1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/inode.c
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)
12 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
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>
43 #include "ext4_jbd2.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
52 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
53 struct ext4_inode_info *ei)
55 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 int offset = offsetof(struct ext4_inode, i_checksum_lo);
59 unsigned int csum_size = sizeof(dummy_csum);
61 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
62 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
64 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
65 EXT4_GOOD_OLD_INODE_SIZE - offset);
67 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
68 offset = offsetof(struct ext4_inode, i_checksum_hi);
69 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
70 EXT4_GOOD_OLD_INODE_SIZE,
71 offset - EXT4_GOOD_OLD_INODE_SIZE);
72 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
73 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
77 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
78 EXT4_INODE_SIZE(inode->i_sb) - offset);
84 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
85 struct ext4_inode_info *ei)
87 __u32 provided, calculated;
89 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
90 cpu_to_le32(EXT4_OS_LINUX) ||
91 !ext4_has_metadata_csum(inode->i_sb))
94 provided = le16_to_cpu(raw->i_checksum_lo);
95 calculated = ext4_inode_csum(inode, raw, ei);
96 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
97 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
98 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
100 calculated &= 0xFFFF;
102 return provided == calculated;
105 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
106 struct ext4_inode_info *ei)
110 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
111 cpu_to_le32(EXT4_OS_LINUX) ||
112 !ext4_has_metadata_csum(inode->i_sb))
115 csum = ext4_inode_csum(inode, raw, ei);
116 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
117 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
118 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
119 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
122 static inline int ext4_begin_ordered_truncate(struct inode *inode,
125 trace_ext4_begin_ordered_truncate(inode, new_size);
127 * If jinode is zero, then we never opened the file for
128 * writing, so there's no need to call
129 * jbd2_journal_begin_ordered_truncate() since there's no
130 * outstanding writes we need to flush.
132 if (!EXT4_I(inode)->jinode)
134 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
135 EXT4_I(inode)->jinode,
139 static void ext4_invalidatepage(struct page *page, unsigned int offset,
140 unsigned int length);
141 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
142 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
143 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
147 * Test whether an inode is a fast symlink.
148 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
150 int ext4_inode_is_fast_symlink(struct inode *inode)
152 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
153 int ea_blocks = EXT4_I(inode)->i_file_acl ?
154 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
156 if (ext4_has_inline_data(inode))
159 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
161 return S_ISLNK(inode->i_mode) && inode->i_size &&
162 (inode->i_size < EXT4_N_BLOCKS * 4);
166 * Restart the transaction associated with *handle. This does a commit,
167 * so before we call here everything must be consistently dirtied against
170 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
176 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
177 * moment, get_block can be called only for blocks inside i_size since
178 * page cache has been already dropped and writes are blocked by
179 * i_mutex. So we can safely drop the i_data_sem here.
181 BUG_ON(EXT4_JOURNAL(inode) == NULL);
182 jbd_debug(2, "restarting handle %p\n", handle);
183 up_write(&EXT4_I(inode)->i_data_sem);
184 ret = ext4_journal_restart(handle, nblocks);
185 down_write(&EXT4_I(inode)->i_data_sem);
186 ext4_discard_preallocations(inode);
192 * Called at the last iput() if i_nlink is zero.
194 void ext4_evict_inode(struct inode *inode)
199 * Credits for final inode cleanup and freeing:
200 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
201 * (xattr block freeing), bitmap, group descriptor (inode freeing)
203 int extra_credits = 6;
204 struct ext4_xattr_inode_array *ea_inode_array = NULL;
205 bool freeze_protected = false;
207 trace_ext4_evict_inode(inode);
209 if (inode->i_nlink) {
211 * When journalling data dirty buffers are tracked only in the
212 * journal. So although mm thinks everything is clean and
213 * ready for reaping the inode might still have some pages to
214 * write in the running transaction or waiting to be
215 * checkpointed. Thus calling jbd2_journal_invalidatepage()
216 * (via truncate_inode_pages()) to discard these buffers can
217 * cause data loss. Also even if we did not discard these
218 * buffers, we would have no way to find them after the inode
219 * is reaped and thus user could see stale data if he tries to
220 * read them before the transaction is checkpointed. So be
221 * careful and force everything to disk here... We use
222 * ei->i_datasync_tid to store the newest transaction
223 * containing inode's data.
225 * Note that directories do not have this problem because they
226 * don't use page cache.
228 if (inode->i_ino != EXT4_JOURNAL_INO &&
229 ext4_should_journal_data(inode) &&
230 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
231 inode->i_data.nrpages) {
232 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
233 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
235 jbd2_complete_transaction(journal, commit_tid);
236 filemap_write_and_wait(&inode->i_data);
238 truncate_inode_pages_final(&inode->i_data);
243 if (is_bad_inode(inode))
245 dquot_initialize(inode);
247 if (ext4_should_order_data(inode))
248 ext4_begin_ordered_truncate(inode, 0);
249 truncate_inode_pages_final(&inode->i_data);
252 * Protect us against freezing - iput() caller didn't have to have any
253 * protection against it. When we are in a running transaction though,
254 * we are already protected against freezing and we cannot grab further
255 * protection due to lock ordering constraints.
257 if (!ext4_journal_current_handle()) {
258 sb_start_intwrite(inode->i_sb);
259 freeze_protected = true;
262 if (!IS_NOQUOTA(inode))
263 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
266 * Block bitmap, group descriptor, and inode are accounted in both
267 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
269 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
270 ext4_blocks_for_truncate(inode) + extra_credits - 3);
271 if (IS_ERR(handle)) {
272 ext4_std_error(inode->i_sb, PTR_ERR(handle));
274 * If we're going to skip the normal cleanup, we still need to
275 * make sure that the in-core orphan linked list is properly
278 ext4_orphan_del(NULL, inode);
279 if (freeze_protected)
280 sb_end_intwrite(inode->i_sb);
285 ext4_handle_sync(handle);
288 * Set inode->i_size to 0 before calling ext4_truncate(). We need
289 * special handling of symlinks here because i_size is used to
290 * determine whether ext4_inode_info->i_data contains symlink data or
291 * block mappings. Setting i_size to 0 will remove its fast symlink
292 * status. Erase i_data so that it becomes a valid empty block map.
294 if (ext4_inode_is_fast_symlink(inode))
295 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
297 err = ext4_mark_inode_dirty(handle, inode);
299 ext4_warning(inode->i_sb,
300 "couldn't mark inode dirty (err %d)", err);
303 if (inode->i_blocks) {
304 err = ext4_truncate(inode);
306 ext4_error(inode->i_sb,
307 "couldn't truncate inode %lu (err %d)",
313 /* Remove xattr references. */
314 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
317 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
319 ext4_journal_stop(handle);
320 ext4_orphan_del(NULL, inode);
321 if (freeze_protected)
322 sb_end_intwrite(inode->i_sb);
323 ext4_xattr_inode_array_free(ea_inode_array);
328 * Kill off the orphan record which ext4_truncate created.
329 * AKPM: I think this can be inside the above `if'.
330 * Note that ext4_orphan_del() has to be able to cope with the
331 * deletion of a non-existent orphan - this is because we don't
332 * know if ext4_truncate() actually created an orphan record.
333 * (Well, we could do this if we need to, but heck - it works)
335 ext4_orphan_del(handle, inode);
336 EXT4_I(inode)->i_dtime = get_seconds();
339 * One subtle ordering requirement: if anything has gone wrong
340 * (transaction abort, IO errors, whatever), then we can still
341 * do these next steps (the fs will already have been marked as
342 * having errors), but we can't free the inode if the mark_dirty
345 if (ext4_mark_inode_dirty(handle, inode))
346 /* If that failed, just do the required in-core inode clear. */
347 ext4_clear_inode(inode);
349 ext4_free_inode(handle, inode);
350 ext4_journal_stop(handle);
351 if (freeze_protected)
352 sb_end_intwrite(inode->i_sb);
353 ext4_xattr_inode_array_free(ea_inode_array);
356 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
360 qsize_t *ext4_get_reserved_space(struct inode *inode)
362 return &EXT4_I(inode)->i_reserved_quota;
367 * Called with i_data_sem down, which is important since we can call
368 * ext4_discard_preallocations() from here.
370 void ext4_da_update_reserve_space(struct inode *inode,
371 int used, int quota_claim)
373 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
374 struct ext4_inode_info *ei = EXT4_I(inode);
376 spin_lock(&ei->i_block_reservation_lock);
377 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
378 if (unlikely(used > ei->i_reserved_data_blocks)) {
379 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
380 "with only %d reserved data blocks",
381 __func__, inode->i_ino, used,
382 ei->i_reserved_data_blocks);
384 used = ei->i_reserved_data_blocks;
387 /* Update per-inode reservations */
388 ei->i_reserved_data_blocks -= used;
389 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
391 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
393 /* Update quota subsystem for data blocks */
395 dquot_claim_block(inode, EXT4_C2B(sbi, used));
398 * We did fallocate with an offset that is already delayed
399 * allocated. So on delayed allocated writeback we should
400 * not re-claim the quota for fallocated blocks.
402 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
406 * If we have done all the pending block allocations and if
407 * there aren't any writers on the inode, we can discard the
408 * inode's preallocations.
410 if ((ei->i_reserved_data_blocks == 0) &&
411 (atomic_read(&inode->i_writecount) == 0))
412 ext4_discard_preallocations(inode);
415 static int __check_block_validity(struct inode *inode, const char *func,
417 struct ext4_map_blocks *map)
419 if (ext4_has_feature_journal(inode->i_sb) &&
421 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
423 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
424 ext4_error_inode(inode, func, line, map->m_pblk,
425 "lblock %lu mapped to illegal pblock %llu "
426 "(length %d)", (unsigned long) map->m_lblk,
427 map->m_pblk, map->m_len);
428 return -EFSCORRUPTED;
433 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
438 if (ext4_encrypted_inode(inode))
439 return fscrypt_zeroout_range(inode, lblk, pblk, len);
441 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
448 #define check_block_validity(inode, map) \
449 __check_block_validity((inode), __func__, __LINE__, (map))
451 #ifdef ES_AGGRESSIVE_TEST
452 static void ext4_map_blocks_es_recheck(handle_t *handle,
454 struct ext4_map_blocks *es_map,
455 struct ext4_map_blocks *map,
462 * There is a race window that the result is not the same.
463 * e.g. xfstests #223 when dioread_nolock enables. The reason
464 * is that we lookup a block mapping in extent status tree with
465 * out taking i_data_sem. So at the time the unwritten extent
466 * could be converted.
468 down_read(&EXT4_I(inode)->i_data_sem);
469 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
470 retval = ext4_ext_map_blocks(handle, inode, map, flags &
471 EXT4_GET_BLOCKS_KEEP_SIZE);
473 retval = ext4_ind_map_blocks(handle, inode, map, flags &
474 EXT4_GET_BLOCKS_KEEP_SIZE);
476 up_read((&EXT4_I(inode)->i_data_sem));
479 * We don't check m_len because extent will be collpased in status
480 * tree. So the m_len might not equal.
482 if (es_map->m_lblk != map->m_lblk ||
483 es_map->m_flags != map->m_flags ||
484 es_map->m_pblk != map->m_pblk) {
485 printk("ES cache assertion failed for inode: %lu "
486 "es_cached ex [%d/%d/%llu/%x] != "
487 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
488 inode->i_ino, es_map->m_lblk, es_map->m_len,
489 es_map->m_pblk, es_map->m_flags, map->m_lblk,
490 map->m_len, map->m_pblk, map->m_flags,
494 #endif /* ES_AGGRESSIVE_TEST */
497 * The ext4_map_blocks() function tries to look up the requested blocks,
498 * and returns if the blocks are already mapped.
500 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
501 * and store the allocated blocks in the result buffer head and mark it
504 * If file type is extents based, it will call ext4_ext_map_blocks(),
505 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
508 * On success, it returns the number of blocks being mapped or allocated. if
509 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
510 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
512 * It returns 0 if plain look up failed (blocks have not been allocated), in
513 * that case, @map is returned as unmapped but we still do fill map->m_len to
514 * indicate the length of a hole starting at map->m_lblk.
516 * It returns the error in case of allocation failure.
518 int ext4_map_blocks(handle_t *handle, struct inode *inode,
519 struct ext4_map_blocks *map, int flags)
521 struct extent_status es;
524 #ifdef ES_AGGRESSIVE_TEST
525 struct ext4_map_blocks orig_map;
527 memcpy(&orig_map, map, sizeof(*map));
531 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
532 "logical block %lu\n", inode->i_ino, flags, map->m_len,
533 (unsigned long) map->m_lblk);
536 * ext4_map_blocks returns an int, and m_len is an unsigned int
538 if (unlikely(map->m_len > INT_MAX))
539 map->m_len = INT_MAX;
541 /* We can handle the block number less than EXT_MAX_BLOCKS */
542 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
543 return -EFSCORRUPTED;
545 /* Lookup extent status tree firstly */
546 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
547 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
548 map->m_pblk = ext4_es_pblock(&es) +
549 map->m_lblk - es.es_lblk;
550 map->m_flags |= ext4_es_is_written(&es) ?
551 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
552 retval = es.es_len - (map->m_lblk - es.es_lblk);
553 if (retval > map->m_len)
556 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
558 retval = es.es_len - (map->m_lblk - es.es_lblk);
559 if (retval > map->m_len)
566 #ifdef ES_AGGRESSIVE_TEST
567 ext4_map_blocks_es_recheck(handle, inode, map,
574 * Try to see if we can get the block without requesting a new
577 down_read(&EXT4_I(inode)->i_data_sem);
578 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
579 retval = ext4_ext_map_blocks(handle, inode, map, flags &
580 EXT4_GET_BLOCKS_KEEP_SIZE);
582 retval = ext4_ind_map_blocks(handle, inode, map, flags &
583 EXT4_GET_BLOCKS_KEEP_SIZE);
588 if (unlikely(retval != map->m_len)) {
589 ext4_warning(inode->i_sb,
590 "ES len assertion failed for inode "
591 "%lu: retval %d != map->m_len %d",
592 inode->i_ino, retval, map->m_len);
596 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
597 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
598 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
599 !(status & EXTENT_STATUS_WRITTEN) &&
600 ext4_find_delalloc_range(inode, map->m_lblk,
601 map->m_lblk + map->m_len - 1))
602 status |= EXTENT_STATUS_DELAYED;
603 ret = ext4_es_insert_extent(inode, map->m_lblk,
604 map->m_len, map->m_pblk, status);
608 up_read((&EXT4_I(inode)->i_data_sem));
611 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
612 ret = check_block_validity(inode, map);
617 /* If it is only a block(s) look up */
618 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
622 * Returns if the blocks have already allocated
624 * Note that if blocks have been preallocated
625 * ext4_ext_get_block() returns the create = 0
626 * with buffer head unmapped.
628 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
630 * If we need to convert extent to unwritten
631 * we continue and do the actual work in
632 * ext4_ext_map_blocks()
634 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
638 * Here we clear m_flags because after allocating an new extent,
639 * it will be set again.
641 map->m_flags &= ~EXT4_MAP_FLAGS;
644 * New blocks allocate and/or writing to unwritten extent
645 * will possibly result in updating i_data, so we take
646 * the write lock of i_data_sem, and call get_block()
647 * with create == 1 flag.
649 down_write(&EXT4_I(inode)->i_data_sem);
652 * We need to check for EXT4 here because migrate
653 * could have changed the inode type in between
655 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
656 retval = ext4_ext_map_blocks(handle, inode, map, flags);
658 retval = ext4_ind_map_blocks(handle, inode, map, flags);
660 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
662 * We allocated new blocks which will result in
663 * i_data's format changing. Force the migrate
664 * to fail by clearing migrate flags
666 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
673 if (unlikely(retval != map->m_len)) {
674 ext4_warning(inode->i_sb,
675 "ES len assertion failed for inode "
676 "%lu: retval %d != map->m_len %d",
677 inode->i_ino, retval, map->m_len);
682 * We have to zeroout blocks before inserting them into extent
683 * status tree. Otherwise someone could look them up there and
684 * use them before they are really zeroed. We also have to
685 * unmap metadata before zeroing as otherwise writeback can
686 * overwrite zeros with stale data from block device.
688 if (flags & EXT4_GET_BLOCKS_ZERO &&
689 map->m_flags & EXT4_MAP_MAPPED &&
690 map->m_flags & EXT4_MAP_NEW) {
691 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
693 ret = ext4_issue_zeroout(inode, map->m_lblk,
694 map->m_pblk, map->m_len);
702 * If the extent has been zeroed out, we don't need to update
703 * extent status tree.
705 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
706 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
707 if (ext4_es_is_written(&es))
710 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
711 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
712 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
713 !(status & EXTENT_STATUS_WRITTEN) &&
714 ext4_find_delalloc_range(inode, map->m_lblk,
715 map->m_lblk + map->m_len - 1))
716 status |= EXTENT_STATUS_DELAYED;
717 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
718 map->m_pblk, status);
726 up_write((&EXT4_I(inode)->i_data_sem));
727 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
728 ret = check_block_validity(inode, map);
733 * Inodes with freshly allocated blocks where contents will be
734 * visible after transaction commit must be on transaction's
737 if (map->m_flags & EXT4_MAP_NEW &&
738 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
739 !(flags & EXT4_GET_BLOCKS_ZERO) &&
740 !ext4_is_quota_file(inode) &&
741 ext4_should_order_data(inode)) {
743 (loff_t)map->m_lblk << inode->i_blkbits;
744 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
746 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
747 ret = ext4_jbd2_inode_add_wait(handle, inode,
750 ret = ext4_jbd2_inode_add_write(handle, inode,
760 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
761 * we have to be careful as someone else may be manipulating b_state as well.
763 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
765 unsigned long old_state;
766 unsigned long new_state;
768 flags &= EXT4_MAP_FLAGS;
770 /* Dummy buffer_head? Set non-atomically. */
772 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
776 * Someone else may be modifying b_state. Be careful! This is ugly but
777 * once we get rid of using bh as a container for mapping information
778 * to pass to / from get_block functions, this can go away.
781 old_state = READ_ONCE(bh->b_state);
782 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
784 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
787 static int _ext4_get_block(struct inode *inode, sector_t iblock,
788 struct buffer_head *bh, int flags)
790 struct ext4_map_blocks map;
793 if (ext4_has_inline_data(inode))
797 map.m_len = bh->b_size >> inode->i_blkbits;
799 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
802 map_bh(bh, inode->i_sb, map.m_pblk);
803 ext4_update_bh_state(bh, map.m_flags);
804 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
806 } else if (ret == 0) {
807 /* hole case, need to fill in bh->b_size */
808 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
813 int ext4_get_block(struct inode *inode, sector_t iblock,
814 struct buffer_head *bh, int create)
816 return _ext4_get_block(inode, iblock, bh,
817 create ? EXT4_GET_BLOCKS_CREATE : 0);
821 * Get block function used when preparing for buffered write if we require
822 * creating an unwritten extent if blocks haven't been allocated. The extent
823 * will be converted to written after the IO is complete.
