1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/ext4/inode.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * from
11 *
12 * linux/fs/minix/inode.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
18 *
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
20 */
22 #include <linux/fs.h>
23 #include <linux/time.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/dax.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/bitops.h>
41 #include <linux/iomap.h>
42 #include <linux/iversion.h>
44 #include "ext4_jbd2.h"
45 #include "xattr.h"
46 #include "acl.h"
47 #include "truncate.h"
49 #include <trace/events/ext4.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
54 struct ext4_inode_info *ei)
55 {
56 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
57 __u32 csum;
58 __u16 dummy_csum = 0;
59 int offset = offsetof(struct ext4_inode, i_checksum_lo);
60 unsigned int csum_size = sizeof(dummy_csum);
62 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
63 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
64 offset += csum_size;
65 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
66 EXT4_GOOD_OLD_INODE_SIZE - offset);
68 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
69 offset = offsetof(struct ext4_inode, i_checksum_hi);
70 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
71 EXT4_GOOD_OLD_INODE_SIZE,
72 offset - EXT4_GOOD_OLD_INODE_SIZE);
73 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
74 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
75 csum_size);
76 offset += csum_size;
77 }
78 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
79 EXT4_INODE_SIZE(inode->i_sb) - offset);
80 }
82 return csum;
83 }
85 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
86 struct ext4_inode_info *ei)
87 {
88 __u32 provided, calculated;
90 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
91 cpu_to_le32(EXT4_OS_LINUX) ||
92 !ext4_has_metadata_csum(inode->i_sb))
93 return 1;
95 provided = le16_to_cpu(raw->i_checksum_lo);
96 calculated = ext4_inode_csum(inode, raw, ei);
97 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
98 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
99 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
100 else
101 calculated &= 0xFFFF;
103 return provided == calculated;
104 }
106 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
107 struct ext4_inode_info *ei)
108 {
109 __u32 csum;
111 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
112 cpu_to_le32(EXT4_OS_LINUX) ||
113 !ext4_has_metadata_csum(inode->i_sb))
114 return;
116 csum = ext4_inode_csum(inode, raw, ei);
117 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
118 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
119 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
120 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
121 }
123 static inline int ext4_begin_ordered_truncate(struct inode *inode,
124 loff_t new_size)
125 {
126 trace_ext4_begin_ordered_truncate(inode, new_size);
127 /*
128 * If jinode is zero, then we never opened the file for
129 * writing, so there's no need to call
130 * jbd2_journal_begin_ordered_truncate() since there's no
131 * outstanding writes we need to flush.
132 */
133 if (!EXT4_I(inode)->jinode)
134 return 0;
135 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
136 EXT4_I(inode)->jinode,
137 new_size);
138 }
140 static void ext4_invalidatepage(struct page *page, unsigned int offset,
141 unsigned int length);
142 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
143 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
144 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
145 int pextents);
147 /*
148 * Test whether an inode is a fast symlink.
149 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
150 */
151 int ext4_inode_is_fast_symlink(struct inode *inode)
152 {
153 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
154 int ea_blocks = EXT4_I(inode)->i_file_acl ?
155 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
157 if (ext4_has_inline_data(inode))
158 return 0;
160 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
161 }
162 return S_ISLNK(inode->i_mode) && inode->i_size &&
163 (inode->i_size < EXT4_N_BLOCKS * 4);
164 }
166 /*
167 * Restart the transaction associated with *handle. This does a commit,
168 * so before we call here everything must be consistently dirtied against
169 * this transaction.
170 */
171 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
172 int nblocks)
173 {
174 int ret;
176 /*
177 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
178 * moment, get_block can be called only for blocks inside i_size since
179 * page cache has been already dropped and writes are blocked by
180 * i_mutex. So we can safely drop the i_data_sem here.
181 */
182 BUG_ON(EXT4_JOURNAL(inode) == NULL);
183 jbd_debug(2, "restarting handle %p\n", handle);
184 up_write(&EXT4_I(inode)->i_data_sem);
185 ret = ext4_journal_restart(handle, nblocks);
186 down_write(&EXT4_I(inode)->i_data_sem);
187 ext4_discard_preallocations(inode);
189 return ret;
190 }
192 /*
193 * Called at the last iput() if i_nlink is zero.
194 */
195 void ext4_evict_inode(struct inode *inode)
196 {
197 handle_t *handle;
198 int err;
199 int extra_credits = 3;
200 struct ext4_xattr_inode_array *ea_inode_array = NULL;
202 trace_ext4_evict_inode(inode);
204 if (inode->i_nlink) {
205 /*
206 * When journalling data dirty buffers are tracked only in the
207 * journal. So although mm thinks everything is clean and
208 * ready for reaping the inode might still have some pages to
209 * write in the running transaction or waiting to be
210 * checkpointed. Thus calling jbd2_journal_invalidatepage()
211 * (via truncate_inode_pages()) to discard these buffers can
212 * cause data loss. Also even if we did not discard these
213 * buffers, we would have no way to find them after the inode
214 * is reaped and thus user could see stale data if he tries to
215 * read them before the transaction is checkpointed. So be
216 * careful and force everything to disk here... We use
217 * ei->i_datasync_tid to store the newest transaction
218 * containing inode's data.
219 *
220 * Note that directories do not have this problem because they
221 * don't use page cache.
222 */
223 if (inode->i_ino != EXT4_JOURNAL_INO &&
224 ext4_should_journal_data(inode) &&
225 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
226 inode->i_data.nrpages) {
227 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
228 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
230 jbd2_complete_transaction(journal, commit_tid);
231 filemap_write_and_wait(&inode->i_data);
232 }
233 truncate_inode_pages_final(&inode->i_data);
235 goto no_delete;
236 }
238 if (is_bad_inode(inode))
239 goto no_delete;
240 dquot_initialize(inode);
242 if (ext4_should_order_data(inode))
243 ext4_begin_ordered_truncate(inode, 0);
244 truncate_inode_pages_final(&inode->i_data);
246 /*
247 * Protect us against freezing - iput() caller didn't have to have any
248 * protection against it
249 */
250 sb_start_intwrite(inode->i_sb);
252 if (!IS_NOQUOTA(inode))
253 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
255 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
256 ext4_blocks_for_truncate(inode)+extra_credits);
257 if (IS_ERR(handle)) {
258 ext4_std_error(inode->i_sb, PTR_ERR(handle));
259 /*
260 * If we're going to skip the normal cleanup, we still need to
261 * make sure that the in-core orphan linked list is properly
262 * cleaned up.
263 */
264 ext4_orphan_del(NULL, inode);
265 sb_end_intwrite(inode->i_sb);
266 goto no_delete;
267 }
269 if (IS_SYNC(inode))
270 ext4_handle_sync(handle);
272 /*
273 * Set inode->i_size to 0 before calling ext4_truncate(). We need
274 * special handling of symlinks here because i_size is used to
275 * determine whether ext4_inode_info->i_data contains symlink data or
276 * block mappings. Setting i_size to 0 will remove its fast symlink
277 * status. Erase i_data so that it becomes a valid empty block map.
278 */
279 if (ext4_inode_is_fast_symlink(inode))
280 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
281 inode->i_size = 0;
282 err = ext4_mark_inode_dirty(handle, inode);
283 if (err) {
284 ext4_warning(inode->i_sb,
285 "couldn't mark inode dirty (err %d)", err);
286 goto stop_handle;
287 }
288 if (inode->i_blocks) {
289 err = ext4_truncate(inode);
290 if (err) {
291 ext4_error(inode->i_sb,
292 "couldn't truncate inode %lu (err %d)",
293 inode->i_ino, err);
294 goto stop_handle;
295 }
296 }
298 /* Remove xattr references. */
299 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
300 extra_credits);
301 if (err) {
302 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
303 stop_handle:
304 ext4_journal_stop(handle);
305 ext4_orphan_del(NULL, inode);
306 sb_end_intwrite(inode->i_sb);
307 ext4_xattr_inode_array_free(ea_inode_array);
308 goto no_delete;
309 }
311 /*
312 * Kill off the orphan record which ext4_truncate created.
313 * AKPM: I think this can be inside the above `if'.
314 * Note that ext4_orphan_del() has to be able to cope with the
315 * deletion of a non-existent orphan - this is because we don't
316 * know if ext4_truncate() actually created an orphan record.
317 * (Well, we could do this if we need to, but heck - it works)
318 */
319 ext4_orphan_del(handle, inode);
320 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
322 /*
323 * One subtle ordering requirement: if anything has gone wrong
324 * (transaction abort, IO errors, whatever), then we can still
325 * do these next steps (the fs will already have been marked as
326 * having errors), but we can't free the inode if the mark_dirty
327 * fails.
328 */
329 if (ext4_mark_inode_dirty(handle, inode))
330 /* If that failed, just do the required in-core inode clear. */
331 ext4_clear_inode(inode);
332 else
333 ext4_free_inode(handle, inode);
334 ext4_journal_stop(handle);
335 sb_end_intwrite(inode->i_sb);
336 ext4_xattr_inode_array_free(ea_inode_array);
337 return;
338 no_delete:
339 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
340 }
342 #ifdef CONFIG_QUOTA
343 qsize_t *ext4_get_reserved_space(struct inode *inode)
344 {
345 return &EXT4_I(inode)->i_reserved_quota;
346 }
347 #endif
349 /*
350 * Called with i_data_sem down, which is important since we can call
351 * ext4_discard_preallocations() from here.
352 */
353 void ext4_da_update_reserve_space(struct inode *inode,
354 int used, int quota_claim)
355 {
356 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
357 struct ext4_inode_info *ei = EXT4_I(inode);
359 spin_lock(&ei->i_block_reservation_lock);
360 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
361 if (unlikely(used > ei->i_reserved_data_blocks)) {
362 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
363 "with only %d reserved data blocks",
364 __func__, inode->i_ino, used,
365 ei->i_reserved_data_blocks);
366 WARN_ON(1);
367 used = ei->i_reserved_data_blocks;
368 }
370 /* Update per-inode reservations */
371 ei->i_reserved_data_blocks -= used;
372 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
374 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
376 /* Update quota subsystem for data blocks */
377 if (quota_claim)
378 dquot_claim_block(inode, EXT4_C2B(sbi, used));
379 else {
380 /*
381 * We did fallocate with an offset that is already delayed
382 * allocated. So on delayed allocated writeback we should
383 * not re-claim the quota for fallocated blocks.
384 */
385 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
386 }
388 /*
389 * If we have done all the pending block allocations and if
390 * there aren't any writers on the inode, we can discard the
391 * inode's preallocations.
392 */
393 if ((ei->i_reserved_data_blocks == 0) &&
394 (atomic_read(&inode->i_writecount) == 0))
395 ext4_discard_preallocations(inode);
396 }
398 static int __check_block_validity(struct inode *inode, const char *func,
399 unsigned int line,
400 struct ext4_map_blocks *map)
401 {
402 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
403 map->m_len)) {
404 ext4_error_inode(inode, func, line, map->m_pblk,
405 "lblock %lu mapped to illegal pblock %llu "
406 "(length %d)", (unsigned long) map->m_lblk,
407 map->m_pblk, map->m_len);
408 return -EFSCORRUPTED;
409 }
410 return 0;
411 }
413 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
414 ext4_lblk_t len)
415 {
416 int ret;
418 if (ext4_encrypted_inode(inode))
419 return fscrypt_zeroout_range(inode, lblk, pblk, len);
421 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
422 if (ret > 0)
423 ret = 0;
425 return ret;
426 }
428 #define check_block_validity(inode, map) \
429 __check_block_validity((inode), __func__, __LINE__, (map))
431 #ifdef ES_AGGRESSIVE_TEST
432 static void ext4_map_blocks_es_recheck(handle_t *handle,
433 struct inode *inode,
434 struct ext4_map_blocks *es_map,
435 struct ext4_map_blocks *map,
436 int flags)
437 {
438 int retval;
440 map->m_flags = 0;
441 /*
442 * There is a race window that the result is not the same.
443 * e.g. xfstests #223 when dioread_nolock enables. The reason
444 * is that we lookup a block mapping in extent status tree with
445 * out taking i_data_sem. So at the time the unwritten extent
446 * could be converted.
447 */
448 down_read(&EXT4_I(inode)->i_data_sem);
449 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
450 retval = ext4_ext_map_blocks(handle, inode, map, flags &
451 EXT4_GET_BLOCKS_KEEP_SIZE);
452 } else {
453 retval = ext4_ind_map_blocks(handle, inode, map, flags &
454 EXT4_GET_BLOCKS_KEEP_SIZE);
455 }
456 up_read((&EXT4_I(inode)->i_data_sem));
458 /*
459 * We don't check m_len because extent will be collpased in status
460 * tree. So the m_len might not equal.
461 */
462 if (es_map->m_lblk != map->m_lblk ||
463 es_map->m_flags != map->m_flags ||
464 es_map->m_pblk != map->m_pblk) {
465 printk("ES cache assertion failed for inode: %lu "
466 "es_cached ex [%d/%d/%llu/%x] != "
467 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
468 inode->i_ino, es_map->m_lblk, es_map->m_len,
469 es_map->m_pblk, es_map->m_flags, map->m_lblk,
470 map->m_len, map->m_pblk, map->m_flags,
471 retval, flags);
472 }
473 }
474 #endif /* ES_AGGRESSIVE_TEST */
476 /*
477 * The ext4_map_blocks() function tries to look up the requested blocks,
478 * and returns if the blocks are already mapped.
479 *
480 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
481 * and store the allocated blocks in the result buffer head and mark it
482 * mapped.
483 *
484 * If file type is extents based, it will call ext4_ext_map_blocks(),
485 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
486 * based files
487 *
488 * On success, it returns the number of blocks being mapped or allocated. if
489 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
490 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
491 *
492 * It returns 0 if plain look up failed (blocks have not been allocated), in
493 * that case, @map is returned as unmapped but we still do fill map->m_len to
494 * indicate the length of a hole starting at map->m_lblk.
495 *
496 * It returns the error in case of allocation failure.
497 */
498 int ext4_map_blocks(handle_t *handle, struct inode *inode,
499 struct ext4_map_blocks *map, int flags)
500 {
501 struct extent_status es;
502 int retval;
503 int ret = 0;
504 #ifdef ES_AGGRESSIVE_TEST
505 struct ext4_map_blocks orig_map;
507 memcpy(&orig_map, map, sizeof(*map));
508 #endif
510 map->m_flags = 0;
511 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
512 "logical block %lu\n", inode->i_ino, flags, map->m_len,
513 (unsigned long) map->m_lblk);
515 /*
516 * ext4_map_blocks returns an int, and m_len is an unsigned int
517 */
518 if (unlikely(map->m_len > INT_MAX))
519 map->m_len = INT_MAX;
521 /* We can handle the block number less than EXT_MAX_BLOCKS */
522 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
523 return -EFSCORRUPTED;
525 /* Lookup extent status tree firstly */
526 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
527 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
528 map->m_pblk = ext4_es_pblock(&es) +
529 map->m_lblk - es.es_lblk;
530 map->m_flags |= ext4_es_is_written(&es) ?
531 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
532 retval = es.es_len - (map->m_lblk - es.es_lblk);
533 if (retval > map->m_len)
534 retval = map->m_len;
535 map->m_len = retval;
536 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
537 map->m_pblk = 0;
538 retval = es.es_len - (map->m_lblk - es.es_lblk);
539 if (retval > map->m_len)
540 retval = map->m_len;
541 map->m_len = retval;
542 retval = 0;
543 } else {
544 BUG_ON(1);
545 }
546 #ifdef ES_AGGRESSIVE_TEST
547 ext4_map_blocks_es_recheck(handle, inode, map,
548 &orig_map, flags);
549 #endif
550 goto found;
551 }
553 /*
554 * Try to see if we can get the block without requesting a new
555 * file system block.
556 */
557 down_read(&EXT4_I(inode)->i_data_sem);
558 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
559 retval = ext4_ext_map_blocks(handle, inode, map, flags &
560 EXT4_GET_BLOCKS_KEEP_SIZE);
561 } else {
562 retval = ext4_ind_map_blocks(handle, inode, map, flags &
563 EXT4_GET_BLOCKS_KEEP_SIZE);
564 }
565 if (retval > 0) {
566 unsigned int status;
568 if (unlikely(retval != map->m_len)) {
569 ext4_warning(inode->i_sb,
570 "ES len assertion failed for inode "
571 "%lu: retval %d != map->m_len %d",
572 inode->i_ino, retval, map->m_len);
573 WARN_ON(1);
574 }
576 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
577 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
578 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
579 !(status & EXTENT_STATUS_WRITTEN) &&
580 ext4_find_delalloc_range(inode, map->m_lblk,
581 map->m_lblk + map->m_len - 1))
582 status |= EXTENT_STATUS_DELAYED;
583 ret = ext4_es_insert_extent(inode, map->m_lblk,
584 map->m_len, map->m_pblk, status);
585 if (ret < 0)
586 retval = ret;
587 }
588 up_read((&EXT4_I(inode)->i_data_sem));
590 found:
591 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
592 ret = check_block_validity(inode, map);
593 if (ret != 0)
594 return ret;
595 }
597 /* If it is only a block(s) look up */
598 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
599 return retval;
601 /*
602 * Returns if the blocks have already allocated
603 *
604 * Note that if blocks have been preallocated
605 * ext4_ext_get_block() returns the create = 0
606 * with buffer head unmapped.
607 */
608 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
609 /*
610 * If we need to convert extent to unwritten
611 * we continue and do the actual work in
612 * ext4_ext_map_blocks()
613 */
614 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
615 return retval;
617 /*
618 * Here we clear m_flags because after allocating an new extent,
619 * it will be set again.
620 */
621 map->m_flags &= ~EXT4_MAP_FLAGS;
623 /*
624 * New blocks allocate and/or writing to unwritten extent
625 * will possibly result in updating i_data, so we take
626 * the write lock of i_data_sem, and call get_block()
627 * with create == 1 flag.
628 */
629 down_write(&EXT4_I(inode)->i_data_sem);
631 /*
632 * We need to check for EXT4 here because migrate
633 * could have changed the inode type in between
634 */
635 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
636 retval = ext4_ext_map_blocks(handle, inode, map, flags);
637 } else {
638 retval = ext4_ind_map_blocks(handle, inode, map, flags);
640 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
641 /*
642 * We allocated new blocks which will result in
643 * i_data's format changing. Force the migrate
644 * to fail by clearing migrate flags
645 */
646 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
647 }
649 /*
650 * Update reserved blocks/metadata blocks after successful
651 * block allocation which had been deferred till now. We don't
652 * support fallocate for non extent files. So we can update
653 * reserve space here.
654 */
655 if ((retval > 0) &&
656 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
657 ext4_da_update_reserve_space(inode, retval, 1);
658 }
660 if (retval > 0) {
661 unsigned int status;
663 if (unlikely(retval != map->m_len)) {
664 ext4_warning(inode->i_sb,
665 "ES len assertion failed for inode "
666 "%lu: retval %d != map->m_len %d",
667 inode->i_ino, retval, map->m_len);
668 WARN_ON(1);
669 }
671 /*
672 * We have to zeroout blocks before inserting them into extent
673 * status tree. Otherwise someone could look them up there and
674 * use them before they are really zeroed. We also have to
675 * unmap metadata before zeroing as otherwise writeback can
676 * overwrite zeros with stale data from block device.
677 */
678 if (flags & EXT4_GET_BLOCKS_ZERO &&
679 map->m_flags & EXT4_MAP_MAPPED &&
680 map->m_flags & EXT4_MAP_NEW) {
681 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
682 map->m_len);
683 ret = ext4_issue_zeroout(inode, map->m_lblk,
684 map->m_pblk, map->m_len);
685 if (ret) {
686 retval = ret;
687 goto out_sem;
688 }
689 }
691 /*
692 * If the extent has been zeroed out, we don't need to update
693 * extent status tree.
694 */
695 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
696 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
697 if (ext4_es_is_written(&es))
698 goto out_sem;
699 }
700 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
701 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
702 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
703 !(status & EXTENT_STATUS_WRITTEN) &&
704 ext4_find_delalloc_range(inode, map->m_lblk,
705 map->m_lblk + map->m_len - 1))
706 status |= EXTENT_STATUS_DELAYED;
707 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
708 map->m_pblk, status);
709 if (ret < 0) {
710 retval = ret;
711 goto out_sem;
712 }
713 }
715 out_sem:
716 up_write((&EXT4_I(inode)->i_data_sem));
717 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
718 ret = check_block_validity(inode, map);
719 if (ret != 0)
720 return ret;
722 /*
723 * Inodes with freshly allocated blocks where contents will be
724 * visible after transaction commit must be on transaction's
725 * ordered data list.
726 */
727 if (map->m_flags & EXT4_MAP_NEW &&
728 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
729 !(flags & EXT4_GET_BLOCKS_ZERO) &&
730 !ext4_is_quota_file(inode) &&
731 ext4_should_order_data(inode)) {
732 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
733 ret = ext4_jbd2_inode_add_wait(handle, inode);
734 else
735 ret = ext4_jbd2_inode_add_write(handle, inode);
736 if (ret)
737 return ret;
738 }
739 }
740 return retval;
741 }
743 /*
744 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
745 * we have to be careful as someone else may be manipulating b_state as well.
746 */
747 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
748 {
749 unsigned long old_state;
750 unsigned long new_state;
752 flags &= EXT4_MAP_FLAGS;
754 /* Dummy buffer_head? Set non-atomically. */
755 if (!bh->b_page) {
756 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
757 return;
758 }
759 /*
760 * Someone else may be modifying b_state. Be careful! This is ugly but
761 * once we get rid of using bh as a container for mapping information
762 * to pass to / from get_block functions, this can go away.
763 */
764 do {
765 old_state = READ_ONCE(bh->b_state);
766 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
767 } while (unlikely(
768 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
769 }
771 static int _ext4_get_block(struct inode *inode, sector_t iblock,
772 struct buffer_head *bh, int flags)
773 {
774 struct ext4_map_blocks map;
775 int ret = 0;
777 if (ext4_has_inline_data(inode))
778 return -ERANGE;
780 map.m_lblk = iblock;
781 map.m_len = bh->b_size >> inode->i_blkbits;
783 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
784 flags);
785 if (ret > 0) {
786 map_bh(bh, inode->i_sb, map.m_pblk);
787 ext4_update_bh_state(bh, map.m_flags);
788 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
789 ret = 0;
790 } else if (ret == 0) {
791 /* hole case, need to fill in bh->b_size */
792 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
793 }
794 return ret;
795 }
797 int ext4_get_block(struct inode *inode, sector_t iblock,
798 struct buffer_head *bh, int create)
799 {
800 return _ext4_get_block(inode, iblock, bh,
801 create ? EXT4_GET_BLOCKS_CREATE : 0);
802 }
804 /*
805 * Get block function used when preparing for buffered write if we require
806 * creating an unwritten extent if blocks haven't been allocated. The extent
807 * will be converted to written after the IO is complete.
808 */
809 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
810 struct buffer_head *bh_result, int create)
811 {
812 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
813 inode->i_ino, create);
814 return _ext4_get_block(inode, iblock, bh_result,
815 EXT4_GET_BLOCKS_IO_CREATE_EXT);
816 }
818 /* Maximum number of blocks we map for direct IO at once. */
819 #define DIO_MAX_BLOCKS 4096
821 /*
822 * Get blocks function for the cases that need to start a transaction -
823 * generally difference cases of direct IO and DAX IO. It also handles retries
824 * in case of ENOSPC.
825 */
826 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
827 struct buffer_head *bh_result, int flags)
828 {
829 int dio_credits;
830 handle_t *handle;
831 int retries = 0;
832 int ret;
834 /* Trim mapping request to maximum we can map at once for DIO */
835 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
836 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
837 dio_credits = ext4_chunk_trans_blocks(inode,
838 bh_result->b_size >> inode->i_blkbits);
839 retry:
840 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
841 if (IS_ERR(handle))
842 return PTR_ERR(handle);
844 ret = _ext4_get_block(inode, iblock, bh_result, flags);
845 ext4_journal_stop(handle);
847 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
848 goto retry;
849 return ret;
850 }
852 /* Get block function for DIO reads and writes to inodes without extents */
853 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
854 struct buffer_head *bh, int create)
855 {
856 /* We don't expect handle for direct IO */
857 WARN_ON_ONCE(ext4_journal_current_handle());
859 if (!create)
860 return _ext4_get_block(inode, iblock, bh, 0);
861 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
862 }
864 /*
865 * Get block function for AIO DIO writes when we create unwritten extent if
866 * blocks are not allocated yet. The extent will be converted to written
867 * after IO is complete.
