]>
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" | |
d2a17637 | 45 | #include "ext4_extents.h" |
9f125d64 | 46 | #include "truncate.h" |
ac27a0ec | 47 | |
9bffad1e TT |
48 | #include <trace/events/ext4.h> |
49 | ||
a1d6cc56 AK |
50 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
51 | ||
678aaf48 JK |
52 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
53 | loff_t new_size) | |
54 | { | |
7ff9c073 | 55 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
56 | /* |
57 | * If jinode is zero, then we never opened the file for | |
58 | * writing, so there's no need to call | |
59 | * jbd2_journal_begin_ordered_truncate() since there's no | |
60 | * outstanding writes we need to flush. | |
61 | */ | |
62 | if (!EXT4_I(inode)->jinode) | |
63 | return 0; | |
64 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
65 | EXT4_I(inode)->jinode, | |
66 | new_size); | |
678aaf48 JK |
67 | } |
68 | ||
64769240 | 69 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
cb20d518 TT |
70 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, |
71 | struct buffer_head *bh_result, int create); | |
72 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode); | |
73 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate); | |
74 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); | |
75 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
64769240 | 76 | |
ac27a0ec DK |
77 | /* |
78 | * Test whether an inode is a fast symlink. | |
79 | */ | |
617ba13b | 80 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 81 | { |
617ba13b | 82 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
83 | (inode->i_sb->s_blocksize >> 9) : 0; |
84 | ||
85 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
86 | } | |
87 | ||
ac27a0ec DK |
88 | /* |
89 | * Restart the transaction associated with *handle. This does a commit, | |
90 | * so before we call here everything must be consistently dirtied against | |
91 | * this transaction. | |
92 | */ | |
fa5d1113 | 93 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 94 | int nblocks) |
ac27a0ec | 95 | { |
487caeef JK |
96 | int ret; |
97 | ||
98 | /* | |
e35fd660 | 99 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
100 | * moment, get_block can be called only for blocks inside i_size since |
101 | * page cache has been already dropped and writes are blocked by | |
102 | * i_mutex. So we can safely drop the i_data_sem here. | |
103 | */ | |
0390131b | 104 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 105 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 106 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 107 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 108 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 109 | ext4_discard_preallocations(inode); |
487caeef JK |
110 | |
111 | return ret; | |
ac27a0ec DK |
112 | } |
113 | ||
114 | /* | |
115 | * Called at the last iput() if i_nlink is zero. | |
116 | */ | |
0930fcc1 | 117 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
118 | { |
119 | handle_t *handle; | |
bc965ab3 | 120 | int err; |
ac27a0ec | 121 | |
7ff9c073 | 122 | trace_ext4_evict_inode(inode); |
2581fdc8 | 123 | |
2581fdc8 JZ |
124 | ext4_ioend_wait(inode); |
125 | ||
0930fcc1 | 126 | if (inode->i_nlink) { |
2d859db3 JK |
127 | /* |
128 | * When journalling data dirty buffers are tracked only in the | |
129 | * journal. So although mm thinks everything is clean and | |
130 | * ready for reaping the inode might still have some pages to | |
131 | * write in the running transaction or waiting to be | |
132 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
133 | * (via truncate_inode_pages()) to discard these buffers can | |
134 | * cause data loss. Also even if we did not discard these | |
135 | * buffers, we would have no way to find them after the inode | |
136 | * is reaped and thus user could see stale data if he tries to | |
137 | * read them before the transaction is checkpointed. So be | |
138 | * careful and force everything to disk here... We use | |
139 | * ei->i_datasync_tid to store the newest transaction | |
140 | * containing inode's data. | |
141 | * | |
142 | * Note that directories do not have this problem because they | |
143 | * don't use page cache. | |
144 | */ | |
145 | if (ext4_should_journal_data(inode) && | |
146 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) { | |
147 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
148 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
149 | ||
150 | jbd2_log_start_commit(journal, commit_tid); | |
151 | jbd2_log_wait_commit(journal, commit_tid); | |
152 | filemap_write_and_wait(&inode->i_data); | |
153 | } | |
0930fcc1 AV |
154 | truncate_inode_pages(&inode->i_data, 0); |
155 | goto no_delete; | |
156 | } | |
157 | ||
907f4554 | 158 | if (!is_bad_inode(inode)) |
871a2931 | 159 | dquot_initialize(inode); |
907f4554 | 160 | |
678aaf48 JK |
161 | if (ext4_should_order_data(inode)) |
162 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec DK |
163 | truncate_inode_pages(&inode->i_data, 0); |
164 | ||
165 | if (is_bad_inode(inode)) | |
166 | goto no_delete; | |
167 | ||
9f125d64 | 168 | handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3); |
ac27a0ec | 169 | if (IS_ERR(handle)) { |
bc965ab3 | 170 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
171 | /* |
172 | * If we're going to skip the normal cleanup, we still need to | |
173 | * make sure that the in-core orphan linked list is properly | |
174 | * cleaned up. | |
175 | */ | |
617ba13b | 176 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
177 | goto no_delete; |
178 | } | |
179 | ||
180 | if (IS_SYNC(inode)) | |
0390131b | 181 | ext4_handle_sync(handle); |
ac27a0ec | 182 | inode->i_size = 0; |
bc965ab3 TT |
183 | err = ext4_mark_inode_dirty(handle, inode); |
184 | if (err) { | |
12062ddd | 185 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
186 | "couldn't mark inode dirty (err %d)", err); |
187 | goto stop_handle; | |
188 | } | |
ac27a0ec | 189 | if (inode->i_blocks) |
617ba13b | 190 | ext4_truncate(inode); |
bc965ab3 TT |
191 | |
192 | /* | |
193 | * ext4_ext_truncate() doesn't reserve any slop when it | |
194 | * restarts journal transactions; therefore there may not be | |
195 | * enough credits left in the handle to remove the inode from | |
196 | * the orphan list and set the dtime field. | |
197 | */ | |
0390131b | 198 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
199 | err = ext4_journal_extend(handle, 3); |
200 | if (err > 0) | |
201 | err = ext4_journal_restart(handle, 3); | |
202 | if (err != 0) { | |
12062ddd | 203 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
204 | "couldn't extend journal (err %d)", err); |
205 | stop_handle: | |
206 | ext4_journal_stop(handle); | |
45388219 | 207 | ext4_orphan_del(NULL, inode); |
bc965ab3 TT |
208 | goto no_delete; |
209 | } | |
210 | } | |
211 | ||
ac27a0ec | 212 | /* |
617ba13b | 213 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 214 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 215 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 216 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 217 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
218 | * (Well, we could do this if we need to, but heck - it works) |
219 | */ | |
617ba13b MC |
220 | ext4_orphan_del(handle, inode); |
221 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
222 | |
223 | /* | |
224 | * One subtle ordering requirement: if anything has gone wrong | |
225 | * (transaction abort, IO errors, whatever), then we can still | |
226 | * do these next steps (the fs will already have been marked as | |
227 | * having errors), but we can't free the inode if the mark_dirty | |
228 | * fails. | |
229 | */ | |
617ba13b | 230 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 231 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 232 | ext4_clear_inode(inode); |
ac27a0ec | 233 | else |
617ba13b MC |
234 | ext4_free_inode(handle, inode); |
235 | ext4_journal_stop(handle); | |
ac27a0ec DK |
236 | return; |
237 | no_delete: | |
0930fcc1 | 238 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
239 | } |
240 | ||
a9e7f447 DM |
241 | #ifdef CONFIG_QUOTA |
242 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 243 | { |
a9e7f447 | 244 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 245 | } |
a9e7f447 | 246 | #endif |
9d0be502 | 247 | |
12219aea AK |
248 | /* |
249 | * Calculate the number of metadata blocks need to reserve | |
9d0be502 | 250 | * to allocate a block located at @lblock |
12219aea | 251 | */ |
01f49d0b | 252 | static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
12219aea | 253 | { |
12e9b892 | 254 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
9d0be502 | 255 | return ext4_ext_calc_metadata_amount(inode, lblock); |
12219aea | 256 | |
8bb2b247 | 257 | return ext4_ind_calc_metadata_amount(inode, lblock); |
12219aea AK |
258 | } |
259 | ||
0637c6f4 TT |
260 | /* |
261 | * Called with i_data_sem down, which is important since we can call | |
262 | * ext4_discard_preallocations() from here. | |
263 | */ | |
5f634d06 AK |
264 | void ext4_da_update_reserve_space(struct inode *inode, |
265 | int used, int quota_claim) | |
12219aea AK |
266 | { |
267 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 268 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
269 | |
270 | spin_lock(&ei->i_block_reservation_lock); | |
f8ec9d68 | 271 | trace_ext4_da_update_reserve_space(inode, used); |
0637c6f4 TT |
272 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
273 | ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d " | |
274 | "with only %d reserved data blocks\n", | |
275 | __func__, inode->i_ino, used, | |
276 | ei->i_reserved_data_blocks); | |
277 | WARN_ON(1); | |
278 | used = ei->i_reserved_data_blocks; | |
279 | } | |
12219aea | 280 | |
0637c6f4 TT |
281 | /* Update per-inode reservations */ |
282 | ei->i_reserved_data_blocks -= used; | |
0637c6f4 | 283 | ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; |
57042651 | 284 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 285 | used + ei->i_allocated_meta_blocks); |
0637c6f4 | 286 | ei->i_allocated_meta_blocks = 0; |
6bc6e63f | 287 | |
0637c6f4 TT |
288 | if (ei->i_reserved_data_blocks == 0) { |
289 | /* | |
290 | * We can release all of the reserved metadata blocks | |
291 | * only when we have written all of the delayed | |
292 | * allocation blocks. | |
293 | */ | |
57042651 | 294 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 295 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 296 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 297 | ei->i_da_metadata_calc_len = 0; |
6bc6e63f | 298 | } |
12219aea | 299 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 300 | |
72b8ab9d ES |
301 | /* Update quota subsystem for data blocks */ |
302 | if (quota_claim) | |
7b415bf6 | 303 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 304 | else { |
5f634d06 AK |
305 | /* |
306 | * We did fallocate with an offset that is already delayed | |
307 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 308 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 309 | */ |
7b415bf6 | 310 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 311 | } |
d6014301 AK |
312 | |
313 | /* | |
314 | * If we have done all the pending block allocations and if | |
315 | * there aren't any writers on the inode, we can discard the | |
316 | * inode's preallocations. | |
317 | */ | |
0637c6f4 TT |
318 | if ((ei->i_reserved_data_blocks == 0) && |
319 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 320 | ext4_discard_preallocations(inode); |
12219aea AK |
321 | } |
322 | ||
e29136f8 | 323 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
324 | unsigned int line, |
325 | struct ext4_map_blocks *map) | |
6fd058f7 | 326 | { |
24676da4 TT |
327 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
328 | map->m_len)) { | |
c398eda0 TT |
329 | ext4_error_inode(inode, func, line, map->m_pblk, |
330 | "lblock %lu mapped to illegal pblock " | |
331 | "(length %d)", (unsigned long) map->m_lblk, | |
332 | map->m_len); | |
6fd058f7 TT |
333 | return -EIO; |
334 | } | |
335 | return 0; | |
336 | } | |
337 | ||
e29136f8 | 338 | #define check_block_validity(inode, map) \ |
c398eda0 | 339 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 340 | |
55138e0b | 341 | /* |
1f94533d TT |
342 | * Return the number of contiguous dirty pages in a given inode |
343 | * starting at page frame idx. | |
55138e0b TT |
344 | */ |
345 | static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx, | |
346 | unsigned int max_pages) | |
347 | { | |
348 | struct address_space *mapping = inode->i_mapping; | |
349 | pgoff_t index; | |
350 | struct pagevec pvec; | |
351 | pgoff_t num = 0; | |
352 | int i, nr_pages, done = 0; | |
353 | ||
354 | if (max_pages == 0) | |
355 | return 0; | |
356 | pagevec_init(&pvec, 0); | |
357 | while (!done) { | |
358 | index = idx; | |
359 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
360 | PAGECACHE_TAG_DIRTY, | |
361 | (pgoff_t)PAGEVEC_SIZE); | |
362 | if (nr_pages == 0) | |
363 | break; | |
364 | for (i = 0; i < nr_pages; i++) { | |
365 | struct page *page = pvec.pages[i]; | |
366 | struct buffer_head *bh, *head; | |
367 | ||
368 | lock_page(page); | |
369 | if (unlikely(page->mapping != mapping) || | |
370 | !PageDirty(page) || | |
371 | PageWriteback(page) || | |
372 | page->index != idx) { | |
373 | done = 1; | |
374 | unlock_page(page); | |
375 | break; | |
376 | } | |
1f94533d TT |
377 | if (page_has_buffers(page)) { |
378 | bh = head = page_buffers(page); | |
379 | do { | |
380 | if (!buffer_delay(bh) && | |
381 | !buffer_unwritten(bh)) | |
382 | done = 1; | |
383 | bh = bh->b_this_page; | |
384 | } while (!done && (bh != head)); | |
385 | } | |
55138e0b TT |
386 | unlock_page(page); |
387 | if (done) | |
388 | break; | |
389 | idx++; | |
390 | num++; | |
659c6009 ES |
391 | if (num >= max_pages) { |
392 | done = 1; | |
55138e0b | 393 | break; |
659c6009 | 394 | } |
55138e0b TT |
395 | } |
396 | pagevec_release(&pvec); | |
397 | } | |
398 | return num; | |
399 | } | |
400 | ||
f5ab0d1f | 401 | /* |
e35fd660 | 402 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 403 | * and returns if the blocks are already mapped. |
f5ab0d1f | 404 | * |
f5ab0d1f MC |
405 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
406 | * and store the allocated blocks in the result buffer head and mark it | |
407 | * mapped. | |
408 | * | |
e35fd660 TT |
409 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
410 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
411 | * based files |
412 | * | |
413 | * On success, it returns the number of blocks being mapped or allocate. | |
414 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
415 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
416 | * the buffer head is mapped. | |
417 | * | |
418 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
419 | * that casem, buffer head is unmapped | |
420 | * | |
421 | * It returns the error in case of allocation failure. | |
422 | */ | |
e35fd660 TT |
423 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
424 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 AK |
425 | { |
426 | int retval; | |
f5ab0d1f | 427 | |
e35fd660 TT |
428 | map->m_flags = 0; |
429 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
430 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
431 | (unsigned long) map->m_lblk); | |
4df3d265 | 432 | /* |
b920c755 TT |
433 | * Try to see if we can get the block without requesting a new |
434 | * file system block. | |
4df3d265 AK |
435 | */ |
436 | down_read((&EXT4_I(inode)->i_data_sem)); | |
12e9b892 | 437 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 438 | retval = ext4_ext_map_blocks(handle, inode, map, 0); |
0e855ac8 | 439 | } else { |
e35fd660 | 440 | retval = ext4_ind_map_blocks(handle, inode, map, 0); |
0e855ac8 | 441 | } |
4df3d265 | 442 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 443 | |
e35fd660 | 444 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 445 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
446 | if (ret != 0) |
447 | return ret; | |
448 | } | |
449 | ||
f5ab0d1f | 450 | /* If it is only a block(s) look up */ |
c2177057 | 451 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
452 | return retval; |
453 | ||
454 | /* | |
455 | * Returns if the blocks have already allocated | |
456 | * | |
457 | * Note that if blocks have been preallocated | |
458 | * ext4_ext_get_block() returns th create = 0 | |
459 | * with buffer head unmapped. | |
460 | */ | |
e35fd660 | 461 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
4df3d265 AK |
462 | return retval; |
463 | ||
2a8964d6 AK |
464 | /* |
465 | * When we call get_blocks without the create flag, the | |
466 | * BH_Unwritten flag could have gotten set if the blocks | |
467 | * requested were part of a uninitialized extent. We need to | |
468 | * clear this flag now that we are committed to convert all or | |
469 | * part of the uninitialized extent to be an initialized | |
470 | * extent. This is because we need to avoid the combination | |
471 | * of BH_Unwritten and BH_Mapped flags being simultaneously | |
472 | * set on the buffer_head. | |
473 | */ | |
e35fd660 | 474 | map->m_flags &= ~EXT4_MAP_UNWRITTEN; |
2a8964d6 | 475 | |
4df3d265 | 476 | /* |
f5ab0d1f MC |
477 | * New blocks allocate and/or writing to uninitialized extent |
478 | * will possibly result in updating i_data, so we take | |
479 | * the write lock of i_data_sem, and call get_blocks() | |
480 | * with create == 1 flag. | |
4df3d265 AK |
481 | */ |
482 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
483 | |
484 | /* | |
485 | * if the caller is from delayed allocation writeout path | |
486 | * we have already reserved fs blocks for allocation | |
487 | * let the underlying get_block() function know to | |
488 | * avoid double accounting | |
489 | */ | |
c2177057 | 490 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 491 | ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
4df3d265 AK |
492 | /* |
493 | * We need to check for EXT4 here because migrate | |
494 | * could have changed the inode type in between | |
495 | */ | |
12e9b892 | 496 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 497 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 498 | } else { |
e35fd660 | 499 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 500 | |
e35fd660 | 501 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
502 | /* |
503 | * We allocated new blocks which will result in | |
504 | * i_data's format changing. Force the migrate | |
505 | * to fail by clearing migrate flags | |
506 | */ | |
19f5fb7a | 507 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 508 | } |
d2a17637 | 509 | |
5f634d06 AK |
510 | /* |
511 | * Update reserved blocks/metadata blocks after successful | |
512 | * block allocation which had been deferred till now. We don't | |
513 | * support fallocate for non extent files. So we can update | |
514 | * reserve space here. | |
515 | */ | |
516 | if ((retval > 0) && | |
1296cc85 | 517 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
518 | ext4_da_update_reserve_space(inode, retval, 1); |
519 | } | |
2ac3b6e0 | 520 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 521 | ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
2ac3b6e0 | 522 | |
4df3d265 | 523 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 524 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 525 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
526 | if (ret != 0) |
527 | return ret; | |
528 | } | |
0e855ac8 AK |
529 | return retval; |
530 | } | |
531 | ||
f3bd1f3f MC |
532 | /* Maximum number of blocks we map for direct IO at once. */ |
533 | #define DIO_MAX_BLOCKS 4096 | |
534 | ||
2ed88685 TT |
535 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
536 | struct buffer_head *bh, int flags) | |
ac27a0ec | 537 | { |
3e4fdaf8 | 538 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 539 | struct ext4_map_blocks map; |
7fb5409d | 540 | int ret = 0, started = 0; |
f3bd1f3f | 541 | int dio_credits; |
ac27a0ec | 542 | |
2ed88685 TT |
543 | map.m_lblk = iblock; |
544 | map.m_len = bh->b_size >> inode->i_blkbits; | |
545 | ||
546 | if (flags && !handle) { | |
7fb5409d | 547 | /* Direct IO write... */ |
2ed88685 TT |
548 | if (map.m_len > DIO_MAX_BLOCKS) |
549 | map.m_len = DIO_MAX_BLOCKS; | |
550 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
f3bd1f3f | 551 | handle = ext4_journal_start(inode, dio_credits); |
7fb5409d | 552 | if (IS_ERR(handle)) { |
ac27a0ec | 553 | ret = PTR_ERR(handle); |
2ed88685 | 554 | return ret; |
ac27a0ec | 555 | } |
7fb5409d | 556 | started = 1; |
ac27a0ec DK |
557 | } |
558 | ||
2ed88685 | 559 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 560 | if (ret > 0) { |
2ed88685 TT |
561 | map_bh(bh, inode->i_sb, map.m_pblk); |
562 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
563 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
7fb5409d | 564 | ret = 0; |
ac27a0ec | 565 | } |
7fb5409d JK |
566 | if (started) |
567 | ext4_journal_stop(handle); | |
ac27a0ec DK |
568 | return ret; |
569 | } | |
570 | ||
2ed88685 TT |
571 | int ext4_get_block(struct inode *inode, sector_t iblock, |
572 | struct buffer_head *bh, int create) | |
573 | { | |
574 | return _ext4_get_block(inode, iblock, bh, | |
575 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
576 | } | |
577 | ||
ac27a0ec DK |
578 | /* |
579 | * `handle' can be NULL if create is zero | |
580 | */ | |
617ba13b | 581 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 582 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec | 583 | { |
2ed88685 TT |
584 | struct ext4_map_blocks map; |
585 | struct buffer_head *bh; | |
ac27a0ec DK |
586 | int fatal = 0, err; |
587 | ||
588 | J_ASSERT(handle != NULL || create == 0); | |
589 | ||
2ed88685 TT |
590 | map.m_lblk = block; |
591 | map.m_len = 1; | |
592 | err = ext4_map_blocks(handle, inode, &map, | |
593 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
ac27a0ec | 594 | |
2ed88685 TT |
595 | if (err < 0) |
596 | *errp = err; | |
597 | if (err <= 0) | |
598 | return NULL; | |
599 | *errp = 0; | |
600 | ||
601 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
602 | if (!bh) { | |
603 | *errp = -EIO; | |
604 | return NULL; | |
ac27a0ec | 605 | } |
2ed88685 TT |
606 | if (map.m_flags & EXT4_MAP_NEW) { |
607 | J_ASSERT(create != 0); | |
608 | J_ASSERT(handle != NULL); | |
ac27a0ec | 609 | |
2ed88685 TT |
610 | /* |
611 | * Now that we do not always journal data, we should | |
612 | * keep in mind whether this should always journal the | |
613 | * new buffer as metadata. For now, regular file | |
614 | * writes use ext4_get_block instead, so it's not a | |
615 | * problem. | |
616 | */ | |
617 | lock_buffer(bh); | |
618 | BUFFER_TRACE(bh, "call get_create_access"); | |
619 | fatal = ext4_journal_get_create_access(handle, bh); | |
620 | if (!fatal && !buffer_uptodate(bh)) { | |
621 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); | |
622 | set_buffer_uptodate(bh); | |
ac27a0ec | 623 | } |
2ed88685 TT |
624 | unlock_buffer(bh); |
625 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
626 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
627 | if (!fatal) | |
628 | fatal = err; | |
629 | } else { | |
630 | BUFFER_TRACE(bh, "not a new buffer"); | |
ac27a0ec | 631 | } |
2ed88685 TT |
632 | if (fatal) { |
633 | *errp = fatal; | |
634 | brelse(bh); | |
635 | bh = NULL; | |
636 | } | |
637 | return bh; | |
ac27a0ec DK |
638 | } |
639 | ||
617ba13b | 640 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 641 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 642 | { |
