed689 CM |
805 | ram_size = ins.offset; |
806 | em->len = ins.offset; | |
c8b97818 | 807 | |
e6dcd2dc | 808 | em->block_start = ins.objectid; |
c8b97818 | 809 | em->block_len = ins.offset; |
e6dcd2dc | 810 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 811 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 812 | |
d397712b | 813 | while (1) { |
890871be | 814 | write_lock(&em_tree->lock); |
e6dcd2dc | 815 | ret = add_extent_mapping(em_tree, em); |
890871be | 816 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
817 | if (ret != -EEXIST) { |
818 | free_extent_map(em); | |
819 | break; | |
820 | } | |
821 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 822 | start + ram_size - 1, 0); |
e6dcd2dc CM |
823 | } |
824 | ||
98d20f67 | 825 | cur_alloc_size = ins.offset; |
e6dcd2dc | 826 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 827 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 828 | BUG_ON(ret); |
c8b97818 | 829 | |
17d217fe YZ |
830 | if (root->root_key.objectid == |
831 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
832 | ret = btrfs_reloc_clone_csums(inode, start, | |
833 | cur_alloc_size); | |
834 | BUG_ON(ret); | |
835 | } | |
836 | ||
d397712b | 837 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 838 | break; |
d397712b | 839 | |
c8b97818 CM |
840 | /* we're not doing compressed IO, don't unlock the first |
841 | * page (which the caller expects to stay locked), don't | |
842 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
843 | * |
844 | * Do set the Private2 bit so we know this page was properly | |
845 | * setup for writepage | |
c8b97818 | 846 | */ |
a791e35e CM |
847 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
848 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
849 | EXTENT_SET_PRIVATE2; | |
850 | ||
c8b97818 CM |
851 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
852 | start, start + ram_size - 1, | |
a791e35e | 853 | locked_page, op); |
c8b97818 | 854 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
855 | num_bytes -= cur_alloc_size; |
856 | alloc_hint = ins.objectid + ins.offset; | |
857 | start += cur_alloc_size; | |
b888db2b | 858 | } |
b888db2b | 859 | out: |
771ed689 | 860 | ret = 0; |
b888db2b | 861 | btrfs_end_transaction(trans, root); |
c8b97818 | 862 | |
be20aa9d | 863 | return ret; |
771ed689 | 864 | } |
c8b97818 | 865 | |
771ed689 CM |
866 | /* |
867 | * work queue call back to started compression on a file and pages | |
868 | */ | |
869 | static noinline void async_cow_start(struct btrfs_work *work) | |
870 | { | |
871 | struct async_cow *async_cow; | |
872 | int num_added = 0; | |
873 | async_cow = container_of(work, struct async_cow, work); | |
874 | ||
875 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
876 | async_cow->start, async_cow->end, async_cow, | |
877 | &num_added); | |
878 | if (num_added == 0) | |
879 | async_cow->inode = NULL; | |
880 | } | |
881 | ||
882 | /* | |
883 | * work queue call back to submit previously compressed pages | |
884 | */ | |
885 | static noinline void async_cow_submit(struct btrfs_work *work) | |
886 | { | |
887 | struct async_cow *async_cow; | |
888 | struct btrfs_root *root; | |
889 | unsigned long nr_pages; | |
890 | ||
891 | async_cow = container_of(work, struct async_cow, work); | |
892 | ||
893 | root = async_cow->root; | |
894 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
895 | PAGE_CACHE_SHIFT; | |
896 | ||
897 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
898 | ||
899 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
900 | 5 * 1042 * 1024 && | |
901 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
902 | wake_up(&root->fs_info->async_submit_wait); | |
903 | ||
d397712b | 904 | if (async_cow->inode) |
771ed689 | 905 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 906 | } |
c8b97818 | 907 | |
771ed689 CM |
908 | static noinline void async_cow_free(struct btrfs_work *work) |
909 | { | |
910 | struct async_cow *async_cow; | |
911 | async_cow = container_of(work, struct async_cow, work); | |
912 | kfree(async_cow); | |
913 | } | |
914 | ||
915 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
916 | u64 start, u64 end, int *page_started, | |
917 | unsigned long *nr_written) | |
918 | { | |
919 | struct async_cow *async_cow; | |
920 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
921 | unsigned long nr_pages; | |
922 | u64 cur_end; | |
923 | int limit = 10 * 1024 * 1042; | |
924 | ||
a3429ab7 CM |
925 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
926 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 927 | while (start < end) { |
771ed689 CM |
928 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
929 | async_cow->inode = inode; | |
930 | async_cow->root = root; | |
931 | async_cow->locked_page = locked_page; | |
932 | async_cow->start = start; | |
933 | ||
6cbff00f | 934 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
935 | cur_end = end; |
936 | else | |
937 | cur_end = min(end, start + 512 * 1024 - 1); | |
938 | ||
939 | async_cow->end = cur_end; | |
940 | INIT_LIST_HEAD(&async_cow->extents); | |
941 | ||
942 | async_cow->work.func = async_cow_start; | |
943 | async_cow->work.ordered_func = async_cow_submit; | |
944 | async_cow->work.ordered_free = async_cow_free; | |
945 | async_cow->work.flags = 0; | |
946 | ||
771ed689 CM |
947 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
948 | PAGE_CACHE_SHIFT; | |
949 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
950 | ||
951 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
952 | &async_cow->work); | |
953 | ||
954 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
955 | wait_event(root->fs_info->async_submit_wait, | |
956 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
957 | limit)); | |
958 | } | |
959 | ||
d397712b | 960 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
961 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
962 | wait_event(root->fs_info->async_submit_wait, | |
963 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
964 | 0)); | |
965 | } | |
966 | ||
967 | *nr_written += nr_pages; | |
968 | start = cur_end + 1; | |
969 | } | |
970 | *page_started = 1; | |
971 | return 0; | |
be20aa9d CM |
972 | } |
973 | ||
d397712b | 974 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
975 | u64 bytenr, u64 num_bytes) |
976 | { | |
977 | int ret; | |
978 | struct btrfs_ordered_sum *sums; | |
979 | LIST_HEAD(list); | |
980 | ||
07d400a6 YZ |
981 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
982 | bytenr + num_bytes - 1, &list); | |
17d217fe YZ |
983 | if (ret == 0 && list_empty(&list)) |
984 | return 0; | |
985 | ||
986 | while (!list_empty(&list)) { | |
987 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
988 | list_del(&sums->list); | |
989 | kfree(sums); | |
990 | } | |
991 | return 1; | |
992 | } | |
993 | ||
d352ac68 CM |
994 | /* |
995 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
996 | * of the extents that exist in the file, and COWs the file as required. | |
997 | * | |
998 | * If no cow copies or snapshots exist, we write directly to the existing | |
999 | * blocks on disk | |
1000 | */ | |
7f366cfe CM |
1001 | static noinline int run_delalloc_nocow(struct inode *inode, |
1002 | struct page *locked_page, | |
771ed689 CM |
1003 | u64 start, u64 end, int *page_started, int force, |
1004 | unsigned long *nr_written) | |
be20aa9d | 1005 | { |
be20aa9d | 1006 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1007 | struct btrfs_trans_handle *trans; |
be20aa9d | 1008 | struct extent_buffer *leaf; |
be20aa9d | 1009 | struct btrfs_path *path; |
80ff3856 | 1010 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1011 | struct btrfs_key found_key; |
80ff3856 YZ |
1012 | u64 cow_start; |
1013 | u64 cur_offset; | |
1014 | u64 extent_end; | |
5d4f98a2 | 1015 | u64 extent_offset; |
80ff3856 YZ |
1016 | u64 disk_bytenr; |
1017 | u64 num_bytes; | |
1018 | int extent_type; | |
1019 | int ret; | |
d899e052 | 1020 | int type; |
80ff3856 YZ |
1021 | int nocow; |
1022 | int check_prev = 1; | |
be20aa9d CM |
1023 | |
1024 | path = btrfs_alloc_path(); | |
1025 | BUG_ON(!path); | |
7ea394f1 YZ |
1026 | trans = btrfs_join_transaction(root, 1); |
1027 | BUG_ON(!trans); | |
be20aa9d | 1028 | |
80ff3856 YZ |
1029 | cow_start = (u64)-1; |
1030 | cur_offset = start; | |
1031 | while (1) { | |
1032 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
1033 | cur_offset, 0); | |
1034 | BUG_ON(ret < 0); | |
1035 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
1036 | leaf = path->nodes[0]; | |
1037 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1038 | path->slots[0] - 1); | |
1039 | if (found_key.objectid == inode->i_ino && | |
1040 | found_key.type == BTRFS_EXTENT_DATA_KEY) | |
1041 | path->slots[0]--; | |
1042 | } | |
1043 | check_prev = 0; | |
1044 | next_slot: | |
1045 | leaf = path->nodes[0]; | |
1046 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1047 | ret = btrfs_next_leaf(root, path); | |
1048 | if (ret < 0) | |
1049 | BUG_ON(1); | |
1050 | if (ret > 0) | |
1051 | break; | |
1052 | leaf = path->nodes[0]; | |
1053 | } | |
be20aa9d | 1054 | |
80ff3856 YZ |
1055 | nocow = 0; |
1056 | disk_bytenr = 0; | |
17d217fe | 1057 | num_bytes = 0; |
80ff3856 YZ |
1058 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1059 | ||
1060 | if (found_key.objectid > inode->i_ino || | |
1061 | found_key.type > BTRFS_EXTENT_DATA_KEY || | |
1062 | found_key.offset > end) | |
1063 | break; | |
1064 | ||
1065 | if (found_key.offset > cur_offset) { | |
1066 | extent_end = found_key.offset; | |
e9061e21 | 1067 | extent_type = 0; |
80ff3856 YZ |
1068 | goto out_check; |
1069 | } | |
1070 | ||
1071 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1072 | struct btrfs_file_extent_item); | |
1073 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1074 | ||
d899e052 YZ |
1075 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1076 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1077 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1078 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1079 | extent_end = found_key.offset + |
1080 | btrfs_file_extent_num_bytes(leaf, fi); | |
1081 | if (extent_end <= start) { | |
1082 | path->slots[0]++; | |
1083 | goto next_slot; | |
1084 | } | |
17d217fe YZ |
1085 | if (disk_bytenr == 0) |
1086 | goto out_check; | |
80ff3856 YZ |
1087 | if (btrfs_file_extent_compression(leaf, fi) || |
1088 | btrfs_file_extent_encryption(leaf, fi) || | |
1089 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1090 | goto out_check; | |
d899e052 YZ |
1091 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1092 | goto out_check; | |
d2fb3437 | 1093 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1094 | goto out_check; |
17d217fe | 1095 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, |
5d4f98a2 YZ |
1096 | found_key.offset - |
1097 | extent_offset, disk_bytenr)) | |
17d217fe | 1098 | goto out_check; |
5d4f98a2 | 1099 | disk_bytenr += extent_offset; |
17d217fe YZ |
1100 | disk_bytenr += cur_offset - found_key.offset; |
1101 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1102 | /* | |
1103 | * force cow if csum exists in the range. | |
1104 | * this ensure that csum for a given extent are | |
1105 | * either valid or do not exist. | |
1106 | */ | |
1107 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1108 | goto out_check; | |
80ff3856 YZ |
1109 | nocow = 1; |
1110 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1111 | extent_end = found_key.offset + | |
1112 | btrfs_file_extent_inline_len(leaf, fi); | |
1113 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1114 | } else { | |
1115 | BUG_ON(1); | |
1116 | } | |
1117 | out_check: | |
1118 | if (extent_end <= start) { | |
1119 | path->slots[0]++; | |
1120 | goto next_slot; | |
1121 | } | |
1122 | if (!nocow) { | |
1123 | if (cow_start == (u64)-1) | |
1124 | cow_start = cur_offset; | |
1125 | cur_offset = extent_end; | |
1126 | if (cur_offset > end) | |
1127 | break; | |
1128 | path->slots[0]++; | |
1129 | goto next_slot; | |
7ea394f1 YZ |
1130 | } |
1131 | ||
1132 | btrfs_release_path(root, path); | |
80ff3856 YZ |
1133 | if (cow_start != (u64)-1) { |
1134 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1135 | found_key.offset - 1, page_started, |
1136 | nr_written, 1); | |
80ff3856 YZ |
1137 | BUG_ON(ret); |
1138 | cow_start = (u64)-1; | |
7ea394f1 | 1139 | } |
80ff3856 | 1140 | |
d899e052 YZ |
1141 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1142 | struct extent_map *em; | |
1143 | struct extent_map_tree *em_tree; | |
1144 | em_tree = &BTRFS_I(inode)->extent_tree; | |
1145 | em = alloc_extent_map(GFP_NOFS); | |
1146 | em->start = cur_offset; | |
445a6944 | 1147 | em->orig_start = em->start; |
d899e052 YZ |
1148 | em->len = num_bytes; |
1149 | em->block_len = num_bytes; | |
1150 | em->block_start = disk_bytenr; | |
1151 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1152 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1153 | while (1) { | |
890871be | 1154 | write_lock(&em_tree->lock); |
d899e052 | 1155 | ret = add_extent_mapping(em_tree, em); |
890871be | 1156 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1157 | if (ret != -EEXIST) { |
1158 | free_extent_map(em); | |
1159 | break; | |
1160 | } | |
1161 | btrfs_drop_extent_cache(inode, em->start, | |
1162 | em->start + em->len - 1, 0); | |
1163 | } | |
1164 | type = BTRFS_ORDERED_PREALLOC; | |
1165 | } else { | |
1166 | type = BTRFS_ORDERED_NOCOW; | |
1167 | } | |
80ff3856 YZ |
1168 | |
1169 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1170 | num_bytes, num_bytes, type); |
1171 | BUG_ON(ret); | |
771ed689 | 1172 | |
d899e052 | 1173 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1174 | cur_offset, cur_offset + num_bytes - 1, |
1175 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1176 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1177 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1178 | cur_offset = extent_end; |
1179 | if (cur_offset > end) | |
1180 | break; | |
be20aa9d | 1181 | } |
80ff3856 YZ |
1182 | btrfs_release_path(root, path); |
1183 | ||
1184 | if (cur_offset <= end && cow_start == (u64)-1) | |
1185 | cow_start = cur_offset; | |
1186 | if (cow_start != (u64)-1) { | |
1187 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1188 | page_started, nr_written, 1); |
80ff3856 YZ |
1189 | BUG_ON(ret); |
1190 | } | |
1191 | ||
1192 | ret = btrfs_end_transaction(trans, root); | |
1193 | BUG_ON(ret); | |
7ea394f1 | 1194 | btrfs_free_path(path); |
80ff3856 | 1195 | return 0; |
be20aa9d CM |
1196 | } |
1197 | ||
d352ac68 CM |
1198 | /* |
1199 | * extent_io.c call back to do delayed allocation processing | |
1200 | */ | |
c8b97818 | 1201 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1202 | u64 start, u64 end, int *page_started, |
1203 | unsigned long *nr_written) | |
be20aa9d | 1204 | { |
be20aa9d | 1205 | int ret; |
7f366cfe | 1206 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1207 | |
6cbff00f | 1208 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1209 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1210 | page_started, 1, nr_written); |
6cbff00f | 1211 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1212 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1213 | page_started, 0, nr_written); |
7f366cfe CM |
1214 | else if (!btrfs_test_opt(root, COMPRESS)) |
1215 | ret = cow_file_range(inode, locked_page, start, end, | |
1216 | page_started, nr_written, 1); | |
be20aa9d | 1217 | else |
771ed689 | 1218 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1219 | page_started, nr_written); |
b888db2b CM |
1220 | return ret; |
1221 | } | |
1222 | ||
9ed74f2d JB |
1223 | static int btrfs_split_extent_hook(struct inode *inode, |
1224 | struct extent_state *orig, u64 split) | |
1225 | { | |
1226 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1227 | u64 size; | |
1228 | ||
1229 | if (!(orig->state & EXTENT_DELALLOC)) | |
1230 | return 0; | |
1231 | ||
1232 | size = orig->end - orig->start + 1; | |
1233 | if (size > root->fs_info->max_extent) { | |
1234 | u64 num_extents; | |
1235 | u64 new_size; | |
1236 | ||
1237 | new_size = orig->end - split + 1; | |
1238 | num_extents = div64_u64(size + root->fs_info->max_extent - 1, | |
1239 | root->fs_info->max_extent); | |
1240 | ||
1241 | /* | |
32c00aff JB |
1242 | * if we break a large extent up then leave oustanding_extents |
1243 | * be, since we've already accounted for the large extent. | |
9ed74f2d JB |
1244 | */ |
1245 | if (div64_u64(new_size + root->fs_info->max_extent - 1, | |
1246 | root->fs_info->max_extent) < num_extents) | |
1247 | return 0; | |
1248 | } | |
1249 | ||
32c00aff JB |
1250 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1251 | BTRFS_I(inode)->outstanding_extents++; | |
1252 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1253 | |
1254 | return 0; | |
1255 | } | |
1256 | ||
1257 | /* | |
1258 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1259 | * extents so we can keep track of new extents that are just merged onto old | |
1260 | * extents, such as when we are doing sequential writes, so we can properly | |
1261 | * account for the metadata space we'll need. | |
1262 | */ | |
1263 | static int btrfs_merge_extent_hook(struct inode *inode, | |
1264 | struct extent_state *new, | |
1265 | struct extent_state *other) | |
1266 | { | |
1267 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1268 | u64 new_size, old_size; | |
1269 | u64 num_extents; | |
1270 | ||
1271 | /* not delalloc, ignore it */ | |
1272 | if (!(other->state & EXTENT_DELALLOC)) | |
1273 | return 0; | |
1274 | ||
1275 | old_size = other->end - other->start + 1; | |
1276 | if (new->start < other->start) | |
1277 | new_size = other->end - new->start + 1; | |
1278 | else | |
1279 | new_size = new->end - other->start + 1; | |
1280 | ||
1281 | /* we're not bigger than the max, unreserve the space and go */ | |
1282 | if (new_size <= root->fs_info->max_extent) { | |
32c00aff JB |
1283 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1284 | BTRFS_I(inode)->outstanding_extents--; | |
1285 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1286 | return 0; |
1287 | } | |
1288 | ||
1289 | /* | |
1290 | * If we grew by another max_extent, just return, we want to keep that | |
1291 | * reserved amount. | |
1292 | */ | |
1293 | num_extents = div64_u64(old_size + root->fs_info->max_extent - 1, | |
1294 | root->fs_info->max_extent); | |
1295 | if (div64_u64(new_size + root->fs_info->max_extent - 1, | |
1296 | root->fs_info->max_extent) > num_extents) | |
1297 | return 0; | |
1298 | ||
32c00aff JB |
1299 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1300 | BTRFS_I(inode)->outstanding_extents--; | |
1301 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1302 | |
1303 | return 0; | |
1304 | } | |
1305 | ||
d352ac68 CM |
1306 | /* |
1307 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1308 | * bytes in this file, and to maintain the list of inodes that | |
1309 | * have pending delalloc work to be done. | |
1310 | */ | |
b2950863 | 1311 | static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end, |
b0c68f8b | 1312 | unsigned long old, unsigned long bits) |
291d673e | 1313 | { |
9ed74f2d | 1314 | |
75eff68e CM |
1315 | /* |
1316 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1317 | * but in this case, we are only testeing for the DELALLOC | |
1318 | * bit, which is only set or cleared with irqs on | |
1319 | */ | |
b0c68f8b | 1320 | if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1321 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9ed74f2d | 1322 | |
32c00aff JB |
1323 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1324 | BTRFS_I(inode)->outstanding_extents++; | |
1325 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
6a63209f | 1326 | btrfs_delalloc_reserve_space(root, inode, end - start + 1); |
75eff68e | 1327 | spin_lock(&root->fs_info->delalloc_lock); |
9069218d | 1328 | BTRFS_I(inode)->delalloc_bytes += end - start + 1; |
291d673e | 1329 | root->fs_info->delalloc_bytes += end - start + 1; |
ea8c2819 CM |
1330 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1331 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1332 | &root->fs_info->delalloc_inodes); | |
1333 | } | |
75eff68e | 1334 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1335 | } |
1336 | return 0; | |
1337 | } | |
1338 | ||
d352ac68 CM |
1339 | /* |
1340 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1341 | */ | |
9ed74f2d JB |
1342 | static int btrfs_clear_bit_hook(struct inode *inode, |
1343 | struct extent_state *state, unsigned long bits) | |
291d673e | 1344 | { |
75eff68e CM |
1345 | /* |
1346 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1347 | * but in this case, we are only testeing for the DELALLOC | |
1348 | * bit, which is only set or cleared with irqs on | |
1349 | */ | |
9ed74f2d | 1350 | if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1351 | struct btrfs_root *root = BTRFS_I(inode)->root; |
bcbfce8a | 1352 | |
32c00aff JB |
1353 | if (bits & EXTENT_DO_ACCOUNTING) { |
1354 | spin_lock(&BTRFS_I(inode)->accounting_lock); | |
1355 | BTRFS_I(inode)->outstanding_extents--; | |
1356 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
1357 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
1358 | } | |
9ed74f2d | 1359 | |
75eff68e | 1360 | spin_lock(&root->fs_info->delalloc_lock); |
9ed74f2d JB |
1361 | if (state->end - state->start + 1 > |
1362 | root->fs_info->delalloc_bytes) { | |
d397712b CM |
1363 | printk(KERN_INFO "btrfs warning: delalloc account " |
1364 | "%llu %llu\n", | |
9ed74f2d JB |
1365 | (unsigned long long) |
1366 | state->end - state->start + 1, | |
d397712b CM |
1367 | (unsigned long long) |
1368 | root->fs_info->delalloc_bytes); | |
6a63209f | 1369 | btrfs_delalloc_free_space(root, inode, (u64)-1); |
b0c68f8b | 1370 | root->fs_info->delalloc_bytes = 0; |
9069218d | 1371 | BTRFS_I(inode)->delalloc_bytes = 0; |
b0c68f8b | 1372 | } else { |
6a63209f | 1373 | btrfs_delalloc_free_space(root, inode, |
9ed74f2d JB |
1374 | state->end - |
1375 | state->start + 1); | |
1376 | root->fs_info->delalloc_bytes -= state->end - | |
1377 | state->start + 1; | |
1378 | BTRFS_I(inode)->delalloc_bytes -= state->end - | |
1379 | state->start + 1; | |
b0c68f8b | 1380 | } |
ea8c2819 CM |
1381 | if (BTRFS_I(inode)->delalloc_bytes == 0 && |
1382 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1383 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1384 | } | |
75eff68e | 1385 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1386 | } |
1387 | return 0; | |
1388 | } | |
1389 | ||
d352ac68 CM |
1390 | /* |
1391 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1392 | * we don't create bios that span stripes or chunks | |
1393 | */ | |
239b14b3 | 1394 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1395 | size_t size, struct bio *bio, |
1396 | unsigned long bio_flags) | |
239b14b3 CM |
1397 | { |
1398 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1399 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1400 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1401 | u64 length = 0; |
1402 | u64 map_length; | |
239b14b3 CM |
1403 | int ret; |
1404 | ||
771ed689 CM |
1405 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1406 | return 0; | |
1407 | ||
f2d8d74d | 1408 | length = bio->bi_size; |
239b14b3 CM |
1409 | map_tree = &root->fs_info->mapping_tree; |
1410 | map_length = length; | |
cea9e445 | 1411 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1412 | &map_length, NULL, 0); |
cea9e445 | 1413 | |
d397712b | 1414 | if (map_length < length + size) |
239b14b3 | 1415 | return 1; |
239b14b3 CM |
1416 | return 0; |
1417 | } | |
1418 | ||
d352ac68 CM |
1419 | /* |
1420 | * in order to insert checksums into the metadata in large chunks, | |
1421 | * we wait until bio submission time. All the pages in the bio are | |
1422 | * checksummed and sums are attached onto the ordered extent record. | |
1423 | * | |
1424 | * At IO completion time the cums attached on the ordered extent record | |
1425 | * are inserted into the btree | |
1426 | */ | |
d397712b CM |
1427 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1428 | struct bio *bio, int mirror_num, | |
1429 | unsigned long bio_flags) | |
065631f6 | 1430 | { |
065631f6 | 1431 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1432 | int ret = 0; |
e015640f | 1433 | |
d20f7043 | 1434 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1435 | BUG_ON(ret); |
4a69a410 CM |
1436 | return 0; |
1437 | } | |
e015640f | 1438 | |
4a69a410 CM |
1439 | /* |
1440 | * in order to insert checksums into the metadata in large chunks, | |
1441 | * we wait until bio submission time. All the pages in the bio are | |
1442 | * checksummed and sums are attached onto the ordered extent record. | |
1443 | * | |
1444 | * At IO completion time the cums attached on the ordered extent record | |
1445 | * are inserted into the btree | |
1446 | */ | |
b2950863 | 1447 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
4a69a410 CM |
1448 | int mirror_num, unsigned long bio_flags) |
1449 | { | |
1450 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1451 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1452 | } |
1453 | ||
d352ac68 | 1454 | /* |