825 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
826 struct buffer_head *bh_result, int create)
828 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
829 inode->i_ino, create);
830 return _ext4_get_block(inode, iblock, bh_result,
831 EXT4_GET_BLOCKS_IO_CREATE_EXT);
834 /* Maximum number of blocks we map for direct IO at once. */
835 #define DIO_MAX_BLOCKS 4096
838 * Get blocks function for the cases that need to start a transaction -
839 * generally difference cases of direct IO and DAX IO. It also handles retries
842 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
843 struct buffer_head *bh_result, int flags)
850 /* Trim mapping request to maximum we can map at once for DIO */
851 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
852 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
853 dio_credits = ext4_chunk_trans_blocks(inode,
854 bh_result->b_size >> inode->i_blkbits);
856 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
858 return PTR_ERR(handle);
860 ret = _ext4_get_block(inode, iblock, bh_result, flags);
861 ext4_journal_stop(handle);
863 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
868 /* Get block function for DIO reads and writes to inodes without extents */
869 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
870 struct buffer_head *bh, int create)
872 /* We don't expect handle for direct IO */
873 WARN_ON_ONCE(ext4_journal_current_handle());
876 return _ext4_get_block(inode, iblock, bh, 0);
877 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
881 * Get block function for AIO DIO writes when we create unwritten extent if
882 * blocks are not allocated yet. The extent will be converted to written
883 * after IO is complete.
885 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
886 sector_t iblock, struct buffer_head *bh_result, int create)
890 /* We don't expect handle for direct IO */
891 WARN_ON_ONCE(ext4_journal_current_handle());
893 ret = ext4_get_block_trans(inode, iblock, bh_result,
894 EXT4_GET_BLOCKS_IO_CREATE_EXT);
897 * When doing DIO using unwritten extents, we need io_end to convert
898 * unwritten extents to written on IO completion. We allocate io_end
899 * once we spot unwritten extent and store it in b_private. Generic
900 * DIO code keeps b_private set and furthermore passes the value to
901 * our completion callback in 'private' argument.
903 if (!ret && buffer_unwritten(bh_result)) {
904 if (!bh_result->b_private) {
905 ext4_io_end_t *io_end;
907 io_end = ext4_init_io_end(inode, GFP_KERNEL);
910 bh_result->b_private = io_end;
911 ext4_set_io_unwritten_flag(inode, io_end);
913 set_buffer_defer_completion(bh_result);
920 * Get block function for non-AIO DIO writes when we create unwritten extent if
921 * blocks are not allocated yet. The extent will be converted to written
922 * after IO is complete by ext4_direct_IO_write().
924 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
925 sector_t iblock, struct buffer_head *bh_result, int create)
929 /* We don't expect handle for direct IO */
930 WARN_ON_ONCE(ext4_journal_current_handle());
932 ret = ext4_get_block_trans(inode, iblock, bh_result,
933 EXT4_GET_BLOCKS_IO_CREATE_EXT);
936 * Mark inode as having pending DIO writes to unwritten extents.
937 * ext4_direct_IO_write() checks this flag and converts extents to
940 if (!ret && buffer_unwritten(bh_result))
941 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
946 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
947 struct buffer_head *bh_result, int create)
951 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
952 inode->i_ino, create);
953 /* We don't expect handle for direct IO */
954 WARN_ON_ONCE(ext4_journal_current_handle());
956 ret = _ext4_get_block(inode, iblock, bh_result, 0);
958 * Blocks should have been preallocated! ext4_file_write_iter() checks
961 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
968 * `handle' can be NULL if create is zero
970 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
971 ext4_lblk_t block, int map_flags)
973 struct ext4_map_blocks map;
974 struct buffer_head *bh;
975 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
978 J_ASSERT(handle != NULL || create == 0);
982 err = ext4_map_blocks(handle, inode, &map, map_flags);
985 return create ? ERR_PTR(-ENOSPC) : NULL;
989 bh = sb_getblk(inode->i_sb, map.m_pblk);
991 return ERR_PTR(-ENOMEM);
992 if (map.m_flags & EXT4_MAP_NEW) {
993 J_ASSERT(create != 0);
994 J_ASSERT(handle != NULL);
997 * Now that we do not always journal data, we should
998 * keep in mind whether this should always journal the
999 * new buffer as metadata. For now, regular file
1000 * writes use ext4_get_block instead, so it's not a
1004 BUFFER_TRACE(bh, "call get_create_access");
1005 err = ext4_journal_get_create_access(handle, bh);
1006 if (unlikely(err)) {
1010 if (!buffer_uptodate(bh)) {
1011 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1012 set_buffer_uptodate(bh);
1015 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1016 err = ext4_handle_dirty_metadata(handle, inode, bh);
1020 BUFFER_TRACE(bh, "not a new buffer");
1024 return ERR_PTR(err);
1027 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1028 ext4_lblk_t block, int map_flags)
1030 struct buffer_head *bh;
1032 bh = ext4_getblk(handle, inode, block, map_flags);
1035 if (!bh || buffer_uptodate(bh))
1037 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1039 if (buffer_uptodate(bh))
1042 return ERR_PTR(-EIO);
1045 /* Read a contiguous batch of blocks. */
1046 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1047 bool wait, struct buffer_head **bhs)
1051 for (i = 0; i < bh_count; i++) {
1052 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1053 if (IS_ERR(bhs[i])) {
1054 err = PTR_ERR(bhs[i]);
1060 for (i = 0; i < bh_count; i++)
1061 /* Note that NULL bhs[i] is valid because of holes. */
1062 if (bhs[i] && !buffer_uptodate(bhs[i]))
1063 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1069 for (i = 0; i < bh_count; i++)
1071 wait_on_buffer(bhs[i]);
1073 for (i = 0; i < bh_count; i++) {
1074 if (bhs[i] && !buffer_uptodate(bhs[i])) {
1082 for (i = 0; i < bh_count; i++) {
1089 int ext4_walk_page_buffers(handle_t *handle,
1090 struct buffer_head *head,
1094 int (*fn)(handle_t *handle,
1095 struct buffer_head *bh))
1097 struct buffer_head *bh;
1098 unsigned block_start, block_end;
1099 unsigned blocksize = head->b_size;
1101 struct buffer_head *next;
1103 for (bh = head, block_start = 0;
1104 ret == 0 && (bh != head || !block_start);
1105 block_start = block_end, bh = next) {
1106 next = bh->b_this_page;
1107 block_end = block_start + blocksize;
1108 if (block_end <= from || block_start >= to) {
1109 if (partial && !buffer_uptodate(bh))
1113 err = (*fn)(handle, bh);
1121 * To preserve ordering, it is essential that the hole instantiation and
1122 * the data write be encapsulated in a single transaction. We cannot
1123 * close off a transaction and start a new one between the ext4_get_block()
1124 * and the commit_write(). So doing the jbd2_journal_start at the start of
1125 * prepare_write() is the right place.
1127 * Also, this function can nest inside ext4_writepage(). In that case, we
1128 * *know* that ext4_writepage() has generated enough buffer credits to do the
1129 * whole page. So we won't block on the journal in that case, which is good,
1130 * because the caller may be PF_MEMALLOC.
1132 * By accident, ext4 can be reentered when a transaction is open via
1133 * quota file writes. If we were to commit the transaction while thus
1134 * reentered, there can be a deadlock - we would be holding a quota
1135 * lock, and the commit would never complete if another thread had a
1136 * transaction open and was blocking on the quota lock - a ranking
1139 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1140 * will _not_ run commit under these circumstances because handle->h_ref
1141 * is elevated. We'll still have enough credits for the tiny quotafile
1144 int do_journal_get_write_access(handle_t *handle,
1145 struct buffer_head *bh)
1147 int dirty = buffer_dirty(bh);
1150 if (!buffer_mapped(bh) || buffer_freed(bh))
1153 * __block_write_begin() could have dirtied some buffers. Clean
1154 * the dirty bit as jbd2_journal_get_write_access() could complain
1155 * otherwise about fs integrity issues. Setting of the dirty bit
1156 * by __block_write_begin() isn't a real problem here as we clear
1157 * the bit before releasing a page lock and thus writeback cannot
1158 * ever write the buffer.
1161 clear_buffer_dirty(bh);
1162 BUFFER_TRACE(bh, "get write access");
1163 ret = ext4_journal_get_write_access(handle, bh);
1165 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1169 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1170 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1171 get_block_t *get_block)
1173 unsigned from = pos & (PAGE_SIZE - 1);
1174 unsigned to = from + len;
1175 struct inode *inode = page->mapping->host;
1176 unsigned block_start, block_end;
1179 unsigned blocksize = inode->i_sb->s_blocksize;
1181 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1182 bool decrypt = false;
1184 BUG_ON(!PageLocked(page));
1185 BUG_ON(from > PAGE_SIZE);
1186 BUG_ON(to > PAGE_SIZE);
1189 if (!page_has_buffers(page))
1190 create_empty_buffers(page, blocksize, 0);
1191 head = page_buffers(page);
1192 bbits = ilog2(blocksize);
1193 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1195 for (bh = head, block_start = 0; bh != head || !block_start;
1196 block++, block_start = block_end, bh = bh->b_this_page) {
1197 block_end = block_start + blocksize;
1198 if (block_end <= from || block_start >= to) {
1199 if (PageUptodate(page)) {
1200 if (!buffer_uptodate(bh))
1201 set_buffer_uptodate(bh);
1206 clear_buffer_new(bh);
1207 if (!buffer_mapped(bh)) {
1208 WARN_ON(bh->b_size != blocksize);
1209 err = get_block(inode, block, bh, 1);
1212 if (buffer_new(bh)) {
1213 clean_bdev_bh_alias(bh);
1214 if (PageUptodate(page)) {
1215 clear_buffer_new(bh);
1216 set_buffer_uptodate(bh);
1217 mark_buffer_dirty(bh);
1220 if (block_end > to || block_start < from)
1221 zero_user_segments(page, to, block_end,
1226 if (PageUptodate(page)) {
1227 if (!buffer_uptodate(bh))
1228 set_buffer_uptodate(bh);
1231 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1232 !buffer_unwritten(bh) &&
1233 (block_start < from || block_end > to)) {
1234 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1236 decrypt = ext4_encrypted_inode(inode) &&
1237 S_ISREG(inode->i_mode);
1241 * If we issued read requests, let them complete.
1243 while (wait_bh > wait) {
1244 wait_on_buffer(*--wait_bh);
1245 if (!buffer_uptodate(*wait_bh))
1249 page_zero_new_buffers(page, from, to);
1251 err = fscrypt_decrypt_page(page->mapping->host, page,
1252 PAGE_SIZE, 0, page->index);
1257 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1258 loff_t pos, unsigned len, unsigned flags,
1259 struct page **pagep, void **fsdata)
1261 struct inode *inode = mapping->host;
1262 int ret, needed_blocks;
1269 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1272 trace_ext4_write_begin(inode, pos, len, flags);
1274 * Reserve one block more for addition to orphan list in case
1275 * we allocate blocks but write fails for some reason
1277 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1278 index = pos >> PAGE_SHIFT;
1279 from = pos & (PAGE_SIZE - 1);
1282 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1283 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1292 * grab_cache_page_write_begin() can take a long time if the
1293 * system is thrashing due to memory pressure, or if the page
1294 * is being written back. So grab it first before we start
1295 * the transaction handle. This also allows us to allocate
1296 * the page (if needed) without using GFP_NOFS.
1299 page = grab_cache_page_write_begin(mapping, index, flags);
1303 * The same as page allocation, we prealloc buffer heads before
1304 * starting the handle.
1306 if (!page_has_buffers(page))
1307 create_empty_buffers(page, inode->i_sb->s_blocksize, 0);
1312 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1313 if (IS_ERR(handle)) {
1315 return PTR_ERR(handle);
1319 if (page->mapping != mapping) {
1320 /* The page got truncated from under us */
1323 ext4_journal_stop(handle);
1326 /* In case writeback began while the page was unlocked */
1327 wait_for_stable_page(page);
1329 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1330 if (ext4_should_dioread_nolock(inode))
1331 ret = ext4_block_write_begin(page, pos, len,
1332 ext4_get_block_unwritten);
1334 ret = ext4_block_write_begin(page, pos, len,
1337 if (ext4_should_dioread_nolock(inode))
1338 ret = __block_write_begin(page, pos, len,
1339 ext4_get_block_unwritten);
1341 ret = __block_write_begin(page, pos, len, ext4_get_block);
1343 if (!ret && ext4_should_journal_data(inode)) {
1344 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1346 do_journal_get_write_access);
1352 * __block_write_begin may have instantiated a few blocks
1353 * outside i_size. Trim these off again. Don't need
1354 * i_size_read because we hold i_mutex.
1356 * Add inode to orphan list in case we crash before
1359 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1360 ext4_orphan_add(handle, inode);
1362 ext4_journal_stop(handle);
1363 if (pos + len > inode->i_size) {
1364 ext4_truncate_failed_write(inode);
1366 * If truncate failed early the inode might
1367 * still be on the orphan list; we need to
1368 * make sure the inode is removed from the
1369 * orphan list in that case.
1372 ext4_orphan_del(NULL, inode);
1375 if (ret == -ENOSPC &&
1376 ext4_should_retry_alloc(inode->i_sb, &retries))
1385 /* For write_end() in data=journal mode */
1386 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1389 if (!buffer_mapped(bh) || buffer_freed(bh))
1391 set_buffer_uptodate(bh);
1392 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1393 clear_buffer_meta(bh);
1394 clear_buffer_prio(bh);
1399 * We need to pick up the new inode size which generic_commit_write gave us
1400 * `file' can be NULL - eg, when called from page_symlink().
1402 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1403 * buffers are managed internally.
1405 static int ext4_write_end(struct file *file,
1406 struct address_space *mapping,
1407 loff_t pos, unsigned len, unsigned copied,
1408 struct page *page, void *fsdata)
1410 handle_t *handle = ext4_journal_current_handle();
1411 struct inode *inode = mapping->host;
1412 loff_t old_size = inode->i_size;
1414 int i_size_changed = 0;
1415 int inline_data = ext4_has_inline_data(inode);
1417 trace_ext4_write_end(inode, pos, len, copied);
1419 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1420 ret = ext4_write_inline_data_end(inode, pos, len,
1429 copied = block_write_end(file, mapping, pos,
1430 len, copied, page, fsdata);
1432 * it's important to update i_size while still holding page lock:
1433 * page writeout could otherwise come in and zero beyond i_size.
1435 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1440 pagecache_isize_extended(inode, old_size, pos);
1442 * Don't mark the inode dirty under page lock. First, it unnecessarily
1443 * makes the holding time of page lock longer. Second, it forces lock
1444 * ordering of page lock and transaction start for journaling
1447 if (i_size_changed || inline_data)
1448 ext4_mark_inode_dirty(handle, inode);
1450 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1451 /* if we have allocated more blocks and copied
1452 * less. We will have blocks allocated outside
1453 * inode->i_size. So truncate them
1455 ext4_orphan_add(handle, inode);
1457 ret2 = ext4_journal_stop(handle);
1461 if (pos + len > inode->i_size) {
1462 ext4_truncate_failed_write(inode);
1464 * If truncate failed early the inode might still be
1465 * on the orphan list; we need to make sure the inode
1466 * is removed from the orphan list in that case.
1469 ext4_orphan_del(NULL, inode);
1472 return ret ? ret : copied;
1476 * This is a private version of page_zero_new_buffers() which doesn't
1477 * set the buffer to be dirty, since in data=journalled mode we need
1478 * to call ext4_handle_dirty_metadata() instead.
1480 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1482 unsigned from, unsigned to)
1484 unsigned int block_start = 0, block_end;
1485 struct buffer_head *head, *bh;
1487 bh = head = page_buffers(page);
1489 block_end = block_start + bh->b_size;
1490 if (buffer_new(bh)) {
1491 if (block_end > from && block_start < to) {
1492 if (!PageUptodate(page)) {
1493 unsigned start, size;
1495 start = max(from, block_start);
1496 size = min(to, block_end) - start;
1498 zero_user(page, start, size);
1499 write_end_fn(handle, bh);
1501 clear_buffer_new(bh);
1504 block_start = block_end;
1505 bh = bh->b_this_page;
1506 } while (bh != head);
1509 static int ext4_journalled_write_end(struct file *file,
1510 struct address_space *mapping,
1511 loff_t pos, unsigned len, unsigned copied,
1512 struct page *page, void *fsdata)
1514 handle_t *handle = ext4_journal_current_handle();
1515 struct inode *inode = mapping->host;
1516 loff_t old_size = inode->i_size;
1520 int size_changed = 0;
1521 int inline_data = ext4_has_inline_data(inode);
1523 trace_ext4_journalled_write_end(inode, pos, len, copied);
1524 from = pos & (PAGE_SIZE - 1);
1527 BUG_ON(!ext4_handle_valid(handle));
1530 ret = ext4_write_inline_data_end(inode, pos, len,
1538 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1540 ext4_journalled_zero_new_buffers(handle, page, from, to);
1542 if (unlikely(copied < len))
1543 ext4_journalled_zero_new_buffers(handle, page,
1545 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1546 from + copied, &partial,
1549 SetPageUptodate(page);
1551 size_changed = ext4_update_inode_size(inode, pos + copied);
1552 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1553 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1558 pagecache_isize_extended(inode, old_size, pos);
1560 if (size_changed || inline_data) {
1561 ret2 = ext4_mark_inode_dirty(handle, inode);
1566 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1567 /* if we have allocated more blocks and copied
1568 * less. We will have blocks allocated outside
1569 * inode->i_size. So truncate them
1571 ext4_orphan_add(handle, inode);
1574 ret2 = ext4_journal_stop(handle);
1577 if (pos + len > inode->i_size) {
1578 ext4_truncate_failed_write(inode);
1580 * If truncate failed early the inode might still be
1581 * on the orphan list; we need to make sure the inode
1582 * is removed from the orphan list in that case.
1585 ext4_orphan_del(NULL, inode);
1588 return ret ? ret : copied;
1592 * Reserve space for a single cluster
1594 static int ext4_da_reserve_space(struct inode *inode)
1596 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1597 struct ext4_inode_info *ei = EXT4_I(inode);
1601 * We will charge metadata quota at writeout time; this saves
1602 * us from metadata over-estimation, though we may go over by
1603 * a small amount in the end. Here we just reserve for data.
1605 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1609 spin_lock(&ei->i_block_reservation_lock);
1610 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1611 spin_unlock(&ei->i_block_reservation_lock);
1612 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1615 ei->i_reserved_data_blocks++;
1616 trace_ext4_da_reserve_space(inode);
1617 spin_unlock(&ei->i_block_reservation_lock);
1619 return 0; /* success */
1622 static void ext4_da_release_space(struct inode *inode, int to_free)
1624 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1625 struct ext4_inode_info *ei = EXT4_I(inode);
1628 return; /* Nothing to release, exit */
1630 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1632 trace_ext4_da_release_space(inode, to_free);
1633 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1635 * if there aren't enough reserved blocks, then the
1636 * counter is messed up somewhere. Since this
1637 * function is called from invalidate page, it's
1638 * harmless to return without any action.