af5bc92d | 643 | struct buffer_head *bh; |
ac27a0ec | 644 | |
617ba13b | 645 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
646 | if (!bh) |
647 | return bh; | |
648 | if (buffer_uptodate(bh)) | |
649 | return bh; | |
650 | ll_rw_block(READ_META, 1, &bh); | |
651 | wait_on_buffer(bh); | |
652 | if (buffer_uptodate(bh)) | |
653 | return bh; | |
654 | put_bh(bh); | |
655 | *err = -EIO; | |
656 | return NULL; | |
657 | } | |
658 | ||
af5bc92d TT |
659 | static int walk_page_buffers(handle_t *handle, |
660 | struct buffer_head *head, | |
661 | unsigned from, | |
662 | unsigned to, | |
663 | int *partial, | |
664 | int (*fn)(handle_t *handle, | |
665 | struct buffer_head *bh)) | |
ac27a0ec DK |
666 | { |
667 | struct buffer_head *bh; | |
668 | unsigned block_start, block_end; | |
669 | unsigned blocksize = head->b_size; | |
670 | int err, ret = 0; | |
671 | struct buffer_head *next; | |
672 | ||
af5bc92d TT |
673 | for (bh = head, block_start = 0; |
674 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 675 | block_start = block_end, bh = next) { |
ac27a0ec DK |
676 | next = bh->b_this_page; |
677 | block_end = block_start + blocksize; | |
678 | if (block_end <= from || block_start >= to) { | |
679 | if (partial && !buffer_uptodate(bh)) | |
680 | *partial = 1; | |
681 | continue; | |
682 | } | |
683 | err = (*fn)(handle, bh); | |
684 | if (!ret) | |
685 | ret = err; | |
686 | } | |
687 | return ret; | |
688 | } | |
689 | ||
690 | /* | |
691 | * To preserve ordering, it is essential that the hole instantiation and | |
692 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 693 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 694 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
695 | * prepare_write() is the right place. |
696 | * | |
617ba13b MC |
697 | * Also, this function can nest inside ext4_writepage() -> |
698 | * block_write_full_page(). In that case, we *know* that ext4_writepage() | |
ac27a0ec DK |
699 | * has generated enough buffer credits to do the whole page. So we won't |
700 | * block on the journal in that case, which is good, because the caller may | |
701 | * be PF_MEMALLOC. | |
702 | * | |
617ba13b | 703 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
704 | * quota file writes. If we were to commit the transaction while thus |
705 | * reentered, there can be a deadlock - we would be holding a quota | |
706 | * lock, and the commit would never complete if another thread had a | |
707 | * transaction open and was blocking on the quota lock - a ranking | |
708 | * violation. | |
709 | * | |
dab291af | 710 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
711 | * will _not_ run commit under these circumstances because handle->h_ref |
712 | * is elevated. We'll still have enough credits for the tiny quotafile | |
713 | * write. | |
714 | */ | |
715 | static int do_journal_get_write_access(handle_t *handle, | |
de9a55b8 | 716 | struct buffer_head *bh) |
ac27a0ec | 717 | { |
56d35a4c JK |
718 | int dirty = buffer_dirty(bh); |
719 | int ret; | |
720 | ||
ac27a0ec DK |
721 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
722 | return 0; | |
56d35a4c | 723 | /* |
ebdec241 | 724 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
725 | * the dirty bit as jbd2_journal_get_write_access() could complain |
726 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 727 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
728 | * the bit before releasing a page lock and thus writeback cannot |
729 | * ever write the buffer. | |
730 | */ | |
731 | if (dirty) | |
732 | clear_buffer_dirty(bh); | |
733 | ret = ext4_journal_get_write_access(handle, bh); | |
734 | if (!ret && dirty) | |
735 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
736 | return ret; | |
ac27a0ec DK |
737 | } |
738 | ||
744692dc JZ |
739 | static int ext4_get_block_write(struct inode *inode, sector_t iblock, |
740 | struct buffer_head *bh_result, int create); | |
bfc1af65 | 741 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
742 | loff_t pos, unsigned len, unsigned flags, |
743 | struct page **pagep, void **fsdata) | |
ac27a0ec | 744 | { |
af5bc92d | 745 | struct inode *inode = mapping->host; |
1938a150 | 746 | int ret, needed_blocks; |
ac27a0ec DK |
747 | handle_t *handle; |
748 | int retries = 0; | |
af5bc92d | 749 | struct page *page; |
de9a55b8 | 750 | pgoff_t index; |
af5bc92d | 751 | unsigned from, to; |
bfc1af65 | 752 | |
9bffad1e | 753 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
754 | /* |
755 | * Reserve one block more for addition to orphan list in case | |
756 | * we allocate blocks but write fails for some reason | |
757 | */ | |
758 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 759 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
760 | from = pos & (PAGE_CACHE_SIZE - 1); |
761 | to = from + len; | |
ac27a0ec DK |
762 | |
763 | retry: | |
af5bc92d TT |
764 | handle = ext4_journal_start(inode, needed_blocks); |
765 | if (IS_ERR(handle)) { | |
766 | ret = PTR_ERR(handle); | |
767 | goto out; | |
7479d2b9 | 768 | } |
ac27a0ec | 769 | |
ebd3610b JK |
770 | /* We cannot recurse into the filesystem as the transaction is already |
771 | * started */ | |
772 | flags |= AOP_FLAG_NOFS; | |
773 | ||
54566b2c | 774 | page = grab_cache_page_write_begin(mapping, index, flags); |
cf108bca JK |
775 | if (!page) { |
776 | ext4_journal_stop(handle); | |
777 | ret = -ENOMEM; | |
778 | goto out; | |
779 | } | |
780 | *pagep = page; | |
781 | ||
744692dc | 782 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 783 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 784 | else |
6e1db88d | 785 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
bfc1af65 NP |
786 | |
787 | if (!ret && ext4_should_journal_data(inode)) { | |
ac27a0ec DK |
788 | ret = walk_page_buffers(handle, page_buffers(page), |
789 | from, to, NULL, do_journal_get_write_access); | |
790 | } | |
bfc1af65 NP |
791 | |
792 | if (ret) { | |
af5bc92d | 793 | unlock_page(page); |
af5bc92d | 794 | page_cache_release(page); |
ae4d5372 | 795 | /* |
6e1db88d | 796 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
797 | * outside i_size. Trim these off again. Don't need |
798 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
799 | * |
800 | * Add inode to orphan list in case we crash before | |
801 | * truncate finishes | |
ae4d5372 | 802 | */ |
ffacfa7a | 803 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
804 | ext4_orphan_add(handle, inode); |
805 | ||
806 | ext4_journal_stop(handle); | |
807 | if (pos + len > inode->i_size) { | |
b9a4207d | 808 | ext4_truncate_failed_write(inode); |
de9a55b8 | 809 | /* |
ffacfa7a | 810 | * If truncate failed early the inode might |
1938a150 AK |
811 | * still be on the orphan list; we need to |
812 | * make sure the inode is removed from the | |
813 | * orphan list in that case. | |
814 | */ | |
815 | if (inode->i_nlink) | |
816 | ext4_orphan_del(NULL, inode); | |
817 | } | |
bfc1af65 NP |
818 | } |
819 | ||
617ba13b | 820 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
ac27a0ec | 821 | goto retry; |
7479d2b9 | 822 | out: |
ac27a0ec DK |
823 | return ret; |
824 | } | |
825 | ||
bfc1af65 NP |
826 | /* For write_end() in data=journal mode */ |
827 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec DK |
828 | { |
829 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
830 | return 0; | |
831 | set_buffer_uptodate(bh); | |
0390131b | 832 | return ext4_handle_dirty_metadata(handle, NULL, bh); |
ac27a0ec DK |
833 | } |
834 | ||
f8514083 | 835 | static int ext4_generic_write_end(struct file *file, |
de9a55b8 TT |
836 | struct address_space *mapping, |
837 | loff_t pos, unsigned len, unsigned copied, | |
838 | struct page *page, void *fsdata) | |
f8514083 AK |
839 | { |
840 | int i_size_changed = 0; | |
841 | struct inode *inode = mapping->host; | |
842 | handle_t *handle = ext4_journal_current_handle(); | |
843 | ||
844 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
845 | ||
846 | /* | |
847 | * No need to use i_size_read() here, the i_size | |
848 | * cannot change under us because we hold i_mutex. | |
849 | * | |
850 | * But it's important to update i_size while still holding page lock: | |
851 | * page writeout could otherwise come in and zero beyond i_size. | |
852 | */ | |
853 | if (pos + copied > inode->i_size) { | |
854 | i_size_write(inode, pos + copied); | |
855 | i_size_changed = 1; | |
856 | } | |
857 | ||
858 | if (pos + copied > EXT4_I(inode)->i_disksize) { | |
859 | /* We need to mark inode dirty even if | |
860 | * new_i_size is less that inode->i_size | |
861 | * bu greater than i_disksize.(hint delalloc) | |
862 | */ | |
863 | ext4_update_i_disksize(inode, (pos + copied)); | |
864 | i_size_changed = 1; | |
865 | } | |
866 | unlock_page(page); | |
867 | page_cache_release(page); | |
868 | ||
869 | /* | |
870 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
871 | * makes the holding time of page lock longer. Second, it forces lock | |
872 | * ordering of page lock and transaction start for journaling | |
873 | * filesystems. | |
874 | */ | |
875 | if (i_size_changed) | |
876 | ext4_mark_inode_dirty(handle, inode); | |
877 | ||
878 | return copied; | |
879 | } | |
880 | ||
ac27a0ec DK |
881 | /* |
882 | * We need to pick up the new inode size which generic_commit_write gave us | |
883 | * `file' can be NULL - eg, when called from page_symlink(). | |
884 | * | |
617ba13b | 885 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
ac27a0ec DK |
886 | * buffers are managed internally. |
887 | */ | |
bfc1af65 | 888 | static int ext4_ordered_write_end(struct file *file, |
de9a55b8 TT |
889 | struct address_space *mapping, |
890 | loff_t pos, unsigned len, unsigned copied, | |
891 | struct page *page, void *fsdata) | |
ac27a0ec | 892 | { |
617ba13b | 893 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 894 | struct inode *inode = mapping->host; |
ac27a0ec DK |
895 | int ret = 0, ret2; |
896 | ||
9bffad1e | 897 | trace_ext4_ordered_write_end(inode, pos, len, copied); |
678aaf48 | 898 | ret = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
899 | |
900 | if (ret == 0) { | |
f8514083 | 901 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 902 | page, fsdata); |
f8a87d89 | 903 | copied = ret2; |
ffacfa7a | 904 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
905 | /* if we have allocated more blocks and copied |
906 | * less. We will have blocks allocated outside | |
907 | * inode->i_size. So truncate them | |
908 | */ | |
909 | ext4_orphan_add(handle, inode); | |
f8a87d89 RK |
910 | if (ret2 < 0) |
911 | ret = ret2; | |
ac27a0ec | 912 | } |
617ba13b | 913 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
914 | if (!ret) |
915 | ret = ret2; | |
bfc1af65 | 916 | |
f8514083 | 917 | if (pos + len > inode->i_size) { |
b9a4207d | 918 | ext4_truncate_failed_write(inode); |
de9a55b8 | 919 | /* |
ffacfa7a | 920 | * If truncate failed early the inode might still be |
f8514083 AK |
921 | * on the orphan list; we need to make sure the inode |
922 | * is removed from the orphan list in that case. | |
923 | */ | |
924 | if (inode->i_nlink) | |
925 | ext4_orphan_del(NULL, inode); | |
926 | } | |
927 | ||
928 | ||
bfc1af65 | 929 | return ret ? ret : copied; |
ac27a0ec DK |
930 | } |
931 | ||
bfc1af65 | 932 | static int ext4_writeback_write_end(struct file *file, |
de9a55b8 TT |
933 | struct address_space *mapping, |
934 | loff_t pos, unsigned len, unsigned copied, | |
935 | struct page *page, void *fsdata) | |
ac27a0ec | 936 | { |
617ba13b | 937 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 938 | struct inode *inode = mapping->host; |
ac27a0ec | 939 | int ret = 0, ret2; |
ac27a0ec | 940 | |
9bffad1e | 941 | trace_ext4_writeback_write_end(inode, pos, len, copied); |
f8514083 | 942 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 943 | page, fsdata); |
f8a87d89 | 944 | copied = ret2; |
ffacfa7a | 945 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
946 | /* if we have allocated more blocks and copied |
947 | * less. We will have blocks allocated outside | |
948 | * inode->i_size. So truncate them | |
949 | */ | |
950 | ext4_orphan_add(handle, inode); | |
951 | ||
f8a87d89 RK |
952 | if (ret2 < 0) |
953 | ret = ret2; | |
ac27a0ec | 954 | |
617ba13b | 955 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
956 | if (!ret) |
957 | ret = ret2; | |
bfc1af65 | 958 | |
f8514083 | 959 | if (pos + len > inode->i_size) { |
b9a4207d | 960 | ext4_truncate_failed_write(inode); |
de9a55b8 | 961 | /* |
ffacfa7a | 962 | * If truncate failed early the inode might still be |
f8514083 AK |
963 | * on the orphan list; we need to make sure the inode |
964 | * is removed from the orphan list in that case. | |
965 | */ | |
966 | if (inode->i_nlink) | |
967 | ext4_orphan_del(NULL, inode); | |
968 | } | |
969 | ||
bfc1af65 | 970 | return ret ? ret : copied; |
ac27a0ec DK |
971 | } |
972 | ||
bfc1af65 | 973 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
974 | struct address_space *mapping, |
975 | loff_t pos, unsigned len, unsigned copied, | |
976 | struct page *page, void *fsdata) | |
ac27a0ec | 977 | { |
617ba13b | 978 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 979 | struct inode *inode = mapping->host; |
ac27a0ec DK |
980 | int ret = 0, ret2; |
981 | int partial = 0; | |
bfc1af65 | 982 | unsigned from, to; |
cf17fea6 | 983 | loff_t new_i_size; |
ac27a0ec | 984 | |
9bffad1e | 985 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
986 | from = pos & (PAGE_CACHE_SIZE - 1); |
987 | to = from + len; | |
988 | ||
441c8508 CW |
989 | BUG_ON(!ext4_handle_valid(handle)); |
990 | ||
bfc1af65 NP |
991 | if (copied < len) { |
992 | if (!PageUptodate(page)) | |
993 | copied = 0; | |
994 | page_zero_new_buffers(page, from+copied, to); | |
995 | } | |
ac27a0ec DK |
996 | |
997 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
bfc1af65 | 998 | to, &partial, write_end_fn); |
ac27a0ec DK |
999 | if (!partial) |
1000 | SetPageUptodate(page); | |
cf17fea6 AK |
1001 | new_i_size = pos + copied; |
1002 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1003 | i_size_write(inode, pos+copied); |
19f5fb7a | 1004 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1005 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
cf17fea6 AK |
1006 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1007 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1008 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1009 | if (!ret) |
1010 | ret = ret2; | |
1011 | } | |
bfc1af65 | 1012 | |
cf108bca | 1013 | unlock_page(page); |
f8514083 | 1014 | page_cache_release(page); |
ffacfa7a | 1015 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1016 | /* if we have allocated more blocks and copied |
1017 | * less. We will have blocks allocated outside | |
1018 | * inode->i_size. So truncate them | |
1019 | */ | |
1020 | ext4_orphan_add(handle, inode); | |
1021 | ||
617ba13b | 1022 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1023 | if (!ret) |
1024 | ret = ret2; | |
f8514083 | 1025 | if (pos + len > inode->i_size) { |
b9a4207d | 1026 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1027 | /* |
ffacfa7a | 1028 | * If truncate failed early the inode might still be |
f8514083 AK |
1029 | * on the orphan list; we need to make sure the inode |
1030 | * is removed from the orphan list in that case. | |
1031 | */ | |
1032 | if (inode->i_nlink) | |
1033 | ext4_orphan_del(NULL, inode); | |
1034 | } | |
bfc1af65 NP |
1035 | |
1036 | return ret ? ret : copied; | |
ac27a0ec | 1037 | } |
d2a17637 | 1038 | |
9d0be502 | 1039 | /* |
7b415bf6 | 1040 | * Reserve a single cluster located at lblock |
9d0be502 | 1041 | */ |
7b415bf6 | 1042 | int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 1043 | { |
030ba6bc | 1044 | int retries = 0; |
60e58e0f | 1045 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1046 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 | 1047 | unsigned int md_needed; |
5dd4056d | 1048 | int ret; |
d2a17637 MC |
1049 | |
1050 | /* | |
1051 | * recalculate the amount of metadata blocks to reserve | |
1052 | * in order to allocate nrblocks | |
1053 | * worse case is one extent per block | |
1054 | */ | |
030ba6bc | 1055 | repeat: |
0637c6f4 | 1056 | spin_lock(&ei->i_block_reservation_lock); |
7b415bf6 AK |
1057 | md_needed = EXT4_NUM_B2C(sbi, |
1058 | ext4_calc_metadata_amount(inode, lblock)); | |
f8ec9d68 | 1059 | trace_ext4_da_reserve_space(inode, md_needed); |
0637c6f4 | 1060 | spin_unlock(&ei->i_block_reservation_lock); |
d2a17637 | 1061 | |
60e58e0f | 1062 | /* |
72b8ab9d ES |
1063 | * We will charge metadata quota at writeout time; this saves |
1064 | * us from metadata over-estimation, though we may go over by | |
1065 | * a small amount in the end. Here we just reserve for data. | |
60e58e0f | 1066 | */ |
7b415bf6 | 1067 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); |
5dd4056d CH |
1068 | if (ret) |
1069 | return ret; | |
72b8ab9d ES |
1070 | /* |
1071 | * We do still charge estimated metadata to the sb though; | |
1072 | * we cannot afford to run out of free blocks. | |
1073 | */ | |
e7d5f315 | 1074 | if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) { |
7b415bf6 | 1075 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
030ba6bc AK |
1076 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
1077 | yield(); | |
1078 | goto repeat; | |
1079 | } | |
d2a17637 MC |
1080 | return -ENOSPC; |
1081 | } | |
0637c6f4 | 1082 | spin_lock(&ei->i_block_reservation_lock); |
9d0be502 | 1083 | ei->i_reserved_data_blocks++; |
0637c6f4 TT |
1084 | ei->i_reserved_meta_blocks += md_needed; |
1085 | spin_unlock(&ei->i_block_reservation_lock); | |
39bc680a | 1086 | |
d2a17637 MC |
1087 | return 0; /* success */ |
1088 | } | |
1089 | ||
12219aea | 1090 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1091 | { |
1092 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1093 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1094 | |
cd213226 MC |
1095 | if (!to_free) |
1096 | return; /* Nothing to release, exit */ | |
1097 | ||
d2a17637 | 1098 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1099 | |
5a58ec87 | 1100 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1101 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1102 | /* |
0637c6f4 TT |
1103 | * if there aren't enough reserved blocks, then the |
1104 | * counter is messed up somewhere. Since this | |
1105 | * function is called from invalidate page, it's | |
1106 | * harmless to return without any action. | |
cd213226 | 1107 | */ |
0637c6f4 TT |
1108 | ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: " |
1109 | "ino %lu, to_free %d with only %d reserved " | |
1110 | "data blocks\n", inode->i_ino, to_free, | |
1111 | ei->i_reserved_data_blocks); | |
1112 | WARN_ON(1); | |
1113 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1114 | } |
0637c6f4 | 1115 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1116 | |
0637c6f4 TT |
1117 | if (ei->i_reserved_data_blocks == 0) { |
1118 | /* | |
1119 | * We can release all of the reserved metadata blocks | |
1120 | * only when we have written all of the delayed | |
1121 | * allocation blocks. | |
7b415bf6 AK |
1122 | * Note that in case of bigalloc, i_reserved_meta_blocks, |
1123 | * i_reserved_data_blocks, etc. refer to number of clusters. | |
0637c6f4 | 1124 | */ |
57042651 | 1125 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 1126 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 1127 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1128 | ei->i_da_metadata_calc_len = 0; |
0637c6f4 | 1129 | } |
d2a17637 | 1130 | |
72b8ab9d | 1131 | /* update fs dirty data blocks counter */ |
57042651 | 1132 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1133 | |
d2a17637 | 1134 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1135 | |
7b415bf6 | 1136 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1137 | } |
1138 | ||
1139 | static void ext4_da_page_release_reservation(struct page *page, | |
de9a55b8 | 1140 | unsigned long offset) |
d2a17637 MC |
1141 | { |
1142 | int to_release = 0; | |
1143 | struct buffer_head *head, *bh; | |
1144 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1145 | struct inode *inode = page->mapping->host; |
1146 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1147 | int num_clusters; | |
d2a17637 MC |
1148 | |
1149 | head = page_buffers(page); | |
1150 | bh = head; | |
1151 | do { | |
1152 | unsigned int next_off = curr_off + bh->b_size; | |
1153 | ||
1154 | if ((offset <= curr_off) && (buffer_delay(bh))) { | |
1155 | to_release++; | |
1156 | clear_buffer_delay(bh); | |
1157 | } | |
1158 | curr_off = next_off; | |
1159 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 AK |
1160 | |
1161 | /* If we have released all the blocks belonging to a cluster, then we | |
1162 | * need to release the reserved space for that cluster. */ | |
1163 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1164 | while (num_clusters > 0) { | |
1165 | ext4_fsblk_t lblk; | |
1166 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + | |
1167 | ((num_clusters - 1) << sbi->s_cluster_bits); | |
1168 | if (sbi->s_cluster_ratio == 1 || | |
1169 | !ext4_find_delalloc_cluster(inode, lblk, 1)) | |
1170 | ext4_da_release_space(inode, 1); | |
1171 | ||
1172 | num_clusters--; | |
1173 | } | |
d2a17637 | 1174 | } |
ac27a0ec | 1175 | |
64769240 AT |
1176 | /* |
1177 | * Delayed allocation stuff | |
1178 | */ | |
1179 | ||
64769240 AT |
1180 | /* |
1181 | * mpage_da_submit_io - walks through extent of pages and try to write | |
a1d6cc56 | 1182 | * them with writepage() call back |
64769240 AT |
1183 | * |
1184 | * @mpd->inode: inode | |
1185 | * @mpd->first_page: first page of the extent | |
1186 | * @mpd->next_page: page after the last page of the extent | |
64769240 AT |
1187 | * |
1188 | * By the time mpage_da_submit_io() is called we expect all blocks | |
1189 | * to be allocated. this may be wrong if allocation failed. | |
1190 | * | |
1191 | * As pages are already locked by write_cache_pages(), we can't use it | |
1192 | */ | |
1de3e3df TT |
1193 | static int mpage_da_submit_io(struct mpage_da_data *mpd, |
1194 | struct ext4_map_blocks *map) | |
64769240 | 1195 | { |
791b7f08 AK |
1196 | struct pagevec pvec; |
1197 | unsigned long index, end; | |
1198 | int ret = 0, err, nr_pages, i; | |
1199 | struct inode *inode = mpd->inode; | |
1200 | struct address_space *mapping = inode->i_mapping; | |
cb20d518 | 1201 | loff_t size = i_size_read(inode); |
3ecdb3a1 TT |
1202 | unsigned int len, block_start; |
1203 | struct buffer_head *bh, *page_bufs = NULL; | |
cb20d518 | 1204 | int journal_data = ext4_should_journal_data(inode); |
1de3e3df | 1205 | sector_t pblock = 0, cur_logical = 0; |
bd2d0210 | 1206 | struct ext4_io_submit io_submit; |
64769240 AT |
1207 | |
1208 | BUG_ON(mpd->next_page <= mpd->first_page); | |
bd2d0210 | 1209 | memset(&io_submit, 0, sizeof(io_submit)); |
791b7f08 AK |
1210 | /* |
1211 | * We need to start from the first_page to the next_page - 1 | |
1212 | * to make sure we also write the mapped dirty buffer_heads. | |
8dc207c0 | 1213 | * If we look at mpd->b_blocknr we would only be looking |
791b7f08 AK |
1214 | * at the currently mapped buffer_heads. |
1215 | */ | |
64769240 AT |
1216 | index = mpd->first_page; |
1217 | end = mpd->next_page - 1; | |
1218 | ||
791b7f08 | 1219 | pagevec_init(&pvec, 0); |
64769240 | 1220 | while (index <= end) { |
791b7f08 | 1221 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
64769240 AT |
1222 | if (nr_pages == 0) |
1223 | break; | |
1224 | for (i = 0; i < nr_pages; i++) { | |
97498956 | 1225 | int commit_write = 0, skip_page = 0; |
64769240 AT |
1226 | struct page *page = pvec.pages[i]; |
1227 | ||
