Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/inode.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * from | |
10 | * | |
11 | * linux/fs/minix/inode.c | |
12 | * | |
13 | * Copyright (C) 1991, 1992 Linus Torvalds | |
14 | * | |
ac27a0ec DK |
15 | * 64-bit file support on 64-bit platforms by Jakub Jelinek |
16 | * (jj@sunsite.ms.mff.cuni.cz) | |
17 | * | |
617ba13b | 18 | * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
ac27a0ec DK |
19 | */ |
20 | ||
21 | #include <linux/module.h> | |
22 | #include <linux/fs.h> | |
23 | #include <linux/time.h> | |
dab291af | 24 | #include <linux/jbd2.h> |
ac27a0ec DK |
25 | #include <linux/highuid.h> |
26 | #include <linux/pagemap.h> | |
27 | #include <linux/quotaops.h> | |
28 | #include <linux/string.h> | |
29 | #include <linux/buffer_head.h> | |
30 | #include <linux/writeback.h> | |
64769240 | 31 | #include <linux/pagevec.h> |
ac27a0ec | 32 | #include <linux/mpage.h> |
e83c1397 | 33 | #include <linux/namei.h> |
ac27a0ec DK |
34 | #include <linux/uio.h> |
35 | #include <linux/bio.h> | |
4c0425ff | 36 | #include <linux/workqueue.h> |
744692dc | 37 | #include <linux/kernel.h> |
6db26ffc | 38 | #include <linux/printk.h> |
5a0e3ad6 | 39 | #include <linux/slab.h> |
a8901d34 | 40 | #include <linux/ratelimit.h> |
9bffad1e | 41 | |
3dcf5451 | 42 | #include "ext4_jbd2.h" |
ac27a0ec DK |
43 | #include "xattr.h" |
44 | #include "acl.h" | |
9f125d64 | 45 | #include "truncate.h" |
ac27a0ec | 46 | |
9bffad1e TT |
47 | #include <trace/events/ext4.h> |
48 | ||
a1d6cc56 AK |
49 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
50 | ||
678aaf48 JK |
51 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
52 | loff_t new_size) | |
53 | { | |
7ff9c073 | 54 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
55 | /* |
56 | * If jinode is zero, then we never opened the file for | |
57 | * writing, so there's no need to call | |
58 | * jbd2_journal_begin_ordered_truncate() since there's no | |
59 | * outstanding writes we need to flush. | |
60 | */ | |
61 | if (!EXT4_I(inode)->jinode) | |
62 | return 0; | |
63 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
64 | EXT4_I(inode)->jinode, | |
65 | new_size); | |
678aaf48 JK |
66 | } |
67 | ||
64769240 | 68 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
cb20d518 TT |
69 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, |
70 | struct buffer_head *bh_result, int create); | |
71 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode); | |
72 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate); | |
73 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); | |
74 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
64769240 | 75 | |
ac27a0ec DK |
76 | /* |
77 | * Test whether an inode is a fast symlink. | |
78 | */ | |
617ba13b | 79 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 80 | { |
617ba13b | 81 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
82 | (inode->i_sb->s_blocksize >> 9) : 0; |
83 | ||
84 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
85 | } | |
86 | ||
ac27a0ec DK |
87 | /* |
88 | * Restart the transaction associated with *handle. This does a commit, | |
89 | * so before we call here everything must be consistently dirtied against | |
90 | * this transaction. | |
91 | */ | |
fa5d1113 | 92 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 93 | int nblocks) |
ac27a0ec | 94 | { |
487caeef JK |
95 | int ret; |
96 | ||
97 | /* | |
e35fd660 | 98 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
99 | * moment, get_block can be called only for blocks inside i_size since |
100 | * page cache has been already dropped and writes are blocked by | |
101 | * i_mutex. So we can safely drop the i_data_sem here. | |
102 | */ | |
0390131b | 103 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 104 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 105 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 106 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 107 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 108 | ext4_discard_preallocations(inode); |
487caeef JK |
109 | |
110 | return ret; | |
ac27a0ec DK |
111 | } |
112 | ||
113 | /* | |
114 | * Called at the last iput() if i_nlink is zero. | |
115 | */ | |
0930fcc1 | 116 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
117 | { |
118 | handle_t *handle; | |
bc965ab3 | 119 | int err; |
ac27a0ec | 120 | |
7ff9c073 | 121 | trace_ext4_evict_inode(inode); |
2581fdc8 | 122 | |
2581fdc8 JZ |
123 | ext4_ioend_wait(inode); |
124 | ||
0930fcc1 | 125 | if (inode->i_nlink) { |
2d859db3 JK |
126 | /* |
127 | * When journalling data dirty buffers are tracked only in the | |
128 | * journal. So although mm thinks everything is clean and | |
129 | * ready for reaping the inode might still have some pages to | |
130 | * write in the running transaction or waiting to be | |
131 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
132 | * (via truncate_inode_pages()) to discard these buffers can | |
133 | * cause data loss. Also even if we did not discard these | |
134 | * buffers, we would have no way to find them after the inode | |
135 | * is reaped and thus user could see stale data if he tries to | |
136 | * read them before the transaction is checkpointed. So be | |
137 | * careful and force everything to disk here... We use | |
138 | * ei->i_datasync_tid to store the newest transaction | |
139 | * containing inode's data. | |
140 | * | |
141 | * Note that directories do not have this problem because they | |
142 | * don't use page cache. | |
143 | */ | |
144 | if (ext4_should_journal_data(inode) && | |
145 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) { | |
146 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
147 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
148 | ||
149 | jbd2_log_start_commit(journal, commit_tid); | |
150 | jbd2_log_wait_commit(journal, commit_tid); | |
151 | filemap_write_and_wait(&inode->i_data); | |
152 | } | |
0930fcc1 AV |
153 | truncate_inode_pages(&inode->i_data, 0); |
154 | goto no_delete; | |
155 | } | |
156 | ||
907f4554 | 157 | if (!is_bad_inode(inode)) |
871a2931 | 158 | dquot_initialize(inode); |
907f4554 | 159 | |
678aaf48 JK |
160 | if (ext4_should_order_data(inode)) |
161 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec DK |
162 | truncate_inode_pages(&inode->i_data, 0); |
163 | ||
164 | if (is_bad_inode(inode)) | |
165 | goto no_delete; | |
166 | ||
9f125d64 | 167 | handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3); |
ac27a0ec | 168 | if (IS_ERR(handle)) { |
bc965ab3 | 169 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
170 | /* |
171 | * If we're going to skip the normal cleanup, we still need to | |
172 | * make sure that the in-core orphan linked list is properly | |
173 | * cleaned up. | |
174 | */ | |
617ba13b | 175 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
176 | goto no_delete; |
177 | } | |
178 | ||
179 | if (IS_SYNC(inode)) | |
0390131b | 180 | ext4_handle_sync(handle); |
ac27a0ec | 181 | inode->i_size = 0; |
bc965ab3 TT |
182 | err = ext4_mark_inode_dirty(handle, inode); |
183 | if (err) { | |
12062ddd | 184 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
185 | "couldn't mark inode dirty (err %d)", err); |
186 | goto stop_handle; | |
187 | } | |
ac27a0ec | 188 | if (inode->i_blocks) |
617ba13b | 189 | ext4_truncate(inode); |
bc965ab3 TT |
190 | |
191 | /* | |
192 | * ext4_ext_truncate() doesn't reserve any slop when it | |
193 | * restarts journal transactions; therefore there may not be | |
194 | * enough credits left in the handle to remove the inode from | |
195 | * the orphan list and set the dtime field. | |
196 | */ | |
0390131b | 197 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
198 | err = ext4_journal_extend(handle, 3); |
199 | if (err > 0) | |
200 | err = ext4_journal_restart(handle, 3); | |
201 | if (err != 0) { | |
12062ddd | 202 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
203 | "couldn't extend journal (err %d)", err); |
204 | stop_handle: | |
205 | ext4_journal_stop(handle); | |
45388219 | 206 | ext4_orphan_del(NULL, inode); |
bc965ab3 TT |
207 | goto no_delete; |
208 | } | |
209 | } | |
210 | ||
ac27a0ec | 211 | /* |
617ba13b | 212 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 213 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 214 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 215 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 216 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
217 | * (Well, we could do this if we need to, but heck - it works) |
218 | */ | |
617ba13b MC |
219 | ext4_orphan_del(handle, inode); |
220 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
221 | |
222 | /* | |
223 | * One subtle ordering requirement: if anything has gone wrong | |
224 | * (transaction abort, IO errors, whatever), then we can still | |
225 | * do these next steps (the fs will already have been marked as | |
226 | * having errors), but we can't free the inode if the mark_dirty | |
227 | * fails. | |
228 | */ | |
617ba13b | 229 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 230 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 231 | ext4_clear_inode(inode); |
ac27a0ec | 232 | else |
617ba13b MC |
233 | ext4_free_inode(handle, inode); |
234 | ext4_journal_stop(handle); | |
ac27a0ec DK |
235 | return; |
236 | no_delete: | |
0930fcc1 | 237 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
238 | } |
239 | ||
a9e7f447 DM |
240 | #ifdef CONFIG_QUOTA |
241 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 242 | { |
a9e7f447 | 243 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 244 | } |
a9e7f447 | 245 | #endif |
9d0be502 | 246 | |
12219aea AK |
247 | /* |
248 | * Calculate the number of metadata blocks need to reserve | |
9d0be502 | 249 | * to allocate a block located at @lblock |
12219aea | 250 | */ |
01f49d0b | 251 | static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
12219aea | 252 | { |
12e9b892 | 253 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
9d0be502 | 254 | return ext4_ext_calc_metadata_amount(inode, lblock); |
12219aea | 255 | |
8bb2b247 | 256 | return ext4_ind_calc_metadata_amount(inode, lblock); |
12219aea AK |
257 | } |
258 | ||
0637c6f4 TT |
259 | /* |
260 | * Called with i_data_sem down, which is important since we can call | |
261 | * ext4_discard_preallocations() from here. | |
262 | */ | |
5f634d06 AK |
263 | void ext4_da_update_reserve_space(struct inode *inode, |
264 | int used, int quota_claim) | |
12219aea AK |
265 | { |
266 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 267 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
268 | |
269 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 270 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 TT |
271 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
272 | ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d " | |
273 | "with only %d reserved data blocks\n", | |
274 | __func__, inode->i_ino, used, | |
275 | ei->i_reserved_data_blocks); | |
276 | WARN_ON(1); | |
277 | used = ei->i_reserved_data_blocks; | |
278 | } | |
12219aea | 279 | |
0637c6f4 TT |
280 | /* Update per-inode reservations */ |
281 | ei->i_reserved_data_blocks -= used; | |
0637c6f4 | 282 | ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; |
57042651 | 283 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 284 | used + ei->i_allocated_meta_blocks); |
0637c6f4 | 285 | ei->i_allocated_meta_blocks = 0; |
6bc6e63f | 286 | |
0637c6f4 TT |
287 | if (ei->i_reserved_data_blocks == 0) { |
288 | /* | |
289 | * We can release all of the reserved metadata blocks | |
290 | * only when we have written all of the delayed | |
291 | * allocation blocks. | |
292 | */ | |
57042651 | 293 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 294 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 295 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 296 | ei->i_da_metadata_calc_len = 0; |
6bc6e63f | 297 | } |
12219aea | 298 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 299 | |
72b8ab9d ES |
300 | /* Update quota subsystem for data blocks */ |
301 | if (quota_claim) | |
7b415bf6 | 302 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 303 | else { |
5f634d06 AK |
304 | /* |
305 | * We did fallocate with an offset that is already delayed | |
306 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 307 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 308 | */ |
7b415bf6 | 309 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 310 | } |
d6014301 AK |
311 | |
312 | /* | |
313 | * If we have done all the pending block allocations and if | |
314 | * there aren't any writers on the inode, we can discard the | |
315 | * inode's preallocations. | |
316 | */ | |
0637c6f4 TT |
317 | if ((ei->i_reserved_data_blocks == 0) && |
318 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 319 | ext4_discard_preallocations(inode); |
12219aea AK |
320 | } |
321 | ||
e29136f8 | 322 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
323 | unsigned int line, |
324 | struct ext4_map_blocks *map) | |
6fd058f7 | 325 | { |
24676da4 TT |
326 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
327 | map->m_len)) { | |
c398eda0 TT |
328 | ext4_error_inode(inode, func, line, map->m_pblk, |
329 | "lblock %lu mapped to illegal pblock " | |
330 | "(length %d)", (unsigned long) map->m_lblk, | |
331 | map->m_len); | |
6fd058f7 TT |
332 | return -EIO; |
333 | } | |
334 | return 0; | |
335 | } | |
336 | ||
e29136f8 | 337 | #define check_block_validity(inode, map) \ |
c398eda0 | 338 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 339 | |
55138e0b | 340 | /* |
1f94533d TT |
341 | * Return the number of contiguous dirty pages in a given inode |
342 | * starting at page frame idx. | |
55138e0b TT |
343 | */ |
344 | static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx, | |
345 | unsigned int max_pages) | |
346 | { | |
347 | struct address_space *mapping = inode->i_mapping; | |
348 | pgoff_t index; | |
349 | struct pagevec pvec; | |
350 | pgoff_t num = 0; | |
351 | int i, nr_pages, done = 0; | |
352 | ||
353 | if (max_pages == 0) | |
354 | return 0; | |
355 | pagevec_init(&pvec, 0); | |
356 | while (!done) { | |
357 | index = idx; | |
358 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
359 | PAGECACHE_TAG_DIRTY, | |
360 | (pgoff_t)PAGEVEC_SIZE); | |
361 | if (nr_pages == 0) | |
362 | break; | |
363 | for (i = 0; i < nr_pages; i++) { | |
364 | struct page *page = pvec.pages[i]; | |
365 | struct buffer_head *bh, *head; | |
366 | ||
367 | lock_page(page); | |
368 | if (unlikely(page->mapping != mapping) || | |
369 | !PageDirty(page) || | |
370 | PageWriteback(page) || | |
371 | page->index != idx) { | |
372 | done = 1; | |
373 | unlock_page(page); | |
374 | break; | |
375 | } | |
1f94533d TT |
376 | if (page_has_buffers(page)) { |
377 | bh = head = page_buffers(page); | |
378 | do { | |
379 | if (!buffer_delay(bh) && | |
380 | !buffer_unwritten(bh)) | |
381 | done = 1; | |
382 | bh = bh->b_this_page; | |
383 | } while (!done && (bh != head)); | |
384 | } | |
55138e0b TT |
385 | unlock_page(page); |
386 | if (done) | |
387 | break; | |
388 | idx++; | |
389 | num++; | |
659c6009 ES |
390 | if (num >= max_pages) { |
391 | done = 1; | |
55138e0b | 392 | break; |
659c6009 | 393 | } |
55138e0b TT |
394 | } |
395 | pagevec_release(&pvec); | |
396 | } | |
397 | return num; | |
398 | } | |
399 | ||
5356f261 AK |
400 | /* |
401 | * Sets the BH_Da_Mapped bit on the buffer heads corresponding to the given map. | |
402 | */ | |
403 | static void set_buffers_da_mapped(struct inode *inode, | |
404 | struct ext4_map_blocks *map) | |
405 | { | |
406 | struct address_space *mapping = inode->i_mapping; | |
407 | struct pagevec pvec; | |
408 | int i, nr_pages; | |
409 | pgoff_t index, end; | |
410 | ||
411 | index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
412 | end = (map->m_lblk + map->m_len - 1) >> | |
413 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
414 | ||
415 | pagevec_init(&pvec, 0); | |
416 | while (index <= end) { | |
417 | nr_pages = pagevec_lookup(&pvec, mapping, index, | |
418 | min(end - index + 1, | |
419 | (pgoff_t)PAGEVEC_SIZE)); | |
420 | if (nr_pages == 0) | |
421 | break; | |
422 | for (i = 0; i < nr_pages; i++) { | |
423 | struct page *page = pvec.pages[i]; | |
424 | struct buffer_head *bh, *head; | |
425 | ||
426 | if (unlikely(page->mapping != mapping) || | |
427 | !PageDirty(page)) | |
428 | break; | |
429 | ||
430 | if (page_has_buffers(page)) { | |
431 | bh = head = page_buffers(page); | |
432 | do { | |
433 | set_buffer_da_mapped(bh); | |
434 | bh = bh->b_this_page; | |
435 | } while (bh != head); | |
436 | } | |
437 | index++; | |
438 | } | |
439 | pagevec_release(&pvec); | |
440 | } | |
441 | } | |
442 | ||
f5ab0d1f | 443 | /* |
e35fd660 | 444 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 445 | * and returns if the blocks are already mapped. |
f5ab0d1f | 446 | * |
f5ab0d1f MC |
447 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
448 | * and store the allocated blocks in the result buffer head and mark it | |
449 | * mapped. | |
450 | * | |
e35fd660 TT |
451 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
452 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
453 | * based files |
454 | * | |
455 | * On success, it returns the number of blocks being mapped or allocate. | |
456 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
457 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
458 | * the buffer head is mapped. | |
459 | * | |
460 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
df3ab170 | 461 | * that case, buffer head is unmapped |
f5ab0d1f MC |
462 | * |
463 | * It returns the error in case of allocation failure. | |
464 | */ | |
e35fd660 TT |
465 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
466 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 AK |
467 | { |
468 | int retval; | |
f5ab0d1f | 469 | |
e35fd660 TT |
470 | map->m_flags = 0; |
471 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
472 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
473 | (unsigned long) map->m_lblk); | |
4df3d265 | 474 | /* |
b920c755 TT |
475 | * Try to see if we can get the block without requesting a new |
476 | * file system block. | |
4df3d265 AK |
477 | */ |
478 | down_read((&EXT4_I(inode)->i_data_sem)); | |
12e9b892 | 479 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 480 | retval = ext4_ext_map_blocks(handle, inode, map, 0); |
0e855ac8 | 481 | } else { |
e35fd660 | 482 | retval = ext4_ind_map_blocks(handle, inode, map, 0); |
0e855ac8 | 483 | } |
4df3d265 | 484 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 485 | |
e35fd660 | 486 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 487 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
488 | if (ret != 0) |
489 | return ret; | |
490 | } | |
491 | ||
f5ab0d1f | 492 | /* If it is only a block(s) look up */ |
c2177057 | 493 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
494 | return retval; |
495 | ||
496 | /* | |
497 | * Returns if the blocks have already allocated | |
498 | * | |
499 | * Note that if blocks have been preallocated | |
df3ab170 | 500 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
501 | * with buffer head unmapped. |
502 | */ | |
e35fd660 | 503 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
4df3d265 AK |
504 | return retval; |
505 | ||
2a8964d6 AK |
506 | /* |
507 | * When we call get_blocks without the create flag, the | |
508 | * BH_Unwritten flag could have gotten set if the blocks | |
509 | * requested were part of a uninitialized extent. We need to | |
510 | * clear this flag now that we are committed to convert all or | |
511 | * part of the uninitialized extent to be an initialized | |
512 | * extent. This is because we need to avoid the combination | |
513 | * of BH_Unwritten and BH_Mapped flags being simultaneously | |
514 | * set on the buffer_head. | |
515 | */ | |
e35fd660 | 516 | map->m_flags &= ~EXT4_MAP_UNWRITTEN; |
2a8964d6 | 517 | |
4df3d265 | 518 | /* |
f5ab0d1f MC |
519 | * New blocks allocate and/or writing to uninitialized extent |
520 | * will possibly result in updating i_data, so we take | |
521 | * the write lock of i_data_sem, and call get_blocks() | |
522 | * with create == 1 flag. | |
4df3d265 AK |
523 | */ |
524 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
525 | |
526 | /* | |
527 | * if the caller is from delayed allocation writeout path | |
528 | * we have already reserved fs blocks for allocation | |
529 | * let the underlying get_block() function know to | |
530 | * avoid double accounting | |
531 | */ | |
c2177057 | 532 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 533 | ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
4df3d265 AK |
534 | /* |
535 | * We need to check for EXT4 here because migrate | |
536 | * could have changed the inode type in between | |
537 | */ | |
12e9b892 | 538 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 539 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 540 | } else { |
e35fd660 | 541 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 542 | |
e35fd660 | 543 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
544 | /* |
545 | * We allocated new blocks which will result in | |
546 | * i_data's format changing. Force the migrate | |
547 | * to fail by clearing migrate flags | |
548 | */ | |
19f5fb7a | 549 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 550 | } |
d2a17637 | 551 | |
5f634d06 AK |
552 | /* |
553 | * Update reserved blocks/metadata blocks after successful | |
554 | * block allocation which had been deferred till now. We don't | |
555 | * support fallocate for non extent files. So we can update | |
556 | * reserve space here. | |
557 | */ | |
558 | if ((retval > 0) && | |
1296cc85 | 559 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
560 | ext4_da_update_reserve_space(inode, retval, 1); |
561 | } | |
5356f261 | 562 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
f2321097 | 563 | ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
2ac3b6e0 | 564 | |
5356f261 AK |
565 | /* If we have successfully mapped the delayed allocated blocks, |
566 | * set the BH_Da_Mapped bit on them. Its important to do this | |
567 | * under the protection of i_data_sem. | |
568 | */ | |
569 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) | |
570 | set_buffers_da_mapped(inode, map); | |
571 | } | |
572 | ||
4df3d265 | 573 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 574 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 575 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
576 | if (ret != 0) |
577 | return ret; | |
578 | } | |
0e855ac8 AK |
579 | return retval; |
580 | } | |
581 | ||
f3bd1f3f MC |
582 | /* Maximum number of blocks we map for direct IO at once. */ |
583 | #define DIO_MAX_BLOCKS 4096 | |
584 | ||
2ed88685 TT |
585 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
586 | struct buffer_head *bh, int flags) | |
ac27a0ec | 587 | { |
3e4fdaf8 | 588 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 589 | struct ext4_map_blocks map; |
7fb5409d | 590 | int ret = 0, started = 0; |
f3bd1f3f | 591 | int dio_credits; |
ac27a0ec | 592 | |
2ed88685 TT |
593 | map.m_lblk = iblock; |
594 | map.m_len = bh->b_size >> inode->i_blkbits; | |
595 | ||
596 | if (flags && !handle) { | |
7fb5409d | 597 | /* Direct IO write... */ |
2ed88685 TT |
598 | if (map.m_len > DIO_MAX_BLOCKS) |
599 | map.m_len = DIO_MAX_BLOCKS; | |
600 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
f3bd1f3f | 601 | handle = ext4_journal_start(inode, dio_credits); |
7fb5409d | 602 | if (IS_ERR(handle)) { |
ac27a0ec | 603 | ret = PTR_ERR(handle); |
2ed88685 | 604 | return ret; |
ac27a0ec | 605 | } |
7fb5409d | 606 | started = 1; |
ac27a0ec DK |
607 | } |
608 | ||
2ed88685 | 609 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 610 | if (ret > 0) { |
2ed88685 TT |
611 | map_bh(bh, inode->i_sb, map.m_pblk); |
612 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
613 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
7fb5409d | 614 | ret = 0; |
ac27a0ec | 615 | } |
7fb5409d JK |
616 | if (started) |
617 | ext4_journal_stop(handle); | |
ac27a0ec DK |
618 | return ret; |
619 | } | |
620 | ||
2ed88685 TT |