868 */
869 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
870 sector_t iblock, struct buffer_head *bh_result, int create)
871 {
872 int ret;
874 /* We don't expect handle for direct IO */
875 WARN_ON_ONCE(ext4_journal_current_handle());
877 ret = ext4_get_block_trans(inode, iblock, bh_result,
878 EXT4_GET_BLOCKS_IO_CREATE_EXT);
880 /*
881 * When doing DIO using unwritten extents, we need io_end to convert
882 * unwritten extents to written on IO completion. We allocate io_end
883 * once we spot unwritten extent and store it in b_private. Generic
884 * DIO code keeps b_private set and furthermore passes the value to
885 * our completion callback in 'private' argument.
886 */
887 if (!ret && buffer_unwritten(bh_result)) {
888 if (!bh_result->b_private) {
889 ext4_io_end_t *io_end;
891 io_end = ext4_init_io_end(inode, GFP_KERNEL);
892 if (!io_end)
893 return -ENOMEM;
894 bh_result->b_private = io_end;
895 ext4_set_io_unwritten_flag(inode, io_end);
896 }
897 set_buffer_defer_completion(bh_result);
898 }
900 return ret;
901 }
903 /*
904 * Get block function for non-AIO DIO writes when we create unwritten extent if
905 * blocks are not allocated yet. The extent will be converted to written
906 * after IO is complete by ext4_direct_IO_write().
907 */
908 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
909 sector_t iblock, struct buffer_head *bh_result, int create)
910 {
911 int ret;
913 /* We don't expect handle for direct IO */
914 WARN_ON_ONCE(ext4_journal_current_handle());
916 ret = ext4_get_block_trans(inode, iblock, bh_result,
917 EXT4_GET_BLOCKS_IO_CREATE_EXT);
919 /*
920 * Mark inode as having pending DIO writes to unwritten extents.
921 * ext4_direct_IO_write() checks this flag and converts extents to
922 * written.
923 */
924 if (!ret && buffer_unwritten(bh_result))
925 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
927 return ret;
928 }
930 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
931 struct buffer_head *bh_result, int create)
932 {
933 int ret;
935 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
936 inode->i_ino, create);
937 /* We don't expect handle for direct IO */
938 WARN_ON_ONCE(ext4_journal_current_handle());
940 ret = _ext4_get_block(inode, iblock, bh_result, 0);
941 /*
942 * Blocks should have been preallocated! ext4_file_write_iter() checks
943 * that.
944 */
945 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
947 return ret;
948 }
951 /*
952 * `handle' can be NULL if create is zero
953 */
954 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
955 ext4_lblk_t block, int map_flags)
956 {
957 struct ext4_map_blocks map;
958 struct buffer_head *bh;
959 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
960 int err;
962 J_ASSERT(handle != NULL || create == 0);
964 map.m_lblk = block;
965 map.m_len = 1;
966 err = ext4_map_blocks(handle, inode, &map, map_flags);
968 if (err == 0)
969 return create ? ERR_PTR(-ENOSPC) : NULL;
970 if (err < 0)
971 return ERR_PTR(err);
973 bh = sb_getblk(inode->i_sb, map.m_pblk);
974 if (unlikely(!bh))
975 return ERR_PTR(-ENOMEM);
976 if (map.m_flags & EXT4_MAP_NEW) {
977 J_ASSERT(create != 0);
978 J_ASSERT(handle != NULL);
980 /*
981 * Now that we do not always journal data, we should
982 * keep in mind whether this should always journal the
983 * new buffer as metadata. For now, regular file
984 * writes use ext4_get_block instead, so it's not a
985 * problem.
986 */
987 lock_buffer(bh);
988 BUFFER_TRACE(bh, "call get_create_access");
989 err = ext4_journal_get_create_access(handle, bh);
990 if (unlikely(err)) {
991 unlock_buffer(bh);
992 goto errout;
993 }
994 if (!buffer_uptodate(bh)) {
995 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
996 set_buffer_uptodate(bh);
997 }
998 unlock_buffer(bh);
999 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1000 err = ext4_handle_dirty_metadata(handle, inode, bh);
1001 if (unlikely(err))
1002 goto errout;
1003 } else
1004 BUFFER_TRACE(bh, "not a new buffer");
1005 return bh;
1006 errout:
1007 brelse(bh);
1008 return ERR_PTR(err);
1009 }
1011 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1012 ext4_lblk_t block, int map_flags)
1013 {
1014 struct buffer_head *bh;
1016 bh = ext4_getblk(handle, inode, block, map_flags);
1017 if (IS_ERR(bh))
1018 return bh;
1019 if (!bh || buffer_uptodate(bh))
1020 return bh;
1021 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1022 wait_on_buffer(bh);
1023 if (buffer_uptodate(bh))
1024 return bh;
1025 put_bh(bh);
1026 return ERR_PTR(-EIO);
1027 }
1029 /* Read a contiguous batch of blocks. */
1030 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1031 bool wait, struct buffer_head **bhs)
1032 {
1033 int i, err;
1035 for (i = 0; i < bh_count; i++) {
1036 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1037 if (IS_ERR(bhs[i])) {
1038 err = PTR_ERR(bhs[i]);
1039 bh_count = i;
1040 goto out_brelse;
1041 }
1042 }
1044 for (i = 0; i < bh_count; i++)
1045 /* Note that NULL bhs[i] is valid because of holes. */
1046 if (bhs[i] && !buffer_uptodate(bhs[i]))
1047 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1048 &bhs[i]);
1050 if (!wait)
1051 return 0;
1053 for (i = 0; i < bh_count; i++)
1054 if (bhs[i])
1055 wait_on_buffer(bhs[i]);
1057 for (i = 0; i < bh_count; i++) {
1058 if (bhs[i] && !buffer_uptodate(bhs[i])) {
1059 err = -EIO;
1060 goto out_brelse;
1061 }
1062 }
1063 return 0;
1065 out_brelse:
1066 for (i = 0; i < bh_count; i++) {
1067 brelse(bhs[i]);
1068 bhs[i] = NULL;
1069 }
1070 return err;
1071 }
1073 int ext4_walk_page_buffers(handle_t *handle,
1074 struct buffer_head *head,
1075 unsigned from,
1076 unsigned to,
1077 int *partial,
1078 int (*fn)(handle_t *handle,
1079 struct buffer_head *bh))
1080 {
1081 struct buffer_head *bh;
1082 unsigned block_start, block_end;
1083 unsigned blocksize = head->b_size;
1084 int err, ret = 0;
1085 struct buffer_head *next;
1087 for (bh = head, block_start = 0;
1088 ret == 0 && (bh != head || !block_start);
1089 block_start = block_end, bh = next) {
1090 next = bh->b_this_page;
1091 block_end = block_start + blocksize;
1092 if (block_end <= from || block_start >= to) {
1093 if (partial && !buffer_uptodate(bh))
1094 *partial = 1;
1095 continue;
1096 }
1097 err = (*fn)(handle, bh);
1098 if (!ret)
1099 ret = err;
1100 }
1101 return ret;
1102 }
1104 /*
1105 * To preserve ordering, it is essential that the hole instantiation and
1106 * the data write be encapsulated in a single transaction. We cannot
1107 * close off a transaction and start a new one between the ext4_get_block()
1108 * and the commit_write(). So doing the jbd2_journal_start at the start of
1109 * prepare_write() is the right place.
1110 *
1111 * Also, this function can nest inside ext4_writepage(). In that case, we
1112 * *know* that ext4_writepage() has generated enough buffer credits to do the
1113 * whole page. So we won't block on the journal in that case, which is good,
1114 * because the caller may be PF_MEMALLOC.
1115 *
1116 * By accident, ext4 can be reentered when a transaction is open via
1117 * quota file writes. If we were to commit the transaction while thus
1118 * reentered, there can be a deadlock - we would be holding a quota
1119 * lock, and the commit would never complete if another thread had a
1120 * transaction open and was blocking on the quota lock - a ranking
1121 * violation.
1122 *
1123 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1124 * will _not_ run commit under these circumstances because handle->h_ref
1125 * is elevated. We'll still have enough credits for the tiny quotafile
1126 * write.
1127 */
1128 int do_journal_get_write_access(handle_t *handle,
1129 struct buffer_head *bh)
1130 {
1131 int dirty = buffer_dirty(bh);
1132 int ret;
1134 if (!buffer_mapped(bh) || buffer_freed(bh))
1135 return 0;
1136 /*
1137 * __block_write_begin() could have dirtied some buffers. Clean
1138 * the dirty bit as jbd2_journal_get_write_access() could complain
1139 * otherwise about fs integrity issues. Setting of the dirty bit
1140 * by __block_write_begin() isn't a real problem here as we clear
1141 * the bit before releasing a page lock and thus writeback cannot
1142 * ever write the buffer.
1143 */
1144 if (dirty)
1145 clear_buffer_dirty(bh);
1146 BUFFER_TRACE(bh, "get write access");
1147 ret = ext4_journal_get_write_access(handle, bh);
1148 if (!ret && dirty)
1149 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1150 return ret;
1151 }
1153 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1154 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1155 get_block_t *get_block)
1156 {
1157 unsigned from = pos & (PAGE_SIZE - 1);
1158 unsigned to = from + len;
1159 struct inode *inode = page->mapping->host;
1160 unsigned block_start, block_end;
1161 sector_t block;
1162 int err = 0;
1163 unsigned blocksize = inode->i_sb->s_blocksize;
1164 unsigned bbits;
1165 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1166 bool decrypt = false;
1168 BUG_ON(!PageLocked(page));
1169 BUG_ON(from > PAGE_SIZE);
1170 BUG_ON(to > PAGE_SIZE);
1171 BUG_ON(from > to);
1173 if (!page_has_buffers(page))
1174 create_empty_buffers(page, blocksize, 0);
1175 head = page_buffers(page);
1176 bbits = ilog2(blocksize);
1177 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1179 for (bh = head, block_start = 0; bh != head || !block_start;
1180 block++, block_start = block_end, bh = bh->b_this_page) {
1181 block_end = block_start + blocksize;
1182 if (block_end <= from || block_start >= to) {
1183 if (PageUptodate(page)) {
1184 if (!buffer_uptodate(bh))
1185 set_buffer_uptodate(bh);
1186 }
1187 continue;
1188 }
1189 if (buffer_new(bh))
1190 clear_buffer_new(bh);
1191 if (!buffer_mapped(bh)) {
1192 WARN_ON(bh->b_size != blocksize);
1193 err = get_block(inode, block, bh, 1);
1194 if (err)
1195 break;
1196 if (buffer_new(bh)) {
1197 clean_bdev_bh_alias(bh);
1198 if (PageUptodate(page)) {
1199 clear_buffer_new(bh);
1200 set_buffer_uptodate(bh);
1201 mark_buffer_dirty(bh);
1202 continue;
1203 }
1204 if (block_end > to || block_start < from)
1205 zero_user_segments(page, to, block_end,
1206 block_start, from);
1207 continue;
1208 }
1209 }
1210 if (PageUptodate(page)) {
1211 if (!buffer_uptodate(bh))
1212 set_buffer_uptodate(bh);
1213 continue;
1214 }
1215 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1216 !buffer_unwritten(bh) &&
1217 (block_start < from || block_end > to)) {
1218 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1219 *wait_bh++ = bh;
1220 decrypt = ext4_encrypted_inode(inode) &&
1221 S_ISREG(inode->i_mode);
1222 }
1223 }
1224 /*
1225 * If we issued read requests, let them complete.
1226 */
1227 while (wait_bh > wait) {
1228 wait_on_buffer(*--wait_bh);
1229 if (!buffer_uptodate(*wait_bh))
1230 err = -EIO;
1231 }
1232 if (unlikely(err))
1233 page_zero_new_buffers(page, from, to);
1234 else if (decrypt)
1235 err = fscrypt_decrypt_page(page->mapping->host, page,
1236 PAGE_SIZE, 0, page->index);
1237 return err;
1238 }
1239 #endif
1241 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1242 loff_t pos, unsigned len, unsigned flags,
1243 struct page **pagep, void **fsdata)
1244 {
1245 struct inode *inode = mapping->host;
1246 int ret, needed_blocks;
1247 handle_t *handle;
1248 int retries = 0;
1249 struct page *page;
1250 pgoff_t index;
1251 unsigned from, to;
1253 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1254 return -EIO;
1256 trace_ext4_write_begin(inode, pos, len, flags);
1257 /*
1258 * Reserve one block more for addition to orphan list in case
1259 * we allocate blocks but write fails for some reason
1260 */
1261 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1262 index = pos >> PAGE_SHIFT;
1263 from = pos & (PAGE_SIZE - 1);
1264 to = from + len;
1266 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1267 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1268 flags, pagep);
1269 if (ret < 0)
1270 return ret;
1271 if (ret == 1)
1272 return 0;
1273 }
1275 /*
1276 * grab_cache_page_write_begin() can take a long time if the
1277 * system is thrashing due to memory pressure, or if the page
1278 * is being written back. So grab it first before we start
1279 * the transaction handle. This also allows us to allocate
1280 * the page (if needed) without using GFP_NOFS.
1281 */
1282 retry_grab:
1283 page = grab_cache_page_write_begin(mapping, index, flags);
1284 if (!page)
1285 return -ENOMEM;
1286 unlock_page(page);
1288 retry_journal:
1289 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1290 if (IS_ERR(handle)) {
1291 put_page(page);
1292 return PTR_ERR(handle);
1293 }
1295 lock_page(page);
1296 if (page->mapping != mapping) {
1297 /* The page got truncated from under us */
1298 unlock_page(page);
1299 put_page(page);
1300 ext4_journal_stop(handle);
1301 goto retry_grab;
1302 }
1303 /* In case writeback began while the page was unlocked */
1304 wait_for_stable_page(page);
1306 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1307 if (ext4_should_dioread_nolock(inode))
1308 ret = ext4_block_write_begin(page, pos, len,
1309 ext4_get_block_unwritten);
1310 else
1311 ret = ext4_block_write_begin(page, pos, len,
1312 ext4_get_block);
1313 #else
1314 if (ext4_should_dioread_nolock(inode))
1315 ret = __block_write_begin(page, pos, len,
1316 ext4_get_block_unwritten);
1317 else
1318 ret = __block_write_begin(page, pos, len, ext4_get_block);
1319 #endif
1320 if (!ret && ext4_should_journal_data(inode)) {
1321 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1322 from, to, NULL,
1323 do_journal_get_write_access);
1324 }
1326 if (ret) {
1327 unlock_page(page);
1328 /*
1329 * __block_write_begin may have instantiated a few blocks
1330 * outside i_size. Trim these off again. Don't need
1331 * i_size_read because we hold i_mutex.
1332 *
1333 * Add inode to orphan list in case we crash before
1334 * truncate finishes
1335 */
1336 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1337 ext4_orphan_add(handle, inode);
1339 ext4_journal_stop(handle);
1340 if (pos + len > inode->i_size) {
1341 ext4_truncate_failed_write(inode);
1342 /*
1343 * If truncate failed early the inode might
1344 * still be on the orphan list; we need to
1345 * make sure the inode is removed from the
1346 * orphan list in that case.
1347 */
1348 if (inode->i_nlink)
1349 ext4_orphan_del(NULL, inode);
1350 }
1352 if (ret == -ENOSPC &&
1353 ext4_should_retry_alloc(inode->i_sb, &retries))
1354 goto retry_journal;
1355 put_page(page);
1356 return ret;
1357 }
1358 *pagep = page;
1359 return ret;
1360 }
1362 /* For write_end() in data=journal mode */
1363 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1364 {
1365 int ret;
1366 if (!buffer_mapped(bh) || buffer_freed(bh))
1367 return 0;
1368 set_buffer_uptodate(bh);
1369 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1370 clear_buffer_meta(bh);
1371 clear_buffer_prio(bh);
1372 return ret;
1373 }
1375 /*
1376 * We need to pick up the new inode size which generic_commit_write gave us
1377 * `file' can be NULL - eg, when called from page_symlink().
1378 *
1379 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1380 * buffers are managed internally.
1381 */
1382 static int ext4_write_end(struct file *file,
1383 struct address_space *mapping,
1384 loff_t pos, unsigned len, unsigned copied,
1385 struct page *page, void *fsdata)
1386 {
1387 handle_t *handle = ext4_journal_current_handle();
1388 struct inode *inode = mapping->host;
1389 loff_t old_size = inode->i_size;
1390 int ret = 0, ret2;
1391 int i_size_changed = 0;
1392 int inline_data = ext4_has_inline_data(inode);
1394 trace_ext4_write_end(inode, pos, len, copied);
1395 if (inline_data) {
1396 ret = ext4_write_inline_data_end(inode, pos, len,
1397 copied, page);
1398 if (ret < 0) {
1399 unlock_page(page);
1400 put_page(page);
1401 goto errout;
1402 }
1403 copied = ret;
1404 } else
1405 copied = block_write_end(file, mapping, pos,
1406 len, copied, page, fsdata);
1407 /*
1408 * it's important to update i_size while still holding page lock:
1409 * page writeout could otherwise come in and zero beyond i_size.
1410 */
1411 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1412 unlock_page(page);
1413 put_page(page);
1415 if (old_size < pos)
1416 pagecache_isize_extended(inode, old_size, pos);
1417 /*
1418 * Don't mark the inode dirty under page lock. First, it unnecessarily
1419 * makes the holding time of page lock longer. Second, it forces lock
1420 * ordering of page lock and transaction start for journaling
1421 * filesystems.
1422 */
1423 if (i_size_changed || inline_data)
1424 ext4_mark_inode_dirty(handle, inode);
1426 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1427 /* if we have allocated more blocks and copied
1428 * less. We will have blocks allocated outside
1429 * inode->i_size. So truncate them
1430 */
1431 ext4_orphan_add(handle, inode);
1432 errout:
1433 ret2 = ext4_journal_stop(handle);
1434 if (!ret)
1435 ret = ret2;
1437 if (pos + len > inode->i_size) {
1438 ext4_truncate_failed_write(inode);
1439 /*
1440 * If truncate failed early the inode might still be
1441 * on the orphan list; we need to make sure the inode
1442 * is removed from the orphan list in that case.
1443 */
1444 if (inode->i_nlink)
1445 ext4_orphan_del(NULL, inode);
1446 }
1448 return ret ? ret : copied;
1449 }
1451 /*
1452 * This is a private version of page_zero_new_buffers() which doesn't
1453 * set the buffer to be dirty, since in data=journalled mode we need
1454 * to call ext4_handle_dirty_metadata() instead.
1455 */
1456 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1457 struct page *page,
1458 unsigned from, unsigned to)
1459 {
1460 unsigned int block_start = 0, block_end;
1461 struct buffer_head *head, *bh;
1463 bh = head = page_buffers(page);
1464 do {
1465 block_end = block_start + bh->b_size;
1466 if (buffer_new(bh)) {
1467 if (block_end > from && block_start < to) {
1468 if (!PageUptodate(page)) {
1469 unsigned start, size;
1471 start = max(from, block_start);
1472 size = min(to, block_end) - start;
1474 zero_user(page, start, size);
1475 write_end_fn(handle, bh);
1476 }
1477 clear_buffer_new(bh);
1478 }
1479 }
1480 block_start = block_end;
1481 bh = bh->b_this_page;
1482 } while (bh != head);
1483 }
1485 static int ext4_journalled_write_end(struct file *file,
1486 struct address_space *mapping,
1487 loff_t pos, unsigned len, unsigned copied,
1488 struct page *page, void *fsdata)
1489 {
1490 handle_t *handle = ext4_journal_current_handle();
1491 struct inode *inode = mapping->host;
1492 loff_t old_size = inode->i_size;
1493 int ret = 0, ret2;
1494 int partial = 0;
1495 unsigned from, to;
1496 int size_changed = 0;
1497 int inline_data = ext4_has_inline_data(inode);
1499 trace_ext4_journalled_write_end(inode, pos, len, copied);
1500 from = pos & (PAGE_SIZE - 1);
1501 to = from + len;
1503 BUG_ON(!ext4_handle_valid(handle));
1505 if (inline_data) {
1506 ret = ext4_write_inline_data_end(inode, pos, len,
1507 copied, page);
1508 if (ret < 0) {
1509 unlock_page(page);
1510 put_page(page);
1511 goto errout;
1512 }
1513 copied = ret;
1514 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1515 copied = 0;
1516 ext4_journalled_zero_new_buffers(handle, page, from, to);
1517 } else {
1518 if (unlikely(copied < len))
1519 ext4_journalled_zero_new_buffers(handle, page,
1520 from + copied, to);
1521 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1522 from + copied, &partial,
1523 write_end_fn);
1524 if (!partial)
1525 SetPageUptodate(page);
1526 }
1527 size_changed = ext4_update_inode_size(inode, pos + copied);
1528 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1529 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1530 unlock_page(page);
1531 put_page(page);
1533 if (old_size < pos)
1534 pagecache_isize_extended(inode, old_size, pos);
1536 if (size_changed || inline_data) {
1537 ret2 = ext4_mark_inode_dirty(handle, inode);
1538 if (!ret)
1539 ret = ret2;
1540 }
1542 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1543 /* if we have allocated more blocks and copied
1544 * less. We will have blocks allocated outside
1545 * inode->i_size. So truncate them
1546 */
1547 ext4_orphan_add(handle, inode);
1549 errout:
1550 ret2 = ext4_journal_stop(handle);
1551 if (!ret)
1552 ret = ret2;
1553 if (pos + len > inode->i_size) {
1554 ext4_truncate_failed_write(inode);
1555 /*
1556 * If truncate failed early the inode might still be
1557 * on the orphan list; we need to make sure the inode
1558 * is removed from the orphan list in that case.
1559 */
1560 if (inode->i_nlink)
1561 ext4_orphan_del(NULL, inode);
1562 }
1564 return ret ? ret : copied;
1565 }
1567 /*
1568 * Reserve space for a single cluster
1569 */
1570 static int ext4_da_reserve_space(struct inode *inode)
1571 {
1572 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1573 struct ext4_inode_info *ei = EXT4_I(inode);
1574 int ret;
1576 /*
1577 * We will charge metadata quota at writeout time; this saves
1578 * us from metadata over-estimation, though we may go over by
1579 * a small amount in the end. Here we just reserve for data.
1580 */
1581 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1582 if (ret)
1583 return ret;
1585 spin_lock(&ei->i_block_reservation_lock);
1586 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1587 spin_unlock(&ei->i_block_reservation_lock);
1588 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1589 return -ENOSPC;
1590 }
1591 ei->i_reserved_data_blocks++;
1592 trace_ext4_da_reserve_space(inode);
1593 spin_unlock(&ei->i_block_reservation_lock);
1595 return 0; /* success */
1596 }
1598 static void ext4_da_release_space(struct inode *inode, int to_free)
1599 {
1600 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1601 struct ext4_inode_info *ei = EXT4_I(inode);
1603 if (!to_free)
1604 return; /* Nothing to release, exit */
1606 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1608 trace_ext4_da_release_space(inode, to_free);
1609 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1610 /*
1611 * if there aren't enough reserved blocks, then the
1612 * counter is messed up somewhere. Since this
1613 * function is called from invalidate page, it's
1614 * harmless to return without any action.