cad321ad CM |
1455 | * extent_io.c submission hook. This does the right thing for csum calculation |
1456 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1457 | */ |
b2950863 | 1458 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
c8b97818 | 1459 | int mirror_num, unsigned long bio_flags) |
44b8bd7e CM |
1460 | { |
1461 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1462 | int ret = 0; | |
19b9bdb0 | 1463 | int skip_sum; |
44b8bd7e | 1464 | |
6cbff00f | 1465 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1466 | |
e6dcd2dc CM |
1467 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
1468 | BUG_ON(ret); | |
065631f6 | 1469 | |
4d1b5fb4 | 1470 | if (!(rw & (1 << BIO_RW))) { |
d20f7043 | 1471 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1472 | return btrfs_submit_compressed_read(inode, bio, |
1473 | mirror_num, bio_flags); | |
d20f7043 CM |
1474 | } else if (!skip_sum) |
1475 | btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
4d1b5fb4 | 1476 | goto mapit; |
19b9bdb0 | 1477 | } else if (!skip_sum) { |
17d217fe YZ |
1478 | /* csum items have already been cloned */ |
1479 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1480 | goto mapit; | |
19b9bdb0 CM |
1481 | /* we're doing a write, do the async checksumming */ |
1482 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1483 | inode, rw, bio, mirror_num, |
4a69a410 CM |
1484 | bio_flags, __btrfs_submit_bio_start, |
1485 | __btrfs_submit_bio_done); | |
19b9bdb0 CM |
1486 | } |
1487 | ||
0b86a832 | 1488 | mapit: |
8b712842 | 1489 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1490 | } |
6885f308 | 1491 | |
d352ac68 CM |
1492 | /* |
1493 | * given a list of ordered sums record them in the inode. This happens | |
1494 | * at IO completion time based on sums calculated at bio submission time. | |
1495 | */ | |
ba1da2f4 | 1496 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1497 | struct inode *inode, u64 file_offset, |
1498 | struct list_head *list) | |
1499 | { | |
e6dcd2dc CM |
1500 | struct btrfs_ordered_sum *sum; |
1501 | ||
1502 | btrfs_set_trans_block_group(trans, inode); | |
c6e30871 QF |
1503 | |
1504 | list_for_each_entry(sum, list, list) { | |
d20f7043 CM |
1505 | btrfs_csum_file_blocks(trans, |
1506 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1507 | } |
1508 | return 0; | |
1509 | } | |
1510 | ||
ea8c2819 CM |
1511 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end) |
1512 | { | |
d397712b | 1513 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1514 | WARN_ON(1); |
ea8c2819 CM |
1515 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
1516 | GFP_NOFS); | |
1517 | } | |
1518 | ||
d352ac68 | 1519 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1520 | struct btrfs_writepage_fixup { |
1521 | struct page *page; | |
1522 | struct btrfs_work work; | |
1523 | }; | |
1524 | ||
b2950863 | 1525 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1526 | { |
1527 | struct btrfs_writepage_fixup *fixup; | |
1528 | struct btrfs_ordered_extent *ordered; | |
1529 | struct page *page; | |
1530 | struct inode *inode; | |
1531 | u64 page_start; | |
1532 | u64 page_end; | |
1533 | ||
1534 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1535 | page = fixup->page; | |
4a096752 | 1536 | again: |
247e743c CM |
1537 | lock_page(page); |
1538 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1539 | ClearPageChecked(page); | |
1540 | goto out_page; | |
1541 | } | |
1542 | ||
1543 | inode = page->mapping->host; | |
1544 | page_start = page_offset(page); | |
1545 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1546 | ||
1547 | lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
4a096752 CM |
1548 | |
1549 | /* already ordered? We're done */ | |
8b62b72b | 1550 | if (PagePrivate2(page)) |
247e743c | 1551 | goto out; |
4a096752 CM |
1552 | |
1553 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1554 | if (ordered) { | |
1555 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, | |
1556 | page_end, GFP_NOFS); | |
1557 | unlock_page(page); | |
1558 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1559 | goto again; | |
1560 | } | |
247e743c | 1561 | |
ea8c2819 | 1562 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
247e743c CM |
1563 | ClearPageChecked(page); |
1564 | out: | |
1565 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
1566 | out_page: | |
1567 | unlock_page(page); | |
1568 | page_cache_release(page); | |
1569 | } | |
1570 | ||
1571 | /* | |
1572 | * There are a few paths in the higher layers of the kernel that directly | |
1573 | * set the page dirty bit without asking the filesystem if it is a | |
1574 | * good idea. This causes problems because we want to make sure COW | |
1575 | * properly happens and the data=ordered rules are followed. | |
1576 | * | |
c8b97818 | 1577 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1578 | * hasn't been properly setup for IO. We kick off an async process |
1579 | * to fix it up. The async helper will wait for ordered extents, set | |
1580 | * the delalloc bit and make it safe to write the page. | |
1581 | */ | |
b2950863 | 1582 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1583 | { |
1584 | struct inode *inode = page->mapping->host; | |
1585 | struct btrfs_writepage_fixup *fixup; | |
1586 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1587 | |
8b62b72b CM |
1588 | /* this page is properly in the ordered list */ |
1589 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1590 | return 0; |
1591 | ||
1592 | if (PageChecked(page)) | |
1593 | return -EAGAIN; | |
1594 | ||
1595 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1596 | if (!fixup) | |
1597 | return -EAGAIN; | |
f421950f | 1598 | |
247e743c CM |
1599 | SetPageChecked(page); |
1600 | page_cache_get(page); | |
1601 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1602 | fixup->page = page; | |
1603 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1604 | return -EAGAIN; | |
1605 | } | |
1606 | ||
d899e052 YZ |
1607 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1608 | struct inode *inode, u64 file_pos, | |
1609 | u64 disk_bytenr, u64 disk_num_bytes, | |
1610 | u64 num_bytes, u64 ram_bytes, | |
1611 | u8 compression, u8 encryption, | |
1612 | u16 other_encoding, int extent_type) | |
1613 | { | |
1614 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1615 | struct btrfs_file_extent_item *fi; | |
1616 | struct btrfs_path *path; | |
1617 | struct extent_buffer *leaf; | |
1618 | struct btrfs_key ins; | |
1619 | u64 hint; | |
1620 | int ret; | |
1621 | ||
1622 | path = btrfs_alloc_path(); | |
1623 | BUG_ON(!path); | |
1624 | ||
b9473439 | 1625 | path->leave_spinning = 1; |
a1ed835e CM |
1626 | |
1627 | /* | |
1628 | * we may be replacing one extent in the tree with another. | |
1629 | * The new extent is pinned in the extent map, and we don't want | |
1630 | * to drop it from the cache until it is completely in the btree. | |
1631 | * | |
1632 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1633 | * the caller is expected to unpin it and allow it to be merged | |
1634 | * with the others. | |
1635 | */ | |
920bbbfb YZ |
1636 | ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes, |
1637 | &hint, 0); | |
d899e052 YZ |
1638 | BUG_ON(ret); |
1639 | ||
1640 | ins.objectid = inode->i_ino; | |
1641 | ins.offset = file_pos; | |
1642 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1643 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1644 | BUG_ON(ret); | |
1645 | leaf = path->nodes[0]; | |
1646 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1647 | struct btrfs_file_extent_item); | |
1648 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1649 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1650 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1651 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1652 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1653 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1654 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1655 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1656 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1657 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1658 | |
1659 | btrfs_unlock_up_safe(path, 1); | |
1660 | btrfs_set_lock_blocking(leaf); | |
1661 | ||
d899e052 YZ |
1662 | btrfs_mark_buffer_dirty(leaf); |
1663 | ||
1664 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1665 | |
1666 | ins.objectid = disk_bytenr; | |
1667 | ins.offset = disk_num_bytes; | |
1668 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1669 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1670 | root->root_key.objectid, | |
1671 | inode->i_ino, file_pos, &ins); | |
d899e052 | 1672 | BUG_ON(ret); |
d899e052 | 1673 | btrfs_free_path(path); |
b9473439 | 1674 | |
d899e052 YZ |
1675 | return 0; |
1676 | } | |
1677 | ||
5d13a98f CM |
1678 | /* |
1679 | * helper function for btrfs_finish_ordered_io, this | |
1680 | * just reads in some of the csum leaves to prime them into ram | |
1681 | * before we start the transaction. It limits the amount of btree | |
1682 | * reads required while inside the transaction. | |
1683 | */ | |
1684 | static noinline void reada_csum(struct btrfs_root *root, | |
1685 | struct btrfs_path *path, | |
1686 | struct btrfs_ordered_extent *ordered_extent) | |
1687 | { | |
1688 | struct btrfs_ordered_sum *sum; | |
1689 | u64 bytenr; | |
1690 | ||
1691 | sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum, | |
1692 | list); | |
1693 | bytenr = sum->sums[0].bytenr; | |
1694 | ||
1695 | /* | |
1696 | * we don't care about the results, the point of this search is | |
1697 | * just to get the btree leaves into ram | |
1698 | */ | |
1699 | btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0); | |
1700 | } | |
1701 | ||
d352ac68 CM |
1702 | /* as ordered data IO finishes, this gets called so we can finish |
1703 | * an ordered extent if the range of bytes in the file it covers are | |
1704 | * fully written. | |
1705 | */ | |
211f90e6 | 1706 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1707 | { |
e6dcd2dc CM |
1708 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1709 | struct btrfs_trans_handle *trans; | |
5d13a98f | 1710 | struct btrfs_ordered_extent *ordered_extent = NULL; |
e6dcd2dc | 1711 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
b7ec40d7 | 1712 | struct btrfs_path *path; |
d899e052 | 1713 | int compressed = 0; |
e6dcd2dc CM |
1714 | int ret; |
1715 | ||
1716 | ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1); | |
ba1da2f4 | 1717 | if (!ret) |
e6dcd2dc | 1718 | return 0; |
e6dcd2dc | 1719 | |
b7ec40d7 CM |
1720 | /* |
1721 | * before we join the transaction, try to do some of our IO. | |
1722 | * This will limit the amount of IO that we have to do with | |
1723 | * the transaction running. We're unlikely to need to do any | |
1724 | * IO if the file extents are new, the disk_i_size checks | |
1725 | * covers the most common case. | |
1726 | */ | |
1727 | if (start < BTRFS_I(inode)->disk_i_size) { | |
1728 | path = btrfs_alloc_path(); | |
1729 | if (path) { | |
1730 | ret = btrfs_lookup_file_extent(NULL, root, path, | |
1731 | inode->i_ino, | |
1732 | start, 0); | |
5d13a98f CM |
1733 | ordered_extent = btrfs_lookup_ordered_extent(inode, |
1734 | start); | |
1735 | if (!list_empty(&ordered_extent->list)) { | |
1736 | btrfs_release_path(root, path); | |
1737 | reada_csum(root, path, ordered_extent); | |
1738 | } | |
b7ec40d7 CM |
1739 | btrfs_free_path(path); |
1740 | } | |
1741 | } | |
1742 | ||
5d13a98f CM |
1743 | if (!ordered_extent) |
1744 | ordered_extent = btrfs_lookup_ordered_extent(inode, start); | |
e6dcd2dc | 1745 | BUG_ON(!ordered_extent); |
c2167754 YZ |
1746 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
1747 | BUG_ON(!list_empty(&ordered_extent->list)); | |
1748 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); | |
1749 | if (!ret) { | |
1750 | trans = btrfs_join_transaction(root, 1); | |
1751 | ret = btrfs_update_inode(trans, root, inode); | |
1752 | BUG_ON(ret); | |
1753 | btrfs_end_transaction(trans, root); | |
1754 | } | |
1755 | goto out; | |
1756 | } | |
e6dcd2dc CM |
1757 | |
1758 | lock_extent(io_tree, ordered_extent->file_offset, | |
1759 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1760 | GFP_NOFS); | |
1761 | ||
c2167754 YZ |
1762 | trans = btrfs_join_transaction(root, 1); |
1763 | ||
c8b97818 | 1764 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
d899e052 YZ |
1765 | compressed = 1; |
1766 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { | |
1767 | BUG_ON(compressed); | |
920bbbfb | 1768 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1769 | ordered_extent->file_offset, |
1770 | ordered_extent->file_offset + | |
1771 | ordered_extent->len); | |
1772 | BUG_ON(ret); | |
1773 | } else { | |
1774 | ret = insert_reserved_file_extent(trans, inode, | |
1775 | ordered_extent->file_offset, | |
1776 | ordered_extent->start, | |
1777 | ordered_extent->disk_len, | |
1778 | ordered_extent->len, | |
1779 | ordered_extent->len, | |
1780 | compressed, 0, 0, | |
1781 | BTRFS_FILE_EXTENT_REG); | |
a1ed835e CM |
1782 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1783 | ordered_extent->file_offset, | |
1784 | ordered_extent->len); | |
d899e052 YZ |
1785 | BUG_ON(ret); |
1786 | } | |
e6dcd2dc CM |
1787 | unlock_extent(io_tree, ordered_extent->file_offset, |
1788 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1789 | GFP_NOFS); | |
1790 | add_pending_csums(trans, inode, ordered_extent->file_offset, | |
1791 | &ordered_extent->list); | |
1792 | ||
c2167754 YZ |
1793 | /* this also removes the ordered extent from the tree */ |
1794 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); | |
1795 | ret = btrfs_update_inode(trans, root, inode); | |
1796 | BUG_ON(ret); | |
1797 | btrfs_end_transaction(trans, root); | |
1798 | out: | |
e6dcd2dc CM |
1799 | /* once for us */ |
1800 | btrfs_put_ordered_extent(ordered_extent); | |
1801 | /* once for the tree */ | |
1802 | btrfs_put_ordered_extent(ordered_extent); | |
1803 | ||
e6dcd2dc CM |
1804 | return 0; |
1805 | } | |
1806 | ||
b2950863 | 1807 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1808 | struct extent_state *state, int uptodate) |
1809 | { | |
8b62b72b | 1810 | ClearPagePrivate2(page); |
211f90e6 CM |
1811 | return btrfs_finish_ordered_io(page->mapping->host, start, end); |
1812 | } | |
1813 | ||
d352ac68 CM |
1814 | /* |
1815 | * When IO fails, either with EIO or csum verification fails, we | |
1816 | * try other mirrors that might have a good copy of the data. This | |
1817 | * io_failure_record is used to record state as we go through all the | |
1818 | * mirrors. If another mirror has good data, the page is set up to date | |
1819 | * and things continue. If a good mirror can't be found, the original | |
1820 | * bio end_io callback is called to indicate things have failed. | |
1821 | */ | |
7e38326f CM |
1822 | struct io_failure_record { |
1823 | struct page *page; | |
1824 | u64 start; | |
1825 | u64 len; | |
1826 | u64 logical; | |
d20f7043 | 1827 | unsigned long bio_flags; |
7e38326f CM |
1828 | int last_mirror; |
1829 | }; | |
1830 | ||
b2950863 | 1831 | static int btrfs_io_failed_hook(struct bio *failed_bio, |
1259ab75 CM |
1832 | struct page *page, u64 start, u64 end, |
1833 | struct extent_state *state) | |
7e38326f CM |
1834 | { |
1835 | struct io_failure_record *failrec = NULL; | |
1836 | u64 private; | |
1837 | struct extent_map *em; | |
1838 | struct inode *inode = page->mapping->host; | |
1839 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1840 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1841 | struct bio *bio; |
1842 | int num_copies; | |
1843 | int ret; | |
1259ab75 | 1844 | int rw; |
7e38326f CM |
1845 | u64 logical; |
1846 | ||
1847 | ret = get_state_private(failure_tree, start, &private); | |
1848 | if (ret) { | |
7e38326f CM |
1849 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1850 | if (!failrec) | |
1851 | return -ENOMEM; | |
1852 | failrec->start = start; | |
1853 | failrec->len = end - start + 1; | |
1854 | failrec->last_mirror = 0; | |
d20f7043 | 1855 | failrec->bio_flags = 0; |
7e38326f | 1856 | |
890871be | 1857 | read_lock(&em_tree->lock); |
3b951516 CM |
1858 | em = lookup_extent_mapping(em_tree, start, failrec->len); |
1859 | if (em->start > start || em->start + em->len < start) { | |
1860 | free_extent_map(em); | |
1861 | em = NULL; | |
1862 | } | |
890871be | 1863 | read_unlock(&em_tree->lock); |
7e38326f CM |
1864 | |
1865 | if (!em || IS_ERR(em)) { | |
1866 | kfree(failrec); | |
1867 | return -EIO; | |
1868 | } | |
1869 | logical = start - em->start; | |
1870 | logical = em->block_start + logical; | |
d20f7043 CM |
1871 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
1872 | logical = em->block_start; | |
1873 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
1874 | } | |
7e38326f CM |
1875 | failrec->logical = logical; |
1876 | free_extent_map(em); | |
1877 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1878 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1879 | set_state_private(failure_tree, start, |
1880 | (u64)(unsigned long)failrec); | |
7e38326f | 1881 | } else { |
587f7704 | 1882 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1883 | } |
1884 | num_copies = btrfs_num_copies( | |
1885 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1886 | failrec->logical, failrec->len); | |
1887 | failrec->last_mirror++; | |
1888 | if (!state) { | |
cad321ad | 1889 | spin_lock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1890 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, |
1891 | failrec->start, | |
1892 | EXTENT_LOCKED); | |
1893 | if (state && state->start != failrec->start) | |
1894 | state = NULL; | |
cad321ad | 1895 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1896 | } |
1897 | if (!state || failrec->last_mirror > num_copies) { | |
1898 | set_state_private(failure_tree, failrec->start, 0); | |
1899 | clear_extent_bits(failure_tree, failrec->start, | |
1900 | failrec->start + failrec->len - 1, | |
1901 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1902 | kfree(failrec); | |
1903 | return -EIO; | |
1904 | } | |
1905 | bio = bio_alloc(GFP_NOFS, 1); | |
1906 | bio->bi_private = state; | |
1907 | bio->bi_end_io = failed_bio->bi_end_io; | |
1908 | bio->bi_sector = failrec->logical >> 9; | |
1909 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1910 | bio->bi_size = 0; |
d20f7043 | 1911 | |
7e38326f | 1912 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
1259ab75 CM |
1913 | if (failed_bio->bi_rw & (1 << BIO_RW)) |
1914 | rw = WRITE; | |
1915 | else | |
1916 | rw = READ; | |
1917 | ||
1918 | BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, | |
c8b97818 | 1919 | failrec->last_mirror, |
d20f7043 | 1920 | failrec->bio_flags); |
1259ab75 CM |
1921 | return 0; |
1922 | } | |
1923 | ||
d352ac68 CM |
1924 | /* |
1925 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1926 | * to see if we need to process or clean up an io_failure_record | |
1927 | */ | |
b2950863 | 1928 | static int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1259ab75 CM |
1929 | { |
1930 | u64 private; | |
1931 | u64 private_failure; | |
1932 | struct io_failure_record *failure; | |
1933 | int ret; | |
1934 | ||
1935 | private = 0; | |
1936 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1937 | (u64)-1, 1, EXTENT_DIRTY)) { | |
1938 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1939 | start, &private_failure); | |
1940 | if (ret == 0) { | |
1941 | failure = (struct io_failure_record *)(unsigned long) | |
1942 | private_failure; | |
1943 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1944 | failure->start, 0); | |
1945 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1946 | failure->start, | |
1947 | failure->start + failure->len - 1, | |
1948 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1949 | GFP_NOFS); | |
1950 | kfree(failure); | |
1951 | } | |
1952 | } | |
7e38326f CM |
1953 | return 0; |
1954 | } | |
1955 | ||
d352ac68 CM |
1956 | /* |
1957 | * when reads are done, we need to check csums to verify the data is correct | |
1958 | * if there's a match, we allow the bio to finish. If not, we go through | |
1959 | * the io_failure_record routines to find good copies | |
1960 | */ | |
b2950863 | 1961 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1962 | struct extent_state *state) |
07157aac | 1963 | { |
35ebb934 | 1964 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1965 | struct inode *inode = page->mapping->host; |
d1310b2e | 1966 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1967 | char *kaddr; |
aadfeb6e | 1968 | u64 private = ~(u32)0; |
07157aac | 1969 | int ret; |
ff79f819 CM |
1970 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1971 | u32 csum = ~(u32)0; | |
d1310b2e | 1972 | |
d20f7043 CM |
1973 | if (PageChecked(page)) { |
1974 | ClearPageChecked(page); | |
1975 | goto good; | |
1976 | } | |
6cbff00f CH |
1977 | |
1978 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
17d217fe YZ |
1979 | return 0; |
1980 | ||
1981 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 1982 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
1983 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
1984 | GFP_NOFS); | |
b6cda9bc | 1985 | return 0; |
17d217fe | 1986 | } |
d20f7043 | 1987 | |
c2e639f0 | 1988 | if (state && state->start == start) { |
70dec807 CM |
1989 | private = state->private; |
1990 | ret = 0; | |
1991 | } else { | |
1992 | ret = get_state_private(io_tree, start, &private); | |
1993 | } | |
9ab86c8e | 1994 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 1995 | if (ret) |
07157aac | 1996 | goto zeroit; |
d397712b | 1997 | |
ff79f819 CM |
1998 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1999 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 2000 | if (csum != private) |
07157aac | 2001 | goto zeroit; |
d397712b | 2002 | |
9ab86c8e | 2003 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 2004 | good: |
7e38326f CM |
2005 | /* if the io failure tree for this inode is non-empty, |
2006 | * check to see if we've recovered from a failed IO | |
2007 | */ | |
1259ab75 | 2008 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
2009 | return 0; |
2010 | ||
2011 | zeroit: | |
193f284d CM |
2012 | if (printk_ratelimit()) { |
2013 | printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u " | |
2014 | "private %llu\n", page->mapping->host->i_ino, | |
2015 | (unsigned long long)start, csum, | |
2016 | (unsigned long long)private); | |
2017 | } | |
db94535d CM |
2018 | memset(kaddr + offset, 1, end - start + 1); |
2019 | flush_dcache_page(page); | |
9ab86c8e | 2020 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
2021 | if (private == 0) |
2022 | return 0; | |
7e38326f | 2023 | return -EIO; |
07157aac | 2024 | } |
b888db2b | 2025 | |
24bbcf04 YZ |
2026 | struct delayed_iput { |
2027 | struct list_head list; | |
2028 | struct inode *inode; | |
2029 | }; | |
2030 | ||
2031 | void btrfs_add_delayed_iput(struct inode *inode) | |
2032 | { | |
2033 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2034 | struct delayed_iput *delayed; | |
2035 | ||
2036 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2037 | return; | |
2038 | ||
2039 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2040 | delayed->inode = inode; | |
2041 | ||
2042 | spin_lock(&fs_info->delayed_iput_lock); | |
2043 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2044 | spin_unlock(&fs_info->delayed_iput_lock); | |
2045 | } | |
2046 | ||
2047 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2048 | { | |
2049 | LIST_HEAD(list); | |
2050 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2051 | struct delayed_iput *delayed; | |
2052 | int empty; | |
2053 | ||
2054 | spin_lock(&fs_info->delayed_iput_lock); | |
2055 | empty = list_empty(&fs_info->delayed_iputs); | |
2056 | spin_unlock(&fs_info->delayed_iput_lock); | |
2057 | if (empty) | |
2058 | return; | |
2059 | ||
2060 | down_read(&root->fs_info->cleanup_work_sem); | |
2061 | spin_lock(&fs_info->delayed_iput_lock); | |
2062 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2063 | spin_unlock(&fs_info->delayed_iput_lock); | |
2064 | ||
2065 | while (!list_empty(&list)) { | |
2066 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2067 | list_del(&delayed->list); | |
2068 | iput(delayed->inode); | |
2069 | kfree(delayed); | |
2070 | } | |
2071 | up_read(&root->fs_info->cleanup_work_sem); | |
2072 | } | |
2073 | ||
7b128766 JB |
2074 | /* |
2075 | * This creates an orphan entry for the given inode in case something goes | |
2076 | * wrong in the middle of an unlink/truncate. | |
2077 | */ | |
2078 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2079 | { | |
2080 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2081 | int ret = 0; | |
2082 | ||
bcc63abb | 2083 | spin_lock(&root->list_lock); |
7b128766 JB |
2084 | |
2085 | /* already on the orphan list, we're good */ | |
2086 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 2087 | spin_unlock(&root->list_lock); |
7b128766 JB |
2088 | return 0; |
2089 | } | |
2090 | ||
2091 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2092 | ||
bcc63abb | 2093 | spin_unlock(&root->list_lock); |
7b128766 JB |
2094 | |
2095 | /* | |