1640 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1641 "ino %lu, to_free %d with only %d reserved "
1642 "data blocks", inode->i_ino, to_free,
1643 ei->i_reserved_data_blocks);
1645 to_free = ei->i_reserved_data_blocks;
1647 ei->i_reserved_data_blocks -= to_free;
1649 /* update fs dirty data blocks counter */
1650 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1652 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1654 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1657 static void ext4_da_page_release_reservation(struct page *page,
1658 unsigned int offset,
1659 unsigned int length)
1661 int to_release = 0, contiguous_blks = 0;
1662 struct buffer_head *head, *bh;
1663 unsigned int curr_off = 0;
1664 struct inode *inode = page->mapping->host;
1665 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1666 unsigned int stop = offset + length;
1670 BUG_ON(stop > PAGE_SIZE || stop < length);
1672 head = page_buffers(page);
1675 unsigned int next_off = curr_off + bh->b_size;
1677 if (next_off > stop)
1680 if ((offset <= curr_off) && (buffer_delay(bh))) {
1683 clear_buffer_delay(bh);
1684 } else if (contiguous_blks) {
1685 lblk = page->index <<
1686 (PAGE_SHIFT - inode->i_blkbits);
1687 lblk += (curr_off >> inode->i_blkbits) -
1689 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1690 contiguous_blks = 0;
1692 curr_off = next_off;
1693 } while ((bh = bh->b_this_page) != head);
1695 if (contiguous_blks) {
1696 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1697 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1698 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1701 /* If we have released all the blocks belonging to a cluster, then we
1702 * need to release the reserved space for that cluster. */
1703 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1704 while (num_clusters > 0) {
1705 lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1706 ((num_clusters - 1) << sbi->s_cluster_bits);
1707 if (sbi->s_cluster_ratio == 1 ||
1708 !ext4_find_delalloc_cluster(inode, lblk))
1709 ext4_da_release_space(inode, 1);
1716 * Delayed allocation stuff
1719 struct mpage_da_data {
1720 struct inode *inode;
1721 struct writeback_control *wbc;
1723 pgoff_t first_page; /* The first page to write */
1724 pgoff_t next_page; /* Current page to examine */
1725 pgoff_t last_page; /* Last page to examine */
1727 * Extent to map - this can be after first_page because that can be
1728 * fully mapped. We somewhat abuse m_flags to store whether the extent
1729 * is delalloc or unwritten.
1731 struct ext4_map_blocks map;
1732 struct ext4_io_submit io_submit; /* IO submission data */
1733 unsigned int do_map:1;
1736 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1741 struct pagevec pvec;
1742 struct inode *inode = mpd->inode;
1743 struct address_space *mapping = inode->i_mapping;
1745 /* This is necessary when next_page == 0. */
1746 if (mpd->first_page >= mpd->next_page)
1749 index = mpd->first_page;
1750 end = mpd->next_page - 1;
1752 ext4_lblk_t start, last;
1753 start = index << (PAGE_SHIFT - inode->i_blkbits);
1754 last = end << (PAGE_SHIFT - inode->i_blkbits);
1757 * avoid racing with extent status tree scans made by
1758 * ext4_insert_delayed_block()
1760 down_write(&EXT4_I(inode)->i_data_sem);
1761 ext4_es_remove_extent(inode, start, last - start + 1);
1762 up_write(&EXT4_I(inode)->i_data_sem);
1765 pagevec_init(&pvec, 0);
1766 while (index <= end) {
1767 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1770 for (i = 0; i < nr_pages; i++) {
1771 struct page *page = pvec.pages[i];
1773 BUG_ON(!PageLocked(page));
1774 BUG_ON(PageWriteback(page));
1776 if (page_mapped(page))
1777 clear_page_dirty_for_io(page);
1778 block_invalidatepage(page, 0, PAGE_SIZE);
1779 ClearPageUptodate(page);
1783 pagevec_release(&pvec);
1787 static void ext4_print_free_blocks(struct inode *inode)
1789 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1790 struct super_block *sb = inode->i_sb;
1791 struct ext4_inode_info *ei = EXT4_I(inode);
1793 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1794 EXT4_C2B(EXT4_SB(inode->i_sb),
1795 ext4_count_free_clusters(sb)));
1796 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1797 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1798 (long long) EXT4_C2B(EXT4_SB(sb),
1799 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1800 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1801 (long long) EXT4_C2B(EXT4_SB(sb),
1802 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1803 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1804 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1805 ei->i_reserved_data_blocks);
1809 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1811 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1815 * This function is grabs code from the very beginning of
1816 * ext4_map_blocks, but assumes that the caller is from delayed write
1817 * time. This function looks up the requested blocks and sets the
1818 * buffer delay bit under the protection of i_data_sem.
1820 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1821 struct ext4_map_blocks *map,
1822 struct buffer_head *bh)
1824 struct extent_status es;
1826 sector_t invalid_block = ~((sector_t) 0xffff);
1827 #ifdef ES_AGGRESSIVE_TEST
1828 struct ext4_map_blocks orig_map;
1830 memcpy(&orig_map, map, sizeof(*map));
1833 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1837 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1838 "logical block %lu\n", inode->i_ino, map->m_len,
1839 (unsigned long) map->m_lblk);
1841 /* Lookup extent status tree firstly */
1842 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1843 if (ext4_es_is_hole(&es)) {
1845 down_read(&EXT4_I(inode)->i_data_sem);
1850 * Delayed extent could be allocated by fallocate.
1851 * So we need to check it.
1853 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1854 map_bh(bh, inode->i_sb, invalid_block);
1856 set_buffer_delay(bh);
1860 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1861 retval = es.es_len - (iblock - es.es_lblk);
1862 if (retval > map->m_len)
1863 retval = map->m_len;
1864 map->m_len = retval;
1865 if (ext4_es_is_written(&es))
1866 map->m_flags |= EXT4_MAP_MAPPED;
1867 else if (ext4_es_is_unwritten(&es))
1868 map->m_flags |= EXT4_MAP_UNWRITTEN;
1872 #ifdef ES_AGGRESSIVE_TEST
1873 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1879 * Try to see if we can get the block without requesting a new
1880 * file system block.
1882 down_read(&EXT4_I(inode)->i_data_sem);
1883 if (ext4_has_inline_data(inode))
1885 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1886 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1888 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1894 * XXX: __block_prepare_write() unmaps passed block,
1898 * If the block was allocated from previously allocated cluster,
1899 * then we don't need to reserve it again. However we still need
1900 * to reserve metadata for every block we're going to write.
1902 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1903 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1904 ret = ext4_da_reserve_space(inode);
1906 /* not enough space to reserve */
1912 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1913 ~0, EXTENT_STATUS_DELAYED);
1919 map_bh(bh, inode->i_sb, invalid_block);
1921 set_buffer_delay(bh);
1922 } else if (retval > 0) {
1924 unsigned int status;
1926 if (unlikely(retval != map->m_len)) {
1927 ext4_warning(inode->i_sb,
1928 "ES len assertion failed for inode "
1929 "%lu: retval %d != map->m_len %d",
1930 inode->i_ino, retval, map->m_len);
1934 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1935 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1936 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1937 map->m_pblk, status);
1943 up_read((&EXT4_I(inode)->i_data_sem));
1949 * This is a special get_block_t callback which is used by
1950 * ext4_da_write_begin(). It will either return mapped block or
1951 * reserve space for a single block.
1953 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1954 * We also have b_blocknr = -1 and b_bdev initialized properly
1956 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1957 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1958 * initialized properly.
1960 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1961 struct buffer_head *bh, int create)
1963 struct ext4_map_blocks map;
1966 BUG_ON(create == 0);
1967 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1969 map.m_lblk = iblock;
1973 * first, we need to know whether the block is allocated already
1974 * preallocated blocks are unmapped but should treated
1975 * the same as allocated blocks.
1977 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1981 map_bh(bh, inode->i_sb, map.m_pblk);
1982 ext4_update_bh_state(bh, map.m_flags);
1984 if (buffer_unwritten(bh)) {
1985 /* A delayed write to unwritten bh should be marked
1986 * new and mapped. Mapped ensures that we don't do
1987 * get_block multiple times when we write to the same
1988 * offset and new ensures that we do proper zero out
1989 * for partial write.
1992 set_buffer_mapped(bh);
1997 static int bget_one(handle_t *handle, struct buffer_head *bh)
2003 static int bput_one(handle_t *handle, struct buffer_head *bh)
2009 static int __ext4_journalled_writepage(struct page *page,
2012 struct address_space *mapping = page->mapping;
2013 struct inode *inode = mapping->host;
2014 struct buffer_head *page_bufs = NULL;
2015 handle_t *handle = NULL;
2016 int ret = 0, err = 0;
2017 int inline_data = ext4_has_inline_data(inode);
2018 struct buffer_head *inode_bh = NULL;
2020 ClearPageChecked(page);
2023 BUG_ON(page->index != 0);
2024 BUG_ON(len > ext4_get_max_inline_size(inode));
2025 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2026 if (inode_bh == NULL)
2029 page_bufs = page_buffers(page);
2034 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2038 * We need to release the page lock before we start the
2039 * journal, so grab a reference so the page won't disappear
2040 * out from under us.
2045 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2046 ext4_writepage_trans_blocks(inode));
2047 if (IS_ERR(handle)) {
2048 ret = PTR_ERR(handle);
2050 goto out_no_pagelock;
2052 BUG_ON(!ext4_handle_valid(handle));
2056 if (page->mapping != mapping) {
2057 /* The page got truncated from under us */
2058 ext4_journal_stop(handle);
2064 ret = ext4_mark_inode_dirty(handle, inode);
2066 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2067 do_journal_get_write_access);
2069 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2074 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2075 err = ext4_journal_stop(handle);
2079 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2083 if (!inline_data && page_bufs)
2084 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2091 * Note that we don't need to start a transaction unless we're journaling data
2092 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2093 * need to file the inode to the transaction's list in ordered mode because if
2094 * we are writing back data added by write(), the inode is already there and if
2095 * we are writing back data modified via mmap(), no one guarantees in which
2096 * transaction the data will hit the disk. In case we are journaling data, we
2097 * cannot start transaction directly because transaction start ranks above page
2098 * lock so we have to do some magic.
2100 * This function can get called via...
2101 * - ext4_writepages after taking page lock (have journal handle)
2102 * - journal_submit_inode_data_buffers (no journal handle)
2103 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2104 * - grab_page_cache when doing write_begin (have journal handle)
2106 * We don't do any block allocation in this function. If we have page with
2107 * multiple blocks we need to write those buffer_heads that are mapped. This
2108 * is important for mmaped based write. So if we do with blocksize 1K
2109 * truncate(f, 1024);
2110 * a = mmap(f, 0, 4096);
2112 * truncate(f, 4096);
2113 * we have in the page first buffer_head mapped via page_mkwrite call back
2114 * but other buffer_heads would be unmapped but dirty (dirty done via the
2115 * do_wp_page). So writepage should write the first block. If we modify
2116 * the mmap area beyond 1024 we will again get a page_fault and the
2117 * page_mkwrite callback will do the block allocation and mark the
2118 * buffer_heads mapped.
2120 * We redirty the page if we have any buffer_heads that is either delay or
2121 * unwritten in the page.
2123 * We can get recursively called as show below.
2125 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2128 * But since we don't do any block allocation we should not deadlock.
2129 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2131 static int ext4_writepage(struct page *page,
2132 struct writeback_control *wbc)
2137 struct buffer_head *page_bufs = NULL;
2138 struct inode *inode = page->mapping->host;
2139 struct ext4_io_submit io_submit;
2140 bool keep_towrite = false;
2142 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2143 inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
2148 trace_ext4_writepage(page);
2149 size = i_size_read(inode);
2150 if (page->index == size >> PAGE_SHIFT)
2151 len = size & ~PAGE_MASK;
2155 /* Should never happen but for bugs in other kernel subsystems */
2156 if (!page_has_buffers(page)) {
2157 ext4_warning_inode(inode,
2158 "page %lu does not have buffers attached", page->index);
2159 ClearPageDirty(page);
2164 page_bufs = page_buffers(page);
2166 * We cannot do block allocation or other extent handling in this
2167 * function. If there are buffers needing that, we have to redirty
2168 * the page. But we may reach here when we do a journal commit via
2169 * journal_submit_inode_data_buffers() and in that case we must write
2170 * allocated buffers to achieve data=ordered mode guarantees.
2172 * Also, if there is only one buffer per page (the fs block
2173 * size == the page size), if one buffer needs block
2174 * allocation or needs to modify the extent tree to clear the
2175 * unwritten flag, we know that the page can't be written at
2176 * all, so we might as well refuse the write immediately.
2177 * Unfortunately if the block size != page size, we can't as
2178 * easily detect this case using ext4_walk_page_buffers(), but
2179 * for the extremely common case, this is an optimization that
2180 * skips a useless round trip through ext4_bio_write_page().
2182 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2183 ext4_bh_delay_or_unwritten)) {
2184 redirty_page_for_writepage(wbc, page);
2185 if ((current->flags & PF_MEMALLOC) ||
2186 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2188 * For memory cleaning there's no point in writing only
2189 * some buffers. So just bail out. Warn if we came here
2190 * from direct reclaim.
2192 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2197 keep_towrite = true;
2200 if (PageChecked(page) && ext4_should_journal_data(inode))
2202 * It's mmapped pagecache. Add buffers and journal it. There
2203 * doesn't seem much point in redirtying the page here.
2205 return __ext4_journalled_writepage(page, len);
2207 ext4_io_submit_init(&io_submit, wbc);
2208 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2209 if (!io_submit.io_end) {
2210 redirty_page_for_writepage(wbc, page);
2214 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2215 ext4_io_submit(&io_submit);
2216 /* Drop io_end reference we got from init */
2217 ext4_put_io_end_defer(io_submit.io_end);
2221 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2227 BUG_ON(page->index != mpd->first_page);
2228 clear_page_dirty_for_io(page);
2230 * We have to be very careful here! Nothing protects writeback path
2231 * against i_size changes and the page can be writeably mapped into
2232 * page tables. So an application can be growing i_size and writing
2233 * data through mmap while writeback runs. clear_page_dirty_for_io()
2234 * write-protects our page in page tables and the page cannot get
2235 * written to again until we release page lock. So only after
2236 * clear_page_dirty_for_io() we are safe to sample i_size for
2237 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2238 * on the barrier provided by TestClearPageDirty in
2239 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2240 * after page tables are updated.
2242 size = i_size_read(mpd->inode);
2243 if (page->index == size >> PAGE_SHIFT)
2244 len = size & ~PAGE_MASK;
2247 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2249 mpd->wbc->nr_to_write--;
2255 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2258 * mballoc gives us at most this number of blocks...
2259 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2260 * The rest of mballoc seems to handle chunks up to full group size.
2262 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2265 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2267 * @mpd - extent of blocks
2268 * @lblk - logical number of the block in the file
2269 * @bh - buffer head we want to add to the extent
2271 * The function is used to collect contig. blocks in the same state. If the
2272 * buffer doesn't require mapping for writeback and we haven't started the
2273 * extent of buffers to map yet, the function returns 'true' immediately - the
2274 * caller can write the buffer right away. Otherwise the function returns true
2275 * if the block has been added to the extent, false if the block couldn't be
2278 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2279 struct buffer_head *bh)
2281 struct ext4_map_blocks *map = &mpd->map;
2283 /* Buffer that doesn't need mapping for writeback? */
2284 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2285 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2286 /* So far no extent to map => we write the buffer right away */
2287 if (map->m_len == 0)
2292 /* First block in the extent? */
2293 if (map->m_len == 0) {
2294 /* We cannot map unless handle is started... */
2299 map->m_flags = bh->b_state & BH_FLAGS;
2303 /* Don't go larger than mballoc is willing to allocate */
2304 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2307 /* Can we merge the block to our big extent? */
2308 if (lblk == map->m_lblk + map->m_len &&
2309 (bh->b_state & BH_FLAGS) == map->m_flags) {
2317 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2319 * @mpd - extent of blocks for mapping
2320 * @head - the first buffer in the page
2321 * @bh - buffer we should start processing from
2322 * @lblk - logical number of the block in the file corresponding to @bh
2324 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2325 * the page for IO if all buffers in this page were mapped and there's no
2326 * accumulated extent of buffers to map or add buffers in the page to the
2327 * extent of buffers to map. The function returns 1 if the caller can continue
2328 * by processing the next page, 0 if it should stop adding buffers to the
2329 * extent to map because we cannot extend it anymore. It can also return value
2330 * < 0 in case of error during IO submission.
2332 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2333 struct buffer_head *head,
2334 struct buffer_head *bh,
2337 struct inode *inode = mpd->inode;
2339 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2340 >> inode->i_blkbits;
2343 BUG_ON(buffer_locked(bh));
2345 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2346 /* Found extent to map? */
2349 /* Buffer needs mapping and handle is not started? */
2352 /* Everything mapped so far and we hit EOF */
2355 } while (lblk++, (bh = bh->b_this_page) != head);
2356 /* So far everything mapped? Submit the page for IO. */
2357 if (mpd->map.m_len == 0) {
2358 err = mpage_submit_page(mpd, head->b_page);
2362 return lblk < blocks;
2366 * mpage_map_buffers - update buffers corresponding to changed extent and
2367 * submit fully mapped pages for IO
2369 * @mpd - description of extent to map, on return next extent to map
2371 * Scan buffers corresponding to changed extent (we expect corresponding pages
2372 * to be already locked) and update buffer state according to new extent state.
2373 * We map delalloc buffers to their physical location, clear unwritten bits,
2374 * and mark buffers as uninit when we perform writes to unwritten extents
2375 * and do extent conversion after IO is finished. If the last page is not fully
2376 * mapped, we update @map to the next extent in the last page that needs
2377 * mapping. Otherwise we submit the page for IO.
2379 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2381 struct pagevec pvec;
2383 struct inode *inode = mpd->inode;
2384 struct buffer_head *head, *bh;
2385 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2391 start = mpd->map.m_lblk >> bpp_bits;
2392 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2393 lblk = start << bpp_bits;
2394 pblock = mpd->map.m_pblk;
2396 pagevec_init(&pvec, 0);
2397 while (start <= end) {
2398 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2402 for (i = 0; i < nr_pages; i++) {
2403 struct page *page = pvec.pages[i];
2405 bh = head = page_buffers(page);
2407 if (lblk < mpd->map.m_lblk)
2409 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2411 * Buffer after end of mapped extent.
2412 * Find next buffer in the page to map.
2415 mpd->map.m_flags = 0;
2417 * FIXME: If dioread_nolock supports
2418 * blocksize < pagesize, we need to make
2419 * sure we add size mapped so far to
2420 * io_end->size as the following call
2421 * can submit the page for IO.
2423 err = mpage_process_page_bufs(mpd, head,
2425 pagevec_release(&pvec);
2430 if (buffer_delay(bh)) {
2431 clear_buffer_delay(bh);
2432 bh->b_blocknr = pblock++;
2434 clear_buffer_unwritten(bh);
2435 } while (lblk++, (bh = bh->b_this_page) != head);
2438 * FIXME: This is going to break if dioread_nolock
2439 * supports blocksize < pagesize as we will try to
2440 * convert potentially unmapped parts of inode.
