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 return S_ISLNK(inode->i_mode) && inode->i_size &&
153 (inode->i_size < EXT4_N_BLOCKS * 4);
157 * Restart the transaction associated with *handle. This does a commit,
158 * so before we call here everything must be consistently dirtied against
161 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
167 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
168 * moment, get_block can be called only for blocks inside i_size since
169 * page cache has been already dropped and writes are blocked by
170 * i_mutex. So we can safely drop the i_data_sem here.
172 BUG_ON(EXT4_JOURNAL(inode) == NULL);
173 jbd_debug(2, "restarting handle %p\n", handle);
174 up_write(&EXT4_I(inode)->i_data_sem);
175 ret = ext4_journal_restart(handle, nblocks);
176 down_write(&EXT4_I(inode)->i_data_sem);
177 ext4_discard_preallocations(inode);
183 * Called at the last iput() if i_nlink is zero.
185 void ext4_evict_inode(struct inode *inode)
189 int extra_credits = 3;
190 struct ext4_xattr_inode_array *ea_inode_array = NULL;
192 trace_ext4_evict_inode(inode);
194 if (inode->i_nlink) {
196 * When journalling data dirty buffers are tracked only in the
197 * journal. So although mm thinks everything is clean and
198 * ready for reaping the inode might still have some pages to
199 * write in the running transaction or waiting to be
200 * checkpointed. Thus calling jbd2_journal_invalidatepage()
201 * (via truncate_inode_pages()) to discard these buffers can
202 * cause data loss. Also even if we did not discard these
203 * buffers, we would have no way to find them after the inode
204 * is reaped and thus user could see stale data if he tries to
205 * read them before the transaction is checkpointed. So be
206 * careful and force everything to disk here... We use
207 * ei->i_datasync_tid to store the newest transaction
208 * containing inode's data.
210 * Note that directories do not have this problem because they
211 * don't use page cache.
213 if (inode->i_ino != EXT4_JOURNAL_INO &&
214 ext4_should_journal_data(inode) &&
215 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
216 inode->i_data.nrpages) {
217 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
218 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
220 jbd2_complete_transaction(journal, commit_tid);
221 filemap_write_and_wait(&inode->i_data);
223 truncate_inode_pages_final(&inode->i_data);
228 if (is_bad_inode(inode))
230 dquot_initialize(inode);
232 if (ext4_should_order_data(inode))
233 ext4_begin_ordered_truncate(inode, 0);
234 truncate_inode_pages_final(&inode->i_data);
237 * Protect us against freezing - iput() caller didn't have to have any
238 * protection against it
240 sb_start_intwrite(inode->i_sb);
242 if (!IS_NOQUOTA(inode))
243 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
245 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
246 ext4_blocks_for_truncate(inode)+extra_credits);
247 if (IS_ERR(handle)) {
248 ext4_std_error(inode->i_sb, PTR_ERR(handle));
250 * If we're going to skip the normal cleanup, we still need to
251 * make sure that the in-core orphan linked list is properly
254 ext4_orphan_del(NULL, inode);
255 sb_end_intwrite(inode->i_sb);
260 ext4_handle_sync(handle);
263 * Set inode->i_size to 0 before calling ext4_truncate(). We need
264 * special handling of symlinks here because i_size is used to
265 * determine whether ext4_inode_info->i_data contains symlink data or
266 * block mappings. Setting i_size to 0 will remove its fast symlink
267 * status. Erase i_data so that it becomes a valid empty block map.
269 if (ext4_inode_is_fast_symlink(inode))
270 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
272 err = ext4_mark_inode_dirty(handle, inode);
274 ext4_warning(inode->i_sb,
275 "couldn't mark inode dirty (err %d)", err);
278 if (inode->i_blocks) {
279 err = ext4_truncate(inode);
281 ext4_error(inode->i_sb,
282 "couldn't truncate inode %lu (err %d)",
288 /* Remove xattr references. */
289 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
292 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
294 ext4_journal_stop(handle);
295 ext4_orphan_del(NULL, inode);
296 sb_end_intwrite(inode->i_sb);
297 ext4_xattr_inode_array_free(ea_inode_array);
302 * Kill off the orphan record which ext4_truncate created.