791b7f08 AK |
1228 | index = page->index; |
1229 | if (index > end) | |
1230 | break; | |
cb20d518 TT |
1231 | |
1232 | if (index == size >> PAGE_CACHE_SHIFT) | |
1233 | len = size & ~PAGE_CACHE_MASK; | |
1234 | else | |
1235 | len = PAGE_CACHE_SIZE; | |
1de3e3df TT |
1236 | if (map) { |
1237 | cur_logical = index << (PAGE_CACHE_SHIFT - | |
1238 | inode->i_blkbits); | |
1239 | pblock = map->m_pblk + (cur_logical - | |
1240 | map->m_lblk); | |
1241 | } | |
791b7f08 AK |
1242 | index++; |
1243 | ||
1244 | BUG_ON(!PageLocked(page)); | |
1245 | BUG_ON(PageWriteback(page)); | |
1246 | ||
64769240 | 1247 | /* |
cb20d518 TT |
1248 | * If the page does not have buffers (for |
1249 | * whatever reason), try to create them using | |
a107e5a3 | 1250 | * __block_write_begin. If this fails, |
97498956 | 1251 | * skip the page and move on. |
64769240 | 1252 | */ |
cb20d518 | 1253 | if (!page_has_buffers(page)) { |
a107e5a3 | 1254 | if (__block_write_begin(page, 0, len, |
cb20d518 | 1255 | noalloc_get_block_write)) { |
97498956 | 1256 | skip_page: |
cb20d518 TT |
1257 | unlock_page(page); |
1258 | continue; | |
1259 | } | |
1260 | commit_write = 1; | |
1261 | } | |
64769240 | 1262 | |
3ecdb3a1 TT |
1263 | bh = page_bufs = page_buffers(page); |
1264 | block_start = 0; | |
64769240 | 1265 | do { |
1de3e3df | 1266 | if (!bh) |
97498956 | 1267 | goto skip_page; |
1de3e3df TT |
1268 | if (map && (cur_logical >= map->m_lblk) && |
1269 | (cur_logical <= (map->m_lblk + | |
1270 | (map->m_len - 1)))) { | |
29fa89d0 AK |
1271 | if (buffer_delay(bh)) { |
1272 | clear_buffer_delay(bh); | |
1273 | bh->b_blocknr = pblock; | |
29fa89d0 | 1274 | } |
1de3e3df TT |
1275 | if (buffer_unwritten(bh) || |
1276 | buffer_mapped(bh)) | |
1277 | BUG_ON(bh->b_blocknr != pblock); | |
1278 | if (map->m_flags & EXT4_MAP_UNINIT) | |
1279 | set_buffer_uninit(bh); | |
1280 | clear_buffer_unwritten(bh); | |
1281 | } | |
29fa89d0 | 1282 | |
97498956 | 1283 | /* skip page if block allocation undone */ |
1de3e3df | 1284 | if (buffer_delay(bh) || buffer_unwritten(bh)) |
97498956 | 1285 | skip_page = 1; |
3ecdb3a1 TT |
1286 | bh = bh->b_this_page; |
1287 | block_start += bh->b_size; | |
64769240 AT |
1288 | cur_logical++; |
1289 | pblock++; | |
1de3e3df TT |
1290 | } while (bh != page_bufs); |
1291 | ||
97498956 TT |
1292 | if (skip_page) |
1293 | goto skip_page; | |
cb20d518 TT |
1294 | |
1295 | if (commit_write) | |
1296 | /* mark the buffer_heads as dirty & uptodate */ | |
1297 | block_commit_write(page, 0, len); | |
1298 | ||
97498956 | 1299 | clear_page_dirty_for_io(page); |
bd2d0210 TT |
1300 | /* |
1301 | * Delalloc doesn't support data journalling, | |
1302 | * but eventually maybe we'll lift this | |
1303 | * restriction. | |
1304 | */ | |
1305 | if (unlikely(journal_data && PageChecked(page))) | |
cb20d518 | 1306 | err = __ext4_journalled_writepage(page, len); |
1449032b | 1307 | else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT)) |
bd2d0210 TT |
1308 | err = ext4_bio_write_page(&io_submit, page, |
1309 | len, mpd->wbc); | |
9dd75f1f TT |
1310 | else if (buffer_uninit(page_bufs)) { |
1311 | ext4_set_bh_endio(page_bufs, inode); | |
1312 | err = block_write_full_page_endio(page, | |
1313 | noalloc_get_block_write, | |
1314 | mpd->wbc, ext4_end_io_buffer_write); | |
1315 | } else | |
1449032b TT |
1316 | err = block_write_full_page(page, |
1317 | noalloc_get_block_write, mpd->wbc); | |
cb20d518 TT |
1318 | |
1319 | if (!err) | |
a1d6cc56 | 1320 | mpd->pages_written++; |
64769240 AT |
1321 | /* |
1322 | * In error case, we have to continue because | |
1323 | * remaining pages are still locked | |
64769240 AT |
1324 | */ |
1325 | if (ret == 0) | |
1326 | ret = err; | |
64769240 AT |
1327 | } |
1328 | pagevec_release(&pvec); | |
1329 | } | |
bd2d0210 | 1330 | ext4_io_submit(&io_submit); |
64769240 | 1331 | return ret; |
64769240 AT |
1332 | } |
1333 | ||
c7f5938a | 1334 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd) |
c4a0c46e AK |
1335 | { |
1336 | int nr_pages, i; | |
1337 | pgoff_t index, end; | |
1338 | struct pagevec pvec; | |
1339 | struct inode *inode = mpd->inode; | |
1340 | struct address_space *mapping = inode->i_mapping; | |
1341 | ||
c7f5938a CW |
1342 | index = mpd->first_page; |
1343 | end = mpd->next_page - 1; | |
c4a0c46e AK |
1344 | while (index <= end) { |
1345 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1346 | if (nr_pages == 0) | |
1347 | break; | |
1348 | for (i = 0; i < nr_pages; i++) { | |
1349 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1350 | if (page->index > end) |
c4a0c46e | 1351 | break; |
c4a0c46e AK |
1352 | BUG_ON(!PageLocked(page)); |
1353 | BUG_ON(PageWriteback(page)); | |
1354 | block_invalidatepage(page, 0); | |
1355 | ClearPageUptodate(page); | |
1356 | unlock_page(page); | |
1357 | } | |
9b1d0998 JK |
1358 | index = pvec.pages[nr_pages - 1]->index + 1; |
1359 | pagevec_release(&pvec); | |
c4a0c46e AK |
1360 | } |
1361 | return; | |
1362 | } | |
1363 | ||
df22291f AK |
1364 | static void ext4_print_free_blocks(struct inode *inode) |
1365 | { | |
1366 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1693918e | 1367 | printk(KERN_CRIT "Total free blocks count %lld\n", |
5dee5437 TT |
1368 | EXT4_C2B(EXT4_SB(inode->i_sb), |
1369 | ext4_count_free_clusters(inode->i_sb))); | |
1693918e TT |
1370 | printk(KERN_CRIT "Free/Dirty block details\n"); |
1371 | printk(KERN_CRIT "free_blocks=%lld\n", | |
57042651 TT |
1372 | (long long) EXT4_C2B(EXT4_SB(inode->i_sb), |
1373 | percpu_counter_sum(&sbi->s_freeclusters_counter))); | |
1693918e | 1374 | printk(KERN_CRIT "dirty_blocks=%lld\n", |
7b415bf6 AK |
1375 | (long long) EXT4_C2B(EXT4_SB(inode->i_sb), |
1376 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); | |
1693918e TT |
1377 | printk(KERN_CRIT "Block reservation details\n"); |
1378 | printk(KERN_CRIT "i_reserved_data_blocks=%u\n", | |
1379 | EXT4_I(inode)->i_reserved_data_blocks); | |
1380 | printk(KERN_CRIT "i_reserved_meta_blocks=%u\n", | |
1381 | EXT4_I(inode)->i_reserved_meta_blocks); | |
df22291f AK |
1382 | return; |
1383 | } | |
1384 | ||
64769240 | 1385 | /* |
5a87b7a5 TT |
1386 | * mpage_da_map_and_submit - go through given space, map them |
1387 | * if necessary, and then submit them for I/O | |
64769240 | 1388 | * |
8dc207c0 | 1389 | * @mpd - bh describing space |
64769240 AT |
1390 | * |
1391 | * The function skips space we know is already mapped to disk blocks. | |
1392 | * | |
64769240 | 1393 | */ |
5a87b7a5 | 1394 | static void mpage_da_map_and_submit(struct mpage_da_data *mpd) |
64769240 | 1395 | { |
2ac3b6e0 | 1396 | int err, blks, get_blocks_flags; |
1de3e3df | 1397 | struct ext4_map_blocks map, *mapp = NULL; |
2fa3cdfb TT |
1398 | sector_t next = mpd->b_blocknr; |
1399 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; | |
1400 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; | |
1401 | handle_t *handle = NULL; | |
64769240 AT |
1402 | |
1403 | /* | |
5a87b7a5 TT |
1404 | * If the blocks are mapped already, or we couldn't accumulate |
1405 | * any blocks, then proceed immediately to the submission stage. | |
2fa3cdfb | 1406 | */ |
5a87b7a5 TT |
1407 | if ((mpd->b_size == 0) || |
1408 | ((mpd->b_state & (1 << BH_Mapped)) && | |
1409 | !(mpd->b_state & (1 << BH_Delay)) && | |
1410 | !(mpd->b_state & (1 << BH_Unwritten)))) | |
1411 | goto submit_io; | |
2fa3cdfb TT |
1412 | |
1413 | handle = ext4_journal_current_handle(); | |
1414 | BUG_ON(!handle); | |
1415 | ||
79ffab34 | 1416 | /* |
79e83036 | 1417 | * Call ext4_map_blocks() to allocate any delayed allocation |
2ac3b6e0 TT |
1418 | * blocks, or to convert an uninitialized extent to be |
1419 | * initialized (in the case where we have written into | |
1420 | * one or more preallocated blocks). | |
1421 | * | |
1422 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to | |
1423 | * indicate that we are on the delayed allocation path. This | |
1424 | * affects functions in many different parts of the allocation | |
1425 | * call path. This flag exists primarily because we don't | |
79e83036 | 1426 | * want to change *many* call functions, so ext4_map_blocks() |
f2321097 | 1427 | * will set the EXT4_STATE_DELALLOC_RESERVED flag once the |
2ac3b6e0 TT |
1428 | * inode's allocation semaphore is taken. |
1429 | * | |
1430 | * If the blocks in questions were delalloc blocks, set | |
1431 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting | |
1432 | * variables are updated after the blocks have been allocated. | |
79ffab34 | 1433 | */ |
2ed88685 TT |
1434 | map.m_lblk = next; |
1435 | map.m_len = max_blocks; | |
1296cc85 | 1436 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE; |
744692dc JZ |
1437 | if (ext4_should_dioread_nolock(mpd->inode)) |
1438 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
2ac3b6e0 | 1439 | if (mpd->b_state & (1 << BH_Delay)) |
1296cc85 AK |
1440 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
1441 | ||
2ed88685 | 1442 | blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags); |
2fa3cdfb | 1443 | if (blks < 0) { |
e3570639 ES |
1444 | struct super_block *sb = mpd->inode->i_sb; |
1445 | ||
2fa3cdfb | 1446 | err = blks; |
ed5bde0b | 1447 | /* |
5a87b7a5 | 1448 | * If get block returns EAGAIN or ENOSPC and there |
97498956 TT |
1449 | * appears to be free blocks we will just let |
1450 | * mpage_da_submit_io() unlock all of the pages. | |
c4a0c46e AK |
1451 | */ |
1452 | if (err == -EAGAIN) | |
5a87b7a5 | 1453 | goto submit_io; |
df22291f | 1454 | |
5dee5437 | 1455 | if (err == -ENOSPC && ext4_count_free_clusters(sb)) { |
df22291f | 1456 | mpd->retval = err; |
5a87b7a5 | 1457 | goto submit_io; |
df22291f AK |
1458 | } |
1459 | ||
c4a0c46e | 1460 | /* |
ed5bde0b TT |
1461 | * get block failure will cause us to loop in |
1462 | * writepages, because a_ops->writepage won't be able | |
1463 | * to make progress. The page will be redirtied by | |
1464 | * writepage and writepages will again try to write | |
1465 | * the same. | |
c4a0c46e | 1466 | */ |
e3570639 ES |
1467 | if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
1468 | ext4_msg(sb, KERN_CRIT, | |
1469 | "delayed block allocation failed for inode %lu " | |
1470 | "at logical offset %llu with max blocks %zd " | |
1471 | "with error %d", mpd->inode->i_ino, | |
1472 | (unsigned long long) next, | |
1473 | mpd->b_size >> mpd->inode->i_blkbits, err); | |
1474 | ext4_msg(sb, KERN_CRIT, | |
1475 | "This should not happen!! Data will be lost\n"); | |
1476 | if (err == -ENOSPC) | |
1477 | ext4_print_free_blocks(mpd->inode); | |
030ba6bc | 1478 | } |
2fa3cdfb | 1479 | /* invalidate all the pages */ |
c7f5938a | 1480 | ext4_da_block_invalidatepages(mpd); |
e0fd9b90 CW |
1481 | |
1482 | /* Mark this page range as having been completed */ | |
1483 | mpd->io_done = 1; | |
5a87b7a5 | 1484 | return; |
c4a0c46e | 1485 | } |
2fa3cdfb TT |
1486 | BUG_ON(blks == 0); |
1487 | ||
1de3e3df | 1488 | mapp = ↦ |
2ed88685 TT |
1489 | if (map.m_flags & EXT4_MAP_NEW) { |
1490 | struct block_device *bdev = mpd->inode->i_sb->s_bdev; | |
1491 | int i; | |
64769240 | 1492 | |
2ed88685 TT |
1493 | for (i = 0; i < map.m_len; i++) |
1494 | unmap_underlying_metadata(bdev, map.m_pblk + i); | |
64769240 | 1495 | |
decbd919 TT |
1496 | if (ext4_should_order_data(mpd->inode)) { |
1497 | err = ext4_jbd2_file_inode(handle, mpd->inode); | |
1498 | if (err) | |
1499 | /* Only if the journal is aborted */ | |
1500 | return; | |
1501 | } | |
2fa3cdfb TT |
1502 | } |
1503 | ||
1504 | /* | |
03f5d8bc | 1505 | * Update on-disk size along with block allocation. |
2fa3cdfb TT |
1506 | */ |
1507 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; | |
1508 | if (disksize > i_size_read(mpd->inode)) | |
1509 | disksize = i_size_read(mpd->inode); | |
1510 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { | |
1511 | ext4_update_i_disksize(mpd->inode, disksize); | |
5a87b7a5 TT |
1512 | err = ext4_mark_inode_dirty(handle, mpd->inode); |
1513 | if (err) | |
1514 | ext4_error(mpd->inode->i_sb, | |
1515 | "Failed to mark inode %lu dirty", | |
1516 | mpd->inode->i_ino); | |
2fa3cdfb TT |
1517 | } |
1518 | ||
5a87b7a5 | 1519 | submit_io: |
1de3e3df | 1520 | mpage_da_submit_io(mpd, mapp); |
5a87b7a5 | 1521 | mpd->io_done = 1; |
64769240 AT |
1522 | } |
1523 | ||
bf068ee2 AK |
1524 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
1525 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
1526 | |
1527 | /* | |
1528 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
1529 | * | |
1530 | * @mpd->lbh - extent of blocks | |
1531 | * @logical - logical number of the block in the file | |
1532 | * @bh - bh of the block (used to access block's state) | |
1533 | * | |
1534 | * the function is used to collect contig. blocks in same state | |
1535 | */ | |
1536 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, | |
8dc207c0 TT |
1537 | sector_t logical, size_t b_size, |
1538 | unsigned long b_state) | |
64769240 | 1539 | { |
64769240 | 1540 | sector_t next; |
8dc207c0 | 1541 | int nrblocks = mpd->b_size >> mpd->inode->i_blkbits; |
64769240 | 1542 | |
c445e3e0 ES |
1543 | /* |
1544 | * XXX Don't go larger than mballoc is willing to allocate | |
1545 | * This is a stopgap solution. We eventually need to fold | |
1546 | * mpage_da_submit_io() into this function and then call | |
79e83036 | 1547 | * ext4_map_blocks() multiple times in a loop |
c445e3e0 ES |
1548 | */ |
1549 | if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize) | |
1550 | goto flush_it; | |
1551 | ||
525f4ed8 | 1552 | /* check if thereserved journal credits might overflow */ |
12e9b892 | 1553 | if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) { |
525f4ed8 MC |
1554 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { |
1555 | /* | |
1556 | * With non-extent format we are limited by the journal | |
1557 | * credit available. Total credit needed to insert | |
1558 | * nrblocks contiguous blocks is dependent on the | |
1559 | * nrblocks. So limit nrblocks. | |
1560 | */ | |
1561 | goto flush_it; | |
1562 | } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) > | |
1563 | EXT4_MAX_TRANS_DATA) { | |
1564 | /* | |
1565 | * Adding the new buffer_head would make it cross the | |
1566 | * allowed limit for which we have journal credit | |
1567 | * reserved. So limit the new bh->b_size | |
1568 | */ | |
1569 | b_size = (EXT4_MAX_TRANS_DATA - nrblocks) << | |
1570 | mpd->inode->i_blkbits; | |
1571 | /* we will do mpage_da_submit_io in the next loop */ | |
1572 | } | |
1573 | } | |
64769240 AT |
1574 | /* |
1575 | * First block in the extent | |
1576 | */ | |
8dc207c0 TT |
1577 | if (mpd->b_size == 0) { |
1578 | mpd->b_blocknr = logical; | |
1579 | mpd->b_size = b_size; | |
1580 | mpd->b_state = b_state & BH_FLAGS; | |
64769240 AT |
1581 | return; |
1582 | } | |
1583 | ||
8dc207c0 | 1584 | next = mpd->b_blocknr + nrblocks; |
64769240 AT |
1585 | /* |
1586 | * Can we merge the block to our big extent? | |
1587 | */ | |
8dc207c0 TT |
1588 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
1589 | mpd->b_size += b_size; | |
64769240 AT |
1590 | return; |
1591 | } | |
1592 | ||
525f4ed8 | 1593 | flush_it: |
64769240 AT |
1594 | /* |
1595 | * We couldn't merge the block to our extent, so we | |
1596 | * need to flush current extent and start new one | |
1597 | */ | |
5a87b7a5 | 1598 | mpage_da_map_and_submit(mpd); |
a1d6cc56 | 1599 | return; |
64769240 AT |
1600 | } |
1601 | ||
c364b22c | 1602 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1603 | { |
c364b22c | 1604 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1605 | } |
1606 | ||
64769240 | 1607 | /* |
b920c755 TT |
1608 | * This is a special get_blocks_t callback which is used by |
1609 | * ext4_da_write_begin(). It will either return mapped block or | |
1610 | * reserve space for a single block. | |
29fa89d0 AK |
1611 | * |
1612 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1613 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1614 | * | |
1615 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1616 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1617 | * initialized properly. | |
64769240 AT |
1618 | */ |
1619 | static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, | |
2ed88685 | 1620 | struct buffer_head *bh, int create) |
64769240 | 1621 | { |
2ed88685 | 1622 | struct ext4_map_blocks map; |
64769240 | 1623 | int ret = 0; |
33b9817e AK |
1624 | sector_t invalid_block = ~((sector_t) 0xffff); |
1625 | ||
1626 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1627 | invalid_block = ~0; | |
64769240 AT |
1628 | |
1629 | BUG_ON(create == 0); | |
2ed88685 TT |
1630 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1631 | ||
1632 | map.m_lblk = iblock; | |
1633 | map.m_len = 1; | |
64769240 AT |
1634 | |
1635 | /* | |
1636 | * first, we need to know whether the block is allocated already | |
1637 | * preallocated blocks are unmapped but should treated | |
1638 | * the same as allocated blocks. | |
1639 | */ | |
2ed88685 TT |
1640 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
1641 | if (ret < 0) | |
1642 | return ret; | |
1643 | if (ret == 0) { | |
1644 | if (buffer_delay(bh)) | |
1645 | return 0; /* Not sure this could or should happen */ | |
64769240 | 1646 | /* |
ebdec241 | 1647 | * XXX: __block_write_begin() unmaps passed block, is it OK? |
64769240 | 1648 | */ |
7b415bf6 AK |
1649 | /* If the block was allocated from previously allocated cluster, |
1650 | * then we dont need to reserve it again. */ | |
1651 | if (!(map.m_flags & EXT4_MAP_FROM_CLUSTER)) { | |
1652 | ret = ext4_da_reserve_space(inode, iblock); | |
1653 | if (ret) | |
1654 | /* not enough space to reserve */ | |
1655 | return ret; | |
1656 | } | |
d2a17637 | 1657 | |
2ed88685 TT |
1658 | map_bh(bh, inode->i_sb, invalid_block); |
1659 | set_buffer_new(bh); | |
1660 | set_buffer_delay(bh); | |
1661 | return 0; | |
64769240 AT |
1662 | } |
1663 | ||
2ed88685 TT |
1664 | map_bh(bh, inode->i_sb, map.m_pblk); |
1665 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
1666 | ||
1667 | if (buffer_unwritten(bh)) { | |
1668 | /* A delayed write to unwritten bh should be marked | |
1669 | * new and mapped. Mapped ensures that we don't do | |
1670 | * get_block multiple times when we write to the same | |
1671 | * offset and new ensures that we do proper zero out | |
1672 | * for partial write. | |
1673 | */ | |
1674 | set_buffer_new(bh); | |
c8205636 | 1675 | set_buffer_mapped(bh); |
2ed88685 TT |
1676 | } |
1677 | return 0; | |
64769240 | 1678 | } |
61628a3f | 1679 | |
b920c755 TT |
1680 | /* |
1681 | * This function is used as a standard get_block_t calback function | |
1682 | * when there is no desire to allocate any blocks. It is used as a | |
ebdec241 | 1683 | * callback function for block_write_begin() and block_write_full_page(). |
206f7ab4 | 1684 | * These functions should only try to map a single block at a time. |
b920c755 TT |
1685 | * |
1686 | * Since this function doesn't do block allocations even if the caller | |
1687 | * requests it by passing in create=1, it is critically important that | |
1688 | * any caller checks to make sure that any buffer heads are returned | |
1689 | * by this function are either all already mapped or marked for | |
206f7ab4 CH |
1690 | * delayed allocation before calling block_write_full_page(). Otherwise, |
1691 | * b_blocknr could be left unitialized, and the page write functions will | |
1692 | * be taken by surprise. | |
b920c755 TT |
1693 | */ |
1694 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, | |
f0e6c985 AK |
1695 | struct buffer_head *bh_result, int create) |
1696 | { | |
a2dc52b5 | 1697 | BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); |
2ed88685 | 1698 | return _ext4_get_block(inode, iblock, bh_result, 0); |
61628a3f MC |
1699 | } |
1700 | ||
62e086be AK |
1701 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1702 | { | |
1703 | get_bh(bh); | |
1704 | return 0; | |
1705 | } | |
1706 | ||
1707 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1708 | { | |
1709 | put_bh(bh); | |
1710 | return 0; | |
1711 | } | |
1712 | ||
1713 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1714 | unsigned int len) |
1715 | { | |
1716 | struct address_space *mapping = page->mapping; | |
1717 | struct inode *inode = mapping->host; | |
1718 | struct buffer_head *page_bufs; | |
1719 | handle_t *handle = NULL; | |
1720 | int ret = 0; | |
1721 | int err; | |
1722 | ||
cb20d518 | 1723 | ClearPageChecked(page); |
62e086be AK |
1724 | page_bufs = page_buffers(page); |
1725 | BUG_ON(!page_bufs); | |
1726 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one); | |
1727 | /* As soon as we unlock the page, it can go away, but we have | |
1728 | * references to buffers so we are safe */ | |
1729 | unlock_page(page); | |
1730 | ||
1731 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); | |
1732 | if (IS_ERR(handle)) { | |
1733 | ret = PTR_ERR(handle); | |
1734 | goto out; | |
1735 | } | |
1736 | ||
441c8508 CW |
1737 | BUG_ON(!ext4_handle_valid(handle)); |
1738 | ||
62e086be AK |
1739 | ret = walk_page_buffers(handle, page_bufs, 0, len, NULL, |
1740 | do_journal_get_write_access); | |
1741 | ||
1742 | err = walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1743 | write_end_fn); | |
1744 | if (ret == 0) | |
1745 | ret = err; | |
2d859db3 | 1746 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1747 | err = ext4_journal_stop(handle); |
1748 | if (!ret) | |
1749 | ret = err; | |
1750 | ||
1751 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one); | |
19f5fb7a | 1752 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be AK |
1753 | out: |
1754 | return ret; | |
1755 | } | |
1756 | ||
744692dc JZ |
1757 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode); |
1758 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate); | |
1759 | ||
61628a3f | 1760 | /* |
43ce1d23 AK |
1761 | * Note that we don't need to start a transaction unless we're journaling data |
1762 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1763 | * need to file the inode to the transaction's list in ordered mode because if | |
1764 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1765 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1766 | * transaction the data will hit the disk. In case we are journaling data, we |
1767 | * cannot start transaction directly because transaction start ranks above page | |
1768 | * lock so we have to do some magic. | |
1769 | * | |
b920c755 TT |
1770 | * This function can get called via... |
1771 | * - ext4_da_writepages after taking page lock (have journal handle) | |
1772 | * - journal_submit_inode_data_buffers (no journal handle) | |
1773 | * - shrink_page_list via pdflush (no journal handle) | |
1774 | * - grab_page_cache when doing write_begin (have journal handle) | |
43ce1d23 AK |
1775 | * |
1776 | * We don't do any block allocation in this function. If we have page with | |
1777 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1778 | * is important for mmaped based write. So if we do with blocksize 1K | |
1779 | * truncate(f, 1024); | |
1780 | * a = mmap(f, 0, 4096); | |
1781 | * a[0] = 'a'; | |
1782 | * truncate(f, 4096); | |
1783 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
1784 | * but other bufer_heads would be unmapped but dirty(dirty done via the | |
1785 | * do_wp_page). So writepage should write the first block. If we modify | |
1786 | * the mmap area beyond 1024 we will again get a page_fault and the | |
1787 | * page_mkwrite callback will do the block allocation and mark the | |
1788 | * buffer_heads mapped. | |
1789 | * | |
1790 | * We redirty the page if we have any buffer_heads that is either delay or | |
1791 | * unwritten in the page. | |
1792 | * | |
1793 | * We can get recursively called as show below. | |