2442 mpd->io_submit.io_end->size += PAGE_SIZE;
2443 /* Page fully mapped - let IO run! */
2444 err = mpage_submit_page(mpd, page);
2446 pagevec_release(&pvec);
2450 pagevec_release(&pvec);
2452 /* Extent fully mapped and matches with page boundary. We are done. */
2454 mpd->map.m_flags = 0;
2458 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2460 struct inode *inode = mpd->inode;
2461 struct ext4_map_blocks *map = &mpd->map;
2462 int get_blocks_flags;
2463 int err, dioread_nolock;
2465 trace_ext4_da_write_pages_extent(inode, map);
2467 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2468 * to convert an unwritten extent to be initialized (in the case
2469 * where we have written into one or more preallocated blocks). It is
2470 * possible that we're going to need more metadata blocks than
2471 * previously reserved. However we must not fail because we're in
2472 * writeback and there is nothing we can do about it so it might result
2473 * in data loss. So use reserved blocks to allocate metadata if
2476 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2477 * the blocks in question are delalloc blocks. This indicates
2478 * that the blocks and quotas has already been checked when
2479 * the data was copied into the page cache.
2481 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2482 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2483 EXT4_GET_BLOCKS_IO_SUBMIT;
2484 dioread_nolock = ext4_should_dioread_nolock(inode);
2486 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2487 if (map->m_flags & (1 << BH_Delay))
2488 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2490 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2493 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2494 if (!mpd->io_submit.io_end->handle &&
2495 ext4_handle_valid(handle)) {
2496 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2497 handle->h_rsv_handle = NULL;
2499 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2502 BUG_ON(map->m_len == 0);
2503 if (map->m_flags & EXT4_MAP_NEW) {
2504 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
2511 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2512 * mpd->len and submit pages underlying it for IO
2514 * @handle - handle for journal operations
2515 * @mpd - extent to map
2516 * @give_up_on_write - we set this to true iff there is a fatal error and there
2517 * is no hope of writing the data. The caller should discard
2518 * dirty pages to avoid infinite loops.
2520 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2521 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2522 * them to initialized or split the described range from larger unwritten
2523 * extent. Note that we need not map all the described range since allocation
2524 * can return less blocks or the range is covered by more unwritten extents. We
2525 * cannot map more because we are limited by reserved transaction credits. On
2526 * the other hand we always make sure that the last touched page is fully
2527 * mapped so that it can be written out (and thus forward progress is
2528 * guaranteed). After mapping we submit all mapped pages for IO.
2530 static int mpage_map_and_submit_extent(handle_t *handle,
2531 struct mpage_da_data *mpd,
2532 bool *give_up_on_write)
2534 struct inode *inode = mpd->inode;
2535 struct ext4_map_blocks *map = &mpd->map;
2540 mpd->io_submit.io_end->offset =
2541 ((loff_t)map->m_lblk) << inode->i_blkbits;
2543 err = mpage_map_one_extent(handle, mpd);
2545 struct super_block *sb = inode->i_sb;
2547 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2548 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2549 goto invalidate_dirty_pages;
2551 * Let the uper layers retry transient errors.
2552 * In the case of ENOSPC, if ext4_count_free_blocks()
2553 * is non-zero, a commit should free up blocks.
2555 if ((err == -ENOMEM) ||
2556 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2558 goto update_disksize;
2561 ext4_msg(sb, KERN_CRIT,
2562 "Delayed block allocation failed for "
2563 "inode %lu at logical offset %llu with"
2564 " max blocks %u with error %d",
2566 (unsigned long long)map->m_lblk,
2567 (unsigned)map->m_len, -err);
2568 ext4_msg(sb, KERN_CRIT,
2569 "This should not happen!! Data will "
2572 ext4_print_free_blocks(inode);
2573 invalidate_dirty_pages:
2574 *give_up_on_write = true;
2579 * Update buffer state, submit mapped pages, and get us new
2582 err = mpage_map_and_submit_buffers(mpd);
2584 goto update_disksize;
2585 } while (map->m_len);
2589 * Update on-disk size after IO is submitted. Races with
2590 * truncate are avoided by checking i_size under i_data_sem.
2592 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2593 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2597 down_write(&EXT4_I(inode)->i_data_sem);
2598 i_size = i_size_read(inode);
2599 if (disksize > i_size)
2601 if (disksize > EXT4_I(inode)->i_disksize)
2602 EXT4_I(inode)->i_disksize = disksize;
2603 up_write(&EXT4_I(inode)->i_data_sem);
2604 err2 = ext4_mark_inode_dirty(handle, inode);
2606 ext4_error(inode->i_sb,
2607 "Failed to mark inode %lu dirty",
2616 * Calculate the total number of credits to reserve for one writepages
2617 * iteration. This is called from ext4_writepages(). We map an extent of
2618 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2619 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2620 * bpp - 1 blocks in bpp different extents.
2622 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2624 int bpp = ext4_journal_blocks_per_page(inode);
2626 return ext4_meta_trans_blocks(inode,
2627 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2631 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2632 * and underlying extent to map
2634 * @mpd - where to look for pages
2636 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2637 * IO immediately. When we find a page which isn't mapped we start accumulating
2638 * extent of buffers underlying these pages that needs mapping (formed by
2639 * either delayed or unwritten buffers). We also lock the pages containing
2640 * these buffers. The extent found is returned in @mpd structure (starting at
2641 * mpd->lblk with length mpd->len blocks).
2643 * Note that this function can attach bios to one io_end structure which are
2644 * neither logically nor physically contiguous. Although it may seem as an
2645 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2646 * case as we need to track IO to all buffers underlying a page in one io_end.
2648 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2650 struct address_space *mapping = mpd->inode->i_mapping;
2651 struct pagevec pvec;
2652 unsigned int nr_pages;
2653 long left = mpd->wbc->nr_to_write;
2654 pgoff_t index = mpd->first_page;
2655 pgoff_t end = mpd->last_page;
2658 int blkbits = mpd->inode->i_blkbits;
2660 struct buffer_head *head;
2662 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2663 tag = PAGECACHE_TAG_TOWRITE;
2665 tag = PAGECACHE_TAG_DIRTY;
2667 pagevec_init(&pvec, 0);
2669 mpd->next_page = index;
2670 while (index <= end) {
2671 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2672 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2676 for (i = 0; i < nr_pages; i++) {
2677 struct page *page = pvec.pages[i];
2680 * At this point, the page may be truncated or
2681 * invalidated (changing page->mapping to NULL), or
2682 * even swizzled back from swapper_space to tmpfs file
2683 * mapping. However, page->index will not change
2684 * because we have a reference on the page.
2686 if (page->index > end)
2690 * Accumulated enough dirty pages? This doesn't apply
2691 * to WB_SYNC_ALL mode. For integrity sync we have to
2692 * keep going because someone may be concurrently
2693 * dirtying pages, and we might have synced a lot of
2694 * newly appeared dirty pages, but have not synced all
2695 * of the old dirty pages.
2697 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2700 /* If we can't merge this page, we are done. */
2701 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2706 * If the page is no longer dirty, or its mapping no
2707 * longer corresponds to inode we are writing (which
2708 * means it has been truncated or invalidated), or the
2709 * page is already under writeback and we are not doing
2710 * a data integrity writeback, skip the page
2712 if (!PageDirty(page) ||
2713 (PageWriteback(page) &&
2714 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2715 unlikely(page->mapping != mapping)) {
2720 wait_on_page_writeback(page);
2721 BUG_ON(PageWriteback(page));
2724 * Should never happen but for buggy code in
2725 * other subsystems that call
2726 * set_page_dirty() without properly warning
2727 * the file system first. See [1] for more
2730 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2732 if (!page_has_buffers(page)) {
2733 ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", page->index);
2734 ClearPageDirty(page);
2739 if (mpd->map.m_len == 0)
2740 mpd->first_page = page->index;
2741 mpd->next_page = page->index + 1;
2742 /* Add all dirty buffers to mpd */
2743 lblk = ((ext4_lblk_t)page->index) <<
2744 (PAGE_SHIFT - blkbits);
2745 head = page_buffers(page);
2746 err = mpage_process_page_bufs(mpd, head, head, lblk);
2752 pagevec_release(&pvec);
2757 pagevec_release(&pvec);
2761 static int __writepage(struct page *page, struct writeback_control *wbc,
2764 struct address_space *mapping = data;
2765 int ret = ext4_writepage(page, wbc);
2766 mapping_set_error(mapping, ret);
2770 static int ext4_writepages(struct address_space *mapping,
2771 struct writeback_control *wbc)
2773 pgoff_t writeback_index = 0;
2774 long nr_to_write = wbc->nr_to_write;
2775 int range_whole = 0;
2777 handle_t *handle = NULL;
2778 struct mpage_da_data mpd;
2779 struct inode *inode = mapping->host;
2780 int needed_blocks, rsv_blocks = 0, ret = 0;
2781 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2783 struct blk_plug plug;
2784 bool give_up_on_write = false;
2786 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2789 percpu_down_read(&sbi->s_writepages_rwsem);
2790 trace_ext4_writepages(inode, wbc);
2792 if (dax_mapping(mapping)) {
2793 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev,
2795 goto out_writepages;
2799 * No pages to write? This is mainly a kludge to avoid starting
2800 * a transaction for special inodes like journal inode on last iput()
2801 * because that could violate lock ordering on umount
2803 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2804 goto out_writepages;
2806 if (ext4_should_journal_data(inode)) {
2807 struct blk_plug plug;
2809 blk_start_plug(&plug);
2810 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2811 blk_finish_plug(&plug);
2812 goto out_writepages;
2816 * If the filesystem has aborted, it is read-only, so return
2817 * right away instead of dumping stack traces later on that
2818 * will obscure the real source of the problem. We test
2819 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2820 * the latter could be true if the filesystem is mounted
2821 * read-only, and in that case, ext4_writepages should
2822 * *never* be called, so if that ever happens, we would want
2825 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2826 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2828 goto out_writepages;
2831 if (ext4_should_dioread_nolock(inode)) {
2833 * We may need to convert up to one extent per block in
2834 * the page and we may dirty the inode.
2836 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2837 PAGE_SIZE >> inode->i_blkbits);
2841 * If we have inline data and arrive here, it means that
2842 * we will soon create the block for the 1st page, so
2843 * we'd better clear the inline data here.
2845 if (ext4_has_inline_data(inode)) {
2846 /* Just inode will be modified... */
2847 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2848 if (IS_ERR(handle)) {
2849 ret = PTR_ERR(handle);
2850 goto out_writepages;
2852 BUG_ON(ext4_test_inode_state(inode,
2853 EXT4_STATE_MAY_INLINE_DATA));
2854 ext4_destroy_inline_data(handle, inode);
2855 ext4_journal_stop(handle);
2858 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2861 if (wbc->range_cyclic) {
2862 writeback_index = mapping->writeback_index;
2863 if (writeback_index)
2865 mpd.first_page = writeback_index;
2868 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2869 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2874 ext4_io_submit_init(&mpd.io_submit, wbc);
2876 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2877 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2879 blk_start_plug(&plug);
2882 * First writeback pages that don't need mapping - we can avoid
2883 * starting a transaction unnecessarily and also avoid being blocked
2884 * in the block layer on device congestion while having transaction
2888 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2889 if (!mpd.io_submit.io_end) {
2893 ret = mpage_prepare_extent_to_map(&mpd);
2894 /* Submit prepared bio */
2895 ext4_io_submit(&mpd.io_submit);
2896 ext4_put_io_end_defer(mpd.io_submit.io_end);
2897 mpd.io_submit.io_end = NULL;
2898 /* Unlock pages we didn't use */
2899 mpage_release_unused_pages(&mpd, false);
2903 while (!done && mpd.first_page <= mpd.last_page) {
2904 /* For each extent of pages we use new io_end */
2905 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2906 if (!mpd.io_submit.io_end) {
2912 * We have two constraints: We find one extent to map and we
2913 * must always write out whole page (makes a difference when
2914 * blocksize < pagesize) so that we don't block on IO when we
2915 * try to write out the rest of the page. Journalled mode is
2916 * not supported by delalloc.
2918 BUG_ON(ext4_should_journal_data(inode));
2919 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2921 /* start a new transaction */
2922 handle = ext4_journal_start_with_reserve(inode,
2923 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2924 if (IS_ERR(handle)) {
2925 ret = PTR_ERR(handle);
2926 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2927 "%ld pages, ino %lu; err %d", __func__,
2928 wbc->nr_to_write, inode->i_ino, ret);
2929 /* Release allocated io_end */
2930 ext4_put_io_end(mpd.io_submit.io_end);
2931 mpd.io_submit.io_end = NULL;
2936 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2937 ret = mpage_prepare_extent_to_map(&mpd);
2940 ret = mpage_map_and_submit_extent(handle, &mpd,
2944 * We scanned the whole range (or exhausted
2945 * nr_to_write), submitted what was mapped and
2946 * didn't find anything needing mapping. We are
2953 * Caution: If the handle is synchronous,
2954 * ext4_journal_stop() can wait for transaction commit
2955 * to finish which may depend on writeback of pages to
2956 * complete or on page lock to be released. In that
2957 * case, we have to wait until after after we have
2958 * submitted all the IO, released page locks we hold,
2959 * and dropped io_end reference (for extent conversion
2960 * to be able to complete) before stopping the handle.
2962 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2963 ext4_journal_stop(handle);
2967 /* Submit prepared bio */
2968 ext4_io_submit(&mpd.io_submit);
2969 /* Unlock pages we didn't use */
2970 mpage_release_unused_pages(&mpd, give_up_on_write);
2972 * Drop our io_end reference we got from init. We have
2973 * to be careful and use deferred io_end finishing if
2974 * we are still holding the transaction as we can
2975 * release the last reference to io_end which may end
2976 * up doing unwritten extent conversion.
2979 ext4_put_io_end_defer(mpd.io_submit.io_end);
2980 ext4_journal_stop(handle);
2982 ext4_put_io_end(mpd.io_submit.io_end);
2983 mpd.io_submit.io_end = NULL;
2985 if (ret == -ENOSPC && sbi->s_journal) {
2987 * Commit the transaction which would
2988 * free blocks released in the transaction
2991 jbd2_journal_force_commit_nested(sbi->s_journal);
2995 /* Fatal error - ENOMEM, EIO... */
3000 blk_finish_plug(&plug);
3001 if (!ret && !cycled && wbc->nr_to_write > 0) {
3003 mpd.last_page = writeback_index - 1;
3009 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3011 * Set the writeback_index so that range_cyclic
3012 * mode will write it back later
3014 mapping->writeback_index = mpd.first_page;
3017 trace_ext4_writepages_result(inode, wbc, ret,
3018 nr_to_write - wbc->nr_to_write);
3019 percpu_up_read(&sbi->s_writepages_rwsem);
3023 static int ext4_nonda_switch(struct super_block *sb)
3025 s64 free_clusters, dirty_clusters;
3026 struct ext4_sb_info *sbi = EXT4_SB(sb);
3029 * switch to non delalloc mode if we are running low
3030 * on free block. The free block accounting via percpu
3031 * counters can get slightly wrong with percpu_counter_batch getting
3032 * accumulated on each CPU without updating global counters
3033 * Delalloc need an accurate free block accounting. So switch
3034 * to non delalloc when we are near to error range.
3037 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
3039 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
3041 * Start pushing delalloc when 1/2 of free blocks are dirty.
3043 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
3044 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
3046 if (2 * free_clusters < 3 * dirty_clusters ||
3047 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
3049 * free block count is less than 150% of dirty blocks
3050 * or free blocks is less than watermark
3057 /* We always reserve for an inode update; the superblock could be there too */
3058 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
3060 if (likely(ext4_has_feature_large_file(inode->i_sb)))
3063 if (pos + len <= 0x7fffffffULL)
3066 /* We might need to update the superblock to set LARGE_FILE */
3070 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3071 loff_t pos, unsigned len, unsigned flags,
3072 struct page **pagep, void **fsdata)
3074 int ret, retries = 0;
3077 struct inode *inode = mapping->host;
3080 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3083 index = pos >> PAGE_SHIFT;
3085 if (ext4_nonda_switch(inode->i_sb) ||
3086 S_ISLNK(inode->i_mode)) {
3087 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3088 return ext4_write_begin(file, mapping, pos,
3089 len, flags, pagep, fsdata);
3091 *fsdata = (void *)0;
3092 trace_ext4_da_write_begin(inode, pos, len, flags);
3094 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3095 ret = ext4_da_write_inline_data_begin(mapping, inode,
3105 * grab_cache_page_write_begin() can take a long time if the
3106 * system is thrashing due to memory pressure, or if the page
3107 * is being written back. So grab it first before we start
3108 * the transaction handle. This also allows us to allocate
3109 * the page (if needed) without using GFP_NOFS.
3112 page = grab_cache_page_write_begin(mapping, index, flags);
3118 * With delayed allocation, we don't log the i_disksize update
3119 * if there is delayed block allocation. But we still need
3120 * to journalling the i_disksize update if writes to the end
3121 * of file which has an already mapped buffer.
3124 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3125 ext4_da_write_credits(inode, pos, len));
3126 if (IS_ERR(handle)) {
3128 return PTR_ERR(handle);
3132 if (page->mapping != mapping) {
3133 /* The page got truncated from under us */
3136 ext4_journal_stop(handle);
3139 /* In case writeback began while the page was unlocked */
3140 wait_for_stable_page(page);
3142 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3143 ret = ext4_block_write_begin(page, pos, len,
3144 ext4_da_get_block_prep);
3146 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3150 ext4_journal_stop(handle);
3152 * block_write_begin may have instantiated a few blocks
3153 * outside i_size. Trim these off again. Don't need
3154 * i_size_read because we hold i_mutex.
3156 if (pos + len > inode->i_size)
3157 ext4_truncate_failed_write(inode);
3159 if (ret == -ENOSPC &&
3160 ext4_should_retry_alloc(inode->i_sb, &retries))
3172 * Check if we should update i_disksize
3173 * when write to the end of file but not require block allocation
3175 static int ext4_da_should_update_i_disksize(struct page *page,
3176 unsigned long offset)
3178 struct buffer_head *bh;
3179 struct inode *inode = page->mapping->host;
3183 bh = page_buffers(page);
3184 idx = offset >> inode->i_blkbits;
3186 for (i = 0; i < idx; i++)
3187 bh = bh->b_this_page;
3189 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3194 static int ext4_da_write_end(struct file *file,
3195 struct address_space *mapping,
3196 loff_t pos, unsigned len, unsigned copied,
3197 struct page *page, void *fsdata)
3199 struct inode *inode = mapping->host;
3201 handle_t *handle = ext4_journal_current_handle();
3203 unsigned long start, end;
3204 int write_mode = (int)(unsigned long)fsdata;
3206 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3207 return ext4_write_end(file, mapping, pos,
3208 len, copied, page, fsdata);
3210 trace_ext4_da_write_end(inode, pos, len, copied);
3211 start = pos & (PAGE_SIZE - 1);
3212 end = start + copied - 1;
3215 * generic_write_end() will run mark_inode_dirty() if i_size
3216 * changes. So let's piggyback the i_disksize mark_inode_dirty
3219 new_i_size = pos + copied;
3220 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3221 if (ext4_has_inline_data(inode) ||
3222 ext4_da_should_update_i_disksize(page, end)) {
3223 ext4_update_i_disksize(inode, new_i_size);
3224 /* We need to mark inode dirty even if
3225 * new_i_size is less that inode->i_size
3226 * bu greater than i_disksize.(hint delalloc)
3228 ext4_mark_inode_dirty(handle, inode);
3232 if (write_mode != CONVERT_INLINE_DATA &&
3233 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3234 ext4_has_inline_data(inode))
3235 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3238 ret2 = generic_write_end(file, mapping, pos, len, copied,
3244 ret2 = ext4_journal_stop(handle);
3248 return ret ? ret : copied;
3251 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3252 unsigned int length)
3255 * Drop reserved blocks
3257 BUG_ON(!PageLocked(page));
3258 if (!page_has_buffers(page))
3261 ext4_da_page_release_reservation(page, offset, length);
3264 ext4_invalidatepage(page, offset, length);
3270 * Force all delayed allocation blocks to be allocated for a given inode.