1615 */
1616 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1617 "ino %lu, to_free %d with only %d reserved "
1618 "data blocks", inode->i_ino, to_free,
1619 ei->i_reserved_data_blocks);
1620 WARN_ON(1);
1621 to_free = ei->i_reserved_data_blocks;
1622 }
1623 ei->i_reserved_data_blocks -= to_free;
1625 /* update fs dirty data blocks counter */
1626 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1628 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1630 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1631 }
1633 static void ext4_da_page_release_reservation(struct page *page,
1634 unsigned int offset,
1635 unsigned int length)
1636 {
1637 int to_release = 0, contiguous_blks = 0;
1638 struct buffer_head *head, *bh;
1639 unsigned int curr_off = 0;
1640 struct inode *inode = page->mapping->host;
1641 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1642 unsigned int stop = offset + length;
1643 int num_clusters;
1644 ext4_fsblk_t lblk;
1646 BUG_ON(stop > PAGE_SIZE || stop < length);
1648 head = page_buffers(page);
1649 bh = head;
1650 do {
1651 unsigned int next_off = curr_off + bh->b_size;
1653 if (next_off > stop)
1654 break;
1656 if ((offset <= curr_off) && (buffer_delay(bh))) {
1657 to_release++;
1658 contiguous_blks++;
1659 clear_buffer_delay(bh);
1660 } else if (contiguous_blks) {
1661 lblk = page->index <<
1662 (PAGE_SHIFT - inode->i_blkbits);
1663 lblk += (curr_off >> inode->i_blkbits) -
1664 contiguous_blks;
1665 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1666 contiguous_blks = 0;
1667 }
1668 curr_off = next_off;
1669 } while ((bh = bh->b_this_page) != head);
1671 if (contiguous_blks) {
1672 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1673 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1674 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1675 }
1677 /* If we have released all the blocks belonging to a cluster, then we
1678 * need to release the reserved space for that cluster. */
1679 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1680 while (num_clusters > 0) {
1681 lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1682 ((num_clusters - 1) << sbi->s_cluster_bits);
1683 if (sbi->s_cluster_ratio == 1 ||
1684 !ext4_find_delalloc_cluster(inode, lblk))
1685 ext4_da_release_space(inode, 1);
1687 num_clusters--;
1688 }
1689 }
1691 /*
1692 * Delayed allocation stuff
1693 */
1695 struct mpage_da_data {
1696 struct inode *inode;
1697 struct writeback_control *wbc;
1699 pgoff_t first_page; /* The first page to write */
1700 pgoff_t next_page; /* Current page to examine */
1701 pgoff_t last_page; /* Last page to examine */
1702 /*
1703 * Extent to map - this can be after first_page because that can be
1704 * fully mapped. We somewhat abuse m_flags to store whether the extent
1705 * is delalloc or unwritten.
1706 */
1707 struct ext4_map_blocks map;
1708 struct ext4_io_submit io_submit; /* IO submission data */
1709 unsigned int do_map:1;
1710 };
1712 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1713 bool invalidate)
1714 {
1715 int nr_pages, i;
1716 pgoff_t index, end;
1717 struct pagevec pvec;
1718 struct inode *inode = mpd->inode;
1719 struct address_space *mapping = inode->i_mapping;
1721 /* This is necessary when next_page == 0. */
1722 if (mpd->first_page >= mpd->next_page)
1723 return;
1725 index = mpd->first_page;
1726 end = mpd->next_page - 1;
1727 if (invalidate) {
1728 ext4_lblk_t start, last;
1729 start = index << (PAGE_SHIFT - inode->i_blkbits);
1730 last = end << (PAGE_SHIFT - inode->i_blkbits);
1731 ext4_es_remove_extent(inode, start, last - start + 1);
1732 }
1734 pagevec_init(&pvec);
1735 while (index <= end) {
1736 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1737 if (nr_pages == 0)
1738 break;
1739 for (i = 0; i < nr_pages; i++) {
1740 struct page *page = pvec.pages[i];
1742 BUG_ON(!PageLocked(page));
1743 BUG_ON(PageWriteback(page));
1744 if (invalidate) {
1745 if (page_mapped(page))
1746 clear_page_dirty_for_io(page);
1747 block_invalidatepage(page, 0, PAGE_SIZE);
1748 ClearPageUptodate(page);
1749 }
1750 unlock_page(page);
1751 }
1752 pagevec_release(&pvec);
1753 }
1754 }
1756 static void ext4_print_free_blocks(struct inode *inode)
1757 {
1758 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1759 struct super_block *sb = inode->i_sb;
1760 struct ext4_inode_info *ei = EXT4_I(inode);
1762 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1763 EXT4_C2B(EXT4_SB(inode->i_sb),
1764 ext4_count_free_clusters(sb)));
1765 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1766 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1767 (long long) EXT4_C2B(EXT4_SB(sb),
1768 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1769 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1770 (long long) EXT4_C2B(EXT4_SB(sb),
1771 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1772 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1773 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1774 ei->i_reserved_data_blocks);
1775 return;
1776 }
1778 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1779 {
1780 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1781 }
1783 /*
1784 * This function is grabs code from the very beginning of
1785 * ext4_map_blocks, but assumes that the caller is from delayed write
1786 * time. This function looks up the requested blocks and sets the
1787 * buffer delay bit under the protection of i_data_sem.
1788 */
1789 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1790 struct ext4_map_blocks *map,
1791 struct buffer_head *bh)
1792 {
1793 struct extent_status es;
1794 int retval;
1795 sector_t invalid_block = ~((sector_t) 0xffff);
1796 #ifdef ES_AGGRESSIVE_TEST
1797 struct ext4_map_blocks orig_map;
1799 memcpy(&orig_map, map, sizeof(*map));
1800 #endif
1802 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1803 invalid_block = ~0;
1805 map->m_flags = 0;
1806 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1807 "logical block %lu\n", inode->i_ino, map->m_len,
1808 (unsigned long) map->m_lblk);
1810 /* Lookup extent status tree firstly */
1811 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1812 if (ext4_es_is_hole(&es)) {
1813 retval = 0;
1814 down_read(&EXT4_I(inode)->i_data_sem);
1815 goto add_delayed;
1816 }
1818 /*
1819 * Delayed extent could be allocated by fallocate.
1820 * So we need to check it.
1821 */
1822 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1823 map_bh(bh, inode->i_sb, invalid_block);
1824 set_buffer_new(bh);
1825 set_buffer_delay(bh);
1826 return 0;
1827 }
1829 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1830 retval = es.es_len - (iblock - es.es_lblk);
1831 if (retval > map->m_len)
1832 retval = map->m_len;
1833 map->m_len = retval;
1834 if (ext4_es_is_written(&es))
1835 map->m_flags |= EXT4_MAP_MAPPED;
1836 else if (ext4_es_is_unwritten(&es))
1837 map->m_flags |= EXT4_MAP_UNWRITTEN;
1838 else
1839 BUG_ON(1);
1841 #ifdef ES_AGGRESSIVE_TEST
1842 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1843 #endif
1844 return retval;
1845 }
1847 /*
1848 * Try to see if we can get the block without requesting a new
1849 * file system block.
1850 */
1851 down_read(&EXT4_I(inode)->i_data_sem);
1852 if (ext4_has_inline_data(inode))
1853 retval = 0;
1854 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1855 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1856 else
1857 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1859 add_delayed:
1860 if (retval == 0) {
1861 int ret;
1862 /*
1863 * XXX: __block_prepare_write() unmaps passed block,
1864 * is it OK?
1865 */
1866 /*
1867 * If the block was allocated from previously allocated cluster,
1868 * then we don't need to reserve it again. However we still need
1869 * to reserve metadata for every block we're going to write.
1870 */
1871 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1872 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1873 ret = ext4_da_reserve_space(inode);
1874 if (ret) {
1875 /* not enough space to reserve */
1876 retval = ret;
1877 goto out_unlock;
1878 }
1879 }
1881 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1882 ~0, EXTENT_STATUS_DELAYED);
1883 if (ret) {
1884 retval = ret;
1885 goto out_unlock;
1886 }
1888 map_bh(bh, inode->i_sb, invalid_block);
1889 set_buffer_new(bh);
1890 set_buffer_delay(bh);
1891 } else if (retval > 0) {
1892 int ret;
1893 unsigned int status;
1895 if (unlikely(retval != map->m_len)) {
1896 ext4_warning(inode->i_sb,
1897 "ES len assertion failed for inode "
1898 "%lu: retval %d != map->m_len %d",
1899 inode->i_ino, retval, map->m_len);
1900 WARN_ON(1);
1901 }
1903 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1904 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1905 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1906 map->m_pblk, status);
1907 if (ret != 0)
1908 retval = ret;
1909 }
1911 out_unlock:
1912 up_read((&EXT4_I(inode)->i_data_sem));
1914 return retval;
1915 }
1917 /*
1918 * This is a special get_block_t callback which is used by
1919 * ext4_da_write_begin(). It will either return mapped block or
1920 * reserve space for a single block.
1921 *
1922 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1923 * We also have b_blocknr = -1 and b_bdev initialized properly
1924 *
1925 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1926 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1927 * initialized properly.
1928 */
1929 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1930 struct buffer_head *bh, int create)
1931 {
1932 struct ext4_map_blocks map;
1933 int ret = 0;
1935 BUG_ON(create == 0);
1936 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1938 map.m_lblk = iblock;
1939 map.m_len = 1;
1941 /*
1942 * first, we need to know whether the block is allocated already
1943 * preallocated blocks are unmapped but should treated
1944 * the same as allocated blocks.
1945 */
1946 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1947 if (ret <= 0)
1948 return ret;
1950 map_bh(bh, inode->i_sb, map.m_pblk);
1951 ext4_update_bh_state(bh, map.m_flags);
1953 if (buffer_unwritten(bh)) {
1954 /* A delayed write to unwritten bh should be marked
1955 * new and mapped. Mapped ensures that we don't do
1956 * get_block multiple times when we write to the same
1957 * offset and new ensures that we do proper zero out
1958 * for partial write.
1959 */
1960 set_buffer_new(bh);
1961 set_buffer_mapped(bh);
1962 }
1963 return 0;
1964 }
1966 static int bget_one(handle_t *handle, struct buffer_head *bh)
1967 {
1968 get_bh(bh);
1969 return 0;
1970 }
1972 static int bput_one(handle_t *handle, struct buffer_head *bh)
1973 {
1974 put_bh(bh);
1975 return 0;
1976 }
1978 static int __ext4_journalled_writepage(struct page *page,
1979 unsigned int len)
1980 {
1981 struct address_space *mapping = page->mapping;
1982 struct inode *inode = mapping->host;
1983 struct buffer_head *page_bufs = NULL;
1984 handle_t *handle = NULL;
1985 int ret = 0, err = 0;
1986 int inline_data = ext4_has_inline_data(inode);
1987 struct buffer_head *inode_bh = NULL;
1989 ClearPageChecked(page);
1991 if (inline_data) {
1992 BUG_ON(page->index != 0);
1993 BUG_ON(len > ext4_get_max_inline_size(inode));
1994 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1995 if (inode_bh == NULL)
1996 goto out;
1997 } else {
1998 page_bufs = page_buffers(page);
1999 if (!page_bufs) {
2000 BUG();
2001 goto out;
2002 }
2003 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2004 NULL, bget_one);
2005 }
2006 /*
2007 * We need to release the page lock before we start the
2008 * journal, so grab a reference so the page won't disappear
2009 * out from under us.
2010 */
2011 get_page(page);
2012 unlock_page(page);
2014 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2015 ext4_writepage_trans_blocks(inode));
2016 if (IS_ERR(handle)) {
2017 ret = PTR_ERR(handle);
2018 put_page(page);
2019 goto out_no_pagelock;
2020 }
2021 BUG_ON(!ext4_handle_valid(handle));
2023 lock_page(page);
2024 put_page(page);
2025 if (page->mapping != mapping) {
2026 /* The page got truncated from under us */
2027 ext4_journal_stop(handle);
2028 ret = 0;
2029 goto out;
2030 }
2032 if (inline_data) {
2033 ret = ext4_mark_inode_dirty(handle, inode);
2034 } else {
2035 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2036 do_journal_get_write_access);
2038 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2039 write_end_fn);
2040 }
2041 if (ret == 0)
2042 ret = err;
2043 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2044 err = ext4_journal_stop(handle);
2045 if (!ret)
2046 ret = err;
2048 if (!ext4_has_inline_data(inode))
2049 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2050 NULL, bput_one);
2051 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2052 out:
2053 unlock_page(page);
2054 out_no_pagelock:
2055 brelse(inode_bh);
2056 return ret;
2057 }
2059 /*
2060 * Note that we don't need to start a transaction unless we're journaling data
2061 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2062 * need to file the inode to the transaction's list in ordered mode because if
2063 * we are writing back data added by write(), the inode is already there and if
2064 * we are writing back data modified via mmap(), no one guarantees in which
2065 * transaction the data will hit the disk. In case we are journaling data, we
2066 * cannot start transaction directly because transaction start ranks above page
2067 * lock so we have to do some magic.
2068 *
2069 * This function can get called via...
2070 * - ext4_writepages after taking page lock (have journal handle)
2071 * - journal_submit_inode_data_buffers (no journal handle)
2072 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2073 * - grab_page_cache when doing write_begin (have journal handle)
2074 *
2075 * We don't do any block allocation in this function. If we have page with
2076 * multiple blocks we need to write those buffer_heads that are mapped. This
2077 * is important for mmaped based write. So if we do with blocksize 1K
2078 * truncate(f, 1024);
2079 * a = mmap(f, 0, 4096);
2080 * a[0] = 'a';
2081 * truncate(f, 4096);
2082 * we have in the page first buffer_head mapped via page_mkwrite call back
2083 * but other buffer_heads would be unmapped but dirty (dirty done via the
2084 * do_wp_page). So writepage should write the first block. If we modify
2085 * the mmap area beyond 1024 we will again get a page_fault and the
2086 * page_mkwrite callback will do the block allocation and mark the
2087 * buffer_heads mapped.
2088 *
2089 * We redirty the page if we have any buffer_heads that is either delay or
2090 * unwritten in the page.
2091 *
2092 * We can get recursively called as show below.
2093 *
2094 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2095 * ext4_writepage()
2096 *
2097 * But since we don't do any block allocation we should not deadlock.
2098 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2099 */
2100 static int ext4_writepage(struct page *page,
2101 struct writeback_control *wbc)
2102 {
2103 int ret = 0;
2104 loff_t size;
2105 unsigned int len;
2106 struct buffer_head *page_bufs = NULL;
2107 struct inode *inode = page->mapping->host;
2108 struct ext4_io_submit io_submit;
2109 bool keep_towrite = false;
2111 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2112 ext4_invalidatepage(page, 0, PAGE_SIZE);
2113 unlock_page(page);
2114 return -EIO;
2115 }
2117 trace_ext4_writepage(page);
2118 size = i_size_read(inode);
2119 if (page->index == size >> PAGE_SHIFT)
2120 len = size & ~PAGE_MASK;
2121 else
2122 len = PAGE_SIZE;
2124 page_bufs = page_buffers(page);
2125 /*
2126 * We cannot do block allocation or other extent handling in this
2127 * function. If there are buffers needing that, we have to redirty
2128 * the page. But we may reach here when we do a journal commit via
2129 * journal_submit_inode_data_buffers() and in that case we must write
2130 * allocated buffers to achieve data=ordered mode guarantees.
2131 *
2132 * Also, if there is only one buffer per page (the fs block
2133 * size == the page size), if one buffer needs block
2134 * allocation or needs to modify the extent tree to clear the
2135 * unwritten flag, we know that the page can't be written at
2136 * all, so we might as well refuse the write immediately.
2137 * Unfortunately if the block size != page size, we can't as
2138 * easily detect this case using ext4_walk_page_buffers(), but
2139 * for the extremely common case, this is an optimization that
2140 * skips a useless round trip through ext4_bio_write_page().
2141 */
2142 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2143 ext4_bh_delay_or_unwritten)) {
2144 redirty_page_for_writepage(wbc, page);
2145 if ((current->flags & PF_MEMALLOC) ||
2146 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2147 /*
2148 * For memory cleaning there's no point in writing only
2149 * some buffers. So just bail out. Warn if we came here
2150 * from direct reclaim.
2151 */
2152 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2153 == PF_MEMALLOC);
2154 unlock_page(page);
2155 return 0;
2156 }
2157 keep_towrite = true;
2158 }
2160 if (PageChecked(page) && ext4_should_journal_data(inode))
2161 /*
2162 * It's mmapped pagecache. Add buffers and journal it. There
2163 * doesn't seem much point in redirtying the page here.
2164 */
2165 return __ext4_journalled_writepage(page, len);
2167 ext4_io_submit_init(&io_submit, wbc);
2168 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2169 if (!io_submit.io_end) {
2170 redirty_page_for_writepage(wbc, page);
2171 unlock_page(page);
2172 return -ENOMEM;
2173 }
2174 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2175 ext4_io_submit(&io_submit);
2176 /* Drop io_end reference we got from init */
2177 ext4_put_io_end_defer(io_submit.io_end);
2178 return ret;
2179 }
2181 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2182 {
2183 int len;
2184 loff_t size;
2185 int err;
2187 BUG_ON(page->index != mpd->first_page);
2188 clear_page_dirty_for_io(page);
2189 /*
2190 * We have to be very careful here! Nothing protects writeback path
2191 * against i_size changes and the page can be writeably mapped into
2192 * page tables. So an application can be growing i_size and writing
2193 * data through mmap while writeback runs. clear_page_dirty_for_io()
2194 * write-protects our page in page tables and the page cannot get
2195 * written to again until we release page lock. So only after
2196 * clear_page_dirty_for_io() we are safe to sample i_size for
2197 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2198 * on the barrier provided by TestClearPageDirty in
2199 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2200 * after page tables are updated.
2201 */
2202 size = i_size_read(mpd->inode);
2203 if (page->index == size >> PAGE_SHIFT)
2204 len = size & ~PAGE_MASK;
2205 else
2206 len = PAGE_SIZE;
2207 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2208 if (!err)
2209 mpd->wbc->nr_to_write--;
2210 mpd->first_page++;
2212 return err;
2213 }
2215 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2217 /*
2218 * mballoc gives us at most this number of blocks...
2219 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2220 * The rest of mballoc seems to handle chunks up to full group size.
2221 */
2222 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2224 /*
2225 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2226 *
2227 * @mpd - extent of blocks
2228 * @lblk - logical number of the block in the file
2229 * @bh - buffer head we want to add to the extent
2230 *
2231 * The function is used to collect contig. blocks in the same state. If the
2232 * buffer doesn't require mapping for writeback and we haven't started the
2233 * extent of buffers to map yet, the function returns 'true' immediately - the
2234 * caller can write the buffer right away. Otherwise the function returns true
2235 * if the block has been added to the extent, false if the block couldn't be
2236 * added.
2237 */
2238 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2239 struct buffer_head *bh)
2240 {
2241 struct ext4_map_blocks *map = &mpd->map;
2243 /* Buffer that doesn't need mapping for writeback? */
2244 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2245 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2246 /* So far no extent to map => we write the buffer right away */
2247 if (map->m_len == 0)
2248 return true;
2249 return false;
2250 }
2252 /* First block in the extent? */
2253 if (map->m_len == 0) {
2254 /* We cannot map unless handle is started... */
2255 if (!mpd->do_map)
2256 return false;
2257 map->m_lblk = lblk;
2258 map->m_len = 1;
2259 map->m_flags = bh->b_state & BH_FLAGS;
2260 return true;
2261 }
2263 /* Don't go larger than mballoc is willing to allocate */
2264 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2265 return false;
2267 /* Can we merge the block to our big extent? */
2268 if (lblk == map->m_lblk + map->m_len &&
2269 (bh->b_state & BH_FLAGS) == map->m_flags) {
2270 map->m_len++;
2271 return true;
2272 }
2273 return false;
2274 }
2276 /*
2277 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2278 *
2279 * @mpd - extent of blocks for mapping
2280 * @head - the first buffer in the page
2281 * @bh - buffer we should start processing from
2282 * @lblk - logical number of the block in the file corresponding to @bh
2283 *
2284 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2285 * the page for IO if all buffers in this page were mapped and there's no
2286 * accumulated extent of buffers to map or add buffers in the page to the
2287 * extent of buffers to map. The function returns 1 if the caller can continue
2288 * by processing the next page, 0 if it should stop adding buffers to the
2289 * extent to map because we cannot extend it anymore. It can also return value
2290 * < 0 in case of error during IO submission.
2291 */
2292 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2293 struct buffer_head *head,
2294 struct buffer_head *bh,
2295 ext4_lblk_t lblk)
2296 {
2297 struct inode *inode = mpd->inode;
2298 int err;
2299 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2300 >> inode->i_blkbits;
2302 do {
2303 BUG_ON(buffer_locked(bh));
2305 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2306 /* Found extent to map? */
2307 if (mpd->map.m_len)
2308 return 0;
2309 /* Buffer needs mapping and handle is not started? */
2310 if (!mpd->do_map)
2311 return 0;
2312 /* Everything mapped so far and we hit EOF */
2313 break;
2314 }
2315 } while (lblk++, (bh = bh->b_this_page) != head);
2316 /* So far everything mapped? Submit the page for IO. */
2317 if (mpd->map.m_len == 0) {
2318 err = mpage_submit_page(mpd, head->b_page);
2319 if (err < 0)
2320 return err;
2321 }
2322 return lblk < blocks;
2323 }
2325 /*
2326 * mpage_map_buffers - update buffers corresponding to changed extent and
2327 * submit fully mapped pages for IO
2328 *
2329 * @mpd - description of extent to map, on return next extent to map
2330 *
2331 * Scan buffers corresponding to changed extent (we expect corresponding pages
2332 * to be already locked) and update buffer state according to new extent state.
2333 * We map delalloc buffers to their physical location, clear unwritten bits,
2334 * and mark buffers as uninit when we perform writes to unwritten extents
2335 * and do extent conversion after IO is finished. If the last page is not fully
2336 * mapped, we update @map to the next extent in the last page that needs
2337 * mapping. Otherwise we submit the page for IO.
2338 */
2339 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2340 {
2341 struct pagevec pvec;
2342 int nr_pages, i;
2343 struct inode *inode = mpd->inode;
2344 struct buffer_head *head, *bh;
2345 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2346 pgoff_t start, end;
2347 ext4_lblk_t lblk;
2348 sector_t pblock;
2349 int err;
2351 start = mpd->map.m_lblk >> bpp_bits;
2352 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2353 lblk = start << bpp_bits;
2354 pblock = mpd->map.m_pblk;
2356 pagevec_init(&pvec);
2357 while (start <= end) {
2358 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2359 &start, end);
2360 if (nr_pages == 0)
2361 break;
2362 for (i = 0; i < nr_pages; i++) {
2363 struct page *page = pvec.pages[i];
2365 bh = head = page_buffers(page);
2366 do {
2367 if (lblk < mpd->map.m_lblk)
2368 continue;
2369 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2370 /*
2371 * Buffer after end of mapped extent.
2372 * Find next buffer in the page to map.
2373 */
2374 mpd->map.m_len = 0;
2375 mpd->map.m_flags = 0;
2376 /*
2377 * FIXME: If dioread_nolock supports
2378 * blocksize < pagesize, we need to make
2379 * sure we add size mapped so far to
2380 * io_end->size as the following call
2381 * can submit the page for IO.
2382 */
2383 err = mpage_process_page_bufs(mpd, head,
2384 bh, lblk);
2385 pagevec_release(&pvec);
2386 if (err > 0)
2387 err = 0;
2388 return err;
2389 }
2390 if (buffer_delay(bh)) {
2391 clear_buffer_delay(bh);
2392 bh->b_blocknr = pblock++;
2393 }
2394 clear_buffer_unwritten(bh);
2395 } while (lblk++, (bh = bh->b_this_page) != head);
2397 /*
2398 * FIXME: This is going to break if dioread_nolock
2399 * supports blocksize < pagesize as we will try to
2400 * convert potentially unmapped parts of inode.
2401 */
2402 mpd->io_submit.io_end->size += PAGE_SIZE;
2403 /* Page fully mapped - let IO run! */
2404 err = mpage_submit_page(mpd, page);
2405 if (err < 0) {
2406 pagevec_release(&pvec);
2407 return err;
2408 }
2409 }
2410 pagevec_release(&pvec);
2411 }
2412 /* Extent fully mapped and matches with page boundary. We are done. */
2413 mpd->map.m_len = 0;
2414 mpd->map.m_flags = 0;
2415 return 0;
2416 }
2418 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2419 {
2420 struct inode *inode = mpd->inode;
2421 struct ext4_map_blocks *map = &mpd->map;
2422 int get_blocks_flags;
2423 int err, dioread_nolock;
2425 trace_ext4_da_write_pages_extent(inode, map);
2426 /*
2427 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2428 * to convert an unwritten extent to be initialized (in the case
2429 * where we have written into one or more preallocated blocks). It is
2430 * possible that we're going to need more metadata blocks than
2431 * previously reserved. However we must not fail because we're in
2432 * writeback and there is nothing we can do about it so it might result
2433 * in data loss. So use reserved blocks to allocate metadata if
2434 * possible.