1794 | * | |
1795 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1796 | * ext4_writepage() | |
1797 | * | |
1798 | * But since we don't do any block allocation we should not deadlock. | |
1799 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 1800 | */ |
43ce1d23 | 1801 | static int ext4_writepage(struct page *page, |
62e086be | 1802 | struct writeback_control *wbc) |
64769240 | 1803 | { |
a42afc5f | 1804 | int ret = 0, commit_write = 0; |
61628a3f | 1805 | loff_t size; |
498e5f24 | 1806 | unsigned int len; |
744692dc | 1807 | struct buffer_head *page_bufs = NULL; |
61628a3f MC |
1808 | struct inode *inode = page->mapping->host; |
1809 | ||
a9c667f8 | 1810 | trace_ext4_writepage(page); |
f0e6c985 AK |
1811 | size = i_size_read(inode); |
1812 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
1813 | len = size & ~PAGE_CACHE_MASK; | |
1814 | else | |
1815 | len = PAGE_CACHE_SIZE; | |
64769240 | 1816 | |
a42afc5f TT |
1817 | /* |
1818 | * If the page does not have buffers (for whatever reason), | |
a107e5a3 | 1819 | * try to create them using __block_write_begin. If this |
a42afc5f TT |
1820 | * fails, redirty the page and move on. |
1821 | */ | |
b1142e8f | 1822 | if (!page_has_buffers(page)) { |
a107e5a3 | 1823 | if (__block_write_begin(page, 0, len, |
a42afc5f TT |
1824 | noalloc_get_block_write)) { |
1825 | redirty_page: | |
f0e6c985 AK |
1826 | redirty_page_for_writepage(wbc, page); |
1827 | unlock_page(page); | |
1828 | return 0; | |
1829 | } | |
a42afc5f TT |
1830 | commit_write = 1; |
1831 | } | |
1832 | page_bufs = page_buffers(page); | |
1833 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, | |
1834 | ext4_bh_delay_or_unwritten)) { | |
f0e6c985 | 1835 | /* |
b1142e8f TT |
1836 | * We don't want to do block allocation, so redirty |
1837 | * the page and return. We may reach here when we do | |
1838 | * a journal commit via journal_submit_inode_data_buffers. | |
1839 | * We can also reach here via shrink_page_list | |
f0e6c985 | 1840 | */ |
a42afc5f TT |
1841 | goto redirty_page; |
1842 | } | |
1843 | if (commit_write) | |
ed9b3e33 | 1844 | /* now mark the buffer_heads as dirty and uptodate */ |
b767e78a | 1845 | block_commit_write(page, 0, len); |
64769240 | 1846 | |
cb20d518 | 1847 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
1848 | /* |
1849 | * It's mmapped pagecache. Add buffers and journal it. There | |
1850 | * doesn't seem much point in redirtying the page here. | |
1851 | */ | |
3f0ca309 | 1852 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 1853 | |
a42afc5f | 1854 | if (buffer_uninit(page_bufs)) { |
744692dc JZ |
1855 | ext4_set_bh_endio(page_bufs, inode); |
1856 | ret = block_write_full_page_endio(page, noalloc_get_block_write, | |
1857 | wbc, ext4_end_io_buffer_write); | |
1858 | } else | |
b920c755 TT |
1859 | ret = block_write_full_page(page, noalloc_get_block_write, |
1860 | wbc); | |
64769240 | 1861 | |
64769240 AT |
1862 | return ret; |
1863 | } | |
1864 | ||
61628a3f | 1865 | /* |
525f4ed8 | 1866 | * This is called via ext4_da_writepages() to |
25985edc | 1867 | * calculate the total number of credits to reserve to fit |
525f4ed8 MC |
1868 | * a single extent allocation into a single transaction, |
1869 | * ext4_da_writpeages() will loop calling this before | |
1870 | * the block allocation. | |
61628a3f | 1871 | */ |
525f4ed8 MC |
1872 | |
1873 | static int ext4_da_writepages_trans_blocks(struct inode *inode) | |
1874 | { | |
1875 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
1876 | ||
1877 | /* | |
1878 | * With non-extent format the journal credit needed to | |
1879 | * insert nrblocks contiguous block is dependent on | |
1880 | * number of contiguous block. So we will limit | |
1881 | * number of contiguous block to a sane value | |
1882 | */ | |
12e9b892 | 1883 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) && |
525f4ed8 MC |
1884 | (max_blocks > EXT4_MAX_TRANS_DATA)) |
1885 | max_blocks = EXT4_MAX_TRANS_DATA; | |
1886 | ||
1887 | return ext4_chunk_trans_blocks(inode, max_blocks); | |
1888 | } | |
61628a3f | 1889 | |
8e48dcfb TT |
1890 | /* |
1891 | * write_cache_pages_da - walk the list of dirty pages of the given | |
8eb9e5ce | 1892 | * address space and accumulate pages that need writing, and call |
168fc022 TT |
1893 | * mpage_da_map_and_submit to map a single contiguous memory region |
1894 | * and then write them. | |
8e48dcfb TT |
1895 | */ |
1896 | static int write_cache_pages_da(struct address_space *mapping, | |
1897 | struct writeback_control *wbc, | |
72f84e65 ES |
1898 | struct mpage_da_data *mpd, |
1899 | pgoff_t *done_index) | |
8e48dcfb | 1900 | { |
4f01b02c | 1901 | struct buffer_head *bh, *head; |
168fc022 | 1902 | struct inode *inode = mapping->host; |
4f01b02c TT |
1903 | struct pagevec pvec; |
1904 | unsigned int nr_pages; | |
1905 | sector_t logical; | |
1906 | pgoff_t index, end; | |
1907 | long nr_to_write = wbc->nr_to_write; | |
1908 | int i, tag, ret = 0; | |
8e48dcfb | 1909 | |
168fc022 TT |
1910 | memset(mpd, 0, sizeof(struct mpage_da_data)); |
1911 | mpd->wbc = wbc; | |
1912 | mpd->inode = inode; | |
8e48dcfb TT |
1913 | pagevec_init(&pvec, 0); |
1914 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
1915 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
1916 | ||
6e6938b6 | 1917 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
1918 | tag = PAGECACHE_TAG_TOWRITE; |
1919 | else | |
1920 | tag = PAGECACHE_TAG_DIRTY; | |
1921 | ||
72f84e65 | 1922 | *done_index = index; |
4f01b02c | 1923 | while (index <= end) { |
5b41d924 | 1924 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
1925 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
1926 | if (nr_pages == 0) | |
4f01b02c | 1927 | return 0; |
8e48dcfb TT |
1928 | |
1929 | for (i = 0; i < nr_pages; i++) { | |
1930 | struct page *page = pvec.pages[i]; | |
1931 | ||
1932 | /* | |
1933 | * At this point, the page may be truncated or | |
1934 | * invalidated (changing page->mapping to NULL), or | |
1935 | * even swizzled back from swapper_space to tmpfs file | |
1936 | * mapping. However, page->index will not change | |
1937 | * because we have a reference on the page. | |
1938 | */ | |
4f01b02c TT |
1939 | if (page->index > end) |
1940 | goto out; | |
8e48dcfb | 1941 | |
72f84e65 ES |
1942 | *done_index = page->index + 1; |
1943 | ||
78aaced3 TT |
1944 | /* |
1945 | * If we can't merge this page, and we have | |
1946 | * accumulated an contiguous region, write it | |
1947 | */ | |
1948 | if ((mpd->next_page != page->index) && | |
1949 | (mpd->next_page != mpd->first_page)) { | |
1950 | mpage_da_map_and_submit(mpd); | |
1951 | goto ret_extent_tail; | |
1952 | } | |
1953 | ||
8e48dcfb TT |
1954 | lock_page(page); |
1955 | ||
1956 | /* | |
4f01b02c TT |
1957 | * If the page is no longer dirty, or its |
1958 | * mapping no longer corresponds to inode we | |
1959 | * are writing (which means it has been | |
1960 | * truncated or invalidated), or the page is | |
1961 | * already under writeback and we are not | |
1962 | * doing a data integrity writeback, skip the page | |
8e48dcfb | 1963 | */ |
4f01b02c TT |
1964 | if (!PageDirty(page) || |
1965 | (PageWriteback(page) && | |
1966 | (wbc->sync_mode == WB_SYNC_NONE)) || | |
1967 | unlikely(page->mapping != mapping)) { | |
8e48dcfb TT |
1968 | unlock_page(page); |
1969 | continue; | |
1970 | } | |
1971 | ||
7cb1a535 | 1972 | wait_on_page_writeback(page); |
8e48dcfb | 1973 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 1974 | |
168fc022 | 1975 | if (mpd->next_page != page->index) |
8eb9e5ce | 1976 | mpd->first_page = page->index; |
8eb9e5ce TT |
1977 | mpd->next_page = page->index + 1; |
1978 | logical = (sector_t) page->index << | |
1979 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1980 | ||
1981 | if (!page_has_buffers(page)) { | |
4f01b02c TT |
1982 | mpage_add_bh_to_extent(mpd, logical, |
1983 | PAGE_CACHE_SIZE, | |
8eb9e5ce | 1984 | (1 << BH_Dirty) | (1 << BH_Uptodate)); |
4f01b02c TT |
1985 | if (mpd->io_done) |
1986 | goto ret_extent_tail; | |
8eb9e5ce TT |
1987 | } else { |
1988 | /* | |
4f01b02c TT |
1989 | * Page with regular buffer heads, |
1990 | * just add all dirty ones | |
8eb9e5ce TT |
1991 | */ |
1992 | head = page_buffers(page); | |
1993 | bh = head; | |
1994 | do { | |
1995 | BUG_ON(buffer_locked(bh)); | |
1996 | /* | |
1997 | * We need to try to allocate | |
1998 | * unmapped blocks in the same page. | |
1999 | * Otherwise we won't make progress | |
2000 | * with the page in ext4_writepage | |
2001 | */ | |
2002 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { | |
2003 | mpage_add_bh_to_extent(mpd, logical, | |
2004 | bh->b_size, | |
2005 | bh->b_state); | |
4f01b02c TT |
2006 | if (mpd->io_done) |
2007 | goto ret_extent_tail; | |
8eb9e5ce TT |
2008 | } else if (buffer_dirty(bh) && (buffer_mapped(bh))) { |
2009 | /* | |
4f01b02c TT |
2010 | * mapped dirty buffer. We need |
2011 | * to update the b_state | |
2012 | * because we look at b_state | |
2013 | * in mpage_da_map_blocks. We | |
2014 | * don't update b_size because | |
2015 | * if we find an unmapped | |
2016 | * buffer_head later we need to | |
2017 | * use the b_state flag of that | |
2018 | * buffer_head. | |
8eb9e5ce TT |
2019 | */ |
2020 | if (mpd->b_size == 0) | |
2021 | mpd->b_state = bh->b_state & BH_FLAGS; | |
2022 | } | |
2023 | logical++; | |
2024 | } while ((bh = bh->b_this_page) != head); | |
8e48dcfb TT |
2025 | } |
2026 | ||
2027 | if (nr_to_write > 0) { | |
2028 | nr_to_write--; | |
2029 | if (nr_to_write == 0 && | |
4f01b02c | 2030 | wbc->sync_mode == WB_SYNC_NONE) |
8e48dcfb TT |
2031 | /* |
2032 | * We stop writing back only if we are | |
2033 | * not doing integrity sync. In case of | |
2034 | * integrity sync we have to keep going | |
2035 | * because someone may be concurrently | |
2036 | * dirtying pages, and we might have | |
2037 | * synced a lot of newly appeared dirty | |
2038 | * pages, but have not synced all of the | |
2039 | * old dirty pages. | |
2040 | */ | |
4f01b02c | 2041 | goto out; |
8e48dcfb TT |
2042 | } |
2043 | } | |
2044 | pagevec_release(&pvec); | |
2045 | cond_resched(); | |
2046 | } | |
4f01b02c TT |
2047 | return 0; |
2048 | ret_extent_tail: | |
2049 | ret = MPAGE_DA_EXTENT_TAIL; | |
8eb9e5ce TT |
2050 | out: |
2051 | pagevec_release(&pvec); | |
2052 | cond_resched(); | |
8e48dcfb TT |
2053 | return ret; |
2054 | } | |
2055 | ||
2056 | ||
64769240 | 2057 | static int ext4_da_writepages(struct address_space *mapping, |
a1d6cc56 | 2058 | struct writeback_control *wbc) |
64769240 | 2059 | { |
22208ded AK |
2060 | pgoff_t index; |
2061 | int range_whole = 0; | |
61628a3f | 2062 | handle_t *handle = NULL; |
df22291f | 2063 | struct mpage_da_data mpd; |
5e745b04 | 2064 | struct inode *inode = mapping->host; |
498e5f24 | 2065 | int pages_written = 0; |
55138e0b | 2066 | unsigned int max_pages; |
2acf2c26 | 2067 | int range_cyclic, cycled = 1, io_done = 0; |
55138e0b TT |
2068 | int needed_blocks, ret = 0; |
2069 | long desired_nr_to_write, nr_to_writebump = 0; | |
de89de6e | 2070 | loff_t range_start = wbc->range_start; |
5e745b04 | 2071 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
72f84e65 | 2072 | pgoff_t done_index = 0; |
5b41d924 | 2073 | pgoff_t end; |
61628a3f | 2074 | |
9bffad1e | 2075 | trace_ext4_da_writepages(inode, wbc); |
ba80b101 | 2076 | |
61628a3f MC |
2077 | /* |
2078 | * No pages to write? This is mainly a kludge to avoid starting | |
2079 | * a transaction for special inodes like journal inode on last iput() | |
2080 | * because that could violate lock ordering on umount | |
2081 | */ | |
a1d6cc56 | 2082 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
61628a3f | 2083 | return 0; |
2a21e37e TT |
2084 | |
2085 | /* | |
2086 | * If the filesystem has aborted, it is read-only, so return | |
2087 | * right away instead of dumping stack traces later on that | |
2088 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2089 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e TT |
2090 | * the latter could be true if the filesystem is mounted |
2091 | * read-only, and in that case, ext4_da_writepages should | |
2092 | * *never* be called, so if that ever happens, we would want | |
2093 | * the stack trace. | |
2094 | */ | |
4ab2f15b | 2095 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
2a21e37e TT |
2096 | return -EROFS; |
2097 | ||
22208ded AK |
2098 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2099 | range_whole = 1; | |
61628a3f | 2100 | |
2acf2c26 AK |
2101 | range_cyclic = wbc->range_cyclic; |
2102 | if (wbc->range_cyclic) { | |
22208ded | 2103 | index = mapping->writeback_index; |
2acf2c26 AK |
2104 | if (index) |
2105 | cycled = 0; | |
2106 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2107 | wbc->range_end = LLONG_MAX; | |
2108 | wbc->range_cyclic = 0; | |
5b41d924 ES |
2109 | end = -1; |
2110 | } else { | |
22208ded | 2111 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
5b41d924 ES |
2112 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
2113 | } | |
a1d6cc56 | 2114 | |
55138e0b TT |
2115 | /* |
2116 | * This works around two forms of stupidity. The first is in | |
2117 | * the writeback code, which caps the maximum number of pages | |
2118 | * written to be 1024 pages. This is wrong on multiple | |
2119 | * levels; different architectues have a different page size, | |
2120 | * which changes the maximum amount of data which gets | |
2121 | * written. Secondly, 4 megabytes is way too small. XFS | |
2122 | * forces this value to be 16 megabytes by multiplying | |
2123 | * nr_to_write parameter by four, and then relies on its | |
2124 | * allocator to allocate larger extents to make them | |
2125 | * contiguous. Unfortunately this brings us to the second | |
2126 | * stupidity, which is that ext4's mballoc code only allocates | |
2127 | * at most 2048 blocks. So we force contiguous writes up to | |
2128 | * the number of dirty blocks in the inode, or | |
2129 | * sbi->max_writeback_mb_bump whichever is smaller. | |
2130 | */ | |
2131 | max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); | |
b443e733 ES |
2132 | if (!range_cyclic && range_whole) { |
2133 | if (wbc->nr_to_write == LONG_MAX) | |
2134 | desired_nr_to_write = wbc->nr_to_write; | |
2135 | else | |
2136 | desired_nr_to_write = wbc->nr_to_write * 8; | |
2137 | } else | |
55138e0b TT |
2138 | desired_nr_to_write = ext4_num_dirty_pages(inode, index, |
2139 | max_pages); | |
2140 | if (desired_nr_to_write > max_pages) | |
2141 | desired_nr_to_write = max_pages; | |
2142 | ||
2143 | if (wbc->nr_to_write < desired_nr_to_write) { | |
2144 | nr_to_writebump = desired_nr_to_write - wbc->nr_to_write; | |
2145 | wbc->nr_to_write = desired_nr_to_write; | |
2146 | } | |
2147 | ||
2acf2c26 | 2148 | retry: |
6e6938b6 | 2149 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2150 | tag_pages_for_writeback(mapping, index, end); |
2151 | ||
22208ded | 2152 | while (!ret && wbc->nr_to_write > 0) { |
a1d6cc56 AK |
2153 | |
2154 | /* | |
2155 | * we insert one extent at a time. So we need | |
2156 | * credit needed for single extent allocation. | |
2157 | * journalled mode is currently not supported | |
2158 | * by delalloc | |
2159 | */ | |
2160 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2161 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2162 | |
61628a3f MC |
2163 | /* start a new transaction*/ |
2164 | handle = ext4_journal_start(inode, needed_blocks); | |
2165 | if (IS_ERR(handle)) { | |
2166 | ret = PTR_ERR(handle); | |
1693918e | 2167 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2168 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2169 | wbc->nr_to_write, inode->i_ino, ret); |
61628a3f MC |
2170 | goto out_writepages; |
2171 | } | |
f63e6005 TT |
2172 | |
2173 | /* | |
8eb9e5ce | 2174 | * Now call write_cache_pages_da() to find the next |
f63e6005 | 2175 | * contiguous region of logical blocks that need |
8eb9e5ce | 2176 | * blocks to be allocated by ext4 and submit them. |
f63e6005 | 2177 | */ |
72f84e65 | 2178 | ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index); |
f63e6005 | 2179 | /* |
af901ca1 | 2180 | * If we have a contiguous extent of pages and we |
f63e6005 TT |
2181 | * haven't done the I/O yet, map the blocks and submit |
2182 | * them for I/O. | |
2183 | */ | |
2184 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
5a87b7a5 | 2185 | mpage_da_map_and_submit(&mpd); |
f63e6005 TT |
2186 | ret = MPAGE_DA_EXTENT_TAIL; |
2187 | } | |
b3a3ca8c | 2188 | trace_ext4_da_write_pages(inode, &mpd); |
f63e6005 | 2189 | wbc->nr_to_write -= mpd.pages_written; |
df22291f | 2190 | |
61628a3f | 2191 | ext4_journal_stop(handle); |
df22291f | 2192 | |
8f64b32e | 2193 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
22208ded AK |
2194 | /* commit the transaction which would |
2195 | * free blocks released in the transaction | |
2196 | * and try again | |
2197 | */ | |
df22291f | 2198 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded AK |
2199 | ret = 0; |
2200 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | |
a1d6cc56 AK |
2201 | /* |
2202 | * got one extent now try with | |
2203 | * rest of the pages | |
2204 | */ | |
22208ded | 2205 | pages_written += mpd.pages_written; |
a1d6cc56 | 2206 | ret = 0; |
2acf2c26 | 2207 | io_done = 1; |
22208ded | 2208 | } else if (wbc->nr_to_write) |
61628a3f MC |
2209 | /* |
2210 | * There is no more writeout needed | |
2211 | * or we requested for a noblocking writeout | |
2212 | * and we found the device congested | |
2213 | */ | |
61628a3f | 2214 | break; |
a1d6cc56 | 2215 | } |
2acf2c26 AK |
2216 | if (!io_done && !cycled) { |
2217 | cycled = 1; | |
2218 | index = 0; | |
2219 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2220 | wbc->range_end = mapping->writeback_index - 1; | |
2221 | goto retry; | |
2222 | } | |
22208ded AK |
2223 | |
2224 | /* Update index */ | |
2acf2c26 | 2225 | wbc->range_cyclic = range_cyclic; |
22208ded AK |
2226 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2227 | /* | |
2228 | * set the writeback_index so that range_cyclic | |
2229 | * mode will write it back later | |
2230 | */ | |
72f84e65 | 2231 | mapping->writeback_index = done_index; |
a1d6cc56 | 2232 | |
61628a3f | 2233 | out_writepages: |
2faf2e19 | 2234 | wbc->nr_to_write -= nr_to_writebump; |
de89de6e | 2235 | wbc->range_start = range_start; |
9bffad1e | 2236 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
61628a3f | 2237 | return ret; |
64769240 AT |
2238 | } |
2239 | ||
79f0be8d AK |
2240 | #define FALL_BACK_TO_NONDELALLOC 1 |
2241 | static int ext4_nonda_switch(struct super_block *sb) | |
2242 | { | |
2243 | s64 free_blocks, dirty_blocks; | |
2244 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2245 | ||
2246 | /* | |
2247 | * switch to non delalloc mode if we are running low | |
2248 | * on free block. The free block accounting via percpu | |
179f7ebf | 2249 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2250 | * accumulated on each CPU without updating global counters |
2251 | * Delalloc need an accurate free block accounting. So switch | |
2252 | * to non delalloc when we are near to error range. | |
2253 | */ | |
57042651 TT |
2254 | free_blocks = EXT4_C2B(sbi, |
2255 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); | |
2256 | dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
79f0be8d | 2257 | if (2 * free_blocks < 3 * dirty_blocks || |
df55c99d | 2258 | free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) { |
79f0be8d | 2259 | /* |
c8afb446 ES |
2260 | * free block count is less than 150% of dirty blocks |
2261 | * or free blocks is less than watermark | |
79f0be8d AK |
2262 | */ |
2263 | return 1; | |
2264 | } | |
c8afb446 ES |
2265 | /* |
2266 | * Even if we don't switch but are nearing capacity, | |
2267 | * start pushing delalloc when 1/2 of free blocks are dirty. | |
2268 | */ | |
2269 | if (free_blocks < 2 * dirty_blocks) | |
2270 | writeback_inodes_sb_if_idle(sb); | |
2271 | ||
79f0be8d AK |
2272 | return 0; |
2273 | } | |
2274 | ||
64769240 | 2275 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2276 | loff_t pos, unsigned len, unsigned flags, |
2277 | struct page **pagep, void **fsdata) | |
64769240 | 2278 | { |
72b8ab9d | 2279 | int ret, retries = 0; |
64769240 AT |
2280 | struct page *page; |
2281 | pgoff_t index; | |
64769240 AT |
2282 | struct inode *inode = mapping->host; |
2283 | handle_t *handle; | |
02fac129 | 2284 | loff_t page_len; |
64769240 AT |
2285 | |
2286 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
2287 | |
2288 | if (ext4_nonda_switch(inode->i_sb)) { | |
2289 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2290 | return ext4_write_begin(file, mapping, pos, | |
2291 | len, flags, pagep, fsdata); | |
2292 | } | |
2293 | *fsdata = (void *)0; | |
9bffad1e | 2294 | trace_ext4_da_write_begin(inode, pos, len, flags); |
d2a17637 | 2295 | retry: |
64769240 AT |
2296 | /* |
2297 | * With delayed allocation, we don't log the i_disksize update | |
2298 | * if there is delayed block allocation. But we still need | |
2299 | * to journalling the i_disksize update if writes to the end | |
2300 | * of file which has an already mapped buffer. | |
2301 | */ | |
2302 | handle = ext4_journal_start(inode, 1); | |
2303 | if (IS_ERR(handle)) { | |
2304 | ret = PTR_ERR(handle); | |
2305 | goto out; | |
2306 | } | |
ebd3610b JK |
2307 | /* We cannot recurse into the filesystem as the transaction is already |
2308 | * started */ | |
2309 | flags |= AOP_FLAG_NOFS; | |
64769240 | 2310 | |
54566b2c | 2311 | page = grab_cache_page_write_begin(mapping, index, flags); |
d5a0d4f7 ES |
2312 | if (!page) { |
2313 | ext4_journal_stop(handle); | |
2314 | ret = -ENOMEM; | |
2315 | goto out; | |
2316 | } | |
64769240 AT |
2317 | *pagep = page; |
2318 | ||
6e1db88d | 2319 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
64769240 AT |
2320 | if (ret < 0) { |
2321 | unlock_page(page); | |
2322 | ext4_journal_stop(handle); | |
2323 | page_cache_release(page); | |
ae4d5372 AK |
2324 | /* |
2325 | * block_write_begin may have instantiated a few blocks | |
2326 | * outside i_size. Trim these off again. Don't need | |
2327 | * i_size_read because we hold i_mutex. | |
2328 | */ | |
2329 | if (pos + len > inode->i_size) | |
b9a4207d | 2330 | ext4_truncate_failed_write(inode); |
02fac129 AH |
2331 | } else { |
2332 | page_len = pos & (PAGE_CACHE_SIZE - 1); | |
2333 | if (page_len > 0) { | |
2334 | ret = ext4_discard_partial_page_buffers_no_lock(handle, | |
2335 | inode, page, pos - page_len, page_len, | |
2336 | EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED); | |
2337 | } | |
64769240 AT |
2338 | } |
2339 | ||
d2a17637 MC |
2340 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
2341 | goto retry; | |
64769240 AT |
2342 | out: |
2343 | return ret; | |
2344 | } | |
2345 | ||
632eaeab MC |
2346 | /* |
2347 | * Check if we should update i_disksize | |
2348 | * when write to the end of file but not require block allocation | |
2349 | */ | |
2350 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2351 | unsigned long offset) |
632eaeab MC |
2352 | { |
2353 | struct buffer_head *bh; | |
2354 | struct inode *inode = page->mapping->host; | |
2355 | unsigned int idx; | |
2356 | int i; | |
2357 | ||
2358 | bh = page_buffers(page); | |
2359 | idx = offset >> inode->i_blkbits; | |
2360 | ||
af5bc92d | 2361 | for (i = 0; i < idx; i++) |
632eaeab MC |
2362 | bh = bh->b_this_page; |
2363 | ||
29fa89d0 | 2364 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2365 | return 0; |
2366 | return 1; | |
2367 | } | |
2368 | ||
64769240 | 2369 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2370 | struct address_space *mapping, |
2371 | loff_t pos, unsigned len, unsigned copied, | |