3272 int ext4_alloc_da_blocks(struct inode *inode)
3274 trace_ext4_alloc_da_blocks(inode);
3276 if (!EXT4_I(inode)->i_reserved_data_blocks)
3280 * We do something simple for now. The filemap_flush() will
3281 * also start triggering a write of the data blocks, which is
3282 * not strictly speaking necessary (and for users of
3283 * laptop_mode, not even desirable). However, to do otherwise
3284 * would require replicating code paths in:
3286 * ext4_writepages() ->
3287 * write_cache_pages() ---> (via passed in callback function)
3288 * __mpage_da_writepage() -->
3289 * mpage_add_bh_to_extent()
3290 * mpage_da_map_blocks()
3292 * The problem is that write_cache_pages(), located in
3293 * mm/page-writeback.c, marks pages clean in preparation for
3294 * doing I/O, which is not desirable if we're not planning on
3297 * We could call write_cache_pages(), and then redirty all of
3298 * the pages by calling redirty_page_for_writepage() but that
3299 * would be ugly in the extreme. So instead we would need to
3300 * replicate parts of the code in the above functions,
3301 * simplifying them because we wouldn't actually intend to
3302 * write out the pages, but rather only collect contiguous
3303 * logical block extents, call the multi-block allocator, and
3304 * then update the buffer heads with the block allocations.
3306 * For now, though, we'll cheat by calling filemap_flush(),
3307 * which will map the blocks, and start the I/O, but not
3308 * actually wait for the I/O to complete.
3310 return filemap_flush(inode->i_mapping);
3314 * bmap() is special. It gets used by applications such as lilo and by
3315 * the swapper to find the on-disk block of a specific piece of data.
3317 * Naturally, this is dangerous if the block concerned is still in the
3318 * journal. If somebody makes a swapfile on an ext4 data-journaling
3319 * filesystem and enables swap, then they may get a nasty shock when the
3320 * data getting swapped to that swapfile suddenly gets overwritten by
3321 * the original zero's written out previously to the journal and
3322 * awaiting writeback in the kernel's buffer cache.
3324 * So, if we see any bmap calls here on a modified, data-journaled file,
3325 * take extra steps to flush any blocks which might be in the cache.
3327 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3329 struct inode *inode = mapping->host;
3334 * We can get here for an inline file via the FIBMAP ioctl
3336 if (ext4_has_inline_data(inode))
3339 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3340 test_opt(inode->i_sb, DELALLOC)) {
3342 * With delalloc we want to sync the file
3343 * so that we can make sure we allocate
3346 filemap_write_and_wait(mapping);
3349 if (EXT4_JOURNAL(inode) &&
3350 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3352 * This is a REALLY heavyweight approach, but the use of
3353 * bmap on dirty files is expected to be extremely rare:
3354 * only if we run lilo or swapon on a freshly made file
3355 * do we expect this to happen.
3357 * (bmap requires CAP_SYS_RAWIO so this does not
3358 * represent an unprivileged user DOS attack --- we'd be
3359 * in trouble if mortal users could trigger this path at
3362 * NB. EXT4_STATE_JDATA is not set on files other than
3363 * regular files. If somebody wants to bmap a directory
3364 * or symlink and gets confused because the buffer
3365 * hasn't yet been flushed to disk, they deserve
3366 * everything they get.
3369 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3370 journal = EXT4_JOURNAL(inode);
3371 jbd2_journal_lock_updates(journal);
3372 err = jbd2_journal_flush(journal);
3373 jbd2_journal_unlock_updates(journal);
3379 return generic_block_bmap(mapping, block, ext4_get_block);
3382 static int ext4_readpage(struct file *file, struct page *page)
3385 struct inode *inode = page->mapping->host;
3387 trace_ext4_readpage(page);
3389 if (ext4_has_inline_data(inode))
3390 ret = ext4_readpage_inline(inode, page);
3393 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3399 ext4_readpages(struct file *file, struct address_space *mapping,
3400 struct list_head *pages, unsigned nr_pages)
3402 struct inode *inode = mapping->host;
3404 /* If the file has inline data, no need to do readpages. */
3405 if (ext4_has_inline_data(inode))
3408 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3411 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3412 unsigned int length)
3414 trace_ext4_invalidatepage(page, offset, length);
3416 /* No journalling happens on data buffers when this function is used */
3417 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3419 block_invalidatepage(page, offset, length);
3422 static int __ext4_journalled_invalidatepage(struct page *page,
3423 unsigned int offset,
3424 unsigned int length)
3426 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3428 trace_ext4_journalled_invalidatepage(page, offset, length);
3431 * If it's a full truncate we just forget about the pending dirtying
3433 if (offset == 0 && length == PAGE_SIZE)
3434 ClearPageChecked(page);
3436 return jbd2_journal_invalidatepage(journal, page, offset, length);
3439 /* Wrapper for aops... */
3440 static void ext4_journalled_invalidatepage(struct page *page,
3441 unsigned int offset,
3442 unsigned int length)
3444 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3447 static int ext4_releasepage(struct page *page, gfp_t wait)
3449 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3451 trace_ext4_releasepage(page);
3453 /* Page has dirty journalled data -> cannot release */
3454 if (PageChecked(page))
3457 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3459 return try_to_free_buffers(page);
3462 #ifdef CONFIG_FS_DAX
3463 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3464 unsigned flags, struct iomap *iomap)
3466 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3467 unsigned int blkbits = inode->i_blkbits;
3468 unsigned long first_block, last_block;
3469 struct ext4_map_blocks map;
3472 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3474 first_block = offset >> blkbits;
3475 last_block = min_t(loff_t, (offset + length - 1) >> blkbits,
3476 EXT4_MAX_LOGICAL_BLOCK);
3478 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3481 map.m_lblk = first_block;
3482 map.m_len = last_block - first_block + 1;
3484 if (!(flags & IOMAP_WRITE)) {
3485 ret = ext4_map_blocks(NULL, inode, &map, 0);
3491 /* Trim mapping request to maximum we can map at once for DIO */
3492 if (map.m_len > DIO_MAX_BLOCKS)
3493 map.m_len = DIO_MAX_BLOCKS;
3494 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3497 * Either we allocate blocks and then we don't get unwritten
3498 * extent so we have reserved enough credits, or the blocks
3499 * are already allocated and unwritten and in that case
3500 * extent conversion fits in the credits as well.
3502 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3505 return PTR_ERR(handle);
3507 ret = ext4_map_blocks(handle, inode, &map,
3508 EXT4_GET_BLOCKS_CREATE_ZERO);
3510 ext4_journal_stop(handle);
3511 if (ret == -ENOSPC &&
3512 ext4_should_retry_alloc(inode->i_sb, &retries))
3518 * If we added blocks beyond i_size, we need to make sure they
3519 * will get truncated if we crash before updating i_size in
3520 * ext4_iomap_end(). For faults we don't need to do that (and
3521 * even cannot because for orphan list operations inode_lock is
3522 * required) - if we happen to instantiate block beyond i_size,
3523 * it is because we race with truncate which has already added
3524 * the inode to the orphan list.
3526 if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3527 (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3530 err = ext4_orphan_add(handle, inode);
3532 ext4_journal_stop(handle);
3536 ext4_journal_stop(handle);
3540 iomap->bdev = inode->i_sb->s_bdev;
3541 iomap->dax_dev = sbi->s_daxdev;
3542 iomap->offset = first_block << blkbits;
3545 iomap->type = IOMAP_HOLE;
3546 iomap->blkno = IOMAP_NULL_BLOCK;
3547 iomap->length = (u64)map.m_len << blkbits;
3549 if (map.m_flags & EXT4_MAP_MAPPED) {
3550 iomap->type = IOMAP_MAPPED;
3551 } else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3552 iomap->type = IOMAP_UNWRITTEN;
3557 iomap->blkno = (sector_t)map.m_pblk << (blkbits - 9);
3558 iomap->length = (u64)map.m_len << blkbits;
3561 if (map.m_flags & EXT4_MAP_NEW)
3562 iomap->flags |= IOMAP_F_NEW;
3566 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3567 ssize_t written, unsigned flags, struct iomap *iomap)
3571 int blkbits = inode->i_blkbits;
3572 bool truncate = false;
3574 if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3577 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3578 if (IS_ERR(handle)) {
3579 ret = PTR_ERR(handle);
3582 if (ext4_update_inode_size(inode, offset + written))
3583 ext4_mark_inode_dirty(handle, inode);
3585 * We may need to truncate allocated but not written blocks beyond EOF.
3587 if (iomap->offset + iomap->length >
3588 ALIGN(inode->i_size, 1 << blkbits)) {
3589 ext4_lblk_t written_blk, end_blk;
3591 written_blk = (offset + written) >> blkbits;
3592 end_blk = (offset + length) >> blkbits;
3593 if (written_blk < end_blk && ext4_can_truncate(inode))
3597 * Remove inode from orphan list if we were extending a inode and
3598 * everything went fine.
3600 if (!truncate && inode->i_nlink &&
3601 !list_empty(&EXT4_I(inode)->i_orphan))
3602 ext4_orphan_del(handle, inode);
3603 ext4_journal_stop(handle);
3605 ext4_truncate_failed_write(inode);
3608 * If truncate failed early the inode might still be on the
3609 * orphan list; we need to make sure the inode is removed from
3610 * the orphan list in that case.
3613 ext4_orphan_del(NULL, inode);
3618 const struct iomap_ops ext4_iomap_ops = {
3619 .iomap_begin = ext4_iomap_begin,
3620 .iomap_end = ext4_iomap_end,
3625 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3626 ssize_t size, void *private)
3628 ext4_io_end_t *io_end = private;
3630 /* if not async direct IO just return */
3634 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3635 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3636 io_end, io_end->inode->i_ino, iocb, offset, size);
3639 * Error during AIO DIO. We cannot convert unwritten extents as the
3640 * data was not written. Just clear the unwritten flag and drop io_end.
3643 ext4_clear_io_unwritten_flag(io_end);
3646 io_end->offset = offset;
3647 io_end->size = size;
3648 ext4_put_io_end(io_end);
3654 * Handling of direct IO writes.
3656 * For ext4 extent files, ext4 will do direct-io write even to holes,
3657 * preallocated extents, and those write extend the file, no need to
3658 * fall back to buffered IO.
3660 * For holes, we fallocate those blocks, mark them as unwritten
3661 * If those blocks were preallocated, we mark sure they are split, but
3662 * still keep the range to write as unwritten.
3664 * The unwritten extents will be converted to written when DIO is completed.
3665 * For async direct IO, since the IO may still pending when return, we
3666 * set up an end_io call back function, which will do the conversion
3667 * when async direct IO completed.
3669 * If the O_DIRECT write will extend the file then add this inode to the
3670 * orphan list. So recovery will truncate it back to the original size
3671 * if the machine crashes during the write.
3674 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3676 struct file *file = iocb->ki_filp;
3677 struct inode *inode = file->f_mapping->host;
3679 loff_t offset = iocb->ki_pos;
3680 size_t count = iov_iter_count(iter);
3682 get_block_t *get_block_func = NULL;
3684 loff_t final_size = offset + count;
3688 if (final_size > inode->i_size) {
3689 /* Credits for sb + inode write */
3690 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3691 if (IS_ERR(handle)) {
3692 ret = PTR_ERR(handle);
3695 ret = ext4_orphan_add(handle, inode);
3697 ext4_journal_stop(handle);
3701 ext4_update_i_disksize(inode, inode->i_size);
3702 ext4_journal_stop(handle);
3705 BUG_ON(iocb->private == NULL);
3708 * Make all waiters for direct IO properly wait also for extent
3709 * conversion. This also disallows race between truncate() and
3710 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3712 inode_dio_begin(inode);
3714 /* If we do a overwrite dio, i_mutex locking can be released */
3715 overwrite = *((int *)iocb->private);
3718 inode_unlock(inode);
3721 * For extent mapped files we could direct write to holes and fallocate.
3723 * Allocated blocks to fill the hole are marked as unwritten to prevent
3724 * parallel buffered read to expose the stale data before DIO complete
3727 * As to previously fallocated extents, ext4 get_block will just simply
3728 * mark the buffer mapped but still keep the extents unwritten.
3730 * For non AIO case, we will convert those unwritten extents to written
3731 * after return back from blockdev_direct_IO. That way we save us from
3732 * allocating io_end structure and also the overhead of offloading
3733 * the extent convertion to a workqueue.
3735 * For async DIO, the conversion needs to be deferred when the
3736 * IO is completed. The ext4 end_io callback function will be
3737 * called to take care of the conversion work. Here for async
3738 * case, we allocate an io_end structure to hook to the iocb.
3740 iocb->private = NULL;
3742 get_block_func = ext4_dio_get_block_overwrite;
3743 else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3744 round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3745 get_block_func = ext4_dio_get_block;
3746 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3747 } else if (is_sync_kiocb(iocb)) {
3748 get_block_func = ext4_dio_get_block_unwritten_sync;
3749 dio_flags = DIO_LOCKING;
3751 get_block_func = ext4_dio_get_block_unwritten_async;
3752 dio_flags = DIO_LOCKING;
3754 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3755 get_block_func, ext4_end_io_dio, NULL,
3758 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3759 EXT4_STATE_DIO_UNWRITTEN)) {
3762 * for non AIO case, since the IO is already
3763 * completed, we could do the conversion right here
3765 err = ext4_convert_unwritten_extents(NULL, inode,
3769 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3772 inode_dio_end(inode);
3773 /* take i_mutex locking again if we do a ovewrite dio */
3777 if (ret < 0 && final_size > inode->i_size)
3778 ext4_truncate_failed_write(inode);
3780 /* Handle extending of i_size after direct IO write */
3784 /* Credits for sb + inode write */
3785 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3786 if (IS_ERR(handle)) {
3788 * We wrote the data but cannot extend
3789 * i_size. Bail out. In async io case, we do
3790 * not return error here because we have
3791 * already submmitted the corresponding
3792 * bio. Returning error here makes the caller
3793 * think that this IO is done and failed
3794 * resulting in race with bio's completion
3798 ret = PTR_ERR(handle);
3800 ext4_orphan_del(NULL, inode);
3805 ext4_orphan_del(handle, inode);
3807 loff_t end = offset + ret;
3808 if (end > inode->i_size) {
3809 ext4_update_i_disksize(inode, end);
3810 i_size_write(inode, end);
3812 * We're going to return a positive `ret'
3813 * here due to non-zero-length I/O, so there's
3814 * no way of reporting error returns from
3815 * ext4_mark_inode_dirty() to userspace. So
3818 ext4_mark_inode_dirty(handle, inode);
3821 err = ext4_journal_stop(handle);
3829 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3831 struct address_space *mapping = iocb->ki_filp->f_mapping;
3832 struct inode *inode = mapping->host;
3833 size_t count = iov_iter_count(iter);
3835 loff_t offset = iocb->ki_pos;
3836 loff_t size = i_size_read(inode);
3842 * Shared inode_lock is enough for us - it protects against concurrent
3843 * writes & truncates and since we take care of writing back page cache,
3844 * we are protected against page writeback as well.
3846 if (iocb->ki_flags & IOCB_NOWAIT) {
3847 if (!inode_trylock_shared(inode))
3850 inode_lock_shared(inode);
3853 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3854 iocb->ki_pos + count - 1);
3857 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3858 iter, ext4_dio_get_block, NULL, NULL, 0);
3860 inode_unlock_shared(inode);
3864 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3866 struct file *file = iocb->ki_filp;
3867 struct inode *inode = file->f_mapping->host;
3868 size_t count = iov_iter_count(iter);
3869 loff_t offset = iocb->ki_pos;
3872 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3873 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3878 * If we are doing data journalling we don't support O_DIRECT
3880 if (ext4_should_journal_data(inode))
3883 /* Let buffer I/O handle the inline data case. */
3884 if (ext4_has_inline_data(inode))
3887 /* DAX uses iomap path now */
3888 if (WARN_ON_ONCE(IS_DAX(inode)))
3891 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3892 if (iov_iter_rw(iter) == READ)
3893 ret = ext4_direct_IO_read(iocb, iter);
3895 ret = ext4_direct_IO_write(iocb, iter);
3896 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3901 * Pages can be marked dirty completely asynchronously from ext4's journalling
3902 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3903 * much here because ->set_page_dirty is called under VFS locks. The page is
3904 * not necessarily locked.
3906 * We cannot just dirty the page and leave attached buffers clean, because the
3907 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3908 * or jbddirty because all the journalling code will explode.
3910 * So what we do is to mark the page "pending dirty" and next time writepage
3911 * is called, propagate that into the buffers appropriately.