303 * AKPM: I think this can be inside the above `if'.
304 * Note that ext4_orphan_del() has to be able to cope with the
305 * deletion of a non-existent orphan - this is because we don't
306 * know if ext4_truncate() actually created an orphan record.
307 * (Well, we could do this if we need to, but heck - it works)
309 ext4_orphan_del(handle, inode);
310 EXT4_I(inode)->i_dtime = get_seconds();
313 * One subtle ordering requirement: if anything has gone wrong
314 * (transaction abort, IO errors, whatever), then we can still
315 * do these next steps (the fs will already have been marked as
316 * having errors), but we can't free the inode if the mark_dirty
319 if (ext4_mark_inode_dirty(handle, inode))
320 /* If that failed, just do the required in-core inode clear. */
321 ext4_clear_inode(inode);
323 ext4_free_inode(handle, inode);
324 ext4_journal_stop(handle);
325 sb_end_intwrite(inode->i_sb);
326 ext4_xattr_inode_array_free(ea_inode_array);
329 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
333 qsize_t *ext4_get_reserved_space(struct inode *inode)
335 return &EXT4_I(inode)->i_reserved_quota;
340 * Called with i_data_sem down, which is important since we can call
341 * ext4_discard_preallocations() from here.
343 void ext4_da_update_reserve_space(struct inode *inode,
344 int used, int quota_claim)
346 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
347 struct ext4_inode_info *ei = EXT4_I(inode);
349 spin_lock(&ei->i_block_reservation_lock);
350 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
351 if (unlikely(used > ei->i_reserved_data_blocks)) {
352 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
353 "with only %d reserved data blocks",
354 __func__, inode->i_ino, used,
355 ei->i_reserved_data_blocks);
357 used = ei->i_reserved_data_blocks;
360 /* Update per-inode reservations */
361 ei->i_reserved_data_blocks -= used;
362 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
364 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
366 /* Update quota subsystem for data blocks */
368 dquot_claim_block(inode, EXT4_C2B(sbi, used));
371 * We did fallocate with an offset that is already delayed
372 * allocated. So on delayed allocated writeback we should
373 * not re-claim the quota for fallocated blocks.
375 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
379 * If we have done all the pending block allocations and if
380 * there aren't any writers on the inode, we can discard the
381 * inode's preallocations.
383 if ((ei->i_reserved_data_blocks == 0) &&
384 (atomic_read(&inode->i_writecount) == 0))
385 ext4_discard_preallocations(inode);
388 static int __check_block_validity(struct inode *inode, const char *func,
390 struct ext4_map_blocks *map)
392 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
394 ext4_error_inode(inode, func, line, map->m_pblk,
395 "lblock %lu mapped to illegal pblock "
396 "(length %d)", (unsigned long) map->m_lblk,
398 return -EFSCORRUPTED;
403 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
408 if (ext4_encrypted_inode(inode))
409 return fscrypt_zeroout_range(inode, lblk, pblk, len);
411 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
418 #define check_block_validity(inode, map) \
419 __check_block_validity((inode), __func__, __LINE__, (map))
421 #ifdef ES_AGGRESSIVE_TEST
422 static void ext4_map_blocks_es_recheck(handle_t *handle,
424 struct ext4_map_blocks *es_map,
425 struct ext4_map_blocks *map,
432 * There is a race window that the result is not the same.
433 * e.g. xfstests #223 when dioread_nolock enables. The reason
434 * is that we lookup a block mapping in extent status tree with
435 * out taking i_data_sem. So at the time the unwritten extent
436 * could be converted.