621 | int ext4_get_block(struct inode *inode, sector_t iblock, |
622 | struct buffer_head *bh, int create) | |
623 | { | |
624 | return _ext4_get_block(inode, iblock, bh, | |
625 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
626 | } | |
627 | ||
ac27a0ec DK |
628 | /* |
629 | * `handle' can be NULL if create is zero | |
630 | */ | |
617ba13b | 631 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 632 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec | 633 | { |
2ed88685 TT |
634 | struct ext4_map_blocks map; |
635 | struct buffer_head *bh; | |
ac27a0ec DK |
636 | int fatal = 0, err; |
637 | ||
638 | J_ASSERT(handle != NULL || create == 0); | |
639 | ||
2ed88685 TT |
640 | map.m_lblk = block; |
641 | map.m_len = 1; | |
642 | err = ext4_map_blocks(handle, inode, &map, | |
643 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
ac27a0ec | 644 | |
2ed88685 TT |
645 | if (err < 0) |
646 | *errp = err; | |
647 | if (err <= 0) | |
648 | return NULL; | |
649 | *errp = 0; | |
650 | ||
651 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
652 | if (!bh) { | |
653 | *errp = -EIO; | |
654 | return NULL; | |
ac27a0ec | 655 | } |
2ed88685 TT |
656 | if (map.m_flags & EXT4_MAP_NEW) { |
657 | J_ASSERT(create != 0); | |
658 | J_ASSERT(handle != NULL); | |
ac27a0ec | 659 | |
2ed88685 TT |
660 | /* |
661 | * Now that we do not always journal data, we should | |
662 | * keep in mind whether this should always journal the | |
663 | * new buffer as metadata. For now, regular file | |
664 | * writes use ext4_get_block instead, so it's not a | |
665 | * problem. | |
666 | */ | |
667 | lock_buffer(bh); | |
668 | BUFFER_TRACE(bh, "call get_create_access"); | |
669 | fatal = ext4_journal_get_create_access(handle, bh); | |
670 | if (!fatal && !buffer_uptodate(bh)) { | |
671 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); | |
672 | set_buffer_uptodate(bh); | |
ac27a0ec | 673 | } |
2ed88685 TT |
674 | unlock_buffer(bh); |
675 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
676 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
677 | if (!fatal) | |
678 | fatal = err; | |
679 | } else { | |
680 | BUFFER_TRACE(bh, "not a new buffer"); | |
ac27a0ec | 681 | } |
2ed88685 TT |
682 | if (fatal) { |
683 | *errp = fatal; | |
684 | brelse(bh); | |
685 | bh = NULL; | |
686 | } | |
687 | return bh; | |
ac27a0ec DK |
688 | } |
689 | ||
617ba13b | 690 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 691 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 692 | { |
af5bc92d | 693 | struct buffer_head *bh; |
ac27a0ec | 694 | |
617ba13b | 695 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
696 | if (!bh) |
697 | return bh; | |
698 | if (buffer_uptodate(bh)) | |
699 | return bh; | |
700 | ll_rw_block(READ_META, 1, &bh); | |
701 | wait_on_buffer(bh); | |
702 | if (buffer_uptodate(bh)) | |
703 | return bh; | |
704 | put_bh(bh); | |
705 | *err = -EIO; | |
706 | return NULL; | |
707 | } | |
708 | ||
af5bc92d TT |
709 | static int walk_page_buffers(handle_t *handle, |
710 | struct buffer_head *head, | |
711 | unsigned from, | |
712 | unsigned to, | |
713 | int *partial, | |
714 | int (*fn)(handle_t *handle, | |
715 | struct buffer_head *bh)) | |
ac27a0ec DK |
716 | { |
717 | struct buffer_head *bh; | |
718 | unsigned block_start, block_end; | |
719 | unsigned blocksize = head->b_size; | |
720 | int err, ret = 0; | |
721 | struct buffer_head *next; | |
722 | ||
af5bc92d TT |
723 | for (bh = head, block_start = 0; |
724 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 725 | block_start = block_end, bh = next) { |
ac27a0ec DK |
726 | next = bh->b_this_page; |
727 | block_end = block_start + blocksize; | |
728 | if (block_end <= from || block_start >= to) { | |
729 | if (partial && !buffer_uptodate(bh)) | |
730 | *partial = 1; | |
731 | continue; | |
732 | } | |
733 | err = (*fn)(handle, bh); | |
734 | if (!ret) | |
735 | ret = err; | |
736 | } | |
737 | return ret; | |
738 | } | |
739 | ||
740 | /* | |
741 | * To preserve ordering, it is essential that the hole instantiation and | |
742 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 743 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 744 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
745 | * prepare_write() is the right place. |
746 | * | |
617ba13b MC |
747 | * Also, this function can nest inside ext4_writepage() -> |
748 | * block_write_full_page(). In that case, we *know* that ext4_writepage() | |
ac27a0ec DK |
749 | * has generated enough buffer credits to do the whole page. So we won't |
750 | * block on the journal in that case, which is good, because the caller may | |
751 | * be PF_MEMALLOC. | |
752 | * | |
617ba13b | 753 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
754 | * quota file writes. If we were to commit the transaction while thus |
755 | * reentered, there can be a deadlock - we would be holding a quota | |
756 | * lock, and the commit would never complete if another thread had a | |
757 | * transaction open and was blocking on the quota lock - a ranking | |
758 | * violation. | |
759 | * | |
dab291af | 760 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
761 | * will _not_ run commit under these circumstances because handle->h_ref |
762 | * is elevated. We'll still have enough credits for the tiny quotafile | |
763 | * write. | |
764 | */ | |
765 | static int do_journal_get_write_access(handle_t *handle, | |
de9a55b8 | 766 | struct buffer_head *bh) |
ac27a0ec | 767 | { |
56d35a4c JK |
768 | int dirty = buffer_dirty(bh); |
769 | int ret; | |
770 | ||
ac27a0ec DK |
771 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
772 | return 0; | |
56d35a4c | 773 | /* |
ebdec241 | 774 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
775 | * the dirty bit as jbd2_journal_get_write_access() could complain |
776 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 777 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
778 | * the bit before releasing a page lock and thus writeback cannot |
779 | * ever write the buffer. | |
780 | */ | |
781 | if (dirty) | |
782 | clear_buffer_dirty(bh); | |
783 | ret = ext4_journal_get_write_access(handle, bh); | |
784 | if (!ret && dirty) | |
785 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
786 | return ret; | |
ac27a0ec DK |
787 | } |
788 | ||
744692dc JZ |
789 | static int ext4_get_block_write(struct inode *inode, sector_t iblock, |
790 | struct buffer_head *bh_result, int create); | |
bfc1af65 | 791 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
792 | loff_t pos, unsigned len, unsigned flags, |
793 | struct page **pagep, void **fsdata) | |
ac27a0ec | 794 | { |
af5bc92d | 795 | struct inode *inode = mapping->host; |
1938a150 | 796 | int ret, needed_blocks; |
ac27a0ec DK |
797 | handle_t *handle; |
798 | int retries = 0; | |
af5bc92d | 799 | struct page *page; |
de9a55b8 | 800 | pgoff_t index; |
af5bc92d | 801 | unsigned from, to; |
bfc1af65 | 802 | |
9bffad1e | 803 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
804 | /* |
805 | * Reserve one block more for addition to orphan list in case | |
806 | * we allocate blocks but write fails for some reason | |
807 | */ | |
808 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 809 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
810 | from = pos & (PAGE_CACHE_SIZE - 1); |
811 | to = from + len; | |
ac27a0ec DK |
812 | |
813 | retry: | |
af5bc92d TT |
814 | handle = ext4_journal_start(inode, needed_blocks); |
815 | if (IS_ERR(handle)) { | |
816 | ret = PTR_ERR(handle); | |
817 | goto out; | |
7479d2b9 | 818 | } |
ac27a0ec | 819 | |
ebd3610b JK |
820 | /* We cannot recurse into the filesystem as the transaction is already |
821 | * started */ | |
822 | flags |= AOP_FLAG_NOFS; | |
823 | ||
54566b2c | 824 | page = grab_cache_page_write_begin(mapping, index, flags); |
cf108bca JK |
825 | if (!page) { |
826 | ext4_journal_stop(handle); | |
827 | ret = -ENOMEM; | |
828 | goto out; | |
829 | } | |
830 | *pagep = page; | |
831 | ||
744692dc | 832 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 833 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 834 | else |
6e1db88d | 835 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
bfc1af65 NP |
836 | |
837 | if (!ret && ext4_should_journal_data(inode)) { | |
ac27a0ec DK |
838 | ret = walk_page_buffers(handle, page_buffers(page), |
839 | from, to, NULL, do_journal_get_write_access); | |
840 | } | |
bfc1af65 NP |
841 | |
842 | if (ret) { | |
af5bc92d | 843 | unlock_page(page); |
af5bc92d | 844 | page_cache_release(page); |
ae4d5372 | 845 | /* |
6e1db88d | 846 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
847 | * outside i_size. Trim these off again. Don't need |
848 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
849 | * |
850 | * Add inode to orphan list in case we crash before | |
851 | * truncate finishes | |
ae4d5372 | 852 | */ |
ffacfa7a | 853 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
854 | ext4_orphan_add(handle, inode); |
855 | ||
856 | ext4_journal_stop(handle); | |
857 | if (pos + len > inode->i_size) { | |
b9a4207d | 858 | ext4_truncate_failed_write(inode); |
de9a55b8 | 859 | /* |
ffacfa7a | 860 | * If truncate failed early the inode might |
1938a150 AK |
861 | * still be on the orphan list; we need to |
862 | * make sure the inode is removed from the | |
863 | * orphan list in that case. | |
864 | */ | |
865 | if (inode->i_nlink) | |
866 | ext4_orphan_del(NULL, inode); | |
867 | } | |
bfc1af65 NP |
868 | } |
869 | ||
617ba13b | 870 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
ac27a0ec | 871 | goto retry; |
7479d2b9 | 872 | out: |
ac27a0ec DK |
873 | return ret; |
874 | } | |
875 | ||
bfc1af65 NP |
876 | /* For write_end() in data=journal mode */ |
877 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec DK |
878 | { |
879 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
880 | return 0; | |
881 | set_buffer_uptodate(bh); | |
0390131b | 882 | return ext4_handle_dirty_metadata(handle, NULL, bh); |
ac27a0ec DK |
883 | } |
884 | ||
f8514083 | 885 | static int ext4_generic_write_end(struct file *file, |
de9a55b8 TT |
886 | struct address_space *mapping, |
887 | loff_t pos, unsigned len, unsigned copied, | |
888 | struct page *page, void *fsdata) | |
f8514083 AK |
889 | { |
890 | int i_size_changed = 0; | |
891 | struct inode *inode = mapping->host; | |
892 | handle_t *handle = ext4_journal_current_handle(); | |
893 | ||
894 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
895 | ||
896 | /* | |
897 | * No need to use i_size_read() here, the i_size | |
898 | * cannot change under us because we hold i_mutex. | |
899 | * | |
900 | * But it's important to update i_size while still holding page lock: | |
901 | * page writeout could otherwise come in and zero beyond i_size. | |
902 | */ | |
903 | if (pos + copied > inode->i_size) { | |
904 | i_size_write(inode, pos + copied); | |
905 | i_size_changed = 1; | |
906 | } | |
907 | ||
908 | if (pos + copied > EXT4_I(inode)->i_disksize) { | |
909 | /* We need to mark inode dirty even if | |
910 | * new_i_size is less that inode->i_size | |
911 | * bu greater than i_disksize.(hint delalloc) | |
912 | */ | |
913 | ext4_update_i_disksize(inode, (pos + copied)); | |
914 | i_size_changed = 1; | |
915 | } | |
916 | unlock_page(page); | |
917 | page_cache_release(page); | |
918 | ||
919 | /* | |
920 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
921 | * makes the holding time of page lock longer. Second, it forces lock | |
922 | * ordering of page lock and transaction start for journaling | |
923 | * filesystems. | |
924 | */ | |
925 | if (i_size_changed) | |
926 | ext4_mark_inode_dirty(handle, inode); | |
927 | ||
928 | return copied; | |
929 | } | |
930 | ||
ac27a0ec DK |
931 | /* |
932 | * We need to pick up the new inode size which generic_commit_write gave us | |
933 | * `file' can be NULL - eg, when called from page_symlink(). | |
934 | * | |
617ba13b | 935 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
ac27a0ec DK |
936 | * buffers are managed internally. |
937 | */ | |
bfc1af65 | 938 | static int ext4_ordered_write_end(struct file *file, |
de9a55b8 TT |
939 | struct address_space *mapping, |
940 | loff_t pos, unsigned len, unsigned copied, | |
941 | struct page *page, void *fsdata) | |
ac27a0ec | 942 | { |
617ba13b | 943 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 944 | struct inode *inode = mapping->host; |
ac27a0ec DK |
945 | int ret = 0, ret2; |
946 | ||
9bffad1e | 947 | trace_ext4_ordered_write_end(inode, pos, len, copied); |
678aaf48 | 948 | ret = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
949 | |
950 | if (ret == 0) { | |
f8514083 | 951 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 952 | page, fsdata); |
f8a87d89 | 953 | copied = ret2; |
ffacfa7a | 954 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
955 | /* if we have allocated more blocks and copied |
956 | * less. We will have blocks allocated outside | |
957 | * inode->i_size. So truncate them | |
958 | */ | |
959 | ext4_orphan_add(handle, inode); | |
f8a87d89 RK |
960 | if (ret2 < 0) |
961 | ret = ret2; | |
ac27a0ec | 962 | } |
617ba13b | 963 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
964 | if (!ret) |
965 | ret = ret2; | |
bfc1af65 | 966 | |
f8514083 | 967 | if (pos + len > inode->i_size) { |
b9a4207d | 968 | ext4_truncate_failed_write(inode); |
de9a55b8 | 969 | /* |
ffacfa7a | 970 | * If truncate failed early the inode might still be |
f8514083 AK |
971 | * on the orphan list; we need to make sure the inode |
972 | * is removed from the orphan list in that case. | |
973 | */ | |
974 | if (inode->i_nlink) | |
975 | ext4_orphan_del(NULL, inode); | |
976 | } | |
977 | ||
978 | ||
bfc1af65 | 979 | return ret ? ret : copied; |
ac27a0ec DK |
980 | } |
981 | ||
bfc1af65 | 982 | static int ext4_writeback_write_end(struct file *file, |
de9a55b8 TT |
983 | struct address_space *mapping, |
984 | loff_t pos, unsigned len, unsigned copied, | |
985 | struct page *page, void *fsdata) | |
ac27a0ec | 986 | { |
617ba13b | 987 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 988 | struct inode *inode = mapping->host; |
ac27a0ec | 989 | int ret = 0, ret2; |
ac27a0ec | 990 | |
9bffad1e | 991 | trace_ext4_writeback_write_end(inode, pos, len, copied); |
f8514083 | 992 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 993 | page, fsdata); |
f8a87d89 | 994 | copied = ret2; |
ffacfa7a | 995 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
996 | /* if we have allocated more blocks and copied |
997 | * less. We will have blocks allocated outside | |
998 | * inode->i_size. So truncate them | |
999 | */ | |
1000 | ext4_orphan_add(handle, inode); | |
1001 | ||
f8a87d89 RK |
1002 | if (ret2 < 0) |
1003 | ret = ret2; | |
ac27a0ec | 1004 | |
617ba13b | 1005 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1006 | if (!ret) |
1007 | ret = ret2; | |
bfc1af65 | 1008 | |
f8514083 | 1009 | if (pos + len > inode->i_size) { |
b9a4207d | 1010 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1011 | /* |
ffacfa7a | 1012 | * If truncate failed early the inode might still be |
f8514083 AK |
1013 | * on the orphan list; we need to make sure the inode |
1014 | * is removed from the orphan list in that case. | |
1015 | */ | |
1016 | if (inode->i_nlink) | |
1017 | ext4_orphan_del(NULL, inode); | |
1018 | } | |
1019 | ||
bfc1af65 | 1020 | return ret ? ret : copied; |
ac27a0ec DK |
1021 | } |
1022 | ||
bfc1af65 | 1023 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1024 | struct address_space *mapping, |
1025 | loff_t pos, unsigned len, unsigned copied, | |
1026 | struct page *page, void *fsdata) | |
ac27a0ec | 1027 | { |
617ba13b | 1028 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1029 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1030 | int ret = 0, ret2; |
1031 | int partial = 0; | |
bfc1af65 | 1032 | unsigned from, to; |
cf17fea6 | 1033 | loff_t new_i_size; |
ac27a0ec | 1034 | |
9bffad1e | 1035 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1036 | from = pos & (PAGE_CACHE_SIZE - 1); |
1037 | to = from + len; | |
1038 | ||
441c8508 CW |
1039 | BUG_ON(!ext4_handle_valid(handle)); |
1040 | ||
bfc1af65 NP |
1041 | if (copied < len) { |
1042 | if (!PageUptodate(page)) | |
1043 | copied = 0; | |
1044 | page_zero_new_buffers(page, from+copied, to); | |
1045 | } | |
ac27a0ec DK |
1046 | |
1047 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
bfc1af65 | 1048 | to, &partial, write_end_fn); |
ac27a0ec DK |
1049 | if (!partial) |
1050 | SetPageUptodate(page); | |
cf17fea6 AK |
1051 | new_i_size = pos + copied; |
1052 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1053 | i_size_write(inode, pos+copied); |
19f5fb7a | 1054 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1055 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
cf17fea6 AK |
1056 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1057 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1058 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1059 | if (!ret) |
1060 | ret = ret2; | |
1061 | } | |
bfc1af65 | 1062 | |
cf108bca | 1063 | unlock_page(page); |
f8514083 | 1064 | page_cache_release(page); |
ffacfa7a | 1065 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1066 | /* if we have allocated more blocks and copied |
1067 | * less. We will have blocks allocated outside | |
1068 | * inode->i_size. So truncate them | |
1069 | */ | |
1070 | ext4_orphan_add(handle, inode); | |
1071 | ||
617ba13b | 1072 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1073 | if (!ret) |
1074 | ret = ret2; | |
f8514083 | 1075 | if (pos + len > inode->i_size) { |
b9a4207d | 1076 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1077 | /* |
ffacfa7a | 1078 | * If truncate failed early the inode might still be |
f8514083 AK |
1079 | * on the orphan list; we need to make sure the inode |
1080 | * is removed from the orphan list in that case. | |
1081 | */ | |
1082 | if (inode->i_nlink) | |
1083 | ext4_orphan_del(NULL, inode); | |
1084 | } | |
bfc1af65 NP |
1085 | |
1086 | return ret ? ret : copied; | |
ac27a0ec | 1087 | } |
d2a17637 | 1088 | |
9d0be502 | 1089 | /* |
7b415bf6 | 1090 | * Reserve a single cluster located at lblock |
9d0be502 | 1091 | */ |
5356f261 | 1092 | static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 1093 | { |
030ba6bc | 1094 | int retries = 0; |
60e58e0f | 1095 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1096 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 | 1097 | unsigned int md_needed; |
5dd4056d | 1098 | int ret; |
d2a17637 MC |
1099 | |
1100 | /* | |
1101 | * recalculate the amount of metadata blocks to reserve | |
1102 | * in order to allocate nrblocks | |
1103 | * worse case is one extent per block | |
1104 | */ | |
030ba6bc | 1105 | repeat: |
0637c6f4 | 1106 | spin_lock(&ei->i_block_reservation_lock); |
7b415bf6 AK |
1107 | md_needed = EXT4_NUM_B2C(sbi, |
1108 | ext4_calc_metadata_amount(inode, lblock)); | |
f8ec9d68 | 1109 | trace_ext4_da_reserve_space(inode, md_needed); |
0637c6f4 | 1110 | spin_unlock(&ei->i_block_reservation_lock); |
d2a17637 | 1111 | |
60e58e0f | 1112 | /* |
72b8ab9d ES |
1113 | * We will charge metadata quota at writeout time; this saves |
1114 | * us from metadata over-estimation, though we may go over by | |
1115 | * a small amount in the end. Here we just reserve for data. | |
60e58e0f | 1116 | */ |
7b415bf6 | 1117 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); |
5dd4056d CH |
1118 | if (ret) |
1119 | return ret; | |
72b8ab9d ES |
1120 | /* |
1121 | * We do still charge estimated metadata to the sb though; | |
1122 | * we cannot afford to run out of free blocks. | |
1123 | */ | |
e7d5f315 | 1124 | if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) { |
7b415bf6 | 1125 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
030ba6bc AK |
1126 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
1127 | yield(); | |
1128 | goto repeat; | |
1129 | } | |
d2a17637 MC |
1130 | return -ENOSPC; |
1131 | } | |
0637c6f4 | 1132 | spin_lock(&ei->i_block_reservation_lock); |
9d0be502 | 1133 | ei->i_reserved_data_blocks++; |
0637c6f4 TT |
1134 | ei->i_reserved_meta_blocks += md_needed; |
1135 | spin_unlock(&ei->i_block_reservation_lock); | |
39bc680a | 1136 | |
d2a17637 MC |
1137 | return 0; /* success */ |
1138 | } | |
1139 | ||
12219aea | 1140 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1141 | { |
1142 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1143 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1144 | |
cd213226 MC |
1145 | if (!to_free) |
1146 | return; /* Nothing to release, exit */ | |
1147 | ||
d2a17637 | 1148 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1149 | |
5a58ec87 | 1150 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1151 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1152 | /* |
0637c6f4 TT |
1153 | * if there aren't enough reserved blocks, then the |
1154 | * counter is messed up somewhere. Since this | |
1155 | * function is called from invalidate page, it's | |
1156 | * harmless to return without any action. | |
cd213226 | 1157 | */ |
0637c6f4 TT |
1158 | ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: " |
1159 | "ino %lu, to_free %d with only %d reserved " | |
1160 | "data blocks\n", inode->i_ino, to_free, | |
1161 | ei->i_reserved_data_blocks); | |
1162 | WARN_ON(1); | |
1163 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1164 | } |
0637c6f4 | 1165 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1166 | |
0637c6f4 TT |
1167 | if (ei->i_reserved_data_blocks == 0) { |
1168 | /* | |
1169 | * We can release all of the reserved metadata blocks | |
1170 | * only when we have written all of the delayed | |
1171 | * allocation blocks. | |
7b415bf6 AK |
1172 | * Note that in case of bigalloc, i_reserved_meta_blocks, |
1173 | * i_reserved_data_blocks, etc. refer to number of clusters. | |
0637c6f4 | 1174 | */ |
57042651 | 1175 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 1176 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 1177 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1178 | ei->i_da_metadata_calc_len = 0; |
0637c6f4 | 1179 | } |
d2a17637 | 1180 | |
72b8ab9d | 1181 | /* update fs dirty data blocks counter */ |
57042651 | 1182 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1183 | |
d2a17637 | 1184 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1185 | |
7b415bf6 | 1186 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1187 | } |
1188 | ||
1189 | static void ext4_da_page_release_reservation(struct page *page, | |
de9a55b8 | 1190 | unsigned long offset) |
d2a17637 MC |
1191 | { |
1192 | int to_release = 0; | |
1193 | struct buffer_head *head, *bh; | |
1194 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1195 | struct inode *inode = page->mapping->host; |
1196 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1197 | int num_clusters; | |
d2a17637 MC |
1198 | |
1199 | head = page_buffers(page); | |
1200 | bh = head; | |
1201 | do { | |
1202 | unsigned int next_off = curr_off + bh->b_size; | |
1203 | ||
1204 | if ((offset <= curr_off) && (buffer_delay(bh))) { | |
1205 | to_release++; | |
1206 | clear_buffer_delay(bh); | |
5356f261 | 1207 | clear_buffer_da_mapped(bh); |
d2a17637 MC |
1208 | } |
1209 | curr_off = next_off; | |
1210 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 AK |
1211 | |
1212 | /* If we have released all the blocks belonging to a cluster, then we | |
1213 | * need to release the reserved space for that cluster. */ | |
1214 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1215 | while (num_clusters > 0) { | |
1216 | ext4_fsblk_t lblk; | |
1217 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + | |
1218 | ((num_clusters - 1) << sbi->s_cluster_bits); | |
1219 | if (sbi->s_cluster_ratio == 1 || | |
1220 | !ext4_find_delalloc_cluster(inode, lblk, 1)) | |
1221 | ext4_da_release_space(inode, 1); | |
1222 | ||
1223 | num_clusters--; | |
1224 | } | |
d2a17637 | 1225 | } |
ac27a0ec | 1226 | |
64769240 AT |
1227 | /* |
1228 | * Delayed allocation stuff | |
1229 | */ | |
1230 | ||
64769240 AT |
1231 | /* |
1232 | * mpage_da_submit_io - walks through extent of pages and try to write | |
a1d6cc56 | 1233 | * them with writepage() call back |
64769240 AT |
1234 | * |
1235 | * @mpd->inode: inode | |