2435 *
2436 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2437 * the blocks in question are delalloc blocks. This indicates
2438 * that the blocks and quotas has already been checked when
2439 * the data was copied into the page cache.
2440 */
2441 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2442 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2443 EXT4_GET_BLOCKS_IO_SUBMIT;
2444 dioread_nolock = ext4_should_dioread_nolock(inode);
2445 if (dioread_nolock)
2446 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2447 if (map->m_flags & (1 << BH_Delay))
2448 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2450 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2451 if (err < 0)
2452 return err;
2453 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2454 if (!mpd->io_submit.io_end->handle &&
2455 ext4_handle_valid(handle)) {
2456 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2457 handle->h_rsv_handle = NULL;
2458 }
2459 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2460 }
2462 BUG_ON(map->m_len == 0);
2463 if (map->m_flags & EXT4_MAP_NEW) {
2464 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
2465 map->m_len);
2466 }
2467 return 0;
2468 }
2470 /*
2471 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2472 * mpd->len and submit pages underlying it for IO
2473 *
2474 * @handle - handle for journal operations
2475 * @mpd - extent to map
2476 * @give_up_on_write - we set this to true iff there is a fatal error and there
2477 * is no hope of writing the data. The caller should discard
2478 * dirty pages to avoid infinite loops.
2479 *
2480 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2481 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2482 * them to initialized or split the described range from larger unwritten
2483 * extent. Note that we need not map all the described range since allocation
2484 * can return less blocks or the range is covered by more unwritten extents. We
2485 * cannot map more because we are limited by reserved transaction credits. On
2486 * the other hand we always make sure that the last touched page is fully
2487 * mapped so that it can be written out (and thus forward progress is
2488 * guaranteed). After mapping we submit all mapped pages for IO.
2489 */
2490 static int mpage_map_and_submit_extent(handle_t *handle,
2491 struct mpage_da_data *mpd,
2492 bool *give_up_on_write)
2493 {
2494 struct inode *inode = mpd->inode;
2495 struct ext4_map_blocks *map = &mpd->map;
2496 int err;
2497 loff_t disksize;
2498 int progress = 0;
2500 mpd->io_submit.io_end->offset =
2501 ((loff_t)map->m_lblk) << inode->i_blkbits;
2502 do {
2503 err = mpage_map_one_extent(handle, mpd);
2504 if (err < 0) {
2505 struct super_block *sb = inode->i_sb;
2507 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2508 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2509 goto invalidate_dirty_pages;
2510 /*
2511 * Let the uper layers retry transient errors.
2512 * In the case of ENOSPC, if ext4_count_free_blocks()
2513 * is non-zero, a commit should free up blocks.
2514 */
2515 if ((err == -ENOMEM) ||
2516 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2517 if (progress)
2518 goto update_disksize;
2519 return err;
2520 }
2521 ext4_msg(sb, KERN_CRIT,
2522 "Delayed block allocation failed for "
2523 "inode %lu at logical offset %llu with"
2524 " max blocks %u with error %d",
2525 inode->i_ino,
2526 (unsigned long long)map->m_lblk,
2527 (unsigned)map->m_len, -err);
2528 ext4_msg(sb, KERN_CRIT,
2529 "This should not happen!! Data will "
2530 "be lost\n");
2531 if (err == -ENOSPC)
2532 ext4_print_free_blocks(inode);
2533 invalidate_dirty_pages:
2534 *give_up_on_write = true;
2535 return err;
2536 }
2537 progress = 1;
2538 /*
2539 * Update buffer state, submit mapped pages, and get us new
2540 * extent to map
2541 */
2542 err = mpage_map_and_submit_buffers(mpd);
2543 if (err < 0)
2544 goto update_disksize;
2545 } while (map->m_len);
2547 update_disksize:
2548 /*
2549 * Update on-disk size after IO is submitted. Races with
2550 * truncate are avoided by checking i_size under i_data_sem.
2551 */
2552 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2553 if (disksize > EXT4_I(inode)->i_disksize) {
2554 int err2;
2555 loff_t i_size;
2557 down_write(&EXT4_I(inode)->i_data_sem);
2558 i_size = i_size_read(inode);
2559 if (disksize > i_size)
2560 disksize = i_size;
2561 if (disksize > EXT4_I(inode)->i_disksize)
2562 EXT4_I(inode)->i_disksize = disksize;
2563 up_write(&EXT4_I(inode)->i_data_sem);
2564 err2 = ext4_mark_inode_dirty(handle, inode);
2565 if (err2)
2566 ext4_error(inode->i_sb,
2567 "Failed to mark inode %lu dirty",
2568 inode->i_ino);
2569 if (!err)
2570 err = err2;
2571 }
2572 return err;
2573 }
2575 /*
2576 * Calculate the total number of credits to reserve for one writepages
2577 * iteration. This is called from ext4_writepages(). We map an extent of
2578 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2579 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2580 * bpp - 1 blocks in bpp different extents.
2581 */
2582 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2583 {
2584 int bpp = ext4_journal_blocks_per_page(inode);
2586 return ext4_meta_trans_blocks(inode,
2587 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2588 }
2590 /*
2591 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2592 * and underlying extent to map
2593 *
2594 * @mpd - where to look for pages
2595 *
2596 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2597 * IO immediately. When we find a page which isn't mapped we start accumulating
2598 * extent of buffers underlying these pages that needs mapping (formed by
2599 * either delayed or unwritten buffers). We also lock the pages containing
2600 * these buffers. The extent found is returned in @mpd structure (starting at
2601 * mpd->lblk with length mpd->len blocks).
2602 *
2603 * Note that this function can attach bios to one io_end structure which are
2604 * neither logically nor physically contiguous. Although it may seem as an
2605 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2606 * case as we need to track IO to all buffers underlying a page in one io_end.
2607 */
2608 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2609 {
2610 struct address_space *mapping = mpd->inode->i_mapping;
2611 struct pagevec pvec;
2612 unsigned int nr_pages;
2613 long left = mpd->wbc->nr_to_write;
2614 pgoff_t index = mpd->first_page;
2615 pgoff_t end = mpd->last_page;
2616 int tag;
2617 int i, err = 0;
2618 int blkbits = mpd->inode->i_blkbits;
2619 ext4_lblk_t lblk;
2620 struct buffer_head *head;
2622 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2623 tag = PAGECACHE_TAG_TOWRITE;
2624 else
2625 tag = PAGECACHE_TAG_DIRTY;
2627 pagevec_init(&pvec);
2628 mpd->map.m_len = 0;
2629 mpd->next_page = index;
2630 while (index <= end) {
2631 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2632 tag);
2633 if (nr_pages == 0)
2634 goto out;
2636 for (i = 0; i < nr_pages; i++) {
2637 struct page *page = pvec.pages[i];
2639 /*
2640 * Accumulated enough dirty pages? This doesn't apply
2641 * to WB_SYNC_ALL mode. For integrity sync we have to
2642 * keep going because someone may be concurrently
2643 * dirtying pages, and we might have synced a lot of
2644 * newly appeared dirty pages, but have not synced all
2645 * of the old dirty pages.
2646 */
2647 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2648 goto out;
2650 /* If we can't merge this page, we are done. */
2651 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2652 goto out;
2654 lock_page(page);
2655 /*
2656 * If the page is no longer dirty, or its mapping no
2657 * longer corresponds to inode we are writing (which
2658 * means it has been truncated or invalidated), or the
2659 * page is already under writeback and we are not doing
2660 * a data integrity writeback, skip the page
2661 */
2662 if (!PageDirty(page) ||
2663 (PageWriteback(page) &&
2664 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2665 unlikely(page->mapping != mapping)) {
2666 unlock_page(page);
2667 continue;
2668 }
2670 wait_on_page_writeback(page);
2671 BUG_ON(PageWriteback(page));
2673 if (mpd->map.m_len == 0)
2674 mpd->first_page = page->index;
2675 mpd->next_page = page->index + 1;
2676 /* Add all dirty buffers to mpd */
2677 lblk = ((ext4_lblk_t)page->index) <<
2678 (PAGE_SHIFT - blkbits);
2679 head = page_buffers(page);
2680 err = mpage_process_page_bufs(mpd, head, head, lblk);
2681 if (err <= 0)
2682 goto out;
2683 err = 0;
2684 left--;
2685 }
2686 pagevec_release(&pvec);
2687 cond_resched();
2688 }
2689 return 0;
2690 out:
2691 pagevec_release(&pvec);
2692 return err;
2693 }
2695 static int ext4_writepages(struct address_space *mapping,
2696 struct writeback_control *wbc)
2697 {
2698 pgoff_t writeback_index = 0;
2699 long nr_to_write = wbc->nr_to_write;
2700 int range_whole = 0;
2701 int cycled = 1;
2702 handle_t *handle = NULL;
2703 struct mpage_da_data mpd;
2704 struct inode *inode = mapping->host;
2705 int needed_blocks, rsv_blocks = 0, ret = 0;
2706 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2707 bool done;
2708 struct blk_plug plug;
2709 bool give_up_on_write = false;
2711 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2712 return -EIO;
2714 percpu_down_read(&sbi->s_journal_flag_rwsem);
2715 trace_ext4_writepages(inode, wbc);
2717 /*
2718 * No pages to write? This is mainly a kludge to avoid starting
2719 * a transaction for special inodes like journal inode on last iput()
2720 * because that could violate lock ordering on umount
2721 */
2722 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2723 goto out_writepages;
2725 if (ext4_should_journal_data(inode)) {
2726 ret = generic_writepages(mapping, wbc);
2727 goto out_writepages;
2728 }
2730 /*
2731 * If the filesystem has aborted, it is read-only, so return
2732 * right away instead of dumping stack traces later on that
2733 * will obscure the real source of the problem. We test
2734 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2735 * the latter could be true if the filesystem is mounted
2736 * read-only, and in that case, ext4_writepages should
2737 * *never* be called, so if that ever happens, we would want
2738 * the stack trace.
2739 */
2740 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2741 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2742 ret = -EROFS;
2743 goto out_writepages;
2744 }
2746 if (ext4_should_dioread_nolock(inode)) {
2747 /*
2748 * We may need to convert up to one extent per block in
2749 * the page and we may dirty the inode.
2750 */
2751 rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits);
2752 }
2754 /*
2755 * If we have inline data and arrive here, it means that
2756 * we will soon create the block for the 1st page, so
2757 * we'd better clear the inline data here.
2758 */
2759 if (ext4_has_inline_data(inode)) {
2760 /* Just inode will be modified... */
2761 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2762 if (IS_ERR(handle)) {
2763 ret = PTR_ERR(handle);
2764 goto out_writepages;
2765 }
2766 BUG_ON(ext4_test_inode_state(inode,
2767 EXT4_STATE_MAY_INLINE_DATA));
2768 ext4_destroy_inline_data(handle, inode);
2769 ext4_journal_stop(handle);
2770 }
2772 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2773 range_whole = 1;
2775 if (wbc->range_cyclic) {
2776 writeback_index = mapping->writeback_index;
2777 if (writeback_index)
2778 cycled = 0;
2779 mpd.first_page = writeback_index;
2780 mpd.last_page = -1;
2781 } else {
2782 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2783 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2784 }
2786 mpd.inode = inode;
2787 mpd.wbc = wbc;
2788 ext4_io_submit_init(&mpd.io_submit, wbc);
2789 retry:
2790 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2791 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2792 done = false;
2793 blk_start_plug(&plug);
2795 /*
2796 * First writeback pages that don't need mapping - we can avoid
2797 * starting a transaction unnecessarily and also avoid being blocked
2798 * in the block layer on device congestion while having transaction
2799 * started.
2800 */
2801 mpd.do_map = 0;
2802 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2803 if (!mpd.io_submit.io_end) {
2804 ret = -ENOMEM;
2805 goto unplug;
2806 }
2807 ret = mpage_prepare_extent_to_map(&mpd);
2808 /* Submit prepared bio */
2809 ext4_io_submit(&mpd.io_submit);
2810 ext4_put_io_end_defer(mpd.io_submit.io_end);
2811 mpd.io_submit.io_end = NULL;
2812 /* Unlock pages we didn't use */
2813 mpage_release_unused_pages(&mpd, false);
2814 if (ret < 0)
2815 goto unplug;
2817 while (!done && mpd.first_page <= mpd.last_page) {
2818 /* For each extent of pages we use new io_end */
2819 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2820 if (!mpd.io_submit.io_end) {
2821 ret = -ENOMEM;
2822 break;
2823 }
2825 /*
2826 * We have two constraints: We find one extent to map and we
2827 * must always write out whole page (makes a difference when
2828 * blocksize < pagesize) so that we don't block on IO when we
2829 * try to write out the rest of the page. Journalled mode is
2830 * not supported by delalloc.
2831 */
2832 BUG_ON(ext4_should_journal_data(inode));
2833 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2835 /* start a new transaction */
2836 handle = ext4_journal_start_with_reserve(inode,
2837 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2838 if (IS_ERR(handle)) {
2839 ret = PTR_ERR(handle);
2840 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2841 "%ld pages, ino %lu; err %d", __func__,
2842 wbc->nr_to_write, inode->i_ino, ret);
2843 /* Release allocated io_end */
2844 ext4_put_io_end(mpd.io_submit.io_end);
2845 mpd.io_submit.io_end = NULL;
2846 break;
2847 }
2848 mpd.do_map = 1;
2850 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2851 ret = mpage_prepare_extent_to_map(&mpd);
2852 if (!ret) {
2853 if (mpd.map.m_len)
2854 ret = mpage_map_and_submit_extent(handle, &mpd,
2855 &give_up_on_write);
2856 else {
2857 /*
2858 * We scanned the whole range (or exhausted
2859 * nr_to_write), submitted what was mapped and
2860 * didn't find anything needing mapping. We are
2861 * done.
2862 */
2863 done = true;
2864 }
2865 }
2866 /*
2867 * Caution: If the handle is synchronous,
2868 * ext4_journal_stop() can wait for transaction commit
2869 * to finish which may depend on writeback of pages to
2870 * complete or on page lock to be released. In that
2871 * case, we have to wait until after after we have
2872 * submitted all the IO, released page locks we hold,
2873 * and dropped io_end reference (for extent conversion
2874 * to be able to complete) before stopping the handle.
2875 */
2876 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2877 ext4_journal_stop(handle);
2878 handle = NULL;
2879 mpd.do_map = 0;
2880 }
2881 /* Submit prepared bio */
2882 ext4_io_submit(&mpd.io_submit);
2883 /* Unlock pages we didn't use */
2884 mpage_release_unused_pages(&mpd, give_up_on_write);
2885 /*
2886 * Drop our io_end reference we got from init. We have
2887 * to be careful and use deferred io_end finishing if
2888 * we are still holding the transaction as we can
2889 * release the last reference to io_end which may end
2890 * up doing unwritten extent conversion.
2891 */
2892 if (handle) {
2893 ext4_put_io_end_defer(mpd.io_submit.io_end);
2894 ext4_journal_stop(handle);
2895 } else
2896 ext4_put_io_end(mpd.io_submit.io_end);
2897 mpd.io_submit.io_end = NULL;
2899 if (ret == -ENOSPC && sbi->s_journal) {
2900 /*
2901 * Commit the transaction which would
2902 * free blocks released in the transaction
2903 * and try again
2904 */
2905 jbd2_journal_force_commit_nested(sbi->s_journal);
2906 ret = 0;
2907 continue;
2908 }
2909 /* Fatal error - ENOMEM, EIO... */
2910 if (ret)
2911 break;
2912 }
2913 unplug:
2914 blk_finish_plug(&plug);
2915 if (!ret && !cycled && wbc->nr_to_write > 0) {
2916 cycled = 1;
2917 mpd.last_page = writeback_index - 1;
2918 mpd.first_page = 0;
2919 goto retry;
2920 }
2922 /* Update index */
2923 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2924 /*
2925 * Set the writeback_index so that range_cyclic
2926 * mode will write it back later
2927 */
2928 mapping->writeback_index = mpd.first_page;
2930 out_writepages:
2931 trace_ext4_writepages_result(inode, wbc, ret,
2932 nr_to_write - wbc->nr_to_write);
2933 percpu_up_read(&sbi->s_journal_flag_rwsem);
2934 return ret;
2935 }
2937 static int ext4_dax_writepages(struct address_space *mapping,
2938 struct writeback_control *wbc)
2939 {
2940 int ret;
2941 long nr_to_write = wbc->nr_to_write;
2942 struct inode *inode = mapping->host;
2943 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2945 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2946 return -EIO;
2948 percpu_down_read(&sbi->s_journal_flag_rwsem);
2949 trace_ext4_writepages(inode, wbc);
2951 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, wbc);
2952 trace_ext4_writepages_result(inode, wbc, ret,
2953 nr_to_write - wbc->nr_to_write);
2954 percpu_up_read(&sbi->s_journal_flag_rwsem);
2955 return ret;
2956 }
2958 static int ext4_nonda_switch(struct super_block *sb)
2959 {
2960 s64 free_clusters, dirty_clusters;
2961 struct ext4_sb_info *sbi = EXT4_SB(sb);
2963 /*
2964 * switch to non delalloc mode if we are running low
2965 * on free block. The free block accounting via percpu
2966 * counters can get slightly wrong with percpu_counter_batch getting
2967 * accumulated on each CPU without updating global counters
2968 * Delalloc need an accurate free block accounting. So switch
2969 * to non delalloc when we are near to error range.
2970 */
2971 free_clusters =
2972 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2973 dirty_clusters =
2974 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2975 /*
2976 * Start pushing delalloc when 1/2 of free blocks are dirty.
2977 */
2978 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2979 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2981 if (2 * free_clusters < 3 * dirty_clusters ||
2982 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2983 /*
2984 * free block count is less than 150% of dirty blocks
2985 * or free blocks is less than watermark
2986 */
2987 return 1;
2988 }
2989 return 0;
2990 }
2992 /* We always reserve for an inode update; the superblock could be there too */
2993 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2994 {
2995 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2996 return 1;
2998 if (pos + len <= 0x7fffffffULL)
2999 return 1;
3001 /* We might need to update the superblock to set LARGE_FILE */
3002 return 2;
3003 }
3005 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3006 loff_t pos, unsigned len, unsigned flags,
3007 struct page **pagep, void **fsdata)
3008 {
3009 int ret, retries = 0;
3010 struct page *page;
3011 pgoff_t index;
3012 struct inode *inode = mapping->host;
3013 handle_t *handle;
3015 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3016 return -EIO;
3018 index = pos >> PAGE_SHIFT;
3020 if (ext4_nonda_switch(inode->i_sb) ||
3021 S_ISLNK(inode->i_mode)) {
3022 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3023 return ext4_write_begin(file, mapping, pos,
3024 len, flags, pagep, fsdata);
3025 }
3026 *fsdata = (void *)0;
3027 trace_ext4_da_write_begin(inode, pos, len, flags);
3029 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3030 ret = ext4_da_write_inline_data_begin(mapping, inode,
3031 pos, len, flags,
3032 pagep, fsdata);
3033 if (ret < 0)
3034 return ret;
3035 if (ret == 1)
3036 return 0;
3037 }
3039 /*
3040 * grab_cache_page_write_begin() can take a long time if the
3041 * system is thrashing due to memory pressure, or if the page
3042 * is being written back. So grab it first before we start
3043 * the transaction handle. This also allows us to allocate
3044 * the page (if needed) without using GFP_NOFS.
3045 */
3046 retry_grab:
3047 page = grab_cache_page_write_begin(mapping, index, flags);
3048 if (!page)
3049 return -ENOMEM;
3050 unlock_page(page);
3052 /*
3053 * With delayed allocation, we don't log the i_disksize update
3054 * if there is delayed block allocation. But we still need
3055 * to journalling the i_disksize update if writes to the end
3056 * of file which has an already mapped buffer.
3057 */
3058 retry_journal:
3059 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3060 ext4_da_write_credits(inode, pos, len));
3061 if (IS_ERR(handle)) {
3062 put_page(page);
3063 return PTR_ERR(handle);
3064 }
3066 lock_page(page);
3067 if (page->mapping != mapping) {
3068 /* The page got truncated from under us */
3069 unlock_page(page);
3070 put_page(page);
3071 ext4_journal_stop(handle);
3072 goto retry_grab;
3073 }
3074 /* In case writeback began while the page was unlocked */
3075 wait_for_stable_page(page);
3077 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3078 ret = ext4_block_write_begin(page, pos, len,
3079 ext4_da_get_block_prep);
3080 #else
3081 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3082 #endif
3083 if (ret < 0) {
3084 unlock_page(page);
3085 ext4_journal_stop(handle);
3086 /*
3087 * block_write_begin may have instantiated a few blocks
3088 * outside i_size. Trim these off again. Don't need
3089 * i_size_read because we hold i_mutex.
3090 */
3091 if (pos + len > inode->i_size)
3092 ext4_truncate_failed_write(inode);
3094 if (ret == -ENOSPC &&
3095 ext4_should_retry_alloc(inode->i_sb, &retries))
3096 goto retry_journal;
3098 put_page(page);
3099 return ret;
3100 }
3102 *pagep = page;
3103 return ret;
3104 }
3106 /*
3107 * Check if we should update i_disksize
3108 * when write to the end of file but not require block allocation
3109 */
3110 static int ext4_da_should_update_i_disksize(struct page *page,
3111 unsigned long offset)
3112 {
3113 struct buffer_head *bh;
3114 struct inode *inode = page->mapping->host;
3115 unsigned int idx;
3116 int i;
3118 bh = page_buffers(page);
3119 idx = offset >> inode->i_blkbits;
3121 for (i = 0; i < idx; i++)
3122 bh = bh->b_this_page;
3124 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3125 return 0;
3126 return 1;
3127 }
3129 static int ext4_da_write_end(struct file *file,
3130 struct address_space *mapping,
3131 loff_t pos, unsigned len, unsigned copied,
3132 struct page *page, void *fsdata)
3133 {
3134 struct inode *inode = mapping->host;
3135 int ret = 0, ret2;
3136 handle_t *handle = ext4_journal_current_handle();
3137 loff_t new_i_size;
3138 unsigned long start, end;
3139 int write_mode = (int)(unsigned long)fsdata;
3141 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3142 return ext4_write_end(file, mapping, pos,
3143 len, copied, page, fsdata);
3145 trace_ext4_da_write_end(inode, pos, len, copied);
3146 start = pos & (PAGE_SIZE - 1);
3147 end = start + copied - 1;
3149 /*
3150 * generic_write_end() will run mark_inode_dirty() if i_size
3151 * changes. So let's piggyback the i_disksize mark_inode_dirty
3152 * into that.
3153 */
3154 new_i_size = pos + copied;
3155 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3156 if (ext4_has_inline_data(inode) ||
3157 ext4_da_should_update_i_disksize(page, end)) {
3158 ext4_update_i_disksize(inode, new_i_size);
3159 /* We need to mark inode dirty even if
3160 * new_i_size is less that inode->i_size
3161 * bu greater than i_disksize.(hint delalloc)
3162 */
3163 ext4_mark_inode_dirty(handle, inode);
3164 }
3165 }
3167 if (write_mode != CONVERT_INLINE_DATA &&
3168 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3169 ext4_has_inline_data(inode))
3170 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3171 page);
3172 else
3173 ret2 = generic_write_end(file, mapping, pos, len, copied,
3174 page, fsdata);
3176 copied = ret2;
3177 if (ret2 < 0)
3178 ret = ret2;
3179 ret2 = ext4_journal_stop(handle);
3180 if (!ret)
3181 ret = ret2;
3183 return ret ? ret : copied;
3184 }
3186 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3187 unsigned int length)
3188 {
3189 /*
3190 * Drop reserved blocks
3191 */
3192 BUG_ON(!PageLocked(page));
3193 if (!page_has_buffers(page))
3194 goto out;
3196 ext4_da_page_release_reservation(page, offset, length);
3198 out:
3199 ext4_invalidatepage(page, offset, length);
3201 return;
3202 }
3204 /*
3205 * Force all delayed allocation blocks to be allocated for a given inode.