438 down_read(&EXT4_I(inode)->i_data_sem);
439 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
440 retval = ext4_ext_map_blocks(handle, inode, map, flags &
441 EXT4_GET_BLOCKS_KEEP_SIZE);
443 retval = ext4_ind_map_blocks(handle, inode, map, flags &
444 EXT4_GET_BLOCKS_KEEP_SIZE);
446 up_read((&EXT4_I(inode)->i_data_sem));
449 * We don't check m_len because extent will be collpased in status
450 * tree. So the m_len might not equal.
452 if (es_map->m_lblk != map->m_lblk ||
453 es_map->m_flags != map->m_flags ||
454 es_map->m_pblk != map->m_pblk) {
455 printk("ES cache assertion failed for inode: %lu "
456 "es_cached ex [%d/%d/%llu/%x] != "
457 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
458 inode->i_ino, es_map->m_lblk, es_map->m_len,
459 es_map->m_pblk, es_map->m_flags, map->m_lblk,
460 map->m_len, map->m_pblk, map->m_flags,
464 #endif /* ES_AGGRESSIVE_TEST */
467 * The ext4_map_blocks() function tries to look up the requested blocks,
468 * and returns if the blocks are already mapped.
470 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
471 * and store the allocated blocks in the result buffer head and mark it
474 * If file type is extents based, it will call ext4_ext_map_blocks(),
475 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
478 * On success, it returns the number of blocks being mapped or allocated. if
479 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
480 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
482 * It returns 0 if plain look up failed (blocks have not been allocated), in
483 * that case, @map is returned as unmapped but we still do fill map->m_len to
484 * indicate the length of a hole starting at map->m_lblk.
486 * It returns the error in case of allocation failure.
488 int ext4_map_blocks(handle_t *handle, struct inode *inode,
489 struct ext4_map_blocks *map, int flags)
491 struct extent_status es;
494 #ifdef ES_AGGRESSIVE_TEST
495 struct ext4_map_blocks orig_map;
497 memcpy(&orig_map, map, sizeof(*map));
501 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
502 "logical block %lu\n", inode->i_ino, flags, map->m_len,
503 (unsigned long) map->m_lblk);
506 * ext4_map_blocks returns an int, and m_len is an unsigned int
508 if (unlikely(map->m_len > INT_MAX))
509 map->m_len = INT_MAX;
511 /* We can handle the block number less than EXT_MAX_BLOCKS */
512 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
513 return -EFSCORRUPTED;
515 /* Lookup extent status tree firstly */
516 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
517 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
518 map->m_pblk = ext4_es_pblock(&es) +
519 map->m_lblk - es.es_lblk;
520 map->m_flags |= ext4_es_is_written(&es) ?
521 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
522 retval = es.es_len - (map->m_lblk - es.es_lblk);
523 if (retval > map->m_len)
526 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
528 retval = es.es_len - (map->m_lblk - es.es_lblk);
529 if (retval > map->m_len)
536 #ifdef ES_AGGRESSIVE_TEST
537 ext4_map_blocks_es_recheck(handle, inode, map,
544 * Try to see if we can get the block without requesting a new
547 down_read(&EXT4_I(inode)->i_data_sem);
548 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
549 retval = ext4_ext_map_blocks(handle, inode, map, flags &
550 EXT4_GET_BLOCKS_KEEP_SIZE);
552 retval = ext4_ind_map_blocks(handle, inode, map, flags &
553 EXT4_GET_BLOCKS_KEEP_SIZE);
558 if (unlikely(retval != map->m_len)) {
559 ext4_warning(inode->i_sb,
560 "ES len assertion failed for inode "
561 "%lu: retval %d != map->m_len %d",
562 inode->i_ino, retval, map->m_len);
566 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
567 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
568 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
569 !(status & EXTENT_STATUS_WRITTEN) &&
570 ext4_find_delalloc_range(inode, map->m_lblk,
571 map->m_lblk + map->m_len - 1))
572 status |= EXTENT_STATUS_DELAYED;
573 ret = ext4_es_insert_extent(inode, map->m_lblk,
574 map->m_len, map->m_pblk, status);
578 up_read((&EXT4_I(inode)->i_data_sem));
581 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
582 ret = check_block_validity(inode, map);
587 /* If it is only a block(s) look up */
588 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
592 * Returns if the blocks have already allocated
594 * Note that if blocks have been preallocated
595 * ext4_ext_get_block() returns the create = 0
596 * with buffer head unmapped.