2096 | * insert an orphan item to track this unlinked/truncated file | |
2097 | */ | |
2098 | ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); | |
2099 | ||
2100 | return ret; | |
2101 | } | |
2102 | ||
2103 | /* | |
2104 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2105 | * item for this particular inode. | |
2106 | */ | |
2107 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2108 | { | |
2109 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2110 | int ret = 0; | |
2111 | ||
bcc63abb | 2112 | spin_lock(&root->list_lock); |
7b128766 JB |
2113 | |
2114 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 2115 | spin_unlock(&root->list_lock); |
7b128766 JB |
2116 | return 0; |
2117 | } | |
2118 | ||
2119 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
2120 | if (!trans) { | |
bcc63abb | 2121 | spin_unlock(&root->list_lock); |
7b128766 JB |
2122 | return 0; |
2123 | } | |
2124 | ||
bcc63abb | 2125 | spin_unlock(&root->list_lock); |
7b128766 JB |
2126 | |
2127 | ret = btrfs_del_orphan_item(trans, root, inode->i_ino); | |
2128 | ||
2129 | return ret; | |
2130 | } | |
2131 | ||
2132 | /* | |
2133 | * this cleans up any orphans that may be left on the list from the last use | |
2134 | * of this root. | |
2135 | */ | |
2136 | void btrfs_orphan_cleanup(struct btrfs_root *root) | |
2137 | { | |
2138 | struct btrfs_path *path; | |
2139 | struct extent_buffer *leaf; | |
2140 | struct btrfs_item *item; | |
2141 | struct btrfs_key key, found_key; | |
2142 | struct btrfs_trans_handle *trans; | |
2143 | struct inode *inode; | |
2144 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
2145 | ||
c71bf099 | 2146 | if (!xchg(&root->clean_orphans, 0)) |
7b128766 | 2147 | return; |
c71bf099 YZ |
2148 | |
2149 | path = btrfs_alloc_path(); | |
2150 | BUG_ON(!path); | |
7b128766 JB |
2151 | path->reada = -1; |
2152 | ||
2153 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2154 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2155 | key.offset = (u64)-1; | |
2156 | ||
7b128766 JB |
2157 | while (1) { |
2158 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2159 | if (ret < 0) { | |
2160 | printk(KERN_ERR "Error searching slot for orphan: %d" | |
2161 | "\n", ret); | |
2162 | break; | |
2163 | } | |
2164 | ||
2165 | /* | |
2166 | * if ret == 0 means we found what we were searching for, which | |
2167 | * is weird, but possible, so only screw with path if we didnt | |
2168 | * find the key and see if we have stuff that matches | |
2169 | */ | |
2170 | if (ret > 0) { | |
2171 | if (path->slots[0] == 0) | |
2172 | break; | |
2173 | path->slots[0]--; | |
2174 | } | |
2175 | ||
2176 | /* pull out the item */ | |
2177 | leaf = path->nodes[0]; | |
2178 | item = btrfs_item_nr(leaf, path->slots[0]); | |
2179 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2180 | ||
2181 | /* make sure the item matches what we want */ | |
2182 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2183 | break; | |
2184 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2185 | break; | |
2186 | ||
2187 | /* release the path since we're done with it */ | |
2188 | btrfs_release_path(root, path); | |
2189 | ||
2190 | /* | |
2191 | * this is where we are basically btrfs_lookup, without the | |
2192 | * crossing root thing. we store the inode number in the | |
2193 | * offset of the orphan item. | |
2194 | */ | |
5d4f98a2 YZ |
2195 | found_key.objectid = found_key.offset; |
2196 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2197 | found_key.offset = 0; | |
2198 | inode = btrfs_iget(root->fs_info->sb, &found_key, root); | |
2199 | if (IS_ERR(inode)) | |
7b128766 JB |
2200 | break; |
2201 | ||
7b128766 JB |
2202 | /* |
2203 | * add this inode to the orphan list so btrfs_orphan_del does | |
2204 | * the proper thing when we hit it | |
2205 | */ | |
bcc63abb | 2206 | spin_lock(&root->list_lock); |
7b128766 | 2207 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
bcc63abb | 2208 | spin_unlock(&root->list_lock); |
7b128766 JB |
2209 | |
2210 | /* | |
2211 | * if this is a bad inode, means we actually succeeded in | |
2212 | * removing the inode, but not the orphan record, which means | |
2213 | * we need to manually delete the orphan since iput will just | |
2214 | * do a destroy_inode | |
2215 | */ | |
2216 | if (is_bad_inode(inode)) { | |
5b21f2ed | 2217 | trans = btrfs_start_transaction(root, 1); |
7b128766 | 2218 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 2219 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2220 | iput(inode); |
2221 | continue; | |
2222 | } | |
2223 | ||
2224 | /* if we have links, this was a truncate, lets do that */ | |
2225 | if (inode->i_nlink) { | |
2226 | nr_truncate++; | |
2227 | btrfs_truncate(inode); | |
2228 | } else { | |
2229 | nr_unlink++; | |
2230 | } | |
2231 | ||
2232 | /* this will do delete_inode and everything for us */ | |
2233 | iput(inode); | |
2234 | } | |
2235 | ||
2236 | if (nr_unlink) | |
2237 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2238 | if (nr_truncate) | |
2239 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
2240 | ||
2241 | btrfs_free_path(path); | |
7b128766 JB |
2242 | } |
2243 | ||
46a53cca CM |
2244 | /* |
2245 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2246 | * don't find any xattrs, we know there can't be any acls. | |
2247 | * | |
2248 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2249 | */ | |
2250 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2251 | int slot, u64 objectid) | |
2252 | { | |
2253 | u32 nritems = btrfs_header_nritems(leaf); | |
2254 | struct btrfs_key found_key; | |
2255 | int scanned = 0; | |
2256 | ||
2257 | slot++; | |
2258 | while (slot < nritems) { | |
2259 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2260 | ||
2261 | /* we found a different objectid, there must not be acls */ | |
2262 | if (found_key.objectid != objectid) | |
2263 | return 0; | |
2264 | ||
2265 | /* we found an xattr, assume we've got an acl */ | |
2266 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2267 | return 1; | |
2268 | ||
2269 | /* | |
2270 | * we found a key greater than an xattr key, there can't | |
2271 | * be any acls later on | |
2272 | */ | |
2273 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2274 | return 0; | |
2275 | ||
2276 | slot++; | |
2277 | scanned++; | |
2278 | ||
2279 | /* | |
2280 | * it goes inode, inode backrefs, xattrs, extents, | |
2281 | * so if there are a ton of hard links to an inode there can | |
2282 | * be a lot of backrefs. Don't waste time searching too hard, | |
2283 | * this is just an optimization | |
2284 | */ | |
2285 | if (scanned >= 8) | |
2286 | break; | |
2287 | } | |
2288 | /* we hit the end of the leaf before we found an xattr or | |
2289 | * something larger than an xattr. We have to assume the inode | |
2290 | * has acls | |
2291 | */ | |
2292 | return 1; | |
2293 | } | |
2294 | ||
d352ac68 CM |
2295 | /* |
2296 | * read an inode from the btree into the in-memory inode | |
2297 | */ | |
5d4f98a2 | 2298 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2299 | { |
2300 | struct btrfs_path *path; | |
5f39d397 | 2301 | struct extent_buffer *leaf; |
39279cc3 | 2302 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2303 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2304 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2305 | struct btrfs_key location; | |
46a53cca | 2306 | int maybe_acls; |
39279cc3 | 2307 | u64 alloc_group_block; |
618e21d5 | 2308 | u32 rdev; |
39279cc3 CM |
2309 | int ret; |
2310 | ||
2311 | path = btrfs_alloc_path(); | |
2312 | BUG_ON(!path); | |
39279cc3 | 2313 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2314 | |
39279cc3 | 2315 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2316 | if (ret) |
39279cc3 | 2317 | goto make_bad; |
39279cc3 | 2318 | |
5f39d397 CM |
2319 | leaf = path->nodes[0]; |
2320 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
2321 | struct btrfs_inode_item); | |
2322 | ||
2323 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
2324 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
2325 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
2326 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2327 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2328 | |
2329 | tspec = btrfs_inode_atime(inode_item); | |
2330 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2331 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2332 | ||
2333 | tspec = btrfs_inode_mtime(inode_item); | |
2334 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2335 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2336 | ||
2337 | tspec = btrfs_inode_ctime(inode_item); | |
2338 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2339 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2340 | ||
a76a3cd4 | 2341 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2342 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2343 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2344 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2345 | inode->i_rdev = 0; |
5f39d397 CM |
2346 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2347 | ||
aec7477b | 2348 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2349 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
aec7477b | 2350 | |
5f39d397 | 2351 | alloc_group_block = btrfs_inode_block_group(leaf, inode_item); |
b4ce94de | 2352 | |
46a53cca CM |
2353 | /* |
2354 | * try to precache a NULL acl entry for files that don't have | |
2355 | * any xattrs or acls | |
2356 | */ | |
2357 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino); | |
72c04902 AV |
2358 | if (!maybe_acls) |
2359 | cache_no_acl(inode); | |
46a53cca | 2360 | |
d2fb3437 YZ |
2361 | BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, |
2362 | alloc_group_block, 0); | |
39279cc3 CM |
2363 | btrfs_free_path(path); |
2364 | inode_item = NULL; | |
2365 | ||
39279cc3 | 2366 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2367 | case S_IFREG: |
2368 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2369 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2370 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2371 | inode->i_fop = &btrfs_file_operations; |
2372 | inode->i_op = &btrfs_file_inode_operations; | |
2373 | break; | |
2374 | case S_IFDIR: | |
2375 | inode->i_fop = &btrfs_dir_file_operations; | |
2376 | if (root == root->fs_info->tree_root) | |
2377 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2378 | else | |
2379 | inode->i_op = &btrfs_dir_inode_operations; | |
2380 | break; | |
2381 | case S_IFLNK: | |
2382 | inode->i_op = &btrfs_symlink_inode_operations; | |
2383 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2384 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2385 | break; |
618e21d5 | 2386 | default: |
0279b4cd | 2387 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2388 | init_special_inode(inode, inode->i_mode, rdev); |
2389 | break; | |
39279cc3 | 2390 | } |
6cbff00f CH |
2391 | |
2392 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2393 | return; |
2394 | ||
2395 | make_bad: | |
39279cc3 | 2396 | btrfs_free_path(path); |
39279cc3 CM |
2397 | make_bad_inode(inode); |
2398 | } | |
2399 | ||
d352ac68 CM |
2400 | /* |
2401 | * given a leaf and an inode, copy the inode fields into the leaf | |
2402 | */ | |
e02119d5 CM |
2403 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2404 | struct extent_buffer *leaf, | |
5f39d397 | 2405 | struct btrfs_inode_item *item, |
39279cc3 CM |
2406 | struct inode *inode) |
2407 | { | |
5f39d397 CM |
2408 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2409 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2410 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2411 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2412 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2413 | ||
2414 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2415 | inode->i_atime.tv_sec); | |
2416 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2417 | inode->i_atime.tv_nsec); | |
2418 | ||
2419 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2420 | inode->i_mtime.tv_sec); | |
2421 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2422 | inode->i_mtime.tv_nsec); | |
2423 | ||
2424 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2425 | inode->i_ctime.tv_sec); | |
2426 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2427 | inode->i_ctime.tv_nsec); | |
2428 | ||
a76a3cd4 | 2429 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2430 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2431 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2432 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2433 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2434 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d2fb3437 | 2435 | btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group); |
39279cc3 CM |
2436 | } |
2437 | ||
d352ac68 CM |
2438 | /* |
2439 | * copy everything in the in-memory inode into the btree. | |
2440 | */ | |
d397712b CM |
2441 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2442 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2443 | { |
2444 | struct btrfs_inode_item *inode_item; | |
2445 | struct btrfs_path *path; | |
5f39d397 | 2446 | struct extent_buffer *leaf; |
39279cc3 CM |
2447 | int ret; |
2448 | ||
2449 | path = btrfs_alloc_path(); | |
2450 | BUG_ON(!path); | |
b9473439 | 2451 | path->leave_spinning = 1; |
39279cc3 CM |
2452 | ret = btrfs_lookup_inode(trans, root, path, |
2453 | &BTRFS_I(inode)->location, 1); | |
2454 | if (ret) { | |
2455 | if (ret > 0) | |
2456 | ret = -ENOENT; | |
2457 | goto failed; | |
2458 | } | |
2459 | ||
b4ce94de | 2460 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2461 | leaf = path->nodes[0]; |
2462 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
39279cc3 CM |
2463 | struct btrfs_inode_item); |
2464 | ||
e02119d5 | 2465 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2466 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2467 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2468 | ret = 0; |
2469 | failed: | |
39279cc3 CM |
2470 | btrfs_free_path(path); |
2471 | return ret; | |
2472 | } | |
2473 | ||
2474 | ||
d352ac68 CM |
2475 | /* |
2476 | * unlink helper that gets used here in inode.c and in the tree logging | |
2477 | * recovery code. It remove a link in a directory with a given name, and | |
2478 | * also drops the back refs in the inode to the directory | |
2479 | */ | |
e02119d5 CM |
2480 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2481 | struct btrfs_root *root, | |
2482 | struct inode *dir, struct inode *inode, | |
2483 | const char *name, int name_len) | |
39279cc3 CM |
2484 | { |
2485 | struct btrfs_path *path; | |
39279cc3 | 2486 | int ret = 0; |
5f39d397 | 2487 | struct extent_buffer *leaf; |
39279cc3 | 2488 | struct btrfs_dir_item *di; |
5f39d397 | 2489 | struct btrfs_key key; |
aec7477b | 2490 | u64 index; |
39279cc3 CM |
2491 | |
2492 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2493 | if (!path) { |
2494 | ret = -ENOMEM; | |
2495 | goto err; | |
2496 | } | |
2497 | ||
b9473439 | 2498 | path->leave_spinning = 1; |
39279cc3 CM |
2499 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, |
2500 | name, name_len, -1); | |
2501 | if (IS_ERR(di)) { | |
2502 | ret = PTR_ERR(di); | |
2503 | goto err; | |
2504 | } | |
2505 | if (!di) { | |
2506 | ret = -ENOENT; | |
2507 | goto err; | |
2508 | } | |
5f39d397 CM |
2509 | leaf = path->nodes[0]; |
2510 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2511 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2512 | if (ret) |
2513 | goto err; | |
39279cc3 CM |
2514 | btrfs_release_path(root, path); |
2515 | ||
aec7477b | 2516 | ret = btrfs_del_inode_ref(trans, root, name, name_len, |
e02119d5 CM |
2517 | inode->i_ino, |
2518 | dir->i_ino, &index); | |
aec7477b | 2519 | if (ret) { |
d397712b | 2520 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
aec7477b | 2521 | "inode %lu parent %lu\n", name_len, name, |
e02119d5 | 2522 | inode->i_ino, dir->i_ino); |
aec7477b JB |
2523 | goto err; |
2524 | } | |
2525 | ||
39279cc3 | 2526 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, |
aec7477b | 2527 | index, name, name_len, -1); |
39279cc3 CM |
2528 | if (IS_ERR(di)) { |
2529 | ret = PTR_ERR(di); | |
2530 | goto err; | |
2531 | } | |
2532 | if (!di) { | |
2533 | ret = -ENOENT; | |
2534 | goto err; | |
2535 | } | |
2536 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
925baedd | 2537 | btrfs_release_path(root, path); |
39279cc3 | 2538 | |
e02119d5 CM |
2539 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
2540 | inode, dir->i_ino); | |
49eb7e46 | 2541 | BUG_ON(ret != 0 && ret != -ENOENT); |
e02119d5 CM |
2542 | |
2543 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2544 | dir, index); | |
2545 | BUG_ON(ret); | |
39279cc3 CM |
2546 | err: |
2547 | btrfs_free_path(path); | |
e02119d5 CM |
2548 | if (ret) |
2549 | goto out; | |
2550 | ||
2551 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2552 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2553 | btrfs_update_inode(trans, root, dir); | |
2554 | btrfs_drop_nlink(inode); | |
2555 | ret = btrfs_update_inode(trans, root, inode); | |
e02119d5 | 2556 | out: |
39279cc3 CM |
2557 | return ret; |
2558 | } | |
2559 | ||
2560 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2561 | { | |
2562 | struct btrfs_root *root; | |
2563 | struct btrfs_trans_handle *trans; | |
7b128766 | 2564 | struct inode *inode = dentry->d_inode; |
39279cc3 | 2565 | int ret; |
1832a6d5 | 2566 | unsigned long nr = 0; |
39279cc3 CM |
2567 | |
2568 | root = BTRFS_I(dir)->root; | |
1832a6d5 | 2569 | |
5df6a9f6 JB |
2570 | /* |
2571 | * 5 items for unlink inode | |
2572 | * 1 for orphan | |
2573 | */ | |
2574 | ret = btrfs_reserve_metadata_space(root, 6); | |
2575 | if (ret) | |
2576 | return ret; | |
2577 | ||
39279cc3 | 2578 | trans = btrfs_start_transaction(root, 1); |
5df6a9f6 JB |
2579 | if (IS_ERR(trans)) { |
2580 | btrfs_unreserve_metadata_space(root, 6); | |
2581 | return PTR_ERR(trans); | |
2582 | } | |
5f39d397 | 2583 | |
39279cc3 | 2584 | btrfs_set_trans_block_group(trans, dir); |
12fcfd22 CM |
2585 | |
2586 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); | |
2587 | ||
e02119d5 CM |
2588 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2589 | dentry->d_name.name, dentry->d_name.len); | |
7b128766 JB |
2590 | |
2591 | if (inode->i_nlink == 0) | |
2592 | ret = btrfs_orphan_add(trans, inode); | |
2593 | ||
d3c2fdcf | 2594 | nr = trans->blocks_used; |
5f39d397 | 2595 | |
89ce8a63 | 2596 | btrfs_end_transaction_throttle(trans, root); |
5df6a9f6 | 2597 | btrfs_unreserve_metadata_space(root, 6); |
d3c2fdcf | 2598 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2599 | return ret; |
2600 | } | |
2601 | ||
4df27c4d YZ |
2602 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
2603 | struct btrfs_root *root, | |
2604 | struct inode *dir, u64 objectid, | |
2605 | const char *name, int name_len) | |
2606 | { | |
2607 | struct btrfs_path *path; | |
2608 | struct extent_buffer *leaf; | |
2609 | struct btrfs_dir_item *di; | |
2610 | struct btrfs_key key; | |
2611 | u64 index; | |
2612 | int ret; | |
2613 | ||
2614 | path = btrfs_alloc_path(); | |
2615 | if (!path) | |
2616 | return -ENOMEM; | |
2617 | ||
2618 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
2619 | name, name_len, -1); | |
2620 | BUG_ON(!di || IS_ERR(di)); | |
2621 | ||
2622 | leaf = path->nodes[0]; | |
2623 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2624 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2625 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2626 | BUG_ON(ret); | |
2627 | btrfs_release_path(root, path); | |
2628 | ||
2629 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
2630 | objectid, root->root_key.objectid, | |
2631 | dir->i_ino, &index, name, name_len); | |
2632 | if (ret < 0) { | |
2633 | BUG_ON(ret != -ENOENT); | |
2634 | di = btrfs_search_dir_index_item(root, path, dir->i_ino, | |
2635 | name, name_len); | |
2636 | BUG_ON(!di || IS_ERR(di)); | |
2637 | ||
2638 | leaf = path->nodes[0]; | |
2639 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2640 | btrfs_release_path(root, path); | |
2641 | index = key.offset; | |
2642 | } | |
2643 | ||
2644 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, | |
2645 | index, name, name_len, -1); | |
2646 | BUG_ON(!di || IS_ERR(di)); | |
2647 | ||
2648 | leaf = path->nodes[0]; | |
2649 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2650 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2651 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2652 | BUG_ON(ret); | |
2653 | btrfs_release_path(root, path); | |
2654 | ||
2655 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2656 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2657 | ret = btrfs_update_inode(trans, root, dir); | |
2658 | BUG_ON(ret); | |
2659 | dir->i_sb->s_dirt = 1; | |
2660 | ||
2661 | btrfs_free_path(path); | |
2662 | return 0; | |
2663 | } | |
2664 | ||
39279cc3 CM |
2665 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
2666 | { | |
2667 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 2668 | int err = 0; |
39279cc3 CM |
2669 | int ret; |
2670 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
39279cc3 | 2671 | struct btrfs_trans_handle *trans; |
1832a6d5 | 2672 | unsigned long nr = 0; |
39279cc3 | 2673 | |
3394e160 | 2674 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
4df27c4d | 2675 | inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
2676 | return -ENOTEMPTY; |
2677 | ||
5df6a9f6 JB |
2678 | ret = btrfs_reserve_metadata_space(root, 5); |
2679 | if (ret) | |
2680 | return ret; | |
2681 | ||
39279cc3 | 2682 | trans = btrfs_start_transaction(root, 1); |
5df6a9f6 JB |
2683 | if (IS_ERR(trans)) { |
2684 | btrfs_unreserve_metadata_space(root, 5); | |
2685 | return PTR_ERR(trans); | |
2686 | } | |
2687 | ||
39279cc3 | 2688 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 | 2689 | |
4df27c4d YZ |
2690 | if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
2691 | err = btrfs_unlink_subvol(trans, root, dir, | |
2692 | BTRFS_I(inode)->location.objectid, | |
2693 | dentry->d_name.name, | |
2694 | dentry->d_name.len); | |
2695 | goto out; | |
2696 | } | |
2697 | ||
7b128766 JB |
2698 | err = btrfs_orphan_add(trans, inode); |
2699 | if (err) | |
4df27c4d | 2700 | goto out; |
7b128766 | 2701 | |
39279cc3 | 2702 | /* now the directory is empty */ |
e02119d5 CM |
2703 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2704 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 2705 | if (!err) |
dbe674a9 | 2706 | btrfs_i_size_write(inode, 0); |
4df27c4d | 2707 | out: |
d3c2fdcf | 2708 | nr = trans->blocks_used; |
89ce8a63 | 2709 | ret = btrfs_end_transaction_throttle(trans, root); |
5df6a9f6 | 2710 | btrfs_unreserve_metadata_space(root, 5); |
d3c2fdcf | 2711 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 2712 | |
39279cc3 CM |
2713 | if (ret && !err) |
2714 | err = ret; | |
2715 | return err; | |
2716 | } | |
2717 | ||
d20f7043 | 2718 | #if 0 |
323ac95b CM |
2719 | /* |
2720 | * when truncating bytes in a file, it is possible to avoid reading | |
2721 | * the leaves that contain only checksum items. This can be the | |
2722 | * majority of the IO required to delete a large file, but it must | |
2723 | * be done carefully. | |
2724 | * | |
2725 | * The keys in the level just above the leaves are checked to make sure | |
2726 | * the lowest key in a given leaf is a csum key, and starts at an offset | |
2727 | * after the new size. | |
2728 | * | |
2729 | * Then the key for the next leaf is checked to make sure it also has | |
2730 | * a checksum item for the same file. If it does, we know our target leaf | |
2731 | * contains only checksum items, and it can be safely freed without reading | |
2732 | * it. | |
2733 | * | |
2734 | * This is just an optimization targeted at large files. It may do | |
2735 | * nothing. It will return 0 unless things went badly. | |
2736 | */ | |
2737 | static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, | |
2738 | struct btrfs_root *root, | |
2739 | struct btrfs_path *path, | |
2740 | struct inode *inode, u64 new_size) | |
2741 | { | |
2742 | struct btrfs_key key; | |
2743 | int ret; | |
2744 | int nritems; | |
2745 | struct btrfs_key found_key; | |
2746 | struct btrfs_key other_key; | |
5b84e8d6 YZ |
2747 | struct btrfs_leaf_ref *ref; |
2748 | u64 leaf_gen; | |
2749 | u64 leaf_start; | |
323ac95b CM |
2750 | |
2751 | path->lowest_level = 1; | |
2752 | key.objectid = inode->i_ino; | |
2753 | key.type = BTRFS_CSUM_ITEM_KEY; | |
2754 | key.offset = new_size; | |
2755 | again: | |
2756 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2757 | if (ret < 0) | |
2758 | goto out; | |
2759 | ||
2760 | if (path->nodes[1] == NULL) { | |
2761 | ret = 0; | |
2762 | goto out; | |
2763 | } | |
2764 | ret = 0; | |
2765 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); | |
2766 | nritems = btrfs_header_nritems(path->nodes[1]); | |
2767 | ||
2768 | if (!nritems) | |
2769 | goto out; | |
2770 | ||
2771 | if (path->slots[1] >= nritems) | |
2772 | goto next_node; | |
2773 | ||
2774 | /* did we find a key greater than anything we want to delete? */ | |