3913 static int ext4_journalled_set_page_dirty(struct page *page)
3915 SetPageChecked(page);
3916 return __set_page_dirty_nobuffers(page);
3919 static int ext4_set_page_dirty(struct page *page)
3921 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3922 WARN_ON_ONCE(!page_has_buffers(page));
3923 return __set_page_dirty_buffers(page);
3926 static const struct address_space_operations ext4_aops = {
3927 .readpage = ext4_readpage,
3928 .readpages = ext4_readpages,
3929 .writepage = ext4_writepage,
3930 .writepages = ext4_writepages,
3931 .write_begin = ext4_write_begin,
3932 .write_end = ext4_write_end,
3933 .set_page_dirty = ext4_set_page_dirty,
3935 .invalidatepage = ext4_invalidatepage,
3936 .releasepage = ext4_releasepage,
3937 .direct_IO = ext4_direct_IO,
3938 .migratepage = buffer_migrate_page,
3939 .is_partially_uptodate = block_is_partially_uptodate,
3940 .error_remove_page = generic_error_remove_page,
3943 static const struct address_space_operations ext4_journalled_aops = {
3944 .readpage = ext4_readpage,
3945 .readpages = ext4_readpages,
3946 .writepage = ext4_writepage,
3947 .writepages = ext4_writepages,
3948 .write_begin = ext4_write_begin,
3949 .write_end = ext4_journalled_write_end,
3950 .set_page_dirty = ext4_journalled_set_page_dirty,
3952 .invalidatepage = ext4_journalled_invalidatepage,
3953 .releasepage = ext4_releasepage,
3954 .direct_IO = ext4_direct_IO,
3955 .is_partially_uptodate = block_is_partially_uptodate,
3956 .error_remove_page = generic_error_remove_page,
3959 static const struct address_space_operations ext4_da_aops = {
3960 .readpage = ext4_readpage,
3961 .readpages = ext4_readpages,
3962 .writepage = ext4_writepage,
3963 .writepages = ext4_writepages,
3964 .write_begin = ext4_da_write_begin,
3965 .write_end = ext4_da_write_end,
3966 .set_page_dirty = ext4_set_page_dirty,
3968 .invalidatepage = ext4_da_invalidatepage,
3969 .releasepage = ext4_releasepage,
3970 .direct_IO = ext4_direct_IO,
3971 .migratepage = buffer_migrate_page,
3972 .is_partially_uptodate = block_is_partially_uptodate,
3973 .error_remove_page = generic_error_remove_page,
3976 void ext4_set_aops(struct inode *inode)
3978 switch (ext4_inode_journal_mode(inode)) {
3979 case EXT4_INODE_ORDERED_DATA_MODE:
3980 case EXT4_INODE_WRITEBACK_DATA_MODE:
3982 case EXT4_INODE_JOURNAL_DATA_MODE:
3983 inode->i_mapping->a_ops = &ext4_journalled_aops;
3988 if (test_opt(inode->i_sb, DELALLOC))
3989 inode->i_mapping->a_ops = &ext4_da_aops;
3991 inode->i_mapping->a_ops = &ext4_aops;
3994 static int __ext4_block_zero_page_range(handle_t *handle,
3995 struct address_space *mapping, loff_t from, loff_t length)
3997 ext4_fsblk_t index = from >> PAGE_SHIFT;
3998 unsigned offset = from & (PAGE_SIZE-1);
3999 unsigned blocksize, pos;
4001 struct inode *inode = mapping->host;
4002 struct buffer_head *bh;
4006 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
4007 mapping_gfp_constraint(mapping, ~__GFP_FS));
4011 blocksize = inode->i_sb->s_blocksize;
4013 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
4015 if (!page_has_buffers(page))
4016 create_empty_buffers(page, blocksize, 0);
4018 /* Find the buffer that contains "offset" */
4019 bh = page_buffers(page);
4021 while (offset >= pos) {
4022 bh = bh->b_this_page;
4026 if (buffer_freed(bh)) {
4027 BUFFER_TRACE(bh, "freed: skip");
4030 if (!buffer_mapped(bh)) {
4031 BUFFER_TRACE(bh, "unmapped");
4032 ext4_get_block(inode, iblock, bh, 0);
4033 /* unmapped? It's a hole - nothing to do */
4034 if (!buffer_mapped(bh)) {
4035 BUFFER_TRACE(bh, "still unmapped");
4040 /* Ok, it's mapped. Make sure it's up-to-date */
4041 if (PageUptodate(page))
4042 set_buffer_uptodate(bh);
4044 if (!buffer_uptodate(bh)) {
4046 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
4048 /* Uhhuh. Read error. Complain and punt. */
4049 if (!buffer_uptodate(bh))
4051 if (S_ISREG(inode->i_mode) &&
4052 ext4_encrypted_inode(inode)) {
4053 /* We expect the key to be set. */
4054 BUG_ON(!fscrypt_has_encryption_key(inode));
4055 BUG_ON(blocksize != PAGE_SIZE);
4056 WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host,
4057 page, PAGE_SIZE, 0, page->index));
4060 if (ext4_should_journal_data(inode)) {
4061 BUFFER_TRACE(bh, "get write access");
4062 err = ext4_journal_get_write_access(handle, bh);
4066 zero_user(page, offset, length);
4067 BUFFER_TRACE(bh, "zeroed end of block");
4069 if (ext4_should_journal_data(inode)) {
4070 err = ext4_handle_dirty_metadata(handle, inode, bh);
4073 mark_buffer_dirty(bh);
4074 if (ext4_should_order_data(inode))
4075 err = ext4_jbd2_inode_add_write(handle, inode, from,
4086 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4087 * starting from file offset 'from'. The range to be zero'd must
4088 * be contained with in one block. If the specified range exceeds
4089 * the end of the block it will be shortened to end of the block
4090 * that cooresponds to 'from'
4092 static int ext4_block_zero_page_range(handle_t *handle,
4093 struct address_space *mapping, loff_t from, loff_t length)
4095 struct inode *inode = mapping->host;
4096 unsigned offset = from & (PAGE_SIZE-1);
4097 unsigned blocksize = inode->i_sb->s_blocksize;
4098 unsigned max = blocksize - (offset & (blocksize - 1));
4101 * correct length if it does not fall between
4102 * 'from' and the end of the block
4104 if (length > max || length < 0)
4107 if (IS_DAX(inode)) {
4108 return iomap_zero_range(inode, from, length, NULL,
4111 return __ext4_block_zero_page_range(handle, mapping, from, length);
4115 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4116 * up to the end of the block which corresponds to `from'.
4117 * This required during truncate. We need to physically zero the tail end
4118 * of that block so it doesn't yield old data if the file is later grown.
4120 static int ext4_block_truncate_page(handle_t *handle,
4121 struct address_space *mapping, loff_t from)
4123 unsigned offset = from & (PAGE_SIZE-1);
4126 struct inode *inode = mapping->host;
4128 /* If we are processing an encrypted inode during orphan list handling */
4129 if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode))
4132 blocksize = inode->i_sb->s_blocksize;
4133 length = blocksize - (offset & (blocksize - 1));
4135 return ext4_block_zero_page_range(handle, mapping, from, length);
4138 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4139 loff_t lstart, loff_t length)
4141 struct super_block *sb = inode->i_sb;
4142 struct address_space *mapping = inode->i_mapping;
4143 unsigned partial_start, partial_end;
4144 ext4_fsblk_t start, end;
4145 loff_t byte_end = (lstart + length - 1);
4148 partial_start = lstart & (sb->s_blocksize - 1);
4149 partial_end = byte_end & (sb->s_blocksize - 1);
4151 start = lstart >> sb->s_blocksize_bits;
4152 end = byte_end >> sb->s_blocksize_bits;
4154 /* Handle partial zero within the single block */
4156 (partial_start || (partial_end != sb->s_blocksize - 1))) {
4157 err = ext4_block_zero_page_range(handle, mapping,
4161 /* Handle partial zero out on the start of the range */
4162 if (partial_start) {
4163 err = ext4_block_zero_page_range(handle, mapping,
4164 lstart, sb->s_blocksize);
4168 /* Handle partial zero out on the end of the range */
4169 if (partial_end != sb->s_blocksize - 1)
4170 err = ext4_block_zero_page_range(handle, mapping,
4171 byte_end - partial_end,
4176 int ext4_can_truncate(struct inode *inode)
4178 if (S_ISREG(inode->i_mode))
4180 if (S_ISDIR(inode->i_mode))
4182 if (S_ISLNK(inode->i_mode))
4183 return !ext4_inode_is_fast_symlink(inode);
4188 * We have to make sure i_disksize gets properly updated before we truncate
4189 * page cache due to hole punching or zero range. Otherwise i_disksize update
4190 * can get lost as it may have been postponed to submission of writeback but
4191 * that will never happen after we truncate page cache.
4193 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4197 loff_t size = i_size_read(inode);
4199 WARN_ON(!inode_is_locked(inode));
4200 if (offset > size || offset + len < size)
4203 if (EXT4_I(inode)->i_disksize >= size)
4206 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4208 return PTR_ERR(handle);
4209 ext4_update_i_disksize(inode, size);
4210 ext4_mark_inode_dirty(handle, inode);
4211 ext4_journal_stop(handle);
4217 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4218 * associated with the given offset and length
4220 * @inode: File inode
4221 * @offset: The offset where the hole will begin
4222 * @len: The length of the hole
4224 * Returns: 0 on success or negative on failure
4227 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4229 struct super_block *sb = inode->i_sb;
4230 ext4_lblk_t first_block, stop_block;
4231 struct address_space *mapping = inode->i_mapping;
4232 loff_t first_block_offset, last_block_offset, max_length;
4233 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4235 unsigned int credits;
4238 if (!S_ISREG(inode->i_mode))
4241 trace_ext4_punch_hole(inode, offset, length, 0);
4243 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4244 if (ext4_has_inline_data(inode)) {
4245 down_write(&EXT4_I(inode)->i_mmap_sem);
4246 ret = ext4_convert_inline_data(inode);
4247 up_write(&EXT4_I(inode)->i_mmap_sem);
4253 * Write out all dirty pages to avoid race conditions
4254 * Then release them.
4256 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4257 ret = filemap_write_and_wait_range(mapping, offset,
4258 offset + length - 1);
4265 /* No need to punch hole beyond i_size */
4266 if (offset >= inode->i_size)
4270 * If the hole extends beyond i_size, set the hole
4271 * to end after the page that contains i_size
4273 if (offset + length > inode->i_size) {
4274 length = inode->i_size +
4275 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4280 * For punch hole the length + offset needs to be within one block
4281 * before last range. Adjust the length if it goes beyond that limit.
4283 max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
4284 if (offset + length > max_length)
4285 length = max_length - offset;
4287 if (offset & (sb->s_blocksize - 1) ||
4288 (offset + length) & (sb->s_blocksize - 1)) {
4290 * Attach jinode to inode for jbd2 if we do any zeroing of
4293 ret = ext4_inode_attach_jinode(inode);
4299 /* Wait all existing dio workers, newcomers will block on i_mutex */
4300 ext4_inode_block_unlocked_dio(inode);
4301 inode_dio_wait(inode);
4304 * Prevent page faults from reinstantiating pages we have released from
4307 down_write(&EXT4_I(inode)->i_mmap_sem);
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);
4316 truncate_pagecache_range(inode, first_block_offset,
4320 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4321 credits = ext4_writepage_trans_blocks(inode);
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);
4331 ret = ext4_zero_partial_blocks(handle, inode, offset,
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);
4349 up_write(&EXT4_I(inode)->i_data_sem);
4353 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4354 ret = ext4_ext_remove_space(inode, first_block,
4357 ret = ext4_ind_remove_space(handle, inode, first_block,
4360 up_write(&EXT4_I(inode)->i_data_sem);
4363 ext4_handle_sync(handle);
4365 inode->i_mtime = inode->i_ctime = current_time(inode);
4366 ext4_mark_inode_dirty(handle, inode);
4368 ext4_update_inode_fsync_trans(handle, inode, 1);
4370 ext4_journal_stop(handle);
4372 up_write(&EXT4_I(inode)->i_mmap_sem);
4373 ext4_inode_resume_unlocked_dio(inode);
4375 inode_unlock(inode);
4379 int ext4_inode_attach_jinode(struct inode *inode)
4381 struct ext4_inode_info *ei = EXT4_I(inode);
4382 struct jbd2_inode *jinode;
4384 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4387 jinode = jbd2_alloc_inode(GFP_KERNEL);
4388 spin_lock(&inode->i_lock);
4391 spin_unlock(&inode->i_lock);
4394 ei->jinode = jinode;
4395 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4398 spin_unlock(&inode->i_lock);
4399 if (unlikely(jinode != NULL))
4400 jbd2_free_inode(jinode);
4407 * We block out ext4_get_block() block instantiations across the entire
4408 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4409 * simultaneously on behalf of the same inode.
4411 * As we work through the truncate and commit bits of it to the journal there
4412 * is one core, guiding principle: the file's tree must always be consistent on
4413 * disk. We must be able to restart the truncate after a crash.
4415 * The file's tree may be transiently inconsistent in memory (although it
4416 * probably isn't), but whenever we close off and commit a journal transaction,
4417 * the contents of (the filesystem + the journal) must be consistent and
4418 * restartable. It's pretty simple, really: bottom up, right to left (although
4419 * left-to-right works OK too).
4421 * Note that at recovery time, journal replay occurs *before* the restart of
4422 * truncate against the orphan inode list.
4424 * The committed inode has the new, desired i_size (which is the same as
4425 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4426 * that this inode's truncate did not complete and it will again call
4427 * ext4_truncate() to have another go. So there will be instantiated blocks
4428 * to the right of the truncation point in a crashed ext4 filesystem. But
4429 * that's fine - as long as they are linked from the inode, the post-crash
4430 * ext4_truncate() run will find them and release them.
4432 int ext4_truncate(struct inode *inode)
4434 struct ext4_inode_info *ei = EXT4_I(inode);
4435 unsigned int credits;
4438 struct address_space *mapping = inode->i_mapping;
4441 * There is a possibility that we're either freeing the inode
4442 * or it's a completely new inode. In those cases we might not
4443 * have i_mutex locked because it's not necessary.
4445 if (!(inode->i_state & (I_NEW|I_FREEING)))
4446 WARN_ON(!inode_is_locked(inode));
4447 trace_ext4_truncate_enter(inode);
4449 if (!ext4_can_truncate(inode))
4452 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4454 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4455 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4457 if (ext4_has_inline_data(inode)) {
4460 err = ext4_inline_data_truncate(inode, &has_inline);
4467 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4468 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4469 if (ext4_inode_attach_jinode(inode) < 0)
4473 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4474 credits = ext4_writepage_trans_blocks(inode);
4476 credits = ext4_blocks_for_truncate(inode);
4478 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4480 return PTR_ERR(handle);
4482 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4483 ext4_block_truncate_page(handle, mapping, inode->i_size);
4486 * We add the inode to the orphan list, so that if this
4487 * truncate spans multiple transactions, and we crash, we will
4488 * resume the truncate when the filesystem recovers. It also
4489 * marks the inode dirty, to catch the new size.
4491 * Implication: the file must always be in a sane, consistent
4492 * truncatable state while each transaction commits.
4494 err = ext4_orphan_add(handle, inode);
4498 down_write(&EXT4_I(inode)->i_data_sem);
4500 ext4_discard_preallocations(inode);
4502 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4503 err = ext4_ext_truncate(handle, inode);
4505 ext4_ind_truncate(handle, inode);
4507 up_write(&ei->i_data_sem);
4512 ext4_handle_sync(handle);
4516 * If this was a simple ftruncate() and the file will remain alive,
4517 * then we need to clear up the orphan record which we created above.
4518 * However, if this was a real unlink then we were called by
4519 * ext4_evict_inode(), and we allow that function to clean up the
4520 * orphan info for us.
4523 ext4_orphan_del(handle, inode);
4525 inode->i_mtime = inode->i_ctime = current_time(inode);
4526 ext4_mark_inode_dirty(handle, inode);
4527 ext4_journal_stop(handle);
4529 trace_ext4_truncate_exit(inode);
4534 * ext4_get_inode_loc returns with an extra refcount against the inode's
4535 * underlying buffer_head on success. If 'in_mem' is true, we have all
4536 * data in memory that is needed to recreate the on-disk version of this
4539 static int __ext4_get_inode_loc(struct inode *inode,
4540 struct ext4_iloc *iloc, int in_mem)
4542 struct ext4_group_desc *gdp;
4543 struct buffer_head *bh;
4544 struct super_block *sb = inode->i_sb;
4546 int inodes_per_block, inode_offset;
4549 if (inode->i_ino < EXT4_ROOT_INO ||
4550 inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4551 return -EFSCORRUPTED;
4553 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4554 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4559 * Figure out the offset within the block group inode table
4561 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4562 inode_offset = ((inode->i_ino - 1) %
4563 EXT4_INODES_PER_GROUP(sb));
4564 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4566 block = ext4_inode_table(sb, gdp);
4567 if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4568 (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4569 ext4_error(sb, "Invalid inode table block %llu in "
4570 "block_group %u", block, iloc->block_group);
4571 return -EFSCORRUPTED;
4573 block += (inode_offset / inodes_per_block);
4575 bh = sb_getblk(sb, block);
4578 if (!buffer_uptodate(bh)) {
4582 * If the buffer has the write error flag, we have failed
4583 * to write out another inode in the same block. In this
4584 * case, we don't have to read the block because we may
4585 * read the old inode data successfully.
4587 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4588 set_buffer_uptodate(bh);
4590 if (buffer_uptodate(bh)) {
4591 /* someone brought it uptodate while we waited */
4597 * If we have all information of the inode in memory and this
4598 * is the only valid inode in the block, we need not read the
4602 struct buffer_head *bitmap_bh;
4605 start = inode_offset & ~(inodes_per_block - 1);
4607 /* Is the inode bitmap in cache? */
4608 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4609 if (unlikely(!bitmap_bh))
4613 * If the inode bitmap isn't in cache then the
4614 * optimisation may end up performing two reads instead
4615 * of one, so skip it.
4617 if (!buffer_uptodate(bitmap_bh)) {
4621 for (i = start; i < start + inodes_per_block; i++) {
4622 if (i == inode_offset)
4624 if (ext4_test_bit(i, bitmap_bh->b_data))
4628 if (i == start + inodes_per_block) {
4629 /* all other inodes are free, so skip I/O */
4630 memset(bh->b_data, 0, bh->b_size);
4631 set_buffer_uptodate(bh);
4639 * If we need to do any I/O, try to pre-readahead extra
4640 * blocks from the inode table.
4642 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4643 ext4_fsblk_t b, end, table;
4645 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4647 table = ext4_inode_table(sb, gdp);
4648 /* s_inode_readahead_blks is always a power of 2 */
4649 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4653 num = EXT4_INODES_PER_GROUP(sb);
4654 if (ext4_has_group_desc_csum(sb))
4655 num -= ext4_itable_unused_count(sb, gdp);
4656 table += num / inodes_per_block;
4660 sb_breadahead_unmovable(sb, b++);
4664 * There are other valid inodes in the buffer, this inode
4665 * has in-inode xattrs, or we don't have this inode in memory.
4666 * Read the block from disk.
4668 trace_ext4_load_inode(inode);
4670 bh->b_end_io = end_buffer_read_sync;
4671 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4673 if (!buffer_uptodate(bh)) {
4674 EXT4_ERROR_INODE_BLOCK(inode, block,
4675 "unable to read itable block");
4685 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4687 /* We have all inode data except xattrs in memory here. */
4688 return __ext4_get_inode_loc(inode, iloc,
4689 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4692 void ext4_set_inode_flags(struct inode *inode)
4694 unsigned int flags = EXT4_I(inode)->i_flags;
4695 unsigned int new_fl = 0;
4697 if (flags & EXT4_SYNC_FL)
4699 if (flags & EXT4_APPEND_FL)
4701 if (flags & EXT4_IMMUTABLE_FL)
4702 new_fl |= S_IMMUTABLE;
4703 if (flags & EXT4_NOATIME_FL)
4704 new_fl |= S_NOATIME;
4705 if (flags & EXT4_DIRSYNC_FL)
4706 new_fl |= S_DIRSYNC;
4707 if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode) &&
4708 !ext4_should_journal_data(inode) && !ext4_has_inline_data(inode) &&
4709 !ext4_encrypted_inode(inode))
4711 inode_set_flags(inode, new_fl,
4712 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4715 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4716 struct ext4_inode_info *ei)
4719 struct inode *inode = &(ei->vfs_inode);
4720 struct super_block *sb = inode->i_sb;
4722 if (ext4_has_feature_huge_file(sb)) {
4723 /* we are using combined 48 bit field */
4724 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4725 le32_to_cpu(raw_inode->i_blocks_lo);
4726 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4727 /* i_blocks represent file system block size */
4728 return i_blocks << (inode->i_blkbits - 9);
4733 return le32_to_cpu(raw_inode->i_blocks_lo);
4737 static inline int ext4_iget_extra_inode(struct inode *inode,
4738 struct ext4_inode *raw_inode,
4739 struct ext4_inode_info *ei)
4741 __le32 *magic = (void *)raw_inode +
4742 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4744 if (EXT4_INODE_HAS_XATTR_SPACE(inode) &&
4745 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4748 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4749 err = ext4_find_inline_data_nolock(inode);
4750 if (!err && ext4_has_inline_data(inode))
4751 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4754 EXT4_I(inode)->i_inline_off = 0;
4758 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4760 if (!ext4_has_feature_project(inode->i_sb))
4762 *projid = EXT4_I(inode)->i_projid;
4766 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4767 ext4_iget_flags flags, const char *function,
4770 struct ext4_iloc iloc;
4771 struct ext4_inode *raw_inode;
4772 struct ext4_inode_info *ei;
4773 struct inode *inode;
4774 journal_t *journal = EXT4_SB(sb)->s_journal;
4782 if ((!(flags & EXT4_IGET_SPECIAL) &&
4783 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4784 (ino < EXT4_ROOT_INO) ||
4785 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4786 if (flags & EXT4_IGET_HANDLE)
4787 return ERR_PTR(-ESTALE);
4788 __ext4_error(sb, function, line,
4789 "inode #%lu: comm %s: iget: illegal inode #",
4790 ino, current->comm);
4791 return ERR_PTR(-EFSCORRUPTED);
4794 inode = iget_locked(sb, ino);
4796 return ERR_PTR(-ENOMEM);
4797 if (!(inode->i_state & I_NEW))
4803 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4806 raw_inode = ext4_raw_inode(&iloc);
4808 if ((flags & EXT4_IGET_HANDLE) &&
4809 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4814 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4815 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4816 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4817 EXT4_INODE_SIZE(inode->i_sb) ||
4818 (ei->i_extra_isize & 3)) {
4819 ext4_error_inode(inode, function, line, 0,
4820 "iget: bad extra_isize %u "
4823 EXT4_INODE_SIZE(inode->i_sb));
4824 ret = -EFSCORRUPTED;
4828 ei->i_extra_isize = 0;
4830 /* Precompute checksum seed for inode metadata */
4831 if (ext4_has_metadata_csum(sb)) {
4832 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4834 __le32 inum = cpu_to_le32(inode->i_ino);
4835 __le32 gen = raw_inode->i_generation;
4836 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4838 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4842 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4843 ext4_error_inode(inode, function, line, 0,
4844 "iget: checksum invalid");
4849 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4850 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4851 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4852 if (ext4_has_feature_project(sb) &&
4853 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4854 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4855 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4857 i_projid = EXT4_DEF_PROJID;
4859 if (!(test_opt(inode->i_sb, NO_UID32))) {
4860 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4861 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4863 i_uid_write(inode, i_uid);
4864 i_gid_write(inode, i_gid);
4865 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4866 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4868 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4869 ei->i_inline_off = 0;
4870 ei->i_dir_start_lookup = 0;
4871 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4872 /* We now have enough fields to check if the inode was active or not.