598 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
600 * If we need to convert extent to unwritten
601 * we continue and do the actual work in
602 * ext4_ext_map_blocks()
604 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
608 * Here we clear m_flags because after allocating an new extent,
609 * it will be set again.
611 map->m_flags &= ~EXT4_MAP_FLAGS;
614 * New blocks allocate and/or writing to unwritten extent
615 * will possibly result in updating i_data, so we take
616 * the write lock of i_data_sem, and call get_block()
617 * with create == 1 flag.
619 down_write(&EXT4_I(inode)->i_data_sem);
622 * We need to check for EXT4 here because migrate
623 * could have changed the inode type in between
625 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
626 retval = ext4_ext_map_blocks(handle, inode, map, flags);
628 retval = ext4_ind_map_blocks(handle, inode, map, flags);
630 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
632 * We allocated new blocks which will result in
633 * i_data's format changing. Force the migrate
634 * to fail by clearing migrate flags
636 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
640 * Update reserved blocks/metadata blocks after successful
641 * block allocation which had been deferred till now. We don't
642 * support fallocate for non extent files. So we can update
643 * reserve space here.
646 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
647 ext4_da_update_reserve_space(inode, retval, 1);
653 if (unlikely(retval != map->m_len)) {
654 ext4_warning(inode->i_sb,
655 "ES len assertion failed for inode "
656 "%lu: retval %d != map->m_len %d",
657 inode->i_ino, retval, map->m_len);
662 * We have to zeroout blocks before inserting them into extent
663 * status tree. Otherwise someone could look them up there and
664 * use them before they are really zeroed. We also have to
665 * unmap metadata before zeroing as otherwise writeback can
666 * overwrite zeros with stale data from block device.
668 if (flags & EXT4_GET_BLOCKS_ZERO &&
669 map->m_flags & EXT4_MAP_MAPPED &&
670 map->m_flags & EXT4_MAP_NEW) {
671 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
673 ret = ext4_issue_zeroout(inode, map->m_lblk,
674 map->m_pblk, map->m_len);
682 * If the extent has been zeroed out, we don't need to update
683 * extent status tree.
685 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
686 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
687 if (ext4_es_is_written(&es))
690 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
691 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
692 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
693 !(status & EXTENT_STATUS_WRITTEN) &&
694 ext4_find_delalloc_range(inode, map->m_lblk,
695 map->m_lblk + map->m_len - 1))
696 status |= EXTENT_STATUS_DELAYED;
697 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
698 map->m_pblk, status);
706 up_write((&EXT4_I(inode)->i_data_sem));
707 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
708 ret = check_block_validity(inode, map);
713 * Inodes with freshly allocated blocks where contents will be
714 * visible after transaction commit must be on transaction's
717 if (map->m_flags & EXT4_MAP_NEW &&
718 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
719 !(flags & EXT4_GET_BLOCKS_ZERO) &&
720 !ext4_is_quota_file(inode) &&
721 ext4_should_order_data(inode)) {
722 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
723 ret = ext4_jbd2_inode_add_wait(handle, inode);
725 ret = ext4_jbd2_inode_add_write(handle, inode);
734 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
735 * we have to be careful as someone else may be manipulating b_state as well.