2372 | struct page *page, void *fsdata) | |
64769240 AT |
2373 | { |
2374 | struct inode *inode = mapping->host; | |
2375 | int ret = 0, ret2; | |
2376 | handle_t *handle = ext4_journal_current_handle(); | |
2377 | loff_t new_i_size; | |
632eaeab | 2378 | unsigned long start, end; |
79f0be8d | 2379 | int write_mode = (int)(unsigned long)fsdata; |
02fac129 | 2380 | loff_t page_len; |
79f0be8d AK |
2381 | |
2382 | if (write_mode == FALL_BACK_TO_NONDELALLOC) { | |
2383 | if (ext4_should_order_data(inode)) { | |
2384 | return ext4_ordered_write_end(file, mapping, pos, | |
2385 | len, copied, page, fsdata); | |
2386 | } else if (ext4_should_writeback_data(inode)) { | |
2387 | return ext4_writeback_write_end(file, mapping, pos, | |
2388 | len, copied, page, fsdata); | |
2389 | } else { | |
2390 | BUG(); | |
2391 | } | |
2392 | } | |
632eaeab | 2393 | |
9bffad1e | 2394 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 2395 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2396 | end = start + copied - 1; |
64769240 AT |
2397 | |
2398 | /* | |
2399 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2400 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2401 | * into that. | |
2402 | */ | |
2403 | ||
2404 | new_i_size = pos + copied; | |
632eaeab MC |
2405 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
2406 | if (ext4_da_should_update_i_disksize(page, end)) { | |
2407 | down_write(&EXT4_I(inode)->i_data_sem); | |
2408 | if (new_i_size > EXT4_I(inode)->i_disksize) { | |
2409 | /* | |
2410 | * Updating i_disksize when extending file | |
2411 | * without needing block allocation | |
2412 | */ | |
2413 | if (ext4_should_order_data(inode)) | |
2414 | ret = ext4_jbd2_file_inode(handle, | |
2415 | inode); | |
64769240 | 2416 | |
632eaeab MC |
2417 | EXT4_I(inode)->i_disksize = new_i_size; |
2418 | } | |
2419 | up_write(&EXT4_I(inode)->i_data_sem); | |
cf17fea6 AK |
2420 | /* We need to mark inode dirty even if |
2421 | * new_i_size is less that inode->i_size | |
2422 | * bu greater than i_disksize.(hint delalloc) | |
2423 | */ | |
2424 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2425 | } |
632eaeab | 2426 | } |
64769240 AT |
2427 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
2428 | page, fsdata); | |
02fac129 AH |
2429 | |
2430 | page_len = PAGE_CACHE_SIZE - | |
2431 | ((pos + copied - 1) & (PAGE_CACHE_SIZE - 1)); | |
2432 | ||
2433 | if (page_len > 0) { | |
2434 | ret = ext4_discard_partial_page_buffers_no_lock(handle, | |
2435 | inode, page, pos + copied - 1, page_len, | |
2436 | EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED); | |
2437 | } | |
2438 | ||
64769240 AT |
2439 | copied = ret2; |
2440 | if (ret2 < 0) | |
2441 | ret = ret2; | |
2442 | ret2 = ext4_journal_stop(handle); | |
2443 | if (!ret) | |
2444 | ret = ret2; | |
2445 | ||
2446 | return ret ? ret : copied; | |
2447 | } | |
2448 | ||
2449 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) | |
2450 | { | |
64769240 AT |
2451 | /* |
2452 | * Drop reserved blocks | |
2453 | */ | |
2454 | BUG_ON(!PageLocked(page)); | |
2455 | if (!page_has_buffers(page)) | |
2456 | goto out; | |
2457 | ||
d2a17637 | 2458 | ext4_da_page_release_reservation(page, offset); |
64769240 AT |
2459 | |
2460 | out: | |
2461 | ext4_invalidatepage(page, offset); | |
2462 | ||
2463 | return; | |
2464 | } | |
2465 | ||
ccd2506b TT |
2466 | /* |
2467 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2468 | */ | |
2469 | int ext4_alloc_da_blocks(struct inode *inode) | |
2470 | { | |
fb40ba0d TT |
2471 | trace_ext4_alloc_da_blocks(inode); |
2472 | ||
ccd2506b TT |
2473 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
2474 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
2475 | return 0; | |
2476 | ||
2477 | /* | |
2478 | * We do something simple for now. The filemap_flush() will | |
2479 | * also start triggering a write of the data blocks, which is | |
2480 | * not strictly speaking necessary (and for users of | |
2481 | * laptop_mode, not even desirable). However, to do otherwise | |
2482 | * would require replicating code paths in: | |
de9a55b8 | 2483 | * |
ccd2506b TT |
2484 | * ext4_da_writepages() -> |
2485 | * write_cache_pages() ---> (via passed in callback function) | |
2486 | * __mpage_da_writepage() --> | |
2487 | * mpage_add_bh_to_extent() | |
2488 | * mpage_da_map_blocks() | |
2489 | * | |
2490 | * The problem is that write_cache_pages(), located in | |
2491 | * mm/page-writeback.c, marks pages clean in preparation for | |
2492 | * doing I/O, which is not desirable if we're not planning on | |
2493 | * doing I/O at all. | |
2494 | * | |
2495 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 2496 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
2497 | * would be ugly in the extreme. So instead we would need to |
2498 | * replicate parts of the code in the above functions, | |
25985edc | 2499 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
2500 | * write out the pages, but rather only collect contiguous |
2501 | * logical block extents, call the multi-block allocator, and | |
2502 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 2503 | * |
ccd2506b TT |
2504 | * For now, though, we'll cheat by calling filemap_flush(), |
2505 | * which will map the blocks, and start the I/O, but not | |
2506 | * actually wait for the I/O to complete. | |
2507 | */ | |
2508 | return filemap_flush(inode->i_mapping); | |
2509 | } | |
64769240 | 2510 | |
ac27a0ec DK |
2511 | /* |
2512 | * bmap() is special. It gets used by applications such as lilo and by | |
2513 | * the swapper to find the on-disk block of a specific piece of data. | |
2514 | * | |
2515 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 2516 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
2517 | * filesystem and enables swap, then they may get a nasty shock when the |
2518 | * data getting swapped to that swapfile suddenly gets overwritten by | |
2519 | * the original zero's written out previously to the journal and | |
2520 | * awaiting writeback in the kernel's buffer cache. | |
2521 | * | |
2522 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
2523 | * take extra steps to flush any blocks which might be in the cache. | |
2524 | */ | |
617ba13b | 2525 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
2526 | { |
2527 | struct inode *inode = mapping->host; | |
2528 | journal_t *journal; | |
2529 | int err; | |
2530 | ||
64769240 AT |
2531 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
2532 | test_opt(inode->i_sb, DELALLOC)) { | |
2533 | /* | |
2534 | * With delalloc we want to sync the file | |
2535 | * so that we can make sure we allocate | |
2536 | * blocks for file | |
2537 | */ | |
2538 | filemap_write_and_wait(mapping); | |
2539 | } | |
2540 | ||
19f5fb7a TT |
2541 | if (EXT4_JOURNAL(inode) && |
2542 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
2543 | /* |
2544 | * This is a REALLY heavyweight approach, but the use of | |
2545 | * bmap on dirty files is expected to be extremely rare: | |
2546 | * only if we run lilo or swapon on a freshly made file | |
2547 | * do we expect this to happen. | |
2548 | * | |
2549 | * (bmap requires CAP_SYS_RAWIO so this does not | |
2550 | * represent an unprivileged user DOS attack --- we'd be | |
2551 | * in trouble if mortal users could trigger this path at | |
2552 | * will.) | |
2553 | * | |
617ba13b | 2554 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
2555 | * regular files. If somebody wants to bmap a directory |
2556 | * or symlink and gets confused because the buffer | |
2557 | * hasn't yet been flushed to disk, they deserve | |
2558 | * everything they get. | |
2559 | */ | |
2560 | ||
19f5fb7a | 2561 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 2562 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
2563 | jbd2_journal_lock_updates(journal); |
2564 | err = jbd2_journal_flush(journal); | |
2565 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
2566 | |
2567 | if (err) | |
2568 | return 0; | |
2569 | } | |
2570 | ||
af5bc92d | 2571 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
2572 | } |
2573 | ||
617ba13b | 2574 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 2575 | { |
0562e0ba | 2576 | trace_ext4_readpage(page); |
617ba13b | 2577 | return mpage_readpage(page, ext4_get_block); |
ac27a0ec DK |
2578 | } |
2579 | ||
2580 | static int | |
617ba13b | 2581 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
2582 | struct list_head *pages, unsigned nr_pages) |
2583 | { | |
617ba13b | 2584 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
2585 | } |
2586 | ||
744692dc JZ |
2587 | static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset) |
2588 | { | |
2589 | struct buffer_head *head, *bh; | |
2590 | unsigned int curr_off = 0; | |
2591 | ||
2592 | if (!page_has_buffers(page)) | |
2593 | return; | |
2594 | head = bh = page_buffers(page); | |
2595 | do { | |
2596 | if (offset <= curr_off && test_clear_buffer_uninit(bh) | |
2597 | && bh->b_private) { | |
2598 | ext4_free_io_end(bh->b_private); | |
2599 | bh->b_private = NULL; | |
2600 | bh->b_end_io = NULL; | |
2601 | } | |
2602 | curr_off = curr_off + bh->b_size; | |
2603 | bh = bh->b_this_page; | |
2604 | } while (bh != head); | |
2605 | } | |
2606 | ||
617ba13b | 2607 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 2608 | { |
617ba13b | 2609 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 2610 | |
0562e0ba JZ |
2611 | trace_ext4_invalidatepage(page, offset); |
2612 | ||
744692dc JZ |
2613 | /* |
2614 | * free any io_end structure allocated for buffers to be discarded | |
2615 | */ | |
2616 | if (ext4_should_dioread_nolock(page->mapping->host)) | |
2617 | ext4_invalidatepage_free_endio(page, offset); | |
ac27a0ec DK |
2618 | /* |
2619 | * If it's a full truncate we just forget about the pending dirtying | |
2620 | */ | |
2621 | if (offset == 0) | |
2622 | ClearPageChecked(page); | |
2623 | ||
0390131b FM |
2624 | if (journal) |
2625 | jbd2_journal_invalidatepage(journal, page, offset); | |
2626 | else | |
2627 | block_invalidatepage(page, offset); | |
ac27a0ec DK |
2628 | } |
2629 | ||
617ba13b | 2630 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 2631 | { |
617ba13b | 2632 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 2633 | |
0562e0ba JZ |
2634 | trace_ext4_releasepage(page); |
2635 | ||
ac27a0ec DK |
2636 | WARN_ON(PageChecked(page)); |
2637 | if (!page_has_buffers(page)) | |
2638 | return 0; | |
0390131b FM |
2639 | if (journal) |
2640 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
2641 | else | |
2642 | return try_to_free_buffers(page); | |
ac27a0ec DK |
2643 | } |
2644 | ||
2ed88685 TT |
2645 | /* |
2646 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
2647 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
2648 | * The extent will be converted to initialized after the IO is complete. | |
2649 | */ | |
c7064ef1 | 2650 | static int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
2651 | struct buffer_head *bh_result, int create) |
2652 | { | |
c7064ef1 | 2653 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 2654 | inode->i_ino, create); |
2ed88685 TT |
2655 | return _ext4_get_block(inode, iblock, bh_result, |
2656 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
2657 | } |
2658 | ||
4c0425ff | 2659 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
552ef802 CH |
2660 | ssize_t size, void *private, int ret, |
2661 | bool is_async) | |
4c0425ff | 2662 | { |
72c5052d | 2663 | struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode; |
4c0425ff MC |
2664 | ext4_io_end_t *io_end = iocb->private; |
2665 | struct workqueue_struct *wq; | |
744692dc JZ |
2666 | unsigned long flags; |
2667 | struct ext4_inode_info *ei; | |
4c0425ff | 2668 | |
4b70df18 M |
2669 | /* if not async direct IO or dio with 0 bytes write, just return */ |
2670 | if (!io_end || !size) | |
552ef802 | 2671 | goto out; |
4b70df18 | 2672 | |
8d5d02e6 MC |
2673 | ext_debug("ext4_end_io_dio(): io_end 0x%p" |
2674 | "for inode %lu, iocb 0x%p, offset %llu, size %llu\n", | |
2675 | iocb->private, io_end->inode->i_ino, iocb, offset, | |
2676 | size); | |
8d5d02e6 MC |
2677 | |
2678 | /* if not aio dio with unwritten extents, just free io and return */ | |
bd2d0210 | 2679 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { |
8d5d02e6 MC |
2680 | ext4_free_io_end(io_end); |
2681 | iocb->private = NULL; | |
5b3ff237 JZ |
2682 | out: |
2683 | if (is_async) | |
2684 | aio_complete(iocb, ret, 0); | |
72c5052d | 2685 | inode_dio_done(inode); |
5b3ff237 | 2686 | return; |
8d5d02e6 MC |
2687 | } |
2688 | ||
4c0425ff MC |
2689 | io_end->offset = offset; |
2690 | io_end->size = size; | |
5b3ff237 JZ |
2691 | if (is_async) { |
2692 | io_end->iocb = iocb; | |
2693 | io_end->result = ret; | |
2694 | } | |
4c0425ff MC |
2695 | wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq; |
2696 | ||
8d5d02e6 | 2697 | /* Add the io_end to per-inode completed aio dio list*/ |
744692dc JZ |
2698 | ei = EXT4_I(io_end->inode); |
2699 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); | |
2700 | list_add_tail(&io_end->list, &ei->i_completed_io_list); | |
2701 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); | |
c999af2b ES |
2702 | |
2703 | /* queue the work to convert unwritten extents to written */ | |
2704 | queue_work(wq, &io_end->work); | |
4c0425ff | 2705 | iocb->private = NULL; |
72c5052d CH |
2706 | |
2707 | /* XXX: probably should move into the real I/O completion handler */ | |
2708 | inode_dio_done(inode); | |
4c0425ff | 2709 | } |
c7064ef1 | 2710 | |
744692dc JZ |
2711 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate) |
2712 | { | |
2713 | ext4_io_end_t *io_end = bh->b_private; | |
2714 | struct workqueue_struct *wq; | |
2715 | struct inode *inode; | |
2716 | unsigned long flags; | |
2717 | ||
2718 | if (!test_clear_buffer_uninit(bh) || !io_end) | |
2719 | goto out; | |
2720 | ||
2721 | if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) { | |
2722 | printk("sb umounted, discard end_io request for inode %lu\n", | |
2723 | io_end->inode->i_ino); | |
2724 | ext4_free_io_end(io_end); | |
2725 | goto out; | |
2726 | } | |
2727 | ||
32c80b32 TM |
2728 | /* |
2729 | * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now, | |
2730 | * but being more careful is always safe for the future change. | |
2731 | */ | |
744692dc | 2732 | inode = io_end->inode; |
32c80b32 TM |
2733 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { |
2734 | io_end->flag |= EXT4_IO_END_UNWRITTEN; | |
2735 | atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten); | |
2736 | } | |
744692dc JZ |
2737 | |
2738 | /* Add the io_end to per-inode completed io list*/ | |
2739 | spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags); | |
2740 | list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list); | |
2741 | spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags); | |
2742 | ||
2743 | wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq; | |
2744 | /* queue the work to convert unwritten extents to written */ | |
2745 | queue_work(wq, &io_end->work); | |
2746 | out: | |
2747 | bh->b_private = NULL; | |
2748 | bh->b_end_io = NULL; | |
2749 | clear_buffer_uninit(bh); | |
2750 | end_buffer_async_write(bh, uptodate); | |
2751 | } | |
2752 | ||
2753 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode) | |
2754 | { | |
2755 | ext4_io_end_t *io_end; | |
2756 | struct page *page = bh->b_page; | |
2757 | loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT; | |
2758 | size_t size = bh->b_size; | |
2759 | ||
2760 | retry: | |
2761 | io_end = ext4_init_io_end(inode, GFP_ATOMIC); | |
2762 | if (!io_end) { | |
6db26ffc | 2763 | pr_warn_ratelimited("%s: allocation fail\n", __func__); |
744692dc JZ |
2764 | schedule(); |
2765 | goto retry; | |
2766 | } | |
2767 | io_end->offset = offset; | |
2768 | io_end->size = size; | |
2769 | /* | |
2770 | * We need to hold a reference to the page to make sure it | |
2771 | * doesn't get evicted before ext4_end_io_work() has a chance | |
2772 | * to convert the extent from written to unwritten. | |
2773 | */ | |
2774 | io_end->page = page; | |
2775 | get_page(io_end->page); | |
2776 | ||
2777 | bh->b_private = io_end; | |
2778 | bh->b_end_io = ext4_end_io_buffer_write; | |
2779 | return 0; | |
2780 | } | |
2781 | ||
4c0425ff MC |
2782 | /* |
2783 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
2784 | * preallocated extents, and those write extend the file, no need to | |
2785 | * fall back to buffered IO. | |
2786 | * | |
b595076a | 2787 | * For holes, we fallocate those blocks, mark them as uninitialized |
4c0425ff | 2788 | * If those blocks were preallocated, we mark sure they are splited, but |
b595076a | 2789 | * still keep the range to write as uninitialized. |
4c0425ff | 2790 | * |
8d5d02e6 MC |
2791 | * The unwrritten extents will be converted to written when DIO is completed. |
2792 | * For async direct IO, since the IO may still pending when return, we | |
25985edc | 2793 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 2794 | * when async direct IO completed. |
4c0425ff MC |
2795 | * |
2796 | * If the O_DIRECT write will extend the file then add this inode to the | |
2797 | * orphan list. So recovery will truncate it back to the original size | |
2798 | * if the machine crashes during the write. | |
2799 | * | |
2800 | */ | |
2801 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, | |
2802 | const struct iovec *iov, loff_t offset, | |
2803 | unsigned long nr_segs) | |
2804 | { | |
2805 | struct file *file = iocb->ki_filp; | |
2806 | struct inode *inode = file->f_mapping->host; | |
2807 | ssize_t ret; | |
2808 | size_t count = iov_length(iov, nr_segs); | |
2809 | ||
2810 | loff_t final_size = offset + count; | |
2811 | if (rw == WRITE && final_size <= inode->i_size) { | |
2812 | /* | |
8d5d02e6 MC |
2813 | * We could direct write to holes and fallocate. |
2814 | * | |
2815 | * Allocated blocks to fill the hole are marked as uninitialized | |
25985edc | 2816 | * to prevent parallel buffered read to expose the stale data |
4c0425ff | 2817 | * before DIO complete the data IO. |
8d5d02e6 MC |
2818 | * |
2819 | * As to previously fallocated extents, ext4 get_block | |
4c0425ff MC |
2820 | * will just simply mark the buffer mapped but still |
2821 | * keep the extents uninitialized. | |
2822 | * | |
8d5d02e6 MC |
2823 | * for non AIO case, we will convert those unwritten extents |
2824 | * to written after return back from blockdev_direct_IO. | |
2825 | * | |
2826 | * for async DIO, the conversion needs to be defered when | |
2827 | * the IO is completed. The ext4 end_io callback function | |
2828 | * will be called to take care of the conversion work. | |
2829 | * Here for async case, we allocate an io_end structure to | |
2830 | * hook to the iocb. | |
4c0425ff | 2831 | */ |
8d5d02e6 MC |
2832 | iocb->private = NULL; |
2833 | EXT4_I(inode)->cur_aio_dio = NULL; | |
2834 | if (!is_sync_kiocb(iocb)) { | |
744692dc | 2835 | iocb->private = ext4_init_io_end(inode, GFP_NOFS); |
8d5d02e6 MC |
2836 | if (!iocb->private) |
2837 | return -ENOMEM; | |
2838 | /* | |
2839 | * we save the io structure for current async | |
79e83036 | 2840 | * direct IO, so that later ext4_map_blocks() |
8d5d02e6 MC |
2841 | * could flag the io structure whether there |
2842 | * is a unwritten extents needs to be converted | |
2843 | * when IO is completed. | |
2844 | */ | |
2845 | EXT4_I(inode)->cur_aio_dio = iocb->private; | |
2846 | } | |
2847 | ||
aacfc19c | 2848 | ret = __blockdev_direct_IO(rw, iocb, inode, |
4c0425ff MC |
2849 | inode->i_sb->s_bdev, iov, |
2850 | offset, nr_segs, | |
c7064ef1 | 2851 | ext4_get_block_write, |
aacfc19c CH |
2852 | ext4_end_io_dio, |
2853 | NULL, | |
2854 | DIO_LOCKING | DIO_SKIP_HOLES); | |
8d5d02e6 MC |
2855 | if (iocb->private) |
2856 | EXT4_I(inode)->cur_aio_dio = NULL; | |
2857 | /* | |
2858 | * The io_end structure takes a reference to the inode, | |
2859 | * that structure needs to be destroyed and the | |
2860 | * reference to the inode need to be dropped, when IO is | |
2861 | * complete, even with 0 byte write, or failed. | |
2862 | * | |
2863 | * In the successful AIO DIO case, the io_end structure will be | |
2864 | * desctroyed and the reference to the inode will be dropped | |
2865 | * after the end_io call back function is called. | |
2866 | * | |
2867 | * In the case there is 0 byte write, or error case, since | |
2868 | * VFS direct IO won't invoke the end_io call back function, | |
2869 | * we need to free the end_io structure here. | |
2870 | */ | |
2871 | if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) { | |
2872 | ext4_free_io_end(iocb->private); | |
2873 | iocb->private = NULL; | |
19f5fb7a TT |
2874 | } else if (ret > 0 && ext4_test_inode_state(inode, |
2875 | EXT4_STATE_DIO_UNWRITTEN)) { | |
109f5565 | 2876 | int err; |
8d5d02e6 MC |
2877 | /* |
2878 | * for non AIO case, since the IO is already | |
25985edc | 2879 | * completed, we could do the conversion right here |
8d5d02e6 | 2880 | */ |
109f5565 M |
2881 | err = ext4_convert_unwritten_extents(inode, |
2882 | offset, ret); | |
2883 | if (err < 0) | |
2884 | ret = err; | |
19f5fb7a | 2885 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
109f5565 | 2886 | } |
4c0425ff MC |
2887 | return ret; |
2888 | } | |
8d5d02e6 MC |
2889 | |
2890 | /* for write the the end of file case, we fall back to old way */ | |
4c0425ff MC |
2891 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); |
2892 | } | |
2893 | ||
2894 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |
2895 | const struct iovec *iov, loff_t offset, | |
2896 | unsigned long nr_segs) | |
2897 | { | |
2898 | struct file *file = iocb->ki_filp; | |
2899 | struct inode *inode = file->f_mapping->host; | |
0562e0ba | 2900 | ssize_t ret; |
4c0425ff | 2901 | |
84ebd795 TT |
2902 | /* |
2903 | * If we are doing data journalling we don't support O_DIRECT | |
2904 | */ | |
2905 | if (ext4_should_journal_data(inode)) | |
2906 | return 0; | |
2907 | ||
0562e0ba | 2908 | trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); |
12e9b892 | 2909 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
0562e0ba JZ |
2910 | ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); |
2911 | else | |
2912 | ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
2913 | trace_ext4_direct_IO_exit(inode, offset, | |
2914 | iov_length(iov, nr_segs), rw, ret); | |
2915 | return ret; | |
4c0425ff MC |
2916 | } |
2917 | ||
ac27a0ec | 2918 | /* |
617ba13b | 2919 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
2920 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
2921 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
2922 | * not necessarily locked. | |
2923 | * | |
2924 | * We cannot just dirty the page and leave attached buffers clean, because the | |
2925 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
2926 | * or jbddirty because all the journalling code will explode. | |
2927 | * | |
2928 | * So what we do is to mark the page "pending dirty" and next time writepage | |
2929 | * is called, propagate that into the buffers appropriately. | |
2930 | */ | |
617ba13b | 2931 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
2932 | { |
2933 | SetPageChecked(page); | |
2934 | return __set_page_dirty_nobuffers(page); | |
2935 | } | |
2936 | ||
617ba13b | 2937 | static const struct address_space_operations ext4_ordered_aops = { |
8ab22b9a HH |
2938 | .readpage = ext4_readpage, |
2939 | .readpages = ext4_readpages, | |
43ce1d23 | 2940 | .writepage = ext4_writepage, |
8ab22b9a HH |
2941 | .write_begin = ext4_write_begin, |
2942 | .write_end = ext4_ordered_write_end, | |
2943 | .bmap = ext4_bmap, | |
2944 | .invalidatepage = ext4_invalidatepage, | |