3206 */
3207 int ext4_alloc_da_blocks(struct inode *inode)
3208 {
3209 trace_ext4_alloc_da_blocks(inode);
3211 if (!EXT4_I(inode)->i_reserved_data_blocks)
3212 return 0;
3214 /*
3215 * We do something simple for now. The filemap_flush() will
3216 * also start triggering a write of the data blocks, which is
3217 * not strictly speaking necessary (and for users of
3218 * laptop_mode, not even desirable). However, to do otherwise
3219 * would require replicating code paths in:
3220 *
3221 * ext4_writepages() ->
3222 * write_cache_pages() ---> (via passed in callback function)
3223 * __mpage_da_writepage() -->
3224 * mpage_add_bh_to_extent()
3225 * mpage_da_map_blocks()
3226 *
3227 * The problem is that write_cache_pages(), located in
3228 * mm/page-writeback.c, marks pages clean in preparation for
3229 * doing I/O, which is not desirable if we're not planning on
3230 * doing I/O at all.
3231 *
3232 * We could call write_cache_pages(), and then redirty all of
3233 * the pages by calling redirty_page_for_writepage() but that
3234 * would be ugly in the extreme. So instead we would need to
3235 * replicate parts of the code in the above functions,
3236 * simplifying them because we wouldn't actually intend to
3237 * write out the pages, but rather only collect contiguous
3238 * logical block extents, call the multi-block allocator, and
3239 * then update the buffer heads with the block allocations.
3240 *
3241 * For now, though, we'll cheat by calling filemap_flush(),
3242 * which will map the blocks, and start the I/O, but not
3243 * actually wait for the I/O to complete.
3244 */
3245 return filemap_flush(inode->i_mapping);
3246 }
3248 /*
3249 * bmap() is special. It gets used by applications such as lilo and by
3250 * the swapper to find the on-disk block of a specific piece of data.
3251 *
3252 * Naturally, this is dangerous if the block concerned is still in the
3253 * journal. If somebody makes a swapfile on an ext4 data-journaling
3254 * filesystem and enables swap, then they may get a nasty shock when the
3255 * data getting swapped to that swapfile suddenly gets overwritten by
3256 * the original zero's written out previously to the journal and
3257 * awaiting writeback in the kernel's buffer cache.
3258 *
3259 * So, if we see any bmap calls here on a modified, data-journaled file,
3260 * take extra steps to flush any blocks which might be in the cache.
3261 */
3262 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3263 {
3264 struct inode *inode = mapping->host;
3265 journal_t *journal;
3266 int err;
3268 /*
3269 * We can get here for an inline file via the FIBMAP ioctl
3270 */
3271 if (ext4_has_inline_data(inode))
3272 return 0;
3274 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3275 test_opt(inode->i_sb, DELALLOC)) {
3276 /*
3277 * With delalloc we want to sync the file
3278 * so that we can make sure we allocate
3279 * blocks for file
3280 */
3281 filemap_write_and_wait(mapping);
3282 }
3284 if (EXT4_JOURNAL(inode) &&
3285 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3286 /*
3287 * This is a REALLY heavyweight approach, but the use of
3288 * bmap on dirty files is expected to be extremely rare:
3289 * only if we run lilo or swapon on a freshly made file
3290 * do we expect this to happen.
3291 *
3292 * (bmap requires CAP_SYS_RAWIO so this does not
3293 * represent an unprivileged user DOS attack --- we'd be
3294 * in trouble if mortal users could trigger this path at
3295 * will.)
3296 *
3297 * NB. EXT4_STATE_JDATA is not set on files other than
3298 * regular files. If somebody wants to bmap a directory
3299 * or symlink and gets confused because the buffer
3300 * hasn't yet been flushed to disk, they deserve
3301 * everything they get.
3302 */
3304 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3305 journal = EXT4_JOURNAL(inode);
3306 jbd2_journal_lock_updates(journal);
3307 err = jbd2_journal_flush(journal);
3308 jbd2_journal_unlock_updates(journal);
3310 if (err)
3311 return 0;
3312 }
3314 return generic_block_bmap(mapping, block, ext4_get_block);
3315 }
3317 static int ext4_readpage(struct file *file, struct page *page)
3318 {
3319 int ret = -EAGAIN;
3320 struct inode *inode = page->mapping->host;
3322 trace_ext4_readpage(page);
3324 if (ext4_has_inline_data(inode))
3325 ret = ext4_readpage_inline(inode, page);
3327 if (ret == -EAGAIN)
3328 return ext4_mpage_readpages(page->mapping, NULL, page, 1,
3329 false);
3331 return ret;
3332 }
3334 static int
3335 ext4_readpages(struct file *file, struct address_space *mapping,
3336 struct list_head *pages, unsigned nr_pages)
3337 {
3338 struct inode *inode = mapping->host;
3340 /* If the file has inline data, no need to do readpages. */
3341 if (ext4_has_inline_data(inode))
3342 return 0;
3344 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages, true);
3345 }
3347 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3348 unsigned int length)
3349 {
3350 trace_ext4_invalidatepage(page, offset, length);
3352 /* No journalling happens on data buffers when this function is used */
3353 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3355 block_invalidatepage(page, offset, length);
3356 }
3358 static int __ext4_journalled_invalidatepage(struct page *page,
3359 unsigned int offset,
3360 unsigned int length)
3361 {
3362 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3364 trace_ext4_journalled_invalidatepage(page, offset, length);
3366 /*
3367 * If it's a full truncate we just forget about the pending dirtying
3368 */
3369 if (offset == 0 && length == PAGE_SIZE)
3370 ClearPageChecked(page);
3372 return jbd2_journal_invalidatepage(journal, page, offset, length);
3373 }
3375 /* Wrapper for aops... */
3376 static void ext4_journalled_invalidatepage(struct page *page,
3377 unsigned int offset,
3378 unsigned int length)
3379 {
3380 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3381 }
3383 static int ext4_releasepage(struct page *page, gfp_t wait)
3384 {
3385 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3387 trace_ext4_releasepage(page);
3389 /* Page has dirty journalled data -> cannot release */
3390 if (PageChecked(page))
3391 return 0;
3392 if (journal)
3393 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3394 else
3395 return try_to_free_buffers(page);
3396 }
3398 static bool ext4_inode_datasync_dirty(struct inode *inode)
3399 {
3400 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3402 if (journal)
3403 return !jbd2_transaction_committed(journal,
3404 EXT4_I(inode)->i_datasync_tid);
3405 /* Any metadata buffers to write? */
3406 if (!list_empty(&inode->i_mapping->private_list))
3407 return true;
3408 return inode->i_state & I_DIRTY_DATASYNC;
3409 }
3411 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3412 unsigned flags, struct iomap *iomap)
3413 {
3414 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3415 unsigned int blkbits = inode->i_blkbits;
3416 unsigned long first_block, last_block;
3417 struct ext4_map_blocks map;
3418 bool delalloc = false;
3419 int ret;
3421 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3422 return -EINVAL;
3423 first_block = offset >> blkbits;
3424 last_block = min_t(loff_t, (offset + length - 1) >> blkbits,
3425 EXT4_MAX_LOGICAL_BLOCK);
3427 if (flags & IOMAP_REPORT) {
3428 if (ext4_has_inline_data(inode)) {
3429 ret = ext4_inline_data_iomap(inode, iomap);
3430 if (ret != -EAGAIN) {
3431 if (ret == 0 && offset >= iomap->length)
3432 ret = -ENOENT;
3433 return ret;
3434 }
3435 }
3436 } else {
3437 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3438 return -ERANGE;
3439 }
3441 map.m_lblk = first_block;
3442 map.m_len = last_block - first_block + 1;
3444 if (flags & IOMAP_REPORT) {
3445 ret = ext4_map_blocks(NULL, inode, &map, 0);
3446 if (ret < 0)
3447 return ret;
3449 if (ret == 0) {
3450 ext4_lblk_t end = map.m_lblk + map.m_len - 1;
3451 struct extent_status es;
3453 ext4_es_find_delayed_extent_range(inode, map.m_lblk, end, &es);
3455 if (!es.es_len || es.es_lblk > end) {
3456 /* entire range is a hole */
3457 } else if (es.es_lblk > map.m_lblk) {
3458 /* range starts with a hole */
3459 map.m_len = es.es_lblk - map.m_lblk;
3460 } else {
3461 ext4_lblk_t offs = 0;
3463 if (es.es_lblk < map.m_lblk)
3464 offs = map.m_lblk - es.es_lblk;
3465 map.m_lblk = es.es_lblk + offs;
3466 map.m_len = es.es_len - offs;
3467 delalloc = true;
3468 }
3469 }
3470 } else if (flags & IOMAP_WRITE) {
3471 int dio_credits;
3472 handle_t *handle;
3473 int retries = 0;
3475 /* Trim mapping request to maximum we can map at once for DIO */
3476 if (map.m_len > DIO_MAX_BLOCKS)
3477 map.m_len = DIO_MAX_BLOCKS;
3478 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3479 retry:
3480 /*
3481 * Either we allocate blocks and then we don't get unwritten
3482 * extent so we have reserved enough credits, or the blocks
3483 * are already allocated and unwritten and in that case
3484 * extent conversion fits in the credits as well.
3485 */
3486 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3487 dio_credits);
3488 if (IS_ERR(handle))
3489 return PTR_ERR(handle);
3491 ret = ext4_map_blocks(handle, inode, &map,
3492 EXT4_GET_BLOCKS_CREATE_ZERO);
3493 if (ret < 0) {
3494 ext4_journal_stop(handle);
3495 if (ret == -ENOSPC &&
3496 ext4_should_retry_alloc(inode->i_sb, &retries))
3497 goto retry;
3498 return ret;
3499 }
3501 /*
3502 * If we added blocks beyond i_size, we need to make sure they
3503 * will get truncated if we crash before updating i_size in
3504 * ext4_iomap_end(). For faults we don't need to do that (and
3505 * even cannot because for orphan list operations inode_lock is
3506 * required) - if we happen to instantiate block beyond i_size,
3507 * it is because we race with truncate which has already added
3508 * the inode to the orphan list.
3509 */
3510 if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3511 (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3512 int err;
3514 err = ext4_orphan_add(handle, inode);
3515 if (err < 0) {
3516 ext4_journal_stop(handle);
3517 return err;
3518 }
3519 }
3520 ext4_journal_stop(handle);
3521 } else {
3522 ret = ext4_map_blocks(NULL, inode, &map, 0);
3523 if (ret < 0)
3524 return ret;
3525 }
3527 iomap->flags = 0;
3528 if (ext4_inode_datasync_dirty(inode))
3529 iomap->flags |= IOMAP_F_DIRTY;
3530 iomap->bdev = inode->i_sb->s_bdev;
3531 iomap->dax_dev = sbi->s_daxdev;
3532 iomap->offset = (u64)first_block << blkbits;
3533 iomap->length = (u64)map.m_len << blkbits;
3535 if (ret == 0) {
3536 iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE;
3537 iomap->addr = IOMAP_NULL_ADDR;
3538 } else {
3539 if (map.m_flags & EXT4_MAP_MAPPED) {
3540 iomap->type = IOMAP_MAPPED;
3541 } else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3542 iomap->type = IOMAP_UNWRITTEN;
3543 } else {
3544 WARN_ON_ONCE(1);
3545 return -EIO;
3546 }
3547 iomap->addr = (u64)map.m_pblk << blkbits;
3548 }
3550 if (map.m_flags & EXT4_MAP_NEW)
3551 iomap->flags |= IOMAP_F_NEW;
3553 return 0;
3554 }
3556 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3557 ssize_t written, unsigned flags, struct iomap *iomap)
3558 {
3559 int ret = 0;
3560 handle_t *handle;
3561 int blkbits = inode->i_blkbits;
3562 bool truncate = false;
3564 if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3565 return 0;
3567 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3568 if (IS_ERR(handle)) {
3569 ret = PTR_ERR(handle);
3570 goto orphan_del;
3571 }
3572 if (ext4_update_inode_size(inode, offset + written))
3573 ext4_mark_inode_dirty(handle, inode);
3574 /*
3575 * We may need to truncate allocated but not written blocks beyond EOF.
3576 */
3577 if (iomap->offset + iomap->length >
3578 ALIGN(inode->i_size, 1 << blkbits)) {
3579 ext4_lblk_t written_blk, end_blk;
3581 written_blk = (offset + written) >> blkbits;
3582 end_blk = (offset + length) >> blkbits;
3583 if (written_blk < end_blk && ext4_can_truncate(inode))
3584 truncate = true;
3585 }
3586 /*
3587 * Remove inode from orphan list if we were extending a inode and
3588 * everything went fine.
3589 */
3590 if (!truncate && inode->i_nlink &&
3591 !list_empty(&EXT4_I(inode)->i_orphan))
3592 ext4_orphan_del(handle, inode);
3593 ext4_journal_stop(handle);
3594 if (truncate) {
3595 ext4_truncate_failed_write(inode);
3596 orphan_del:
3597 /*
3598 * If truncate failed early the inode might still be on the
3599 * orphan list; we need to make sure the inode is removed from
3600 * the orphan list in that case.
3601 */
3602 if (inode->i_nlink)
3603 ext4_orphan_del(NULL, inode);
3604 }
3605 return ret;
3606 }
3608 const struct iomap_ops ext4_iomap_ops = {
3609 .iomap_begin = ext4_iomap_begin,
3610 .iomap_end = ext4_iomap_end,
3611 };
3613 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3614 ssize_t size, void *private)
3615 {
3616 ext4_io_end_t *io_end = private;
3618 /* if not async direct IO just return */
3619 if (!io_end)
3620 return 0;
3622 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3623 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3624 io_end, io_end->inode->i_ino, iocb, offset, size);
3626 /*
3627 * Error during AIO DIO. We cannot convert unwritten extents as the
3628 * data was not written. Just clear the unwritten flag and drop io_end.
3629 */
3630 if (size <= 0) {
3631 ext4_clear_io_unwritten_flag(io_end);
3632 size = 0;
3633 }
3634 io_end->offset = offset;
3635 io_end->size = size;
3636 ext4_put_io_end(io_end);
3638 return 0;
3639 }
3641 /*
3642 * Handling of direct IO writes.
3643 *
3644 * For ext4 extent files, ext4 will do direct-io write even to holes,
3645 * preallocated extents, and those write extend the file, no need to
3646 * fall back to buffered IO.
3647 *
3648 * For holes, we fallocate those blocks, mark them as unwritten
3649 * If those blocks were preallocated, we mark sure they are split, but
3650 * still keep the range to write as unwritten.
3651 *
3652 * The unwritten extents will be converted to written when DIO is completed.
3653 * For async direct IO, since the IO may still pending when return, we
3654 * set up an end_io call back function, which will do the conversion
3655 * when async direct IO completed.
3656 *
3657 * If the O_DIRECT write will extend the file then add this inode to the
3658 * orphan list. So recovery will truncate it back to the original size
3659 * if the machine crashes during the write.
3660 *
3661 */
3662 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3663 {
3664 struct file *file = iocb->ki_filp;
3665 struct inode *inode = file->f_mapping->host;
3666 struct ext4_inode_info *ei = EXT4_I(inode);
3667 ssize_t ret;
3668 loff_t offset = iocb->ki_pos;
3669 size_t count = iov_iter_count(iter);
3670 int overwrite = 0;
3671 get_block_t *get_block_func = NULL;
3672 int dio_flags = 0;
3673 loff_t final_size = offset + count;
3674 int orphan = 0;
3675 handle_t *handle;
3677 if (final_size > inode->i_size || final_size > ei->i_disksize) {
3678 /* Credits for sb + inode write */
3679 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3680 if (IS_ERR(handle)) {
3681 ret = PTR_ERR(handle);
3682 goto out;
3683 }
3684 ret = ext4_orphan_add(handle, inode);
3685 if (ret) {
3686 ext4_journal_stop(handle);
3687 goto out;
3688 }
3689 orphan = 1;
3690 ext4_update_i_disksize(inode, inode->i_size);
3691 ext4_journal_stop(handle);
3692 }
3694 BUG_ON(iocb->private == NULL);
3696 /*
3697 * Make all waiters for direct IO properly wait also for extent
3698 * conversion. This also disallows race between truncate() and
3699 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3700 */
3701 inode_dio_begin(inode);
3703 /* If we do a overwrite dio, i_mutex locking can be released */
3704 overwrite = *((int *)iocb->private);
3706 if (overwrite)
3707 inode_unlock(inode);
3709 /*
3710 * For extent mapped files we could direct write to holes and fallocate.
3711 *
3712 * Allocated blocks to fill the hole are marked as unwritten to prevent
3713 * parallel buffered read to expose the stale data before DIO complete
3714 * the data IO.
3715 *
3716 * As to previously fallocated extents, ext4 get_block will just simply
3717 * mark the buffer mapped but still keep the extents unwritten.
3718 *
3719 * For non AIO case, we will convert those unwritten extents to written
3720 * after return back from blockdev_direct_IO. That way we save us from
3721 * allocating io_end structure and also the overhead of offloading
3722 * the extent convertion to a workqueue.
3723 *
3724 * For async DIO, the conversion needs to be deferred when the
3725 * IO is completed. The ext4 end_io callback function will be
3726 * called to take care of the conversion work. Here for async
3727 * case, we allocate an io_end structure to hook to the iocb.
3728 */
3729 iocb->private = NULL;
3730 if (overwrite)
3731 get_block_func = ext4_dio_get_block_overwrite;
3732 else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3733 round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3734 get_block_func = ext4_dio_get_block;
3735 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3736 } else if (is_sync_kiocb(iocb)) {
3737 get_block_func = ext4_dio_get_block_unwritten_sync;
3738 dio_flags = DIO_LOCKING;
3739 } else {
3740 get_block_func = ext4_dio_get_block_unwritten_async;
3741 dio_flags = DIO_LOCKING;
3742 }
3743 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3744 get_block_func, ext4_end_io_dio, NULL,
3745 dio_flags);
3747 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3748 EXT4_STATE_DIO_UNWRITTEN)) {
3749 int err;
3750 /*
3751 * for non AIO case, since the IO is already
3752 * completed, we could do the conversion right here
3753 */
3754 err = ext4_convert_unwritten_extents(NULL, inode,
3755 offset, ret);
3756 if (err < 0)
3757 ret = err;
3758 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3759 }
3761 inode_dio_end(inode);
3762 /* take i_mutex locking again if we do a ovewrite dio */
3763 if (overwrite)
3764 inode_lock(inode);
3766 if (ret < 0 && final_size > inode->i_size)
3767 ext4_truncate_failed_write(inode);
3769 /* Handle extending of i_size after direct IO write */
3770 if (orphan) {
3771 int err;
3773 /* Credits for sb + inode write */
3774 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3775 if (IS_ERR(handle)) {
3776 /*
3777 * We wrote the data but cannot extend
3778 * i_size. Bail out. In async io case, we do
3779 * not return error here because we have
3780 * already submmitted the corresponding
3781 * bio. Returning error here makes the caller
3782 * think that this IO is done and failed
3783 * resulting in race with bio's completion
3784 * handler.
3785 */
3786 if (!ret)
3787 ret = PTR_ERR(handle);
3788 if (inode->i_nlink)
3789 ext4_orphan_del(NULL, inode);
3791 goto out;
3792 }
3793 if (inode->i_nlink)
3794 ext4_orphan_del(handle, inode);
3795 if (ret > 0) {
3796 loff_t end = offset + ret;
3797 if (end > inode->i_size || end > ei->i_disksize) {
3798 ext4_update_i_disksize(inode, end);
3799 if (end > inode->i_size)
3800 i_size_write(inode, end);
3801 /*
3802 * We're going to return a positive `ret'
3803 * here due to non-zero-length I/O, so there's
3804 * no way of reporting error returns from
3805 * ext4_mark_inode_dirty() to userspace. So
3806 * ignore it.
3807 */
3808 ext4_mark_inode_dirty(handle, inode);
3809 }
3810 }
3811 err = ext4_journal_stop(handle);
3812 if (ret == 0)
3813 ret = err;
3814 }
3815 out:
3816 return ret;
3817 }
3819 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3820 {
3821 struct address_space *mapping = iocb->ki_filp->f_mapping;
3822 struct inode *inode = mapping->host;
3823 size_t count = iov_iter_count(iter);
3824 ssize_t ret;
3826 /*
3827 * Shared inode_lock is enough for us - it protects against concurrent
3828 * writes & truncates and since we take care of writing back page cache,
3829 * we are protected against page writeback as well.
3830 */
3831 inode_lock_shared(inode);
3832 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3833 iocb->ki_pos + count - 1);
3834 if (ret)
3835 goto out_unlock;
3836 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3837 iter, ext4_dio_get_block, NULL, NULL, 0);
3838 out_unlock:
3839 inode_unlock_shared(inode);
3840 return ret;
3841 }
3843 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3844 {
3845 struct file *file = iocb->ki_filp;
3846 struct inode *inode = file->f_mapping->host;
3847 size_t count = iov_iter_count(iter);
3848 loff_t offset = iocb->ki_pos;
3849 ssize_t ret;
3851 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3852 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3853 return 0;
3854 #endif
3856 /*
3857 * If we are doing data journalling we don't support O_DIRECT
3858 */
3859 if (ext4_should_journal_data(inode))
3860 return 0;
3862 /* Let buffer I/O handle the inline data case. */
3863 if (ext4_has_inline_data(inode))
3864 return 0;
3866 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3867 if (iov_iter_rw(iter) == READ)
3868 ret = ext4_direct_IO_read(iocb, iter);
3869 else
3870 ret = ext4_direct_IO_write(iocb, iter);
3871 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3872 return ret;
3873 }
3875 /*
3876 * Pages can be marked dirty completely asynchronously from ext4's journalling
3877 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3878 * much here because ->set_page_dirty is called under VFS locks. The page is
3879 * not necessarily locked.
3880 *
3881 * We cannot just dirty the page and leave attached buffers clean, because the
3882 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3883 * or jbddirty because all the journalling code will explode.
3884 *
3885 * So what we do is to mark the page "pending dirty" and next time writepage
3886 * is called, propagate that into the buffers appropriately.