1236 | * @mpd->first_page: first page of the extent | |
1237 | * @mpd->next_page: page after the last page of the extent | |
64769240 AT |
1238 | * |
1239 | * By the time mpage_da_submit_io() is called we expect all blocks | |
1240 | * to be allocated. this may be wrong if allocation failed. | |
1241 | * | |
1242 | * As pages are already locked by write_cache_pages(), we can't use it | |
1243 | */ | |
1de3e3df TT |
1244 | static int mpage_da_submit_io(struct mpage_da_data *mpd, |
1245 | struct ext4_map_blocks *map) | |
64769240 | 1246 | { |
791b7f08 AK |
1247 | struct pagevec pvec; |
1248 | unsigned long index, end; | |
1249 | int ret = 0, err, nr_pages, i; | |
1250 | struct inode *inode = mpd->inode; | |
1251 | struct address_space *mapping = inode->i_mapping; | |
cb20d518 | 1252 | loff_t size = i_size_read(inode); |
3ecdb3a1 TT |
1253 | unsigned int len, block_start; |
1254 | struct buffer_head *bh, *page_bufs = NULL; | |
cb20d518 | 1255 | int journal_data = ext4_should_journal_data(inode); |
1de3e3df | 1256 | sector_t pblock = 0, cur_logical = 0; |
bd2d0210 | 1257 | struct ext4_io_submit io_submit; |
64769240 AT |
1258 | |
1259 | BUG_ON(mpd->next_page <= mpd->first_page); | |
bd2d0210 | 1260 | memset(&io_submit, 0, sizeof(io_submit)); |
791b7f08 AK |
1261 | /* |
1262 | * We need to start from the first_page to the next_page - 1 | |
1263 | * to make sure we also write the mapped dirty buffer_heads. | |
8dc207c0 | 1264 | * If we look at mpd->b_blocknr we would only be looking |
791b7f08 AK |
1265 | * at the currently mapped buffer_heads. |
1266 | */ | |
64769240 AT |
1267 | index = mpd->first_page; |
1268 | end = mpd->next_page - 1; | |
1269 | ||
791b7f08 | 1270 | pagevec_init(&pvec, 0); |
64769240 | 1271 | while (index <= end) { |
791b7f08 | 1272 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
64769240 AT |
1273 | if (nr_pages == 0) |
1274 | break; | |
1275 | for (i = 0; i < nr_pages; i++) { | |
97498956 | 1276 | int commit_write = 0, skip_page = 0; |
64769240 AT |
1277 | struct page *page = pvec.pages[i]; |
1278 | ||
791b7f08 AK |
1279 | index = page->index; |
1280 | if (index > end) | |
1281 | break; | |
cb20d518 TT |
1282 | |
1283 | if (index == size >> PAGE_CACHE_SHIFT) | |
1284 | len = size & ~PAGE_CACHE_MASK; | |
1285 | else | |
1286 | len = PAGE_CACHE_SIZE; | |
1de3e3df TT |
1287 | if (map) { |
1288 | cur_logical = index << (PAGE_CACHE_SHIFT - | |
1289 | inode->i_blkbits); | |
1290 | pblock = map->m_pblk + (cur_logical - | |
1291 | map->m_lblk); | |
1292 | } | |
791b7f08 AK |
1293 | index++; |
1294 | ||
1295 | BUG_ON(!PageLocked(page)); | |
1296 | BUG_ON(PageWriteback(page)); | |
1297 | ||
64769240 | 1298 | /* |
cb20d518 TT |
1299 | * If the page does not have buffers (for |
1300 | * whatever reason), try to create them using | |
a107e5a3 | 1301 | * __block_write_begin. If this fails, |
97498956 | 1302 | * skip the page and move on. |
64769240 | 1303 | */ |
cb20d518 | 1304 | if (!page_has_buffers(page)) { |
a107e5a3 | 1305 | if (__block_write_begin(page, 0, len, |
cb20d518 | 1306 | noalloc_get_block_write)) { |
97498956 | 1307 | skip_page: |
cb20d518 TT |
1308 | unlock_page(page); |
1309 | continue; | |
1310 | } | |
1311 | commit_write = 1; | |
1312 | } | |
64769240 | 1313 | |
3ecdb3a1 TT |
1314 | bh = page_bufs = page_buffers(page); |
1315 | block_start = 0; | |
64769240 | 1316 | do { |
1de3e3df | 1317 | if (!bh) |
97498956 | 1318 | goto skip_page; |
1de3e3df TT |
1319 | if (map && (cur_logical >= map->m_lblk) && |
1320 | (cur_logical <= (map->m_lblk + | |
1321 | (map->m_len - 1)))) { | |
29fa89d0 AK |
1322 | if (buffer_delay(bh)) { |
1323 | clear_buffer_delay(bh); | |
1324 | bh->b_blocknr = pblock; | |
29fa89d0 | 1325 | } |
5356f261 AK |
1326 | if (buffer_da_mapped(bh)) |
1327 | clear_buffer_da_mapped(bh); | |
1de3e3df TT |
1328 | if (buffer_unwritten(bh) || |
1329 | buffer_mapped(bh)) | |
1330 | BUG_ON(bh->b_blocknr != pblock); | |
1331 | if (map->m_flags & EXT4_MAP_UNINIT) | |
1332 | set_buffer_uninit(bh); | |
1333 | clear_buffer_unwritten(bh); | |
1334 | } | |
29fa89d0 | 1335 | |
97498956 | 1336 | /* skip page if block allocation undone */ |
1de3e3df | 1337 | if (buffer_delay(bh) || buffer_unwritten(bh)) |
97498956 | 1338 | skip_page = 1; |
3ecdb3a1 TT |
1339 | bh = bh->b_this_page; |
1340 | block_start += bh->b_size; | |
64769240 AT |
1341 | cur_logical++; |
1342 | pblock++; | |
1de3e3df TT |
1343 | } while (bh != page_bufs); |
1344 | ||
97498956 TT |
1345 | if (skip_page) |
1346 | goto skip_page; | |
cb20d518 TT |
1347 | |
1348 | if (commit_write) | |
1349 | /* mark the buffer_heads as dirty & uptodate */ | |
1350 | block_commit_write(page, 0, len); | |
1351 | ||
97498956 | 1352 | clear_page_dirty_for_io(page); |
bd2d0210 TT |
1353 | /* |
1354 | * Delalloc doesn't support data journalling, | |
1355 | * but eventually maybe we'll lift this | |
1356 | * restriction. | |
1357 | */ | |
1358 | if (unlikely(journal_data && PageChecked(page))) | |
cb20d518 | 1359 | err = __ext4_journalled_writepage(page, len); |
1449032b | 1360 | else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT)) |
bd2d0210 TT |
1361 | err = ext4_bio_write_page(&io_submit, page, |
1362 | len, mpd->wbc); | |
9dd75f1f TT |
1363 | else if (buffer_uninit(page_bufs)) { |
1364 | ext4_set_bh_endio(page_bufs, inode); | |
1365 | err = block_write_full_page_endio(page, | |
1366 | noalloc_get_block_write, | |
1367 | mpd->wbc, ext4_end_io_buffer_write); | |
1368 | } else | |
1449032b TT |
1369 | err = block_write_full_page(page, |
1370 | noalloc_get_block_write, mpd->wbc); | |
cb20d518 TT |
1371 | |
1372 | if (!err) | |
a1d6cc56 | 1373 | mpd->pages_written++; |
64769240 AT |
1374 | /* |
1375 | * In error case, we have to continue because | |
1376 | * remaining pages are still locked | |
64769240 AT |
1377 | */ |
1378 | if (ret == 0) | |
1379 | ret = err; | |
64769240 AT |
1380 | } |
1381 | pagevec_release(&pvec); | |
1382 | } | |
bd2d0210 | 1383 | ext4_io_submit(&io_submit); |
64769240 | 1384 | return ret; |
64769240 AT |
1385 | } |
1386 | ||
c7f5938a | 1387 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd) |
c4a0c46e AK |
1388 | { |
1389 | int nr_pages, i; | |
1390 | pgoff_t index, end; | |
1391 | struct pagevec pvec; | |
1392 | struct inode *inode = mpd->inode; | |
1393 | struct address_space *mapping = inode->i_mapping; | |
1394 | ||
c7f5938a CW |
1395 | index = mpd->first_page; |
1396 | end = mpd->next_page - 1; | |
c4a0c46e AK |
1397 | while (index <= end) { |
1398 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1399 | if (nr_pages == 0) | |
1400 | break; | |
1401 | for (i = 0; i < nr_pages; i++) { | |
1402 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1403 | if (page->index > end) |
c4a0c46e | 1404 | break; |
c4a0c46e AK |
1405 | BUG_ON(!PageLocked(page)); |
1406 | BUG_ON(PageWriteback(page)); | |
1407 | block_invalidatepage(page, 0); | |
1408 | ClearPageUptodate(page); | |
1409 | unlock_page(page); | |
1410 | } | |
9b1d0998 JK |
1411 | index = pvec.pages[nr_pages - 1]->index + 1; |
1412 | pagevec_release(&pvec); | |
c4a0c46e AK |
1413 | } |
1414 | return; | |
1415 | } | |
1416 | ||
df22291f AK |
1417 | static void ext4_print_free_blocks(struct inode *inode) |
1418 | { | |
1419 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1693918e | 1420 | printk(KERN_CRIT "Total free blocks count %lld\n", |
5dee5437 TT |
1421 | EXT4_C2B(EXT4_SB(inode->i_sb), |
1422 | ext4_count_free_clusters(inode->i_sb))); | |
1693918e TT |
1423 | printk(KERN_CRIT "Free/Dirty block details\n"); |
1424 | printk(KERN_CRIT "free_blocks=%lld\n", | |
57042651 TT |
1425 | (long long) EXT4_C2B(EXT4_SB(inode->i_sb), |
1426 | percpu_counter_sum(&sbi->s_freeclusters_counter))); | |
1693918e | 1427 | printk(KERN_CRIT "dirty_blocks=%lld\n", |
7b415bf6 AK |
1428 | (long long) EXT4_C2B(EXT4_SB(inode->i_sb), |
1429 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); | |
1693918e TT |
1430 | printk(KERN_CRIT "Block reservation details\n"); |
1431 | printk(KERN_CRIT "i_reserved_data_blocks=%u\n", | |
1432 | EXT4_I(inode)->i_reserved_data_blocks); | |
1433 | printk(KERN_CRIT "i_reserved_meta_blocks=%u\n", | |
1434 | EXT4_I(inode)->i_reserved_meta_blocks); | |
df22291f AK |
1435 | return; |
1436 | } | |
1437 | ||
64769240 | 1438 | /* |
5a87b7a5 TT |
1439 | * mpage_da_map_and_submit - go through given space, map them |
1440 | * if necessary, and then submit them for I/O | |
64769240 | 1441 | * |
8dc207c0 | 1442 | * @mpd - bh describing space |
64769240 AT |
1443 | * |
1444 | * The function skips space we know is already mapped to disk blocks. | |
1445 | * | |
64769240 | 1446 | */ |
5a87b7a5 | 1447 | static void mpage_da_map_and_submit(struct mpage_da_data *mpd) |
64769240 | 1448 | { |
2ac3b6e0 | 1449 | int err, blks, get_blocks_flags; |
1de3e3df | 1450 | struct ext4_map_blocks map, *mapp = NULL; |
2fa3cdfb TT |
1451 | sector_t next = mpd->b_blocknr; |
1452 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; | |
1453 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; | |
1454 | handle_t *handle = NULL; | |
64769240 AT |
1455 | |
1456 | /* | |
5a87b7a5 TT |
1457 | * If the blocks are mapped already, or we couldn't accumulate |
1458 | * any blocks, then proceed immediately to the submission stage. | |
2fa3cdfb | 1459 | */ |
5a87b7a5 TT |
1460 | if ((mpd->b_size == 0) || |
1461 | ((mpd->b_state & (1 << BH_Mapped)) && | |
1462 | !(mpd->b_state & (1 << BH_Delay)) && | |
1463 | !(mpd->b_state & (1 << BH_Unwritten)))) | |
1464 | goto submit_io; | |
2fa3cdfb TT |
1465 | |
1466 | handle = ext4_journal_current_handle(); | |
1467 | BUG_ON(!handle); | |
1468 | ||
79ffab34 | 1469 | /* |
79e83036 | 1470 | * Call ext4_map_blocks() to allocate any delayed allocation |
2ac3b6e0 TT |
1471 | * blocks, or to convert an uninitialized extent to be |
1472 | * initialized (in the case where we have written into | |
1473 | * one or more preallocated blocks). | |
1474 | * | |
1475 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to | |
1476 | * indicate that we are on the delayed allocation path. This | |
1477 | * affects functions in many different parts of the allocation | |
1478 | * call path. This flag exists primarily because we don't | |
79e83036 | 1479 | * want to change *many* call functions, so ext4_map_blocks() |
f2321097 | 1480 | * will set the EXT4_STATE_DELALLOC_RESERVED flag once the |
2ac3b6e0 TT |
1481 | * inode's allocation semaphore is taken. |
1482 | * | |
1483 | * If the blocks in questions were delalloc blocks, set | |
1484 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting | |
1485 | * variables are updated after the blocks have been allocated. | |
79ffab34 | 1486 | */ |
2ed88685 TT |
1487 | map.m_lblk = next; |
1488 | map.m_len = max_blocks; | |
1296cc85 | 1489 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE; |
744692dc JZ |
1490 | if (ext4_should_dioread_nolock(mpd->inode)) |
1491 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
2ac3b6e0 | 1492 | if (mpd->b_state & (1 << BH_Delay)) |
1296cc85 AK |
1493 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
1494 | ||
2ed88685 | 1495 | blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags); |
2fa3cdfb | 1496 | if (blks < 0) { |
e3570639 ES |
1497 | struct super_block *sb = mpd->inode->i_sb; |
1498 | ||
2fa3cdfb | 1499 | err = blks; |
ed5bde0b | 1500 | /* |
5a87b7a5 | 1501 | * If get block returns EAGAIN or ENOSPC and there |
97498956 TT |
1502 | * appears to be free blocks we will just let |
1503 | * mpage_da_submit_io() unlock all of the pages. | |
c4a0c46e AK |
1504 | */ |
1505 | if (err == -EAGAIN) | |
5a87b7a5 | 1506 | goto submit_io; |
df22291f | 1507 | |
5dee5437 | 1508 | if (err == -ENOSPC && ext4_count_free_clusters(sb)) { |
df22291f | 1509 | mpd->retval = err; |
5a87b7a5 | 1510 | goto submit_io; |
df22291f AK |
1511 | } |
1512 | ||
c4a0c46e | 1513 | /* |
ed5bde0b TT |
1514 | * get block failure will cause us to loop in |
1515 | * writepages, because a_ops->writepage won't be able | |
1516 | * to make progress. The page will be redirtied by | |
1517 | * writepage and writepages will again try to write | |
1518 | * the same. | |
c4a0c46e | 1519 | */ |
e3570639 ES |
1520 | if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
1521 | ext4_msg(sb, KERN_CRIT, | |
1522 | "delayed block allocation failed for inode %lu " | |
1523 | "at logical offset %llu with max blocks %zd " | |
1524 | "with error %d", mpd->inode->i_ino, | |
1525 | (unsigned long long) next, | |
1526 | mpd->b_size >> mpd->inode->i_blkbits, err); | |
1527 | ext4_msg(sb, KERN_CRIT, | |
1528 | "This should not happen!! Data will be lost\n"); | |
1529 | if (err == -ENOSPC) | |
1530 | ext4_print_free_blocks(mpd->inode); | |
030ba6bc | 1531 | } |
2fa3cdfb | 1532 | /* invalidate all the pages */ |
c7f5938a | 1533 | ext4_da_block_invalidatepages(mpd); |
e0fd9b90 CW |
1534 | |
1535 | /* Mark this page range as having been completed */ | |
1536 | mpd->io_done = 1; | |
5a87b7a5 | 1537 | return; |
c4a0c46e | 1538 | } |
2fa3cdfb TT |
1539 | BUG_ON(blks == 0); |
1540 | ||
1de3e3df | 1541 | mapp = ↦ |
2ed88685 TT |
1542 | if (map.m_flags & EXT4_MAP_NEW) { |
1543 | struct block_device *bdev = mpd->inode->i_sb->s_bdev; | |
1544 | int i; | |
64769240 | 1545 | |
2ed88685 TT |
1546 | for (i = 0; i < map.m_len; i++) |
1547 | unmap_underlying_metadata(bdev, map.m_pblk + i); | |
64769240 | 1548 | |
decbd919 TT |
1549 | if (ext4_should_order_data(mpd->inode)) { |
1550 | err = ext4_jbd2_file_inode(handle, mpd->inode); | |
1551 | if (err) | |
1552 | /* Only if the journal is aborted */ | |
1553 | return; | |
1554 | } | |
2fa3cdfb TT |
1555 | } |
1556 | ||
1557 | /* | |
03f5d8bc | 1558 | * Update on-disk size along with block allocation. |
2fa3cdfb TT |
1559 | */ |
1560 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; | |
1561 | if (disksize > i_size_read(mpd->inode)) | |
1562 | disksize = i_size_read(mpd->inode); | |
1563 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { | |
1564 | ext4_update_i_disksize(mpd->inode, disksize); | |
5a87b7a5 TT |
1565 | err = ext4_mark_inode_dirty(handle, mpd->inode); |
1566 | if (err) | |
1567 | ext4_error(mpd->inode->i_sb, | |
1568 | "Failed to mark inode %lu dirty", | |
1569 | mpd->inode->i_ino); | |
2fa3cdfb TT |
1570 | } |
1571 | ||
5a87b7a5 | 1572 | submit_io: |
1de3e3df | 1573 | mpage_da_submit_io(mpd, mapp); |
5a87b7a5 | 1574 | mpd->io_done = 1; |
64769240 AT |
1575 | } |
1576 | ||
bf068ee2 AK |
1577 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
1578 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
1579 | |
1580 | /* | |
1581 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
1582 | * | |
1583 | * @mpd->lbh - extent of blocks | |
1584 | * @logical - logical number of the block in the file | |
1585 | * @bh - bh of the block (used to access block's state) | |
1586 | * | |
1587 | * the function is used to collect contig. blocks in same state | |
1588 | */ | |
1589 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, | |
8dc207c0 TT |
1590 | sector_t logical, size_t b_size, |
1591 | unsigned long b_state) | |
64769240 | 1592 | { |
64769240 | 1593 | sector_t next; |
8dc207c0 | 1594 | int nrblocks = mpd->b_size >> mpd->inode->i_blkbits; |
64769240 | 1595 | |
c445e3e0 ES |
1596 | /* |
1597 | * XXX Don't go larger than mballoc is willing to allocate | |
1598 | * This is a stopgap solution. We eventually need to fold | |
1599 | * mpage_da_submit_io() into this function and then call | |
79e83036 | 1600 | * ext4_map_blocks() multiple times in a loop |
c445e3e0 ES |
1601 | */ |
1602 | if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize) | |
1603 | goto flush_it; | |
1604 | ||
525f4ed8 | 1605 | /* check if thereserved journal credits might overflow */ |
12e9b892 | 1606 | if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) { |
525f4ed8 MC |
1607 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { |
1608 | /* | |
1609 | * With non-extent format we are limited by the journal | |
1610 | * credit available. Total credit needed to insert | |
1611 | * nrblocks contiguous blocks is dependent on the | |
1612 | * nrblocks. So limit nrblocks. | |
1613 | */ | |
1614 | goto flush_it; | |
1615 | } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) > | |
1616 | EXT4_MAX_TRANS_DATA) { | |
1617 | /* | |
1618 | * Adding the new buffer_head would make it cross the | |
1619 | * allowed limit for which we have journal credit | |
1620 | * reserved. So limit the new bh->b_size | |
1621 | */ | |
1622 | b_size = (EXT4_MAX_TRANS_DATA - nrblocks) << | |
1623 | mpd->inode->i_blkbits; | |
1624 | /* we will do mpage_da_submit_io in the next loop */ | |
1625 | } | |
1626 | } | |
64769240 AT |
1627 | /* |
1628 | * First block in the extent | |
1629 | */ | |
8dc207c0 TT |
1630 | if (mpd->b_size == 0) { |
1631 | mpd->b_blocknr = logical; | |
1632 | mpd->b_size = b_size; | |
1633 | mpd->b_state = b_state & BH_FLAGS; | |
64769240 AT |
1634 | return; |
1635 | } | |
1636 | ||
8dc207c0 | 1637 | next = mpd->b_blocknr + nrblocks; |
64769240 AT |
1638 | /* |
1639 | * Can we merge the block to our big extent? | |
1640 | */ | |
8dc207c0 TT |
1641 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
1642 | mpd->b_size += b_size; | |
64769240 AT |
1643 | return; |
1644 | } | |
1645 | ||
525f4ed8 | 1646 | flush_it: |
64769240 AT |
1647 | /* |
1648 | * We couldn't merge the block to our extent, so we | |
1649 | * need to flush current extent and start new one | |
1650 | */ | |
5a87b7a5 | 1651 | mpage_da_map_and_submit(mpd); |
a1d6cc56 | 1652 | return; |
64769240 AT |
1653 | } |
1654 | ||
c364b22c | 1655 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1656 | { |
c364b22c | 1657 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1658 | } |
1659 | ||
5356f261 AK |
1660 | /* |
1661 | * This function is grabs code from the very beginning of | |
1662 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1663 | * time. This function looks up the requested blocks and sets the | |
1664 | * buffer delay bit under the protection of i_data_sem. | |
1665 | */ | |
1666 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1667 | struct ext4_map_blocks *map, | |
1668 | struct buffer_head *bh) | |
1669 | { | |
1670 | int retval; | |
1671 | sector_t invalid_block = ~((sector_t) 0xffff); | |
1672 | ||
1673 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1674 | invalid_block = ~0; | |
1675 | ||
1676 | map->m_flags = 0; | |
1677 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1678 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1679 | (unsigned long) map->m_lblk); | |
1680 | /* | |
1681 | * Try to see if we can get the block without requesting a new | |
1682 | * file system block. | |
1683 | */ | |
1684 | down_read((&EXT4_I(inode)->i_data_sem)); | |
1685 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
1686 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); | |
1687 | else | |
1688 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); | |
1689 | ||
1690 | if (retval == 0) { | |
1691 | /* | |
1692 | * XXX: __block_prepare_write() unmaps passed block, | |
1693 | * is it OK? | |
1694 | */ | |
1695 | /* If the block was allocated from previously allocated cluster, | |
1696 | * then we dont need to reserve it again. */ | |
1697 | if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) { | |
1698 | retval = ext4_da_reserve_space(inode, iblock); | |
1699 | if (retval) | |
1700 | /* not enough space to reserve */ | |
1701 | goto out_unlock; | |
1702 | } | |
1703 | ||
1704 | /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served | |
1705 | * and it should not appear on the bh->b_state. | |
1706 | */ | |
1707 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; | |
1708 | ||
1709 | map_bh(bh, inode->i_sb, invalid_block); | |
1710 | set_buffer_new(bh); | |
1711 | set_buffer_delay(bh); | |
1712 | } | |
1713 | ||
1714 | out_unlock: | |
1715 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1716 | ||
1717 | return retval; | |
1718 | } | |
1719 | ||
64769240 | 1720 | /* |
b920c755 TT |
1721 | * This is a special get_blocks_t callback which is used by |
1722 | * ext4_da_write_begin(). It will either return mapped block or | |
1723 | * reserve space for a single block. | |
29fa89d0 AK |
1724 | * |
1725 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1726 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1727 | * | |
1728 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1729 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1730 | * initialized properly. | |
64769240 AT |
1731 | */ |
1732 | static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, | |
2ed88685 | 1733 | struct buffer_head *bh, int create) |
64769240 | 1734 | { |
2ed88685 | 1735 | struct ext4_map_blocks map; |
64769240 AT |
1736 | int ret = 0; |
1737 | ||
1738 | BUG_ON(create == 0); | |
2ed88685 TT |
1739 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1740 | ||
1741 | map.m_lblk = iblock; | |
1742 | map.m_len = 1; | |
64769240 AT |
1743 | |
1744 | /* | |
1745 | * first, we need to know whether the block is allocated already | |
1746 | * preallocated blocks are unmapped but should treated | |
1747 | * the same as allocated blocks. | |
1748 | */ | |
5356f261 AK |
1749 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1750 | if (ret <= 0) | |
2ed88685 | 1751 | return ret; |
64769240 | 1752 | |
2ed88685 TT |
1753 | map_bh(bh, inode->i_sb, map.m_pblk); |
1754 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
1755 | ||
1756 | if (buffer_unwritten(bh)) { | |
1757 | /* A delayed write to unwritten bh should be marked | |
1758 | * new and mapped. Mapped ensures that we don't do | |
1759 | * get_block multiple times when we write to the same | |
1760 | * offset and new ensures that we do proper zero out | |
1761 | * for partial write. | |
1762 | */ | |
1763 | set_buffer_new(bh); | |
c8205636 | 1764 | set_buffer_mapped(bh); |
2ed88685 TT |
1765 | } |
1766 | return 0; | |
64769240 | 1767 | } |
61628a3f | 1768 | |
b920c755 TT |
1769 | /* |
1770 | * This function is used as a standard get_block_t calback function | |
1771 | * when there is no desire to allocate any blocks. It is used as a | |
ebdec241 | 1772 | * callback function for block_write_begin() and block_write_full_page(). |
206f7ab4 | 1773 | * These functions should only try to map a single block at a time. |
b920c755 TT |
1774 | * |
1775 | * Since this function doesn't do block allocations even if the caller | |
1776 | * requests it by passing in create=1, it is critically important that | |
1777 | * any caller checks to make sure that any buffer heads are returned | |
1778 | * by this function are either all already mapped or marked for | |
206f7ab4 CH |
1779 | * delayed allocation before calling block_write_full_page(). Otherwise, |
1780 | * b_blocknr could be left unitialized, and the page write functions will | |
1781 | * be taken by surprise. | |
b920c755 TT |
1782 | */ |
1783 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, | |
f0e6c985 AK |
1784 | struct buffer_head *bh_result, int create) |
1785 | { | |
a2dc52b5 | 1786 | BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); |
2ed88685 | 1787 | return _ext4_get_block(inode, iblock, bh_result, 0); |
61628a3f MC |
1788 | } |
1789 | ||
62e086be AK |
1790 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1791 | { | |
1792 | get_bh(bh); | |
1793 | return 0; | |
1794 | } | |
1795 | ||
1796 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1797 | { | |
1798 | put_bh(bh); | |
1799 | return 0; | |
1800 | } | |
1801 | ||
1802 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1803 | unsigned int len) |
1804 | { | |
1805 | struct address_space *mapping = page->mapping; | |
1806 | struct inode *inode = mapping->host; | |
1807 | struct buffer_head *page_bufs; | |
1808 | handle_t *handle = NULL; | |
1809 | int ret = 0; | |
1810 | int err; | |
1811 | ||
cb20d518 | 1812 | ClearPageChecked(page); |
62e086be AK |
1813 | page_bufs = page_buffers(page); |
1814 | BUG_ON(!page_bufs); | |
1815 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one); | |
1816 | /* As soon as we unlock the page, it can go away, but we have | |
1817 | * references to buffers so we are safe */ | |
1818 | unlock_page(page); | |
1819 | ||
1820 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); | |
1821 | if (IS_ERR(handle)) { | |