2945 | .releasepage = ext4_releasepage, | |
2946 | .direct_IO = ext4_direct_IO, | |
2947 | .migratepage = buffer_migrate_page, | |
2948 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 2949 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
2950 | }; |
2951 | ||
617ba13b | 2952 | static const struct address_space_operations ext4_writeback_aops = { |
8ab22b9a HH |
2953 | .readpage = ext4_readpage, |
2954 | .readpages = ext4_readpages, | |
43ce1d23 | 2955 | .writepage = ext4_writepage, |
8ab22b9a HH |
2956 | .write_begin = ext4_write_begin, |
2957 | .write_end = ext4_writeback_write_end, | |
2958 | .bmap = ext4_bmap, | |
2959 | .invalidatepage = ext4_invalidatepage, | |
2960 | .releasepage = ext4_releasepage, | |
2961 | .direct_IO = ext4_direct_IO, | |
2962 | .migratepage = buffer_migrate_page, | |
2963 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 2964 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
2965 | }; |
2966 | ||
617ba13b | 2967 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
2968 | .readpage = ext4_readpage, |
2969 | .readpages = ext4_readpages, | |
43ce1d23 | 2970 | .writepage = ext4_writepage, |
8ab22b9a HH |
2971 | .write_begin = ext4_write_begin, |
2972 | .write_end = ext4_journalled_write_end, | |
2973 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
2974 | .bmap = ext4_bmap, | |
2975 | .invalidatepage = ext4_invalidatepage, | |
2976 | .releasepage = ext4_releasepage, | |
84ebd795 | 2977 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 2978 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 2979 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
2980 | }; |
2981 | ||
64769240 | 2982 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
2983 | .readpage = ext4_readpage, |
2984 | .readpages = ext4_readpages, | |
43ce1d23 | 2985 | .writepage = ext4_writepage, |
8ab22b9a | 2986 | .writepages = ext4_da_writepages, |
8ab22b9a HH |
2987 | .write_begin = ext4_da_write_begin, |
2988 | .write_end = ext4_da_write_end, | |
2989 | .bmap = ext4_bmap, | |
2990 | .invalidatepage = ext4_da_invalidatepage, | |
2991 | .releasepage = ext4_releasepage, | |
2992 | .direct_IO = ext4_direct_IO, | |
2993 | .migratepage = buffer_migrate_page, | |
2994 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 2995 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
2996 | }; |
2997 | ||
617ba13b | 2998 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 2999 | { |
cd1aac32 AK |
3000 | if (ext4_should_order_data(inode) && |
3001 | test_opt(inode->i_sb, DELALLOC)) | |
3002 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3003 | else if (ext4_should_order_data(inode)) | |
617ba13b | 3004 | inode->i_mapping->a_ops = &ext4_ordered_aops; |
64769240 AT |
3005 | else if (ext4_should_writeback_data(inode) && |
3006 | test_opt(inode->i_sb, DELALLOC)) | |
3007 | inode->i_mapping->a_ops = &ext4_da_aops; | |
617ba13b MC |
3008 | else if (ext4_should_writeback_data(inode)) |
3009 | inode->i_mapping->a_ops = &ext4_writeback_aops; | |
ac27a0ec | 3010 | else |
617ba13b | 3011 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
ac27a0ec DK |
3012 | } |
3013 | ||
4e96b2db AH |
3014 | |
3015 | /* | |
3016 | * ext4_discard_partial_page_buffers() | |
3017 | * Wrapper function for ext4_discard_partial_page_buffers_no_lock. | |
3018 | * This function finds and locks the page containing the offset | |
3019 | * "from" and passes it to ext4_discard_partial_page_buffers_no_lock. | |
3020 | * Calling functions that already have the page locked should call | |
3021 | * ext4_discard_partial_page_buffers_no_lock directly. | |
3022 | */ | |
3023 | int ext4_discard_partial_page_buffers(handle_t *handle, | |
3024 | struct address_space *mapping, loff_t from, | |
3025 | loff_t length, int flags) | |
3026 | { | |
3027 | struct inode *inode = mapping->host; | |
3028 | struct page *page; | |
3029 | int err = 0; | |
3030 | ||
3031 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, | |
3032 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
3033 | if (!page) | |
3034 | return -EINVAL; | |
3035 | ||
3036 | err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page, | |
3037 | from, length, flags); | |
3038 | ||
3039 | unlock_page(page); | |
3040 | page_cache_release(page); | |
3041 | return err; | |
3042 | } | |
3043 | ||
3044 | /* | |
3045 | * ext4_discard_partial_page_buffers_no_lock() | |
3046 | * Zeros a page range of length 'length' starting from offset 'from'. | |
3047 | * Buffer heads that correspond to the block aligned regions of the | |
3048 | * zeroed range will be unmapped. Unblock aligned regions | |
3049 | * will have the corresponding buffer head mapped if needed so that | |
3050 | * that region of the page can be updated with the partial zero out. | |
3051 | * | |
3052 | * This function assumes that the page has already been locked. The | |
3053 | * The range to be discarded must be contained with in the given page. | |
3054 | * If the specified range exceeds the end of the page it will be shortened | |
3055 | * to the end of the page that corresponds to 'from'. This function is | |
3056 | * appropriate for updating a page and it buffer heads to be unmapped and | |
3057 | * zeroed for blocks that have been either released, or are going to be | |
3058 | * released. | |
3059 | * | |
3060 | * handle: The journal handle | |
3061 | * inode: The files inode | |
3062 | * page: A locked page that contains the offset "from" | |
3063 | * from: The starting byte offset (from the begining of the file) | |
3064 | * to begin discarding | |
3065 | * len: The length of bytes to discard | |
3066 | * flags: Optional flags that may be used: | |
3067 | * | |
3068 | * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED | |
3069 | * Only zero the regions of the page whose buffer heads | |
3070 | * have already been unmapped. This flag is appropriate | |
3071 | * for updateing the contents of a page whose blocks may | |
3072 | * have already been released, and we only want to zero | |
3073 | * out the regions that correspond to those released blocks. | |
3074 | * | |
3075 | * Returns zero on sucess or negative on failure. | |
3076 | */ | |
3077 | int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, | |
3078 | struct inode *inode, struct page *page, loff_t from, | |
3079 | loff_t length, int flags) | |
3080 | { | |
3081 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; | |
3082 | unsigned int offset = from & (PAGE_CACHE_SIZE-1); | |
3083 | unsigned int blocksize, max, pos; | |
3084 | unsigned int end_of_block, range_to_discard; | |
3085 | ext4_lblk_t iblock; | |
3086 | struct buffer_head *bh; | |
3087 | int err = 0; | |
3088 | ||
3089 | blocksize = inode->i_sb->s_blocksize; | |
3090 | max = PAGE_CACHE_SIZE - offset; | |
3091 | ||
3092 | if (index != page->index) | |
3093 | return -EINVAL; | |
3094 | ||
3095 | /* | |
3096 | * correct length if it does not fall between | |
3097 | * 'from' and the end of the page | |
3098 | */ | |
3099 | if (length > max || length < 0) | |
3100 | length = max; | |
3101 | ||
3102 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3103 | ||
3104 | if (!page_has_buffers(page)) { | |
3105 | /* | |
3106 | * If the range to be discarded covers a partial block | |
3107 | * we need to get the page buffers. This is because | |
3108 | * partial blocks cannot be released and the page needs | |
3109 | * to be updated with the contents of the block before | |
3110 | * we write the zeros on top of it. | |
3111 | */ | |
3112 | if (!(from & (blocksize - 1)) || | |
3113 | !((from + length) & (blocksize - 1))) { | |
3114 | create_empty_buffers(page, blocksize, 0); | |
3115 | } else { | |
3116 | /* | |
3117 | * If there are no partial blocks, | |
3118 | * there is nothing to update, | |
3119 | * so we can return now | |
3120 | */ | |
3121 | return 0; | |
3122 | } | |
3123 | } | |
3124 | ||
3125 | /* Find the buffer that contains "offset" */ | |
3126 | bh = page_buffers(page); | |
3127 | pos = blocksize; | |
3128 | while (offset >= pos) { | |
3129 | bh = bh->b_this_page; | |
3130 | iblock++; | |
3131 | pos += blocksize; | |
3132 | } | |
3133 | ||
3134 | pos = offset; | |
3135 | while (pos < offset + length) { | |
3136 | err = 0; | |
3137 | ||
3138 | /* The length of space left to zero and unmap */ | |
3139 | range_to_discard = offset + length - pos; | |
3140 | ||
3141 | /* The length of space until the end of the block */ | |
3142 | end_of_block = blocksize - (pos & (blocksize-1)); | |
3143 | ||
3144 | /* | |
3145 | * Do not unmap or zero past end of block | |
3146 | * for this buffer head | |
3147 | */ | |
3148 | if (range_to_discard > end_of_block) | |
3149 | range_to_discard = end_of_block; | |
3150 | ||
3151 | ||
3152 | /* | |
3153 | * Skip this buffer head if we are only zeroing unampped | |
3154 | * regions of the page | |
3155 | */ | |
3156 | if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED && | |
3157 | buffer_mapped(bh)) | |
3158 | goto next; | |
3159 | ||
3160 | /* If the range is block aligned, unmap */ | |
3161 | if (range_to_discard == blocksize) { | |
3162 | clear_buffer_dirty(bh); | |
3163 | bh->b_bdev = NULL; | |
3164 | clear_buffer_mapped(bh); | |
3165 | clear_buffer_req(bh); | |
3166 | clear_buffer_new(bh); | |
3167 | clear_buffer_delay(bh); | |
3168 | clear_buffer_unwritten(bh); | |
3169 | clear_buffer_uptodate(bh); | |
3170 | zero_user(page, pos, range_to_discard); | |
3171 | BUFFER_TRACE(bh, "Buffer discarded"); | |
3172 | goto next; | |
3173 | } | |
3174 | ||
3175 | /* | |
3176 | * If this block is not completely contained in the range | |
3177 | * to be discarded, then it is not going to be released. Because | |
3178 | * we need to keep this block, we need to make sure this part | |
3179 | * of the page is uptodate before we modify it by writeing | |
3180 | * partial zeros on it. | |
3181 | */ | |
3182 | if (!buffer_mapped(bh)) { | |
3183 | /* | |
3184 | * Buffer head must be mapped before we can read | |
3185 | * from the block | |
3186 | */ | |
3187 | BUFFER_TRACE(bh, "unmapped"); | |
3188 | ext4_get_block(inode, iblock, bh, 0); | |
3189 | /* unmapped? It's a hole - nothing to do */ | |
3190 | if (!buffer_mapped(bh)) { | |
3191 | BUFFER_TRACE(bh, "still unmapped"); | |
3192 | goto next; | |
3193 | } | |
3194 | } | |
3195 | ||
3196 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3197 | if (PageUptodate(page)) | |
3198 | set_buffer_uptodate(bh); | |
3199 | ||
3200 | if (!buffer_uptodate(bh)) { | |
3201 | err = -EIO; | |
3202 | ll_rw_block(READ, 1, &bh); | |
3203 | wait_on_buffer(bh); | |
3204 | /* Uhhuh. Read error. Complain and punt.*/ | |
3205 | if (!buffer_uptodate(bh)) | |
3206 | goto next; | |
3207 | } | |
3208 | ||
3209 | if (ext4_should_journal_data(inode)) { | |
3210 | BUFFER_TRACE(bh, "get write access"); | |
3211 | err = ext4_journal_get_write_access(handle, bh); | |
3212 | if (err) | |
3213 | goto next; | |
3214 | } | |
3215 | ||
3216 | zero_user(page, pos, range_to_discard); | |
3217 | ||
3218 | err = 0; | |
3219 | if (ext4_should_journal_data(inode)) { | |
3220 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
decbd919 | 3221 | } else |
4e96b2db | 3222 | mark_buffer_dirty(bh); |
4e96b2db AH |
3223 | |
3224 | BUFFER_TRACE(bh, "Partial buffer zeroed"); | |
3225 | next: | |
3226 | bh = bh->b_this_page; | |
3227 | iblock++; | |
3228 | pos += range_to_discard; | |
3229 | } | |
3230 | ||
3231 | return err; | |
3232 | } | |
3233 | ||
ac27a0ec | 3234 | /* |
617ba13b | 3235 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' |
ac27a0ec DK |
3236 | * up to the end of the block which corresponds to `from'. |
3237 | * This required during truncate. We need to physically zero the tail end | |
3238 | * of that block so it doesn't yield old data if the file is later grown. | |
3239 | */ | |
cf108bca | 3240 | int ext4_block_truncate_page(handle_t *handle, |
ac27a0ec | 3241 | struct address_space *mapping, loff_t from) |
30848851 AH |
3242 | { |
3243 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3244 | unsigned length; | |
3245 | unsigned blocksize; | |
3246 | struct inode *inode = mapping->host; | |
3247 | ||
3248 | blocksize = inode->i_sb->s_blocksize; | |
3249 | length = blocksize - (offset & (blocksize - 1)); | |
3250 | ||
3251 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3252 | } | |
3253 | ||
3254 | /* | |
3255 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3256 | * starting from file offset 'from'. The range to be zero'd must | |
3257 | * be contained with in one block. If the specified range exceeds | |
3258 | * the end of the block it will be shortened to end of the block | |
3259 | * that cooresponds to 'from' | |
3260 | */ | |
3261 | int ext4_block_zero_page_range(handle_t *handle, | |
3262 | struct address_space *mapping, loff_t from, loff_t length) | |
ac27a0ec | 3263 | { |
617ba13b | 3264 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
ac27a0ec | 3265 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
30848851 | 3266 | unsigned blocksize, max, pos; |
725d26d3 | 3267 | ext4_lblk_t iblock; |
ac27a0ec DK |
3268 | struct inode *inode = mapping->host; |
3269 | struct buffer_head *bh; | |
cf108bca | 3270 | struct page *page; |
ac27a0ec | 3271 | int err = 0; |
ac27a0ec | 3272 | |
f4a01017 TT |
3273 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, |
3274 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
cf108bca JK |
3275 | if (!page) |
3276 | return -EINVAL; | |
3277 | ||
ac27a0ec | 3278 | blocksize = inode->i_sb->s_blocksize; |
30848851 AH |
3279 | max = blocksize - (offset & (blocksize - 1)); |
3280 | ||
3281 | /* | |
3282 | * correct length if it does not fall between | |
3283 | * 'from' and the end of the block | |
3284 | */ | |
3285 | if (length > max || length < 0) | |
3286 | length = max; | |
3287 | ||
ac27a0ec DK |
3288 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); |
3289 | ||
ac27a0ec DK |
3290 | if (!page_has_buffers(page)) |
3291 | create_empty_buffers(page, blocksize, 0); | |
3292 | ||
3293 | /* Find the buffer that contains "offset" */ | |
3294 | bh = page_buffers(page); | |
3295 | pos = blocksize; | |
3296 | while (offset >= pos) { | |
3297 | bh = bh->b_this_page; | |
3298 | iblock++; | |
3299 | pos += blocksize; | |
3300 | } | |
3301 | ||
3302 | err = 0; | |
3303 | if (buffer_freed(bh)) { | |
3304 | BUFFER_TRACE(bh, "freed: skip"); | |
3305 | goto unlock; | |
3306 | } | |
3307 | ||
3308 | if (!buffer_mapped(bh)) { | |
3309 | BUFFER_TRACE(bh, "unmapped"); | |
617ba13b | 3310 | ext4_get_block(inode, iblock, bh, 0); |
ac27a0ec DK |
3311 | /* unmapped? It's a hole - nothing to do */ |
3312 | if (!buffer_mapped(bh)) { | |
3313 | BUFFER_TRACE(bh, "still unmapped"); | |
3314 | goto unlock; | |
3315 | } | |
3316 | } | |
3317 | ||
3318 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3319 | if (PageUptodate(page)) | |
3320 | set_buffer_uptodate(bh); | |
3321 | ||
3322 | if (!buffer_uptodate(bh)) { | |
3323 | err = -EIO; | |
3324 | ll_rw_block(READ, 1, &bh); | |
3325 | wait_on_buffer(bh); | |
3326 | /* Uhhuh. Read error. Complain and punt. */ | |
3327 | if (!buffer_uptodate(bh)) | |
3328 | goto unlock; | |
3329 | } | |
3330 | ||
617ba13b | 3331 | if (ext4_should_journal_data(inode)) { |
ac27a0ec | 3332 | BUFFER_TRACE(bh, "get write access"); |
617ba13b | 3333 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
3334 | if (err) |
3335 | goto unlock; | |
3336 | } | |
3337 | ||
eebd2aa3 | 3338 | zero_user(page, offset, length); |
ac27a0ec DK |
3339 | |
3340 | BUFFER_TRACE(bh, "zeroed end of block"); | |
3341 | ||
3342 | err = 0; | |
617ba13b | 3343 | if (ext4_should_journal_data(inode)) { |
0390131b | 3344 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
decbd919 | 3345 | } else |
ac27a0ec | 3346 | mark_buffer_dirty(bh); |
ac27a0ec DK |
3347 | |
3348 | unlock: | |
3349 | unlock_page(page); | |
3350 | page_cache_release(page); | |
3351 | return err; | |
3352 | } | |
3353 | ||
91ef4caf DG |
3354 | int ext4_can_truncate(struct inode *inode) |
3355 | { | |
91ef4caf DG |
3356 | if (S_ISREG(inode->i_mode)) |
3357 | return 1; | |
3358 | if (S_ISDIR(inode->i_mode)) | |
3359 | return 1; | |
3360 | if (S_ISLNK(inode->i_mode)) | |
3361 | return !ext4_inode_is_fast_symlink(inode); | |
3362 | return 0; | |
3363 | } | |
3364 | ||
a4bb6b64 AH |
3365 | /* |
3366 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks | |
3367 | * associated with the given offset and length | |
3368 | * | |
3369 | * @inode: File inode | |
3370 | * @offset: The offset where the hole will begin | |
3371 | * @len: The length of the hole | |
3372 | * | |
3373 | * Returns: 0 on sucess or negative on failure | |
3374 | */ | |
3375 | ||
3376 | int ext4_punch_hole(struct file *file, loff_t offset, loff_t length) | |
3377 | { | |
3378 | struct inode *inode = file->f_path.dentry->d_inode; | |
3379 | if (!S_ISREG(inode->i_mode)) | |
3380 | return -ENOTSUPP; | |
3381 | ||
3382 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
3383 | /* TODO: Add support for non extent hole punching */ | |
3384 | return -ENOTSUPP; | |
3385 | } | |
3386 | ||
bab08ab9 TT |
3387 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) { |
3388 | /* TODO: Add support for bigalloc file systems */ | |
3389 | return -ENOTSUPP; | |
3390 | } | |
3391 | ||
a4bb6b64 AH |
3392 | return ext4_ext_punch_hole(file, offset, length); |
3393 | } | |
3394 | ||
ac27a0ec | 3395 | /* |
617ba13b | 3396 | * ext4_truncate() |
ac27a0ec | 3397 | * |
617ba13b MC |
3398 | * We block out ext4_get_block() block instantiations across the entire |
3399 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
3400 | * simultaneously on behalf of the same inode. |
3401 | * | |
3402 | * As we work through the truncate and commmit bits of it to the journal there | |
3403 | * is one core, guiding principle: the file's tree must always be consistent on | |
3404 | * disk. We must be able to restart the truncate after a crash. | |
3405 | * | |
3406 | * The file's tree may be transiently inconsistent in memory (although it | |
3407 | * probably isn't), but whenever we close off and commit a journal transaction, | |
3408 | * the contents of (the filesystem + the journal) must be consistent and | |
3409 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
3410 | * left-to-right works OK too). | |
3411 | * | |
3412 | * Note that at recovery time, journal replay occurs *before* the restart of | |
3413 | * truncate against the orphan inode list. | |
3414 | * | |
3415 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 3416 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 3417 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
3418 | * ext4_truncate() to have another go. So there will be instantiated blocks |
3419 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 3420 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 3421 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 3422 | */ |
617ba13b | 3423 | void ext4_truncate(struct inode *inode) |
ac27a0ec | 3424 | { |
0562e0ba JZ |
3425 | trace_ext4_truncate_enter(inode); |
3426 | ||
91ef4caf | 3427 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
3428 | return; |
3429 | ||
12e9b892 | 3430 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 3431 | |
5534fb5b | 3432 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 3433 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 3434 | |
ff9893dc | 3435 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
cf108bca | 3436 | ext4_ext_truncate(inode); |
ff9893dc AG |
3437 | else |
3438 | ext4_ind_truncate(inode); | |
ac27a0ec | 3439 | |
0562e0ba | 3440 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
3441 | } |
3442 | ||
ac27a0ec | 3443 | /* |
617ba13b | 3444 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
3445 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
3446 | * data in memory that is needed to recreate the on-disk version of this | |
3447 | * inode. | |
3448 | */ | |
617ba13b MC |
3449 | static int __ext4_get_inode_loc(struct inode *inode, |
3450 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 3451 | { |
240799cd TT |
3452 | struct ext4_group_desc *gdp; |
3453 | struct buffer_head *bh; | |
3454 | struct super_block *sb = inode->i_sb; | |
3455 | ext4_fsblk_t block; | |
3456 | int inodes_per_block, inode_offset; | |
3457 | ||
3a06d778 | 3458 | iloc->bh = NULL; |
240799cd TT |
3459 | if (!ext4_valid_inum(sb, inode->i_ino)) |
3460 | return -EIO; | |
ac27a0ec | 3461 | |
240799cd TT |
3462 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
3463 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
3464 | if (!gdp) | |
ac27a0ec DK |
3465 | return -EIO; |
3466 | ||
240799cd TT |
3467 | /* |
3468 | * Figure out the offset within the block group inode table | |
3469 | */ | |
00d09882 | 3470 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
3471 | inode_offset = ((inode->i_ino - 1) % |
3472 | EXT4_INODES_PER_GROUP(sb)); | |
3473 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
3474 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
3475 | ||
3476 | bh = sb_getblk(sb, block); | |
ac27a0ec | 3477 | if (!bh) { |
c398eda0 TT |
3478 | EXT4_ERROR_INODE_BLOCK(inode, block, |
3479 | "unable to read itable block"); | |
ac27a0ec DK |
3480 | return -EIO; |
3481 | } | |
3482 | if (!buffer_uptodate(bh)) { | |
3483 | lock_buffer(bh); | |
9c83a923 HK |
3484 | |
3485 | /* | |
3486 | * If the buffer has the write error flag, we have failed | |
3487 | * to write out another inode in the same block. In this | |
3488 | * case, we don't have to read the block because we may | |
3489 | * read the old inode data successfully. | |
3490 | */ | |
3491 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
3492 | set_buffer_uptodate(bh); | |
3493 | ||
ac27a0ec DK |
3494 | if (buffer_uptodate(bh)) { |
3495 | /* someone brought it uptodate while we waited */ | |
3496 | unlock_buffer(bh); | |
3497 | goto has_buffer; | |
3498 | } | |
3499 | ||
3500 | /* | |
3501 | * If we have all information of the inode in memory and this | |
3502 | * is the only valid inode in the block, we need not read the | |
3503 | * block. | |
3504 | */ | |
3505 | if (in_mem) { | |
3506 | struct buffer_head *bitmap_bh; | |
240799cd | 3507 | int i, start; |
ac27a0ec | 3508 | |
240799cd | 3509 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 3510 | |
240799cd TT |
3511 | /* Is the inode bitmap in cache? */ |
3512 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
ac27a0ec DK |
3513 | if (!bitmap_bh) |
3514 | goto make_io; | |
3515 | ||
3516 | /* | |
3517 | * If the inode bitmap isn't in cache then the | |
3518 | * optimisation may end up performing two reads instead | |
3519 | * of one, so skip it. | |
3520 | */ | |
3521 | if (!buffer_uptodate(bitmap_bh)) { | |
3522 | brelse(bitmap_bh); | |
3523 | goto make_io; | |
3524 | } | |
240799cd | 3525 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
3526 | if (i == inode_offset) |
3527 | continue; | |
617ba13b | 3528 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
3529 | break; |
3530 | } | |
3531 | brelse(bitmap_bh); | |
240799cd | 3532 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
3533 | /* all other inodes are free, so skip I/O */ |
3534 | memset(bh->b_data, 0, bh->b_size); | |
3535 | set_buffer_uptodate(bh); | |
3536 | unlock_buffer(bh); | |
3537 | goto has_buffer; | |
3538 | } | |
3539 | } | |
3540 | ||
3541 | make_io: | |
240799cd TT |
3542 | /* |
3543 | * If we need to do any I/O, try to pre-readahead extra | |
3544 | * blocks from the inode table. | |
3545 | */ | |
3546 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
3547 | ext4_fsblk_t b, end, table; | |
3548 | unsigned num; | |
3549 | ||
3550 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 3551 | /* s_inode_readahead_blks is always a power of 2 */ |