2775 | if (found_key.objectid > inode->i_ino || | |
2776 | (found_key.objectid == inode->i_ino && found_key.type > key.type)) | |
2777 | goto out; | |
2778 | ||
2779 | /* we check the next key in the node to make sure the leave contains | |
2780 | * only checksum items. This comparison doesn't work if our | |
2781 | * leaf is the last one in the node | |
2782 | */ | |
2783 | if (path->slots[1] + 1 >= nritems) { | |
2784 | next_node: | |
2785 | /* search forward from the last key in the node, this | |
2786 | * will bring us into the next node in the tree | |
2787 | */ | |
2788 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); | |
2789 | ||
2790 | /* unlikely, but we inc below, so check to be safe */ | |
2791 | if (found_key.offset == (u64)-1) | |
2792 | goto out; | |
2793 | ||
2794 | /* search_forward needs a path with locks held, do the | |
2795 | * search again for the original key. It is possible | |
2796 | * this will race with a balance and return a path that | |
2797 | * we could modify, but this drop is just an optimization | |
2798 | * and is allowed to miss some leaves. | |
2799 | */ | |
2800 | btrfs_release_path(root, path); | |
2801 | found_key.offset++; | |
2802 | ||
2803 | /* setup a max key for search_forward */ | |
2804 | other_key.offset = (u64)-1; | |
2805 | other_key.type = key.type; | |
2806 | other_key.objectid = key.objectid; | |
2807 | ||
2808 | path->keep_locks = 1; | |
2809 | ret = btrfs_search_forward(root, &found_key, &other_key, | |
2810 | path, 0, 0); | |
2811 | path->keep_locks = 0; | |
2812 | if (ret || found_key.objectid != key.objectid || | |
2813 | found_key.type != key.type) { | |
2814 | ret = 0; | |
2815 | goto out; | |
2816 | } | |
2817 | ||
2818 | key.offset = found_key.offset; | |
2819 | btrfs_release_path(root, path); | |
2820 | cond_resched(); | |
2821 | goto again; | |
2822 | } | |
2823 | ||
2824 | /* we know there's one more slot after us in the tree, | |
2825 | * read that key so we can verify it is also a checksum item | |
2826 | */ | |
2827 | btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); | |
2828 | ||
2829 | if (found_key.objectid < inode->i_ino) | |
2830 | goto next_key; | |
2831 | ||
2832 | if (found_key.type != key.type || found_key.offset < new_size) | |
2833 | goto next_key; | |
2834 | ||
2835 | /* | |
2836 | * if the key for the next leaf isn't a csum key from this objectid, | |
2837 | * we can't be sure there aren't good items inside this leaf. | |
2838 | * Bail out | |
2839 | */ | |
2840 | if (other_key.objectid != inode->i_ino || other_key.type != key.type) | |
2841 | goto out; | |
2842 | ||
5b84e8d6 YZ |
2843 | leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); |
2844 | leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); | |
323ac95b CM |
2845 | /* |
2846 | * it is safe to delete this leaf, it contains only | |
2847 | * csum items from this inode at an offset >= new_size | |
2848 | */ | |
5b84e8d6 | 2849 | ret = btrfs_del_leaf(trans, root, path, leaf_start); |
323ac95b CM |
2850 | BUG_ON(ret); |
2851 | ||
5b84e8d6 YZ |
2852 | if (root->ref_cows && leaf_gen < trans->transid) { |
2853 | ref = btrfs_alloc_leaf_ref(root, 0); | |
2854 | if (ref) { | |
2855 | ref->root_gen = root->root_key.offset; | |
2856 | ref->bytenr = leaf_start; | |
2857 | ref->owner = 0; | |
2858 | ref->generation = leaf_gen; | |
2859 | ref->nritems = 0; | |
2860 | ||
bd56b302 CM |
2861 | btrfs_sort_leaf_ref(ref); |
2862 | ||
5b84e8d6 YZ |
2863 | ret = btrfs_add_leaf_ref(root, ref, 0); |
2864 | WARN_ON(ret); | |
2865 | btrfs_free_leaf_ref(root, ref); | |
2866 | } else { | |
2867 | WARN_ON(1); | |
2868 | } | |
2869 | } | |
323ac95b CM |
2870 | next_key: |
2871 | btrfs_release_path(root, path); | |
2872 | ||
2873 | if (other_key.objectid == inode->i_ino && | |
2874 | other_key.type == key.type && other_key.offset > key.offset) { | |
2875 | key.offset = other_key.offset; | |
2876 | cond_resched(); | |
2877 | goto again; | |
2878 | } | |
2879 | ret = 0; | |
2880 | out: | |
2881 | /* fixup any changes we've made to the path */ | |
2882 | path->lowest_level = 0; | |
2883 | path->keep_locks = 0; | |
2884 | btrfs_release_path(root, path); | |
2885 | return ret; | |
2886 | } | |
2887 | ||
d20f7043 CM |
2888 | #endif |
2889 | ||
39279cc3 CM |
2890 | /* |
2891 | * this can truncate away extent items, csum items and directory items. | |
2892 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 2893 | * any higher than new_size |
39279cc3 CM |
2894 | * |
2895 | * csum items that cross the new i_size are truncated to the new size | |
2896 | * as well. | |
7b128766 JB |
2897 | * |
2898 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
2899 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 2900 | */ |
8082510e YZ |
2901 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
2902 | struct btrfs_root *root, | |
2903 | struct inode *inode, | |
2904 | u64 new_size, u32 min_type) | |
39279cc3 | 2905 | { |
39279cc3 | 2906 | struct btrfs_path *path; |
5f39d397 | 2907 | struct extent_buffer *leaf; |
39279cc3 | 2908 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
2909 | struct btrfs_key key; |
2910 | struct btrfs_key found_key; | |
39279cc3 | 2911 | u64 extent_start = 0; |
db94535d | 2912 | u64 extent_num_bytes = 0; |
5d4f98a2 | 2913 | u64 extent_offset = 0; |
39279cc3 | 2914 | u64 item_end = 0; |
8082510e YZ |
2915 | u64 mask = root->sectorsize - 1; |
2916 | u32 found_type = (u8)-1; | |
39279cc3 CM |
2917 | int found_extent; |
2918 | int del_item; | |
85e21bac CM |
2919 | int pending_del_nr = 0; |
2920 | int pending_del_slot = 0; | |
179e29e4 | 2921 | int extent_type = -1; |
771ed689 | 2922 | int encoding; |
8082510e YZ |
2923 | int ret; |
2924 | int err = 0; | |
2925 | ||
2926 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 2927 | |
e02119d5 | 2928 | if (root->ref_cows) |
5b21f2ed | 2929 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
8082510e | 2930 | |
39279cc3 CM |
2931 | path = btrfs_alloc_path(); |
2932 | BUG_ON(!path); | |
33c17ad5 | 2933 | path->reada = -1; |
5f39d397 | 2934 | |
39279cc3 CM |
2935 | key.objectid = inode->i_ino; |
2936 | key.offset = (u64)-1; | |
5f39d397 CM |
2937 | key.type = (u8)-1; |
2938 | ||
85e21bac | 2939 | search_again: |
b9473439 | 2940 | path->leave_spinning = 1; |
85e21bac | 2941 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
2942 | if (ret < 0) { |
2943 | err = ret; | |
2944 | goto out; | |
2945 | } | |
d397712b | 2946 | |
85e21bac | 2947 | if (ret > 0) { |
e02119d5 CM |
2948 | /* there are no items in the tree for us to truncate, we're |
2949 | * done | |
2950 | */ | |
8082510e YZ |
2951 | if (path->slots[0] == 0) |
2952 | goto out; | |
85e21bac CM |
2953 | path->slots[0]--; |
2954 | } | |
2955 | ||
d397712b | 2956 | while (1) { |
39279cc3 | 2957 | fi = NULL; |
5f39d397 CM |
2958 | leaf = path->nodes[0]; |
2959 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2960 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 2961 | encoding = 0; |
39279cc3 | 2962 | |
5f39d397 | 2963 | if (found_key.objectid != inode->i_ino) |
39279cc3 | 2964 | break; |
5f39d397 | 2965 | |
85e21bac | 2966 | if (found_type < min_type) |
39279cc3 CM |
2967 | break; |
2968 | ||
5f39d397 | 2969 | item_end = found_key.offset; |
39279cc3 | 2970 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 2971 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 2972 | struct btrfs_file_extent_item); |
179e29e4 | 2973 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
2974 | encoding = btrfs_file_extent_compression(leaf, fi); |
2975 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
2976 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
2977 | ||
179e29e4 | 2978 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 2979 | item_end += |
db94535d | 2980 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 2981 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 2982 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 2983 | fi); |
39279cc3 | 2984 | } |
008630c1 | 2985 | item_end--; |
39279cc3 | 2986 | } |
8082510e YZ |
2987 | if (found_type > min_type) { |
2988 | del_item = 1; | |
2989 | } else { | |
2990 | if (item_end < new_size) | |
b888db2b | 2991 | break; |
8082510e YZ |
2992 | if (found_key.offset >= new_size) |
2993 | del_item = 1; | |
2994 | else | |
2995 | del_item = 0; | |
39279cc3 | 2996 | } |
39279cc3 | 2997 | found_extent = 0; |
39279cc3 | 2998 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
2999 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3000 | goto delete; | |
3001 | ||
3002 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3003 | u64 num_dec; |
db94535d | 3004 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 3005 | if (!del_item && !encoding) { |
db94535d CM |
3006 | u64 orig_num_bytes = |
3007 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3008 | extent_num_bytes = new_size - |
5f39d397 | 3009 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3010 | extent_num_bytes = extent_num_bytes & |
3011 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3012 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3013 | extent_num_bytes); | |
3014 | num_dec = (orig_num_bytes - | |
9069218d | 3015 | extent_num_bytes); |
e02119d5 | 3016 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3017 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3018 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3019 | } else { |
db94535d CM |
3020 | extent_num_bytes = |
3021 | btrfs_file_extent_disk_num_bytes(leaf, | |
3022 | fi); | |
5d4f98a2 YZ |
3023 | extent_offset = found_key.offset - |
3024 | btrfs_file_extent_offset(leaf, fi); | |
3025 | ||
39279cc3 | 3026 | /* FIXME blocksize != 4096 */ |
9069218d | 3027 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3028 | if (extent_start != 0) { |
3029 | found_extent = 1; | |
e02119d5 | 3030 | if (root->ref_cows) |
a76a3cd4 | 3031 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3032 | } |
39279cc3 | 3033 | } |
9069218d | 3034 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3035 | /* |
3036 | * we can't truncate inline items that have had | |
3037 | * special encodings | |
3038 | */ | |
3039 | if (!del_item && | |
3040 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3041 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3042 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3043 | u32 size = new_size - found_key.offset; |
3044 | ||
3045 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3046 | inode_sub_bytes(inode, item_end + 1 - |
3047 | new_size); | |
e02119d5 CM |
3048 | } |
3049 | size = | |
3050 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 3051 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 3052 | size, 1); |
9069218d | 3053 | BUG_ON(ret); |
e02119d5 | 3054 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3055 | inode_sub_bytes(inode, item_end + 1 - |
3056 | found_key.offset); | |
9069218d | 3057 | } |
39279cc3 | 3058 | } |
179e29e4 | 3059 | delete: |
39279cc3 | 3060 | if (del_item) { |
85e21bac CM |
3061 | if (!pending_del_nr) { |
3062 | /* no pending yet, add ourselves */ | |
3063 | pending_del_slot = path->slots[0]; | |
3064 | pending_del_nr = 1; | |
3065 | } else if (pending_del_nr && | |
3066 | path->slots[0] + 1 == pending_del_slot) { | |
3067 | /* hop on the pending chunk */ | |
3068 | pending_del_nr++; | |
3069 | pending_del_slot = path->slots[0]; | |
3070 | } else { | |
d397712b | 3071 | BUG(); |
85e21bac | 3072 | } |
39279cc3 CM |
3073 | } else { |
3074 | break; | |
3075 | } | |
5d4f98a2 | 3076 | if (found_extent && root->ref_cows) { |
b9473439 | 3077 | btrfs_set_path_blocking(path); |
39279cc3 | 3078 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3079 | extent_num_bytes, 0, |
3080 | btrfs_header_owner(leaf), | |
3081 | inode->i_ino, extent_offset); | |
39279cc3 CM |
3082 | BUG_ON(ret); |
3083 | } | |
85e21bac | 3084 | |
8082510e YZ |
3085 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3086 | break; | |
3087 | ||
3088 | if (path->slots[0] == 0 || | |
3089 | path->slots[0] != pending_del_slot) { | |
3090 | if (root->ref_cows) { | |
3091 | err = -EAGAIN; | |
3092 | goto out; | |
3093 | } | |
3094 | if (pending_del_nr) { | |
3095 | ret = btrfs_del_items(trans, root, path, | |
3096 | pending_del_slot, | |
3097 | pending_del_nr); | |
3098 | BUG_ON(ret); | |
3099 | pending_del_nr = 0; | |
3100 | } | |
85e21bac CM |
3101 | btrfs_release_path(root, path); |
3102 | goto search_again; | |
8082510e YZ |
3103 | } else { |
3104 | path->slots[0]--; | |
85e21bac | 3105 | } |
39279cc3 | 3106 | } |
8082510e | 3107 | out: |
85e21bac CM |
3108 | if (pending_del_nr) { |
3109 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3110 | pending_del_nr); | |
3111 | } | |
39279cc3 | 3112 | btrfs_free_path(path); |
8082510e | 3113 | return err; |
39279cc3 CM |
3114 | } |
3115 | ||
3116 | /* | |
3117 | * taken from block_truncate_page, but does cow as it zeros out | |
3118 | * any bytes left in the last page in the file. | |
3119 | */ | |
3120 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3121 | { | |
3122 | struct inode *inode = mapping->host; | |
db94535d | 3123 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3124 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3125 | struct btrfs_ordered_extent *ordered; | |
3126 | char *kaddr; | |
db94535d | 3127 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3128 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3129 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3130 | struct page *page; | |
39279cc3 | 3131 | int ret = 0; |
a52d9a80 | 3132 | u64 page_start; |
e6dcd2dc | 3133 | u64 page_end; |
39279cc3 CM |
3134 | |
3135 | if ((offset & (blocksize - 1)) == 0) | |
3136 | goto out; | |
5d5e103a JB |
3137 | ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); |
3138 | if (ret) | |
3139 | goto out; | |
3140 | ||
3141 | ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); | |
3142 | if (ret) | |
3143 | goto out; | |
39279cc3 CM |
3144 | |
3145 | ret = -ENOMEM; | |
211c17f5 | 3146 | again: |
39279cc3 | 3147 | page = grab_cache_page(mapping, index); |
5d5e103a JB |
3148 | if (!page) { |
3149 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
3150 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
39279cc3 | 3151 | goto out; |
5d5e103a | 3152 | } |
e6dcd2dc CM |
3153 | |
3154 | page_start = page_offset(page); | |
3155 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3156 | ||
39279cc3 | 3157 | if (!PageUptodate(page)) { |
9ebefb18 | 3158 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3159 | lock_page(page); |
211c17f5 CM |
3160 | if (page->mapping != mapping) { |
3161 | unlock_page(page); | |
3162 | page_cache_release(page); | |
3163 | goto again; | |
3164 | } | |
39279cc3 CM |
3165 | if (!PageUptodate(page)) { |
3166 | ret = -EIO; | |
89642229 | 3167 | goto out_unlock; |
39279cc3 CM |
3168 | } |
3169 | } | |
211c17f5 | 3170 | wait_on_page_writeback(page); |
e6dcd2dc CM |
3171 | |
3172 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3173 | set_page_extent_mapped(page); | |
3174 | ||
3175 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3176 | if (ordered) { | |
3177 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3178 | unlock_page(page); | |
3179 | page_cache_release(page); | |
eb84ae03 | 3180 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3181 | btrfs_put_ordered_extent(ordered); |
3182 | goto again; | |
3183 | } | |
3184 | ||
5d5e103a JB |
3185 | clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
3186 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, | |
3187 | GFP_NOFS); | |
3188 | ||
9ed74f2d JB |
3189 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end); |
3190 | if (ret) { | |
3191 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3192 | goto out_unlock; | |
3193 | } | |
3194 | ||
e6dcd2dc CM |
3195 | ret = 0; |
3196 | if (offset != PAGE_CACHE_SIZE) { | |
3197 | kaddr = kmap(page); | |
3198 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3199 | flush_dcache_page(page); | |
3200 | kunmap(page); | |
3201 | } | |
247e743c | 3202 | ClearPageChecked(page); |
e6dcd2dc CM |
3203 | set_page_dirty(page); |
3204 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
39279cc3 | 3205 | |
89642229 | 3206 | out_unlock: |
5d5e103a JB |
3207 | if (ret) |
3208 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
3209 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
39279cc3 CM |
3210 | unlock_page(page); |
3211 | page_cache_release(page); | |
3212 | out: | |
3213 | return ret; | |
3214 | } | |
3215 | ||
9036c102 | 3216 | int btrfs_cont_expand(struct inode *inode, loff_t size) |
39279cc3 | 3217 | { |
9036c102 YZ |
3218 | struct btrfs_trans_handle *trans; |
3219 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3220 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
3221 | struct extent_map *em; | |
3222 | u64 mask = root->sectorsize - 1; | |
3223 | u64 hole_start = (inode->i_size + mask) & ~mask; | |
3224 | u64 block_end = (size + mask) & ~mask; | |
3225 | u64 last_byte; | |
3226 | u64 cur_offset; | |
3227 | u64 hole_size; | |
9ed74f2d | 3228 | int err = 0; |
39279cc3 | 3229 | |
9036c102 YZ |
3230 | if (size <= hole_start) |
3231 | return 0; | |
3232 | ||
9036c102 YZ |
3233 | while (1) { |
3234 | struct btrfs_ordered_extent *ordered; | |
3235 | btrfs_wait_ordered_range(inode, hole_start, | |
3236 | block_end - hole_start); | |
3237 | lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3238 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); | |
3239 | if (!ordered) | |
3240 | break; | |
3241 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3242 | btrfs_put_ordered_extent(ordered); | |
3243 | } | |
39279cc3 | 3244 | |
9036c102 YZ |
3245 | cur_offset = hole_start; |
3246 | while (1) { | |
3247 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3248 | block_end - cur_offset, 0); | |
3249 | BUG_ON(IS_ERR(em) || !em); | |
3250 | last_byte = min(extent_map_end(em), block_end); | |
3251 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3252 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
771ed689 | 3253 | u64 hint_byte = 0; |
9036c102 | 3254 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3255 | |
8082510e | 3256 | err = btrfs_reserve_metadata_space(root, 2); |
9ed74f2d JB |
3257 | if (err) |
3258 | break; | |
3259 | ||
8082510e YZ |
3260 | trans = btrfs_start_transaction(root, 1); |
3261 | btrfs_set_trans_block_group(trans, inode); | |
3262 | ||
3263 | err = btrfs_drop_extents(trans, inode, cur_offset, | |
3264 | cur_offset + hole_size, | |
3265 | &hint_byte, 1); | |
3266 | BUG_ON(err); | |
3267 | ||
9036c102 YZ |
3268 | err = btrfs_insert_file_extent(trans, root, |
3269 | inode->i_ino, cur_offset, 0, | |
3270 | 0, hole_size, 0, hole_size, | |
3271 | 0, 0, 0); | |
8082510e YZ |
3272 | BUG_ON(err); |
3273 | ||
9036c102 YZ |
3274 | btrfs_drop_extent_cache(inode, hole_start, |
3275 | last_byte - 1, 0); | |
8082510e YZ |
3276 | |
3277 | btrfs_end_transaction(trans, root); | |
3278 | btrfs_unreserve_metadata_space(root, 2); | |
9036c102 YZ |
3279 | } |
3280 | free_extent_map(em); | |
3281 | cur_offset = last_byte; | |
8082510e | 3282 | if (cur_offset >= block_end) |
9036c102 YZ |
3283 | break; |
3284 | } | |
1832a6d5 | 3285 | |
9036c102 YZ |
3286 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); |
3287 | return err; | |
3288 | } | |
39279cc3 | 3289 | |
8082510e YZ |
3290 | static int btrfs_setattr_size(struct inode *inode, struct iattr *attr) |
3291 | { | |
3292 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3293 | struct btrfs_trans_handle *trans; | |
3294 | unsigned long nr; | |
3295 | int ret; | |
3296 | ||
3297 | if (attr->ia_size == inode->i_size) | |
3298 | return 0; | |
3299 | ||
3300 | if (attr->ia_size > inode->i_size) { | |
3301 | unsigned long limit; | |
3302 | limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; | |
3303 | if (attr->ia_size > inode->i_sb->s_maxbytes) | |
3304 | return -EFBIG; | |
3305 | if (limit != RLIM_INFINITY && attr->ia_size > limit) { | |
3306 | send_sig(SIGXFSZ, current, 0); | |
3307 | return -EFBIG; | |
3308 | } | |
3309 | } | |
3310 | ||
3311 | ret = btrfs_reserve_metadata_space(root, 1); | |
3312 | if (ret) | |
3313 | return ret; | |
3314 | ||
3315 | trans = btrfs_start_transaction(root, 1); | |
3316 | btrfs_set_trans_block_group(trans, inode); | |
3317 | ||
3318 | ret = btrfs_orphan_add(trans, inode); | |
3319 | BUG_ON(ret); | |
3320 | ||
3321 | nr = trans->blocks_used; | |
3322 | btrfs_end_transaction(trans, root); | |
3323 | btrfs_unreserve_metadata_space(root, 1); | |
3324 | btrfs_btree_balance_dirty(root, nr); | |
3325 | ||
3326 | if (attr->ia_size > inode->i_size) { | |
3327 | ret = btrfs_cont_expand(inode, attr->ia_size); | |
3328 | if (ret) { | |
3329 | btrfs_truncate(inode); | |
3330 | return ret; | |
3331 | } | |
3332 | ||
3333 | i_size_write(inode, attr->ia_size); | |
3334 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); | |
3335 | ||
3336 | trans = btrfs_start_transaction(root, 1); | |
3337 | btrfs_set_trans_block_group(trans, inode); | |
3338 | ||
3339 | ret = btrfs_update_inode(trans, root, inode); | |
3340 | BUG_ON(ret); | |
3341 | if (inode->i_nlink > 0) { | |
3342 | ret = btrfs_orphan_del(trans, inode); | |
3343 | BUG_ON(ret); | |
3344 | } | |
3345 | nr = trans->blocks_used; | |
3346 | btrfs_end_transaction(trans, root); | |
3347 | btrfs_btree_balance_dirty(root, nr); | |
3348 | return 0; | |
3349 | } | |
3350 | ||
3351 | /* | |
3352 | * We're truncating a file that used to have good data down to | |
3353 | * zero. Make sure it gets into the ordered flush list so that | |
3354 | * any new writes get down to disk quickly. | |
3355 | */ | |
3356 | if (attr->ia_size == 0) | |
3357 | BTRFS_I(inode)->ordered_data_close = 1; | |
3358 | ||
3359 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ | |
3360 | ret = vmtruncate(inode, attr->ia_size); | |
3361 | BUG_ON(ret); | |
3362 | ||
3363 | return 0; | |
3364 | } | |
3365 | ||
9036c102 YZ |
3366 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3367 | { | |
3368 | struct inode *inode = dentry->d_inode; | |
3369 | int err; | |
39279cc3 | 3370 | |
9036c102 YZ |
3371 | err = inode_change_ok(inode, attr); |
3372 | if (err) | |
3373 | return err; | |
2bf5a725 | 3374 | |
5a3f23d5 | 3375 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
8082510e YZ |
3376 | err = btrfs_setattr_size(inode, attr); |
3377 | if (err) | |
3378 | return err; | |
39279cc3 | 3379 | } |
8082510e | 3380 | attr->ia_valid &= ~ATTR_SIZE; |
9036c102 | 3381 | |
8082510e YZ |
3382 | if (attr->ia_valid) |
3383 | err = inode_setattr(inode, attr); | |
33268eaf JB |
3384 | |
3385 | if (!err && ((attr->ia_valid & ATTR_MODE))) | |
3386 | err = btrfs_acl_chmod(inode); | |
39279cc3 CM |
3387 | return err; |
3388 | } | |
61295eb8 | 3389 | |
39279cc3 CM |
3390 | void btrfs_delete_inode(struct inode *inode) |
3391 | { | |
3392 | struct btrfs_trans_handle *trans; | |
3393 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 3394 | unsigned long nr; |
39279cc3 CM |
3395 | int ret; |
3396 | ||
3397 | truncate_inode_pages(&inode->i_data, 0); | |
3398 | if (is_bad_inode(inode)) { | |
7b128766 | 3399 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3400 | goto no_delete; |
3401 | } | |
4a096752 | 3402 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3403 | |
c71bf099 YZ |
3404 | if (root->fs_info->log_root_recovering) { |
3405 | BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan)); | |
3406 | goto no_delete; | |
3407 | } | |
3408 | ||
76dda93c YZ |
3409 | if (inode->i_nlink > 0) { |
3410 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3411 | goto no_delete; | |
3412 | } | |
3413 | ||
dbe674a9 | 3414 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3415 | |
8082510e YZ |
3416 | while (1) { |
3417 | trans = btrfs_start_transaction(root, 1); | |
3418 | btrfs_set_trans_block_group(trans, inode); | |
3419 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); | |
7b128766 | 3420 | |
8082510e YZ |
3421 | if (ret != -EAGAIN) |
3422 | break; | |
85e21bac | 3423 | |
8082510e YZ |
3424 | nr = trans->blocks_used; |
3425 | btrfs_end_transaction(trans, root); | |
3426 | trans = NULL; | |
3427 | btrfs_btree_balance_dirty(root, nr); | |
3428 | } | |
5f39d397 | 3429 | |
8082510e YZ |
3430 | if (ret == 0) { |
3431 | ret = btrfs_orphan_del(trans, inode); | |
3432 | BUG_ON(ret); | |
3433 | } | |
54aa1f4d | 3434 | |
d3c2fdcf | 3435 | nr = trans->blocks_used; |
54aa1f4d | 3436 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3437 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3438 | no_delete: |
3439 | clear_inode(inode); | |
8082510e | 3440 | return; |
39279cc3 CM |
3441 | } |
3442 | ||
3443 | /* | |
3444 | * this returns the key found in the dir entry in the location pointer. | |
3445 | * If no dir entries were found, location->objectid is 0. | |
3446 | */ | |
3447 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3448 | struct btrfs_key *location) | |
3449 | { | |
3450 | const char *name = dentry->d_name.name; | |
3451 | int namelen = dentry->d_name.len; | |
3452 | struct btrfs_dir_item *di; | |
3453 | struct btrfs_path *path; | |
3454 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3455 | int ret = 0; |
39279cc3 CM |
3456 | |
3457 | path = btrfs_alloc_path(); | |
3458 | BUG_ON(!path); | |
3954401f | 3459 | |