1822 | ret = PTR_ERR(handle); | |
1823 | goto out; | |
1824 | } | |
1825 | ||
441c8508 CW |
1826 | BUG_ON(!ext4_handle_valid(handle)); |
1827 | ||
62e086be AK |
1828 | ret = walk_page_buffers(handle, page_bufs, 0, len, NULL, |
1829 | do_journal_get_write_access); | |
1830 | ||
1831 | err = walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1832 | write_end_fn); | |
1833 | if (ret == 0) | |
1834 | ret = err; | |
2d859db3 | 1835 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1836 | err = ext4_journal_stop(handle); |
1837 | if (!ret) | |
1838 | ret = err; | |
1839 | ||
1840 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one); | |
19f5fb7a | 1841 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be AK |
1842 | out: |
1843 | return ret; | |
1844 | } | |
1845 | ||
744692dc JZ |
1846 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode); |
1847 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate); | |
1848 | ||
61628a3f | 1849 | /* |
43ce1d23 AK |
1850 | * Note that we don't need to start a transaction unless we're journaling data |
1851 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1852 | * need to file the inode to the transaction's list in ordered mode because if | |
1853 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1854 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1855 | * transaction the data will hit the disk. In case we are journaling data, we |
1856 | * cannot start transaction directly because transaction start ranks above page | |
1857 | * lock so we have to do some magic. | |
1858 | * | |
b920c755 TT |
1859 | * This function can get called via... |
1860 | * - ext4_da_writepages after taking page lock (have journal handle) | |
1861 | * - journal_submit_inode_data_buffers (no journal handle) | |
1862 | * - shrink_page_list via pdflush (no journal handle) | |
1863 | * - grab_page_cache when doing write_begin (have journal handle) | |
43ce1d23 AK |
1864 | * |
1865 | * We don't do any block allocation in this function. If we have page with | |
1866 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1867 | * is important for mmaped based write. So if we do with blocksize 1K | |
1868 | * truncate(f, 1024); | |
1869 | * a = mmap(f, 0, 4096); | |
1870 | * a[0] = 'a'; | |
1871 | * truncate(f, 4096); | |
1872 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
1873 | * but other bufer_heads would be unmapped but dirty(dirty done via the | |
1874 | * do_wp_page). So writepage should write the first block. If we modify | |
1875 | * the mmap area beyond 1024 we will again get a page_fault and the | |
1876 | * page_mkwrite callback will do the block allocation and mark the | |
1877 | * buffer_heads mapped. | |
1878 | * | |
1879 | * We redirty the page if we have any buffer_heads that is either delay or | |
1880 | * unwritten in the page. | |
1881 | * | |
1882 | * We can get recursively called as show below. | |
1883 | * | |
1884 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1885 | * ext4_writepage() | |
1886 | * | |
1887 | * But since we don't do any block allocation we should not deadlock. | |
1888 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 1889 | */ |
43ce1d23 | 1890 | static int ext4_writepage(struct page *page, |
62e086be | 1891 | struct writeback_control *wbc) |
64769240 | 1892 | { |
a42afc5f | 1893 | int ret = 0, commit_write = 0; |
61628a3f | 1894 | loff_t size; |
498e5f24 | 1895 | unsigned int len; |
744692dc | 1896 | struct buffer_head *page_bufs = NULL; |
61628a3f MC |
1897 | struct inode *inode = page->mapping->host; |
1898 | ||
a9c667f8 | 1899 | trace_ext4_writepage(page); |
f0e6c985 AK |
1900 | size = i_size_read(inode); |
1901 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
1902 | len = size & ~PAGE_CACHE_MASK; | |
1903 | else | |
1904 | len = PAGE_CACHE_SIZE; | |
64769240 | 1905 | |
a42afc5f TT |
1906 | /* |
1907 | * If the page does not have buffers (for whatever reason), | |
a107e5a3 | 1908 | * try to create them using __block_write_begin. If this |
a42afc5f TT |
1909 | * fails, redirty the page and move on. |
1910 | */ | |
b1142e8f | 1911 | if (!page_has_buffers(page)) { |
a107e5a3 | 1912 | if (__block_write_begin(page, 0, len, |
a42afc5f TT |
1913 | noalloc_get_block_write)) { |
1914 | redirty_page: | |
f0e6c985 AK |
1915 | redirty_page_for_writepage(wbc, page); |
1916 | unlock_page(page); | |
1917 | return 0; | |
1918 | } | |
a42afc5f TT |
1919 | commit_write = 1; |
1920 | } | |
1921 | page_bufs = page_buffers(page); | |
1922 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, | |
1923 | ext4_bh_delay_or_unwritten)) { | |
f0e6c985 | 1924 | /* |
b1142e8f TT |
1925 | * We don't want to do block allocation, so redirty |
1926 | * the page and return. We may reach here when we do | |
1927 | * a journal commit via journal_submit_inode_data_buffers. | |
1928 | * We can also reach here via shrink_page_list | |
f0e6c985 | 1929 | */ |
a42afc5f TT |
1930 | goto redirty_page; |
1931 | } | |
1932 | if (commit_write) | |
ed9b3e33 | 1933 | /* now mark the buffer_heads as dirty and uptodate */ |
b767e78a | 1934 | block_commit_write(page, 0, len); |
64769240 | 1935 | |
cb20d518 | 1936 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
1937 | /* |
1938 | * It's mmapped pagecache. Add buffers and journal it. There | |
1939 | * doesn't seem much point in redirtying the page here. | |
1940 | */ | |
3f0ca309 | 1941 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 1942 | |
a42afc5f | 1943 | if (buffer_uninit(page_bufs)) { |
744692dc JZ |
1944 | ext4_set_bh_endio(page_bufs, inode); |
1945 | ret = block_write_full_page_endio(page, noalloc_get_block_write, | |
1946 | wbc, ext4_end_io_buffer_write); | |
1947 | } else | |
b920c755 TT |
1948 | ret = block_write_full_page(page, noalloc_get_block_write, |
1949 | wbc); | |
64769240 | 1950 | |
64769240 AT |
1951 | return ret; |
1952 | } | |
1953 | ||
61628a3f | 1954 | /* |
525f4ed8 | 1955 | * This is called via ext4_da_writepages() to |
25985edc | 1956 | * calculate the total number of credits to reserve to fit |
525f4ed8 MC |
1957 | * a single extent allocation into a single transaction, |
1958 | * ext4_da_writpeages() will loop calling this before | |
1959 | * the block allocation. | |
61628a3f | 1960 | */ |
525f4ed8 MC |
1961 | |
1962 | static int ext4_da_writepages_trans_blocks(struct inode *inode) | |
1963 | { | |
1964 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
1965 | ||
1966 | /* | |
1967 | * With non-extent format the journal credit needed to | |
1968 | * insert nrblocks contiguous block is dependent on | |
1969 | * number of contiguous block. So we will limit | |
1970 | * number of contiguous block to a sane value | |
1971 | */ | |
12e9b892 | 1972 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) && |
525f4ed8 MC |
1973 | (max_blocks > EXT4_MAX_TRANS_DATA)) |
1974 | max_blocks = EXT4_MAX_TRANS_DATA; | |
1975 | ||
1976 | return ext4_chunk_trans_blocks(inode, max_blocks); | |
1977 | } | |
61628a3f | 1978 | |
8e48dcfb TT |
1979 | /* |
1980 | * write_cache_pages_da - walk the list of dirty pages of the given | |
8eb9e5ce | 1981 | * address space and accumulate pages that need writing, and call |
168fc022 TT |
1982 | * mpage_da_map_and_submit to map a single contiguous memory region |
1983 | * and then write them. | |
8e48dcfb TT |
1984 | */ |
1985 | static int write_cache_pages_da(struct address_space *mapping, | |
1986 | struct writeback_control *wbc, | |
72f84e65 ES |
1987 | struct mpage_da_data *mpd, |
1988 | pgoff_t *done_index) | |
8e48dcfb | 1989 | { |
4f01b02c | 1990 | struct buffer_head *bh, *head; |
168fc022 | 1991 | struct inode *inode = mapping->host; |
4f01b02c TT |
1992 | struct pagevec pvec; |
1993 | unsigned int nr_pages; | |
1994 | sector_t logical; | |
1995 | pgoff_t index, end; | |
1996 | long nr_to_write = wbc->nr_to_write; | |
1997 | int i, tag, ret = 0; | |
8e48dcfb | 1998 | |
168fc022 TT |
1999 | memset(mpd, 0, sizeof(struct mpage_da_data)); |
2000 | mpd->wbc = wbc; | |
2001 | mpd->inode = inode; | |
8e48dcfb TT |
2002 | pagevec_init(&pvec, 0); |
2003 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2004 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
2005 | ||
6e6938b6 | 2006 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2007 | tag = PAGECACHE_TAG_TOWRITE; |
2008 | else | |
2009 | tag = PAGECACHE_TAG_DIRTY; | |
2010 | ||
72f84e65 | 2011 | *done_index = index; |
4f01b02c | 2012 | while (index <= end) { |
5b41d924 | 2013 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2014 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2015 | if (nr_pages == 0) | |
4f01b02c | 2016 | return 0; |
8e48dcfb TT |
2017 | |
2018 | for (i = 0; i < nr_pages; i++) { | |
2019 | struct page *page = pvec.pages[i]; | |
2020 | ||
2021 | /* | |
2022 | * At this point, the page may be truncated or | |
2023 | * invalidated (changing page->mapping to NULL), or | |
2024 | * even swizzled back from swapper_space to tmpfs file | |
2025 | * mapping. However, page->index will not change | |
2026 | * because we have a reference on the page. | |
2027 | */ | |
4f01b02c TT |
2028 | if (page->index > end) |
2029 | goto out; | |
8e48dcfb | 2030 | |
72f84e65 ES |
2031 | *done_index = page->index + 1; |
2032 | ||
78aaced3 TT |
2033 | /* |
2034 | * If we can't merge this page, and we have | |
2035 | * accumulated an contiguous region, write it | |
2036 | */ | |
2037 | if ((mpd->next_page != page->index) && | |
2038 | (mpd->next_page != mpd->first_page)) { | |
2039 | mpage_da_map_and_submit(mpd); | |
2040 | goto ret_extent_tail; | |
2041 | } | |
2042 | ||
8e48dcfb TT |
2043 | lock_page(page); |
2044 | ||
2045 | /* | |
4f01b02c TT |
2046 | * If the page is no longer dirty, or its |
2047 | * mapping no longer corresponds to inode we | |
2048 | * are writing (which means it has been | |
2049 | * truncated or invalidated), or the page is | |
2050 | * already under writeback and we are not | |
2051 | * doing a data integrity writeback, skip the page | |
8e48dcfb | 2052 | */ |
4f01b02c TT |
2053 | if (!PageDirty(page) || |
2054 | (PageWriteback(page) && | |
2055 | (wbc->sync_mode == WB_SYNC_NONE)) || | |
2056 | unlikely(page->mapping != mapping)) { | |
8e48dcfb TT |
2057 | unlock_page(page); |
2058 | continue; | |
2059 | } | |
2060 | ||
7cb1a535 | 2061 | wait_on_page_writeback(page); |
8e48dcfb | 2062 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2063 | |
168fc022 | 2064 | if (mpd->next_page != page->index) |
8eb9e5ce | 2065 | mpd->first_page = page->index; |
8eb9e5ce TT |
2066 | mpd->next_page = page->index + 1; |
2067 | logical = (sector_t) page->index << | |
2068 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2069 | ||
2070 | if (!page_has_buffers(page)) { | |
4f01b02c TT |
2071 | mpage_add_bh_to_extent(mpd, logical, |
2072 | PAGE_CACHE_SIZE, | |
8eb9e5ce | 2073 | (1 << BH_Dirty) | (1 << BH_Uptodate)); |
4f01b02c TT |
2074 | if (mpd->io_done) |
2075 | goto ret_extent_tail; | |
8eb9e5ce TT |
2076 | } else { |
2077 | /* | |
4f01b02c TT |
2078 | * Page with regular buffer heads, |
2079 | * just add all dirty ones | |
8eb9e5ce TT |
2080 | */ |
2081 | head = page_buffers(page); | |
2082 | bh = head; | |
2083 | do { | |
2084 | BUG_ON(buffer_locked(bh)); | |
2085 | /* | |
2086 | * We need to try to allocate | |
2087 | * unmapped blocks in the same page. | |
2088 | * Otherwise we won't make progress | |
2089 | * with the page in ext4_writepage | |
2090 | */ | |
2091 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { | |
2092 | mpage_add_bh_to_extent(mpd, logical, | |
2093 | bh->b_size, | |
2094 | bh->b_state); | |
4f01b02c TT |
2095 | if (mpd->io_done) |
2096 | goto ret_extent_tail; | |
8eb9e5ce TT |
2097 | } else if (buffer_dirty(bh) && (buffer_mapped(bh))) { |
2098 | /* | |
4f01b02c TT |
2099 | * mapped dirty buffer. We need |
2100 | * to update the b_state | |
2101 | * because we look at b_state | |
2102 | * in mpage_da_map_blocks. We | |
2103 | * don't update b_size because | |
2104 | * if we find an unmapped | |
2105 | * buffer_head later we need to | |
2106 | * use the b_state flag of that | |
2107 | * buffer_head. | |
8eb9e5ce TT |
2108 | */ |
2109 | if (mpd->b_size == 0) | |
2110 | mpd->b_state = bh->b_state & BH_FLAGS; | |
2111 | } | |
2112 | logical++; | |
2113 | } while ((bh = bh->b_this_page) != head); | |
8e48dcfb TT |
2114 | } |
2115 | ||
2116 | if (nr_to_write > 0) { | |
2117 | nr_to_write--; | |
2118 | if (nr_to_write == 0 && | |
4f01b02c | 2119 | wbc->sync_mode == WB_SYNC_NONE) |
8e48dcfb TT |
2120 | /* |
2121 | * We stop writing back only if we are | |
2122 | * not doing integrity sync. In case of | |
2123 | * integrity sync we have to keep going | |
2124 | * because someone may be concurrently | |
2125 | * dirtying pages, and we might have | |
2126 | * synced a lot of newly appeared dirty | |
2127 | * pages, but have not synced all of the | |
2128 | * old dirty pages. | |
2129 | */ | |
4f01b02c | 2130 | goto out; |
8e48dcfb TT |
2131 | } |
2132 | } | |
2133 | pagevec_release(&pvec); | |
2134 | cond_resched(); | |
2135 | } | |
4f01b02c TT |
2136 | return 0; |
2137 | ret_extent_tail: | |
2138 | ret = MPAGE_DA_EXTENT_TAIL; | |
8eb9e5ce TT |
2139 | out: |
2140 | pagevec_release(&pvec); | |
2141 | cond_resched(); | |
8e48dcfb TT |
2142 | return ret; |
2143 | } | |
2144 | ||
2145 | ||
64769240 | 2146 | static int ext4_da_writepages(struct address_space *mapping, |
a1d6cc56 | 2147 | struct writeback_control *wbc) |
64769240 | 2148 | { |
22208ded AK |
2149 | pgoff_t index; |
2150 | int range_whole = 0; | |
61628a3f | 2151 | handle_t *handle = NULL; |
df22291f | 2152 | struct mpage_da_data mpd; |
5e745b04 | 2153 | struct inode *inode = mapping->host; |
498e5f24 | 2154 | int pages_written = 0; |
55138e0b | 2155 | unsigned int max_pages; |
2acf2c26 | 2156 | int range_cyclic, cycled = 1, io_done = 0; |
55138e0b TT |
2157 | int needed_blocks, ret = 0; |
2158 | long desired_nr_to_write, nr_to_writebump = 0; | |
de89de6e | 2159 | loff_t range_start = wbc->range_start; |
5e745b04 | 2160 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
72f84e65 | 2161 | pgoff_t done_index = 0; |
5b41d924 | 2162 | pgoff_t end; |
61628a3f | 2163 | |
9bffad1e | 2164 | trace_ext4_da_writepages(inode, wbc); |
ba80b101 | 2165 | |
61628a3f MC |
2166 | /* |
2167 | * No pages to write? This is mainly a kludge to avoid starting | |
2168 | * a transaction for special inodes like journal inode on last iput() | |
2169 | * because that could violate lock ordering on umount | |
2170 | */ | |
a1d6cc56 | 2171 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
61628a3f | 2172 | return 0; |
2a21e37e TT |
2173 | |
2174 | /* | |
2175 | * If the filesystem has aborted, it is read-only, so return | |
2176 | * right away instead of dumping stack traces later on that | |
2177 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2178 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e TT |
2179 | * the latter could be true if the filesystem is mounted |
2180 | * read-only, and in that case, ext4_da_writepages should | |
2181 | * *never* be called, so if that ever happens, we would want | |
2182 | * the stack trace. | |
2183 | */ | |
4ab2f15b | 2184 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
2a21e37e TT |
2185 | return -EROFS; |
2186 | ||
22208ded AK |
2187 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2188 | range_whole = 1; | |
61628a3f | 2189 | |
2acf2c26 AK |
2190 | range_cyclic = wbc->range_cyclic; |
2191 | if (wbc->range_cyclic) { | |
22208ded | 2192 | index = mapping->writeback_index; |
2acf2c26 AK |
2193 | if (index) |
2194 | cycled = 0; | |
2195 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2196 | wbc->range_end = LLONG_MAX; | |
2197 | wbc->range_cyclic = 0; | |
5b41d924 ES |
2198 | end = -1; |
2199 | } else { | |
22208ded | 2200 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
5b41d924 ES |
2201 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
2202 | } | |
a1d6cc56 | 2203 | |
55138e0b TT |
2204 | /* |
2205 | * This works around two forms of stupidity. The first is in | |
2206 | * the writeback code, which caps the maximum number of pages | |
2207 | * written to be 1024 pages. This is wrong on multiple | |
2208 | * levels; different architectues have a different page size, | |
2209 | * which changes the maximum amount of data which gets | |
2210 | * written. Secondly, 4 megabytes is way too small. XFS | |
2211 | * forces this value to be 16 megabytes by multiplying | |
2212 | * nr_to_write parameter by four, and then relies on its | |
2213 | * allocator to allocate larger extents to make them | |
2214 | * contiguous. Unfortunately this brings us to the second | |
2215 | * stupidity, which is that ext4's mballoc code only allocates | |
2216 | * at most 2048 blocks. So we force contiguous writes up to | |
2217 | * the number of dirty blocks in the inode, or | |
2218 | * sbi->max_writeback_mb_bump whichever is smaller. | |
2219 | */ | |
2220 | max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); | |
b443e733 ES |
2221 | if (!range_cyclic && range_whole) { |
2222 | if (wbc->nr_to_write == LONG_MAX) | |
2223 | desired_nr_to_write = wbc->nr_to_write; | |
2224 | else | |
2225 | desired_nr_to_write = wbc->nr_to_write * 8; | |
2226 | } else | |
55138e0b TT |
2227 | desired_nr_to_write = ext4_num_dirty_pages(inode, index, |
2228 | max_pages); | |
2229 | if (desired_nr_to_write > max_pages) | |
2230 | desired_nr_to_write = max_pages; | |
2231 | ||
2232 | if (wbc->nr_to_write < desired_nr_to_write) { | |
2233 | nr_to_writebump = desired_nr_to_write - wbc->nr_to_write; | |
2234 | wbc->nr_to_write = desired_nr_to_write; | |
2235 | } | |
2236 | ||
2acf2c26 | 2237 | retry: |
6e6938b6 | 2238 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2239 | tag_pages_for_writeback(mapping, index, end); |
2240 | ||
22208ded | 2241 | while (!ret && wbc->nr_to_write > 0) { |
a1d6cc56 AK |
2242 | |
2243 | /* | |
2244 | * we insert one extent at a time. So we need | |
2245 | * credit needed for single extent allocation. | |
2246 | * journalled mode is currently not supported | |
2247 | * by delalloc | |
2248 | */ | |
2249 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2250 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2251 | |
61628a3f MC |
2252 | /* start a new transaction*/ |
2253 | handle = ext4_journal_start(inode, needed_blocks); | |
2254 | if (IS_ERR(handle)) { | |
2255 | ret = PTR_ERR(handle); | |
1693918e | 2256 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2257 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2258 | wbc->nr_to_write, inode->i_ino, ret); |
61628a3f MC |
2259 | goto out_writepages; |
2260 | } | |
f63e6005 TT |
2261 | |
2262 | /* | |
8eb9e5ce | 2263 | * Now call write_cache_pages_da() to find the next |
f63e6005 | 2264 | * contiguous region of logical blocks that need |
8eb9e5ce | 2265 | * blocks to be allocated by ext4 and submit them. |
f63e6005 | 2266 | */ |
72f84e65 | 2267 | ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index); |
f63e6005 | 2268 | /* |
af901ca1 | 2269 | * If we have a contiguous extent of pages and we |
f63e6005 TT |
2270 | * haven't done the I/O yet, map the blocks and submit |
2271 | * them for I/O. | |
2272 | */ | |
2273 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
5a87b7a5 | 2274 | mpage_da_map_and_submit(&mpd); |
f63e6005 TT |
2275 | ret = MPAGE_DA_EXTENT_TAIL; |
2276 | } | |
b3a3ca8c | 2277 | trace_ext4_da_write_pages(inode, &mpd); |
f63e6005 | 2278 | wbc->nr_to_write -= mpd.pages_written; |
df22291f | 2279 | |
61628a3f | 2280 | ext4_journal_stop(handle); |
df22291f | 2281 | |
8f64b32e | 2282 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
22208ded AK |
2283 | /* commit the transaction which would |
2284 | * free blocks released in the transaction | |
2285 | * and try again | |
2286 | */ | |
df22291f | 2287 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded AK |
2288 | ret = 0; |
2289 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | |
a1d6cc56 AK |
2290 | /* |
2291 | * got one extent now try with | |
2292 | * rest of the pages | |
2293 | */ | |
22208ded | 2294 | pages_written += mpd.pages_written; |
a1d6cc56 | 2295 | ret = 0; |
2acf2c26 | 2296 | io_done = 1; |
22208ded | 2297 | } else if (wbc->nr_to_write) |
61628a3f MC |
2298 | /* |
2299 | * There is no more writeout needed | |
2300 | * or we requested for a noblocking writeout | |
2301 | * and we found the device congested | |
2302 | */ | |
61628a3f | 2303 | break; |
a1d6cc56 | 2304 | } |
2acf2c26 AK |
2305 | if (!io_done && !cycled) { |
2306 | cycled = 1; | |
2307 | index = 0; | |
2308 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2309 | wbc->range_end = mapping->writeback_index - 1; | |
2310 | goto retry; | |
2311 | } | |
22208ded AK |
2312 | |
2313 | /* Update index */ | |
2acf2c26 | 2314 | wbc->range_cyclic = range_cyclic; |
22208ded AK |
2315 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2316 | /* | |
2317 | * set the writeback_index so that range_cyclic | |
2318 | * mode will write it back later | |
2319 | */ | |
72f84e65 | 2320 | mapping->writeback_index = done_index; |
a1d6cc56 | 2321 | |
61628a3f | 2322 | out_writepages: |
2faf2e19 | 2323 | wbc->nr_to_write -= nr_to_writebump; |
de89de6e | 2324 | wbc->range_start = range_start; |
9bffad1e | 2325 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
61628a3f | 2326 | return ret; |
64769240 AT |
2327 | } |
2328 | ||
79f0be8d AK |
2329 | #define FALL_BACK_TO_NONDELALLOC 1 |
2330 | static int ext4_nonda_switch(struct super_block *sb) | |
2331 | { | |
2332 | s64 free_blocks, dirty_blocks; | |
2333 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2334 | ||
2335 | /* | |
2336 | * switch to non delalloc mode if we are running low | |
2337 | * on free block. The free block accounting via percpu | |
179f7ebf | 2338 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2339 | * accumulated on each CPU without updating global counters |
2340 | * Delalloc need an accurate free block accounting. So switch | |
2341 | * to non delalloc when we are near to error range. | |
2342 | */ | |
57042651 TT |
2343 | free_blocks = EXT4_C2B(sbi, |
2344 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); | |
2345 | dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
79f0be8d | 2346 | if (2 * free_blocks < 3 * dirty_blocks || |
df55c99d | 2347 | free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) { |
79f0be8d | 2348 | /* |
c8afb446 ES |
2349 | * free block count is less than 150% of dirty blocks |
2350 | * or free blocks is less than watermark | |
79f0be8d AK |
2351 | */ |
2352 | return 1; | |
2353 | } | |
c8afb446 ES |
2354 | /* |
2355 | * Even if we don't switch but are nearing capacity, | |
2356 | * start pushing delalloc when 1/2 of free blocks are dirty. | |
2357 | */ | |
2358 | if (free_blocks < 2 * dirty_blocks) | |
2359 | writeback_inodes_sb_if_idle(sb); | |
2360 | ||
79f0be8d AK |
2361 | return 0; |
2362 | } | |
2363 | ||
64769240 | 2364 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2365 | loff_t pos, unsigned len, unsigned flags, |
2366 | struct page **pagep, void **fsdata) | |
64769240 | 2367 | { |
72b8ab9d | 2368 | int ret, retries = 0; |
64769240 AT |
2369 | struct page *page; |
2370 | pgoff_t index; | |
64769240 AT |
2371 | struct inode *inode = mapping->host; |
2372 | handle_t *handle; | |
02fac129 | 2373 | loff_t page_len; |
64769240 AT |
2374 | |
2375 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
2376 | |
2377 | if (ext4_nonda_switch(inode->i_sb)) { | |
2378 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2379 | return ext4_write_begin(file, mapping, pos, | |
2380 | len, flags, pagep, fsdata); | |
2381 | } | |
2382 | *fsdata = (void *)0; | |
9bffad1e | 2383 | trace_ext4_da_write_begin(inode, pos, len, flags); |
d2a17637 | 2384 | retry: |
64769240 AT |
2385 | /* |
2386 | * With delayed allocation, we don't log the i_disksize update | |
2387 | * if there is delayed block allocation. But we still need | |
2388 | * to journalling the i_disksize update if writes to the end | |
2389 | * of file which has an already mapped buffer. | |
2390 | */ | |
2391 | handle = ext4_journal_start(inode, 1); | |
2392 | if (IS_ERR(handle)) { | |
2393 | ret = PTR_ERR(handle); | |
2394 | goto out; | |
2395 | } | |
ebd3610b JK |
2396 | /* We cannot recurse into the filesystem as the transaction is already |