4873 * This is needed because nfsd might try to access dead inodes
4874 * the test is that same one that e2fsck uses
4875 * NeilBrown 1999oct15
4877 if (inode->i_nlink == 0) {
4878 if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL ||
4879 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4880 ino != EXT4_BOOT_LOADER_INO) {
4881 /* this inode is deleted or unallocated */
4882 if (flags & EXT4_IGET_SPECIAL) {
4883 ext4_error_inode(inode, function, line, 0,
4884 "iget: special inode unallocated");
4885 ret = -EFSCORRUPTED;
4890 /* The only unlinked inodes we let through here have
4891 * valid i_mode and are being read by the orphan
4892 * recovery code: that's fine, we're about to complete
4893 * the process of deleting those.
4894 * OR it is the EXT4_BOOT_LOADER_INO which is
4895 * not initialized on a new filesystem. */
4897 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4898 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4899 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4900 if (ext4_has_feature_64bit(sb))
4902 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4903 inode->i_size = ext4_isize(sb, raw_inode);
4904 if ((size = i_size_read(inode)) < 0) {
4905 ext4_error_inode(inode, function, line, 0,
4906 "iget: bad i_size value: %lld", size);
4907 ret = -EFSCORRUPTED;
4911 * If dir_index is not enabled but there's dir with INDEX flag set,
4912 * we'd normally treat htree data as empty space. But with metadata
4913 * checksumming that corrupts checksums so forbid that.
4915 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4916 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4917 EXT4_ERROR_INODE(inode,
4918 "iget: Dir with htree data on filesystem without dir_index feature.");
4919 ret = -EFSCORRUPTED;
4922 ei->i_disksize = inode->i_size;
4924 ei->i_reserved_quota = 0;
4926 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4927 ei->i_block_group = iloc.block_group;
4928 ei->i_last_alloc_group = ~0;
4930 * NOTE! The in-memory inode i_data array is in little-endian order
4931 * even on big-endian machines: we do NOT byteswap the block numbers!
4933 for (block = 0; block < EXT4_N_BLOCKS; block++)
4934 ei->i_data[block] = raw_inode->i_block[block];
4935 INIT_LIST_HEAD(&ei->i_orphan);
4938 * Set transaction id's of transactions that have to be committed
4939 * to finish f[data]sync. We set them to currently running transaction
4940 * as we cannot be sure that the inode or some of its metadata isn't
4941 * part of the transaction - the inode could have been reclaimed and
4942 * now it is reread from disk.
4945 transaction_t *transaction;
4948 read_lock(&journal->j_state_lock);
4949 if (journal->j_running_transaction)
4950 transaction = journal->j_running_transaction;
4952 transaction = journal->j_committing_transaction;
4954 tid = transaction->t_tid;
4956 tid = journal->j_commit_sequence;
4957 read_unlock(&journal->j_state_lock);
4958 ei->i_sync_tid = tid;
4959 ei->i_datasync_tid = tid;
4962 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4963 if (ei->i_extra_isize == 0) {
4964 /* The extra space is currently unused. Use it. */
4965 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4966 ei->i_extra_isize = sizeof(struct ext4_inode) -
4967 EXT4_GOOD_OLD_INODE_SIZE;
4969 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4975 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4976 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4977 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4978 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4980 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4981 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4982 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4983 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4985 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4990 if (ei->i_file_acl &&
4991 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4992 ext4_error_inode(inode, function, line, 0,
4993 "iget: bad extended attribute block %llu",
4995 ret = -EFSCORRUPTED;
4997 } else if (!ext4_has_inline_data(inode)) {
4998 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4999 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5000 (S_ISLNK(inode->i_mode) &&
5001 !ext4_inode_is_fast_symlink(inode))))
5002 /* Validate extent which is part of inode */
5003 ret = ext4_ext_check_inode(inode);
5004 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5005 (S_ISLNK(inode->i_mode) &&
5006 !ext4_inode_is_fast_symlink(inode))) {
5007 /* Validate block references which are part of inode */
5008 ret = ext4_ind_check_inode(inode);
5014 if (S_ISREG(inode->i_mode)) {
5015 inode->i_op = &ext4_file_inode_operations;
5016 inode->i_fop = &ext4_file_operations;
5017 ext4_set_aops(inode);
5018 } else if (S_ISDIR(inode->i_mode)) {
5019 inode->i_op = &ext4_dir_inode_operations;
5020 inode->i_fop = &ext4_dir_operations;
5021 } else if (S_ISLNK(inode->i_mode)) {
5022 if (ext4_encrypted_inode(inode)) {
5023 inode->i_op = &ext4_encrypted_symlink_inode_operations;
5024 ext4_set_aops(inode);
5025 } else if (ext4_inode_is_fast_symlink(inode)) {
5026 inode->i_link = (char *)ei->i_data;
5027 inode->i_op = &ext4_fast_symlink_inode_operations;
5028 nd_terminate_link(ei->i_data, inode->i_size,
5029 sizeof(ei->i_data) - 1);
5031 inode->i_op = &ext4_symlink_inode_operations;
5032 ext4_set_aops(inode);
5034 inode_nohighmem(inode);
5035 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5036 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5037 inode->i_op = &ext4_special_inode_operations;
5038 if (raw_inode->i_block[0])
5039 init_special_inode(inode, inode->i_mode,
5040 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5042 init_special_inode(inode, inode->i_mode,
5043 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5044 } else if (ino == EXT4_BOOT_LOADER_INO) {
5045 make_bad_inode(inode);
5047 ret = -EFSCORRUPTED;
5048 ext4_error_inode(inode, function, line, 0,
5049 "iget: bogus i_mode (%o)", inode->i_mode);
5053 ext4_set_inode_flags(inode);
5055 unlock_new_inode(inode);
5061 return ERR_PTR(ret);
5064 static int ext4_inode_blocks_set(handle_t *handle,
5065 struct ext4_inode *raw_inode,
5066 struct ext4_inode_info *ei)
5068 struct inode *inode = &(ei->vfs_inode);
5069 u64 i_blocks = READ_ONCE(inode->i_blocks);
5070 struct super_block *sb = inode->i_sb;
5072 if (i_blocks <= ~0U) {
5074 * i_blocks can be represented in a 32 bit variable
5075 * as multiple of 512 bytes
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);
5082 if (!ext4_has_feature_huge_file(sb))
5085 if (i_blocks <= 0xffffffffffffULL) {
5087 * i_blocks can be represented in a 48 bit variable
5088 * as multiple of 512 bytes
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);
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);
5103 struct other_inode {
5104 unsigned long orig_ino;
5105 struct ext4_inode *raw_inode;
5108 static int other_inode_match(struct inode * inode, unsigned long ino,
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 |
5116 ((inode->i_state & I_DIRTY_TIME) == 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;
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);
5136 spin_unlock(&inode->i_lock);
5141 * Opportunistically update the other time fields for other inodes in
5142 * the same inode table block.
5144 static void ext4_update_other_inodes_time(struct super_block *sb,
5145 unsigned long orig_ino, char *buf)
5147 struct other_inode oi;
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;
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).
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)
5162 oi.raw_inode = (struct ext4_inode *) buf;
5163 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
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.
5172 * The caller must have write access to iloc->bh.
5174 static int ext4_do_update_inode(handle_t *handle,
5175 struct inode *inode,
5176 struct ext4_iloc *iloc)
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;
5183 int need_datasync = 0, set_large_file = 0;
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 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5197 spin_unlock(&ei->i_raw_lock);
5201 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5202 i_uid = i_uid_read(inode);
5203 i_gid = i_gid_read(inode);
5204 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5205 if (!(test_opt(inode->i_sb, NO_UID32))) {
5206 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5207 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5209 * Fix up interoperability with old kernels. Otherwise, old inodes get
5210 * re-used with the upper 16 bits of the uid/gid intact
5212 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5213 raw_inode->i_uid_high = 0;
5214 raw_inode->i_gid_high = 0;
5216 raw_inode->i_uid_high =
5217 cpu_to_le16(high_16_bits(i_uid));
5218 raw_inode->i_gid_high =
5219 cpu_to_le16(high_16_bits(i_gid));
5222 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5223 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5224 raw_inode->i_uid_high = 0;
5225 raw_inode->i_gid_high = 0;
5227 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5229 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5230 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5231 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5232 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5234 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5235 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5236 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5237 raw_inode->i_file_acl_high =
5238 cpu_to_le16(ei->i_file_acl >> 32);
5239 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5240 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) {
5241 ext4_isize_set(raw_inode, ei->i_disksize);
5244 if (ei->i_disksize > 0x7fffffffULL) {
5245 if (!ext4_has_feature_large_file(sb) ||
5246 EXT4_SB(sb)->s_es->s_rev_level ==
5247 cpu_to_le32(EXT4_GOOD_OLD_REV))
5250 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5251 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5252 if (old_valid_dev(inode->i_rdev)) {
5253 raw_inode->i_block[0] =
5254 cpu_to_le32(old_encode_dev(inode->i_rdev));
5255 raw_inode->i_block[1] = 0;
5257 raw_inode->i_block[0] = 0;
5258 raw_inode->i_block[1] =
5259 cpu_to_le32(new_encode_dev(inode->i_rdev));
5260 raw_inode->i_block[2] = 0;
5262 } else if (!ext4_has_inline_data(inode)) {
5263 for (block = 0; block < EXT4_N_BLOCKS; block++)
5264 raw_inode->i_block[block] = ei->i_data[block];
5267 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5268 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5269 if (ei->i_extra_isize) {
5270 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5271 raw_inode->i_version_hi =
5272 cpu_to_le32(inode->i_version >> 32);
5273 raw_inode->i_extra_isize =
5274 cpu_to_le16(ei->i_extra_isize);
5278 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5279 i_projid != EXT4_DEF_PROJID);
5281 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5282 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5283 raw_inode->i_projid = cpu_to_le32(i_projid);
5285 ext4_inode_csum_set(inode, raw_inode, ei);
5286 spin_unlock(&ei->i_raw_lock);
5287 if (inode->i_sb->s_flags & MS_LAZYTIME)
5288 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5291 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5292 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5295 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5296 if (set_large_file) {
5297 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5298 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5301 ext4_update_dynamic_rev(sb);
5302 ext4_set_feature_large_file(sb);
5303 ext4_handle_sync(handle);
5304 err = ext4_handle_dirty_super(handle, sb);
5306 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5309 ext4_std_error(inode->i_sb, err);
5314 * ext4_write_inode()
5316 * We are called from a few places:
5318 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5319 * Here, there will be no transaction running. We wait for any running
5320 * transaction to commit.
5322 * - Within flush work (sys_sync(), kupdate and such).
5323 * We wait on commit, if told to.
5325 * - Within iput_final() -> write_inode_now()
5326 * We wait on commit, if told to.
5328 * In all cases it is actually safe for us to return without doing anything,
5329 * because the inode has been copied into a raw inode buffer in
5330 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5333 * Note that we are absolutely dependent upon all inode dirtiers doing the
5334 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5335 * which we are interested.
5337 * It would be a bug for them to not do this. The code:
5339 * mark_inode_dirty(inode)
5341 * inode->i_size = expr;
5343 * is in error because write_inode() could occur while `stuff()' is running,
5344 * and the new i_size will be lost. Plus the inode will no longer be on the
5345 * superblock's dirty inode list.
5347 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5351 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5352 sb_rdonly(inode->i_sb))
5355 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5358 if (EXT4_SB(inode->i_sb)->s_journal) {
5359 if (ext4_journal_current_handle()) {
5360 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5366 * No need to force transaction in WB_SYNC_NONE mode. Also
5367 * ext4_sync_fs() will force the commit after everything is
5370 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5373 err = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5374 EXT4_I(inode)->i_sync_tid);
5376 struct ext4_iloc iloc;
5378 err = __ext4_get_inode_loc(inode, &iloc, 0);
5382 * sync(2) will flush the whole buffer cache. No need to do
5383 * it here separately for each inode.
5385 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5386 sync_dirty_buffer(iloc.bh);
5387 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5388 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5389 "IO error syncing inode");
5398 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5399 * buffers that are attached to a page stradding i_size and are undergoing
5400 * commit. In that case we have to wait for commit to finish and try again.
5402 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5406 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5407 tid_t commit_tid = 0;
5410 offset = inode->i_size & (PAGE_SIZE - 1);
5412 * If the page is fully truncated, we don't need to wait for any commit
5413 * (and we even should not as __ext4_journalled_invalidatepage() may
5414 * strip all buffers from the page but keep the page dirty which can then
5415 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5416 * buffers). Also we don't need to wait for any commit if all buffers in
5417 * the page remain valid. This is most beneficial for the common case of
5418 * blocksize == PAGESIZE.
5420 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5423 page = find_lock_page(inode->i_mapping,
5424 inode->i_size >> PAGE_SHIFT);
5427 ret = __ext4_journalled_invalidatepage(page, offset,
5428 PAGE_SIZE - offset);
5434 read_lock(&journal->j_state_lock);
5435 if (journal->j_committing_transaction)
5436 commit_tid = journal->j_committing_transaction->t_tid;
5437 read_unlock(&journal->j_state_lock);
5439 jbd2_log_wait_commit(journal, commit_tid);
5446 * Called from notify_change.
5448 * We want to trap VFS attempts to truncate the file as soon as
5449 * possible. In particular, we want to make sure that when the VFS
5450 * shrinks i_size, we put the inode on the orphan list and modify
5451 * i_disksize immediately, so that during the subsequent flushing of
5452 * dirty pages and freeing of disk blocks, we can guarantee that any
5453 * commit will leave the blocks being flushed in an unused state on
5454 * disk. (On recovery, the inode will get truncated and the blocks will
5455 * be freed, so we have a strong guarantee that no future commit will
5456 * leave these blocks visible to the user.)
5458 * Another thing we have to assure is that if we are in ordered mode
5459 * and inode is still attached to the committing transaction, we must
5460 * we start writeout of all the dirty pages which are being truncated.
5461 * This way we are sure that all the data written in the previous
5462 * transaction are already on disk (truncate waits for pages under
5465 * Called with inode->i_mutex down.
5467 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5469 struct inode *inode = d_inode(dentry);
5472 const unsigned int ia_valid = attr->ia_valid;
5474 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5477 if (unlikely(IS_IMMUTABLE(inode)))
5480 if (unlikely(IS_APPEND(inode) &&
5481 (ia_valid & (ATTR_MODE | ATTR_UID |
5482 ATTR_GID | ATTR_TIMES_SET))))
5485 error = setattr_prepare(dentry, attr);
5489 if (is_quota_modification(inode, attr)) {
5490 error = dquot_initialize(inode);
5494 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5495 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5498 /* (user+group)*(old+new) structure, inode write (sb,
5499 * inode block, ? - but truncate inode update has it) */
5500 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5501 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5502 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5503 if (IS_ERR(handle)) {
5504 error = PTR_ERR(handle);
5508 /* dquot_transfer() calls back ext4_get_inode_usage() which
5509 * counts xattr inode references.
5511 down_read(&EXT4_I(inode)->xattr_sem);
5512 error = dquot_transfer(inode, attr);
5513 up_read(&EXT4_I(inode)->xattr_sem);
5516 ext4_journal_stop(handle);
5519 /* Update corresponding info in inode so that everything is in
5520 * one transaction */
5521 if (attr->ia_valid & ATTR_UID)
5522 inode->i_uid = attr->ia_uid;
5523 if (attr->ia_valid & ATTR_GID)
5524 inode->i_gid = attr->ia_gid;
5525 error = ext4_mark_inode_dirty(handle, inode);
5526 ext4_journal_stop(handle);
5529 if (attr->ia_valid & ATTR_SIZE) {
5531 loff_t oldsize = inode->i_size;
5532 int shrink = (attr->ia_size <= inode->i_size);
5534 if (ext4_encrypted_inode(inode)) {
5535 error = fscrypt_get_encryption_info(inode);
5538 if (!fscrypt_has_encryption_key(inode))
5542 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5543 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5545 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5548 if (!S_ISREG(inode->i_mode))
5551 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5552 inode_inc_iversion(inode);
5554 if (ext4_should_order_data(inode) &&
5555 (attr->ia_size < inode->i_size)) {
5556 error = ext4_begin_ordered_truncate(inode,
5561 if (attr->ia_size != inode->i_size) {
5562 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5563 if (IS_ERR(handle)) {
5564 error = PTR_ERR(handle);
5567 if (ext4_handle_valid(handle) && shrink) {
5568 error = ext4_orphan_add(handle, inode);
5572 * Update c/mtime on truncate up, ext4_truncate() will
5573 * update c/mtime in shrink case below
5576 inode->i_mtime = current_time(inode);
5577 inode->i_ctime = inode->i_mtime;
5579 down_write(&EXT4_I(inode)->i_data_sem);
5580 EXT4_I(inode)->i_disksize = attr->ia_size;
5581 rc = ext4_mark_inode_dirty(handle, inode);
5585 * We have to update i_size under i_data_sem together
5586 * with i_disksize to avoid races with writeback code
5587 * running ext4_wb_update_i_disksize().