3887 */
3888 static int ext4_journalled_set_page_dirty(struct page *page)
3889 {
3890 SetPageChecked(page);
3891 return __set_page_dirty_nobuffers(page);
3892 }
3894 static int ext4_set_page_dirty(struct page *page)
3895 {
3896 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3897 WARN_ON_ONCE(!page_has_buffers(page));
3898 return __set_page_dirty_buffers(page);
3899 }
3901 static const struct address_space_operations ext4_aops = {
3902 .readpage = ext4_readpage,
3903 .readpages = ext4_readpages,
3904 .writepage = ext4_writepage,
3905 .writepages = ext4_writepages,
3906 .write_begin = ext4_write_begin,
3907 .write_end = ext4_write_end,
3908 .set_page_dirty = ext4_set_page_dirty,
3909 .bmap = ext4_bmap,
3910 .invalidatepage = ext4_invalidatepage,
3911 .releasepage = ext4_releasepage,
3912 .direct_IO = ext4_direct_IO,
3913 .migratepage = buffer_migrate_page,
3914 .is_partially_uptodate = block_is_partially_uptodate,
3915 .error_remove_page = generic_error_remove_page,
3916 };
3918 static const struct address_space_operations ext4_journalled_aops = {
3919 .readpage = ext4_readpage,
3920 .readpages = ext4_readpages,
3921 .writepage = ext4_writepage,
3922 .writepages = ext4_writepages,
3923 .write_begin = ext4_write_begin,
3924 .write_end = ext4_journalled_write_end,
3925 .set_page_dirty = ext4_journalled_set_page_dirty,
3926 .bmap = ext4_bmap,
3927 .invalidatepage = ext4_journalled_invalidatepage,
3928 .releasepage = ext4_releasepage,
3929 .direct_IO = ext4_direct_IO,
3930 .is_partially_uptodate = block_is_partially_uptodate,
3931 .error_remove_page = generic_error_remove_page,
3932 };
3934 static const struct address_space_operations ext4_da_aops = {
3935 .readpage = ext4_readpage,
3936 .readpages = ext4_readpages,
3937 .writepage = ext4_writepage,
3938 .writepages = ext4_writepages,
3939 .write_begin = ext4_da_write_begin,
3940 .write_end = ext4_da_write_end,
3941 .set_page_dirty = ext4_set_page_dirty,
3942 .bmap = ext4_bmap,
3943 .invalidatepage = ext4_da_invalidatepage,
3944 .releasepage = ext4_releasepage,
3945 .direct_IO = ext4_direct_IO,
3946 .migratepage = buffer_migrate_page,
3947 .is_partially_uptodate = block_is_partially_uptodate,
3948 .error_remove_page = generic_error_remove_page,
3949 };
3951 static const struct address_space_operations ext4_dax_aops = {
3952 .writepages = ext4_dax_writepages,
3953 .direct_IO = noop_direct_IO,
3954 .set_page_dirty = noop_set_page_dirty,
3955 .bmap = ext4_bmap,
3956 .invalidatepage = noop_invalidatepage,
3957 };
3959 void ext4_set_aops(struct inode *inode)
3960 {
3961 switch (ext4_inode_journal_mode(inode)) {
3962 case EXT4_INODE_ORDERED_DATA_MODE:
3963 case EXT4_INODE_WRITEBACK_DATA_MODE:
3964 break;
3965 case EXT4_INODE_JOURNAL_DATA_MODE:
3966 inode->i_mapping->a_ops = &ext4_journalled_aops;
3967 return;
3968 default:
3969 BUG();
3970 }
3971 if (IS_DAX(inode))
3972 inode->i_mapping->a_ops = &ext4_dax_aops;
3973 else if (test_opt(inode->i_sb, DELALLOC))
3974 inode->i_mapping->a_ops = &ext4_da_aops;
3975 else
3976 inode->i_mapping->a_ops = &ext4_aops;
3977 }
3979 static int __ext4_block_zero_page_range(handle_t *handle,
3980 struct address_space *mapping, loff_t from, loff_t length)
3981 {
3982 ext4_fsblk_t index = from >> PAGE_SHIFT;
3983 unsigned offset = from & (PAGE_SIZE-1);
3984 unsigned blocksize, pos;
3985 ext4_lblk_t iblock;
3986 struct inode *inode = mapping->host;
3987 struct buffer_head *bh;
3988 struct page *page;
3989 int err = 0;
3991 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3992 mapping_gfp_constraint(mapping, ~__GFP_FS));
3993 if (!page)
3994 return -ENOMEM;
3996 blocksize = inode->i_sb->s_blocksize;
3998 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
4000 if (!page_has_buffers(page))
4001 create_empty_buffers(page, blocksize, 0);
4003 /* Find the buffer that contains "offset" */
4004 bh = page_buffers(page);
4005 pos = blocksize;
4006 while (offset >= pos) {
4007 bh = bh->b_this_page;
4008 iblock++;
4009 pos += blocksize;
4010 }
4011 if (buffer_freed(bh)) {
4012 BUFFER_TRACE(bh, "freed: skip");
4013 goto unlock;
4014 }
4015 if (!buffer_mapped(bh)) {
4016 BUFFER_TRACE(bh, "unmapped");
4017 ext4_get_block(inode, iblock, bh, 0);
4018 /* unmapped? It's a hole - nothing to do */
4019 if (!buffer_mapped(bh)) {
4020 BUFFER_TRACE(bh, "still unmapped");
4021 goto unlock;
4022 }
4023 }
4025 /* Ok, it's mapped. Make sure it's up-to-date */
4026 if (PageUptodate(page))
4027 set_buffer_uptodate(bh);
4029 if (!buffer_uptodate(bh)) {
4030 err = -EIO;
4031 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
4032 wait_on_buffer(bh);
4033 /* Uhhuh. Read error. Complain and punt. */
4034 if (!buffer_uptodate(bh))
4035 goto unlock;
4036 if (S_ISREG(inode->i_mode) &&
4037 ext4_encrypted_inode(inode)) {
4038 /* We expect the key to be set. */
4039 BUG_ON(!fscrypt_has_encryption_key(inode));
4040 BUG_ON(blocksize != PAGE_SIZE);
4041 WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host,
4042 page, PAGE_SIZE, 0, page->index));
4043 }
4044 }
4045 if (ext4_should_journal_data(inode)) {
4046 BUFFER_TRACE(bh, "get write access");
4047 err = ext4_journal_get_write_access(handle, bh);
4048 if (err)
4049 goto unlock;
4050 }
4051 zero_user(page, offset, length);
4052 BUFFER_TRACE(bh, "zeroed end of block");
4054 if (ext4_should_journal_data(inode)) {
4055 err = ext4_handle_dirty_metadata(handle, inode, bh);
4056 } else {
4057 err = 0;
4058 mark_buffer_dirty(bh);
4059 if (ext4_should_order_data(inode))
4060 err = ext4_jbd2_inode_add_write(handle, inode);
4061 }
4063 unlock:
4064 unlock_page(page);
4065 put_page(page);
4066 return err;
4067 }
4069 /*
4070 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4071 * starting from file offset 'from'. The range to be zero'd must
4072 * be contained with in one block. If the specified range exceeds
4073 * the end of the block it will be shortened to end of the block
4074 * that cooresponds to 'from'
4075 */
4076 static int ext4_block_zero_page_range(handle_t *handle,
4077 struct address_space *mapping, loff_t from, loff_t length)
4078 {
4079 struct inode *inode = mapping->host;
4080 unsigned offset = from & (PAGE_SIZE-1);
4081 unsigned blocksize = inode->i_sb->s_blocksize;
4082 unsigned max = blocksize - (offset & (blocksize - 1));
4084 /*
4085 * correct length if it does not fall between
4086 * 'from' and the end of the block
4087 */
4088 if (length > max || length < 0)
4089 length = max;
4091 if (IS_DAX(inode)) {
4092 return iomap_zero_range(inode, from, length, NULL,
4093 &ext4_iomap_ops);
4094 }
4095 return __ext4_block_zero_page_range(handle, mapping, from, length);
4096 }
4098 /*
4099 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4100 * up to the end of the block which corresponds to `from'.
4101 * This required during truncate. We need to physically zero the tail end
4102 * of that block so it doesn't yield old data if the file is later grown.
4103 */
4104 static int ext4_block_truncate_page(handle_t *handle,
4105 struct address_space *mapping, loff_t from)
4106 {
4107 unsigned offset = from & (PAGE_SIZE-1);
4108 unsigned length;
4109 unsigned blocksize;
4110 struct inode *inode = mapping->host;
4112 /* If we are processing an encrypted inode during orphan list handling */
4113 if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode))
4114 return 0;
4116 blocksize = inode->i_sb->s_blocksize;
4117 length = blocksize - (offset & (blocksize - 1));
4119 return ext4_block_zero_page_range(handle, mapping, from, length);
4120 }
4122 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4123 loff_t lstart, loff_t length)
4124 {
4125 struct super_block *sb = inode->i_sb;
4126 struct address_space *mapping = inode->i_mapping;
4127 unsigned partial_start, partial_end;
4128 ext4_fsblk_t start, end;
4129 loff_t byte_end = (lstart + length - 1);
4130 int err = 0;
4132 partial_start = lstart & (sb->s_blocksize - 1);
4133 partial_end = byte_end & (sb->s_blocksize - 1);
4135 start = lstart >> sb->s_blocksize_bits;
4136 end = byte_end >> sb->s_blocksize_bits;
4138 /* Handle partial zero within the single block */
4139 if (start == end &&
4140 (partial_start || (partial_end != sb->s_blocksize - 1))) {
4141 err = ext4_block_zero_page_range(handle, mapping,
4142 lstart, length);
4143 return err;
4144 }
4145 /* Handle partial zero out on the start of the range */
4146 if (partial_start) {
4147 err = ext4_block_zero_page_range(handle, mapping,
4148 lstart, sb->s_blocksize);
4149 if (err)
4150 return err;
4151 }
4152 /* Handle partial zero out on the end of the range */
4153 if (partial_end != sb->s_blocksize - 1)
4154 err = ext4_block_zero_page_range(handle, mapping,
4155 byte_end - partial_end,
4156 partial_end + 1);
4157 return err;
4158 }
4160 int ext4_can_truncate(struct inode *inode)
4161 {
4162 if (S_ISREG(inode->i_mode))
4163 return 1;
4164 if (S_ISDIR(inode->i_mode))
4165 return 1;
4166 if (S_ISLNK(inode->i_mode))
4167 return !ext4_inode_is_fast_symlink(inode);
4168 return 0;
4169 }
4171 /*
4172 * We have to make sure i_disksize gets properly updated before we truncate
4173 * page cache due to hole punching or zero range. Otherwise i_disksize update
4174 * can get lost as it may have been postponed to submission of writeback but
4175 * that will never happen after we truncate page cache.
4176 */
4177 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4178 loff_t len)
4179 {
4180 handle_t *handle;
4181 loff_t size = i_size_read(inode);
4183 WARN_ON(!inode_is_locked(inode));
4184 if (offset > size || offset + len < size)
4185 return 0;
4187 if (EXT4_I(inode)->i_disksize >= size)
4188 return 0;
4190 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4191 if (IS_ERR(handle))
4192 return PTR_ERR(handle);
4193 ext4_update_i_disksize(inode, size);
4194 ext4_mark_inode_dirty(handle, inode);
4195 ext4_journal_stop(handle);
4197 return 0;
4198 }
4200 static void ext4_wait_dax_page(struct ext4_inode_info *ei)
4201 {
4202 up_write(&ei->i_mmap_sem);
4203 schedule();
4204 down_write(&ei->i_mmap_sem);
4205 }
4207 int ext4_break_layouts(struct inode *inode)
4208 {
4209 struct ext4_inode_info *ei = EXT4_I(inode);
4210 struct page *page;
4211 int error;
4213 if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
4214 return -EINVAL;
4216 do {
4217 page = dax_layout_busy_page(inode->i_mapping);
4218 if (!page)
4219 return 0;
4221 error = ___wait_var_event(&page->_refcount,
4222 atomic_read(&page->_refcount) == 1,
4223 TASK_INTERRUPTIBLE, 0, 0,
4224 ext4_wait_dax_page(ei));
4225 } while (error == 0);
4227 return error;
4228 }
4230 /*
4231 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4232 * associated with the given offset and length
4233 *
4234 * @inode: File inode
4235 * @offset: The offset where the hole will begin
4236 * @len: The length of the hole
4237 *
4238 * Returns: 0 on success or negative on failure
4239 */
4241 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4242 {
4243 struct super_block *sb = inode->i_sb;
4244 ext4_lblk_t first_block, stop_block;
4245 struct address_space *mapping = inode->i_mapping;
4246 loff_t first_block_offset, last_block_offset;
4247 handle_t *handle;
4248 unsigned int credits;
4249 int ret = 0;
4251 if (!S_ISREG(inode->i_mode))
4252 return -EOPNOTSUPP;
4254 trace_ext4_punch_hole(inode, offset, length, 0);
4256 /*
4257 * Write out all dirty pages to avoid race conditions
4258 * Then release them.
4259 */
4260 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4261 ret = filemap_write_and_wait_range(mapping, offset,
4262 offset + length - 1);
4263 if (ret)
4264 return ret;
4265 }
4267 inode_lock(inode);
4269 /* No need to punch hole beyond i_size */
4270 if (offset >= inode->i_size)
4271 goto out_mutex;
4273 /*
4274 * If the hole extends beyond i_size, set the hole
4275 * to end after the page that contains i_size
4276 */
4277 if (offset + length > inode->i_size) {
4278 length = inode->i_size +
4279 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4280 offset;
4281 }
4283 if (offset & (sb->s_blocksize - 1) ||
4284 (offset + length) & (sb->s_blocksize - 1)) {
4285 /*
4286 * Attach jinode to inode for jbd2 if we do any zeroing of
4287 * partial block
4288 */
4289 ret = ext4_inode_attach_jinode(inode);
4290 if (ret < 0)
4291 goto out_mutex;
4293 }
4295 /* Wait all existing dio workers, newcomers will block on i_mutex */
4296 inode_dio_wait(inode);
4298 /*
4299 * Prevent page faults from reinstantiating pages we have released from
4300 * page cache.
4301 */
4302 down_write(&EXT4_I(inode)->i_mmap_sem);
4304 ret = ext4_break_layouts(inode);
4305 if (ret)
4306 goto out_dio;
4308 first_block_offset = round_up(offset, sb->s_blocksize);
4309 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4311 /* Now release the pages and zero block aligned part of pages*/
4312 if (last_block_offset > first_block_offset) {
4313 ret = ext4_update_disksize_before_punch(inode, offset, length);
4314 if (ret)
4315 goto out_dio;
4316 truncate_pagecache_range(inode, first_block_offset,
4317 last_block_offset);
4318 }
4320 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4321 credits = ext4_writepage_trans_blocks(inode);
4322 else
4323 credits = ext4_blocks_for_truncate(inode);
4324 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4325 if (IS_ERR(handle)) {
4326 ret = PTR_ERR(handle);
4327 ext4_std_error(sb, ret);
4328 goto out_dio;
4329 }
4331 ret = ext4_zero_partial_blocks(handle, inode, offset,
4332 length);
4333 if (ret)
4334 goto out_stop;
4336 first_block = (offset + sb->s_blocksize - 1) >>
4337 EXT4_BLOCK_SIZE_BITS(sb);
4338 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4340 /* If there are blocks to remove, do it */
4341 if (stop_block > first_block) {
4343 down_write(&EXT4_I(inode)->i_data_sem);
4344 ext4_discard_preallocations(inode);
4346 ret = ext4_es_remove_extent(inode, first_block,
4347 stop_block - first_block);
4348 if (ret) {
4349 up_write(&EXT4_I(inode)->i_data_sem);
4350 goto out_stop;
4351 }
4353 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4354 ret = ext4_ext_remove_space(inode, first_block,
4355 stop_block - 1);
4356 else
4357 ret = ext4_ind_remove_space(handle, inode, first_block,
4358 stop_block);
4360 up_write(&EXT4_I(inode)->i_data_sem);
4361 }
4362 if (IS_SYNC(inode))
4363 ext4_handle_sync(handle);
4365 inode->i_mtime = inode->i_ctime = current_time(inode);
4366 ext4_mark_inode_dirty(handle, inode);
4367 if (ret >= 0)
4368 ext4_update_inode_fsync_trans(handle, inode, 1);
4369 out_stop:
4370 ext4_journal_stop(handle);
4371 out_dio:
4372 up_write(&EXT4_I(inode)->i_mmap_sem);
4373 out_mutex:
4374 inode_unlock(inode);
4375 return ret;
4376 }
4378 int ext4_inode_attach_jinode(struct inode *inode)
4379 {
4380 struct ext4_inode_info *ei = EXT4_I(inode);
4381 struct jbd2_inode *jinode;
4383 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4384 return 0;
4386 jinode = jbd2_alloc_inode(GFP_KERNEL);
4387 spin_lock(&inode->i_lock);
4388 if (!ei->jinode) {
4389 if (!jinode) {
4390 spin_unlock(&inode->i_lock);
4391 return -ENOMEM;
4392 }
4393 ei->jinode = jinode;
4394 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4395 jinode = NULL;
4396 }
4397 spin_unlock(&inode->i_lock);
4398 if (unlikely(jinode != NULL))
4399 jbd2_free_inode(jinode);
4400 return 0;
4401 }
4403 /*
4404 * ext4_truncate()
4405 *
4406 * We block out ext4_get_block() block instantiations across the entire
4407 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4408 * simultaneously on behalf of the same inode.
4409 *
4410 * As we work through the truncate and commit bits of it to the journal there
4411 * is one core, guiding principle: the file's tree must always be consistent on
4412 * disk. We must be able to restart the truncate after a crash.
4413 *
4414 * The file's tree may be transiently inconsistent in memory (although it
4415 * probably isn't), but whenever we close off and commit a journal transaction,
4416 * the contents of (the filesystem + the journal) must be consistent and
4417 * restartable. It's pretty simple, really: bottom up, right to left (although
4418 * left-to-right works OK too).
4419 *
4420 * Note that at recovery time, journal replay occurs *before* the restart of
4421 * truncate against the orphan inode list.
4422 *
4423 * The committed inode has the new, desired i_size (which is the same as
4424 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4425 * that this inode's truncate did not complete and it will again call
4426 * ext4_truncate() to have another go. So there will be instantiated blocks
4427 * to the right of the truncation point in a crashed ext4 filesystem. But
4428 * that's fine - as long as they are linked from the inode, the post-crash
4429 * ext4_truncate() run will find them and release them.
4430 */
4431 int ext4_truncate(struct inode *inode)
4432 {
4433 struct ext4_inode_info *ei = EXT4_I(inode);
4434 unsigned int credits;
4435 int err = 0;
4436 handle_t *handle;
4437 struct address_space *mapping = inode->i_mapping;
4439 /*
4440 * There is a possibility that we're either freeing the inode
4441 * or it's a completely new inode. In those cases we might not
4442 * have i_mutex locked because it's not necessary.
4443 */
4444 if (!(inode->i_state & (I_NEW|I_FREEING)))
4445 WARN_ON(!inode_is_locked(inode));
4446 trace_ext4_truncate_enter(inode);
4448 if (!ext4_can_truncate(inode))
4449 return 0;
4451 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4453 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4454 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4456 if (ext4_has_inline_data(inode)) {
4457 int has_inline = 1;
4459 err = ext4_inline_data_truncate(inode, &has_inline);
4460 if (err)
4461 return err;
4462 if (has_inline)
4463 return 0;
4464 }
4466 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4467 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4468 if (ext4_inode_attach_jinode(inode) < 0)
4469 return 0;
4470 }
4472 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4473 credits = ext4_writepage_trans_blocks(inode);
4474 else
4475 credits = ext4_blocks_for_truncate(inode);
4477 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4478 if (IS_ERR(handle))
4479 return PTR_ERR(handle);
4481 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4482 ext4_block_truncate_page(handle, mapping, inode->i_size);
4484 /*
4485 * We add the inode to the orphan list, so that if this
4486 * truncate spans multiple transactions, and we crash, we will
4487 * resume the truncate when the filesystem recovers. It also
4488 * marks the inode dirty, to catch the new size.
4489 *
4490 * Implication: the file must always be in a sane, consistent
4491 * truncatable state while each transaction commits.
4492 */
4493 err = ext4_orphan_add(handle, inode);
4494 if (err)
4495 goto out_stop;
4497 down_write(&EXT4_I(inode)->i_data_sem);
4499 ext4_discard_preallocations(inode);
4501 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4502 err = ext4_ext_truncate(handle, inode);
4503 else
4504 ext4_ind_truncate(handle, inode);
4506 up_write(&ei->i_data_sem);
4507 if (err)
4508 goto out_stop;
4510 if (IS_SYNC(inode))
4511 ext4_handle_sync(handle);
4513 out_stop:
4514 /*
4515 * If this was a simple ftruncate() and the file will remain alive,
4516 * then we need to clear up the orphan record which we created above.
4517 * However, if this was a real unlink then we were called by
4518 * ext4_evict_inode(), and we allow that function to clean up the
4519 * orphan info for us.
4520 */
4521 if (inode->i_nlink)
4522 ext4_orphan_del(handle, inode);
4524 inode->i_mtime = inode->i_ctime = current_time(inode);
4525 ext4_mark_inode_dirty(handle, inode);
4526 ext4_journal_stop(handle);
4528 trace_ext4_truncate_exit(inode);
4529 return err;
4530 }
4532 /*
4533 * ext4_get_inode_loc returns with an extra refcount against the inode's
4534 * underlying buffer_head on success. If 'in_mem' is true, we have all
4535 * data in memory that is needed to recreate the on-disk version of this
4536 * inode.
4537 */
4538 static int __ext4_get_inode_loc(struct inode *inode,
4539 struct ext4_iloc *iloc, int in_mem)
4540 {
4541 struct ext4_group_desc *gdp;
4542 struct buffer_head *bh;
4543 struct super_block *sb = inode->i_sb;
4544 ext4_fsblk_t block;
4545 int inodes_per_block, inode_offset;
4547 iloc->bh = NULL;
4548 if (inode->i_ino < EXT4_ROOT_INO ||
4549 inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4550 return -EFSCORRUPTED;
4552 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4553 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4554 if (!gdp)
4555 return -EIO;
4557 /*
4558 * Figure out the offset within the block group inode table
4559 */
4560 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4561 inode_offset = ((inode->i_ino - 1) %
4562 EXT4_INODES_PER_GROUP(sb));
4563 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4564 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4566 bh = sb_getblk(sb, block);
4567 if (unlikely(!bh))
4568 return -ENOMEM;
4569 if (!buffer_uptodate(bh)) {
4570 lock_buffer(bh);
4572 /*
4573 * If the buffer has the write error flag, we have failed
4574 * to write out another inode in the same block. In this
4575 * case, we don't have to read the block because we may
4576 * read the old inode data successfully.
4577 */
4578 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4579 set_buffer_uptodate(bh);
4581 if (buffer_uptodate(bh)) {
4582 /* someone brought it uptodate while we waited */
4583 unlock_buffer(bh);
4584 goto has_buffer;
4585 }
4587 /*
4588 * If we have all information of the inode in memory and this
4589 * is the only valid inode in the block, we need not read the
4590 * block.
4591 */
4592 if (in_mem) {
4593 struct buffer_head *bitmap_bh;
4594 int i, start;
4596 start = inode_offset & ~(inodes_per_block - 1);
4598 /* Is the inode bitmap in cache? */
4599 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4600 if (unlikely(!bitmap_bh))
4601 goto make_io;
4603 /*
4604 * If the inode bitmap isn't in cache then the
4605 * optimisation may end up performing two reads instead
4606 * of one, so skip it.
4607 */
4608 if (!buffer_uptodate(bitmap_bh)) {
4609 brelse(bitmap_bh);
4610 goto make_io;
4611 }
4612 for (i = start; i < start + inodes_per_block; i++) {
4613 if (i == inode_offset)
4614 continue;
4615 if (ext4_test_bit(i, bitmap_bh->b_data))
4616 break;
4617 }
4618 brelse(bitmap_bh);
4619 if (i == start + inodes_per_block) {
4620 /* all other inodes are free, so skip I/O */
4621 memset(bh->b_data, 0, bh->b_size);
4622 set_buffer_uptodate(bh);
4623 unlock_buffer(bh);
4624 goto has_buffer;
4625 }
4626 }
4628 make_io:
4629 /*
4630 * If we need to do any I/O, try to pre-readahead extra
4631 * blocks from the inode table.
4632 */
4633 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4634 ext4_fsblk_t b, end, table;
4635 unsigned num;
4636 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4638 table = ext4_inode_table(sb, gdp);
4639 /* s_inode_readahead_blks is always a power of 2 */
4640 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4641 if (table > b)
4642 b = table;
4643 end = b + ra_blks;
4644 num = EXT4_INODES_PER_GROUP(sb);
4645 if (ext4_has_group_desc_csum(sb))
4646 num -= ext4_itable_unused_count(sb, gdp);
4647 table += num / inodes_per_block;
4648 if (end > table)
4649 end = table;
4650 while (b <= end)
4651 sb_breadahead(sb, b++);
4652 }
4654 /*
4655 * There are other valid inodes in the buffer, this inode
4656 * has in-inode xattrs, or we don't have this inode in memory.
4657 * Read the block from disk.