2397 | * started */ | |
2398 | flags |= AOP_FLAG_NOFS; | |
64769240 | 2399 | |
54566b2c | 2400 | page = grab_cache_page_write_begin(mapping, index, flags); |
d5a0d4f7 ES |
2401 | if (!page) { |
2402 | ext4_journal_stop(handle); | |
2403 | ret = -ENOMEM; | |
2404 | goto out; | |
2405 | } | |
64769240 AT |
2406 | *pagep = page; |
2407 | ||
6e1db88d | 2408 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
64769240 AT |
2409 | if (ret < 0) { |
2410 | unlock_page(page); | |
2411 | ext4_journal_stop(handle); | |
2412 | page_cache_release(page); | |
ae4d5372 AK |
2413 | /* |
2414 | * block_write_begin may have instantiated a few blocks | |
2415 | * outside i_size. Trim these off again. Don't need | |
2416 | * i_size_read because we hold i_mutex. | |
2417 | */ | |
2418 | if (pos + len > inode->i_size) | |
b9a4207d | 2419 | ext4_truncate_failed_write(inode); |
02fac129 AH |
2420 | } else { |
2421 | page_len = pos & (PAGE_CACHE_SIZE - 1); | |
2422 | if (page_len > 0) { | |
2423 | ret = ext4_discard_partial_page_buffers_no_lock(handle, | |
2424 | inode, page, pos - page_len, page_len, | |
2425 | EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED); | |
2426 | } | |
64769240 AT |
2427 | } |
2428 | ||
d2a17637 MC |
2429 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
2430 | goto retry; | |
64769240 AT |
2431 | out: |
2432 | return ret; | |
2433 | } | |
2434 | ||
632eaeab MC |
2435 | /* |
2436 | * Check if we should update i_disksize | |
2437 | * when write to the end of file but not require block allocation | |
2438 | */ | |
2439 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2440 | unsigned long offset) |
632eaeab MC |
2441 | { |
2442 | struct buffer_head *bh; | |
2443 | struct inode *inode = page->mapping->host; | |
2444 | unsigned int idx; | |
2445 | int i; | |
2446 | ||
2447 | bh = page_buffers(page); | |
2448 | idx = offset >> inode->i_blkbits; | |
2449 | ||
af5bc92d | 2450 | for (i = 0; i < idx; i++) |
632eaeab MC |
2451 | bh = bh->b_this_page; |
2452 | ||
29fa89d0 | 2453 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2454 | return 0; |
2455 | return 1; | |
2456 | } | |
2457 | ||
64769240 | 2458 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2459 | struct address_space *mapping, |
2460 | loff_t pos, unsigned len, unsigned copied, | |
2461 | struct page *page, void *fsdata) | |
64769240 AT |
2462 | { |
2463 | struct inode *inode = mapping->host; | |
2464 | int ret = 0, ret2; | |
2465 | handle_t *handle = ext4_journal_current_handle(); | |
2466 | loff_t new_i_size; | |
632eaeab | 2467 | unsigned long start, end; |
79f0be8d | 2468 | int write_mode = (int)(unsigned long)fsdata; |
02fac129 | 2469 | loff_t page_len; |
79f0be8d AK |
2470 | |
2471 | if (write_mode == FALL_BACK_TO_NONDELALLOC) { | |
2472 | if (ext4_should_order_data(inode)) { | |
2473 | return ext4_ordered_write_end(file, mapping, pos, | |
2474 | len, copied, page, fsdata); | |
2475 | } else if (ext4_should_writeback_data(inode)) { | |
2476 | return ext4_writeback_write_end(file, mapping, pos, | |
2477 | len, copied, page, fsdata); | |
2478 | } else { | |
2479 | BUG(); | |
2480 | } | |
2481 | } | |
632eaeab | 2482 | |
9bffad1e | 2483 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 2484 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2485 | end = start + copied - 1; |
64769240 AT |
2486 | |
2487 | /* | |
2488 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2489 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2490 | * into that. | |
2491 | */ | |
2492 | ||
2493 | new_i_size = pos + copied; | |
632eaeab MC |
2494 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
2495 | if (ext4_da_should_update_i_disksize(page, end)) { | |
2496 | down_write(&EXT4_I(inode)->i_data_sem); | |
2497 | if (new_i_size > EXT4_I(inode)->i_disksize) { | |
2498 | /* | |
2499 | * Updating i_disksize when extending file | |
2500 | * without needing block allocation | |
2501 | */ | |
2502 | if (ext4_should_order_data(inode)) | |
2503 | ret = ext4_jbd2_file_inode(handle, | |
2504 | inode); | |
64769240 | 2505 | |
632eaeab MC |
2506 | EXT4_I(inode)->i_disksize = new_i_size; |
2507 | } | |
2508 | up_write(&EXT4_I(inode)->i_data_sem); | |
cf17fea6 AK |
2509 | /* We need to mark inode dirty even if |
2510 | * new_i_size is less that inode->i_size | |
2511 | * bu greater than i_disksize.(hint delalloc) | |
2512 | */ | |
2513 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2514 | } |
632eaeab | 2515 | } |
64769240 AT |
2516 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
2517 | page, fsdata); | |
02fac129 AH |
2518 | |
2519 | page_len = PAGE_CACHE_SIZE - | |
2520 | ((pos + copied - 1) & (PAGE_CACHE_SIZE - 1)); | |
2521 | ||
2522 | if (page_len > 0) { | |
2523 | ret = ext4_discard_partial_page_buffers_no_lock(handle, | |
2524 | inode, page, pos + copied - 1, page_len, | |
2525 | EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED); | |
2526 | } | |
2527 | ||
64769240 AT |
2528 | copied = ret2; |
2529 | if (ret2 < 0) | |
2530 | ret = ret2; | |
2531 | ret2 = ext4_journal_stop(handle); | |
2532 | if (!ret) | |
2533 | ret = ret2; | |
2534 | ||
2535 | return ret ? ret : copied; | |
2536 | } | |
2537 | ||
2538 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) | |
2539 | { | |
64769240 AT |
2540 | /* |
2541 | * Drop reserved blocks | |
2542 | */ | |
2543 | BUG_ON(!PageLocked(page)); | |
2544 | if (!page_has_buffers(page)) | |
2545 | goto out; | |
2546 | ||
d2a17637 | 2547 | ext4_da_page_release_reservation(page, offset); |
64769240 AT |
2548 | |
2549 | out: | |
2550 | ext4_invalidatepage(page, offset); | |
2551 | ||
2552 | return; | |
2553 | } | |
2554 | ||
ccd2506b TT |
2555 | /* |
2556 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2557 | */ | |
2558 | int ext4_alloc_da_blocks(struct inode *inode) | |
2559 | { | |
fb40ba0d TT |
2560 | trace_ext4_alloc_da_blocks(inode); |
2561 | ||
ccd2506b TT |
2562 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
2563 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
2564 | return 0; | |
2565 | ||
2566 | /* | |
2567 | * We do something simple for now. The filemap_flush() will | |
2568 | * also start triggering a write of the data blocks, which is | |
2569 | * not strictly speaking necessary (and for users of | |
2570 | * laptop_mode, not even desirable). However, to do otherwise | |
2571 | * would require replicating code paths in: | |
de9a55b8 | 2572 | * |
ccd2506b TT |
2573 | * ext4_da_writepages() -> |
2574 | * write_cache_pages() ---> (via passed in callback function) | |
2575 | * __mpage_da_writepage() --> | |
2576 | * mpage_add_bh_to_extent() | |
2577 | * mpage_da_map_blocks() | |
2578 | * | |
2579 | * The problem is that write_cache_pages(), located in | |
2580 | * mm/page-writeback.c, marks pages clean in preparation for | |
2581 | * doing I/O, which is not desirable if we're not planning on | |
2582 | * doing I/O at all. | |
2583 | * | |
2584 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 2585 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
2586 | * would be ugly in the extreme. So instead we would need to |
2587 | * replicate parts of the code in the above functions, | |
25985edc | 2588 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
2589 | * write out the pages, but rather only collect contiguous |
2590 | * logical block extents, call the multi-block allocator, and | |
2591 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 2592 | * |
ccd2506b TT |
2593 | * For now, though, we'll cheat by calling filemap_flush(), |
2594 | * which will map the blocks, and start the I/O, but not | |
2595 | * actually wait for the I/O to complete. | |
2596 | */ | |
2597 | return filemap_flush(inode->i_mapping); | |
2598 | } | |
64769240 | 2599 | |
ac27a0ec DK |
2600 | /* |
2601 | * bmap() is special. It gets used by applications such as lilo and by | |
2602 | * the swapper to find the on-disk block of a specific piece of data. | |
2603 | * | |
2604 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 2605 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
2606 | * filesystem and enables swap, then they may get a nasty shock when the |
2607 | * data getting swapped to that swapfile suddenly gets overwritten by | |
2608 | * the original zero's written out previously to the journal and | |
2609 | * awaiting writeback in the kernel's buffer cache. | |
2610 | * | |
2611 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
2612 | * take extra steps to flush any blocks which might be in the cache. | |
2613 | */ | |
617ba13b | 2614 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
2615 | { |
2616 | struct inode *inode = mapping->host; | |
2617 | journal_t *journal; | |
2618 | int err; | |
2619 | ||
64769240 AT |
2620 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
2621 | test_opt(inode->i_sb, DELALLOC)) { | |
2622 | /* | |
2623 | * With delalloc we want to sync the file | |
2624 | * so that we can make sure we allocate | |
2625 | * blocks for file | |
2626 | */ | |
2627 | filemap_write_and_wait(mapping); | |
2628 | } | |
2629 | ||
19f5fb7a TT |
2630 | if (EXT4_JOURNAL(inode) && |
2631 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
2632 | /* |
2633 | * This is a REALLY heavyweight approach, but the use of | |
2634 | * bmap on dirty files is expected to be extremely rare: | |
2635 | * only if we run lilo or swapon on a freshly made file | |
2636 | * do we expect this to happen. | |
2637 | * | |
2638 | * (bmap requires CAP_SYS_RAWIO so this does not | |
2639 | * represent an unprivileged user DOS attack --- we'd be | |
2640 | * in trouble if mortal users could trigger this path at | |
2641 | * will.) | |
2642 | * | |
617ba13b | 2643 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
2644 | * regular files. If somebody wants to bmap a directory |
2645 | * or symlink and gets confused because the buffer | |
2646 | * hasn't yet been flushed to disk, they deserve | |
2647 | * everything they get. | |
2648 | */ | |
2649 | ||
19f5fb7a | 2650 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 2651 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
2652 | jbd2_journal_lock_updates(journal); |
2653 | err = jbd2_journal_flush(journal); | |
2654 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
2655 | |
2656 | if (err) | |
2657 | return 0; | |
2658 | } | |
2659 | ||
af5bc92d | 2660 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
2661 | } |
2662 | ||
617ba13b | 2663 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 2664 | { |
0562e0ba | 2665 | trace_ext4_readpage(page); |
617ba13b | 2666 | return mpage_readpage(page, ext4_get_block); |
ac27a0ec DK |
2667 | } |
2668 | ||
2669 | static int | |
617ba13b | 2670 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
2671 | struct list_head *pages, unsigned nr_pages) |
2672 | { | |
617ba13b | 2673 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
2674 | } |
2675 | ||
744692dc JZ |
2676 | static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset) |
2677 | { | |
2678 | struct buffer_head *head, *bh; | |
2679 | unsigned int curr_off = 0; | |
2680 | ||
2681 | if (!page_has_buffers(page)) | |
2682 | return; | |
2683 | head = bh = page_buffers(page); | |
2684 | do { | |
2685 | if (offset <= curr_off && test_clear_buffer_uninit(bh) | |
2686 | && bh->b_private) { | |
2687 | ext4_free_io_end(bh->b_private); | |
2688 | bh->b_private = NULL; | |
2689 | bh->b_end_io = NULL; | |
2690 | } | |
2691 | curr_off = curr_off + bh->b_size; | |
2692 | bh = bh->b_this_page; | |
2693 | } while (bh != head); | |
2694 | } | |
2695 | ||
617ba13b | 2696 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 2697 | { |
617ba13b | 2698 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 2699 | |
0562e0ba JZ |
2700 | trace_ext4_invalidatepage(page, offset); |
2701 | ||
744692dc JZ |
2702 | /* |
2703 | * free any io_end structure allocated for buffers to be discarded | |
2704 | */ | |
2705 | if (ext4_should_dioread_nolock(page->mapping->host)) | |
2706 | ext4_invalidatepage_free_endio(page, offset); | |
ac27a0ec DK |
2707 | /* |
2708 | * If it's a full truncate we just forget about the pending dirtying | |
2709 | */ | |
2710 | if (offset == 0) | |
2711 | ClearPageChecked(page); | |
2712 | ||
0390131b FM |
2713 | if (journal) |
2714 | jbd2_journal_invalidatepage(journal, page, offset); | |
2715 | else | |
2716 | block_invalidatepage(page, offset); | |
ac27a0ec DK |
2717 | } |
2718 | ||
617ba13b | 2719 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 2720 | { |
617ba13b | 2721 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 2722 | |
0562e0ba JZ |
2723 | trace_ext4_releasepage(page); |
2724 | ||
ac27a0ec DK |
2725 | WARN_ON(PageChecked(page)); |
2726 | if (!page_has_buffers(page)) | |
2727 | return 0; | |
0390131b FM |
2728 | if (journal) |
2729 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
2730 | else | |
2731 | return try_to_free_buffers(page); | |
ac27a0ec DK |
2732 | } |
2733 | ||
2ed88685 TT |
2734 | /* |
2735 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
2736 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
2737 | * The extent will be converted to initialized after the IO is complete. | |
2738 | */ | |
c7064ef1 | 2739 | static int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
2740 | struct buffer_head *bh_result, int create) |
2741 | { | |
c7064ef1 | 2742 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 2743 | inode->i_ino, create); |
2ed88685 TT |
2744 | return _ext4_get_block(inode, iblock, bh_result, |
2745 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
2746 | } |
2747 | ||
4c0425ff | 2748 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
552ef802 CH |
2749 | ssize_t size, void *private, int ret, |
2750 | bool is_async) | |
4c0425ff | 2751 | { |
72c5052d | 2752 | struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode; |
4c0425ff MC |
2753 | ext4_io_end_t *io_end = iocb->private; |
2754 | struct workqueue_struct *wq; | |
744692dc JZ |
2755 | unsigned long flags; |
2756 | struct ext4_inode_info *ei; | |
4c0425ff | 2757 | |
4b70df18 M |
2758 | /* if not async direct IO or dio with 0 bytes write, just return */ |
2759 | if (!io_end || !size) | |
552ef802 | 2760 | goto out; |
4b70df18 | 2761 | |
8d5d02e6 MC |
2762 | ext_debug("ext4_end_io_dio(): io_end 0x%p" |
2763 | "for inode %lu, iocb 0x%p, offset %llu, size %llu\n", | |
2764 | iocb->private, io_end->inode->i_ino, iocb, offset, | |
2765 | size); | |
8d5d02e6 MC |
2766 | |
2767 | /* if not aio dio with unwritten extents, just free io and return */ | |
bd2d0210 | 2768 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { |
8d5d02e6 MC |
2769 | ext4_free_io_end(io_end); |
2770 | iocb->private = NULL; | |
5b3ff237 JZ |
2771 | out: |
2772 | if (is_async) | |
2773 | aio_complete(iocb, ret, 0); | |
72c5052d | 2774 | inode_dio_done(inode); |
5b3ff237 | 2775 | return; |
8d5d02e6 MC |
2776 | } |
2777 | ||
4c0425ff MC |
2778 | io_end->offset = offset; |
2779 | io_end->size = size; | |
5b3ff237 JZ |
2780 | if (is_async) { |
2781 | io_end->iocb = iocb; | |
2782 | io_end->result = ret; | |
2783 | } | |
4c0425ff MC |
2784 | wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq; |
2785 | ||
8d5d02e6 | 2786 | /* Add the io_end to per-inode completed aio dio list*/ |
744692dc JZ |
2787 | ei = EXT4_I(io_end->inode); |
2788 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); | |
2789 | list_add_tail(&io_end->list, &ei->i_completed_io_list); | |
2790 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); | |
c999af2b ES |
2791 | |
2792 | /* queue the work to convert unwritten extents to written */ | |
2793 | queue_work(wq, &io_end->work); | |
4c0425ff | 2794 | iocb->private = NULL; |
72c5052d CH |
2795 | |
2796 | /* XXX: probably should move into the real I/O completion handler */ | |
2797 | inode_dio_done(inode); | |
4c0425ff | 2798 | } |
c7064ef1 | 2799 | |
744692dc JZ |
2800 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate) |
2801 | { | |
2802 | ext4_io_end_t *io_end = bh->b_private; | |
2803 | struct workqueue_struct *wq; | |
2804 | struct inode *inode; | |
2805 | unsigned long flags; | |
2806 | ||
2807 | if (!test_clear_buffer_uninit(bh) || !io_end) | |
2808 | goto out; | |
2809 | ||
2810 | if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) { | |
2811 | printk("sb umounted, discard end_io request for inode %lu\n", | |
2812 | io_end->inode->i_ino); | |
2813 | ext4_free_io_end(io_end); | |
2814 | goto out; | |
2815 | } | |
2816 | ||
32c80b32 TM |
2817 | /* |
2818 | * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now, | |
2819 | * but being more careful is always safe for the future change. | |
2820 | */ | |
744692dc | 2821 | inode = io_end->inode; |
32c80b32 TM |
2822 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { |
2823 | io_end->flag |= EXT4_IO_END_UNWRITTEN; | |
2824 | atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten); | |
2825 | } | |
744692dc JZ |
2826 | |
2827 | /* Add the io_end to per-inode completed io list*/ | |
2828 | spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags); | |
2829 | list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list); | |
2830 | spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags); | |
2831 | ||
2832 | wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq; | |
2833 | /* queue the work to convert unwritten extents to written */ | |
2834 | queue_work(wq, &io_end->work); | |
2835 | out: | |
2836 | bh->b_private = NULL; | |
2837 | bh->b_end_io = NULL; | |
2838 | clear_buffer_uninit(bh); | |
2839 | end_buffer_async_write(bh, uptodate); | |
2840 | } | |
2841 | ||
2842 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode) | |
2843 | { | |
2844 | ext4_io_end_t *io_end; | |
2845 | struct page *page = bh->b_page; | |
2846 | loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT; | |
2847 | size_t size = bh->b_size; | |
2848 | ||
2849 | retry: | |
2850 | io_end = ext4_init_io_end(inode, GFP_ATOMIC); | |
2851 | if (!io_end) { | |
6db26ffc | 2852 | pr_warn_ratelimited("%s: allocation fail\n", __func__); |
744692dc JZ |
2853 | schedule(); |
2854 | goto retry; | |
2855 | } | |
2856 | io_end->offset = offset; | |
2857 | io_end->size = size; | |
2858 | /* | |
2859 | * We need to hold a reference to the page to make sure it | |
2860 | * doesn't get evicted before ext4_end_io_work() has a chance | |
2861 | * to convert the extent from written to unwritten. | |
2862 | */ | |
2863 | io_end->page = page; | |
2864 | get_page(io_end->page); | |
2865 | ||
2866 | bh->b_private = io_end; | |
2867 | bh->b_end_io = ext4_end_io_buffer_write; | |
2868 | return 0; | |
2869 | } | |
2870 | ||
4c0425ff MC |
2871 | /* |
2872 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
2873 | * preallocated extents, and those write extend the file, no need to | |
2874 | * fall back to buffered IO. | |
2875 | * | |
b595076a | 2876 | * For holes, we fallocate those blocks, mark them as uninitialized |
4c0425ff | 2877 | * If those blocks were preallocated, we mark sure they are splited, but |
b595076a | 2878 | * still keep the range to write as uninitialized. |
4c0425ff | 2879 | * |
8d5d02e6 MC |
2880 | * The unwrritten extents will be converted to written when DIO is completed. |
2881 | * For async direct IO, since the IO may still pending when return, we | |
25985edc | 2882 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 2883 | * when async direct IO completed. |
4c0425ff MC |
2884 | * |
2885 | * If the O_DIRECT write will extend the file then add this inode to the | |
2886 | * orphan list. So recovery will truncate it back to the original size | |
2887 | * if the machine crashes during the write. | |
2888 | * | |
2889 | */ | |
2890 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, | |
2891 | const struct iovec *iov, loff_t offset, | |
2892 | unsigned long nr_segs) | |
2893 | { | |
2894 | struct file *file = iocb->ki_filp; | |
2895 | struct inode *inode = file->f_mapping->host; | |
2896 | ssize_t ret; | |
2897 | size_t count = iov_length(iov, nr_segs); | |
2898 | ||
2899 | loff_t final_size = offset + count; | |
2900 | if (rw == WRITE && final_size <= inode->i_size) { | |
2901 | /* | |
8d5d02e6 MC |
2902 | * We could direct write to holes and fallocate. |
2903 | * | |
2904 | * Allocated blocks to fill the hole are marked as uninitialized | |
25985edc | 2905 | * to prevent parallel buffered read to expose the stale data |
4c0425ff | 2906 | * before DIO complete the data IO. |
8d5d02e6 MC |
2907 | * |
2908 | * As to previously fallocated extents, ext4 get_block | |
4c0425ff MC |
2909 | * will just simply mark the buffer mapped but still |
2910 | * keep the extents uninitialized. | |
2911 | * | |
8d5d02e6 MC |
2912 | * for non AIO case, we will convert those unwritten extents |
2913 | * to written after return back from blockdev_direct_IO. | |
2914 | * | |
2915 | * for async DIO, the conversion needs to be defered when | |
2916 | * the IO is completed. The ext4 end_io callback function | |
2917 | * will be called to take care of the conversion work. | |
2918 | * Here for async case, we allocate an io_end structure to | |
2919 | * hook to the iocb. | |
4c0425ff | 2920 | */ |
8d5d02e6 MC |
2921 | iocb->private = NULL; |
2922 | EXT4_I(inode)->cur_aio_dio = NULL; | |
2923 | if (!is_sync_kiocb(iocb)) { | |
744692dc | 2924 | iocb->private = ext4_init_io_end(inode, GFP_NOFS); |
8d5d02e6 MC |
2925 | if (!iocb->private) |
2926 | return -ENOMEM; | |
2927 | /* | |
2928 | * we save the io structure for current async | |
79e83036 | 2929 | * direct IO, so that later ext4_map_blocks() |
8d5d02e6 MC |
2930 | * could flag the io structure whether there |
2931 | * is a unwritten extents needs to be converted | |
2932 | * when IO is completed. | |
2933 | */ | |
2934 | EXT4_I(inode)->cur_aio_dio = iocb->private; | |
2935 | } | |
2936 | ||
aacfc19c | 2937 | ret = __blockdev_direct_IO(rw, iocb, inode, |
4c0425ff MC |
2938 | inode->i_sb->s_bdev, iov, |
2939 | offset, nr_segs, | |
c7064ef1 | 2940 | ext4_get_block_write, |
aacfc19c CH |
2941 | ext4_end_io_dio, |
2942 | NULL, | |
2943 | DIO_LOCKING | DIO_SKIP_HOLES); | |
8d5d02e6 MC |
2944 | if (iocb->private) |
2945 | EXT4_I(inode)->cur_aio_dio = NULL; | |
2946 | /* | |
2947 | * The io_end structure takes a reference to the inode, | |
2948 | * that structure needs to be destroyed and the | |
2949 | * reference to the inode need to be dropped, when IO is | |
2950 | * complete, even with 0 byte write, or failed. | |
2951 | * | |
2952 | * In the successful AIO DIO case, the io_end structure will be | |
2953 | * desctroyed and the reference to the inode will be dropped | |
2954 | * after the end_io call back function is called. | |
2955 | * | |
2956 | * In the case there is 0 byte write, or error case, since | |
2957 | * VFS direct IO won't invoke the end_io call back function, | |
2958 | * we need to free the end_io structure here. | |
2959 | */ | |
2960 | if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) { | |
2961 | ext4_free_io_end(iocb->private); | |
2962 | iocb->private = NULL; | |
19f5fb7a TT |
2963 | } else if (ret > 0 && ext4_test_inode_state(inode, |
2964 | EXT4_STATE_DIO_UNWRITTEN)) { | |
109f5565 | 2965 | int err; |
8d5d02e6 MC |
2966 | /* |
2967 | * for non AIO case, since the IO is already | |
25985edc | 2968 | * completed, we could do the conversion right here |
8d5d02e6 | 2969 | */ |
109f5565 M |
2970 | err = ext4_convert_unwritten_extents(inode, |
2971 | offset, ret); | |
2972 | if (err < 0) | |
2973 | ret = err; | |
19f5fb7a | 2974 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
109f5565 | 2975 | } |
4c0425ff MC |
2976 | return ret; |
2977 | } | |
8d5d02e6 MC |
2978 | |
2979 | /* for write the the end of file case, we fall back to old way */ | |
4c0425ff MC |
2980 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); |
2981 | } | |
2982 | ||
2983 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |
2984 | const struct iovec *iov, loff_t offset, | |
2985 | unsigned long nr_segs) | |
2986 | { | |
2987 | struct file *file = iocb->ki_filp; | |
2988 | struct inode *inode = file->f_mapping->host; | |
0562e0ba | 2989 | ssize_t ret; |
4c0425ff | 2990 | |
84ebd795 TT |
2991 | /* |
2992 | * If we are doing data journalling we don't support O_DIRECT | |
2993 | */ | |
2994 | if (ext4_should_journal_data(inode)) | |