240799cd TT |
3552 | b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1); |
3553 | if (table > b) | |
3554 | b = table; | |
3555 | end = b + EXT4_SB(sb)->s_inode_readahead_blks; | |
3556 | num = EXT4_INODES_PER_GROUP(sb); | |
3557 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
3558 | EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) | |
560671a0 | 3559 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
3560 | table += num / inodes_per_block; |
3561 | if (end > table) | |
3562 | end = table; | |
3563 | while (b <= end) | |
3564 | sb_breadahead(sb, b++); | |
3565 | } | |
3566 | ||
ac27a0ec DK |
3567 | /* |
3568 | * There are other valid inodes in the buffer, this inode | |
3569 | * has in-inode xattrs, or we don't have this inode in memory. | |
3570 | * Read the block from disk. | |
3571 | */ | |
0562e0ba | 3572 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
3573 | get_bh(bh); |
3574 | bh->b_end_io = end_buffer_read_sync; | |
3575 | submit_bh(READ_META, bh); | |
3576 | wait_on_buffer(bh); | |
3577 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
3578 | EXT4_ERROR_INODE_BLOCK(inode, block, |
3579 | "unable to read itable block"); | |
ac27a0ec DK |
3580 | brelse(bh); |
3581 | return -EIO; | |
3582 | } | |
3583 | } | |
3584 | has_buffer: | |
3585 | iloc->bh = bh; | |
3586 | return 0; | |
3587 | } | |
3588 | ||
617ba13b | 3589 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
3590 | { |
3591 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 3592 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 3593 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
3594 | } |
3595 | ||
617ba13b | 3596 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 3597 | { |
617ba13b | 3598 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
3599 | |
3600 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 3601 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 3602 | inode->i_flags |= S_SYNC; |
617ba13b | 3603 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 3604 | inode->i_flags |= S_APPEND; |
617ba13b | 3605 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 3606 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 3607 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 3608 | inode->i_flags |= S_NOATIME; |
617ba13b | 3609 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
3610 | inode->i_flags |= S_DIRSYNC; |
3611 | } | |
3612 | ||
ff9ddf7e JK |
3613 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
3614 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
3615 | { | |
84a8dce2 DM |
3616 | unsigned int vfs_fl; |
3617 | unsigned long old_fl, new_fl; | |
3618 | ||
3619 | do { | |
3620 | vfs_fl = ei->vfs_inode.i_flags; | |
3621 | old_fl = ei->i_flags; | |
3622 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
3623 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
3624 | EXT4_DIRSYNC_FL); | |
3625 | if (vfs_fl & S_SYNC) | |
3626 | new_fl |= EXT4_SYNC_FL; | |
3627 | if (vfs_fl & S_APPEND) | |
3628 | new_fl |= EXT4_APPEND_FL; | |
3629 | if (vfs_fl & S_IMMUTABLE) | |
3630 | new_fl |= EXT4_IMMUTABLE_FL; | |
3631 | if (vfs_fl & S_NOATIME) | |
3632 | new_fl |= EXT4_NOATIME_FL; | |
3633 | if (vfs_fl & S_DIRSYNC) | |
3634 | new_fl |= EXT4_DIRSYNC_FL; | |
3635 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 3636 | } |
de9a55b8 | 3637 | |
0fc1b451 | 3638 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 3639 | struct ext4_inode_info *ei) |
0fc1b451 AK |
3640 | { |
3641 | blkcnt_t i_blocks ; | |
8180a562 AK |
3642 | struct inode *inode = &(ei->vfs_inode); |
3643 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
3644 | |
3645 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
3646 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
3647 | /* we are using combined 48 bit field */ | |
3648 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
3649 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 3650 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
3651 | /* i_blocks represent file system block size */ |
3652 | return i_blocks << (inode->i_blkbits - 9); | |
3653 | } else { | |
3654 | return i_blocks; | |
3655 | } | |
0fc1b451 AK |
3656 | } else { |
3657 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
3658 | } | |
3659 | } | |
ff9ddf7e | 3660 | |
1d1fe1ee | 3661 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 3662 | { |
617ba13b MC |
3663 | struct ext4_iloc iloc; |
3664 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 3665 | struct ext4_inode_info *ei; |
1d1fe1ee | 3666 | struct inode *inode; |
b436b9be | 3667 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 3668 | long ret; |
ac27a0ec DK |
3669 | int block; |
3670 | ||
1d1fe1ee DH |
3671 | inode = iget_locked(sb, ino); |
3672 | if (!inode) | |
3673 | return ERR_PTR(-ENOMEM); | |
3674 | if (!(inode->i_state & I_NEW)) | |
3675 | return inode; | |
3676 | ||
3677 | ei = EXT4_I(inode); | |
7dc57615 | 3678 | iloc.bh = NULL; |
ac27a0ec | 3679 | |
1d1fe1ee DH |
3680 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
3681 | if (ret < 0) | |
ac27a0ec | 3682 | goto bad_inode; |
617ba13b | 3683 | raw_inode = ext4_raw_inode(&iloc); |
ac27a0ec DK |
3684 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
3685 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
3686 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 3687 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
3688 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
3689 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
3690 | } | |
3691 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); | |
ac27a0ec | 3692 | |
353eb83c | 3693 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
ac27a0ec DK |
3694 | ei->i_dir_start_lookup = 0; |
3695 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
3696 | /* We now have enough fields to check if the inode was active or not. | |
3697 | * This is needed because nfsd might try to access dead inodes | |
3698 | * the test is that same one that e2fsck uses | |
3699 | * NeilBrown 1999oct15 | |
3700 | */ | |
3701 | if (inode->i_nlink == 0) { | |
3702 | if (inode->i_mode == 0 || | |
617ba13b | 3703 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
ac27a0ec | 3704 | /* this inode is deleted */ |
1d1fe1ee | 3705 | ret = -ESTALE; |
ac27a0ec DK |
3706 | goto bad_inode; |
3707 | } | |
3708 | /* The only unlinked inodes we let through here have | |
3709 | * valid i_mode and are being read by the orphan | |
3710 | * recovery code: that's fine, we're about to complete | |
3711 | * the process of deleting those. */ | |
3712 | } | |
ac27a0ec | 3713 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 3714 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 3715 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 3716 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
3717 | ei->i_file_acl |= |
3718 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 3719 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 3720 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
3721 | #ifdef CONFIG_QUOTA |
3722 | ei->i_reserved_quota = 0; | |
3723 | #endif | |
ac27a0ec DK |
3724 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
3725 | ei->i_block_group = iloc.block_group; | |
a4912123 | 3726 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
3727 | /* |
3728 | * NOTE! The in-memory inode i_data array is in little-endian order | |
3729 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
3730 | */ | |
617ba13b | 3731 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
3732 | ei->i_data[block] = raw_inode->i_block[block]; |
3733 | INIT_LIST_HEAD(&ei->i_orphan); | |
3734 | ||
b436b9be JK |
3735 | /* |
3736 | * Set transaction id's of transactions that have to be committed | |
3737 | * to finish f[data]sync. We set them to currently running transaction | |
3738 | * as we cannot be sure that the inode or some of its metadata isn't | |
3739 | * part of the transaction - the inode could have been reclaimed and | |
3740 | * now it is reread from disk. | |
3741 | */ | |
3742 | if (journal) { | |
3743 | transaction_t *transaction; | |
3744 | tid_t tid; | |
3745 | ||
a931da6a | 3746 | read_lock(&journal->j_state_lock); |
b436b9be JK |
3747 | if (journal->j_running_transaction) |
3748 | transaction = journal->j_running_transaction; | |
3749 | else | |
3750 | transaction = journal->j_committing_transaction; | |
3751 | if (transaction) | |
3752 | tid = transaction->t_tid; | |
3753 | else | |
3754 | tid = journal->j_commit_sequence; | |
a931da6a | 3755 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
3756 | ei->i_sync_tid = tid; |
3757 | ei->i_datasync_tid = tid; | |
3758 | } | |
3759 | ||
0040d987 | 3760 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec | 3761 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
617ba13b | 3762 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
e5d2861f | 3763 | EXT4_INODE_SIZE(inode->i_sb)) { |
1d1fe1ee | 3764 | ret = -EIO; |
ac27a0ec | 3765 | goto bad_inode; |
e5d2861f | 3766 | } |
ac27a0ec DK |
3767 | if (ei->i_extra_isize == 0) { |
3768 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
3769 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
3770 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec DK |
3771 | } else { |
3772 | __le32 *magic = (void *)raw_inode + | |
617ba13b | 3773 | EXT4_GOOD_OLD_INODE_SIZE + |
ac27a0ec | 3774 | ei->i_extra_isize; |
617ba13b | 3775 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) |
19f5fb7a | 3776 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
ac27a0ec DK |
3777 | } |
3778 | } else | |
3779 | ei->i_extra_isize = 0; | |
3780 | ||
ef7f3835 KS |
3781 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
3782 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
3783 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
3784 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
3785 | ||
25ec56b5 JNC |
3786 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
3787 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
3788 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
3789 | inode->i_version |= | |
3790 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
3791 | } | |
3792 | ||
c4b5a614 | 3793 | ret = 0; |
485c26ec | 3794 | if (ei->i_file_acl && |
1032988c | 3795 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
3796 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
3797 | ei->i_file_acl); | |
485c26ec TT |
3798 | ret = -EIO; |
3799 | goto bad_inode; | |
07a03824 | 3800 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
c4b5a614 TT |
3801 | if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
3802 | (S_ISLNK(inode->i_mode) && | |
3803 | !ext4_inode_is_fast_symlink(inode))) | |
3804 | /* Validate extent which is part of inode */ | |
3805 | ret = ext4_ext_check_inode(inode); | |
de9a55b8 | 3806 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
fe2c8191 TN |
3807 | (S_ISLNK(inode->i_mode) && |
3808 | !ext4_inode_is_fast_symlink(inode))) { | |
de9a55b8 | 3809 | /* Validate block references which are part of inode */ |
1f7d1e77 | 3810 | ret = ext4_ind_check_inode(inode); |
fe2c8191 | 3811 | } |
567f3e9a | 3812 | if (ret) |
de9a55b8 | 3813 | goto bad_inode; |
7a262f7c | 3814 | |
ac27a0ec | 3815 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
3816 | inode->i_op = &ext4_file_inode_operations; |
3817 | inode->i_fop = &ext4_file_operations; | |
3818 | ext4_set_aops(inode); | |
ac27a0ec | 3819 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
3820 | inode->i_op = &ext4_dir_inode_operations; |
3821 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 3822 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 3823 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 3824 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
3825 | nd_terminate_link(ei->i_data, inode->i_size, |
3826 | sizeof(ei->i_data) - 1); | |
3827 | } else { | |
617ba13b MC |
3828 | inode->i_op = &ext4_symlink_inode_operations; |
3829 | ext4_set_aops(inode); | |
ac27a0ec | 3830 | } |
563bdd61 TT |
3831 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
3832 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 3833 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
3834 | if (raw_inode->i_block[0]) |
3835 | init_special_inode(inode, inode->i_mode, | |
3836 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
3837 | else | |
3838 | init_special_inode(inode, inode->i_mode, | |
3839 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
563bdd61 | 3840 | } else { |
563bdd61 | 3841 | ret = -EIO; |
24676da4 | 3842 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 3843 | goto bad_inode; |
ac27a0ec | 3844 | } |
af5bc92d | 3845 | brelse(iloc.bh); |
617ba13b | 3846 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
3847 | unlock_new_inode(inode); |
3848 | return inode; | |
ac27a0ec DK |
3849 | |
3850 | bad_inode: | |
567f3e9a | 3851 | brelse(iloc.bh); |
1d1fe1ee DH |
3852 | iget_failed(inode); |
3853 | return ERR_PTR(ret); | |
ac27a0ec DK |
3854 | } |
3855 | ||
0fc1b451 AK |
3856 | static int ext4_inode_blocks_set(handle_t *handle, |
3857 | struct ext4_inode *raw_inode, | |
3858 | struct ext4_inode_info *ei) | |
3859 | { | |
3860 | struct inode *inode = &(ei->vfs_inode); | |
3861 | u64 i_blocks = inode->i_blocks; | |
3862 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
3863 | |
3864 | if (i_blocks <= ~0U) { | |
3865 | /* | |
3866 | * i_blocks can be represnted in a 32 bit variable | |
3867 | * as multiple of 512 bytes | |
3868 | */ | |
8180a562 | 3869 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 3870 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 3871 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
3872 | return 0; |
3873 | } | |
3874 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
3875 | return -EFBIG; | |
3876 | ||
3877 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
3878 | /* |
3879 | * i_blocks can be represented in a 48 bit variable | |
3880 | * as multiple of 512 bytes | |
3881 | */ | |
8180a562 | 3882 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 3883 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 3884 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 3885 | } else { |
84a8dce2 | 3886 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
3887 | /* i_block is stored in file system block size */ |
3888 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
3889 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
3890 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 3891 | } |
f287a1a5 | 3892 | return 0; |
0fc1b451 AK |
3893 | } |
3894 | ||
ac27a0ec DK |
3895 | /* |
3896 | * Post the struct inode info into an on-disk inode location in the | |
3897 | * buffer-cache. This gobbles the caller's reference to the | |
3898 | * buffer_head in the inode location struct. | |
3899 | * | |
3900 | * The caller must have write access to iloc->bh. | |
3901 | */ | |
617ba13b | 3902 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 3903 | struct inode *inode, |
830156c7 | 3904 | struct ext4_iloc *iloc) |
ac27a0ec | 3905 | { |
617ba13b MC |
3906 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
3907 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
3908 | struct buffer_head *bh = iloc->bh; |
3909 | int err = 0, rc, block; | |
3910 | ||
3911 | /* For fields not not tracking in the in-memory inode, | |
3912 | * initialise them to zero for new inodes. */ | |
19f5fb7a | 3913 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 3914 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 3915 | |
ff9ddf7e | 3916 | ext4_get_inode_flags(ei); |
ac27a0ec | 3917 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
af5bc92d | 3918 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
3919 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); |
3920 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
3921 | /* | |
3922 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
3923 | * re-used with the upper 16 bits of the uid/gid intact | |
3924 | */ | |
af5bc92d | 3925 | if (!ei->i_dtime) { |
ac27a0ec DK |
3926 | raw_inode->i_uid_high = |
3927 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
3928 | raw_inode->i_gid_high = | |
3929 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
3930 | } else { | |
3931 | raw_inode->i_uid_high = 0; | |
3932 | raw_inode->i_gid_high = 0; | |
3933 | } | |
3934 | } else { | |
3935 | raw_inode->i_uid_low = | |
3936 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
3937 | raw_inode->i_gid_low = | |
3938 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
3939 | raw_inode->i_uid_high = 0; | |
3940 | raw_inode->i_gid_high = 0; | |
3941 | } | |
3942 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
3943 | |
3944 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
3945 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
3946 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
3947 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
3948 | ||
0fc1b451 AK |
3949 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
3950 | goto out_brelse; | |
ac27a0ec | 3951 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 3952 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
9b8f1f01 MC |
3953 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
3954 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
3955 | raw_inode->i_file_acl_high = |
3956 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 3957 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
a48380f7 AK |
3958 | ext4_isize_set(raw_inode, ei->i_disksize); |
3959 | if (ei->i_disksize > 0x7fffffffULL) { | |
3960 | struct super_block *sb = inode->i_sb; | |
3961 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
3962 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
3963 | EXT4_SB(sb)->s_es->s_rev_level == | |
3964 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
3965 | /* If this is the first large file | |
3966 | * created, add a flag to the superblock. | |
3967 | */ | |
3968 | err = ext4_journal_get_write_access(handle, | |
3969 | EXT4_SB(sb)->s_sbh); | |
3970 | if (err) | |
3971 | goto out_brelse; | |
3972 | ext4_update_dynamic_rev(sb); | |
3973 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 3974 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
a48380f7 | 3975 | sb->s_dirt = 1; |
0390131b | 3976 | ext4_handle_sync(handle); |
73b50c1c | 3977 | err = ext4_handle_dirty_metadata(handle, NULL, |
a48380f7 | 3978 | EXT4_SB(sb)->s_sbh); |
ac27a0ec DK |
3979 | } |
3980 | } | |
3981 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
3982 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
3983 | if (old_valid_dev(inode->i_rdev)) { | |
3984 | raw_inode->i_block[0] = | |
3985 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
3986 | raw_inode->i_block[1] = 0; | |
3987 | } else { | |
3988 | raw_inode->i_block[0] = 0; | |
3989 | raw_inode->i_block[1] = | |
3990 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
3991 | raw_inode->i_block[2] = 0; | |
3992 | } | |
de9a55b8 TT |
3993 | } else |
3994 | for (block = 0; block < EXT4_N_BLOCKS; block++) | |
3995 | raw_inode->i_block[block] = ei->i_data[block]; | |
ac27a0ec | 3996 | |
25ec56b5 JNC |
3997 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
3998 | if (ei->i_extra_isize) { | |
3999 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4000 | raw_inode->i_version_hi = | |
4001 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 4002 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
4003 | } |
4004 | ||
830156c7 | 4005 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 4006 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
4007 | if (!err) |
4008 | err = rc; | |
19f5fb7a | 4009 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
ac27a0ec | 4010 | |
b436b9be | 4011 | ext4_update_inode_fsync_trans(handle, inode, 0); |
ac27a0ec | 4012 | out_brelse: |
af5bc92d | 4013 | brelse(bh); |
617ba13b | 4014 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4015 | return err; |
4016 | } | |
4017 | ||
4018 | /* | |
617ba13b | 4019 | * ext4_write_inode() |
ac27a0ec DK |
4020 | * |
4021 | * We are called from a few places: | |
4022 | * | |
4023 | * - Within generic_file_write() for O_SYNC files. | |
4024 | * Here, there will be no transaction running. We wait for any running | |
4025 | * trasnaction to commit. | |
4026 | * | |
4027 | * - Within sys_sync(), kupdate and such. | |
4028 | * We wait on commit, if tol to. | |
4029 | * | |
4030 | * - Within prune_icache() (PF_MEMALLOC == true) | |
4031 | * Here we simply return. We can't afford to block kswapd on the | |
4032 | * journal commit. | |
4033 | * | |
4034 | * In all cases it is actually safe for us to return without doing anything, | |
4035 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 4036 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
4037 | * knfsd. |
4038 | * | |
4039 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4040 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4041 | * which we are interested. | |
4042 | * | |
4043 | * It would be a bug for them to not do this. The code: | |
4044 | * | |
4045 | * mark_inode_dirty(inode) | |
4046 | * stuff(); | |
4047 | * inode->i_size = expr; | |
4048 | * | |
4049 | * is in error because a kswapd-driven write_inode() could occur while | |
4050 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
4051 | * will no longer be on the superblock's dirty inode list. | |
4052 | */ | |
a9185b41 | 4053 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4054 | { |
91ac6f43 FM |
4055 | int err; |
4056 | ||
ac27a0ec DK |
4057 | if (current->flags & PF_MEMALLOC) |
4058 | return 0; | |
4059 | ||
91ac6f43 FM |
4060 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4061 | if (ext4_journal_current_handle()) { | |
4062 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4063 | dump_stack(); | |
4064 | return -EIO; | |
4065 | } | |
ac27a0ec | 4066 | |
a9185b41 | 4067 | if (wbc->sync_mode != WB_SYNC_ALL) |
91ac6f43 FM |
4068 | return 0; |
4069 | ||
4070 | err = ext4_force_commit(inode->i_sb); | |
4071 | } else { | |
4072 | struct ext4_iloc iloc; | |
ac27a0ec | 4073 | |
8b472d73 | 4074 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4075 | if (err) |
4076 | return err; | |
a9185b41 | 4077 | if (wbc->sync_mode == WB_SYNC_ALL) |
830156c7 FM |
4078 | sync_dirty_buffer(iloc.bh); |
4079 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
4080 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
4081 | "IO error syncing inode"); | |
830156c7 FM |
4082 | err = -EIO; |
4083 | } | |
fd2dd9fb | 4084 | brelse(iloc.bh); |
91ac6f43 FM |
4085 | } |
4086 | return err; | |
ac27a0ec DK |
4087 | } |
4088 | ||
4089 | /* | |
617ba13b | 4090 | * ext4_setattr() |
ac27a0ec DK |
4091 | * |
4092 | * Called from notify_change. | |
4093 | * | |
4094 | * We want to trap VFS attempts to truncate the file as soon as | |
4095 | * possible. In particular, we want to make sure that when the VFS | |
4096 | * shrinks i_size, we put the inode on the orphan list and modify | |
4097 | * i_disksize immediately, so that during the subsequent flushing of | |
4098 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4099 | * commit will leave the blocks being flushed in an unused state on | |
4100 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4101 | * be freed, so we have a strong guarantee that no future commit will | |
4102 | * leave these blocks visible to the user.) | |
4103 | * | |
678aaf48 JK |
4104 | * Another thing we have to assure is that if we are in ordered mode |
4105 | * and inode is still attached to the committing transaction, we must | |
4106 | * we start writeout of all the dirty pages which are being truncated. | |