39279cc3 CM |
3460 | di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, |
3461 | namelen, 0); | |
0d9f7f3e Y |
3462 | if (IS_ERR(di)) |
3463 | ret = PTR_ERR(di); | |
d397712b CM |
3464 | |
3465 | if (!di || IS_ERR(di)) | |
3954401f | 3466 | goto out_err; |
d397712b | 3467 | |
5f39d397 | 3468 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3469 | out: |
39279cc3 CM |
3470 | btrfs_free_path(path); |
3471 | return ret; | |
3954401f CM |
3472 | out_err: |
3473 | location->objectid = 0; | |
3474 | goto out; | |
39279cc3 CM |
3475 | } |
3476 | ||
3477 | /* | |
3478 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3479 | * needs to be changed to reflect the root directory of the tree root. This | |
3480 | * is kind of like crossing a mount point. | |
3481 | */ | |
3482 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3483 | struct inode *dir, |
3484 | struct dentry *dentry, | |
3485 | struct btrfs_key *location, | |
3486 | struct btrfs_root **sub_root) | |
39279cc3 | 3487 | { |
4df27c4d YZ |
3488 | struct btrfs_path *path; |
3489 | struct btrfs_root *new_root; | |
3490 | struct btrfs_root_ref *ref; | |
3491 | struct extent_buffer *leaf; | |
3492 | int ret; | |
3493 | int err = 0; | |
39279cc3 | 3494 | |
4df27c4d YZ |
3495 | path = btrfs_alloc_path(); |
3496 | if (!path) { | |
3497 | err = -ENOMEM; | |
3498 | goto out; | |
3499 | } | |
39279cc3 | 3500 | |
4df27c4d YZ |
3501 | err = -ENOENT; |
3502 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3503 | BTRFS_I(dir)->root->root_key.objectid, | |
3504 | location->objectid); | |
3505 | if (ret) { | |
3506 | if (ret < 0) | |
3507 | err = ret; | |
3508 | goto out; | |
3509 | } | |
39279cc3 | 3510 | |
4df27c4d YZ |
3511 | leaf = path->nodes[0]; |
3512 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
3513 | if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino || | |
3514 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) | |
3515 | goto out; | |
39279cc3 | 3516 | |
4df27c4d YZ |
3517 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3518 | (unsigned long)(ref + 1), | |
3519 | dentry->d_name.len); | |
3520 | if (ret) | |
3521 | goto out; | |
3522 | ||
3523 | btrfs_release_path(root->fs_info->tree_root, path); | |
3524 | ||
3525 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3526 | if (IS_ERR(new_root)) { | |
3527 | err = PTR_ERR(new_root); | |
3528 | goto out; | |
3529 | } | |
3530 | ||
3531 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3532 | err = -ENOENT; | |
3533 | goto out; | |
3534 | } | |
3535 | ||
3536 | *sub_root = new_root; | |
3537 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3538 | location->type = BTRFS_INODE_ITEM_KEY; | |
3539 | location->offset = 0; | |
3540 | err = 0; | |
3541 | out: | |
3542 | btrfs_free_path(path); | |
3543 | return err; | |
39279cc3 CM |
3544 | } |
3545 | ||
5d4f98a2 YZ |
3546 | static void inode_tree_add(struct inode *inode) |
3547 | { | |
3548 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3549 | struct btrfs_inode *entry; | |
03e860bd FNP |
3550 | struct rb_node **p; |
3551 | struct rb_node *parent; | |
03e860bd FNP |
3552 | again: |
3553 | p = &root->inode_tree.rb_node; | |
3554 | parent = NULL; | |
5d4f98a2 | 3555 | |
76dda93c YZ |
3556 | if (hlist_unhashed(&inode->i_hash)) |
3557 | return; | |
3558 | ||
5d4f98a2 YZ |
3559 | spin_lock(&root->inode_lock); |
3560 | while (*p) { | |
3561 | parent = *p; | |
3562 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3563 | ||
3564 | if (inode->i_ino < entry->vfs_inode.i_ino) | |
03e860bd | 3565 | p = &parent->rb_left; |
5d4f98a2 | 3566 | else if (inode->i_ino > entry->vfs_inode.i_ino) |
03e860bd | 3567 | p = &parent->rb_right; |
5d4f98a2 YZ |
3568 | else { |
3569 | WARN_ON(!(entry->vfs_inode.i_state & | |
3570 | (I_WILL_FREE | I_FREEING | I_CLEAR))); | |
03e860bd FNP |
3571 | rb_erase(parent, &root->inode_tree); |
3572 | RB_CLEAR_NODE(parent); | |
3573 | spin_unlock(&root->inode_lock); | |
3574 | goto again; | |
5d4f98a2 YZ |
3575 | } |
3576 | } | |
3577 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3578 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3579 | spin_unlock(&root->inode_lock); | |
3580 | } | |
3581 | ||
3582 | static void inode_tree_del(struct inode *inode) | |
3583 | { | |
3584 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3585 | int empty = 0; |
5d4f98a2 | 3586 | |
03e860bd | 3587 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3588 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3589 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3590 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3591 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3592 | } |
03e860bd | 3593 | spin_unlock(&root->inode_lock); |
76dda93c YZ |
3594 | |
3595 | if (empty && btrfs_root_refs(&root->root_item) == 0) { | |
3596 | synchronize_srcu(&root->fs_info->subvol_srcu); | |
3597 | spin_lock(&root->inode_lock); | |
3598 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3599 | spin_unlock(&root->inode_lock); | |
3600 | if (empty) | |
3601 | btrfs_add_dead_root(root); | |
3602 | } | |
3603 | } | |
3604 | ||
3605 | int btrfs_invalidate_inodes(struct btrfs_root *root) | |
3606 | { | |
3607 | struct rb_node *node; | |
3608 | struct rb_node *prev; | |
3609 | struct btrfs_inode *entry; | |
3610 | struct inode *inode; | |
3611 | u64 objectid = 0; | |
3612 | ||
3613 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3614 | ||
3615 | spin_lock(&root->inode_lock); | |
3616 | again: | |
3617 | node = root->inode_tree.rb_node; | |
3618 | prev = NULL; | |
3619 | while (node) { | |
3620 | prev = node; | |
3621 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3622 | ||
3623 | if (objectid < entry->vfs_inode.i_ino) | |
3624 | node = node->rb_left; | |
3625 | else if (objectid > entry->vfs_inode.i_ino) | |
3626 | node = node->rb_right; | |
3627 | else | |
3628 | break; | |
3629 | } | |
3630 | if (!node) { | |
3631 | while (prev) { | |
3632 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
3633 | if (objectid <= entry->vfs_inode.i_ino) { | |
3634 | node = prev; | |
3635 | break; | |
3636 | } | |
3637 | prev = rb_next(prev); | |
3638 | } | |
3639 | } | |
3640 | while (node) { | |
3641 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3642 | objectid = entry->vfs_inode.i_ino + 1; | |
3643 | inode = igrab(&entry->vfs_inode); | |
3644 | if (inode) { | |
3645 | spin_unlock(&root->inode_lock); | |
3646 | if (atomic_read(&inode->i_count) > 1) | |
3647 | d_prune_aliases(inode); | |
3648 | /* | |
3649 | * btrfs_drop_inode will remove it from | |
3650 | * the inode cache when its usage count | |
3651 | * hits zero. | |
3652 | */ | |
3653 | iput(inode); | |
3654 | cond_resched(); | |
3655 | spin_lock(&root->inode_lock); | |
3656 | goto again; | |
3657 | } | |
3658 | ||
3659 | if (cond_resched_lock(&root->inode_lock)) | |
3660 | goto again; | |
3661 | ||
3662 | node = rb_next(node); | |
3663 | } | |
3664 | spin_unlock(&root->inode_lock); | |
3665 | return 0; | |
5d4f98a2 YZ |
3666 | } |
3667 | ||
e02119d5 | 3668 | static noinline void init_btrfs_i(struct inode *inode) |
39279cc3 | 3669 | { |
e02119d5 CM |
3670 | struct btrfs_inode *bi = BTRFS_I(inode); |
3671 | ||
e02119d5 | 3672 | bi->generation = 0; |
c3027eb5 | 3673 | bi->sequence = 0; |
e02119d5 | 3674 | bi->last_trans = 0; |
257c62e1 | 3675 | bi->last_sub_trans = 0; |
e02119d5 CM |
3676 | bi->logged_trans = 0; |
3677 | bi->delalloc_bytes = 0; | |
6a63209f | 3678 | bi->reserved_bytes = 0; |
e02119d5 CM |
3679 | bi->disk_i_size = 0; |
3680 | bi->flags = 0; | |
3681 | bi->index_cnt = (u64)-1; | |
12fcfd22 | 3682 | bi->last_unlink_trans = 0; |
2757495c | 3683 | bi->ordered_data_close = 0; |
d1310b2e CM |
3684 | extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS); |
3685 | extent_io_tree_init(&BTRFS_I(inode)->io_tree, | |
b888db2b | 3686 | inode->i_mapping, GFP_NOFS); |
7e38326f CM |
3687 | extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree, |
3688 | inode->i_mapping, GFP_NOFS); | |
ea8c2819 | 3689 | INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes); |
5a3f23d5 | 3690 | INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations); |
5d4f98a2 | 3691 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
ba1da2f4 | 3692 | btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree); |
e02119d5 CM |
3693 | mutex_init(&BTRFS_I(inode)->log_mutex); |
3694 | } | |
3695 | ||
3696 | static int btrfs_init_locked_inode(struct inode *inode, void *p) | |
3697 | { | |
3698 | struct btrfs_iget_args *args = p; | |
3699 | inode->i_ino = args->ino; | |
3700 | init_btrfs_i(inode); | |
3701 | BTRFS_I(inode)->root = args->root; | |
6a63209f | 3702 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
3703 | return 0; |
3704 | } | |
3705 | ||
3706 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
3707 | { | |
3708 | struct btrfs_iget_args *args = opaque; | |
d397712b CM |
3709 | return args->ino == inode->i_ino && |
3710 | args->root == BTRFS_I(inode)->root; | |
39279cc3 CM |
3711 | } |
3712 | ||
5d4f98a2 YZ |
3713 | static struct inode *btrfs_iget_locked(struct super_block *s, |
3714 | u64 objectid, | |
3715 | struct btrfs_root *root) | |
39279cc3 CM |
3716 | { |
3717 | struct inode *inode; | |
3718 | struct btrfs_iget_args args; | |
3719 | args.ino = objectid; | |
3720 | args.root = root; | |
3721 | ||
3722 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
3723 | btrfs_init_locked_inode, | |
3724 | (void *)&args); | |
3725 | return inode; | |
3726 | } | |
3727 | ||
1a54ef8c BR |
3728 | /* Get an inode object given its location and corresponding root. |
3729 | * Returns in *is_new if the inode was read from disk | |
3730 | */ | |
3731 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
5d4f98a2 | 3732 | struct btrfs_root *root) |
1a54ef8c BR |
3733 | { |
3734 | struct inode *inode; | |
3735 | ||
3736 | inode = btrfs_iget_locked(s, location->objectid, root); | |
3737 | if (!inode) | |
5d4f98a2 | 3738 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
3739 | |
3740 | if (inode->i_state & I_NEW) { | |
3741 | BTRFS_I(inode)->root = root; | |
3742 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
3743 | btrfs_read_locked_inode(inode); | |
5d4f98a2 YZ |
3744 | |
3745 | inode_tree_add(inode); | |
1a54ef8c | 3746 | unlock_new_inode(inode); |
1a54ef8c BR |
3747 | } |
3748 | ||
3749 | return inode; | |
3750 | } | |
3751 | ||
4df27c4d YZ |
3752 | static struct inode *new_simple_dir(struct super_block *s, |
3753 | struct btrfs_key *key, | |
3754 | struct btrfs_root *root) | |
3755 | { | |
3756 | struct inode *inode = new_inode(s); | |
3757 | ||
3758 | if (!inode) | |
3759 | return ERR_PTR(-ENOMEM); | |
3760 | ||
3761 | init_btrfs_i(inode); | |
3762 | ||
3763 | BTRFS_I(inode)->root = root; | |
3764 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
3765 | BTRFS_I(inode)->dummy_inode = 1; | |
3766 | ||
3767 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
3768 | inode->i_op = &simple_dir_inode_operations; | |
3769 | inode->i_fop = &simple_dir_operations; | |
3770 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
3771 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
3772 | ||
3773 | return inode; | |
3774 | } | |
3775 | ||
3de4586c | 3776 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 3777 | { |
d397712b | 3778 | struct inode *inode; |
4df27c4d | 3779 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
3780 | struct btrfs_root *sub_root = root; |
3781 | struct btrfs_key location; | |
76dda93c | 3782 | int index; |
5d4f98a2 | 3783 | int ret; |
39279cc3 | 3784 | |
76dda93c YZ |
3785 | dentry->d_op = &btrfs_dentry_operations; |
3786 | ||
39279cc3 CM |
3787 | if (dentry->d_name.len > BTRFS_NAME_LEN) |
3788 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 3789 | |
39279cc3 | 3790 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 3791 | |
39279cc3 CM |
3792 | if (ret < 0) |
3793 | return ERR_PTR(ret); | |
5f39d397 | 3794 | |
4df27c4d YZ |
3795 | if (location.objectid == 0) |
3796 | return NULL; | |
3797 | ||
3798 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
3799 | inode = btrfs_iget(dir->i_sb, &location, root); | |
3800 | return inode; | |
3801 | } | |
3802 | ||
3803 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
3804 | ||
76dda93c | 3805 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
3806 | ret = fixup_tree_root_location(root, dir, dentry, |
3807 | &location, &sub_root); | |
3808 | if (ret < 0) { | |
3809 | if (ret != -ENOENT) | |
3810 | inode = ERR_PTR(ret); | |
3811 | else | |
3812 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
3813 | } else { | |
5d4f98a2 | 3814 | inode = btrfs_iget(dir->i_sb, &location, sub_root); |
39279cc3 | 3815 | } |
76dda93c YZ |
3816 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
3817 | ||
c71bf099 YZ |
3818 | if (root != sub_root) { |
3819 | down_read(&root->fs_info->cleanup_work_sem); | |
3820 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
3821 | btrfs_orphan_cleanup(sub_root); | |
3822 | up_read(&root->fs_info->cleanup_work_sem); | |
3823 | } | |
3824 | ||
3de4586c CM |
3825 | return inode; |
3826 | } | |
3827 | ||
76dda93c YZ |
3828 | static int btrfs_dentry_delete(struct dentry *dentry) |
3829 | { | |
3830 | struct btrfs_root *root; | |
3831 | ||
efefb143 YZ |
3832 | if (!dentry->d_inode && !IS_ROOT(dentry)) |
3833 | dentry = dentry->d_parent; | |
76dda93c | 3834 | |
efefb143 YZ |
3835 | if (dentry->d_inode) { |
3836 | root = BTRFS_I(dentry->d_inode)->root; | |
3837 | if (btrfs_root_refs(&root->root_item) == 0) | |
3838 | return 1; | |
3839 | } | |
76dda93c YZ |
3840 | return 0; |
3841 | } | |
3842 | ||
3de4586c CM |
3843 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
3844 | struct nameidata *nd) | |
3845 | { | |
3846 | struct inode *inode; | |
3847 | ||
3de4586c CM |
3848 | inode = btrfs_lookup_dentry(dir, dentry); |
3849 | if (IS_ERR(inode)) | |
3850 | return ERR_CAST(inode); | |
7b128766 | 3851 | |
39279cc3 CM |
3852 | return d_splice_alias(inode, dentry); |
3853 | } | |
3854 | ||
39279cc3 CM |
3855 | static unsigned char btrfs_filetype_table[] = { |
3856 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
3857 | }; | |
3858 | ||
cbdf5a24 DW |
3859 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
3860 | filldir_t filldir) | |
39279cc3 | 3861 | { |
6da6abae | 3862 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
3863 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3864 | struct btrfs_item *item; | |
3865 | struct btrfs_dir_item *di; | |
3866 | struct btrfs_key key; | |
5f39d397 | 3867 | struct btrfs_key found_key; |
39279cc3 CM |
3868 | struct btrfs_path *path; |
3869 | int ret; | |
3870 | u32 nritems; | |
5f39d397 | 3871 | struct extent_buffer *leaf; |
39279cc3 CM |
3872 | int slot; |
3873 | int advance; | |
3874 | unsigned char d_type; | |
3875 | int over = 0; | |
3876 | u32 di_cur; | |
3877 | u32 di_total; | |
3878 | u32 di_len; | |
3879 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
3880 | char tmp_name[32]; |
3881 | char *name_ptr; | |
3882 | int name_len; | |
39279cc3 CM |
3883 | |
3884 | /* FIXME, use a real flag for deciding about the key type */ | |
3885 | if (root->fs_info->tree_root == root) | |
3886 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 3887 | |
3954401f CM |
3888 | /* special case for "." */ |
3889 | if (filp->f_pos == 0) { | |
3890 | over = filldir(dirent, ".", 1, | |
3891 | 1, inode->i_ino, | |
3892 | DT_DIR); | |
3893 | if (over) | |
3894 | return 0; | |
3895 | filp->f_pos = 1; | |
3896 | } | |
3954401f CM |
3897 | /* special case for .., just use the back ref */ |
3898 | if (filp->f_pos == 1) { | |
5ecc7e5d | 3899 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 3900 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 3901 | 2, pino, DT_DIR); |
3954401f | 3902 | if (over) |
49593bfa | 3903 | return 0; |
3954401f CM |
3904 | filp->f_pos = 2; |
3905 | } | |
49593bfa DW |
3906 | path = btrfs_alloc_path(); |
3907 | path->reada = 2; | |
3908 | ||
39279cc3 CM |
3909 | btrfs_set_key_type(&key, key_type); |
3910 | key.offset = filp->f_pos; | |
49593bfa | 3911 | key.objectid = inode->i_ino; |
5f39d397 | 3912 | |
39279cc3 CM |
3913 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3914 | if (ret < 0) | |
3915 | goto err; | |
3916 | advance = 0; | |
49593bfa DW |
3917 | |
3918 | while (1) { | |
5f39d397 CM |
3919 | leaf = path->nodes[0]; |
3920 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3921 | slot = path->slots[0]; |
3922 | if (advance || slot >= nritems) { | |
49593bfa | 3923 | if (slot >= nritems - 1) { |
39279cc3 CM |
3924 | ret = btrfs_next_leaf(root, path); |
3925 | if (ret) | |
3926 | break; | |
5f39d397 CM |
3927 | leaf = path->nodes[0]; |
3928 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3929 | slot = path->slots[0]; |
3930 | } else { | |
3931 | slot++; | |
3932 | path->slots[0]++; | |
3933 | } | |
3934 | } | |
3de4586c | 3935 | |
39279cc3 | 3936 | advance = 1; |
5f39d397 CM |
3937 | item = btrfs_item_nr(leaf, slot); |
3938 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3939 | ||
3940 | if (found_key.objectid != key.objectid) | |
39279cc3 | 3941 | break; |
5f39d397 | 3942 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 3943 | break; |
5f39d397 | 3944 | if (found_key.offset < filp->f_pos) |
39279cc3 | 3945 | continue; |
5f39d397 CM |
3946 | |
3947 | filp->f_pos = found_key.offset; | |
49593bfa | 3948 | |
39279cc3 CM |
3949 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
3950 | di_cur = 0; | |
5f39d397 | 3951 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
3952 | |
3953 | while (di_cur < di_total) { | |
5f39d397 CM |
3954 | struct btrfs_key location; |
3955 | ||
3956 | name_len = btrfs_dir_name_len(leaf, di); | |
49593bfa | 3957 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
3958 | name_ptr = tmp_name; |
3959 | } else { | |
3960 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
3961 | if (!name_ptr) { |
3962 | ret = -ENOMEM; | |
3963 | goto err; | |
3964 | } | |
5f39d397 CM |
3965 | } |
3966 | read_extent_buffer(leaf, name_ptr, | |
3967 | (unsigned long)(di + 1), name_len); | |
3968 | ||
3969 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
3970 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
3971 | |
3972 | /* is this a reference to our own snapshot? If so | |
3973 | * skip it | |
3974 | */ | |
3975 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
3976 | location.objectid == root->root_key.objectid) { | |
3977 | over = 0; | |
3978 | goto skip; | |
3979 | } | |
5f39d397 | 3980 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 3981 | found_key.offset, location.objectid, |
39279cc3 | 3982 | d_type); |
5f39d397 | 3983 | |
3de4586c | 3984 | skip: |
5f39d397 CM |
3985 | if (name_ptr != tmp_name) |
3986 | kfree(name_ptr); | |
3987 | ||
39279cc3 CM |
3988 | if (over) |
3989 | goto nopos; | |
5103e947 | 3990 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 3991 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
3992 | di_cur += di_len; |
3993 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
3994 | } | |
3995 | } | |
49593bfa DW |
3996 | |
3997 | /* Reached end of directory/root. Bump pos past the last item. */ | |
5e591a07 | 3998 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
3999 | /* |
4000 | * 32-bit glibc will use getdents64, but then strtol - | |
4001 | * so the last number we can serve is this. | |
4002 | */ | |
4003 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4004 | else |
4005 | filp->f_pos++; | |
39279cc3 CM |
4006 | nopos: |
4007 | ret = 0; | |
4008 | err: | |
39279cc3 | 4009 | btrfs_free_path(path); |
39279cc3 CM |
4010 | return ret; |
4011 | } | |
4012 | ||
4013 | int btrfs_write_inode(struct inode *inode, int wait) | |
4014 | { | |
4015 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4016 | struct btrfs_trans_handle *trans; | |
4017 | int ret = 0; | |
4018 | ||
c146afad | 4019 | if (root->fs_info->btree_inode == inode) |
4ca8b41e CM |
4020 | return 0; |
4021 | ||
39279cc3 | 4022 | if (wait) { |
f9295749 | 4023 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
4024 | btrfs_set_trans_block_group(trans, inode); |
4025 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
4026 | } |
4027 | return ret; | |
4028 | } | |
4029 | ||
4030 | /* | |
54aa1f4d | 4031 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4032 | * inode changes. But, it is most likely to find the inode in cache. |
4033 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4034 | * to keep or drop this code. | |
4035 | */ | |
4036 | void btrfs_dirty_inode(struct inode *inode) | |
4037 | { | |
4038 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4039 | struct btrfs_trans_handle *trans; | |
4040 | ||
f9295749 | 4041 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
4042 | btrfs_set_trans_block_group(trans, inode); |
4043 | btrfs_update_inode(trans, root, inode); | |
4044 | btrfs_end_transaction(trans, root); | |
39279cc3 CM |
4045 | } |
4046 | ||
d352ac68 CM |
4047 | /* |
4048 | * find the highest existing sequence number in a directory | |
4049 | * and then set the in-memory index_cnt variable to reflect | |
4050 | * free sequence numbers | |
4051 | */ | |
aec7477b JB |
4052 | static int btrfs_set_inode_index_count(struct inode *inode) |
4053 | { | |
4054 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4055 | struct btrfs_key key, found_key; | |
4056 | struct btrfs_path *path; | |
4057 | struct extent_buffer *leaf; | |
4058 | int ret; | |
4059 | ||
4060 | key.objectid = inode->i_ino; | |
4061 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); | |
4062 | key.offset = (u64)-1; | |
4063 | ||
4064 | path = btrfs_alloc_path(); | |
4065 | if (!path) | |
4066 | return -ENOMEM; | |
4067 | ||
4068 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4069 | if (ret < 0) | |
4070 | goto out; | |
4071 | /* FIXME: we should be able to handle this */ | |
4072 | if (ret == 0) | |
4073 | goto out; | |
4074 | ret = 0; | |
4075 | ||
4076 | /* | |
4077 | * MAGIC NUMBER EXPLANATION: | |
4078 | * since we search a directory based on f_pos we have to start at 2 | |
4079 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4080 | * else has to start at 2 | |
4081 | */ | |
4082 | if (path->slots[0] == 0) { | |
4083 | BTRFS_I(inode)->index_cnt = 2; | |
4084 | goto out; | |
4085 | } | |
4086 | ||
4087 | path->slots[0]--; | |
4088 | ||
4089 | leaf = path->nodes[0]; | |
4090 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4091 | ||
4092 | if (found_key.objectid != inode->i_ino || | |
4093 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { | |
4094 | BTRFS_I(inode)->index_cnt = 2; | |
4095 | goto out; | |
4096 | } | |
4097 | ||
4098 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4099 | out: | |
4100 | btrfs_free_path(path); | |
4101 | return ret; | |
4102 | } | |
4103 | ||
d352ac68 CM |
4104 | /* |
4105 | * helper to find a free sequence number in a given directory. This current | |
4106 | * code is very simple, later versions will do smarter things in the btree | |
4107 | */ | |
3de4586c | 4108 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4109 | { |
4110 | int ret = 0; | |
4111 | ||
4112 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
4113 | ret = btrfs_set_inode_index_count(dir); | |
d397712b | 4114 | if (ret) |
aec7477b JB |
4115 | return ret; |
4116 | } | |
4117 | ||
00e4e6b3 | 4118 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4119 | BTRFS_I(dir)->index_cnt++; |
4120 | ||
4121 | return ret; | |
4122 | } | |
4123 | ||
39279cc3 CM |
4124 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4125 | struct btrfs_root *root, | |
aec7477b | 4126 | struct inode *dir, |
9c58309d | 4127 | const char *name, int name_len, |
d2fb3437 YZ |
4128 | u64 ref_objectid, u64 objectid, |
4129 | u64 alloc_hint, int mode, u64 *index) | |
39279cc3 CM |
4130 | { |
4131 | struct inode *inode; | |
5f39d397 | 4132 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4133 | struct btrfs_key *location; |
5f39d397 | 4134 | struct btrfs_path *path; |
9c58309d CM |
4135 | struct btrfs_inode_ref *ref; |
4136 | struct btrfs_key key[2]; | |
4137 | u32 sizes[2]; | |
4138 | unsigned long ptr; | |
39279cc3 CM |
4139 | int ret; |
4140 | int owner; | |
4141 | ||
5f39d397 CM |
4142 | path = btrfs_alloc_path(); |
4143 | BUG_ON(!path); | |
4144 | ||
39279cc3 CM |
4145 | inode = new_inode(root->fs_info->sb); |
4146 | if (!inode) | |
4147 | return ERR_PTR(-ENOMEM); | |
4148 | ||
aec7477b | 4149 | if (dir) { |
3de4586c | 4150 | ret = btrfs_set_inode_index(dir, index); |
09771430 SF |
4151 | if (ret) { |
4152 | iput(inode); | |
aec7477b | 4153 | return ERR_PTR(ret); |
09771430 | 4154 | } |
aec7477b JB |
4155 | } |
4156 | /* | |
4157 | * index_cnt is ignored for everything but a dir, | |
4158 | * btrfs_get_inode_index_count has an explanation for the magic | |
4159 | * number | |
4160 | */ | |
e02119d5 | 4161 | init_btrfs_i(inode); |
aec7477b | 4162 | BTRFS_I(inode)->index_cnt = 2; |
39279cc3 | 4163 | BTRFS_I(inode)->root = root; |
e02119d5 | 4164 | BTRFS_I(inode)->generation = trans->transid; |
6a63209f | 4165 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 4166 | |
39279cc3 CM |
4167 | if (mode & S_IFDIR) |
4168 | owner = 0; | |
4169 | else | |
4170 | owner = 1; | |
d2fb3437 YZ |
4171 | BTRFS_I(inode)->block_group = |