2995 | return 0; | |
2996 | ||
0562e0ba | 2997 | trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); |
12e9b892 | 2998 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
0562e0ba JZ |
2999 | ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); |
3000 | else | |
3001 | ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
3002 | trace_ext4_direct_IO_exit(inode, offset, | |
3003 | iov_length(iov, nr_segs), rw, ret); | |
3004 | return ret; | |
4c0425ff MC |
3005 | } |
3006 | ||
ac27a0ec | 3007 | /* |
617ba13b | 3008 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3009 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3010 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3011 | * not necessarily locked. | |
3012 | * | |
3013 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3014 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3015 | * or jbddirty because all the journalling code will explode. | |
3016 | * | |
3017 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3018 | * is called, propagate that into the buffers appropriately. | |
3019 | */ | |
617ba13b | 3020 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3021 | { |
3022 | SetPageChecked(page); | |
3023 | return __set_page_dirty_nobuffers(page); | |
3024 | } | |
3025 | ||
617ba13b | 3026 | static const struct address_space_operations ext4_ordered_aops = { |
8ab22b9a HH |
3027 | .readpage = ext4_readpage, |
3028 | .readpages = ext4_readpages, | |
43ce1d23 | 3029 | .writepage = ext4_writepage, |
8ab22b9a HH |
3030 | .write_begin = ext4_write_begin, |
3031 | .write_end = ext4_ordered_write_end, | |
3032 | .bmap = ext4_bmap, | |
3033 | .invalidatepage = ext4_invalidatepage, | |
3034 | .releasepage = ext4_releasepage, | |
3035 | .direct_IO = ext4_direct_IO, | |
3036 | .migratepage = buffer_migrate_page, | |
3037 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3038 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3039 | }; |
3040 | ||
617ba13b | 3041 | static const struct address_space_operations ext4_writeback_aops = { |
8ab22b9a HH |
3042 | .readpage = ext4_readpage, |
3043 | .readpages = ext4_readpages, | |
43ce1d23 | 3044 | .writepage = ext4_writepage, |
8ab22b9a HH |
3045 | .write_begin = ext4_write_begin, |
3046 | .write_end = ext4_writeback_write_end, | |
3047 | .bmap = ext4_bmap, | |
3048 | .invalidatepage = ext4_invalidatepage, | |
3049 | .releasepage = ext4_releasepage, | |
3050 | .direct_IO = ext4_direct_IO, | |
3051 | .migratepage = buffer_migrate_page, | |
3052 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3053 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3054 | }; |
3055 | ||
617ba13b | 3056 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3057 | .readpage = ext4_readpage, |
3058 | .readpages = ext4_readpages, | |
43ce1d23 | 3059 | .writepage = ext4_writepage, |
8ab22b9a HH |
3060 | .write_begin = ext4_write_begin, |
3061 | .write_end = ext4_journalled_write_end, | |
3062 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3063 | .bmap = ext4_bmap, | |
3064 | .invalidatepage = ext4_invalidatepage, | |
3065 | .releasepage = ext4_releasepage, | |
84ebd795 | 3066 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3067 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3068 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3069 | }; |
3070 | ||
64769240 | 3071 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3072 | .readpage = ext4_readpage, |
3073 | .readpages = ext4_readpages, | |
43ce1d23 | 3074 | .writepage = ext4_writepage, |
8ab22b9a | 3075 | .writepages = ext4_da_writepages, |
8ab22b9a HH |
3076 | .write_begin = ext4_da_write_begin, |
3077 | .write_end = ext4_da_write_end, | |
3078 | .bmap = ext4_bmap, | |
3079 | .invalidatepage = ext4_da_invalidatepage, | |
3080 | .releasepage = ext4_releasepage, | |
3081 | .direct_IO = ext4_direct_IO, | |
3082 | .migratepage = buffer_migrate_page, | |
3083 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3084 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3085 | }; |
3086 | ||
617ba13b | 3087 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3088 | { |
cd1aac32 AK |
3089 | if (ext4_should_order_data(inode) && |
3090 | test_opt(inode->i_sb, DELALLOC)) | |
3091 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3092 | else if (ext4_should_order_data(inode)) | |
617ba13b | 3093 | inode->i_mapping->a_ops = &ext4_ordered_aops; |
64769240 AT |
3094 | else if (ext4_should_writeback_data(inode) && |
3095 | test_opt(inode->i_sb, DELALLOC)) | |
3096 | inode->i_mapping->a_ops = &ext4_da_aops; | |
617ba13b MC |
3097 | else if (ext4_should_writeback_data(inode)) |
3098 | inode->i_mapping->a_ops = &ext4_writeback_aops; | |
ac27a0ec | 3099 | else |
617ba13b | 3100 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
ac27a0ec DK |
3101 | } |
3102 | ||
4e96b2db AH |
3103 | |
3104 | /* | |
3105 | * ext4_discard_partial_page_buffers() | |
3106 | * Wrapper function for ext4_discard_partial_page_buffers_no_lock. | |
3107 | * This function finds and locks the page containing the offset | |
3108 | * "from" and passes it to ext4_discard_partial_page_buffers_no_lock. | |
3109 | * Calling functions that already have the page locked should call | |
3110 | * ext4_discard_partial_page_buffers_no_lock directly. | |
3111 | */ | |
3112 | int ext4_discard_partial_page_buffers(handle_t *handle, | |
3113 | struct address_space *mapping, loff_t from, | |
3114 | loff_t length, int flags) | |
3115 | { | |
3116 | struct inode *inode = mapping->host; | |
3117 | struct page *page; | |
3118 | int err = 0; | |
3119 | ||
3120 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, | |
3121 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
3122 | if (!page) | |
3123 | return -EINVAL; | |
3124 | ||
3125 | err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page, | |
3126 | from, length, flags); | |
3127 | ||
3128 | unlock_page(page); | |
3129 | page_cache_release(page); | |
3130 | return err; | |
3131 | } | |
3132 | ||
3133 | /* | |
3134 | * ext4_discard_partial_page_buffers_no_lock() | |
3135 | * Zeros a page range of length 'length' starting from offset 'from'. | |
3136 | * Buffer heads that correspond to the block aligned regions of the | |
3137 | * zeroed range will be unmapped. Unblock aligned regions | |
3138 | * will have the corresponding buffer head mapped if needed so that | |
3139 | * that region of the page can be updated with the partial zero out. | |
3140 | * | |
3141 | * This function assumes that the page has already been locked. The | |
3142 | * The range to be discarded must be contained with in the given page. | |
3143 | * If the specified range exceeds the end of the page it will be shortened | |
3144 | * to the end of the page that corresponds to 'from'. This function is | |
3145 | * appropriate for updating a page and it buffer heads to be unmapped and | |
3146 | * zeroed for blocks that have been either released, or are going to be | |
3147 | * released. | |
3148 | * | |
3149 | * handle: The journal handle | |
3150 | * inode: The files inode | |
3151 | * page: A locked page that contains the offset "from" | |
3152 | * from: The starting byte offset (from the begining of the file) | |
3153 | * to begin discarding | |
3154 | * len: The length of bytes to discard | |
3155 | * flags: Optional flags that may be used: | |
3156 | * | |
3157 | * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED | |
3158 | * Only zero the regions of the page whose buffer heads | |
3159 | * have already been unmapped. This flag is appropriate | |
3160 | * for updateing the contents of a page whose blocks may | |
3161 | * have already been released, and we only want to zero | |
3162 | * out the regions that correspond to those released blocks. | |
3163 | * | |
3164 | * Returns zero on sucess or negative on failure. | |
3165 | */ | |
3166 | int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, | |
3167 | struct inode *inode, struct page *page, loff_t from, | |
3168 | loff_t length, int flags) | |
3169 | { | |
3170 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; | |
3171 | unsigned int offset = from & (PAGE_CACHE_SIZE-1); | |
3172 | unsigned int blocksize, max, pos; | |
3173 | unsigned int end_of_block, range_to_discard; | |
3174 | ext4_lblk_t iblock; | |
3175 | struct buffer_head *bh; | |
3176 | int err = 0; | |
3177 | ||
3178 | blocksize = inode->i_sb->s_blocksize; | |
3179 | max = PAGE_CACHE_SIZE - offset; | |
3180 | ||
3181 | if (index != page->index) | |
3182 | return -EINVAL; | |
3183 | ||
3184 | /* | |
3185 | * correct length if it does not fall between | |
3186 | * 'from' and the end of the page | |
3187 | */ | |
3188 | if (length > max || length < 0) | |
3189 | length = max; | |
3190 | ||
3191 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3192 | ||
3193 | if (!page_has_buffers(page)) { | |
3194 | /* | |
3195 | * If the range to be discarded covers a partial block | |
3196 | * we need to get the page buffers. This is because | |
3197 | * partial blocks cannot be released and the page needs | |
3198 | * to be updated with the contents of the block before | |
3199 | * we write the zeros on top of it. | |
3200 | */ | |
3201 | if (!(from & (blocksize - 1)) || | |
3202 | !((from + length) & (blocksize - 1))) { | |
3203 | create_empty_buffers(page, blocksize, 0); | |
3204 | } else { | |
3205 | /* | |
3206 | * If there are no partial blocks, | |
3207 | * there is nothing to update, | |
3208 | * so we can return now | |
3209 | */ | |
3210 | return 0; | |
3211 | } | |
3212 | } | |
3213 | ||
3214 | /* Find the buffer that contains "offset" */ | |
3215 | bh = page_buffers(page); | |
3216 | pos = blocksize; | |
3217 | while (offset >= pos) { | |
3218 | bh = bh->b_this_page; | |
3219 | iblock++; | |
3220 | pos += blocksize; | |
3221 | } | |
3222 | ||
3223 | pos = offset; | |
3224 | while (pos < offset + length) { | |
3225 | err = 0; | |
3226 | ||
3227 | /* The length of space left to zero and unmap */ | |
3228 | range_to_discard = offset + length - pos; | |
3229 | ||
3230 | /* The length of space until the end of the block */ | |
3231 | end_of_block = blocksize - (pos & (blocksize-1)); | |
3232 | ||
3233 | /* | |
3234 | * Do not unmap or zero past end of block | |
3235 | * for this buffer head | |
3236 | */ | |
3237 | if (range_to_discard > end_of_block) | |
3238 | range_to_discard = end_of_block; | |
3239 | ||
3240 | ||
3241 | /* | |
3242 | * Skip this buffer head if we are only zeroing unampped | |
3243 | * regions of the page | |
3244 | */ | |
3245 | if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED && | |
3246 | buffer_mapped(bh)) | |
3247 | goto next; | |
3248 | ||
3249 | /* If the range is block aligned, unmap */ | |
3250 | if (range_to_discard == blocksize) { | |
3251 | clear_buffer_dirty(bh); | |
3252 | bh->b_bdev = NULL; | |
3253 | clear_buffer_mapped(bh); | |
3254 | clear_buffer_req(bh); | |
3255 | clear_buffer_new(bh); | |
3256 | clear_buffer_delay(bh); | |
3257 | clear_buffer_unwritten(bh); | |
3258 | clear_buffer_uptodate(bh); | |
3259 | zero_user(page, pos, range_to_discard); | |
3260 | BUFFER_TRACE(bh, "Buffer discarded"); | |
3261 | goto next; | |
3262 | } | |
3263 | ||
3264 | /* | |
3265 | * If this block is not completely contained in the range | |
3266 | * to be discarded, then it is not going to be released. Because | |
3267 | * we need to keep this block, we need to make sure this part | |
3268 | * of the page is uptodate before we modify it by writeing | |
3269 | * partial zeros on it. | |
3270 | */ | |
3271 | if (!buffer_mapped(bh)) { | |
3272 | /* | |
3273 | * Buffer head must be mapped before we can read | |
3274 | * from the block | |
3275 | */ | |
3276 | BUFFER_TRACE(bh, "unmapped"); | |
3277 | ext4_get_block(inode, iblock, bh, 0); | |
3278 | /* unmapped? It's a hole - nothing to do */ | |
3279 | if (!buffer_mapped(bh)) { | |
3280 | BUFFER_TRACE(bh, "still unmapped"); | |
3281 | goto next; | |
3282 | } | |
3283 | } | |
3284 | ||
3285 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3286 | if (PageUptodate(page)) | |
3287 | set_buffer_uptodate(bh); | |
3288 | ||
3289 | if (!buffer_uptodate(bh)) { | |
3290 | err = -EIO; | |
3291 | ll_rw_block(READ, 1, &bh); | |
3292 | wait_on_buffer(bh); | |
3293 | /* Uhhuh. Read error. Complain and punt.*/ | |
3294 | if (!buffer_uptodate(bh)) | |
3295 | goto next; | |
3296 | } | |
3297 | ||
3298 | if (ext4_should_journal_data(inode)) { | |
3299 | BUFFER_TRACE(bh, "get write access"); | |
3300 | err = ext4_journal_get_write_access(handle, bh); | |
3301 | if (err) | |
3302 | goto next; | |
3303 | } | |
3304 | ||
3305 | zero_user(page, pos, range_to_discard); | |
3306 | ||
3307 | err = 0; | |
3308 | if (ext4_should_journal_data(inode)) { | |
3309 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
decbd919 | 3310 | } else |
4e96b2db | 3311 | mark_buffer_dirty(bh); |
4e96b2db AH |
3312 | |
3313 | BUFFER_TRACE(bh, "Partial buffer zeroed"); | |
3314 | next: | |
3315 | bh = bh->b_this_page; | |
3316 | iblock++; | |
3317 | pos += range_to_discard; | |
3318 | } | |
3319 | ||
3320 | return err; | |
3321 | } | |
3322 | ||
ac27a0ec | 3323 | /* |
617ba13b | 3324 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' |
ac27a0ec DK |
3325 | * up to the end of the block which corresponds to `from'. |
3326 | * This required during truncate. We need to physically zero the tail end | |
3327 | * of that block so it doesn't yield old data if the file is later grown. | |
3328 | */ | |
cf108bca | 3329 | int ext4_block_truncate_page(handle_t *handle, |
ac27a0ec | 3330 | struct address_space *mapping, loff_t from) |
30848851 AH |
3331 | { |
3332 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3333 | unsigned length; | |
3334 | unsigned blocksize; | |
3335 | struct inode *inode = mapping->host; | |
3336 | ||
3337 | blocksize = inode->i_sb->s_blocksize; | |
3338 | length = blocksize - (offset & (blocksize - 1)); | |
3339 | ||
3340 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3341 | } | |
3342 | ||
3343 | /* | |
3344 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3345 | * starting from file offset 'from'. The range to be zero'd must | |
3346 | * be contained with in one block. If the specified range exceeds | |
3347 | * the end of the block it will be shortened to end of the block | |
3348 | * that cooresponds to 'from' | |
3349 | */ | |
3350 | int ext4_block_zero_page_range(handle_t *handle, | |
3351 | struct address_space *mapping, loff_t from, loff_t length) | |
ac27a0ec | 3352 | { |
617ba13b | 3353 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
ac27a0ec | 3354 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
30848851 | 3355 | unsigned blocksize, max, pos; |
725d26d3 | 3356 | ext4_lblk_t iblock; |
ac27a0ec DK |
3357 | struct inode *inode = mapping->host; |
3358 | struct buffer_head *bh; | |
cf108bca | 3359 | struct page *page; |
ac27a0ec | 3360 | int err = 0; |
ac27a0ec | 3361 | |
f4a01017 TT |
3362 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, |
3363 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
cf108bca JK |
3364 | if (!page) |
3365 | return -EINVAL; | |
3366 | ||
ac27a0ec | 3367 | blocksize = inode->i_sb->s_blocksize; |
30848851 AH |
3368 | max = blocksize - (offset & (blocksize - 1)); |
3369 | ||
3370 | /* | |
3371 | * correct length if it does not fall between | |
3372 | * 'from' and the end of the block | |
3373 | */ | |
3374 | if (length > max || length < 0) | |
3375 | length = max; | |
3376 | ||
ac27a0ec DK |
3377 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); |
3378 | ||
ac27a0ec DK |
3379 | if (!page_has_buffers(page)) |
3380 | create_empty_buffers(page, blocksize, 0); | |
3381 | ||
3382 | /* Find the buffer that contains "offset" */ | |
3383 | bh = page_buffers(page); | |
3384 | pos = blocksize; | |
3385 | while (offset >= pos) { | |
3386 | bh = bh->b_this_page; | |
3387 | iblock++; | |
3388 | pos += blocksize; | |
3389 | } | |
3390 | ||
3391 | err = 0; | |
3392 | if (buffer_freed(bh)) { | |
3393 | BUFFER_TRACE(bh, "freed: skip"); | |
3394 | goto unlock; | |
3395 | } | |
3396 | ||
3397 | if (!buffer_mapped(bh)) { | |
3398 | BUFFER_TRACE(bh, "unmapped"); | |
617ba13b | 3399 | ext4_get_block(inode, iblock, bh, 0); |
ac27a0ec DK |
3400 | /* unmapped? It's a hole - nothing to do */ |
3401 | if (!buffer_mapped(bh)) { | |
3402 | BUFFER_TRACE(bh, "still unmapped"); | |
3403 | goto unlock; | |
3404 | } | |
3405 | } | |
3406 | ||
3407 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3408 | if (PageUptodate(page)) | |
3409 | set_buffer_uptodate(bh); | |
3410 | ||
3411 | if (!buffer_uptodate(bh)) { | |
3412 | err = -EIO; | |
3413 | ll_rw_block(READ, 1, &bh); | |
3414 | wait_on_buffer(bh); | |
3415 | /* Uhhuh. Read error. Complain and punt. */ | |
3416 | if (!buffer_uptodate(bh)) | |
3417 | goto unlock; | |
3418 | } | |
3419 | ||
617ba13b | 3420 | if (ext4_should_journal_data(inode)) { |
ac27a0ec | 3421 | BUFFER_TRACE(bh, "get write access"); |
617ba13b | 3422 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
3423 | if (err) |
3424 | goto unlock; | |
3425 | } | |
3426 | ||
eebd2aa3 | 3427 | zero_user(page, offset, length); |
ac27a0ec DK |
3428 | |
3429 | BUFFER_TRACE(bh, "zeroed end of block"); | |
3430 | ||
3431 | err = 0; | |
617ba13b | 3432 | if (ext4_should_journal_data(inode)) { |
0390131b | 3433 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
decbd919 | 3434 | } else |
ac27a0ec | 3435 | mark_buffer_dirty(bh); |
ac27a0ec DK |
3436 | |
3437 | unlock: | |
3438 | unlock_page(page); | |
3439 | page_cache_release(page); | |
3440 | return err; | |
3441 | } | |
3442 | ||
91ef4caf DG |
3443 | int ext4_can_truncate(struct inode *inode) |
3444 | { | |
91ef4caf DG |
3445 | if (S_ISREG(inode->i_mode)) |
3446 | return 1; | |
3447 | if (S_ISDIR(inode->i_mode)) | |
3448 | return 1; | |
3449 | if (S_ISLNK(inode->i_mode)) | |
3450 | return !ext4_inode_is_fast_symlink(inode); | |
3451 | return 0; | |
3452 | } | |
3453 | ||
a4bb6b64 AH |
3454 | /* |
3455 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks | |
3456 | * associated with the given offset and length | |
3457 | * | |
3458 | * @inode: File inode | |
3459 | * @offset: The offset where the hole will begin | |
3460 | * @len: The length of the hole | |
3461 | * | |
3462 | * Returns: 0 on sucess or negative on failure | |
3463 | */ | |
3464 | ||
3465 | int ext4_punch_hole(struct file *file, loff_t offset, loff_t length) | |
3466 | { | |
3467 | struct inode *inode = file->f_path.dentry->d_inode; | |
3468 | if (!S_ISREG(inode->i_mode)) | |
3469 | return -ENOTSUPP; | |
3470 | ||
3471 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
3472 | /* TODO: Add support for non extent hole punching */ | |
3473 | return -ENOTSUPP; | |
3474 | } | |
3475 | ||
bab08ab9 TT |
3476 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) { |
3477 | /* TODO: Add support for bigalloc file systems */ | |
3478 | return -ENOTSUPP; | |
3479 | } | |
3480 | ||
a4bb6b64 AH |
3481 | return ext4_ext_punch_hole(file, offset, length); |
3482 | } | |
3483 | ||
ac27a0ec | 3484 | /* |
617ba13b | 3485 | * ext4_truncate() |
ac27a0ec | 3486 | * |
617ba13b MC |
3487 | * We block out ext4_get_block() block instantiations across the entire |
3488 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
3489 | * simultaneously on behalf of the same inode. |
3490 | * | |
3491 | * As we work through the truncate and commmit bits of it to the journal there | |
3492 | * is one core, guiding principle: the file's tree must always be consistent on | |
3493 | * disk. We must be able to restart the truncate after a crash. | |
3494 | * | |
3495 | * The file's tree may be transiently inconsistent in memory (although it | |
3496 | * probably isn't), but whenever we close off and commit a journal transaction, | |
3497 | * the contents of (the filesystem + the journal) must be consistent and | |
3498 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
3499 | * left-to-right works OK too). | |
3500 | * | |
3501 | * Note that at recovery time, journal replay occurs *before* the restart of | |
3502 | * truncate against the orphan inode list. | |
3503 | * | |
3504 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 3505 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 3506 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
3507 | * ext4_truncate() to have another go. So there will be instantiated blocks |
3508 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 3509 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 3510 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 3511 | */ |
617ba13b | 3512 | void ext4_truncate(struct inode *inode) |
ac27a0ec | 3513 | { |
0562e0ba JZ |
3514 | trace_ext4_truncate_enter(inode); |
3515 | ||
91ef4caf | 3516 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
3517 | return; |
3518 | ||
12e9b892 | 3519 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 3520 | |
5534fb5b | 3521 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 3522 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 3523 | |
ff9893dc | 3524 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
cf108bca | 3525 | ext4_ext_truncate(inode); |
ff9893dc AG |
3526 | else |
3527 | ext4_ind_truncate(inode); | |
ac27a0ec | 3528 | |
0562e0ba | 3529 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
3530 | } |
3531 | ||
ac27a0ec | 3532 | /* |
617ba13b | 3533 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
3534 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
3535 | * data in memory that is needed to recreate the on-disk version of this | |
3536 | * inode. | |
3537 | */ | |
617ba13b MC |
3538 | static int __ext4_get_inode_loc(struct inode *inode, |
3539 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 3540 | { |
240799cd TT |
3541 | struct ext4_group_desc *gdp; |
3542 | struct buffer_head *bh; | |
3543 | struct super_block *sb = inode->i_sb; | |
3544 | ext4_fsblk_t block; | |
3545 | int inodes_per_block, inode_offset; | |
3546 | ||
3a06d778 | 3547 | iloc->bh = NULL; |
240799cd TT |
3548 | if (!ext4_valid_inum(sb, inode->i_ino)) |
3549 | return -EIO; | |
ac27a0ec | 3550 | |
240799cd TT |
3551 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
3552 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
3553 | if (!gdp) | |
ac27a0ec DK |
3554 | return -EIO; |
3555 | ||
240799cd TT |
3556 | /* |
3557 | * Figure out the offset within the block group inode table | |
3558 | */ | |
00d09882 | 3559 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
3560 | inode_offset = ((inode->i_ino - 1) % |
3561 | EXT4_INODES_PER_GROUP(sb)); | |
3562 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
3563 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
3564 | ||
3565 | bh = sb_getblk(sb, block); | |
ac27a0ec | 3566 | if (!bh) { |
c398eda0 TT |
3567 | EXT4_ERROR_INODE_BLOCK(inode, block, |
3568 | "unable to read itable block"); | |
ac27a0ec DK |
3569 | return -EIO; |
3570 | } | |
3571 | if (!buffer_uptodate(bh)) { | |
3572 | lock_buffer(bh); | |
9c83a923 HK |
3573 | |
3574 | /* | |
3575 | * If the buffer has the write error flag, we have failed | |
3576 | * to write out another inode in the same block. In this | |
3577 | * case, we don't have to read the block because we may | |
3578 | * read the old inode data successfully. | |
3579 | */ | |
3580 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
3581 | set_buffer_uptodate(bh); | |
3582 | ||
ac27a0ec DK |
3583 | if (buffer_uptodate(bh)) { |
3584 | /* someone brought it uptodate while we waited */ | |
3585 | unlock_buffer(bh); | |
3586 | goto has_buffer; | |
3587 | } | |
3588 | ||
3589 | /* | |
3590 | * If we have all information of the inode in memory and this | |
3591 | * is the only valid inode in the block, we need not read the | |
3592 | * block. | |
3593 | */ | |
3594 | if (in_mem) { | |
3595 | struct buffer_head *bitmap_bh; | |
240799cd | 3596 | int i, start; |
ac27a0ec | 3597 | |
240799cd | 3598 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 3599 | |
240799cd TT |
3600 | /* Is the inode bitmap in cache? */ |
3601 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
ac27a0ec DK |
3602 | if (!bitmap_bh) |
3603 | goto make_io; | |
3604 | ||
3605 | /* | |
3606 | * If the inode bitmap isn't in cache then the | |
3607 | * optimisation may end up performing two reads instead | |