4658 */
4659 trace_ext4_load_inode(inode);
4660 get_bh(bh);
4661 bh->b_end_io = end_buffer_read_sync;
4662 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4663 wait_on_buffer(bh);
4664 if (!buffer_uptodate(bh)) {
4665 EXT4_ERROR_INODE_BLOCK(inode, block,
4666 "unable to read itable block");
4667 brelse(bh);
4668 return -EIO;
4669 }
4670 }
4671 has_buffer:
4672 iloc->bh = bh;
4673 return 0;
4674 }
4676 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4677 {
4678 /* We have all inode data except xattrs in memory here. */
4679 return __ext4_get_inode_loc(inode, iloc,
4680 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4681 }
4683 static bool ext4_should_use_dax(struct inode *inode)
4684 {
4685 if (!test_opt(inode->i_sb, DAX))
4686 return false;
4687 if (!S_ISREG(inode->i_mode))
4688 return false;
4689 if (ext4_should_journal_data(inode))
4690 return false;
4691 if (ext4_has_inline_data(inode))
4692 return false;
4693 if (ext4_encrypted_inode(inode))
4694 return false;
4695 return true;
4696 }
4698 void ext4_set_inode_flags(struct inode *inode)
4699 {
4700 unsigned int flags = EXT4_I(inode)->i_flags;
4701 unsigned int new_fl = 0;
4703 if (flags & EXT4_SYNC_FL)
4704 new_fl |= S_SYNC;
4705 if (flags & EXT4_APPEND_FL)
4706 new_fl |= S_APPEND;
4707 if (flags & EXT4_IMMUTABLE_FL)
4708 new_fl |= S_IMMUTABLE;
4709 if (flags & EXT4_NOATIME_FL)
4710 new_fl |= S_NOATIME;
4711 if (flags & EXT4_DIRSYNC_FL)
4712 new_fl |= S_DIRSYNC;
4713 if (ext4_should_use_dax(inode))
4714 new_fl |= S_DAX;
4715 if (flags & EXT4_ENCRYPT_FL)
4716 new_fl |= S_ENCRYPTED;
4717 inode_set_flags(inode, new_fl,
4718 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4719 S_ENCRYPTED);
4720 }
4722 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4723 struct ext4_inode_info *ei)
4724 {
4725 blkcnt_t i_blocks ;
4726 struct inode *inode = &(ei->vfs_inode);
4727 struct super_block *sb = inode->i_sb;
4729 if (ext4_has_feature_huge_file(sb)) {
4730 /* we are using combined 48 bit field */
4731 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4732 le32_to_cpu(raw_inode->i_blocks_lo);
4733 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4734 /* i_blocks represent file system block size */
4735 return i_blocks << (inode->i_blkbits - 9);
4736 } else {
4737 return i_blocks;
4738 }
4739 } else {
4740 return le32_to_cpu(raw_inode->i_blocks_lo);
4741 }
4742 }
4744 static inline int ext4_iget_extra_inode(struct inode *inode,
4745 struct ext4_inode *raw_inode,
4746 struct ext4_inode_info *ei)
4747 {
4748 __le32 *magic = (void *)raw_inode +
4749 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4751 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4752 EXT4_INODE_SIZE(inode->i_sb) &&
4753 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4754 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4755 return ext4_find_inline_data_nolock(inode);
4756 } else
4757 EXT4_I(inode)->i_inline_off = 0;
4758 return 0;
4759 }
4761 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4762 {
4763 if (!ext4_has_feature_project(inode->i_sb))
4764 return -EOPNOTSUPP;
4765 *projid = EXT4_I(inode)->i_projid;
4766 return 0;
4767 }
4769 /*
4770 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4771 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4772 * set.
4773 */
4774 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4775 {
4776 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4777 inode_set_iversion_raw(inode, val);
4778 else
4779 inode_set_iversion_queried(inode, val);
4780 }
4781 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4782 {
4783 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4784 return inode_peek_iversion_raw(inode);
4785 else
4786 return inode_peek_iversion(inode);
4787 }
4789 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4790 {
4791 struct ext4_iloc iloc;
4792 struct ext4_inode *raw_inode;
4793 struct ext4_inode_info *ei;
4794 struct inode *inode;
4795 journal_t *journal = EXT4_SB(sb)->s_journal;
4796 long ret;
4797 loff_t size;
4798 int block;
4799 uid_t i_uid;
4800 gid_t i_gid;
4801 projid_t i_projid;
4803 inode = iget_locked(sb, ino);
4804 if (!inode)
4805 return ERR_PTR(-ENOMEM);
4806 if (!(inode->i_state & I_NEW))
4807 return inode;
4809 ei = EXT4_I(inode);
4810 iloc.bh = NULL;
4812 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4813 if (ret < 0)
4814 goto bad_inode;
4815 raw_inode = ext4_raw_inode(&iloc);
4817 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4818 EXT4_ERROR_INODE(inode, "root inode unallocated");
4819 ret = -EFSCORRUPTED;
4820 goto bad_inode;
4821 }
4823 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4824 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4825 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4826 EXT4_INODE_SIZE(inode->i_sb) ||
4827 (ei->i_extra_isize & 3)) {
4828 EXT4_ERROR_INODE(inode,
4829 "bad extra_isize %u (inode size %u)",
4830 ei->i_extra_isize,
4831 EXT4_INODE_SIZE(inode->i_sb));
4832 ret = -EFSCORRUPTED;
4833 goto bad_inode;
4834 }
4835 } else
4836 ei->i_extra_isize = 0;
4838 /* Precompute checksum seed for inode metadata */
4839 if (ext4_has_metadata_csum(sb)) {
4840 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4841 __u32 csum;
4842 __le32 inum = cpu_to_le32(inode->i_ino);
4843 __le32 gen = raw_inode->i_generation;
4844 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4845 sizeof(inum));
4846 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4847 sizeof(gen));
4848 }
4850 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4851 EXT4_ERROR_INODE(inode, "checksum invalid");
4852 ret = -EFSBADCRC;
4853 goto bad_inode;
4854 }
4856 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4857 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4858 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4859 if (ext4_has_feature_project(sb) &&
4860 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4861 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4862 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4863 else
4864 i_projid = EXT4_DEF_PROJID;
4866 if (!(test_opt(inode->i_sb, NO_UID32))) {
4867 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4868 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4869 }
4870 i_uid_write(inode, i_uid);
4871 i_gid_write(inode, i_gid);
4872 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4873 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4875 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4876 ei->i_inline_off = 0;
4877 ei->i_dir_start_lookup = 0;
4878 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4879 /* We now have enough fields to check if the inode was active or not.
4880 * This is needed because nfsd might try to access dead inodes
4881 * the test is that same one that e2fsck uses
4882 * NeilBrown 1999oct15
4883 */
4884 if (inode->i_nlink == 0) {
4885 if ((inode->i_mode == 0 ||
4886 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4887 ino != EXT4_BOOT_LOADER_INO) {
4888 /* this inode is deleted */
4889 ret = -ESTALE;
4890 goto bad_inode;
4891 }
4892 /* The only unlinked inodes we let through here have
4893 * valid i_mode and are being read by the orphan
4894 * recovery code: that's fine, we're about to complete
4895 * the process of deleting those.
4896 * OR it is the EXT4_BOOT_LOADER_INO which is
4897 * not initialized on a new filesystem. */
4898 }
4899 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4900 ext4_set_inode_flags(inode);
4901 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4902 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4903 if (ext4_has_feature_64bit(sb))
4904 ei->i_file_acl |=
4905 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4906 inode->i_size = ext4_isize(sb, raw_inode);
4907 if ((size = i_size_read(inode)) < 0) {
4908 EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size);
4909 ret = -EFSCORRUPTED;
4910 goto bad_inode;
4911 }
4912 ei->i_disksize = inode->i_size;
4913 #ifdef CONFIG_QUOTA
4914 ei->i_reserved_quota = 0;
4915 #endif
4916 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4917 ei->i_block_group = iloc.block_group;
4918 ei->i_last_alloc_group = ~0;
4919 /*
4920 * NOTE! The in-memory inode i_data array is in little-endian order
4921 * even on big-endian machines: we do NOT byteswap the block numbers!
4922 */
4923 for (block = 0; block < EXT4_N_BLOCKS; block++)
4924 ei->i_data[block] = raw_inode->i_block[block];
4925 INIT_LIST_HEAD(&ei->i_orphan);
4927 /*
4928 * Set transaction id's of transactions that have to be committed
4929 * to finish f[data]sync. We set them to currently running transaction
4930 * as we cannot be sure that the inode or some of its metadata isn't
4931 * part of the transaction - the inode could have been reclaimed and
4932 * now it is reread from disk.
4933 */
4934 if (journal) {
4935 transaction_t *transaction;
4936 tid_t tid;
4938 read_lock(&journal->j_state_lock);
4939 if (journal->j_running_transaction)
4940 transaction = journal->j_running_transaction;
4941 else
4942 transaction = journal->j_committing_transaction;
4943 if (transaction)
4944 tid = transaction->t_tid;
4945 else
4946 tid = journal->j_commit_sequence;
4947 read_unlock(&journal->j_state_lock);
4948 ei->i_sync_tid = tid;
4949 ei->i_datasync_tid = tid;
4950 }
4952 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4953 if (ei->i_extra_isize == 0) {
4954 /* The extra space is currently unused. Use it. */
4955 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4956 ei->i_extra_isize = sizeof(struct ext4_inode) -
4957 EXT4_GOOD_OLD_INODE_SIZE;
4958 } else {
4959 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4960 if (ret)
4961 goto bad_inode;
4962 }
4963 }
4965 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4966 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4967 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4968 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4970 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4971 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4973 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4974 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4975 ivers |=
4976 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4977 }
4978 ext4_inode_set_iversion_queried(inode, ivers);
4979 }
4981 ret = 0;
4982 if (ei->i_file_acl &&
4983 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4984 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4985 ei->i_file_acl);
4986 ret = -EFSCORRUPTED;
4987 goto bad_inode;
4988 } else if (!ext4_has_inline_data(inode)) {
4989 /* validate the block references in the inode */
4990 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4991 (S_ISLNK(inode->i_mode) &&
4992 !ext4_inode_is_fast_symlink(inode))) {
4993 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4994 ret = ext4_ext_check_inode(inode);
4995 else
4996 ret = ext4_ind_check_inode(inode);
4997 }
4998 }
4999 if (ret)
5000 goto bad_inode;
5002 if (S_ISREG(inode->i_mode)) {
5003 inode->i_op = &ext4_file_inode_operations;
5004 inode->i_fop = &ext4_file_operations;
5005 ext4_set_aops(inode);
5006 } else if (S_ISDIR(inode->i_mode)) {
5007 inode->i_op = &ext4_dir_inode_operations;
5008 inode->i_fop = &ext4_dir_operations;
5009 } else if (S_ISLNK(inode->i_mode)) {
5010 /* VFS does not allow setting these so must be corruption */
5011 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5012 EXT4_ERROR_INODE(inode,
5013 "immutable or append flags not allowed on symlinks");
5014 ret = -EFSCORRUPTED;
5015 goto bad_inode;
5016 }
5017 if (ext4_encrypted_inode(inode)) {
5018 inode->i_op = &ext4_encrypted_symlink_inode_operations;
5019 ext4_set_aops(inode);
5020 } else if (ext4_inode_is_fast_symlink(inode)) {
5021 inode->i_link = (char *)ei->i_data;
5022 inode->i_op = &ext4_fast_symlink_inode_operations;
5023 nd_terminate_link(ei->i_data, inode->i_size,
5024 sizeof(ei->i_data) - 1);
5025 } else {
5026 inode->i_op = &ext4_symlink_inode_operations;
5027 ext4_set_aops(inode);
5028 }
5029 inode_nohighmem(inode);
5030 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5031 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5032 inode->i_op = &ext4_special_inode_operations;
5033 if (raw_inode->i_block[0])
5034 init_special_inode(inode, inode->i_mode,
5035 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5036 else
5037 init_special_inode(inode, inode->i_mode,
5038 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5039 } else if (ino == EXT4_BOOT_LOADER_INO) {
5040 make_bad_inode(inode);
5041 } else {
5042 ret = -EFSCORRUPTED;
5043 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
5044 goto bad_inode;
5045 }
5046 brelse(iloc.bh);
5048 unlock_new_inode(inode);
5049 return inode;
5051 bad_inode:
5052 brelse(iloc.bh);
5053 iget_failed(inode);
5054 return ERR_PTR(ret);
5055 }
5057 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
5058 {
5059 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
5060 return ERR_PTR(-EFSCORRUPTED);
5061 return ext4_iget(sb, ino);
5062 }
5064 static int ext4_inode_blocks_set(handle_t *handle,
5065 struct ext4_inode *raw_inode,
5066 struct ext4_inode_info *ei)
5067 {
5068 struct inode *inode = &(ei->vfs_inode);
5069 u64 i_blocks = inode->i_blocks;
5070 struct super_block *sb = inode->i_sb;
5072 if (i_blocks <= ~0U) {
5073 /*
5074 * i_blocks can be represented in a 32 bit variable
5075 * as multiple of 512 bytes
5076 */
5077 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5078 raw_inode->i_blocks_high = 0;
5079 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5080 return 0;
5081 }
5082 if (!ext4_has_feature_huge_file(sb))
5083 return -EFBIG;
5085 if (i_blocks <= 0xffffffffffffULL) {
5086 /*
5087 * i_blocks can be represented in a 48 bit variable
5088 * as multiple of 512 bytes
5089 */
5090 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5091 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5092 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5093 } else {
5094 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5095 /* i_block is stored in file system block size */
5096 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5097 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5098 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5099 }
5100 return 0;
5101 }
5103 struct other_inode {
5104 unsigned long orig_ino;
5105 struct ext4_inode *raw_inode;
5106 };
5108 static int other_inode_match(struct inode * inode, unsigned long ino,
5109 void *data)
5110 {
5111 struct other_inode *oi = (struct other_inode *) data;
5113 if ((inode->i_ino != ino) ||
5114 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5115 I_DIRTY_INODE)) ||
5116 ((inode->i_state & I_DIRTY_TIME) == 0))
5117 return 0;
5118 spin_lock(&inode->i_lock);
5119 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5120 I_DIRTY_INODE)) == 0) &&
5121 (inode->i_state & I_DIRTY_TIME)) {
5122 struct ext4_inode_info *ei = EXT4_I(inode);
5124 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
5125 spin_unlock(&inode->i_lock);
5127 spin_lock(&ei->i_raw_lock);
5128 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
5129 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
5130 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
5131 ext4_inode_csum_set(inode, oi->raw_inode, ei);
5132 spin_unlock(&ei->i_raw_lock);
5133 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
5134 return -1;
5135 }
5136 spin_unlock(&inode->i_lock);
5137 return -1;
5138 }
5140 /*
5141 * Opportunistically update the other time fields for other inodes in
5142 * the same inode table block.
5143 */
5144 static void ext4_update_other_inodes_time(struct super_block *sb,
5145 unsigned long orig_ino, char *buf)
5146 {
5147 struct other_inode oi;
5148 unsigned long ino;
5149 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5150 int inode_size = EXT4_INODE_SIZE(sb);
5152 oi.orig_ino = orig_ino;
5153 /*
5154 * Calculate the first inode in the inode table block. Inode
5155 * numbers are one-based. That is, the first inode in a block
5156 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5157 */
5158 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5159 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5160 if (ino == orig_ino)
5161 continue;
5162 oi.raw_inode = (struct ext4_inode *) buf;
5163 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
5164 }
5165 }
5167 /*
5168 * Post the struct inode info into an on-disk inode location in the
5169 * buffer-cache. This gobbles the caller's reference to the
5170 * buffer_head in the inode location struct.
5171 *
5172 * The caller must have write access to iloc->bh.
5173 */
5174 static int ext4_do_update_inode(handle_t *handle,
5175 struct inode *inode,
5176 struct ext4_iloc *iloc)
5177 {
5178 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5179 struct ext4_inode_info *ei = EXT4_I(inode);
5180 struct buffer_head *bh = iloc->bh;
5181 struct super_block *sb = inode->i_sb;
5182 int err = 0, rc, block;
5183 int need_datasync = 0, set_large_file = 0;
5184 uid_t i_uid;
5185 gid_t i_gid;
5186 projid_t i_projid;
5188 spin_lock(&ei->i_raw_lock);
5190 /* For fields not tracked in the in-memory inode,
5191 * initialise them to zero for new inodes. */
5192 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5193 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5195 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5196 i_uid = i_uid_read(inode);
5197 i_gid = i_gid_read(inode);
5198 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5199 if (!(test_opt(inode->i_sb, NO_UID32))) {
5200 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5201 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5202 /*
5203 * Fix up interoperability with old kernels. Otherwise, old inodes get
5204 * re-used with the upper 16 bits of the uid/gid intact
5205 */
5206 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5207 raw_inode->i_uid_high = 0;
5208 raw_inode->i_gid_high = 0;
5209 } else {
5210 raw_inode->i_uid_high =
5211 cpu_to_le16(high_16_bits(i_uid));
5212 raw_inode->i_gid_high =
5213 cpu_to_le16(high_16_bits(i_gid));
5214 }
5215 } else {
5216 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5217 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5218 raw_inode->i_uid_high = 0;
5219 raw_inode->i_gid_high = 0;
5220 }
5221 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5223 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5224 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5225 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5226 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5228 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5229 if (err) {
5230 spin_unlock(&ei->i_raw_lock);
5231 goto out_brelse;
5232 }
5233 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5234 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5235 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5236 raw_inode->i_file_acl_high =
5237 cpu_to_le16(ei->i_file_acl >> 32);
5238 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5239 if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
5240 ext4_isize_set(raw_inode, ei->i_disksize);
5241 need_datasync = 1;
5242 }
5243 if (ei->i_disksize > 0x7fffffffULL) {
5244 if (!ext4_has_feature_large_file(sb) ||
5245 EXT4_SB(sb)->s_es->s_rev_level ==
5246 cpu_to_le32(EXT4_GOOD_OLD_REV))
5247 set_large_file = 1;
5248 }
5249 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5250 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5251 if (old_valid_dev(inode->i_rdev)) {
5252 raw_inode->i_block[0] =
5253 cpu_to_le32(old_encode_dev(inode->i_rdev));
5254 raw_inode->i_block[1] = 0;
5255 } else {
5256 raw_inode->i_block[0] = 0;
5257 raw_inode->i_block[1] =
5258 cpu_to_le32(new_encode_dev(inode->i_rdev));
5259 raw_inode->i_block[2] = 0;
5260 }
5261 } else if (!ext4_has_inline_data(inode)) {
5262 for (block = 0; block < EXT4_N_BLOCKS; block++)
5263 raw_inode->i_block[block] = ei->i_data[block];
5264 }
5266 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5267 u64 ivers = ext4_inode_peek_iversion(inode);
5269 raw_inode->i_disk_version = cpu_to_le32(ivers);
5270 if (ei->i_extra_isize) {
5271 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5272 raw_inode->i_version_hi =
5273 cpu_to_le32(ivers >> 32);
5274 raw_inode->i_extra_isize =
5275 cpu_to_le16(ei->i_extra_isize);
5276 }
5277 }
5279 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5280 i_projid != EXT4_DEF_PROJID);
5282 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5283 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5284 raw_inode->i_projid = cpu_to_le32(i_projid);
5286 ext4_inode_csum_set(inode, raw_inode, ei);
5287 spin_unlock(&ei->i_raw_lock);
5288 if (inode->i_sb->s_flags & SB_LAZYTIME)
5289 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5290 bh->b_data);
5292 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5293 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5294 if (!err)
5295 err = rc;
5296 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5297 if (set_large_file) {
5298 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5299 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5300 if (err)
5301 goto out_brelse;
5302 ext4_update_dynamic_rev(sb);
5303 ext4_set_feature_large_file(sb);
5304 ext4_handle_sync(handle);
5305 err = ext4_handle_dirty_super(handle, sb);
5306 }
5307 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5308 out_brelse:
5309 brelse(bh);
5310 ext4_std_error(inode->i_sb, err);
5311 return err;
5312 }
5314 /*
5315 * ext4_write_inode()
5316 *
5317 * We are called from a few places:
5318 *
5319 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5320 * Here, there will be no transaction running. We wait for any running
5321 * transaction to commit.
5322 *
5323 * - Within flush work (sys_sync(), kupdate and such).
5324 * We wait on commit, if told to.
5325 *
5326 * - Within iput_final() -> write_inode_now()
5327 * We wait on commit, if told to.
5328 *
5329 * In all cases it is actually safe for us to return without doing anything,
5330 * because the inode has been copied into a raw inode buffer in
5331 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5332 * writeback.
5333 *
5334 * Note that we are absolutely dependent upon all inode dirtiers doing the
5335 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5336 * which we are interested.
5337 *
5338 * It would be a bug for them to not do this. The code:
5339 *
5340 * mark_inode_dirty(inode)
5341 * stuff();
5342 * inode->i_size = expr;
5343 *
5344 * is in error because write_inode() could occur while `stuff()' is running,
5345 * and the new i_size will be lost. Plus the inode will no longer be on the
5346 * superblock's dirty inode list.
5347 */
5348 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5349 {
5350 int err;
5352 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5353 return 0;
5355 if (EXT4_SB(inode->i_sb)->s_journal) {
5356 if (ext4_journal_current_handle()) {
5357 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5358 dump_stack();
5359 return -EIO;
5360 }
5362 /*
5363 * No need to force transaction in WB_SYNC_NONE mode. Also
5364 * ext4_sync_fs() will force the commit after everything is
5365 * written.
5366 */
5367 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5368 return 0;
5370 err = ext4_force_commit(inode->i_sb);
5371 } else {
5372 struct ext4_iloc iloc;
5374 err = __ext4_get_inode_loc(inode, &iloc, 0);
5375 if (err)
5376 return err;
5377 /*
5378 * sync(2) will flush the whole buffer cache. No need to do
5379 * it here separately for each inode.
5380 */
5381 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5382 sync_dirty_buffer(iloc.bh);
5383 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5384 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5385 "IO error syncing inode");
5386 err = -EIO;
5387 }
5388 brelse(iloc.bh);
5389 }
5390 return err;
5391 }
5393 /*
5394 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5395 * buffers that are attached to a page stradding i_size and are undergoing
5396 * commit. In that case we have to wait for commit to finish and try again.
5397 */
5398 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5399 {
5400 struct page *page;
5401 unsigned offset;
5402 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5403 tid_t commit_tid = 0;
5404 int ret;
5406 offset = inode->i_size & (PAGE_SIZE - 1);
5407 /*
5408 * All buffers in the last page remain valid? Then there's nothing to
5409 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5410 * blocksize case
5411 */
5412 if (offset > PAGE_SIZE - i_blocksize(inode))
5413 return;
5414 while (1) {
5415 page = find_lock_page(inode->i_mapping,
5416 inode->i_size >> PAGE_SHIFT);
5417 if (!page)
5418 return;
5419 ret = __ext4_journalled_invalidatepage(page, offset,
5420 PAGE_SIZE - offset);
5421 unlock_page(page);
5422 put_page(page);
5423 if (ret != -EBUSY)
5424 return;
5425 commit_tid = 0;
5426 read_lock(&journal->j_state_lock);
5427 if (journal->j_committing_transaction)
5428 commit_tid = journal->j_committing_transaction->t_tid;
5429 read_unlock(&journal->j_state_lock);
5430 if (commit_tid)
5431 jbd2_log_wait_commit(journal, commit_tid);
5432 }
5433 }
5435 /*
5436 * ext4_setattr()
5437 *
5438 * Called from notify_change.
5439 *
5440 * We want to trap VFS attempts to truncate the file as soon as
5441 * possible. In particular, we want to make sure that when the VFS
5442 * shrinks i_size, we put the inode on the orphan list and modify
5443 * i_disksize immediately, so that during the subsequent flushing of
5444 * dirty pages and freeing of disk blocks, we can guarantee that any
5445 * commit will leave the blocks being flushed in an unused state on
5446 * disk. (On recovery, the inode will get truncated and the blocks will
5447 * be freed, so we have a strong guarantee that no future commit will
5448 * leave these blocks visible to the user.)
5449 *
5450 * Another thing we have to assure is that if we are in ordered mode
5451 * and inode is still attached to the committing transaction, we must
5452 * we start writeout of all the dirty pages which are being truncated.
5453 * This way we are sure that all the data written in the previous
5454 * transaction are already on disk (truncate waits for pages under
5455 * writeback).
5456 *
5457 * Called with inode->i_mutex down.