4107 | * This way we are sure that all the data written in the previous | |
4108 | * transaction are already on disk (truncate waits for pages under | |
4109 | * writeback). | |
4110 | * | |
4111 | * Called with inode->i_mutex down. | |
ac27a0ec | 4112 | */ |
617ba13b | 4113 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
4114 | { |
4115 | struct inode *inode = dentry->d_inode; | |
4116 | int error, rc = 0; | |
3d287de3 | 4117 | int orphan = 0; |
ac27a0ec DK |
4118 | const unsigned int ia_valid = attr->ia_valid; |
4119 | ||
4120 | error = inode_change_ok(inode, attr); | |
4121 | if (error) | |
4122 | return error; | |
4123 | ||
12755627 | 4124 | if (is_quota_modification(inode, attr)) |
871a2931 | 4125 | dquot_initialize(inode); |
ac27a0ec DK |
4126 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || |
4127 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
4128 | handle_t *handle; | |
4129 | ||
4130 | /* (user+group)*(old+new) structure, inode write (sb, | |
4131 | * inode block, ? - but truncate inode update has it) */ | |
5aca07eb | 4132 | handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+ |
194074ac | 4133 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3); |
ac27a0ec DK |
4134 | if (IS_ERR(handle)) { |
4135 | error = PTR_ERR(handle); | |
4136 | goto err_out; | |
4137 | } | |
b43fa828 | 4138 | error = dquot_transfer(inode, attr); |
ac27a0ec | 4139 | if (error) { |
617ba13b | 4140 | ext4_journal_stop(handle); |
ac27a0ec DK |
4141 | return error; |
4142 | } | |
4143 | /* Update corresponding info in inode so that everything is in | |
4144 | * one transaction */ | |
4145 | if (attr->ia_valid & ATTR_UID) | |
4146 | inode->i_uid = attr->ia_uid; | |
4147 | if (attr->ia_valid & ATTR_GID) | |
4148 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4149 | error = ext4_mark_inode_dirty(handle, inode); |
4150 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4151 | } |
4152 | ||
e2b46574 | 4153 | if (attr->ia_valid & ATTR_SIZE) { |
562c72aa CH |
4154 | inode_dio_wait(inode); |
4155 | ||
12e9b892 | 4156 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
4157 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4158 | ||
0c095c7f TT |
4159 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
4160 | return -EFBIG; | |
e2b46574 ES |
4161 | } |
4162 | } | |
4163 | ||
ac27a0ec | 4164 | if (S_ISREG(inode->i_mode) && |
c8d46e41 | 4165 | attr->ia_valid & ATTR_SIZE && |
072bd7ea | 4166 | (attr->ia_size < inode->i_size)) { |
ac27a0ec DK |
4167 | handle_t *handle; |
4168 | ||
617ba13b | 4169 | handle = ext4_journal_start(inode, 3); |
ac27a0ec DK |
4170 | if (IS_ERR(handle)) { |
4171 | error = PTR_ERR(handle); | |
4172 | goto err_out; | |
4173 | } | |
3d287de3 DM |
4174 | if (ext4_handle_valid(handle)) { |
4175 | error = ext4_orphan_add(handle, inode); | |
4176 | orphan = 1; | |
4177 | } | |
617ba13b MC |
4178 | EXT4_I(inode)->i_disksize = attr->ia_size; |
4179 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
4180 | if (!error) |
4181 | error = rc; | |
617ba13b | 4182 | ext4_journal_stop(handle); |
678aaf48 JK |
4183 | |
4184 | if (ext4_should_order_data(inode)) { | |
4185 | error = ext4_begin_ordered_truncate(inode, | |
4186 | attr->ia_size); | |
4187 | if (error) { | |
4188 | /* Do as much error cleanup as possible */ | |
4189 | handle = ext4_journal_start(inode, 3); | |
4190 | if (IS_ERR(handle)) { | |
4191 | ext4_orphan_del(NULL, inode); | |
4192 | goto err_out; | |
4193 | } | |
4194 | ext4_orphan_del(handle, inode); | |
3d287de3 | 4195 | orphan = 0; |
678aaf48 JK |
4196 | ext4_journal_stop(handle); |
4197 | goto err_out; | |
4198 | } | |
4199 | } | |
ac27a0ec DK |
4200 | } |
4201 | ||
072bd7ea TT |
4202 | if (attr->ia_valid & ATTR_SIZE) { |
4203 | if (attr->ia_size != i_size_read(inode)) { | |
4204 | truncate_setsize(inode, attr->ia_size); | |
4205 | ext4_truncate(inode); | |
4206 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
4207 | ext4_truncate(inode); | |
4208 | } | |
ac27a0ec | 4209 | |
1025774c CH |
4210 | if (!rc) { |
4211 | setattr_copy(inode, attr); | |
4212 | mark_inode_dirty(inode); | |
4213 | } | |
4214 | ||
4215 | /* | |
4216 | * If the call to ext4_truncate failed to get a transaction handle at | |
4217 | * all, we need to clean up the in-core orphan list manually. | |
4218 | */ | |
3d287de3 | 4219 | if (orphan && inode->i_nlink) |
617ba13b | 4220 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4221 | |
4222 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 4223 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
4224 | |
4225 | err_out: | |
617ba13b | 4226 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4227 | if (!error) |
4228 | error = rc; | |
4229 | return error; | |
4230 | } | |
4231 | ||
3e3398a0 MC |
4232 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4233 | struct kstat *stat) | |
4234 | { | |
4235 | struct inode *inode; | |
4236 | unsigned long delalloc_blocks; | |
4237 | ||
4238 | inode = dentry->d_inode; | |
4239 | generic_fillattr(inode, stat); | |
4240 | ||
4241 | /* | |
4242 | * We can't update i_blocks if the block allocation is delayed | |
4243 | * otherwise in the case of system crash before the real block | |
4244 | * allocation is done, we will have i_blocks inconsistent with | |
4245 | * on-disk file blocks. | |
4246 | * We always keep i_blocks updated together with real | |
4247 | * allocation. But to not confuse with user, stat | |
4248 | * will return the blocks that include the delayed allocation | |
4249 | * blocks for this file. | |
4250 | */ | |
3e3398a0 | 4251 | delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks; |
3e3398a0 MC |
4252 | |
4253 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; | |
4254 | return 0; | |
4255 | } | |
ac27a0ec | 4256 | |
a02908f1 MC |
4257 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
4258 | { | |
12e9b892 | 4259 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
8bb2b247 | 4260 | return ext4_ind_trans_blocks(inode, nrblocks, chunk); |
ac51d837 | 4261 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); |
a02908f1 | 4262 | } |
ac51d837 | 4263 | |
ac27a0ec | 4264 | /* |
a02908f1 MC |
4265 | * Account for index blocks, block groups bitmaps and block group |
4266 | * descriptor blocks if modify datablocks and index blocks | |
4267 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 4268 | * |
a02908f1 | 4269 | * If datablocks are discontiguous, they are possible to spread over |
af901ca1 | 4270 | * different block groups too. If they are contiuguous, with flexbg, |
a02908f1 | 4271 | * they could still across block group boundary. |
ac27a0ec | 4272 | * |
a02908f1 MC |
4273 | * Also account for superblock, inode, quota and xattr blocks |
4274 | */ | |
1f109d5a | 4275 | static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
a02908f1 | 4276 | { |
8df9675f TT |
4277 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
4278 | int gdpblocks; | |
a02908f1 MC |
4279 | int idxblocks; |
4280 | int ret = 0; | |
4281 | ||
4282 | /* | |
4283 | * How many index blocks need to touch to modify nrblocks? | |
4284 | * The "Chunk" flag indicating whether the nrblocks is | |
4285 | * physically contiguous on disk | |
4286 | * | |
4287 | * For Direct IO and fallocate, they calls get_block to allocate | |
4288 | * one single extent at a time, so they could set the "Chunk" flag | |
4289 | */ | |
4290 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); | |
4291 | ||
4292 | ret = idxblocks; | |
4293 | ||
4294 | /* | |
4295 | * Now let's see how many group bitmaps and group descriptors need | |
4296 | * to account | |
4297 | */ | |
4298 | groups = idxblocks; | |
4299 | if (chunk) | |
4300 | groups += 1; | |
4301 | else | |
4302 | groups += nrblocks; | |
4303 | ||
4304 | gdpblocks = groups; | |
8df9675f TT |
4305 | if (groups > ngroups) |
4306 | groups = ngroups; | |
a02908f1 MC |
4307 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
4308 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
4309 | ||
4310 | /* bitmaps and block group descriptor blocks */ | |
4311 | ret += groups + gdpblocks; | |
4312 | ||
4313 | /* Blocks for super block, inode, quota and xattr blocks */ | |
4314 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
4315 | ||
4316 | return ret; | |
4317 | } | |
4318 | ||
4319 | /* | |
25985edc | 4320 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
4321 | * the modification of a single pages into a single transaction, |
4322 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 4323 | * |
525f4ed8 | 4324 | * This could be called via ext4_write_begin() |
ac27a0ec | 4325 | * |
525f4ed8 | 4326 | * We need to consider the worse case, when |
a02908f1 | 4327 | * one new block per extent. |
ac27a0ec | 4328 | */ |
a86c6181 | 4329 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 4330 | { |
617ba13b | 4331 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
4332 | int ret; |
4333 | ||
a02908f1 | 4334 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
a86c6181 | 4335 | |
a02908f1 | 4336 | /* Account for data blocks for journalled mode */ |
617ba13b | 4337 | if (ext4_should_journal_data(inode)) |
a02908f1 | 4338 | ret += bpp; |
ac27a0ec DK |
4339 | return ret; |
4340 | } | |
f3bd1f3f MC |
4341 | |
4342 | /* | |
4343 | * Calculate the journal credits for a chunk of data modification. | |
4344 | * | |
4345 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 4346 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
4347 | * |
4348 | * journal buffers for data blocks are not included here, as DIO | |
4349 | * and fallocate do no need to journal data buffers. | |
4350 | */ | |
4351 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
4352 | { | |
4353 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
4354 | } | |
4355 | ||
ac27a0ec | 4356 | /* |
617ba13b | 4357 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
4358 | * Give this, we know that the caller already has write access to iloc->bh. |
4359 | */ | |
617ba13b | 4360 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 4361 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4362 | { |
4363 | int err = 0; | |
4364 | ||
25ec56b5 JNC |
4365 | if (test_opt(inode->i_sb, I_VERSION)) |
4366 | inode_inc_iversion(inode); | |
4367 | ||
ac27a0ec DK |
4368 | /* the do_update_inode consumes one bh->b_count */ |
4369 | get_bh(iloc->bh); | |
4370 | ||
dab291af | 4371 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 4372 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
4373 | put_bh(iloc->bh); |
4374 | return err; | |
4375 | } | |
4376 | ||
4377 | /* | |
4378 | * On success, We end up with an outstanding reference count against | |
4379 | * iloc->bh. This _must_ be cleaned up later. | |
4380 | */ | |
4381 | ||
4382 | int | |
617ba13b MC |
4383 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
4384 | struct ext4_iloc *iloc) | |
ac27a0ec | 4385 | { |
0390131b FM |
4386 | int err; |
4387 | ||
4388 | err = ext4_get_inode_loc(inode, iloc); | |
4389 | if (!err) { | |
4390 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
4391 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
4392 | if (err) { | |
4393 | brelse(iloc->bh); | |
4394 | iloc->bh = NULL; | |
ac27a0ec DK |
4395 | } |
4396 | } | |
617ba13b | 4397 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4398 | return err; |
4399 | } | |
4400 | ||
6dd4ee7c KS |
4401 | /* |
4402 | * Expand an inode by new_extra_isize bytes. | |
4403 | * Returns 0 on success or negative error number on failure. | |
4404 | */ | |
1d03ec98 AK |
4405 | static int ext4_expand_extra_isize(struct inode *inode, |
4406 | unsigned int new_extra_isize, | |
4407 | struct ext4_iloc iloc, | |
4408 | handle_t *handle) | |
6dd4ee7c KS |
4409 | { |
4410 | struct ext4_inode *raw_inode; | |
4411 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
4412 | |
4413 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
4414 | return 0; | |
4415 | ||
4416 | raw_inode = ext4_raw_inode(&iloc); | |
4417 | ||
4418 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
4419 | |
4420 | /* No extended attributes present */ | |
19f5fb7a TT |
4421 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
4422 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
4423 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
4424 | new_extra_isize); | |
4425 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
4426 | return 0; | |
4427 | } | |
4428 | ||
4429 | /* try to expand with EAs present */ | |
4430 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
4431 | raw_inode, handle); | |
4432 | } | |
4433 | ||
ac27a0ec DK |
4434 | /* |
4435 | * What we do here is to mark the in-core inode as clean with respect to inode | |
4436 | * dirtiness (it may still be data-dirty). | |
4437 | * This means that the in-core inode may be reaped by prune_icache | |
4438 | * without having to perform any I/O. This is a very good thing, | |
4439 | * because *any* task may call prune_icache - even ones which | |
4440 | * have a transaction open against a different journal. | |
4441 | * | |
4442 | * Is this cheating? Not really. Sure, we haven't written the | |
4443 | * inode out, but prune_icache isn't a user-visible syncing function. | |
4444 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
4445 | * we start and wait on commits. | |
4446 | * | |
4447 | * Is this efficient/effective? Well, we're being nice to the system | |
4448 | * by cleaning up our inodes proactively so they can be reaped | |
4449 | * without I/O. But we are potentially leaving up to five seconds' | |
4450 | * worth of inodes floating about which prune_icache wants us to | |
4451 | * write out. One way to fix that would be to get prune_icache() | |
4452 | * to do a write_super() to free up some memory. It has the desired | |
4453 | * effect. | |
4454 | */ | |
617ba13b | 4455 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 4456 | { |
617ba13b | 4457 | struct ext4_iloc iloc; |
6dd4ee7c KS |
4458 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4459 | static unsigned int mnt_count; | |
4460 | int err, ret; | |
ac27a0ec DK |
4461 | |
4462 | might_sleep(); | |
7ff9c073 | 4463 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 4464 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
4465 | if (ext4_handle_valid(handle) && |
4466 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 4467 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
4468 | /* |
4469 | * We need extra buffer credits since we may write into EA block | |
4470 | * with this same handle. If journal_extend fails, then it will | |
4471 | * only result in a minor loss of functionality for that inode. | |
4472 | * If this is felt to be critical, then e2fsck should be run to | |
4473 | * force a large enough s_min_extra_isize. | |
4474 | */ | |
4475 | if ((jbd2_journal_extend(handle, | |
4476 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
4477 | ret = ext4_expand_extra_isize(inode, | |
4478 | sbi->s_want_extra_isize, | |
4479 | iloc, handle); | |
4480 | if (ret) { | |
19f5fb7a TT |
4481 | ext4_set_inode_state(inode, |
4482 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
4483 | if (mnt_count != |
4484 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 4485 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
4486 | "Unable to expand inode %lu. Delete" |
4487 | " some EAs or run e2fsck.", | |
4488 | inode->i_ino); | |
c1bddad9 AK |
4489 | mnt_count = |
4490 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
4491 | } |
4492 | } | |
4493 | } | |
4494 | } | |
ac27a0ec | 4495 | if (!err) |
617ba13b | 4496 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
4497 | return err; |
4498 | } | |
4499 | ||
4500 | /* | |
617ba13b | 4501 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
4502 | * |
4503 | * We're really interested in the case where a file is being extended. | |
4504 | * i_size has been changed by generic_commit_write() and we thus need | |
4505 | * to include the updated inode in the current transaction. | |
4506 | * | |
5dd4056d | 4507 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
4508 | * are allocated to the file. |
4509 | * | |
4510 | * If the inode is marked synchronous, we don't honour that here - doing | |
4511 | * so would cause a commit on atime updates, which we don't bother doing. | |
4512 | * We handle synchronous inodes at the highest possible level. | |
4513 | */ | |
aa385729 | 4514 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 4515 | { |
ac27a0ec DK |
4516 | handle_t *handle; |
4517 | ||
617ba13b | 4518 | handle = ext4_journal_start(inode, 2); |
ac27a0ec DK |
4519 | if (IS_ERR(handle)) |
4520 | goto out; | |
f3dc272f | 4521 | |
f3dc272f CW |
4522 | ext4_mark_inode_dirty(handle, inode); |
4523 | ||
617ba13b | 4524 | ext4_journal_stop(handle); |
ac27a0ec DK |
4525 | out: |
4526 | return; | |
4527 | } | |
4528 | ||
4529 | #if 0 | |
4530 | /* | |
4531 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
4532 | * it from being flushed to disk early. Unlike | |
617ba13b | 4533 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
4534 | * returns no iloc structure, so the caller needs to repeat the iloc |
4535 | * lookup to mark the inode dirty later. | |
4536 | */ | |
617ba13b | 4537 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 4538 | { |
617ba13b | 4539 | struct ext4_iloc iloc; |
ac27a0ec DK |
4540 | |
4541 | int err = 0; | |
4542 | if (handle) { | |
617ba13b | 4543 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
4544 | if (!err) { |
4545 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 4546 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 4547 | if (!err) |
0390131b | 4548 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 4549 | NULL, |
0390131b | 4550 | iloc.bh); |
ac27a0ec DK |
4551 | brelse(iloc.bh); |
4552 | } | |
4553 | } | |
617ba13b | 4554 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4555 | return err; |
4556 | } | |
4557 | #endif | |
4558 | ||
617ba13b | 4559 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
4560 | { |
4561 | journal_t *journal; | |
4562 | handle_t *handle; | |
4563 | int err; | |
4564 | ||
4565 | /* | |
4566 | * We have to be very careful here: changing a data block's | |
4567 | * journaling status dynamically is dangerous. If we write a | |
4568 | * data block to the journal, change the status and then delete | |
4569 | * that block, we risk forgetting to revoke the old log record | |
4570 | * from the journal and so a subsequent replay can corrupt data. | |
4571 | * So, first we make sure that the journal is empty and that | |
4572 | * nobody is changing anything. | |
4573 | */ | |
4574 | ||
617ba13b | 4575 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
4576 | if (!journal) |
4577 | return 0; | |
d699594d | 4578 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
4579 | return -EROFS; |
4580 | ||
dab291af MC |
4581 | jbd2_journal_lock_updates(journal); |
4582 | jbd2_journal_flush(journal); | |
ac27a0ec DK |
4583 | |
4584 | /* | |
4585 | * OK, there are no updates running now, and all cached data is | |
4586 | * synced to disk. We are now in a completely consistent state | |
4587 | * which doesn't have anything in the journal, and we know that | |
4588 | * no filesystem updates are running, so it is safe to modify | |
4589 | * the inode's in-core data-journaling state flag now. | |
4590 | */ | |
4591 | ||
4592 | if (val) | |
12e9b892 | 4593 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
ac27a0ec | 4594 | else |
12e9b892 | 4595 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
617ba13b | 4596 | ext4_set_aops(inode); |
ac27a0ec | 4597 | |
dab291af | 4598 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
4599 | |
4600 | /* Finally we can mark the inode as dirty. */ | |
4601 | ||
617ba13b | 4602 | handle = ext4_journal_start(inode, 1); |
ac27a0ec DK |
4603 | if (IS_ERR(handle)) |
4604 | return PTR_ERR(handle); | |
4605 | ||
617ba13b | 4606 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 4607 | ext4_handle_sync(handle); |
617ba13b MC |
4608 | ext4_journal_stop(handle); |
4609 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
4610 | |
4611 | return err; | |
4612 | } | |
2e9ee850 AK |
4613 | |
4614 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
4615 | { | |
4616 | return !buffer_mapped(bh); | |
4617 | } | |
4618 | ||
c2ec175c | 4619 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 4620 | { |
c2ec175c | 4621 | struct page *page = vmf->page; |
2e9ee850 AK |
4622 | loff_t size; |
4623 | unsigned long len; | |
9ea7df53 | 4624 | int ret; |
2e9ee850 AK |
4625 | struct file *file = vma->vm_file; |
4626 | struct inode *inode = file->f_path.dentry->d_inode; | |
4627 | struct address_space *mapping = inode->i_mapping; | |
9ea7df53 JK |
4628 | handle_t *handle; |
4629 | get_block_t *get_block; | |
4630 | int retries = 0; | |
2e9ee850 AK |
4631 | |
4632 | /* | |
9ea7df53 JK |
4633 | * This check is racy but catches the common case. We rely on |
4634 | * __block_page_mkwrite() to do a reliable check. | |
2e9ee850 | 4635 | */ |
9ea7df53 JK |
4636 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); |
4637 | /* Delalloc case is easy... */ | |
4638 | if (test_opt(inode->i_sb, DELALLOC) && | |
4639 | !ext4_should_journal_data(inode) && | |
4640 | !ext4_nonda_switch(inode->i_sb)) { | |
4641 | do { | |
4642 | ret = __block_page_mkwrite(vma, vmf, | |
4643 | ext4_da_get_block_prep); | |
4644 | } while (ret == -ENOSPC && | |
4645 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
4646 | goto out_ret; | |
2e9ee850 | 4647 | } |
0e499890 DW |
4648 | |
4649 | lock_page(page); | |
9ea7df53 JK |
4650 | size = i_size_read(inode); |
4651 | /* Page got truncated from under us? */ | |
4652 | if (page->mapping != mapping || page_offset(page) > size) { | |
4653 | unlock_page(page); | |
4654 | ret = VM_FAULT_NOPAGE; | |
4655 | goto out; | |
0e499890 | 4656 | } |
2e9ee850 AK |
4657 | |
4658 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
4659 | len = size & ~PAGE_CACHE_MASK; | |
4660 | else | |
4661 | len = PAGE_CACHE_SIZE; | |
a827eaff | 4662 | /* |
9ea7df53 JK |
4663 | * Return if we have all the buffers mapped. This avoids the need to do |
4664 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 4665 | */ |
2e9ee850 | 4666 | if (page_has_buffers(page)) { |
2e9ee850 | 4667 | if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, |
a827eaff | 4668 | ext4_bh_unmapped)) { |
9ea7df53 JK |
4669 | /* Wait so that we don't change page under IO */ |
4670 | wait_on_page_writeback(page); | |
4671 | ret = VM_FAULT_LOCKED; | |
4672 | goto out; | |
a827eaff | 4673 | } |
2e9ee850 | 4674 | } |
a827eaff | 4675 | unlock_page(page); |
9ea7df53 JK |
4676 | /* OK, we need to fill the hole... */ |
4677 | if (ext4_should_dioread_nolock(inode)) | |
4678 | get_block = ext4_get_block_write; | |
4679 | else | |
4680 | get_block = ext4_get_block; | |
4681 | retry_alloc: | |
4682 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); | |
4683 | if (IS_ERR(handle)) { | |
c2ec175c | 4684 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
4685 | goto out; |
4686 | } | |
4687 | ret = __block_page_mkwrite(vma, vmf, get_block); | |
4688 | if (!ret && ext4_should_journal_data(inode)) { | |
4689 | if (walk_page_buffers(handle, page_buffers(page), 0, | |
4690 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { | |
4691 | unlock_page(page); | |
4692 | ret = VM_FAULT_SIGBUS; | |
4693 | goto out; | |
4694 | } | |
4695 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
4696 | } | |
4697 | ext4_journal_stop(handle); | |
4698 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
4699 | goto retry_alloc; | |
4700 | out_ret: | |
4701 | ret = block_page_mkwrite_return(ret); | |
4702 | out: | |
2e9ee850 AK |
4703 | return ret; |
4704 | } |