3608 | * of one, so skip it. | |
3609 | */ | |
3610 | if (!buffer_uptodate(bitmap_bh)) { | |
3611 | brelse(bitmap_bh); | |
3612 | goto make_io; | |
3613 | } | |
240799cd | 3614 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
3615 | if (i == inode_offset) |
3616 | continue; | |
617ba13b | 3617 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
3618 | break; |
3619 | } | |
3620 | brelse(bitmap_bh); | |
240799cd | 3621 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
3622 | /* all other inodes are free, so skip I/O */ |
3623 | memset(bh->b_data, 0, bh->b_size); | |
3624 | set_buffer_uptodate(bh); | |
3625 | unlock_buffer(bh); | |
3626 | goto has_buffer; | |
3627 | } | |
3628 | } | |
3629 | ||
3630 | make_io: | |
240799cd TT |
3631 | /* |
3632 | * If we need to do any I/O, try to pre-readahead extra | |
3633 | * blocks from the inode table. | |
3634 | */ | |
3635 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
3636 | ext4_fsblk_t b, end, table; | |
3637 | unsigned num; | |
3638 | ||
3639 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 3640 | /* s_inode_readahead_blks is always a power of 2 */ |
240799cd TT |
3641 | b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1); |
3642 | if (table > b) | |
3643 | b = table; | |
3644 | end = b + EXT4_SB(sb)->s_inode_readahead_blks; | |
3645 | num = EXT4_INODES_PER_GROUP(sb); | |
3646 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
3647 | EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) | |
560671a0 | 3648 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
3649 | table += num / inodes_per_block; |
3650 | if (end > table) | |
3651 | end = table; | |
3652 | while (b <= end) | |
3653 | sb_breadahead(sb, b++); | |
3654 | } | |
3655 | ||
ac27a0ec DK |
3656 | /* |
3657 | * There are other valid inodes in the buffer, this inode | |
3658 | * has in-inode xattrs, or we don't have this inode in memory. | |
3659 | * Read the block from disk. | |
3660 | */ | |
0562e0ba | 3661 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
3662 | get_bh(bh); |
3663 | bh->b_end_io = end_buffer_read_sync; | |
3664 | submit_bh(READ_META, bh); | |
3665 | wait_on_buffer(bh); | |
3666 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
3667 | EXT4_ERROR_INODE_BLOCK(inode, block, |
3668 | "unable to read itable block"); | |
ac27a0ec DK |
3669 | brelse(bh); |
3670 | return -EIO; | |
3671 | } | |
3672 | } | |
3673 | has_buffer: | |
3674 | iloc->bh = bh; | |
3675 | return 0; | |
3676 | } | |
3677 | ||
617ba13b | 3678 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
3679 | { |
3680 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 3681 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 3682 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
3683 | } |
3684 | ||
617ba13b | 3685 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 3686 | { |
617ba13b | 3687 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
3688 | |
3689 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 3690 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 3691 | inode->i_flags |= S_SYNC; |
617ba13b | 3692 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 3693 | inode->i_flags |= S_APPEND; |
617ba13b | 3694 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 3695 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 3696 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 3697 | inode->i_flags |= S_NOATIME; |
617ba13b | 3698 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
3699 | inode->i_flags |= S_DIRSYNC; |
3700 | } | |
3701 | ||
ff9ddf7e JK |
3702 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
3703 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
3704 | { | |
84a8dce2 DM |
3705 | unsigned int vfs_fl; |
3706 | unsigned long old_fl, new_fl; | |
3707 | ||
3708 | do { | |
3709 | vfs_fl = ei->vfs_inode.i_flags; | |
3710 | old_fl = ei->i_flags; | |
3711 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
3712 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
3713 | EXT4_DIRSYNC_FL); | |
3714 | if (vfs_fl & S_SYNC) | |
3715 | new_fl |= EXT4_SYNC_FL; | |
3716 | if (vfs_fl & S_APPEND) | |
3717 | new_fl |= EXT4_APPEND_FL; | |
3718 | if (vfs_fl & S_IMMUTABLE) | |
3719 | new_fl |= EXT4_IMMUTABLE_FL; | |
3720 | if (vfs_fl & S_NOATIME) | |
3721 | new_fl |= EXT4_NOATIME_FL; | |
3722 | if (vfs_fl & S_DIRSYNC) | |
3723 | new_fl |= EXT4_DIRSYNC_FL; | |
3724 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 3725 | } |
de9a55b8 | 3726 | |
0fc1b451 | 3727 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 3728 | struct ext4_inode_info *ei) |
0fc1b451 AK |
3729 | { |
3730 | blkcnt_t i_blocks ; | |
8180a562 AK |
3731 | struct inode *inode = &(ei->vfs_inode); |
3732 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
3733 | |
3734 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
3735 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
3736 | /* we are using combined 48 bit field */ | |
3737 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
3738 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 3739 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
3740 | /* i_blocks represent file system block size */ |
3741 | return i_blocks << (inode->i_blkbits - 9); | |
3742 | } else { | |
3743 | return i_blocks; | |
3744 | } | |
0fc1b451 AK |
3745 | } else { |
3746 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
3747 | } | |
3748 | } | |
ff9ddf7e | 3749 | |
1d1fe1ee | 3750 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 3751 | { |
617ba13b MC |
3752 | struct ext4_iloc iloc; |
3753 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 3754 | struct ext4_inode_info *ei; |
1d1fe1ee | 3755 | struct inode *inode; |
b436b9be | 3756 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 3757 | long ret; |
ac27a0ec DK |
3758 | int block; |
3759 | ||
1d1fe1ee DH |
3760 | inode = iget_locked(sb, ino); |
3761 | if (!inode) | |
3762 | return ERR_PTR(-ENOMEM); | |
3763 | if (!(inode->i_state & I_NEW)) | |
3764 | return inode; | |
3765 | ||
3766 | ei = EXT4_I(inode); | |
7dc57615 | 3767 | iloc.bh = NULL; |
ac27a0ec | 3768 | |
1d1fe1ee DH |
3769 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
3770 | if (ret < 0) | |
ac27a0ec | 3771 | goto bad_inode; |
617ba13b | 3772 | raw_inode = ext4_raw_inode(&iloc); |
ac27a0ec DK |
3773 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
3774 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
3775 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 3776 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
3777 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
3778 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
3779 | } | |
3780 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); | |
ac27a0ec | 3781 | |
353eb83c | 3782 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
ac27a0ec DK |
3783 | ei->i_dir_start_lookup = 0; |
3784 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
3785 | /* We now have enough fields to check if the inode was active or not. | |
3786 | * This is needed because nfsd might try to access dead inodes | |
3787 | * the test is that same one that e2fsck uses | |
3788 | * NeilBrown 1999oct15 | |
3789 | */ | |
3790 | if (inode->i_nlink == 0) { | |
3791 | if (inode->i_mode == 0 || | |
617ba13b | 3792 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
ac27a0ec | 3793 | /* this inode is deleted */ |
1d1fe1ee | 3794 | ret = -ESTALE; |
ac27a0ec DK |
3795 | goto bad_inode; |
3796 | } | |
3797 | /* The only unlinked inodes we let through here have | |
3798 | * valid i_mode and are being read by the orphan | |
3799 | * recovery code: that's fine, we're about to complete | |
3800 | * the process of deleting those. */ | |
3801 | } | |
ac27a0ec | 3802 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 3803 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 3804 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 3805 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
3806 | ei->i_file_acl |= |
3807 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 3808 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 3809 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
3810 | #ifdef CONFIG_QUOTA |
3811 | ei->i_reserved_quota = 0; | |
3812 | #endif | |
ac27a0ec DK |
3813 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
3814 | ei->i_block_group = iloc.block_group; | |
a4912123 | 3815 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
3816 | /* |
3817 | * NOTE! The in-memory inode i_data array is in little-endian order | |
3818 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
3819 | */ | |
617ba13b | 3820 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
3821 | ei->i_data[block] = raw_inode->i_block[block]; |
3822 | INIT_LIST_HEAD(&ei->i_orphan); | |
3823 | ||
b436b9be JK |
3824 | /* |
3825 | * Set transaction id's of transactions that have to be committed | |
3826 | * to finish f[data]sync. We set them to currently running transaction | |
3827 | * as we cannot be sure that the inode or some of its metadata isn't | |
3828 | * part of the transaction - the inode could have been reclaimed and | |
3829 | * now it is reread from disk. | |
3830 | */ | |
3831 | if (journal) { | |
3832 | transaction_t *transaction; | |
3833 | tid_t tid; | |
3834 | ||
a931da6a | 3835 | read_lock(&journal->j_state_lock); |
b436b9be JK |
3836 | if (journal->j_running_transaction) |
3837 | transaction = journal->j_running_transaction; | |
3838 | else | |
3839 | transaction = journal->j_committing_transaction; | |
3840 | if (transaction) | |
3841 | tid = transaction->t_tid; | |
3842 | else | |
3843 | tid = journal->j_commit_sequence; | |
a931da6a | 3844 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
3845 | ei->i_sync_tid = tid; |
3846 | ei->i_datasync_tid = tid; | |
3847 | } | |
3848 | ||
0040d987 | 3849 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec | 3850 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
617ba13b | 3851 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
e5d2861f | 3852 | EXT4_INODE_SIZE(inode->i_sb)) { |
1d1fe1ee | 3853 | ret = -EIO; |
ac27a0ec | 3854 | goto bad_inode; |
e5d2861f | 3855 | } |
ac27a0ec DK |
3856 | if (ei->i_extra_isize == 0) { |
3857 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
3858 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
3859 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec DK |
3860 | } else { |
3861 | __le32 *magic = (void *)raw_inode + | |
617ba13b | 3862 | EXT4_GOOD_OLD_INODE_SIZE + |
ac27a0ec | 3863 | ei->i_extra_isize; |
617ba13b | 3864 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) |
19f5fb7a | 3865 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
ac27a0ec DK |
3866 | } |
3867 | } else | |
3868 | ei->i_extra_isize = 0; | |
3869 | ||
ef7f3835 KS |
3870 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
3871 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
3872 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
3873 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
3874 | ||
25ec56b5 JNC |
3875 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
3876 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
3877 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
3878 | inode->i_version |= | |
3879 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
3880 | } | |
3881 | ||
c4b5a614 | 3882 | ret = 0; |
485c26ec | 3883 | if (ei->i_file_acl && |
1032988c | 3884 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
3885 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
3886 | ei->i_file_acl); | |
485c26ec TT |
3887 | ret = -EIO; |
3888 | goto bad_inode; | |
07a03824 | 3889 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
c4b5a614 TT |
3890 | if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
3891 | (S_ISLNK(inode->i_mode) && | |
3892 | !ext4_inode_is_fast_symlink(inode))) | |
3893 | /* Validate extent which is part of inode */ | |
3894 | ret = ext4_ext_check_inode(inode); | |
de9a55b8 | 3895 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
fe2c8191 TN |
3896 | (S_ISLNK(inode->i_mode) && |
3897 | !ext4_inode_is_fast_symlink(inode))) { | |
de9a55b8 | 3898 | /* Validate block references which are part of inode */ |
1f7d1e77 | 3899 | ret = ext4_ind_check_inode(inode); |
fe2c8191 | 3900 | } |
567f3e9a | 3901 | if (ret) |
de9a55b8 | 3902 | goto bad_inode; |
7a262f7c | 3903 | |
ac27a0ec | 3904 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
3905 | inode->i_op = &ext4_file_inode_operations; |
3906 | inode->i_fop = &ext4_file_operations; | |
3907 | ext4_set_aops(inode); | |
ac27a0ec | 3908 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
3909 | inode->i_op = &ext4_dir_inode_operations; |
3910 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 3911 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 3912 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 3913 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
3914 | nd_terminate_link(ei->i_data, inode->i_size, |
3915 | sizeof(ei->i_data) - 1); | |
3916 | } else { | |
617ba13b MC |
3917 | inode->i_op = &ext4_symlink_inode_operations; |
3918 | ext4_set_aops(inode); | |
ac27a0ec | 3919 | } |
563bdd61 TT |
3920 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
3921 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 3922 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
3923 | if (raw_inode->i_block[0]) |
3924 | init_special_inode(inode, inode->i_mode, | |
3925 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
3926 | else | |
3927 | init_special_inode(inode, inode->i_mode, | |
3928 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
563bdd61 | 3929 | } else { |
563bdd61 | 3930 | ret = -EIO; |
24676da4 | 3931 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 3932 | goto bad_inode; |
ac27a0ec | 3933 | } |
af5bc92d | 3934 | brelse(iloc.bh); |
617ba13b | 3935 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
3936 | unlock_new_inode(inode); |
3937 | return inode; | |
ac27a0ec DK |
3938 | |
3939 | bad_inode: | |
567f3e9a | 3940 | brelse(iloc.bh); |
1d1fe1ee DH |
3941 | iget_failed(inode); |
3942 | return ERR_PTR(ret); | |
ac27a0ec DK |
3943 | } |
3944 | ||
0fc1b451 AK |
3945 | static int ext4_inode_blocks_set(handle_t *handle, |
3946 | struct ext4_inode *raw_inode, | |
3947 | struct ext4_inode_info *ei) | |
3948 | { | |
3949 | struct inode *inode = &(ei->vfs_inode); | |
3950 | u64 i_blocks = inode->i_blocks; | |
3951 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
3952 | |
3953 | if (i_blocks <= ~0U) { | |
3954 | /* | |
3955 | * i_blocks can be represnted in a 32 bit variable | |
3956 | * as multiple of 512 bytes | |
3957 | */ | |
8180a562 | 3958 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 3959 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 3960 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
3961 | return 0; |
3962 | } | |
3963 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
3964 | return -EFBIG; | |
3965 | ||
3966 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
3967 | /* |
3968 | * i_blocks can be represented in a 48 bit variable | |
3969 | * as multiple of 512 bytes | |
3970 | */ | |
8180a562 | 3971 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 3972 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 3973 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 3974 | } else { |
84a8dce2 | 3975 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
3976 | /* i_block is stored in file system block size */ |
3977 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
3978 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
3979 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 3980 | } |
f287a1a5 | 3981 | return 0; |
0fc1b451 AK |
3982 | } |
3983 | ||
ac27a0ec DK |
3984 | /* |
3985 | * Post the struct inode info into an on-disk inode location in the | |
3986 | * buffer-cache. This gobbles the caller's reference to the | |
3987 | * buffer_head in the inode location struct. | |
3988 | * | |
3989 | * The caller must have write access to iloc->bh. | |
3990 | */ | |
617ba13b | 3991 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 3992 | struct inode *inode, |
830156c7 | 3993 | struct ext4_iloc *iloc) |
ac27a0ec | 3994 | { |
617ba13b MC |
3995 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
3996 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
3997 | struct buffer_head *bh = iloc->bh; |
3998 | int err = 0, rc, block; | |
3999 | ||
4000 | /* For fields not not tracking in the in-memory inode, | |
4001 | * initialise them to zero for new inodes. */ | |
19f5fb7a | 4002 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4003 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 4004 | |
ff9ddf7e | 4005 | ext4_get_inode_flags(ei); |
ac27a0ec | 4006 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
af5bc92d | 4007 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
4008 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); |
4009 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
4010 | /* | |
4011 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4012 | * re-used with the upper 16 bits of the uid/gid intact | |
4013 | */ | |
af5bc92d | 4014 | if (!ei->i_dtime) { |
ac27a0ec DK |
4015 | raw_inode->i_uid_high = |
4016 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
4017 | raw_inode->i_gid_high = | |
4018 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
4019 | } else { | |
4020 | raw_inode->i_uid_high = 0; | |
4021 | raw_inode->i_gid_high = 0; | |
4022 | } | |
4023 | } else { | |
4024 | raw_inode->i_uid_low = | |
4025 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
4026 | raw_inode->i_gid_low = | |
4027 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
4028 | raw_inode->i_uid_high = 0; | |
4029 | raw_inode->i_gid_high = 0; | |
4030 | } | |
4031 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4032 | |
4033 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4034 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4035 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4036 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4037 | ||
0fc1b451 AK |
4038 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
4039 | goto out_brelse; | |
ac27a0ec | 4040 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4041 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
9b8f1f01 MC |
4042 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
4043 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
4044 | raw_inode->i_file_acl_high = |
4045 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4046 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
a48380f7 AK |
4047 | ext4_isize_set(raw_inode, ei->i_disksize); |
4048 | if (ei->i_disksize > 0x7fffffffULL) { | |
4049 | struct super_block *sb = inode->i_sb; | |
4050 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4051 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
4052 | EXT4_SB(sb)->s_es->s_rev_level == | |
4053 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
4054 | /* If this is the first large file | |
4055 | * created, add a flag to the superblock. | |
4056 | */ | |
4057 | err = ext4_journal_get_write_access(handle, | |
4058 | EXT4_SB(sb)->s_sbh); | |
4059 | if (err) | |
4060 | goto out_brelse; | |
4061 | ext4_update_dynamic_rev(sb); | |
4062 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 4063 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
a48380f7 | 4064 | sb->s_dirt = 1; |
0390131b | 4065 | ext4_handle_sync(handle); |
73b50c1c | 4066 | err = ext4_handle_dirty_metadata(handle, NULL, |
a48380f7 | 4067 | EXT4_SB(sb)->s_sbh); |
ac27a0ec DK |
4068 | } |
4069 | } | |
4070 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4071 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4072 | if (old_valid_dev(inode->i_rdev)) { | |
4073 | raw_inode->i_block[0] = | |
4074 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4075 | raw_inode->i_block[1] = 0; | |
4076 | } else { | |
4077 | raw_inode->i_block[0] = 0; | |
4078 | raw_inode->i_block[1] = | |
4079 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4080 | raw_inode->i_block[2] = 0; | |
4081 | } | |
de9a55b8 TT |
4082 | } else |
4083 | for (block = 0; block < EXT4_N_BLOCKS; block++) | |
4084 | raw_inode->i_block[block] = ei->i_data[block]; | |
ac27a0ec | 4085 | |
25ec56b5 JNC |
4086 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4087 | if (ei->i_extra_isize) { | |
4088 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4089 | raw_inode->i_version_hi = | |
4090 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 4091 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
4092 | } |
4093 | ||
830156c7 | 4094 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 4095 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
4096 | if (!err) |
4097 | err = rc; | |
19f5fb7a | 4098 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
ac27a0ec | 4099 | |
b436b9be | 4100 | ext4_update_inode_fsync_trans(handle, inode, 0); |
ac27a0ec | 4101 | out_brelse: |
af5bc92d | 4102 | brelse(bh); |
617ba13b | 4103 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4104 | return err; |
4105 | } | |
4106 | ||
4107 | /* | |
617ba13b | 4108 | * ext4_write_inode() |
ac27a0ec DK |
4109 | * |
4110 | * We are called from a few places: | |
4111 | * | |
4112 | * - Within generic_file_write() for O_SYNC files. | |
4113 | * Here, there will be no transaction running. We wait for any running | |
4114 | * trasnaction to commit. | |
4115 | * | |
4116 | * - Within sys_sync(), kupdate and such. | |
4117 | * We wait on commit, if tol to. | |
4118 | * | |
4119 | * - Within prune_icache() (PF_MEMALLOC == true) | |
4120 | * Here we simply return. We can't afford to block kswapd on the | |
4121 | * journal commit. | |
4122 | * | |
4123 | * In all cases it is actually safe for us to return without doing anything, | |
4124 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 4125 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
4126 | * knfsd. |
4127 | * | |
4128 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4129 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4130 | * which we are interested. | |
4131 | * | |
4132 | * It would be a bug for them to not do this. The code: | |
4133 | * | |
4134 | * mark_inode_dirty(inode) | |
4135 | * stuff(); | |
4136 | * inode->i_size = expr; | |
4137 | * | |
4138 | * is in error because a kswapd-driven write_inode() could occur while | |
4139 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
4140 | * will no longer be on the superblock's dirty inode list. | |
4141 | */ | |
a9185b41 | 4142 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4143 | { |
91ac6f43 FM |
4144 | int err; |
4145 | ||
ac27a0ec DK |
4146 | if (current->flags & PF_MEMALLOC) |
4147 | return 0; | |
4148 | ||
91ac6f43 FM |
4149 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4150 | if (ext4_journal_current_handle()) { | |
4151 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4152 | dump_stack(); | |
4153 | return -EIO; | |
4154 | } | |
ac27a0ec | 4155 | |
a9185b41 | 4156 | if (wbc->sync_mode != WB_SYNC_ALL) |
91ac6f43 FM |
4157 | return 0; |
4158 | ||
4159 | err = ext4_force_commit(inode->i_sb); | |
4160 | } else { | |
4161 | struct ext4_iloc iloc; | |
ac27a0ec | 4162 | |
8b472d73 | 4163 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4164 | if (err) |
4165 | return err; | |
a9185b41 | 4166 | if (wbc->sync_mode == WB_SYNC_ALL) |
830156c7 FM |
4167 | sync_dirty_buffer(iloc.bh); |
4168 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
4169 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
4170 | "IO error syncing inode"); | |
830156c7 FM |
4171 | err = -EIO; |
4172 | } | |
fd2dd9fb | 4173 | brelse(iloc.bh); |
91ac6f43 FM |
4174 | } |
4175 | return err; | |
ac27a0ec DK |
4176 | } |
4177 | ||
4178 | /* | |
617ba13b | 4179 | * ext4_setattr() |
ac27a0ec DK |
4180 | * |
4181 | * Called from notify_change. | |
4182 | * | |
4183 | * We want to trap VFS attempts to truncate the file as soon as | |
4184 | * possible. In particular, we want to make sure that when the VFS | |
4185 | * shrinks i_size, we put the inode on the orphan list and modify | |
4186 | * i_disksize immediately, so that during the subsequent flushing of | |