4172 | btrfs_find_block_group(root, 0, alloc_hint, owner); | |
9c58309d CM |
4173 | |
4174 | key[0].objectid = objectid; | |
4175 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4176 | key[0].offset = 0; | |
4177 | ||
4178 | key[1].objectid = objectid; | |
4179 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4180 | key[1].offset = ref_objectid; | |
4181 | ||
4182 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4183 | sizes[1] = name_len + sizeof(*ref); | |
4184 | ||
b9473439 | 4185 | path->leave_spinning = 1; |
9c58309d CM |
4186 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4187 | if (ret != 0) | |
5f39d397 CM |
4188 | goto fail; |
4189 | ||
79683f2d | 4190 | inode->i_uid = current_fsuid(); |
8c087b51 | 4191 | |
42f15d77 | 4192 | if (dir && (dir->i_mode & S_ISGID)) { |
8c087b51 CB |
4193 | inode->i_gid = dir->i_gid; |
4194 | if (S_ISDIR(mode)) | |
4195 | mode |= S_ISGID; | |
4196 | } else | |
4197 | inode->i_gid = current_fsgid(); | |
4198 | ||
39279cc3 CM |
4199 | inode->i_mode = mode; |
4200 | inode->i_ino = objectid; | |
a76a3cd4 | 4201 | inode_set_bytes(inode, 0); |
39279cc3 | 4202 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4203 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4204 | struct btrfs_inode_item); | |
e02119d5 | 4205 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4206 | |
4207 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4208 | struct btrfs_inode_ref); | |
4209 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4210 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4211 | ptr = (unsigned long)(ref + 1); |
4212 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4213 | ||
5f39d397 CM |
4214 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4215 | btrfs_free_path(path); | |
4216 | ||
39279cc3 CM |
4217 | location = &BTRFS_I(inode)->location; |
4218 | location->objectid = objectid; | |
39279cc3 CM |
4219 | location->offset = 0; |
4220 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4221 | ||
6cbff00f CH |
4222 | btrfs_inherit_iflags(inode, dir); |
4223 | ||
94272164 CM |
4224 | if ((mode & S_IFREG)) { |
4225 | if (btrfs_test_opt(root, NODATASUM)) | |
4226 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
4227 | if (btrfs_test_opt(root, NODATACOW)) | |
4228 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
4229 | } | |
4230 | ||
39279cc3 | 4231 | insert_inode_hash(inode); |
5d4f98a2 | 4232 | inode_tree_add(inode); |
39279cc3 | 4233 | return inode; |
5f39d397 | 4234 | fail: |
aec7477b JB |
4235 | if (dir) |
4236 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4237 | btrfs_free_path(path); |
09771430 | 4238 | iput(inode); |
5f39d397 | 4239 | return ERR_PTR(ret); |
39279cc3 CM |
4240 | } |
4241 | ||
4242 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4243 | { | |
4244 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4245 | } | |
4246 | ||
d352ac68 CM |
4247 | /* |
4248 | * utility function to add 'inode' into 'parent_inode' with | |
4249 | * a give name and a given sequence number. | |
4250 | * if 'add_backref' is true, also insert a backref from the | |
4251 | * inode to the parent directory. | |
4252 | */ | |
e02119d5 CM |
4253 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4254 | struct inode *parent_inode, struct inode *inode, | |
4255 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4256 | { |
4df27c4d | 4257 | int ret = 0; |
39279cc3 | 4258 | struct btrfs_key key; |
e02119d5 | 4259 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
5f39d397 | 4260 | |
4df27c4d YZ |
4261 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4262 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); | |
4263 | } else { | |
4264 | key.objectid = inode->i_ino; | |
4265 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
4266 | key.offset = 0; | |
4267 | } | |
4268 | ||
4269 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4270 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, | |
4271 | key.objectid, root->root_key.objectid, | |
4272 | parent_inode->i_ino, | |
4273 | index, name, name_len); | |
4274 | } else if (add_backref) { | |
4275 | ret = btrfs_insert_inode_ref(trans, root, | |
4276 | name, name_len, inode->i_ino, | |
4277 | parent_inode->i_ino, index); | |
4278 | } | |
39279cc3 | 4279 | |
39279cc3 | 4280 | if (ret == 0) { |
4df27c4d YZ |
4281 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4282 | parent_inode->i_ino, &key, | |
4283 | btrfs_inode_type(inode), index); | |
4284 | BUG_ON(ret); | |
4285 | ||
dbe674a9 | 4286 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 4287 | name_len * 2); |
79c44584 | 4288 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 4289 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
4290 | } |
4291 | return ret; | |
4292 | } | |
4293 | ||
4294 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
9c58309d | 4295 | struct dentry *dentry, struct inode *inode, |
00e4e6b3 | 4296 | int backref, u64 index) |
39279cc3 | 4297 | { |
e02119d5 CM |
4298 | int err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4299 | inode, dentry->d_name.name, | |
4300 | dentry->d_name.len, backref, index); | |
39279cc3 CM |
4301 | if (!err) { |
4302 | d_instantiate(dentry, inode); | |
4303 | return 0; | |
4304 | } | |
4305 | if (err > 0) | |
4306 | err = -EEXIST; | |
4307 | return err; | |
4308 | } | |
4309 | ||
618e21d5 JB |
4310 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
4311 | int mode, dev_t rdev) | |
4312 | { | |
4313 | struct btrfs_trans_handle *trans; | |
4314 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4315 | struct inode *inode = NULL; |
618e21d5 JB |
4316 | int err; |
4317 | int drop_inode = 0; | |
4318 | u64 objectid; | |
1832a6d5 | 4319 | unsigned long nr = 0; |
00e4e6b3 | 4320 | u64 index = 0; |
618e21d5 JB |
4321 | |
4322 | if (!new_valid_dev(rdev)) | |
4323 | return -EINVAL; | |
4324 | ||
9ed74f2d JB |
4325 | /* |
4326 | * 2 for inode item and ref | |
4327 | * 2 for dir items | |
4328 | * 1 for xattr if selinux is on | |
4329 | */ | |
4330 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4331 | if (err) |
9ed74f2d | 4332 | return err; |
1832a6d5 | 4333 | |
618e21d5 | 4334 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4335 | if (!trans) |
4336 | goto fail; | |
618e21d5 JB |
4337 | btrfs_set_trans_block_group(trans, dir); |
4338 | ||
4339 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4340 | if (err) { | |
4341 | err = -ENOSPC; | |
4342 | goto out_unlock; | |
4343 | } | |
4344 | ||
aec7477b | 4345 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4346 | dentry->d_name.len, |
4347 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 | 4348 | BTRFS_I(dir)->block_group, mode, &index); |
618e21d5 JB |
4349 | err = PTR_ERR(inode); |
4350 | if (IS_ERR(inode)) | |
4351 | goto out_unlock; | |
4352 | ||
f34f57a3 | 4353 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4354 | if (err) { |
4355 | drop_inode = 1; | |
4356 | goto out_unlock; | |
4357 | } | |
4358 | ||
618e21d5 | 4359 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4360 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
618e21d5 JB |
4361 | if (err) |
4362 | drop_inode = 1; | |
4363 | else { | |
4364 | inode->i_op = &btrfs_special_inode_operations; | |
4365 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 4366 | btrfs_update_inode(trans, root, inode); |
618e21d5 | 4367 | } |
618e21d5 JB |
4368 | btrfs_update_inode_block_group(trans, inode); |
4369 | btrfs_update_inode_block_group(trans, dir); | |
4370 | out_unlock: | |
d3c2fdcf | 4371 | nr = trans->blocks_used; |
89ce8a63 | 4372 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4373 | fail: |
9ed74f2d | 4374 | btrfs_unreserve_metadata_space(root, 5); |
618e21d5 JB |
4375 | if (drop_inode) { |
4376 | inode_dec_link_count(inode); | |
4377 | iput(inode); | |
4378 | } | |
d3c2fdcf | 4379 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4380 | return err; |
4381 | } | |
4382 | ||
39279cc3 CM |
4383 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4384 | int mode, struct nameidata *nd) | |
4385 | { | |
4386 | struct btrfs_trans_handle *trans; | |
4387 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4388 | struct inode *inode = NULL; |
39279cc3 CM |
4389 | int err; |
4390 | int drop_inode = 0; | |
1832a6d5 | 4391 | unsigned long nr = 0; |
39279cc3 | 4392 | u64 objectid; |
00e4e6b3 | 4393 | u64 index = 0; |
39279cc3 | 4394 | |
9ed74f2d JB |
4395 | /* |
4396 | * 2 for inode item and ref | |
4397 | * 2 for dir items | |
4398 | * 1 for xattr if selinux is on | |
4399 | */ | |
4400 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4401 | if (err) |
9ed74f2d JB |
4402 | return err; |
4403 | ||
39279cc3 | 4404 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4405 | if (!trans) |
4406 | goto fail; | |
39279cc3 CM |
4407 | btrfs_set_trans_block_group(trans, dir); |
4408 | ||
4409 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4410 | if (err) { | |
4411 | err = -ENOSPC; | |
4412 | goto out_unlock; | |
4413 | } | |
4414 | ||
aec7477b | 4415 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4416 | dentry->d_name.len, |
4417 | dentry->d_parent->d_inode->i_ino, | |
00e4e6b3 CM |
4418 | objectid, BTRFS_I(dir)->block_group, mode, |
4419 | &index); | |
39279cc3 CM |
4420 | err = PTR_ERR(inode); |
4421 | if (IS_ERR(inode)) | |
4422 | goto out_unlock; | |
4423 | ||
f34f57a3 | 4424 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4425 | if (err) { |
4426 | drop_inode = 1; | |
4427 | goto out_unlock; | |
4428 | } | |
4429 | ||
39279cc3 | 4430 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4431 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
4432 | if (err) |
4433 | drop_inode = 1; | |
4434 | else { | |
4435 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4436 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4437 | inode->i_fop = &btrfs_file_operations; |
4438 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4439 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 4440 | } |
39279cc3 CM |
4441 | btrfs_update_inode_block_group(trans, inode); |
4442 | btrfs_update_inode_block_group(trans, dir); | |
4443 | out_unlock: | |
d3c2fdcf | 4444 | nr = trans->blocks_used; |
ab78c84d | 4445 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4446 | fail: |
9ed74f2d | 4447 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
4448 | if (drop_inode) { |
4449 | inode_dec_link_count(inode); | |
4450 | iput(inode); | |
4451 | } | |
d3c2fdcf | 4452 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4453 | return err; |
4454 | } | |
4455 | ||
4456 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4457 | struct dentry *dentry) | |
4458 | { | |
4459 | struct btrfs_trans_handle *trans; | |
4460 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4461 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4462 | u64 index; |
1832a6d5 | 4463 | unsigned long nr = 0; |
39279cc3 CM |
4464 | int err; |
4465 | int drop_inode = 0; | |
4466 | ||
4467 | if (inode->i_nlink == 0) | |
4468 | return -ENOENT; | |
4469 | ||
4a8be425 TH |
4470 | /* do not allow sys_link's with other subvols of the same device */ |
4471 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
4472 | return -EPERM; | |
4473 | ||
9ed74f2d JB |
4474 | /* |
4475 | * 1 item for inode ref | |
4476 | * 2 items for dir items | |
4477 | */ | |
4478 | err = btrfs_reserve_metadata_space(root, 3); | |
1832a6d5 | 4479 | if (err) |
9ed74f2d JB |
4480 | return err; |
4481 | ||
4482 | btrfs_inc_nlink(inode); | |
4483 | ||
3de4586c | 4484 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4485 | if (err) |
4486 | goto fail; | |
4487 | ||
39279cc3 | 4488 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 4489 | |
39279cc3 CM |
4490 | btrfs_set_trans_block_group(trans, dir); |
4491 | atomic_inc(&inode->i_count); | |
aec7477b | 4492 | |
00e4e6b3 | 4493 | err = btrfs_add_nondir(trans, dentry, inode, 1, index); |
5f39d397 | 4494 | |
a5719521 | 4495 | if (err) { |
54aa1f4d | 4496 | drop_inode = 1; |
a5719521 YZ |
4497 | } else { |
4498 | btrfs_update_inode_block_group(trans, dir); | |
4499 | err = btrfs_update_inode(trans, root, inode); | |
4500 | BUG_ON(err); | |
4501 | btrfs_log_new_name(trans, inode, NULL, dentry->d_parent); | |
4502 | } | |
39279cc3 | 4503 | |
d3c2fdcf | 4504 | nr = trans->blocks_used; |
ab78c84d | 4505 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4506 | fail: |
9ed74f2d | 4507 | btrfs_unreserve_metadata_space(root, 3); |
39279cc3 CM |
4508 | if (drop_inode) { |
4509 | inode_dec_link_count(inode); | |
4510 | iput(inode); | |
4511 | } | |
d3c2fdcf | 4512 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4513 | return err; |
4514 | } | |
4515 | ||
39279cc3 CM |
4516 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
4517 | { | |
b9d86667 | 4518 | struct inode *inode = NULL; |
39279cc3 CM |
4519 | struct btrfs_trans_handle *trans; |
4520 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4521 | int err = 0; | |
4522 | int drop_on_err = 0; | |
b9d86667 | 4523 | u64 objectid = 0; |
00e4e6b3 | 4524 | u64 index = 0; |
d3c2fdcf | 4525 | unsigned long nr = 1; |
39279cc3 | 4526 | |
9ed74f2d JB |
4527 | /* |
4528 | * 2 items for inode and ref | |
4529 | * 2 items for dir items | |
4530 | * 1 for xattr if selinux is on | |
4531 | */ | |
4532 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4533 | if (err) |
9ed74f2d | 4534 | return err; |
1832a6d5 | 4535 | |
39279cc3 | 4536 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4537 | if (!trans) { |
4538 | err = -ENOMEM; | |
39279cc3 CM |
4539 | goto out_unlock; |
4540 | } | |
9ed74f2d | 4541 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 CM |
4542 | |
4543 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4544 | if (err) { | |
4545 | err = -ENOSPC; | |
4546 | goto out_unlock; | |
4547 | } | |
4548 | ||
aec7477b | 4549 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4550 | dentry->d_name.len, |
4551 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
4552 | BTRFS_I(dir)->block_group, S_IFDIR | mode, |
4553 | &index); | |
39279cc3 CM |
4554 | if (IS_ERR(inode)) { |
4555 | err = PTR_ERR(inode); | |
4556 | goto out_fail; | |
4557 | } | |
5f39d397 | 4558 | |
39279cc3 | 4559 | drop_on_err = 1; |
33268eaf | 4560 | |
f34f57a3 | 4561 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4562 | if (err) |
4563 | goto out_fail; | |
4564 | ||
39279cc3 CM |
4565 | inode->i_op = &btrfs_dir_inode_operations; |
4566 | inode->i_fop = &btrfs_dir_file_operations; | |
4567 | btrfs_set_trans_block_group(trans, inode); | |
4568 | ||
dbe674a9 | 4569 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
4570 | err = btrfs_update_inode(trans, root, inode); |
4571 | if (err) | |
4572 | goto out_fail; | |
5f39d397 | 4573 | |
e02119d5 CM |
4574 | err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4575 | inode, dentry->d_name.name, | |
4576 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
4577 | if (err) |
4578 | goto out_fail; | |
5f39d397 | 4579 | |
39279cc3 CM |
4580 | d_instantiate(dentry, inode); |
4581 | drop_on_err = 0; | |
39279cc3 CM |
4582 | btrfs_update_inode_block_group(trans, inode); |
4583 | btrfs_update_inode_block_group(trans, dir); | |
4584 | ||
4585 | out_fail: | |
d3c2fdcf | 4586 | nr = trans->blocks_used; |
ab78c84d | 4587 | btrfs_end_transaction_throttle(trans, root); |
5f39d397 | 4588 | |
39279cc3 | 4589 | out_unlock: |
9ed74f2d | 4590 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
4591 | if (drop_on_err) |
4592 | iput(inode); | |
d3c2fdcf | 4593 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4594 | return err; |
4595 | } | |
4596 | ||
d352ac68 CM |
4597 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
4598 | * and an extent that you want to insert, deal with overlap and insert | |
4599 | * the new extent into the tree. | |
4600 | */ | |
3b951516 CM |
4601 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
4602 | struct extent_map *existing, | |
e6dcd2dc CM |
4603 | struct extent_map *em, |
4604 | u64 map_start, u64 map_len) | |
3b951516 CM |
4605 | { |
4606 | u64 start_diff; | |
3b951516 | 4607 | |
e6dcd2dc CM |
4608 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
4609 | start_diff = map_start - em->start; | |
4610 | em->start = map_start; | |
4611 | em->len = map_len; | |
c8b97818 CM |
4612 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
4613 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 4614 | em->block_start += start_diff; |
c8b97818 CM |
4615 | em->block_len -= start_diff; |
4616 | } | |
e6dcd2dc | 4617 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
4618 | } |
4619 | ||
c8b97818 CM |
4620 | static noinline int uncompress_inline(struct btrfs_path *path, |
4621 | struct inode *inode, struct page *page, | |
4622 | size_t pg_offset, u64 extent_offset, | |
4623 | struct btrfs_file_extent_item *item) | |
4624 | { | |
4625 | int ret; | |
4626 | struct extent_buffer *leaf = path->nodes[0]; | |
4627 | char *tmp; | |
4628 | size_t max_size; | |
4629 | unsigned long inline_size; | |
4630 | unsigned long ptr; | |
4631 | ||
4632 | WARN_ON(pg_offset != 0); | |
4633 | max_size = btrfs_file_extent_ram_bytes(leaf, item); | |
4634 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
4635 | btrfs_item_nr(leaf, path->slots[0])); | |
4636 | tmp = kmalloc(inline_size, GFP_NOFS); | |
4637 | ptr = btrfs_file_extent_inline_start(item); | |
4638 | ||
4639 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
4640 | ||
5b050f04 | 4641 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
c8b97818 CM |
4642 | ret = btrfs_zlib_decompress(tmp, page, extent_offset, |
4643 | inline_size, max_size); | |
4644 | if (ret) { | |
4645 | char *kaddr = kmap_atomic(page, KM_USER0); | |
4646 | unsigned long copy_size = min_t(u64, | |
4647 | PAGE_CACHE_SIZE - pg_offset, | |
4648 | max_size - extent_offset); | |
4649 | memset(kaddr + pg_offset, 0, copy_size); | |
4650 | kunmap_atomic(kaddr, KM_USER0); | |
4651 | } | |
4652 | kfree(tmp); | |
4653 | return 0; | |
4654 | } | |
4655 | ||
d352ac68 CM |
4656 | /* |
4657 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
4658 | * the ugly parts come from merging extents from the disk with the in-ram |
4659 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
4660 | * where the in-ram extents might be locked pending data=ordered completion. |
4661 | * | |
4662 | * This also copies inline extents directly into the page. | |
4663 | */ | |
d397712b | 4664 | |
a52d9a80 | 4665 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 4666 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
4667 | int create) |
4668 | { | |
4669 | int ret; | |
4670 | int err = 0; | |
db94535d | 4671 | u64 bytenr; |
a52d9a80 CM |
4672 | u64 extent_start = 0; |
4673 | u64 extent_end = 0; | |
4674 | u64 objectid = inode->i_ino; | |
4675 | u32 found_type; | |
f421950f | 4676 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
4677 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4678 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
4679 | struct extent_buffer *leaf; |
4680 | struct btrfs_key found_key; | |
a52d9a80 CM |
4681 | struct extent_map *em = NULL; |
4682 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 4683 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 4684 | struct btrfs_trans_handle *trans = NULL; |
c8b97818 | 4685 | int compressed; |
a52d9a80 | 4686 | |
a52d9a80 | 4687 | again: |
890871be | 4688 | read_lock(&em_tree->lock); |
d1310b2e | 4689 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
4690 | if (em) |
4691 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 4692 | read_unlock(&em_tree->lock); |
d1310b2e | 4693 | |
a52d9a80 | 4694 | if (em) { |
e1c4b745 CM |
4695 | if (em->start > start || em->start + em->len <= start) |
4696 | free_extent_map(em); | |
4697 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
4698 | free_extent_map(em); |
4699 | else | |
4700 | goto out; | |
a52d9a80 | 4701 | } |
d1310b2e | 4702 | em = alloc_extent_map(GFP_NOFS); |
a52d9a80 | 4703 | if (!em) { |
d1310b2e CM |
4704 | err = -ENOMEM; |
4705 | goto out; | |
a52d9a80 | 4706 | } |
e6dcd2dc | 4707 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 4708 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 4709 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 4710 | em->len = (u64)-1; |
c8b97818 | 4711 | em->block_len = (u64)-1; |
f421950f CM |
4712 | |
4713 | if (!path) { | |
4714 | path = btrfs_alloc_path(); | |
4715 | BUG_ON(!path); | |
4716 | } | |
4717 | ||
179e29e4 CM |
4718 | ret = btrfs_lookup_file_extent(trans, root, path, |
4719 | objectid, start, trans != NULL); | |
a52d9a80 CM |
4720 | if (ret < 0) { |
4721 | err = ret; | |
4722 | goto out; | |
4723 | } | |
4724 | ||
4725 | if (ret != 0) { | |
4726 | if (path->slots[0] == 0) | |
4727 | goto not_found; | |
4728 | path->slots[0]--; | |
4729 | } | |
4730 | ||
5f39d397 CM |
4731 | leaf = path->nodes[0]; |
4732 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 4733 | struct btrfs_file_extent_item); |
a52d9a80 | 4734 | /* are we inside the extent that was found? */ |
5f39d397 CM |
4735 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4736 | found_type = btrfs_key_type(&found_key); | |
4737 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
4738 | found_type != BTRFS_EXTENT_DATA_KEY) { |
4739 | goto not_found; | |
4740 | } | |
4741 | ||
5f39d397 CM |
4742 | found_type = btrfs_file_extent_type(leaf, item); |
4743 | extent_start = found_key.offset; | |
c8b97818 | 4744 | compressed = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
4745 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4746 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 4747 | extent_end = extent_start + |
db94535d | 4748 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
4749 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
4750 | size_t size; | |
4751 | size = btrfs_file_extent_inline_len(leaf, item); | |
4752 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
4753 | ~((u64)root->sectorsize - 1); | |
4754 | } | |
4755 | ||
4756 | if (start >= extent_end) { | |
4757 | path->slots[0]++; | |
4758 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
4759 | ret = btrfs_next_leaf(root, path); | |
4760 | if (ret < 0) { | |
4761 | err = ret; | |
4762 | goto out; | |
a52d9a80 | 4763 | } |
9036c102 YZ |
4764 | if (ret > 0) |
4765 | goto not_found; | |
4766 | leaf = path->nodes[0]; | |
a52d9a80 | 4767 | } |
9036c102 YZ |
4768 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4769 | if (found_key.objectid != objectid || | |
4770 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
4771 | goto not_found; | |
4772 | if (start + len <= found_key.offset) | |
4773 | goto not_found; | |
4774 | em->start = start; | |
4775 | em->len = found_key.offset - start; | |
4776 | goto not_found_em; | |
4777 | } | |
4778 | ||
d899e052 YZ |
4779 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4780 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
4781 | em->start = extent_start; |
4782 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
4783 | em->orig_start = extent_start - |
4784 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
4785 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
4786 | if (bytenr == 0) { | |
5f39d397 | 4787 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
4788 | goto insert; |
4789 | } | |
c8b97818 CM |
4790 | if (compressed) { |
4791 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
4792 | em->block_start = bytenr; | |
4793 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
4794 | item); | |
4795 | } else { | |
4796 | bytenr += btrfs_file_extent_offset(leaf, item); | |
4797 | em->block_start = bytenr; | |
4798 | em->block_len = em->len; | |
d899e052 YZ |
4799 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
4800 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 4801 | } |
a52d9a80 CM |
4802 | goto insert; |
4803 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4804 | unsigned long ptr; |
a52d9a80 | 4805 | char *map; |
3326d1b0 CM |
4806 | size_t size; |
4807 | size_t extent_offset; | |
4808 | size_t copy_size; | |
a52d9a80 | 4809 | |
689f9346 | 4810 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 4811 | if (!page || create) { |
689f9346 | 4812 | em->start = extent_start; |
9036c102 | 4813 | em->len = extent_end - extent_start; |
689f9346 Y |
4814 | goto out; |
4815 | } | |
5f39d397 | 4816 | |
9036c102 YZ |
4817 | size = btrfs_file_extent_inline_len(leaf, item); |
4818 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 4819 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 4820 | size - extent_offset); |
3326d1b0 | 4821 | em->start = extent_start + extent_offset; |
70dec807 CM |
4822 | em->len = (copy_size + root->sectorsize - 1) & |
4823 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 4824 | em->orig_start = EXTENT_MAP_INLINE; |
c8b97818 CM |
4825 | if (compressed) |
4826 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
689f9346 | 4827 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 4828 | if (create == 0 && !PageUptodate(page)) { |
c8b97818 CM |
4829 | if (btrfs_file_extent_compression(leaf, item) == |
4830 | BTRFS_COMPRESS_ZLIB) { | |
4831 | ret = uncompress_inline(path, inode, page, | |
4832 | pg_offset, | |
4833 | extent_offset, item); | |
4834 | BUG_ON(ret); | |
4835 | } else { | |
4836 | map = kmap(page); | |