5458 */
5459 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5460 {
5461 struct inode *inode = d_inode(dentry);
5462 int error, rc = 0;
5463 int orphan = 0;
5464 const unsigned int ia_valid = attr->ia_valid;
5466 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5467 return -EIO;
5469 error = setattr_prepare(dentry, attr);
5470 if (error)
5471 return error;
5473 error = fscrypt_prepare_setattr(dentry, attr);
5474 if (error)
5475 return error;
5477 if (is_quota_modification(inode, attr)) {
5478 error = dquot_initialize(inode);
5479 if (error)
5480 return error;
5481 }
5482 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5483 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5484 handle_t *handle;
5486 /* (user+group)*(old+new) structure, inode write (sb,
5487 * inode block, ? - but truncate inode update has it) */
5488 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5489 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5490 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5491 if (IS_ERR(handle)) {
5492 error = PTR_ERR(handle);
5493 goto err_out;
5494 }
5496 /* dquot_transfer() calls back ext4_get_inode_usage() which
5497 * counts xattr inode references.
5498 */
5499 down_read(&EXT4_I(inode)->xattr_sem);
5500 error = dquot_transfer(inode, attr);
5501 up_read(&EXT4_I(inode)->xattr_sem);
5503 if (error) {
5504 ext4_journal_stop(handle);
5505 return error;
5506 }
5507 /* Update corresponding info in inode so that everything is in
5508 * one transaction */
5509 if (attr->ia_valid & ATTR_UID)
5510 inode->i_uid = attr->ia_uid;
5511 if (attr->ia_valid & ATTR_GID)
5512 inode->i_gid = attr->ia_gid;
5513 error = ext4_mark_inode_dirty(handle, inode);
5514 ext4_journal_stop(handle);
5515 }
5517 if (attr->ia_valid & ATTR_SIZE) {
5518 handle_t *handle;
5519 loff_t oldsize = inode->i_size;
5520 int shrink = (attr->ia_size <= inode->i_size);
5522 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5523 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5525 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5526 return -EFBIG;
5527 }
5528 if (!S_ISREG(inode->i_mode))
5529 return -EINVAL;
5531 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5532 inode_inc_iversion(inode);
5534 if (ext4_should_order_data(inode) &&
5535 (attr->ia_size < inode->i_size)) {
5536 error = ext4_begin_ordered_truncate(inode,
5537 attr->ia_size);
5538 if (error)
5539 goto err_out;
5540 }
5541 if (attr->ia_size != inode->i_size) {
5542 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5543 if (IS_ERR(handle)) {
5544 error = PTR_ERR(handle);
5545 goto err_out;
5546 }
5547 if (ext4_handle_valid(handle) && shrink) {
5548 error = ext4_orphan_add(handle, inode);
5549 orphan = 1;
5550 }
5551 /*
5552 * Update c/mtime on truncate up, ext4_truncate() will
5553 * update c/mtime in shrink case below
5554 */
5555 if (!shrink) {
5556 inode->i_mtime = current_time(inode);
5557 inode->i_ctime = inode->i_mtime;
5558 }
5559 down_write(&EXT4_I(inode)->i_data_sem);
5560 EXT4_I(inode)->i_disksize = attr->ia_size;
5561 rc = ext4_mark_inode_dirty(handle, inode);
5562 if (!error)
5563 error = rc;
5564 /*
5565 * We have to update i_size under i_data_sem together
5566 * with i_disksize to avoid races with writeback code
5567 * running ext4_wb_update_i_disksize().
5568 */
5569 if (!error)
5570 i_size_write(inode, attr->ia_size);
5571 up_write(&EXT4_I(inode)->i_data_sem);
5572 ext4_journal_stop(handle);
5573 if (error) {
5574 if (orphan)
5575 ext4_orphan_del(NULL, inode);
5576 goto err_out;
5577 }
5578 }
5579 if (!shrink)
5580 pagecache_isize_extended(inode, oldsize, inode->i_size);
5582 /*
5583 * Blocks are going to be removed from the inode. Wait
5584 * for dio in flight. Temporarily disable
5585 * dioread_nolock to prevent livelock.
5586 */
5587 if (orphan) {
5588 if (!ext4_should_journal_data(inode)) {
5589 inode_dio_wait(inode);
5590 } else
5591 ext4_wait_for_tail_page_commit(inode);
5592 }
5593 down_write(&EXT4_I(inode)->i_mmap_sem);
5595 rc = ext4_break_layouts(inode);
5596 if (rc) {
5597 up_write(&EXT4_I(inode)->i_mmap_sem);
5598 error = rc;
5599 goto err_out;
5600 }
5602 /*
5603 * Truncate pagecache after we've waited for commit
5604 * in data=journal mode to make pages freeable.
5605 */
5606 truncate_pagecache(inode, inode->i_size);
5607 if (shrink) {
5608 rc = ext4_truncate(inode);
5609 if (rc)
5610 error = rc;
5611 }
5612 up_write(&EXT4_I(inode)->i_mmap_sem);
5613 }
5615 if (!error) {
5616 setattr_copy(inode, attr);
5617 mark_inode_dirty(inode);
5618 }
5620 /*
5621 * If the call to ext4_truncate failed to get a transaction handle at
5622 * all, we need to clean up the in-core orphan list manually.
5623 */
5624 if (orphan && inode->i_nlink)
5625 ext4_orphan_del(NULL, inode);
5627 if (!error && (ia_valid & ATTR_MODE))
5628 rc = posix_acl_chmod(inode, inode->i_mode);
5630 err_out:
5631 ext4_std_error(inode->i_sb, error);
5632 if (!error)
5633 error = rc;
5634 return error;
5635 }
5637 int ext4_getattr(const struct path *path, struct kstat *stat,
5638 u32 request_mask, unsigned int query_flags)
5639 {
5640 struct inode *inode = d_inode(path->dentry);
5641 struct ext4_inode *raw_inode;
5642 struct ext4_inode_info *ei = EXT4_I(inode);
5643 unsigned int flags;
5645 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5646 stat->result_mask |= STATX_BTIME;
5647 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5648 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5649 }
5651 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5652 if (flags & EXT4_APPEND_FL)
5653 stat->attributes |= STATX_ATTR_APPEND;
5654 if (flags & EXT4_COMPR_FL)
5655 stat->attributes |= STATX_ATTR_COMPRESSED;
5656 if (flags & EXT4_ENCRYPT_FL)
5657 stat->attributes |= STATX_ATTR_ENCRYPTED;
5658 if (flags & EXT4_IMMUTABLE_FL)
5659 stat->attributes |= STATX_ATTR_IMMUTABLE;
5660 if (flags & EXT4_NODUMP_FL)
5661 stat->attributes |= STATX_ATTR_NODUMP;
5663 stat->attributes_mask |= (STATX_ATTR_APPEND |
5664 STATX_ATTR_COMPRESSED |
5665 STATX_ATTR_ENCRYPTED |
5666 STATX_ATTR_IMMUTABLE |
5667 STATX_ATTR_NODUMP);
5669 generic_fillattr(inode, stat);
5670 return 0;
5671 }
5673 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5674 u32 request_mask, unsigned int query_flags)
5675 {
5676 struct inode *inode = d_inode(path->dentry);
5677 u64 delalloc_blocks;
5679 ext4_getattr(path, stat, request_mask, query_flags);
5681 /*
5682 * If there is inline data in the inode, the inode will normally not
5683 * have data blocks allocated (it may have an external xattr block).
5684 * Report at least one sector for such files, so tools like tar, rsync,
5685 * others don't incorrectly think the file is completely sparse.
5686 */
5687 if (unlikely(ext4_has_inline_data(inode)))
5688 stat->blocks += (stat->size + 511) >> 9;
5690 /*
5691 * We can't update i_blocks if the block allocation is delayed
5692 * otherwise in the case of system crash before the real block
5693 * allocation is done, we will have i_blocks inconsistent with
5694 * on-disk file blocks.
5695 * We always keep i_blocks updated together with real
5696 * allocation. But to not confuse with user, stat
5697 * will return the blocks that include the delayed allocation
5698 * blocks for this file.
5699 */
5700 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5701 EXT4_I(inode)->i_reserved_data_blocks);
5702 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5703 return 0;
5704 }
5706 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5707 int pextents)
5708 {
5709 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5710 return ext4_ind_trans_blocks(inode, lblocks);
5711 return ext4_ext_index_trans_blocks(inode, pextents);
5712 }
5714 /*
5715 * Account for index blocks, block groups bitmaps and block group
5716 * descriptor blocks if modify datablocks and index blocks
5717 * worse case, the indexs blocks spread over different block groups
5718 *
5719 * If datablocks are discontiguous, they are possible to spread over
5720 * different block groups too. If they are contiguous, with flexbg,
5721 * they could still across block group boundary.
5722 *
5723 * Also account for superblock, inode, quota and xattr blocks
5724 */
5725 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5726 int pextents)
5727 {
5728 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5729 int gdpblocks;
5730 int idxblocks;
5731 int ret = 0;
5733 /*
5734 * How many index blocks need to touch to map @lblocks logical blocks
5735 * to @pextents physical extents?
5736 */
5737 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5739 ret = idxblocks;
5741 /*
5742 * Now let's see how many group bitmaps and group descriptors need
5743 * to account
5744 */
5745 groups = idxblocks + pextents;
5746 gdpblocks = groups;
5747 if (groups > ngroups)
5748 groups = ngroups;
5749 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5750 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5752 /* bitmaps and block group descriptor blocks */
5753 ret += groups + gdpblocks;
5755 /* Blocks for super block, inode, quota and xattr blocks */
5756 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5758 return ret;
5759 }
5761 /*
5762 * Calculate the total number of credits to reserve to fit
5763 * the modification of a single pages into a single transaction,
5764 * which may include multiple chunks of block allocations.
5765 *
5766 * This could be called via ext4_write_begin()
5767 *
5768 * We need to consider the worse case, when
5769 * one new block per extent.
5770 */
5771 int ext4_writepage_trans_blocks(struct inode *inode)
5772 {
5773 int bpp = ext4_journal_blocks_per_page(inode);
5774 int ret;
5776 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5778 /* Account for data blocks for journalled mode */
5779 if (ext4_should_journal_data(inode))
5780 ret += bpp;
5781 return ret;
5782 }
5784 /*
5785 * Calculate the journal credits for a chunk of data modification.
5786 *
5787 * This is called from DIO, fallocate or whoever calling
5788 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5789 *
5790 * journal buffers for data blocks are not included here, as DIO
5791 * and fallocate do no need to journal data buffers.
5792 */
5793 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5794 {
5795 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5796 }
5798 /*
5799 * The caller must have previously called ext4_reserve_inode_write().
5800 * Give this, we know that the caller already has write access to iloc->bh.
5801 */
5802 int ext4_mark_iloc_dirty(handle_t *handle,
5803 struct inode *inode, struct ext4_iloc *iloc)
5804 {
5805 int err = 0;
5807 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5808 return -EIO;
5810 if (IS_I_VERSION(inode))
5811 inode_inc_iversion(inode);
5813 /* the do_update_inode consumes one bh->b_count */
5814 get_bh(iloc->bh);
5816 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5817 err = ext4_do_update_inode(handle, inode, iloc);
5818 put_bh(iloc->bh);
5819 return err;
5820 }
5822 /*
5823 * On success, We end up with an outstanding reference count against
5824 * iloc->bh. This _must_ be cleaned up later.
5825 */
5827 int
5828 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5829 struct ext4_iloc *iloc)
5830 {
5831 int err;
5833 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5834 return -EIO;
5836 err = ext4_get_inode_loc(inode, iloc);
5837 if (!err) {
5838 BUFFER_TRACE(iloc->bh, "get_write_access");
5839 err = ext4_journal_get_write_access(handle, iloc->bh);
5840 if (err) {
5841 brelse(iloc->bh);
5842 iloc->bh = NULL;
5843 }
5844 }
5845 ext4_std_error(inode->i_sb, err);
5846 return err;
5847 }
5849 static int __ext4_expand_extra_isize(struct inode *inode,
5850 unsigned int new_extra_isize,
5851 struct ext4_iloc *iloc,
5852 handle_t *handle, int *no_expand)
5853 {
5854 struct ext4_inode *raw_inode;
5855 struct ext4_xattr_ibody_header *header;
5856 int error;
5858 raw_inode = ext4_raw_inode(iloc);
5860 header = IHDR(inode, raw_inode);
5862 /* No extended attributes present */
5863 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5864 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5865 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5866 EXT4_I(inode)->i_extra_isize, 0,
5867 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5868 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5869 return 0;
5870 }
5872 /* try to expand with EAs present */
5873 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5874 raw_inode, handle);
5875 if (error) {
5876 /*
5877 * Inode size expansion failed; don't try again
5878 */
5879 *no_expand = 1;
5880 }
5882 return error;
5883 }
5885 /*
5886 * Expand an inode by new_extra_isize bytes.
5887 * Returns 0 on success or negative error number on failure.
5888 */
5889 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5890 unsigned int new_extra_isize,
5891 struct ext4_iloc iloc,
5892 handle_t *handle)
5893 {
5894 int no_expand;
5895 int error;
5897 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5898 return -EOVERFLOW;
5900 /*
5901 * In nojournal mode, we can immediately attempt to expand
5902 * the inode. When journaled, we first need to obtain extra
5903 * buffer credits since we may write into the EA block
5904 * with this same handle. If journal_extend fails, then it will
5905 * only result in a minor loss of functionality for that inode.
5906 * If this is felt to be critical, then e2fsck should be run to
5907 * force a large enough s_min_extra_isize.
5908 */
5909 if (ext4_handle_valid(handle) &&
5910 jbd2_journal_extend(handle,
5911 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0)
5912 return -ENOSPC;
5914 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5915 return -EBUSY;
5917 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5918 handle, &no_expand);
5919 ext4_write_unlock_xattr(inode, &no_expand);
5921 return error;
5922 }
5924 int ext4_expand_extra_isize(struct inode *inode,
5925 unsigned int new_extra_isize,
5926 struct ext4_iloc *iloc)
5927 {
5928 handle_t *handle;
5929 int no_expand;
5930 int error, rc;
5932 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5933 brelse(iloc->bh);
5934 return -EOVERFLOW;
5935 }
5937 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5938 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5939 if (IS_ERR(handle)) {
5940 error = PTR_ERR(handle);
5941 brelse(iloc->bh);
5942 return error;
5943 }
5945 ext4_write_lock_xattr(inode, &no_expand);
5947 BUFFER_TRACE(iloc.bh, "get_write_access");
5948 error = ext4_journal_get_write_access(handle, iloc->bh);
5949 if (error) {
5950 brelse(iloc->bh);
5951 goto out_stop;
5952 }
5954 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5955 handle, &no_expand);
5957 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5958 if (!error)
5959 error = rc;
5961 ext4_write_unlock_xattr(inode, &no_expand);
5962 out_stop:
5963 ext4_journal_stop(handle);
5964 return error;
5965 }
5967 /*
5968 * What we do here is to mark the in-core inode as clean with respect to inode
5969 * dirtiness (it may still be data-dirty).
5970 * This means that the in-core inode may be reaped by prune_icache
5971 * without having to perform any I/O. This is a very good thing,
5972 * because *any* task may call prune_icache - even ones which
5973 * have a transaction open against a different journal.
5974 *
5975 * Is this cheating? Not really. Sure, we haven't written the
5976 * inode out, but prune_icache isn't a user-visible syncing function.
5977 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5978 * we start and wait on commits.
5979 */
5980 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5981 {
5982 struct ext4_iloc iloc;
5983 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5984 int err;
5986 might_sleep();
5987 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5988 err = ext4_reserve_inode_write(handle, inode, &iloc);
5989 if (err)
5990 return err;
5992 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5993 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5994 iloc, handle);
5996 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5997 }
5999 /*
6000 * ext4_dirty_inode() is called from __mark_inode_dirty()
6001 *
6002 * We're really interested in the case where a file is being extended.
6003 * i_size has been changed by generic_commit_write() and we thus need
6004 * to include the updated inode in the current transaction.
6005 *
6006 * Also, dquot_alloc_block() will always dirty the inode when blocks
6007 * are allocated to the file.
6008 *
6009 * If the inode is marked synchronous, we don't honour that here - doing
6010 * so would cause a commit on atime updates, which we don't bother doing.
6011 * We handle synchronous inodes at the highest possible level.
6012 *
6013 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
6014 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
6015 * to copy into the on-disk inode structure are the timestamp files.
6016 */
6017 void ext4_dirty_inode(struct inode *inode, int flags)
6018 {
6019 handle_t *handle;
6021 if (flags == I_DIRTY_TIME)
6022 return;
6023 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6024 if (IS_ERR(handle))
6025 goto out;
6027 ext4_mark_inode_dirty(handle, inode);
6029 ext4_journal_stop(handle);
6030 out:
6031 return;
6032 }
6034 #if 0
6035 /*
6036 * Bind an inode's backing buffer_head into this transaction, to prevent
6037 * it from being flushed to disk early. Unlike
6038 * ext4_reserve_inode_write, this leaves behind no bh reference and
6039 * returns no iloc structure, so the caller needs to repeat the iloc
6040 * lookup to mark the inode dirty later.
6041 */
6042 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
6043 {
6044 struct ext4_iloc iloc;
6046 int err = 0;
6047 if (handle) {
6048 err = ext4_get_inode_loc(inode, &iloc);
6049 if (!err) {
6050 BUFFER_TRACE(iloc.bh, "get_write_access");
6051 err = jbd2_journal_get_write_access(handle, iloc.bh);
6052 if (!err)
6053 err = ext4_handle_dirty_metadata(handle,
6054 NULL,
6055 iloc.bh);
6056 brelse(iloc.bh);
6057 }
6058 }
6059 ext4_std_error(inode->i_sb, err);
6060 return err;
6061 }
6062 #endif
6064 int ext4_change_inode_journal_flag(struct inode *inode, int val)
6065 {
6066 journal_t *journal;
6067 handle_t *handle;
6068 int err;
6069 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6071 /*
6072 * We have to be very careful here: changing a data block's
6073 * journaling status dynamically is dangerous. If we write a
6074 * data block to the journal, change the status and then delete
6075 * that block, we risk forgetting to revoke the old log record
6076 * from the journal and so a subsequent replay can corrupt data.
6077 * So, first we make sure that the journal is empty and that
6078 * nobody is changing anything.
6079 */
6081 journal = EXT4_JOURNAL(inode);
6082 if (!journal)
6083 return 0;
6084 if (is_journal_aborted(journal))
6085 return -EROFS;
6087 /* Wait for all existing dio workers */
6088 inode_dio_wait(inode);
6090 /*
6091 * Before flushing the journal and switching inode's aops, we have
6092 * to flush all dirty data the inode has. There can be outstanding
6093 * delayed allocations, there can be unwritten extents created by
6094 * fallocate or buffered writes in dioread_nolock mode covered by
6095 * dirty data which can be converted only after flushing the dirty
6096 * data (and journalled aops don't know how to handle these cases).
6097 */
6098 if (val) {
6099 down_write(&EXT4_I(inode)->i_mmap_sem);
6100 err = filemap_write_and_wait(inode->i_mapping);
6101 if (err < 0) {
6102 up_write(&EXT4_I(inode)->i_mmap_sem);
6103 return err;
6104 }
6105 }
6107 percpu_down_write(&sbi->s_journal_flag_rwsem);
6108 jbd2_journal_lock_updates(journal);
6110 /*
6111 * OK, there are no updates running now, and all cached data is
6112 * synced to disk. We are now in a completely consistent state
6113 * which doesn't have anything in the journal, and we know that
6114 * no filesystem updates are running, so it is safe to modify
6115 * the inode's in-core data-journaling state flag now.
6116 */
6118 if (val)
6119 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6120 else {
6121 err = jbd2_journal_flush(journal);
6122 if (err < 0) {
6123 jbd2_journal_unlock_updates(journal);
6124 percpu_up_write(&sbi->s_journal_flag_rwsem);
6125 return err;
6126 }
6127 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6128 }
6129 ext4_set_aops(inode);
6131 jbd2_journal_unlock_updates(journal);
6132 percpu_up_write(&sbi->s_journal_flag_rwsem);
6134 if (val)
6135 up_write(&EXT4_I(inode)->i_mmap_sem);
6137 /* Finally we can mark the inode as dirty. */
6139 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6140 if (IS_ERR(handle))
6141 return PTR_ERR(handle);
6143 err = ext4_mark_inode_dirty(handle, inode);
6144 ext4_handle_sync(handle);
6145 ext4_journal_stop(handle);
6146 ext4_std_error(inode->i_sb, err);
6148 return err;
6149 }
6151 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6152 {
6153 return !buffer_mapped(bh);
6154 }
6156 int ext4_page_mkwrite(struct vm_fault *vmf)
6157 {
6158 struct vm_area_struct *vma = vmf->vma;
6159 struct page *page = vmf->page;
6160 loff_t size;
6161 unsigned long len;
6162 int ret;
6163 struct file *file = vma->vm_file;
6164 struct inode *inode = file_inode(file);
6165 struct address_space *mapping = inode->i_mapping;
6166 handle_t *handle;
6167 get_block_t *get_block;
6168 int retries = 0;
6170 sb_start_pagefault(inode->i_sb);
6171 file_update_time(vma->vm_file);
6173 down_read(&EXT4_I(inode)->i_mmap_sem);
6175 ret = ext4_convert_inline_data(inode);
6176 if (ret)
6177 goto out_ret;
6179 /* Delalloc case is easy... */
6180 if (test_opt(inode->i_sb, DELALLOC) &&
6181 !ext4_should_journal_data(inode) &&
6182 !ext4_nonda_switch(inode->i_sb)) {
6183 do {
6184 ret = block_page_mkwrite(vma, vmf,
6185 ext4_da_get_block_prep);
6186 } while (ret == -ENOSPC &&
6187 ext4_should_retry_alloc(inode->i_sb, &retries));
6188 goto out_ret;
6189 }
6191 lock_page(page);
6192 size = i_size_read(inode);
6193 /* Page got truncated from under us? */
6194 if (page->mapping != mapping || page_offset(page) > size) {
6195 unlock_page(page);
6196 ret = VM_FAULT_NOPAGE;
6197 goto out;
6198 }
6200 if (page->index == size >> PAGE_SHIFT)
6201 len = size & ~PAGE_MASK;
6202 else
6203 len = PAGE_SIZE;
6204 /*
6205 * Return if we have all the buffers mapped. This avoids the need to do
6206 * journal_start/journal_stop which can block and take a long time
6207 */
6208 if (page_has_buffers(page)) {
6209 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6210 0, len, NULL,
6211 ext4_bh_unmapped)) {
6212 /* Wait so that we don't change page under IO */
6213 wait_for_stable_page(page);
6214 ret = VM_FAULT_LOCKED;
6215 goto out;
6216 }
6217 }
6218 unlock_page(page);
6219 /* OK, we need to fill the hole... */
6220 if (ext4_should_dioread_nolock(inode))
6221 get_block = ext4_get_block_unwritten;
6222 else
6223 get_block = ext4_get_block;
6224 retry_alloc:
6225 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6226 ext4_writepage_trans_blocks(inode));
6227 if (IS_ERR(handle)) {
6228 ret = VM_FAULT_SIGBUS;
6229 goto out;
6230 }
6231 ret = block_page_mkwrite(vma, vmf, get_block);
6232 if (!ret && ext4_should_journal_data(inode)) {
6233 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
6234 PAGE_SIZE, NULL, do_journal_get_write_access)) {
6235 unlock_page(page);
6236 ret = VM_FAULT_SIGBUS;
6237 ext4_journal_stop(handle);
6238 goto out;
6239 }
6240 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6241 }
6242 ext4_journal_stop(handle);
6243 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6244 goto retry_alloc;
6245 out_ret:
6246 ret = block_page_mkwrite_return(ret);
6247 out:
6248 up_read(&EXT4_I(inode)->i_mmap_sem);
6249 sb_end_pagefault(inode->i_sb);
6250 return ret;
6251 }
6253 int ext4_filemap_fault(struct vm_fault *vmf)
6254 {
6255 struct inode *inode = file_inode(vmf->vma->vm_file);
6256 int err;
6258 down_read(&EXT4_I(inode)->i_mmap_sem);
6259 err = filemap_fault(vmf);
6260 up_read(&EXT4_I(inode)->i_mmap_sem);
6262 return err;
6263 }