4187 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4188 | * commit will leave the blocks being flushed in an unused state on | |
4189 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4190 | * be freed, so we have a strong guarantee that no future commit will | |
4191 | * leave these blocks visible to the user.) | |
4192 | * | |
678aaf48 JK |
4193 | * Another thing we have to assure is that if we are in ordered mode |
4194 | * and inode is still attached to the committing transaction, we must | |
4195 | * we start writeout of all the dirty pages which are being truncated. | |
4196 | * This way we are sure that all the data written in the previous | |
4197 | * transaction are already on disk (truncate waits for pages under | |
4198 | * writeback). | |
4199 | * | |
4200 | * Called with inode->i_mutex down. | |
ac27a0ec | 4201 | */ |
617ba13b | 4202 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
4203 | { |
4204 | struct inode *inode = dentry->d_inode; | |
4205 | int error, rc = 0; | |
3d287de3 | 4206 | int orphan = 0; |
ac27a0ec DK |
4207 | const unsigned int ia_valid = attr->ia_valid; |
4208 | ||
4209 | error = inode_change_ok(inode, attr); | |
4210 | if (error) | |
4211 | return error; | |
4212 | ||
12755627 | 4213 | if (is_quota_modification(inode, attr)) |
871a2931 | 4214 | dquot_initialize(inode); |
ac27a0ec DK |
4215 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || |
4216 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
4217 | handle_t *handle; | |
4218 | ||
4219 | /* (user+group)*(old+new) structure, inode write (sb, | |
4220 | * inode block, ? - but truncate inode update has it) */ | |
5aca07eb | 4221 | handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+ |
194074ac | 4222 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3); |
ac27a0ec DK |
4223 | if (IS_ERR(handle)) { |
4224 | error = PTR_ERR(handle); | |
4225 | goto err_out; | |
4226 | } | |
b43fa828 | 4227 | error = dquot_transfer(inode, attr); |
ac27a0ec | 4228 | if (error) { |
617ba13b | 4229 | ext4_journal_stop(handle); |
ac27a0ec DK |
4230 | return error; |
4231 | } | |
4232 | /* Update corresponding info in inode so that everything is in | |
4233 | * one transaction */ | |
4234 | if (attr->ia_valid & ATTR_UID) | |
4235 | inode->i_uid = attr->ia_uid; | |
4236 | if (attr->ia_valid & ATTR_GID) | |
4237 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4238 | error = ext4_mark_inode_dirty(handle, inode); |
4239 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4240 | } |
4241 | ||
e2b46574 | 4242 | if (attr->ia_valid & ATTR_SIZE) { |
562c72aa CH |
4243 | inode_dio_wait(inode); |
4244 | ||
12e9b892 | 4245 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
4246 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4247 | ||
0c095c7f TT |
4248 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
4249 | return -EFBIG; | |
e2b46574 ES |
4250 | } |
4251 | } | |
4252 | ||
ac27a0ec | 4253 | if (S_ISREG(inode->i_mode) && |
c8d46e41 | 4254 | attr->ia_valid & ATTR_SIZE && |
072bd7ea | 4255 | (attr->ia_size < inode->i_size)) { |
ac27a0ec DK |
4256 | handle_t *handle; |
4257 | ||
617ba13b | 4258 | handle = ext4_journal_start(inode, 3); |
ac27a0ec DK |
4259 | if (IS_ERR(handle)) { |
4260 | error = PTR_ERR(handle); | |
4261 | goto err_out; | |
4262 | } | |
3d287de3 DM |
4263 | if (ext4_handle_valid(handle)) { |
4264 | error = ext4_orphan_add(handle, inode); | |
4265 | orphan = 1; | |
4266 | } | |
617ba13b MC |
4267 | EXT4_I(inode)->i_disksize = attr->ia_size; |
4268 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
4269 | if (!error) |
4270 | error = rc; | |
617ba13b | 4271 | ext4_journal_stop(handle); |
678aaf48 JK |
4272 | |
4273 | if (ext4_should_order_data(inode)) { | |
4274 | error = ext4_begin_ordered_truncate(inode, | |
4275 | attr->ia_size); | |
4276 | if (error) { | |
4277 | /* Do as much error cleanup as possible */ | |
4278 | handle = ext4_journal_start(inode, 3); | |
4279 | if (IS_ERR(handle)) { | |
4280 | ext4_orphan_del(NULL, inode); | |
4281 | goto err_out; | |
4282 | } | |
4283 | ext4_orphan_del(handle, inode); | |
3d287de3 | 4284 | orphan = 0; |
678aaf48 JK |
4285 | ext4_journal_stop(handle); |
4286 | goto err_out; | |
4287 | } | |
4288 | } | |
ac27a0ec DK |
4289 | } |
4290 | ||
072bd7ea TT |
4291 | if (attr->ia_valid & ATTR_SIZE) { |
4292 | if (attr->ia_size != i_size_read(inode)) { | |
4293 | truncate_setsize(inode, attr->ia_size); | |
4294 | ext4_truncate(inode); | |
4295 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
4296 | ext4_truncate(inode); | |
4297 | } | |
ac27a0ec | 4298 | |
1025774c CH |
4299 | if (!rc) { |
4300 | setattr_copy(inode, attr); | |
4301 | mark_inode_dirty(inode); | |
4302 | } | |
4303 | ||
4304 | /* | |
4305 | * If the call to ext4_truncate failed to get a transaction handle at | |
4306 | * all, we need to clean up the in-core orphan list manually. | |
4307 | */ | |
3d287de3 | 4308 | if (orphan && inode->i_nlink) |
617ba13b | 4309 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4310 | |
4311 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 4312 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
4313 | |
4314 | err_out: | |
617ba13b | 4315 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4316 | if (!error) |
4317 | error = rc; | |
4318 | return error; | |
4319 | } | |
4320 | ||
3e3398a0 MC |
4321 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4322 | struct kstat *stat) | |
4323 | { | |
4324 | struct inode *inode; | |
4325 | unsigned long delalloc_blocks; | |
4326 | ||
4327 | inode = dentry->d_inode; | |
4328 | generic_fillattr(inode, stat); | |
4329 | ||
4330 | /* | |
4331 | * We can't update i_blocks if the block allocation is delayed | |
4332 | * otherwise in the case of system crash before the real block | |
4333 | * allocation is done, we will have i_blocks inconsistent with | |
4334 | * on-disk file blocks. | |
4335 | * We always keep i_blocks updated together with real | |
4336 | * allocation. But to not confuse with user, stat | |
4337 | * will return the blocks that include the delayed allocation | |
4338 | * blocks for this file. | |
4339 | */ | |
3e3398a0 | 4340 | delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks; |
3e3398a0 MC |
4341 | |
4342 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; | |
4343 | return 0; | |
4344 | } | |
ac27a0ec | 4345 | |
a02908f1 MC |
4346 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
4347 | { | |
12e9b892 | 4348 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
8bb2b247 | 4349 | return ext4_ind_trans_blocks(inode, nrblocks, chunk); |
ac51d837 | 4350 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); |
a02908f1 | 4351 | } |
ac51d837 | 4352 | |
ac27a0ec | 4353 | /* |
a02908f1 MC |
4354 | * Account for index blocks, block groups bitmaps and block group |
4355 | * descriptor blocks if modify datablocks and index blocks | |
4356 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 4357 | * |
a02908f1 | 4358 | * If datablocks are discontiguous, they are possible to spread over |
af901ca1 | 4359 | * different block groups too. If they are contiuguous, with flexbg, |
a02908f1 | 4360 | * they could still across block group boundary. |
ac27a0ec | 4361 | * |
a02908f1 MC |
4362 | * Also account for superblock, inode, quota and xattr blocks |
4363 | */ | |
1f109d5a | 4364 | static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
a02908f1 | 4365 | { |
8df9675f TT |
4366 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
4367 | int gdpblocks; | |
a02908f1 MC |
4368 | int idxblocks; |
4369 | int ret = 0; | |
4370 | ||
4371 | /* | |
4372 | * How many index blocks need to touch to modify nrblocks? | |
4373 | * The "Chunk" flag indicating whether the nrblocks is | |
4374 | * physically contiguous on disk | |
4375 | * | |
4376 | * For Direct IO and fallocate, they calls get_block to allocate | |
4377 | * one single extent at a time, so they could set the "Chunk" flag | |
4378 | */ | |
4379 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); | |
4380 | ||
4381 | ret = idxblocks; | |
4382 | ||
4383 | /* | |
4384 | * Now let's see how many group bitmaps and group descriptors need | |
4385 | * to account | |
4386 | */ | |
4387 | groups = idxblocks; | |
4388 | if (chunk) | |
4389 | groups += 1; | |
4390 | else | |
4391 | groups += nrblocks; | |
4392 | ||
4393 | gdpblocks = groups; | |
8df9675f TT |
4394 | if (groups > ngroups) |
4395 | groups = ngroups; | |
a02908f1 MC |
4396 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
4397 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
4398 | ||
4399 | /* bitmaps and block group descriptor blocks */ | |
4400 | ret += groups + gdpblocks; | |
4401 | ||
4402 | /* Blocks for super block, inode, quota and xattr blocks */ | |
4403 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
4404 | ||
4405 | return ret; | |
4406 | } | |
4407 | ||
4408 | /* | |
25985edc | 4409 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
4410 | * the modification of a single pages into a single transaction, |
4411 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 4412 | * |
525f4ed8 | 4413 | * This could be called via ext4_write_begin() |
ac27a0ec | 4414 | * |
525f4ed8 | 4415 | * We need to consider the worse case, when |
a02908f1 | 4416 | * one new block per extent. |
ac27a0ec | 4417 | */ |
a86c6181 | 4418 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 4419 | { |
617ba13b | 4420 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
4421 | int ret; |
4422 | ||
a02908f1 | 4423 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
a86c6181 | 4424 | |
a02908f1 | 4425 | /* Account for data blocks for journalled mode */ |
617ba13b | 4426 | if (ext4_should_journal_data(inode)) |
a02908f1 | 4427 | ret += bpp; |
ac27a0ec DK |
4428 | return ret; |
4429 | } | |
f3bd1f3f MC |
4430 | |
4431 | /* | |
4432 | * Calculate the journal credits for a chunk of data modification. | |
4433 | * | |
4434 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 4435 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
4436 | * |
4437 | * journal buffers for data blocks are not included here, as DIO | |
4438 | * and fallocate do no need to journal data buffers. | |
4439 | */ | |
4440 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
4441 | { | |
4442 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
4443 | } | |
4444 | ||
ac27a0ec | 4445 | /* |
617ba13b | 4446 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
4447 | * Give this, we know that the caller already has write access to iloc->bh. |
4448 | */ | |
617ba13b | 4449 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 4450 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4451 | { |
4452 | int err = 0; | |
4453 | ||
25ec56b5 JNC |
4454 | if (test_opt(inode->i_sb, I_VERSION)) |
4455 | inode_inc_iversion(inode); | |
4456 | ||
ac27a0ec DK |
4457 | /* the do_update_inode consumes one bh->b_count */ |
4458 | get_bh(iloc->bh); | |
4459 | ||
dab291af | 4460 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 4461 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
4462 | put_bh(iloc->bh); |
4463 | return err; | |
4464 | } | |
4465 | ||
4466 | /* | |
4467 | * On success, We end up with an outstanding reference count against | |
4468 | * iloc->bh. This _must_ be cleaned up later. | |
4469 | */ | |
4470 | ||
4471 | int | |
617ba13b MC |
4472 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
4473 | struct ext4_iloc *iloc) | |
ac27a0ec | 4474 | { |
0390131b FM |
4475 | int err; |
4476 | ||
4477 | err = ext4_get_inode_loc(inode, iloc); | |
4478 | if (!err) { | |
4479 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
4480 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
4481 | if (err) { | |
4482 | brelse(iloc->bh); | |
4483 | iloc->bh = NULL; | |
ac27a0ec DK |
4484 | } |
4485 | } | |
617ba13b | 4486 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4487 | return err; |
4488 | } | |
4489 | ||
6dd4ee7c KS |
4490 | /* |
4491 | * Expand an inode by new_extra_isize bytes. | |
4492 | * Returns 0 on success or negative error number on failure. | |
4493 | */ | |
1d03ec98 AK |
4494 | static int ext4_expand_extra_isize(struct inode *inode, |
4495 | unsigned int new_extra_isize, | |
4496 | struct ext4_iloc iloc, | |
4497 | handle_t *handle) | |
6dd4ee7c KS |
4498 | { |
4499 | struct ext4_inode *raw_inode; | |
4500 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
4501 | |
4502 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
4503 | return 0; | |
4504 | ||
4505 | raw_inode = ext4_raw_inode(&iloc); | |
4506 | ||
4507 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
4508 | |
4509 | /* No extended attributes present */ | |
19f5fb7a TT |
4510 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
4511 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
4512 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
4513 | new_extra_isize); | |
4514 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
4515 | return 0; | |
4516 | } | |
4517 | ||
4518 | /* try to expand with EAs present */ | |
4519 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
4520 | raw_inode, handle); | |
4521 | } | |
4522 | ||
ac27a0ec DK |
4523 | /* |
4524 | * What we do here is to mark the in-core inode as clean with respect to inode | |
4525 | * dirtiness (it may still be data-dirty). | |
4526 | * This means that the in-core inode may be reaped by prune_icache | |
4527 | * without having to perform any I/O. This is a very good thing, | |
4528 | * because *any* task may call prune_icache - even ones which | |
4529 | * have a transaction open against a different journal. | |
4530 | * | |
4531 | * Is this cheating? Not really. Sure, we haven't written the | |
4532 | * inode out, but prune_icache isn't a user-visible syncing function. | |
4533 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
4534 | * we start and wait on commits. | |
4535 | * | |
4536 | * Is this efficient/effective? Well, we're being nice to the system | |
4537 | * by cleaning up our inodes proactively so they can be reaped | |
4538 | * without I/O. But we are potentially leaving up to five seconds' | |
4539 | * worth of inodes floating about which prune_icache wants us to | |
4540 | * write out. One way to fix that would be to get prune_icache() | |
4541 | * to do a write_super() to free up some memory. It has the desired | |
4542 | * effect. | |
4543 | */ | |
617ba13b | 4544 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 4545 | { |
617ba13b | 4546 | struct ext4_iloc iloc; |
6dd4ee7c KS |
4547 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4548 | static unsigned int mnt_count; | |
4549 | int err, ret; | |
ac27a0ec DK |
4550 | |
4551 | might_sleep(); | |
7ff9c073 | 4552 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 4553 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
4554 | if (ext4_handle_valid(handle) && |
4555 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 4556 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
4557 | /* |
4558 | * We need extra buffer credits since we may write into EA block | |
4559 | * with this same handle. If journal_extend fails, then it will | |
4560 | * only result in a minor loss of functionality for that inode. | |
4561 | * If this is felt to be critical, then e2fsck should be run to | |
4562 | * force a large enough s_min_extra_isize. | |
4563 | */ | |
4564 | if ((jbd2_journal_extend(handle, | |
4565 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
4566 | ret = ext4_expand_extra_isize(inode, | |
4567 | sbi->s_want_extra_isize, | |
4568 | iloc, handle); | |
4569 | if (ret) { | |
19f5fb7a TT |
4570 | ext4_set_inode_state(inode, |
4571 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
4572 | if (mnt_count != |
4573 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 4574 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
4575 | "Unable to expand inode %lu. Delete" |
4576 | " some EAs or run e2fsck.", | |
4577 | inode->i_ino); | |
c1bddad9 AK |
4578 | mnt_count = |
4579 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
4580 | } |
4581 | } | |
4582 | } | |
4583 | } | |
ac27a0ec | 4584 | if (!err) |
617ba13b | 4585 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
4586 | return err; |
4587 | } | |
4588 | ||
4589 | /* | |
617ba13b | 4590 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
4591 | * |
4592 | * We're really interested in the case where a file is being extended. | |
4593 | * i_size has been changed by generic_commit_write() and we thus need | |
4594 | * to include the updated inode in the current transaction. | |
4595 | * | |
5dd4056d | 4596 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
4597 | * are allocated to the file. |
4598 | * | |
4599 | * If the inode is marked synchronous, we don't honour that here - doing | |
4600 | * so would cause a commit on atime updates, which we don't bother doing. | |
4601 | * We handle synchronous inodes at the highest possible level. | |
4602 | */ | |
aa385729 | 4603 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 4604 | { |
ac27a0ec DK |
4605 | handle_t *handle; |
4606 | ||
617ba13b | 4607 | handle = ext4_journal_start(inode, 2); |
ac27a0ec DK |
4608 | if (IS_ERR(handle)) |
4609 | goto out; | |
f3dc272f | 4610 | |
f3dc272f CW |
4611 | ext4_mark_inode_dirty(handle, inode); |
4612 | ||
617ba13b | 4613 | ext4_journal_stop(handle); |
ac27a0ec DK |
4614 | out: |
4615 | return; | |
4616 | } | |
4617 | ||
4618 | #if 0 | |
4619 | /* | |
4620 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
4621 | * it from being flushed to disk early. Unlike | |
617ba13b | 4622 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
4623 | * returns no iloc structure, so the caller needs to repeat the iloc |
4624 | * lookup to mark the inode dirty later. | |
4625 | */ | |
617ba13b | 4626 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 4627 | { |
617ba13b | 4628 | struct ext4_iloc iloc; |
ac27a0ec DK |
4629 | |
4630 | int err = 0; | |
4631 | if (handle) { | |
617ba13b | 4632 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
4633 | if (!err) { |
4634 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 4635 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 4636 | if (!err) |
0390131b | 4637 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 4638 | NULL, |
0390131b | 4639 | iloc.bh); |
ac27a0ec DK |
4640 | brelse(iloc.bh); |
4641 | } | |
4642 | } | |
617ba13b | 4643 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4644 | return err; |
4645 | } | |
4646 | #endif | |
4647 | ||
617ba13b | 4648 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
4649 | { |
4650 | journal_t *journal; | |
4651 | handle_t *handle; | |
4652 | int err; | |
4653 | ||
4654 | /* | |
4655 | * We have to be very careful here: changing a data block's | |
4656 | * journaling status dynamically is dangerous. If we write a | |
4657 | * data block to the journal, change the status and then delete | |
4658 | * that block, we risk forgetting to revoke the old log record | |
4659 | * from the journal and so a subsequent replay can corrupt data. | |
4660 | * So, first we make sure that the journal is empty and that | |
4661 | * nobody is changing anything. | |
4662 | */ | |
4663 | ||
617ba13b | 4664 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
4665 | if (!journal) |
4666 | return 0; | |
d699594d | 4667 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
4668 | return -EROFS; |
4669 | ||
dab291af MC |
4670 | jbd2_journal_lock_updates(journal); |
4671 | jbd2_journal_flush(journal); | |
ac27a0ec DK |
4672 | |
4673 | /* | |
4674 | * OK, there are no updates running now, and all cached data is | |
4675 | * synced to disk. We are now in a completely consistent state | |
4676 | * which doesn't have anything in the journal, and we know that | |
4677 | * no filesystem updates are running, so it is safe to modify | |
4678 | * the inode's in-core data-journaling state flag now. | |
4679 | */ | |
4680 | ||
4681 | if (val) | |
12e9b892 | 4682 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
ac27a0ec | 4683 | else |
12e9b892 | 4684 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
617ba13b | 4685 | ext4_set_aops(inode); |
ac27a0ec | 4686 | |
dab291af | 4687 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
4688 | |
4689 | /* Finally we can mark the inode as dirty. */ | |
4690 | ||
617ba13b | 4691 | handle = ext4_journal_start(inode, 1); |
ac27a0ec DK |
4692 | if (IS_ERR(handle)) |
4693 | return PTR_ERR(handle); | |
4694 | ||
617ba13b | 4695 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 4696 | ext4_handle_sync(handle); |
617ba13b MC |
4697 | ext4_journal_stop(handle); |
4698 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
4699 | |
4700 | return err; | |
4701 | } | |
2e9ee850 AK |
4702 | |
4703 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
4704 | { | |
4705 | return !buffer_mapped(bh); | |
4706 | } | |
4707 | ||
c2ec175c | 4708 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 4709 | { |
c2ec175c | 4710 | struct page *page = vmf->page; |
2e9ee850 AK |
4711 | loff_t size; |
4712 | unsigned long len; | |
9ea7df53 | 4713 | int ret; |
2e9ee850 AK |
4714 | struct file *file = vma->vm_file; |
4715 | struct inode *inode = file->f_path.dentry->d_inode; | |
4716 | struct address_space *mapping = inode->i_mapping; | |
9ea7df53 JK |
4717 | handle_t *handle; |
4718 | get_block_t *get_block; | |
4719 | int retries = 0; | |
2e9ee850 AK |
4720 | |
4721 | /* | |
9ea7df53 JK |
4722 | * This check is racy but catches the common case. We rely on |
4723 | * __block_page_mkwrite() to do a reliable check. | |
2e9ee850 | 4724 | */ |
9ea7df53 JK |
4725 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); |
4726 | /* Delalloc case is easy... */ | |
4727 | if (test_opt(inode->i_sb, DELALLOC) && | |
4728 | !ext4_should_journal_data(inode) && | |
4729 | !ext4_nonda_switch(inode->i_sb)) { | |
4730 | do { | |
4731 | ret = __block_page_mkwrite(vma, vmf, | |
4732 | ext4_da_get_block_prep); | |
4733 | } while (ret == -ENOSPC && | |
4734 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
4735 | goto out_ret; | |
2e9ee850 | 4736 | } |
0e499890 DW |
4737 | |
4738 | lock_page(page); | |
9ea7df53 JK |
4739 | size = i_size_read(inode); |
4740 | /* Page got truncated from under us? */ | |
4741 | if (page->mapping != mapping || page_offset(page) > size) { | |
4742 | unlock_page(page); | |
4743 | ret = VM_FAULT_NOPAGE; | |
4744 | goto out; | |
0e499890 | 4745 | } |
2e9ee850 AK |
4746 | |
4747 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
4748 | len = size & ~PAGE_CACHE_MASK; | |
4749 | else | |
4750 | len = PAGE_CACHE_SIZE; | |
a827eaff | 4751 | /* |
9ea7df53 JK |
4752 | * Return if we have all the buffers mapped. This avoids the need to do |
4753 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 4754 | */ |
2e9ee850 | 4755 | if (page_has_buffers(page)) { |
2e9ee850 | 4756 | if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, |
a827eaff | 4757 | ext4_bh_unmapped)) { |
9ea7df53 JK |
4758 | /* Wait so that we don't change page under IO */ |
4759 | wait_on_page_writeback(page); | |
4760 | ret = VM_FAULT_LOCKED; | |
4761 | goto out; | |
a827eaff | 4762 | } |
2e9ee850 | 4763 | } |
a827eaff | 4764 | unlock_page(page); |
9ea7df53 JK |
4765 | /* OK, we need to fill the hole... */ |
4766 | if (ext4_should_dioread_nolock(inode)) | |
4767 | get_block = ext4_get_block_write; | |
4768 | else | |
4769 | get_block = ext4_get_block; | |
4770 | retry_alloc: | |
4771 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); | |
4772 | if (IS_ERR(handle)) { | |
c2ec175c | 4773 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
4774 | goto out; |
4775 | } | |
4776 | ret = __block_page_mkwrite(vma, vmf, get_block); | |
4777 | if (!ret && ext4_should_journal_data(inode)) { | |
4778 | if (walk_page_buffers(handle, page_buffers(page), 0, | |
4779 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { | |
4780 | unlock_page(page); | |
4781 | ret = VM_FAULT_SIGBUS; | |
4782 | goto out; | |
4783 | } | |
4784 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
4785 | } | |
4786 | ext4_journal_stop(handle); | |
4787 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
4788 | goto retry_alloc; | |
4789 | out_ret: | |
4790 | ret = block_page_mkwrite_return(ret); | |
4791 | out: | |
2e9ee850 AK |
4792 | return ret; |
4793 | } |