4837 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
4838 | copy_size); | |
93c82d57 CM |
4839 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
4840 | memset(map + pg_offset + copy_size, 0, | |
4841 | PAGE_CACHE_SIZE - pg_offset - | |
4842 | copy_size); | |
4843 | } | |
c8b97818 CM |
4844 | kunmap(page); |
4845 | } | |
179e29e4 CM |
4846 | flush_dcache_page(page); |
4847 | } else if (create && PageUptodate(page)) { | |
4848 | if (!trans) { | |
4849 | kunmap(page); | |
4850 | free_extent_map(em); | |
4851 | em = NULL; | |
4852 | btrfs_release_path(root, path); | |
f9295749 | 4853 | trans = btrfs_join_transaction(root, 1); |
179e29e4 CM |
4854 | goto again; |
4855 | } | |
c8b97818 | 4856 | map = kmap(page); |
70dec807 | 4857 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 4858 | copy_size); |
c8b97818 | 4859 | kunmap(page); |
179e29e4 | 4860 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 4861 | } |
d1310b2e CM |
4862 | set_extent_uptodate(io_tree, em->start, |
4863 | extent_map_end(em) - 1, GFP_NOFS); | |
a52d9a80 CM |
4864 | goto insert; |
4865 | } else { | |
d397712b | 4866 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
4867 | WARN_ON(1); |
4868 | } | |
4869 | not_found: | |
4870 | em->start = start; | |
d1310b2e | 4871 | em->len = len; |
a52d9a80 | 4872 | not_found_em: |
5f39d397 | 4873 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 4874 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 CM |
4875 | insert: |
4876 | btrfs_release_path(root, path); | |
d1310b2e | 4877 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
4878 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
4879 | "[%llu %llu]\n", (unsigned long long)em->start, | |
4880 | (unsigned long long)em->len, | |
4881 | (unsigned long long)start, | |
4882 | (unsigned long long)len); | |
a52d9a80 CM |
4883 | err = -EIO; |
4884 | goto out; | |
4885 | } | |
d1310b2e CM |
4886 | |
4887 | err = 0; | |
890871be | 4888 | write_lock(&em_tree->lock); |
a52d9a80 | 4889 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
4890 | /* it is possible that someone inserted the extent into the tree |
4891 | * while we had the lock dropped. It is also possible that | |
4892 | * an overlapping map exists in the tree | |
4893 | */ | |
a52d9a80 | 4894 | if (ret == -EEXIST) { |
3b951516 | 4895 | struct extent_map *existing; |
e6dcd2dc CM |
4896 | |
4897 | ret = 0; | |
4898 | ||
3b951516 | 4899 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
4900 | if (existing && (existing->start > start || |
4901 | existing->start + existing->len <= start)) { | |
4902 | free_extent_map(existing); | |
4903 | existing = NULL; | |
4904 | } | |
3b951516 CM |
4905 | if (!existing) { |
4906 | existing = lookup_extent_mapping(em_tree, em->start, | |
4907 | em->len); | |
4908 | if (existing) { | |
4909 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
4910 | em, start, |
4911 | root->sectorsize); | |
3b951516 CM |
4912 | free_extent_map(existing); |
4913 | if (err) { | |
4914 | free_extent_map(em); | |
4915 | em = NULL; | |
4916 | } | |
4917 | } else { | |
4918 | err = -EIO; | |
3b951516 CM |
4919 | free_extent_map(em); |
4920 | em = NULL; | |
4921 | } | |
4922 | } else { | |
4923 | free_extent_map(em); | |
4924 | em = existing; | |
e6dcd2dc | 4925 | err = 0; |
a52d9a80 | 4926 | } |
a52d9a80 | 4927 | } |
890871be | 4928 | write_unlock(&em_tree->lock); |
a52d9a80 | 4929 | out: |
f421950f CM |
4930 | if (path) |
4931 | btrfs_free_path(path); | |
a52d9a80 CM |
4932 | if (trans) { |
4933 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 4934 | if (!err) |
a52d9a80 CM |
4935 | err = ret; |
4936 | } | |
a52d9a80 CM |
4937 | if (err) { |
4938 | free_extent_map(em); | |
a52d9a80 CM |
4939 | return ERR_PTR(err); |
4940 | } | |
4941 | return em; | |
4942 | } | |
4943 | ||
16432985 CM |
4944 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
4945 | const struct iovec *iov, loff_t offset, | |
4946 | unsigned long nr_segs) | |
4947 | { | |
e1c4b745 | 4948 | return -EINVAL; |
16432985 CM |
4949 | } |
4950 | ||
1506fcc8 YS |
4951 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4952 | __u64 start, __u64 len) | |
4953 | { | |
4954 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent); | |
4955 | } | |
4956 | ||
a52d9a80 | 4957 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 4958 | { |
d1310b2e CM |
4959 | struct extent_io_tree *tree; |
4960 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 4961 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 4962 | } |
1832a6d5 | 4963 | |
a52d9a80 | 4964 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 4965 | { |
d1310b2e | 4966 | struct extent_io_tree *tree; |
b888db2b CM |
4967 | |
4968 | ||
4969 | if (current->flags & PF_MEMALLOC) { | |
4970 | redirty_page_for_writepage(wbc, page); | |
4971 | unlock_page(page); | |
4972 | return 0; | |
4973 | } | |
d1310b2e | 4974 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 4975 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
4976 | } |
4977 | ||
f421950f CM |
4978 | int btrfs_writepages(struct address_space *mapping, |
4979 | struct writeback_control *wbc) | |
b293f02e | 4980 | { |
d1310b2e | 4981 | struct extent_io_tree *tree; |
771ed689 | 4982 | |
d1310b2e | 4983 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
4984 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
4985 | } | |
4986 | ||
3ab2fb5a CM |
4987 | static int |
4988 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
4989 | struct list_head *pages, unsigned nr_pages) | |
4990 | { | |
d1310b2e CM |
4991 | struct extent_io_tree *tree; |
4992 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
4993 | return extent_readpages(tree, mapping, pages, nr_pages, |
4994 | btrfs_get_extent); | |
4995 | } | |
e6dcd2dc | 4996 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 4997 | { |
d1310b2e CM |
4998 | struct extent_io_tree *tree; |
4999 | struct extent_map_tree *map; | |
a52d9a80 | 5000 | int ret; |
8c2383c3 | 5001 | |
d1310b2e CM |
5002 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
5003 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 5004 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
5005 | if (ret == 1) { |
5006 | ClearPagePrivate(page); | |
5007 | set_page_private(page, 0); | |
5008 | page_cache_release(page); | |
39279cc3 | 5009 | } |
a52d9a80 | 5010 | return ret; |
39279cc3 CM |
5011 | } |
5012 | ||
e6dcd2dc CM |
5013 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
5014 | { | |
98509cfc CM |
5015 | if (PageWriteback(page) || PageDirty(page)) |
5016 | return 0; | |
b335b003 | 5017 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
5018 | } |
5019 | ||
a52d9a80 | 5020 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 5021 | { |
d1310b2e | 5022 | struct extent_io_tree *tree; |
e6dcd2dc CM |
5023 | struct btrfs_ordered_extent *ordered; |
5024 | u64 page_start = page_offset(page); | |
5025 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 5026 | |
8b62b72b CM |
5027 | |
5028 | /* | |
5029 | * we have the page locked, so new writeback can't start, | |
5030 | * and the dirty bit won't be cleared while we are here. | |
5031 | * | |
5032 | * Wait for IO on this page so that we can safely clear | |
5033 | * the PagePrivate2 bit and do ordered accounting | |
5034 | */ | |
e6dcd2dc | 5035 | wait_on_page_writeback(page); |
8b62b72b | 5036 | |
d1310b2e | 5037 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
5038 | if (offset) { |
5039 | btrfs_releasepage(page, GFP_NOFS); | |
5040 | return; | |
5041 | } | |
e6dcd2dc CM |
5042 | lock_extent(tree, page_start, page_end, GFP_NOFS); |
5043 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, | |
5044 | page_offset(page)); | |
5045 | if (ordered) { | |
eb84ae03 CM |
5046 | /* |
5047 | * IO on this page will never be started, so we need | |
5048 | * to account for any ordered extents now | |
5049 | */ | |
e6dcd2dc CM |
5050 | clear_extent_bit(tree, page_start, page_end, |
5051 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff JB |
5052 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
5053 | NULL, GFP_NOFS); | |
8b62b72b CM |
5054 | /* |
5055 | * whoever cleared the private bit is responsible | |
5056 | * for the finish_ordered_io | |
5057 | */ | |
5058 | if (TestClearPagePrivate2(page)) { | |
5059 | btrfs_finish_ordered_io(page->mapping->host, | |
5060 | page_start, page_end); | |
5061 | } | |
e6dcd2dc CM |
5062 | btrfs_put_ordered_extent(ordered); |
5063 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
5064 | } | |
5065 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff JB |
5066 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
5067 | EXTENT_DO_ACCOUNTING, 1, 1, NULL, GFP_NOFS); | |
e6dcd2dc CM |
5068 | __btrfs_releasepage(page, GFP_NOFS); |
5069 | ||
4a096752 | 5070 | ClearPageChecked(page); |
9ad6b7bc | 5071 | if (PagePrivate(page)) { |
9ad6b7bc CM |
5072 | ClearPagePrivate(page); |
5073 | set_page_private(page, 0); | |
5074 | page_cache_release(page); | |
5075 | } | |
39279cc3 CM |
5076 | } |
5077 | ||
9ebefb18 CM |
5078 | /* |
5079 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
5080 | * called from a page fault handler when a page is first dirtied. Hence we must | |
5081 | * be careful to check for EOF conditions here. We set the page up correctly | |
5082 | * for a written page which means we get ENOSPC checking when writing into | |
5083 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
5084 | * support these features. | |
5085 | * | |
5086 | * We are not allowed to take the i_mutex here so we have to play games to | |
5087 | * protect against truncate races as the page could now be beyond EOF. Because | |
5088 | * vmtruncate() writes the inode size before removing pages, once we have the | |
5089 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
5090 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
5091 | * unlock the page. | |
5092 | */ | |
c2ec175c | 5093 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 5094 | { |
c2ec175c | 5095 | struct page *page = vmf->page; |
6da6abae | 5096 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 5097 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
5098 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
5099 | struct btrfs_ordered_extent *ordered; | |
5100 | char *kaddr; | |
5101 | unsigned long zero_start; | |
9ebefb18 | 5102 | loff_t size; |
1832a6d5 | 5103 | int ret; |
a52d9a80 | 5104 | u64 page_start; |
e6dcd2dc | 5105 | u64 page_end; |
9ebefb18 | 5106 | |
6a63209f | 5107 | ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); |
56a76f82 NP |
5108 | if (ret) { |
5109 | if (ret == -ENOMEM) | |
5110 | ret = VM_FAULT_OOM; | |
5111 | else /* -ENOSPC, -EIO, etc */ | |
5112 | ret = VM_FAULT_SIGBUS; | |
1832a6d5 | 5113 | goto out; |
56a76f82 | 5114 | } |
1832a6d5 | 5115 | |
9ed74f2d JB |
5116 | ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); |
5117 | if (ret) { | |
5118 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
5119 | ret = VM_FAULT_SIGBUS; | |
5120 | goto out; | |
5121 | } | |
5122 | ||
56a76f82 | 5123 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 5124 | again: |
9ebefb18 | 5125 | lock_page(page); |
9ebefb18 | 5126 | size = i_size_read(inode); |
e6dcd2dc CM |
5127 | page_start = page_offset(page); |
5128 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 5129 | |
9ebefb18 | 5130 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 5131 | (page_start >= size)) { |
6a63209f | 5132 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); |
9ebefb18 CM |
5133 | /* page got truncated out from underneath us */ |
5134 | goto out_unlock; | |
5135 | } | |
e6dcd2dc CM |
5136 | wait_on_page_writeback(page); |
5137 | ||
5138 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
5139 | set_page_extent_mapped(page); | |
5140 | ||
eb84ae03 CM |
5141 | /* |
5142 | * we can't set the delalloc bits if there are pending ordered | |
5143 | * extents. Drop our locks and wait for them to finish | |
5144 | */ | |
e6dcd2dc CM |
5145 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
5146 | if (ordered) { | |
5147 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
5148 | unlock_page(page); | |
eb84ae03 | 5149 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
5150 | btrfs_put_ordered_extent(ordered); |
5151 | goto again; | |
5152 | } | |
5153 | ||
fbf19087 JB |
5154 | /* |
5155 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
5156 | * if it was already dirty, so for space accounting reasons we need to | |
5157 | * clear any delalloc bits for the range we are fixing to save. There | |
5158 | * is probably a better way to do this, but for now keep consistent with | |
5159 | * prepare_pages in the normal write path. | |
5160 | */ | |
5161 | clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, | |
32c00aff JB |
5162 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
5163 | GFP_NOFS); | |
fbf19087 | 5164 | |
9ed74f2d JB |
5165 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end); |
5166 | if (ret) { | |
5167 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
5168 | ret = VM_FAULT_SIGBUS; | |
fbf19087 | 5169 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); |
9ed74f2d JB |
5170 | goto out_unlock; |
5171 | } | |
e6dcd2dc | 5172 | ret = 0; |
9ebefb18 CM |
5173 | |
5174 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 5175 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 5176 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 5177 | else |
e6dcd2dc | 5178 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 5179 | |
e6dcd2dc CM |
5180 | if (zero_start != PAGE_CACHE_SIZE) { |
5181 | kaddr = kmap(page); | |
5182 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
5183 | flush_dcache_page(page); | |
5184 | kunmap(page); | |
5185 | } | |
247e743c | 5186 | ClearPageChecked(page); |
e6dcd2dc | 5187 | set_page_dirty(page); |
50a9b214 | 5188 | SetPageUptodate(page); |
5a3f23d5 | 5189 | |
257c62e1 CM |
5190 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
5191 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
5192 | ||
e6dcd2dc | 5193 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
9ebefb18 CM |
5194 | |
5195 | out_unlock: | |
9ed74f2d | 5196 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); |
50a9b214 CM |
5197 | if (!ret) |
5198 | return VM_FAULT_LOCKED; | |
9ebefb18 | 5199 | unlock_page(page); |
1832a6d5 | 5200 | out: |
9ebefb18 CM |
5201 | return ret; |
5202 | } | |
5203 | ||
39279cc3 CM |
5204 | static void btrfs_truncate(struct inode *inode) |
5205 | { | |
5206 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5207 | int ret; | |
5208 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 5209 | unsigned long nr; |
dbe674a9 | 5210 | u64 mask = root->sectorsize - 1; |
39279cc3 | 5211 | |
8082510e YZ |
5212 | if (!S_ISREG(inode->i_mode)) { |
5213 | WARN_ON(1); | |
39279cc3 | 5214 | return; |
8082510e | 5215 | } |
39279cc3 | 5216 | |
5d5e103a JB |
5217 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
5218 | if (ret) | |
5219 | return; | |
8082510e | 5220 | |
4a096752 | 5221 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 5222 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 5223 | |
39279cc3 | 5224 | trans = btrfs_start_transaction(root, 1); |
8082510e | 5225 | btrfs_set_trans_block_group(trans, inode); |
5a3f23d5 CM |
5226 | |
5227 | /* | |
5228 | * setattr is responsible for setting the ordered_data_close flag, | |
5229 | * but that is only tested during the last file release. That | |
5230 | * could happen well after the next commit, leaving a great big | |
5231 | * window where new writes may get lost if someone chooses to write | |
5232 | * to this file after truncating to zero | |
5233 | * | |
5234 | * The inode doesn't have any dirty data here, and so if we commit | |
5235 | * this is a noop. If someone immediately starts writing to the inode | |
5236 | * it is very likely we'll catch some of their writes in this | |
5237 | * transaction, and the commit will find this file on the ordered | |
5238 | * data list with good things to send down. | |
5239 | * | |
5240 | * This is a best effort solution, there is still a window where | |
5241 | * using truncate to replace the contents of the file will | |
5242 | * end up with a zero length file after a crash. | |
5243 | */ | |
5244 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
5245 | btrfs_add_ordered_operation(trans, root, inode); | |
5246 | ||
8082510e YZ |
5247 | while (1) { |
5248 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
5249 | inode->i_size, | |
5250 | BTRFS_EXTENT_DATA_KEY); | |
5251 | if (ret != -EAGAIN) | |
5252 | break; | |
39279cc3 | 5253 | |
8082510e YZ |
5254 | ret = btrfs_update_inode(trans, root, inode); |
5255 | BUG_ON(ret); | |
5f39d397 | 5256 | |
8082510e YZ |
5257 | nr = trans->blocks_used; |
5258 | btrfs_end_transaction(trans, root); | |
5259 | btrfs_btree_balance_dirty(root, nr); | |
5260 | ||
5261 | trans = btrfs_start_transaction(root, 1); | |
5262 | btrfs_set_trans_block_group(trans, inode); | |
5263 | } | |
5264 | ||
5265 | if (ret == 0 && inode->i_nlink > 0) { | |
5266 | ret = btrfs_orphan_del(trans, inode); | |
5267 | BUG_ON(ret); | |
5268 | } | |
5269 | ||
5270 | ret = btrfs_update_inode(trans, root, inode); | |
7b128766 JB |
5271 | BUG_ON(ret); |
5272 | ||
7b128766 | 5273 | nr = trans->blocks_used; |
89ce8a63 | 5274 | ret = btrfs_end_transaction_throttle(trans, root); |
39279cc3 | 5275 | BUG_ON(ret); |
d3c2fdcf | 5276 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5277 | } |
5278 | ||
d352ac68 CM |
5279 | /* |
5280 | * create a new subvolume directory/inode (helper for the ioctl). | |
5281 | */ | |
d2fb3437 | 5282 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
76dda93c | 5283 | struct btrfs_root *new_root, |
d2fb3437 | 5284 | u64 new_dirid, u64 alloc_hint) |
39279cc3 | 5285 | { |
39279cc3 | 5286 | struct inode *inode; |
76dda93c | 5287 | int err; |
00e4e6b3 | 5288 | u64 index = 0; |
39279cc3 | 5289 | |
aec7477b | 5290 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d2fb3437 | 5291 | new_dirid, alloc_hint, S_IFDIR | 0700, &index); |
54aa1f4d | 5292 | if (IS_ERR(inode)) |
f46b5a66 | 5293 | return PTR_ERR(inode); |
39279cc3 CM |
5294 | inode->i_op = &btrfs_dir_inode_operations; |
5295 | inode->i_fop = &btrfs_dir_file_operations; | |
5296 | ||
39279cc3 | 5297 | inode->i_nlink = 1; |
dbe674a9 | 5298 | btrfs_i_size_write(inode, 0); |
3b96362c | 5299 | |
76dda93c YZ |
5300 | err = btrfs_update_inode(trans, new_root, inode); |
5301 | BUG_ON(err); | |
cb8e7090 | 5302 | |
76dda93c | 5303 | iput(inode); |
cb8e7090 | 5304 | return 0; |
39279cc3 CM |
5305 | } |
5306 | ||
d352ac68 CM |
5307 | /* helper function for file defrag and space balancing. This |
5308 | * forces readahead on a given range of bytes in an inode | |
5309 | */ | |
edbd8d4e | 5310 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
5311 | struct file_ra_state *ra, struct file *file, |
5312 | pgoff_t offset, pgoff_t last_index) | |
5313 | { | |
8e7bf94f | 5314 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 5315 | |
86479a04 CM |
5316 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
5317 | return offset + req_size; | |
86479a04 CM |
5318 | } |
5319 | ||
39279cc3 CM |
5320 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
5321 | { | |
5322 | struct btrfs_inode *ei; | |
5323 | ||
5324 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
5325 | if (!ei) | |
5326 | return NULL; | |
15ee9bc7 | 5327 | ei->last_trans = 0; |
257c62e1 | 5328 | ei->last_sub_trans = 0; |
e02119d5 | 5329 | ei->logged_trans = 0; |
32c00aff JB |
5330 | ei->outstanding_extents = 0; |
5331 | ei->reserved_extents = 0; | |
a6dbd429 | 5332 | ei->root = NULL; |
32c00aff | 5333 | spin_lock_init(&ei->accounting_lock); |
e6dcd2dc | 5334 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 5335 | INIT_LIST_HEAD(&ei->i_orphan); |
5a3f23d5 | 5336 | INIT_LIST_HEAD(&ei->ordered_operations); |
39279cc3 CM |
5337 | return &ei->vfs_inode; |
5338 | } | |
5339 | ||
5340 | void btrfs_destroy_inode(struct inode *inode) | |
5341 | { | |
e6dcd2dc | 5342 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
5343 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5344 | ||
39279cc3 CM |
5345 | WARN_ON(!list_empty(&inode->i_dentry)); |
5346 | WARN_ON(inode->i_data.nrpages); | |
5347 | ||
a6dbd429 JB |
5348 | /* |
5349 | * This can happen where we create an inode, but somebody else also | |
5350 | * created the same inode and we need to destroy the one we already | |
5351 | * created. | |
5352 | */ | |
5353 | if (!root) | |
5354 | goto free; | |
5355 | ||
5a3f23d5 CM |
5356 | /* |
5357 | * Make sure we're properly removed from the ordered operation | |
5358 | * lists. | |
5359 | */ | |
5360 | smp_mb(); | |
5361 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
5362 | spin_lock(&root->fs_info->ordered_extent_lock); | |
5363 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
5364 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
5365 | } | |
5366 | ||
5367 | spin_lock(&root->list_lock); | |
7b128766 | 5368 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
8082510e YZ |
5369 | printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n", |
5370 | inode->i_ino); | |
5371 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
7b128766 | 5372 | } |
5a3f23d5 | 5373 | spin_unlock(&root->list_lock); |
7b128766 | 5374 | |
d397712b | 5375 | while (1) { |
e6dcd2dc CM |
5376 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
5377 | if (!ordered) | |
5378 | break; | |
5379 | else { | |
d397712b CM |
5380 | printk(KERN_ERR "btrfs found ordered " |
5381 | "extent %llu %llu on inode cleanup\n", | |
5382 | (unsigned long long)ordered->file_offset, | |
5383 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
5384 | btrfs_remove_ordered_extent(inode, ordered); |
5385 | btrfs_put_ordered_extent(ordered); | |
5386 | btrfs_put_ordered_extent(ordered); | |
5387 | } | |
5388 | } | |
5d4f98a2 | 5389 | inode_tree_del(inode); |
5b21f2ed | 5390 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 5391 | free: |
39279cc3 CM |
5392 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
5393 | } | |
5394 | ||
76dda93c YZ |
5395 | void btrfs_drop_inode(struct inode *inode) |
5396 | { | |
5397 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5398 | ||
5399 | if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0) | |
5400 | generic_delete_inode(inode); | |
5401 | else | |
5402 | generic_drop_inode(inode); | |
5403 | } | |
5404 | ||
0ee0fda0 | 5405 | static void init_once(void *foo) |
39279cc3 CM |
5406 | { |
5407 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
5408 | ||
5409 | inode_init_once(&ei->vfs_inode); | |
5410 | } | |
5411 | ||
5412 | void btrfs_destroy_cachep(void) | |
5413 | { | |
5414 | if (btrfs_inode_cachep) | |
5415 | kmem_cache_destroy(btrfs_inode_cachep); | |
5416 | if (btrfs_trans_handle_cachep) | |
5417 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5418 | if (btrfs_transaction_cachep) | |
5419 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
5420 | if (btrfs_path_cachep) |
5421 | kmem_cache_destroy(btrfs_path_cachep); | |
5422 | } | |
5423 | ||
5424 | int btrfs_init_cachep(void) | |
5425 | { | |
9601e3f6 CH |
5426 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
5427 | sizeof(struct btrfs_inode), 0, | |
5428 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
5429 | if (!btrfs_inode_cachep) |
5430 | goto fail; | |
9601e3f6 CH |
5431 | |
5432 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
5433 | sizeof(struct btrfs_trans_handle), 0, | |
5434 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5435 | if (!btrfs_trans_handle_cachep) |
5436 | goto fail; | |
9601e3f6 CH |
5437 | |
5438 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
5439 | sizeof(struct btrfs_transaction), 0, | |
5440 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5441 | if (!btrfs_transaction_cachep) |
5442 | goto fail; | |
9601e3f6 CH |
5443 | |
5444 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
5445 | sizeof(struct btrfs_path), 0, | |
5446 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5447 | if (!btrfs_path_cachep) |
5448 | goto fail; | |
9601e3f6 | 5449 | |
39279cc3 CM |
5450 | return 0; |
5451 | fail: | |
5452 | btrfs_destroy_cachep(); | |
5453 | return -ENOMEM; | |
5454 | } | |
5455 | ||
5456 | static int btrfs_getattr(struct vfsmount *mnt, | |
5457 | struct dentry *dentry, struct kstat *stat) | |
5458 | { | |
5459 | struct inode *inode = dentry->d_inode; | |
5460 | generic_fillattr(inode, stat); | |
3394e160 | 5461 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 5462 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
5463 | stat->blocks = (inode_get_bytes(inode) + |
5464 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
5465 | return 0; |
5466 | } | |
5467 | ||
d397712b CM |
5468 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
5469 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
5470 | { |
5471 | struct btrfs_trans_handle *trans; | |
5472 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 5473 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
5474 | struct inode *new_inode = new_dentry->d_inode; |
5475 | struct inode *old_inode = old_dentry->d_inode; | |
5476 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 5477 | u64 index = 0; |