5590 i_size_write(inode, attr->ia_size);
5591 up_write(&EXT4_I(inode)->i_data_sem);
5592 ext4_journal_stop(handle);
5594 if (orphan && inode->i_nlink)
5595 ext4_orphan_del(NULL, inode);
5600 pagecache_isize_extended(inode, oldsize, inode->i_size);
5603 * Blocks are going to be removed from the inode. Wait
5604 * for dio in flight. Temporarily disable
5605 * dioread_nolock to prevent livelock.
5608 if (!ext4_should_journal_data(inode)) {
5609 ext4_inode_block_unlocked_dio(inode);
5610 inode_dio_wait(inode);
5611 ext4_inode_resume_unlocked_dio(inode);
5613 ext4_wait_for_tail_page_commit(inode);
5615 down_write(&EXT4_I(inode)->i_mmap_sem);
5617 * Truncate pagecache after we've waited for commit
5618 * in data=journal mode to make pages freeable.
5620 truncate_pagecache(inode, inode->i_size);
5622 rc = ext4_truncate(inode);
5626 up_write(&EXT4_I(inode)->i_mmap_sem);
5630 setattr_copy(inode, attr);
5631 mark_inode_dirty(inode);
5635 * If the call to ext4_truncate failed to get a transaction handle at
5636 * all, we need to clean up the in-core orphan list manually.
5638 if (orphan && inode->i_nlink)
5639 ext4_orphan_del(NULL, inode);
5641 if (!error && (ia_valid & ATTR_MODE))
5642 rc = posix_acl_chmod(inode, inode->i_mode);
5645 ext4_std_error(inode->i_sb, error);
5651 int ext4_getattr(const struct path *path, struct kstat *stat,
5652 u32 request_mask, unsigned int query_flags)
5654 struct inode *inode = d_inode(path->dentry);
5655 struct ext4_inode *raw_inode;
5656 struct ext4_inode_info *ei = EXT4_I(inode);
5659 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5660 stat->result_mask |= STATX_BTIME;
5661 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5662 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5665 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5666 if (flags & EXT4_APPEND_FL)
5667 stat->attributes |= STATX_ATTR_APPEND;
5668 if (flags & EXT4_COMPR_FL)
5669 stat->attributes |= STATX_ATTR_COMPRESSED;
5670 if (flags & EXT4_ENCRYPT_FL)
5671 stat->attributes |= STATX_ATTR_ENCRYPTED;
5672 if (flags & EXT4_IMMUTABLE_FL)
5673 stat->attributes |= STATX_ATTR_IMMUTABLE;
5674 if (flags & EXT4_NODUMP_FL)
5675 stat->attributes |= STATX_ATTR_NODUMP;
5677 stat->attributes_mask |= (STATX_ATTR_APPEND |
5678 STATX_ATTR_COMPRESSED |
5679 STATX_ATTR_ENCRYPTED |
5680 STATX_ATTR_IMMUTABLE |
5683 generic_fillattr(inode, stat);
5687 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5688 u32 request_mask, unsigned int query_flags)
5690 struct inode *inode = d_inode(path->dentry);
5691 u64 delalloc_blocks;
5693 ext4_getattr(path, stat, request_mask, query_flags);
5696 * If there is inline data in the inode, the inode will normally not
5697 * have data blocks allocated (it may have an external xattr block).
5698 * Report at least one sector for such files, so tools like tar, rsync,
5699 * others don't incorrectly think the file is completely sparse.
5701 if (unlikely(ext4_has_inline_data(inode)))
5702 stat->blocks += (stat->size + 511) >> 9;
5705 * We can't update i_blocks if the block allocation is delayed
5706 * otherwise in the case of system crash before the real block
5707 * allocation is done, we will have i_blocks inconsistent with
5708 * on-disk file blocks.
5709 * We always keep i_blocks updated together with real
5710 * allocation. But to not confuse with user, stat
5711 * will return the blocks that include the delayed allocation
5712 * blocks for this file.
5714 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5715 EXT4_I(inode)->i_reserved_data_blocks);
5716 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5720 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5723 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5724 return ext4_ind_trans_blocks(inode, lblocks);
5725 return ext4_ext_index_trans_blocks(inode, pextents);
5729 * Account for index blocks, block groups bitmaps and block group
5730 * descriptor blocks if modify datablocks and index blocks
5731 * worse case, the indexs blocks spread over different block groups
5733 * If datablocks are discontiguous, they are possible to spread over
5734 * different block groups too. If they are contiguous, with flexbg,
5735 * they could still across block group boundary.
5737 * Also account for superblock, inode, quota and xattr blocks
5739 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5742 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5748 * How many index blocks need to touch to map @lblocks logical blocks
5749 * to @pextents physical extents?
5751 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5756 * Now let's see how many group bitmaps and group descriptors need
5759 groups = idxblocks + pextents;
5761 if (groups > ngroups)
5763 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5764 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5766 /* bitmaps and block group descriptor blocks */
5767 ret += groups + gdpblocks;
5769 /* Blocks for super block, inode, quota and xattr blocks */
5770 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5776 * Calculate the total number of credits to reserve to fit
5777 * the modification of a single pages into a single transaction,
5778 * which may include multiple chunks of block allocations.
5780 * This could be called via ext4_write_begin()
5782 * We need to consider the worse case, when
5783 * one new block per extent.
5785 int ext4_writepage_trans_blocks(struct inode *inode)
5787 int bpp = ext4_journal_blocks_per_page(inode);
5790 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5792 /* Account for data blocks for journalled mode */
5793 if (ext4_should_journal_data(inode))
5799 * Calculate the journal credits for a chunk of data modification.
5801 * This is called from DIO, fallocate or whoever calling
5802 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5804 * journal buffers for data blocks are not included here, as DIO
5805 * and fallocate do no need to journal data buffers.
5807 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5809 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5813 * The caller must have previously called ext4_reserve_inode_write().
5814 * Give this, we know that the caller already has write access to iloc->bh.
5816 int ext4_mark_iloc_dirty(handle_t *handle,
5817 struct inode *inode, struct ext4_iloc *iloc)
5821 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5825 if (IS_I_VERSION(inode))
5826 inode_inc_iversion(inode);
5828 /* the do_update_inode consumes one bh->b_count */
5831 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5832 err = ext4_do_update_inode(handle, inode, iloc);
5838 * On success, We end up with an outstanding reference count against
5839 * iloc->bh. This _must_ be cleaned up later.
5843 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5844 struct ext4_iloc *iloc)
5848 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5851 err = ext4_get_inode_loc(inode, iloc);
5853 BUFFER_TRACE(iloc->bh, "get_write_access");
5854 err = ext4_journal_get_write_access(handle, iloc->bh);
5860 ext4_std_error(inode->i_sb, err);
5864 static int __ext4_expand_extra_isize(struct inode *inode,
5865 unsigned int new_extra_isize,
5866 struct ext4_iloc *iloc,
5867 handle_t *handle, int *no_expand)
5869 struct ext4_inode *raw_inode;
5870 struct ext4_xattr_ibody_header *header;
5871 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5872 struct ext4_inode_info *ei = EXT4_I(inode);
5875 /* this was checked at iget time, but double check for good measure */
5876 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5877 (ei->i_extra_isize & 3)) {
5878 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5880 EXT4_INODE_SIZE(inode->i_sb));
5881 return -EFSCORRUPTED;
5883 if ((new_extra_isize < ei->i_extra_isize) ||
5884 (new_extra_isize < 4) ||
5885 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5886 return -EINVAL; /* Should never happen */
5888 raw_inode = ext4_raw_inode(iloc);
5890 header = IHDR(inode, raw_inode);
5892 /* No extended attributes present */
5893 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5894 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5895 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5896 EXT4_I(inode)->i_extra_isize, 0,
5897 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5898 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5903 * We may need to allocate external xattr block so we need quotas
5904 * initialized. Here we can be called with various locks held so we
5905 * cannot affort to initialize quotas ourselves. So just bail.
5907 if (dquot_initialize_needed(inode))
5910 /* try to expand with EAs present */
5911 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5915 * Inode size expansion failed; don't try again
5924 * Expand an inode by new_extra_isize bytes.
5925 * Returns 0 on success or negative error number on failure.
5927 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5928 unsigned int new_extra_isize,
5929 struct ext4_iloc iloc,
5935 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5939 * In nojournal mode, we can immediately attempt to expand
5940 * the inode. When journaled, we first need to obtain extra
5941 * buffer credits since we may write into the EA block
5942 * with this same handle. If journal_extend fails, then it will
5943 * only result in a minor loss of functionality for that inode.
5944 * If this is felt to be critical, then e2fsck should be run to
5945 * force a large enough s_min_extra_isize.
5947 if (ext4_handle_valid(handle) &&
5948 jbd2_journal_extend(handle,
5949 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0)
5952 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5955 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5956 handle, &no_expand);
5957 ext4_write_unlock_xattr(inode, &no_expand);
5962 int ext4_expand_extra_isize(struct inode *inode,
5963 unsigned int new_extra_isize,
5964 struct ext4_iloc *iloc)
5970 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5975 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5976 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5977 if (IS_ERR(handle)) {
5978 error = PTR_ERR(handle);
5983 ext4_write_lock_xattr(inode, &no_expand);
5985 BUFFER_TRACE(iloc->bh, "get_write_access");
5986 error = ext4_journal_get_write_access(handle, iloc->bh);
5992 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5993 handle, &no_expand);
5995 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
6000 ext4_write_unlock_xattr(inode, &no_expand);
6001 ext4_journal_stop(handle);
6006 * What we do here is to mark the in-core inode as clean with respect to inode
6007 * dirtiness (it may still be data-dirty).
6008 * This means that the in-core inode may be reaped by prune_icache
6009 * without having to perform any I/O. This is a very good thing,
6010 * because *any* task may call prune_icache - even ones which
6011 * have a transaction open against a different journal.
6013 * Is this cheating? Not really. Sure, we haven't written the
6014 * inode out, but prune_icache isn't a user-visible syncing function.
6015 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6016 * we start and wait on commits.
6018 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
6020 struct ext4_iloc iloc;
6021 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6025 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6026 err = ext4_reserve_inode_write(handle, inode, &iloc);
6030 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6031 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6034 return ext4_mark_iloc_dirty(handle, inode, &iloc);
6038 * ext4_dirty_inode() is called from __mark_inode_dirty()
6040 * We're really interested in the case where a file is being extended.
6041 * i_size has been changed by generic_commit_write() and we thus need
6042 * to include the updated inode in the current transaction.
6044 * Also, dquot_alloc_block() will always dirty the inode when blocks
6045 * are allocated to the file.
6047 * If the inode is marked synchronous, we don't honour that here - doing
6048 * so would cause a commit on atime updates, which we don't bother doing.
6049 * We handle synchronous inodes at the highest possible level.
6051 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
6052 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
6053 * to copy into the on-disk inode structure are the timestamp files.
6055 void ext4_dirty_inode(struct inode *inode, int flags)
6059 if (flags == I_DIRTY_TIME)
6061 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6065 ext4_mark_inode_dirty(handle, inode);
6067 ext4_journal_stop(handle);
6074 * Bind an inode's backing buffer_head into this transaction, to prevent
6075 * it from being flushed to disk early. Unlike
6076 * ext4_reserve_inode_write, this leaves behind no bh reference and
6077 * returns no iloc structure, so the caller needs to repeat the iloc
6078 * lookup to mark the inode dirty later.
6080 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
6082 struct ext4_iloc iloc;
6086 err = ext4_get_inode_loc(inode, &iloc);
6088 BUFFER_TRACE(iloc.bh, "get_write_access");
6089 err = jbd2_journal_get_write_access(handle, iloc.bh);
6091 err = ext4_handle_dirty_metadata(handle,
6097 ext4_std_error(inode->i_sb, err);
6102 int ext4_change_inode_journal_flag(struct inode *inode, int val)
6107 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6110 * We have to be very careful here: changing a data block's
6111 * journaling status dynamically is dangerous. If we write a
6112 * data block to the journal, change the status and then delete
6113 * that block, we risk forgetting to revoke the old log record
6114 * from the journal and so a subsequent replay can corrupt data.
6115 * So, first we make sure that the journal is empty and that
6116 * nobody is changing anything.
6119 journal = EXT4_JOURNAL(inode);
6122 if (is_journal_aborted(journal))
6125 /* Wait for all existing dio workers */
6126 ext4_inode_block_unlocked_dio(inode);
6127 inode_dio_wait(inode);
6130 * Before flushing the journal and switching inode's aops, we have
6131 * to flush all dirty data the inode has. There can be outstanding
6132 * delayed allocations, there can be unwritten extents created by
6133 * fallocate or buffered writes in dioread_nolock mode covered by
6134 * dirty data which can be converted only after flushing the dirty
6135 * data (and journalled aops don't know how to handle these cases).
6138 down_write(&EXT4_I(inode)->i_mmap_sem);
6139 err = filemap_write_and_wait(inode->i_mapping);
6141 up_write(&EXT4_I(inode)->i_mmap_sem);
6142 ext4_inode_resume_unlocked_dio(inode);
6147 percpu_down_write(&sbi->s_writepages_rwsem);
6148 jbd2_journal_lock_updates(journal);
6151 * OK, there are no updates running now, and all cached data is
6152 * synced to disk. We are now in a completely consistent state
6153 * which doesn't have anything in the journal, and we know that
6154 * no filesystem updates are running, so it is safe to modify
6155 * the inode's in-core data-journaling state flag now.
6159 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6161 err = jbd2_journal_flush(journal);
6163 jbd2_journal_unlock_updates(journal);
6164 percpu_up_write(&sbi->s_writepages_rwsem);
6165 ext4_inode_resume_unlocked_dio(inode);
6168 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6170 ext4_set_aops(inode);
6172 jbd2_journal_unlock_updates(journal);
6173 percpu_up_write(&sbi->s_writepages_rwsem);
6176 up_write(&EXT4_I(inode)->i_mmap_sem);
6177 ext4_inode_resume_unlocked_dio(inode);
6179 /* Finally we can mark the inode as dirty. */
6181 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6183 return PTR_ERR(handle);
6185 err = ext4_mark_inode_dirty(handle, inode);
6186 ext4_handle_sync(handle);
6187 ext4_journal_stop(handle);
6188 ext4_std_error(inode->i_sb, err);
6193 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6195 return !buffer_mapped(bh);
6198 int ext4_page_mkwrite(struct vm_fault *vmf)
6200 struct vm_area_struct *vma = vmf->vma;
6201 struct page *page = vmf->page;
6205 struct file *file = vma->vm_file;
6206 struct inode *inode = file_inode(file);
6207 struct address_space *mapping = inode->i_mapping;
6209 get_block_t *get_block;
6212 if (unlikely(IS_IMMUTABLE(inode)))
6213 return VM_FAULT_SIGBUS;
6215 sb_start_pagefault(inode->i_sb);
6216 file_update_time(vma->vm_file);
6218 down_read(&EXT4_I(inode)->i_mmap_sem);
6220 ret = ext4_convert_inline_data(inode);
6224 /* Delalloc case is easy... */
6225 if (test_opt(inode->i_sb, DELALLOC) &&
6226 !ext4_should_journal_data(inode) &&
6227 !ext4_nonda_switch(inode->i_sb)) {
6229 ret = block_page_mkwrite(vma, vmf,
6230 ext4_da_get_block_prep);
6231 } while (ret == -ENOSPC &&
6232 ext4_should_retry_alloc(inode->i_sb, &retries));
6237 size = i_size_read(inode);
6238 /* Page got truncated from under us? */
6239 if (page->mapping != mapping || page_offset(page) > size) {
6241 ret = VM_FAULT_NOPAGE;
6245 if (page->index == size >> PAGE_SHIFT)
6246 len = size & ~PAGE_MASK;
6250 * Return if we have all the buffers mapped. This avoids the need to do
6251 * journal_start/journal_stop which can block and take a long time
6253 if (page_has_buffers(page)) {
6254 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6256 ext4_bh_unmapped)) {
6257 /* Wait so that we don't change page under IO */
6258 wait_for_stable_page(page);
6259 ret = VM_FAULT_LOCKED;
6264 /* OK, we need to fill the hole... */
6265 if (ext4_should_dioread_nolock(inode))
6266 get_block = ext4_get_block_unwritten;
6268 get_block = ext4_get_block;
6270 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6271 ext4_writepage_trans_blocks(inode));
6272 if (IS_ERR(handle)) {
6273 ret = VM_FAULT_SIGBUS;
6276 ret = block_page_mkwrite(vma, vmf, get_block);
6277 if (!ret && ext4_should_journal_data(inode)) {
6278 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
6279 PAGE_SIZE, NULL, do_journal_get_write_access)) {
6281 ret = VM_FAULT_SIGBUS;
6282 ext4_journal_stop(handle);
6285 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6287 ext4_journal_stop(handle);
6288 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6291 ret = block_page_mkwrite_return(ret);
6293 up_read(&EXT4_I(inode)->i_mmap_sem);
6294 sb_end_pagefault(inode->i_sb);
6298 int ext4_filemap_fault(struct vm_fault *vmf)
6300 struct inode *inode = file_inode(vmf->vma->vm_file);
6303 down_read(&EXT4_I(inode)->i_mmap_sem);
6304 err = filemap_fault(vmf);
6305 up_read(&EXT4_I(inode)->i_mmap_sem);
6311 * Find the first extent at or after @lblk in an inode that is not a hole.
6312 * Search for @map_len blocks at most. The extent is returned in @result.
6314 * The function returns 1 if we found an extent. The function returns 0 in
6315 * case there is no extent at or after @lblk and in that case also sets
6316 * @result->es_len to 0. In case of error, the error code is returned.
6318 int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk,
6319 unsigned int map_len, struct extent_status *result)
6321 struct ext4_map_blocks map;
6322 struct extent_status es = {};
6326 map.m_len = map_len;
6329 * For non-extent based files this loop may iterate several times since
6330 * we do not determine full hole size.
6332 while (map.m_len > 0) {
6333 ret = ext4_map_blocks(NULL, inode, &map, 0);
6336 /* There's extent covering m_lblk? Just return it. */
6340 ext4_es_store_pblock(result, map.m_pblk);
6341 result->es_lblk = map.m_lblk;
6342 result->es_len = map.m_len;
6343 if (map.m_flags & EXT4_MAP_UNWRITTEN)
6344 status = EXTENT_STATUS_UNWRITTEN;
6346 status = EXTENT_STATUS_WRITTEN;
6347 ext4_es_store_status(result, status);
6350 ext4_es_find_delayed_extent_range(inode, map.m_lblk,
6351 map.m_lblk + map.m_len - 1,
6353 /* Is delalloc data before next block in extent tree? */
6354 if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) {
6355 ext4_lblk_t offset = 0;
6357 if (es.es_lblk < lblk)
6358 offset = lblk - es.es_lblk;
6359 result->es_lblk = es.es_lblk + offset;
6360 ext4_es_store_pblock(result,
6361 ext4_es_pblock(&es) + offset);
6362 result->es_len = es.es_len - offset;
6363 ext4_es_store_status(result, ext4_es_status(&es));
6367 /* There's a hole at m_lblk, advance us after it */
6368 map.m_lblk += map.m_len;
6369 map_len -= map.m_len;
6370 map.m_len = map_len;