4df27c4d | 5478 | u64 root_objectid; |
39279cc3 CM |
5479 | int ret; |
5480 | ||
f679a840 YZ |
5481 | if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
5482 | return -EPERM; | |
5483 | ||
4df27c4d YZ |
5484 | /* we only allow rename subvolume link between subvolumes */ |
5485 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
3394e160 CM |
5486 | return -EXDEV; |
5487 | ||
4df27c4d YZ |
5488 | if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
5489 | (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 5490 | return -ENOTEMPTY; |
5f39d397 | 5491 | |
4df27c4d YZ |
5492 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
5493 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
5494 | return -ENOTEMPTY; | |
0660b5af | 5495 | |
9ed74f2d | 5496 | /* |
5df6a9f6 JB |
5497 | * We want to reserve the absolute worst case amount of items. So if |
5498 | * both inodes are subvols and we need to unlink them then that would | |
5499 | * require 4 item modifications, but if they are both normal inodes it | |
5500 | * would require 5 item modifications, so we'll assume their normal | |
5501 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
5502 | * should cover the worst case number of items we'll modify. | |
9ed74f2d | 5503 | */ |
5df6a9f6 | 5504 | ret = btrfs_reserve_metadata_space(root, 11); |
1832a6d5 | 5505 | if (ret) |
4df27c4d | 5506 | return ret; |
1832a6d5 | 5507 | |
5a3f23d5 CM |
5508 | /* |
5509 | * we're using rename to replace one file with another. | |
5510 | * and the replacement file is large. Start IO on it now so | |
5511 | * we don't add too much work to the end of the transaction | |
5512 | */ | |
4baf8c92 | 5513 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
5514 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
5515 | filemap_flush(old_inode->i_mapping); | |
5516 | ||
76dda93c YZ |
5517 | /* close the racy window with snapshot create/destroy ioctl */ |
5518 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) | |
5519 | down_read(&root->fs_info->subvol_sem); | |
5520 | ||
39279cc3 | 5521 | trans = btrfs_start_transaction(root, 1); |
a5719521 | 5522 | btrfs_set_trans_block_group(trans, new_dir); |
5f39d397 | 5523 | |
4df27c4d YZ |
5524 | if (dest != root) |
5525 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 5526 | |
a5719521 YZ |
5527 | ret = btrfs_set_inode_index(new_dir, &index); |
5528 | if (ret) | |
5529 | goto out_fail; | |
5a3f23d5 | 5530 | |
a5719521 | 5531 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5532 | /* force full log commit if subvolume involved. */ |
5533 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
5534 | } else { | |
a5719521 YZ |
5535 | ret = btrfs_insert_inode_ref(trans, dest, |
5536 | new_dentry->d_name.name, | |
5537 | new_dentry->d_name.len, | |
5538 | old_inode->i_ino, | |
5539 | new_dir->i_ino, index); | |
5540 | if (ret) | |
5541 | goto out_fail; | |
4df27c4d YZ |
5542 | /* |
5543 | * this is an ugly little race, but the rename is required | |
5544 | * to make sure that if we crash, the inode is either at the | |
5545 | * old name or the new one. pinning the log transaction lets | |
5546 | * us make sure we don't allow a log commit to come in after | |
5547 | * we unlink the name but before we add the new name back in. | |
5548 | */ | |
5549 | btrfs_pin_log_trans(root); | |
5550 | } | |
5a3f23d5 CM |
5551 | /* |
5552 | * make sure the inode gets flushed if it is replacing | |
5553 | * something. | |
5554 | */ | |
5555 | if (new_inode && new_inode->i_size && | |
5556 | old_inode && S_ISREG(old_inode->i_mode)) { | |
5557 | btrfs_add_ordered_operation(trans, root, old_inode); | |
5558 | } | |
5559 | ||
39279cc3 CM |
5560 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
5561 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
5562 | old_inode->i_ctime = ctime; | |
5f39d397 | 5563 | |
12fcfd22 CM |
5564 | if (old_dentry->d_parent != new_dentry->d_parent) |
5565 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
5566 | ||
4df27c4d YZ |
5567 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
5568 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
5569 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
5570 | old_dentry->d_name.name, | |
5571 | old_dentry->d_name.len); | |
5572 | } else { | |
5573 | btrfs_inc_nlink(old_dentry->d_inode); | |
5574 | ret = btrfs_unlink_inode(trans, root, old_dir, | |
5575 | old_dentry->d_inode, | |
5576 | old_dentry->d_name.name, | |
5577 | old_dentry->d_name.len); | |
5578 | } | |
5579 | BUG_ON(ret); | |
39279cc3 CM |
5580 | |
5581 | if (new_inode) { | |
5582 | new_inode->i_ctime = CURRENT_TIME; | |
4df27c4d YZ |
5583 | if (unlikely(new_inode->i_ino == |
5584 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { | |
5585 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
5586 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
5587 | root_objectid, | |
5588 | new_dentry->d_name.name, | |
5589 | new_dentry->d_name.len); | |
5590 | BUG_ON(new_inode->i_nlink == 0); | |
5591 | } else { | |
5592 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
5593 | new_dentry->d_inode, | |
5594 | new_dentry->d_name.name, | |
5595 | new_dentry->d_name.len); | |
5596 | } | |
5597 | BUG_ON(ret); | |
7b128766 | 5598 | if (new_inode->i_nlink == 0) { |
e02119d5 | 5599 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 5600 | BUG_ON(ret); |
7b128766 | 5601 | } |
39279cc3 | 5602 | } |
aec7477b | 5603 | |
4df27c4d YZ |
5604 | ret = btrfs_add_link(trans, new_dir, old_inode, |
5605 | new_dentry->d_name.name, | |
a5719521 | 5606 | new_dentry->d_name.len, 0, index); |
4df27c4d | 5607 | BUG_ON(ret); |
39279cc3 | 5608 | |
4df27c4d YZ |
5609 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) { |
5610 | btrfs_log_new_name(trans, old_inode, old_dir, | |
5611 | new_dentry->d_parent); | |
5612 | btrfs_end_log_trans(root); | |
5613 | } | |
39279cc3 | 5614 | out_fail: |
ab78c84d | 5615 | btrfs_end_transaction_throttle(trans, root); |
4df27c4d | 5616 | |
76dda93c YZ |
5617 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
5618 | up_read(&root->fs_info->subvol_sem); | |
9ed74f2d | 5619 | |
5df6a9f6 | 5620 | btrfs_unreserve_metadata_space(root, 11); |
39279cc3 CM |
5621 | return ret; |
5622 | } | |
5623 | ||
d352ac68 CM |
5624 | /* |
5625 | * some fairly slow code that needs optimization. This walks the list | |
5626 | * of all the inodes with pending delalloc and forces them to disk. | |
5627 | */ | |
24bbcf04 | 5628 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 CM |
5629 | { |
5630 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
5631 | struct btrfs_inode *binode; | |
5b21f2ed | 5632 | struct inode *inode; |
ea8c2819 | 5633 | |
c146afad YZ |
5634 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
5635 | return -EROFS; | |
5636 | ||
75eff68e | 5637 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 5638 | while (!list_empty(head)) { |
ea8c2819 CM |
5639 | binode = list_entry(head->next, struct btrfs_inode, |
5640 | delalloc_inodes); | |
5b21f2ed ZY |
5641 | inode = igrab(&binode->vfs_inode); |
5642 | if (!inode) | |
5643 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 5644 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 5645 | if (inode) { |
8c8bee1d | 5646 | filemap_flush(inode->i_mapping); |
24bbcf04 YZ |
5647 | if (delay_iput) |
5648 | btrfs_add_delayed_iput(inode); | |
5649 | else | |
5650 | iput(inode); | |
5b21f2ed ZY |
5651 | } |
5652 | cond_resched(); | |
75eff68e | 5653 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 5654 | } |
75eff68e | 5655 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
5656 | |
5657 | /* the filemap_flush will queue IO into the worker threads, but | |
5658 | * we have to make sure the IO is actually started and that | |
5659 | * ordered extents get created before we return | |
5660 | */ | |
5661 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 5662 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 5663 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 5664 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
5665 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
5666 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
5667 | } |
5668 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
5669 | return 0; |
5670 | } | |
5671 | ||
39279cc3 CM |
5672 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
5673 | const char *symname) | |
5674 | { | |
5675 | struct btrfs_trans_handle *trans; | |
5676 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5677 | struct btrfs_path *path; | |
5678 | struct btrfs_key key; | |
1832a6d5 | 5679 | struct inode *inode = NULL; |
39279cc3 CM |
5680 | int err; |
5681 | int drop_inode = 0; | |
5682 | u64 objectid; | |
00e4e6b3 | 5683 | u64 index = 0 ; |
39279cc3 CM |
5684 | int name_len; |
5685 | int datasize; | |
5f39d397 | 5686 | unsigned long ptr; |
39279cc3 | 5687 | struct btrfs_file_extent_item *ei; |
5f39d397 | 5688 | struct extent_buffer *leaf; |
1832a6d5 | 5689 | unsigned long nr = 0; |
39279cc3 CM |
5690 | |
5691 | name_len = strlen(symname) + 1; | |
5692 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
5693 | return -ENAMETOOLONG; | |
1832a6d5 | 5694 | |
9ed74f2d JB |
5695 | /* |
5696 | * 2 items for inode item and ref | |
5697 | * 2 items for dir items | |
5698 | * 1 item for xattr if selinux is on | |
5699 | */ | |
5700 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 5701 | if (err) |
9ed74f2d | 5702 | return err; |
1832a6d5 | 5703 | |
39279cc3 | 5704 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
5705 | if (!trans) |
5706 | goto out_fail; | |
39279cc3 CM |
5707 | btrfs_set_trans_block_group(trans, dir); |
5708 | ||
5709 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
5710 | if (err) { | |
5711 | err = -ENOSPC; | |
5712 | goto out_unlock; | |
5713 | } | |
5714 | ||
aec7477b | 5715 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
5716 | dentry->d_name.len, |
5717 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
5718 | BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, |
5719 | &index); | |
39279cc3 CM |
5720 | err = PTR_ERR(inode); |
5721 | if (IS_ERR(inode)) | |
5722 | goto out_unlock; | |
5723 | ||
f34f57a3 | 5724 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
5725 | if (err) { |
5726 | drop_inode = 1; | |
5727 | goto out_unlock; | |
5728 | } | |
5729 | ||
39279cc3 | 5730 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 5731 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
5732 | if (err) |
5733 | drop_inode = 1; | |
5734 | else { | |
5735 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 5736 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
5737 | inode->i_fop = &btrfs_file_operations; |
5738 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 5739 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 5740 | } |
39279cc3 CM |
5741 | btrfs_update_inode_block_group(trans, inode); |
5742 | btrfs_update_inode_block_group(trans, dir); | |
5743 | if (drop_inode) | |
5744 | goto out_unlock; | |
5745 | ||
5746 | path = btrfs_alloc_path(); | |
5747 | BUG_ON(!path); | |
5748 | key.objectid = inode->i_ino; | |
5749 | key.offset = 0; | |
39279cc3 CM |
5750 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
5751 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
5752 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
5753 | datasize); | |
54aa1f4d CM |
5754 | if (err) { |
5755 | drop_inode = 1; | |
5756 | goto out_unlock; | |
5757 | } | |
5f39d397 CM |
5758 | leaf = path->nodes[0]; |
5759 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
5760 | struct btrfs_file_extent_item); | |
5761 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
5762 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 5763 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
5764 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
5765 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
5766 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
5767 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
5768 | ||
39279cc3 | 5769 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
5770 | write_extent_buffer(leaf, symname, ptr, name_len); |
5771 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 5772 | btrfs_free_path(path); |
5f39d397 | 5773 | |
39279cc3 CM |
5774 | inode->i_op = &btrfs_symlink_inode_operations; |
5775 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 5776 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 5777 | inode_set_bytes(inode, name_len); |
dbe674a9 | 5778 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
5779 | err = btrfs_update_inode(trans, root, inode); |
5780 | if (err) | |
5781 | drop_inode = 1; | |
39279cc3 CM |
5782 | |
5783 | out_unlock: | |
d3c2fdcf | 5784 | nr = trans->blocks_used; |
ab78c84d | 5785 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 5786 | out_fail: |
9ed74f2d | 5787 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
5788 | if (drop_inode) { |
5789 | inode_dec_link_count(inode); | |
5790 | iput(inode); | |
5791 | } | |
d3c2fdcf | 5792 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5793 | return err; |
5794 | } | |
16432985 | 5795 | |
5a303d5d | 5796 | static int prealloc_file_range(struct inode *inode, u64 start, u64 end, |
d1ea6a61 | 5797 | u64 alloc_hint, int mode, loff_t actual_len) |
d899e052 | 5798 | { |
5a303d5d | 5799 | struct btrfs_trans_handle *trans; |
d899e052 YZ |
5800 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5801 | struct btrfs_key ins; | |
5802 | u64 alloc_size; | |
5803 | u64 cur_offset = start; | |
5804 | u64 num_bytes = end - start; | |
5805 | int ret = 0; | |
d1ea6a61 | 5806 | u64 i_size; |
d899e052 | 5807 | |
d899e052 YZ |
5808 | while (num_bytes > 0) { |
5809 | alloc_size = min(num_bytes, root->fs_info->max_extent); | |
9ed74f2d | 5810 | |
3a1abec9 CM |
5811 | trans = btrfs_start_transaction(root, 1); |
5812 | ||
d899e052 YZ |
5813 | ret = btrfs_reserve_extent(trans, root, alloc_size, |
5814 | root->sectorsize, 0, alloc_hint, | |
5815 | (u64)-1, &ins, 1); | |
5816 | if (ret) { | |
5817 | WARN_ON(1); | |
3a1abec9 | 5818 | goto stop_trans; |
5a303d5d YZ |
5819 | } |
5820 | ||
5821 | ret = btrfs_reserve_metadata_space(root, 3); | |
5822 | if (ret) { | |
5823 | btrfs_free_reserved_extent(root, ins.objectid, | |
5824 | ins.offset); | |
3a1abec9 | 5825 | goto stop_trans; |
d899e052 | 5826 | } |
5a303d5d | 5827 | |
d899e052 YZ |
5828 | ret = insert_reserved_file_extent(trans, inode, |
5829 | cur_offset, ins.objectid, | |
5830 | ins.offset, ins.offset, | |
920bbbfb | 5831 | ins.offset, 0, 0, 0, |
d899e052 YZ |
5832 | BTRFS_FILE_EXTENT_PREALLOC); |
5833 | BUG_ON(ret); | |
a1ed835e CM |
5834 | btrfs_drop_extent_cache(inode, cur_offset, |
5835 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 5836 | |
d899e052 YZ |
5837 | num_bytes -= ins.offset; |
5838 | cur_offset += ins.offset; | |
5839 | alloc_hint = ins.objectid + ins.offset; | |
5a303d5d | 5840 | |
d899e052 | 5841 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 5842 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 5843 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
5a303d5d | 5844 | cur_offset > inode->i_size) { |
d1ea6a61 AK |
5845 | if (cur_offset > actual_len) |
5846 | i_size = actual_len; | |
5847 | else | |
5848 | i_size = cur_offset; | |
5849 | i_size_write(inode, i_size); | |
5850 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
5851 | } |
5852 | ||
d899e052 YZ |
5853 | ret = btrfs_update_inode(trans, root, inode); |
5854 | BUG_ON(ret); | |
d899e052 | 5855 | |
5a303d5d YZ |
5856 | btrfs_end_transaction(trans, root); |
5857 | btrfs_unreserve_metadata_space(root, 3); | |
5858 | } | |
d899e052 | 5859 | return ret; |
3a1abec9 CM |
5860 | |
5861 | stop_trans: | |
5862 | btrfs_end_transaction(trans, root); | |
5863 | return ret; | |
5864 | ||
d899e052 YZ |
5865 | } |
5866 | ||
5867 | static long btrfs_fallocate(struct inode *inode, int mode, | |
5868 | loff_t offset, loff_t len) | |
5869 | { | |
5870 | u64 cur_offset; | |
5871 | u64 last_byte; | |
5872 | u64 alloc_start; | |
5873 | u64 alloc_end; | |
5874 | u64 alloc_hint = 0; | |
e980b50c | 5875 | u64 locked_end; |
d899e052 YZ |
5876 | u64 mask = BTRFS_I(inode)->root->sectorsize - 1; |
5877 | struct extent_map *em; | |
5878 | int ret; | |
5879 | ||
5880 | alloc_start = offset & ~mask; | |
5881 | alloc_end = (offset + len + mask) & ~mask; | |
5882 | ||
546888da CM |
5883 | /* |
5884 | * wait for ordered IO before we have any locks. We'll loop again | |
5885 | * below with the locks held. | |
5886 | */ | |
5887 | btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start); | |
5888 | ||
d899e052 YZ |
5889 | mutex_lock(&inode->i_mutex); |
5890 | if (alloc_start > inode->i_size) { | |
5891 | ret = btrfs_cont_expand(inode, alloc_start); | |
5892 | if (ret) | |
5893 | goto out; | |
5894 | } | |
5895 | ||
5a303d5d | 5896 | ret = btrfs_check_data_free_space(BTRFS_I(inode)->root, inode, |
a970b0a1 JB |
5897 | alloc_end - alloc_start); |
5898 | if (ret) | |
5899 | goto out; | |
5900 | ||
e980b50c | 5901 | locked_end = alloc_end - 1; |
d899e052 YZ |
5902 | while (1) { |
5903 | struct btrfs_ordered_extent *ordered; | |
546888da | 5904 | |
546888da CM |
5905 | /* the extent lock is ordered inside the running |
5906 | * transaction | |
5907 | */ | |
e980b50c CM |
5908 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
5909 | GFP_NOFS); | |
d899e052 YZ |
5910 | ordered = btrfs_lookup_first_ordered_extent(inode, |
5911 | alloc_end - 1); | |
5912 | if (ordered && | |
5913 | ordered->file_offset + ordered->len > alloc_start && | |
5914 | ordered->file_offset < alloc_end) { | |
5915 | btrfs_put_ordered_extent(ordered); | |
5916 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
e980b50c | 5917 | alloc_start, locked_end, GFP_NOFS); |
546888da CM |
5918 | /* |
5919 | * we can't wait on the range with the transaction | |
5920 | * running or with the extent lock held | |
5921 | */ | |
d899e052 YZ |
5922 | btrfs_wait_ordered_range(inode, alloc_start, |
5923 | alloc_end - alloc_start); | |
5924 | } else { | |
5925 | if (ordered) | |
5926 | btrfs_put_ordered_extent(ordered); | |
5927 | break; | |
5928 | } | |
5929 | } | |
5930 | ||
5931 | cur_offset = alloc_start; | |
5932 | while (1) { | |
5933 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
5934 | alloc_end - cur_offset, 0); | |
5935 | BUG_ON(IS_ERR(em) || !em); | |
5936 | last_byte = min(extent_map_end(em), alloc_end); | |
5937 | last_byte = (last_byte + mask) & ~mask; | |
5a303d5d YZ |
5938 | if (em->block_start == EXTENT_MAP_HOLE || |
5939 | (cur_offset >= inode->i_size && | |
5940 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
5941 | ret = prealloc_file_range(inode, | |
5942 | cur_offset, last_byte, | |
d1ea6a61 | 5943 | alloc_hint, mode, offset+len); |
d899e052 YZ |
5944 | if (ret < 0) { |
5945 | free_extent_map(em); | |
5946 | break; | |
5947 | } | |
5948 | } | |
5949 | if (em->block_start <= EXTENT_MAP_LAST_BYTE) | |
5950 | alloc_hint = em->block_start; | |
5951 | free_extent_map(em); | |
5952 | ||
5953 | cur_offset = last_byte; | |
5954 | if (cur_offset >= alloc_end) { | |
5955 | ret = 0; | |
5956 | break; | |
5957 | } | |
5958 | } | |
e980b50c | 5959 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
d899e052 | 5960 | GFP_NOFS); |
546888da | 5961 | |
5a303d5d YZ |
5962 | btrfs_free_reserved_data_space(BTRFS_I(inode)->root, inode, |
5963 | alloc_end - alloc_start); | |
d899e052 YZ |
5964 | out: |
5965 | mutex_unlock(&inode->i_mutex); | |
5966 | return ret; | |
5967 | } | |
5968 | ||
e6dcd2dc CM |
5969 | static int btrfs_set_page_dirty(struct page *page) |
5970 | { | |
e6dcd2dc CM |
5971 | return __set_page_dirty_nobuffers(page); |
5972 | } | |
5973 | ||
0ee0fda0 | 5974 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 5975 | { |
6cbff00f | 5976 | if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE)) |
fdebe2bd | 5977 | return -EACCES; |
33268eaf | 5978 | return generic_permission(inode, mask, btrfs_check_acl); |
fdebe2bd | 5979 | } |
39279cc3 | 5980 | |
6e1d5dcc | 5981 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 5982 | .getattr = btrfs_getattr, |
39279cc3 CM |
5983 | .lookup = btrfs_lookup, |
5984 | .create = btrfs_create, | |
5985 | .unlink = btrfs_unlink, | |
5986 | .link = btrfs_link, | |
5987 | .mkdir = btrfs_mkdir, | |
5988 | .rmdir = btrfs_rmdir, | |
5989 | .rename = btrfs_rename, | |
5990 | .symlink = btrfs_symlink, | |
5991 | .setattr = btrfs_setattr, | |
618e21d5 | 5992 | .mknod = btrfs_mknod, |
95819c05 CH |
5993 | .setxattr = btrfs_setxattr, |
5994 | .getxattr = btrfs_getxattr, | |
5103e947 | 5995 | .listxattr = btrfs_listxattr, |
95819c05 | 5996 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5997 | .permission = btrfs_permission, |
39279cc3 | 5998 | }; |
6e1d5dcc | 5999 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 6000 | .lookup = btrfs_lookup, |
fdebe2bd | 6001 | .permission = btrfs_permission, |
39279cc3 | 6002 | }; |
76dda93c | 6003 | |
828c0950 | 6004 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
6005 | .llseek = generic_file_llseek, |
6006 | .read = generic_read_dir, | |
cbdf5a24 | 6007 | .readdir = btrfs_real_readdir, |
34287aa3 | 6008 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 6009 | #ifdef CONFIG_COMPAT |
34287aa3 | 6010 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 6011 | #endif |
6bf13c0c | 6012 | .release = btrfs_release_file, |
e02119d5 | 6013 | .fsync = btrfs_sync_file, |
39279cc3 CM |
6014 | }; |
6015 | ||
d1310b2e | 6016 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 6017 | .fill_delalloc = run_delalloc_range, |
065631f6 | 6018 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 6019 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 6020 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 6021 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 6022 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 6023 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
6024 | .set_bit_hook = btrfs_set_bit_hook, |
6025 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
6026 | .merge_extent_hook = btrfs_merge_extent_hook, |
6027 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
6028 | }; |
6029 | ||
35054394 CM |
6030 | /* |
6031 | * btrfs doesn't support the bmap operation because swapfiles | |
6032 | * use bmap to make a mapping of extents in the file. They assume | |
6033 | * these extents won't change over the life of the file and they | |
6034 | * use the bmap result to do IO directly to the drive. | |
6035 | * | |
6036 | * the btrfs bmap call would return logical addresses that aren't | |
6037 | * suitable for IO and they also will change frequently as COW | |
6038 | * operations happen. So, swapfile + btrfs == corruption. | |
6039 | * | |
6040 | * For now we're avoiding this by dropping bmap. | |
6041 | */ | |
7f09410b | 6042 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
6043 | .readpage = btrfs_readpage, |
6044 | .writepage = btrfs_writepage, | |
b293f02e | 6045 | .writepages = btrfs_writepages, |
3ab2fb5a | 6046 | .readpages = btrfs_readpages, |
39279cc3 | 6047 | .sync_page = block_sync_page, |
16432985 | 6048 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
6049 | .invalidatepage = btrfs_invalidatepage, |
6050 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 6051 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 6052 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
6053 | }; |
6054 | ||
7f09410b | 6055 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
6056 | .readpage = btrfs_readpage, |
6057 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
6058 | .invalidatepage = btrfs_invalidatepage, |
6059 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
6060 | }; |
6061 | ||
6e1d5dcc | 6062 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
6063 | .truncate = btrfs_truncate, |
6064 | .getattr = btrfs_getattr, | |
6065 | .setattr = btrfs_setattr, | |
95819c05 CH |
6066 | .setxattr = btrfs_setxattr, |
6067 | .getxattr = btrfs_getxattr, | |
5103e947 | 6068 | .listxattr = btrfs_listxattr, |
95819c05 | 6069 | .removexattr = btrfs_removexattr, |
fdebe2bd | 6070 | .permission = btrfs_permission, |
d899e052 | 6071 | .fallocate = btrfs_fallocate, |
1506fcc8 | 6072 | .fiemap = btrfs_fiemap, |
39279cc3 | 6073 | }; |
6e1d5dcc | 6074 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
6075 | .getattr = btrfs_getattr, |
6076 | .setattr = btrfs_setattr, | |
fdebe2bd | 6077 | .permission = btrfs_permission, |
95819c05 CH |
6078 | .setxattr = btrfs_setxattr, |
6079 | .getxattr = btrfs_getxattr, | |
33268eaf | 6080 | .listxattr = btrfs_listxattr, |
95819c05 | 6081 | .removexattr = btrfs_removexattr, |
618e21d5 | 6082 | }; |
6e1d5dcc | 6083 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
6084 | .readlink = generic_readlink, |
6085 | .follow_link = page_follow_link_light, | |
6086 | .put_link = page_put_link, | |
fdebe2bd | 6087 | .permission = btrfs_permission, |
0279b4cd JO |
6088 | .setxattr = btrfs_setxattr, |
6089 | .getxattr = btrfs_getxattr, | |
6090 | .listxattr = btrfs_listxattr, | |
6091 | .removexattr = btrfs_removexattr, | |
39279cc3 | 6092 | }; |
76dda93c | 6093 | |
82d339d9 | 6094 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
6095 | .d_delete = btrfs_dentry_delete, |
6096 | }; |