737 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
739 unsigned long old_state;
740 unsigned long new_state;
742 flags &= EXT4_MAP_FLAGS;
744 /* Dummy buffer_head? Set non-atomically. */
746 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
750 * Someone else may be modifying b_state. Be careful! This is ugly but
751 * once we get rid of using bh as a container for mapping information
752 * to pass to / from get_block functions, this can go away.
755 old_state = READ_ONCE(bh->b_state);
756 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
758 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
761 static int _ext4_get_block(struct inode *inode, sector_t iblock,
762 struct buffer_head *bh, int flags)
764 struct ext4_map_blocks map;
767 if (ext4_has_inline_data(inode))
771 map.m_len = bh->b_size >> inode->i_blkbits;
773 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
776 map_bh(bh, inode->i_sb, map.m_pblk);
777 ext4_update_bh_state(bh, map.m_flags);
778 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
780 } else if (ret == 0) {
781 /* hole case, need to fill in bh->b_size */
782 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
787 int ext4_get_block(struct inode *inode, sector_t iblock,
788 struct buffer_head *bh, int create)
790 return _ext4_get_block(inode, iblock, bh,
791 create ? EXT4_GET_BLOCKS_CREATE : 0);
795 * Get block function used when preparing for buffered write if we require
796 * creating an unwritten extent if blocks haven't been allocated. The extent
797 * will be converted to written after the IO is complete.
799 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
800 struct buffer_head *bh_result, int create)
802 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
803 inode->i_ino, create);
804 return _ext4_get_block(inode, iblock, bh_result,
805 EXT4_GET_BLOCKS_IO_CREATE_EXT);
808 /* Maximum number of blocks we map for direct IO at once. */
809 #define DIO_MAX_BLOCKS 4096
812 * Get blocks function for the cases that need to start a transaction -
813 * generally difference cases of direct IO and DAX IO. It also handles retries
816 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
817 struct buffer_head *bh_result, int flags)
824 /* Trim mapping request to maximum we can map at once for DIO */
825 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
826 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
827 dio_credits = ext4_chunk_trans_blocks(inode,
828 bh_result->b_size >> inode->i_blkbits);
830 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
832 return PTR_ERR(handle);
834 ret = _ext4_get_block(inode, iblock, bh_result, flags);
835 ext4_journal_stop(handle);
837 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
842 /* Get block function for DIO reads and writes to inodes without extents */
843 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
844 struct buffer_head *bh, int create)
846 /* We don't expect handle for direct IO */
847 WARN_ON_ONCE(ext4_journal_current_handle());
850 return _ext4_get_block(inode, iblock, bh, 0);
851 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
855 * Get block function for AIO DIO writes when we create unwritten extent if
856 * blocks are not allocated yet. The extent will be converted to written
857 * after IO is complete.
859 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
860 sector_t iblock, struct buffer_head *bh_result, int create)
864 /* We don't expect handle for direct IO */
865 WARN_ON_ONCE(ext4_journal_current_handle());
867 ret = ext4_get_block_trans(inode, iblock, bh_result,
868 EXT4_GET_BLOCKS_IO_CREATE_EXT);
871 * When doing DIO using unwritten extents, we need io_end to convert
872 * unwritten extents to written on IO completion. We allocate io_end
873 * once we spot unwritten extent and store it in b_private. Generic
874 * DIO code keeps b_private set and furthermore passes the value to
875 * our completion callback in 'private' argument.
877 if (!ret && buffer_unwritten(bh_result)) {
878 if (!bh_result->b_private) {
879 ext4_io_end_t *io_end;
881 io_end = ext4_init_io_end(inode, GFP_KERNEL);
884 bh_result->b_private = io_end;
885 ext4_set_io_unwritten_flag(inode, io_end);
887 set_buffer_defer_completion(bh_result);
894 * Get block function for non-AIO DIO writes when we create unwritten extent if
895 * blocks are not allocated yet. The extent will be converted to written
896 * after IO is complete by ext4_direct_IO_write().
898 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
899 sector_t iblock, struct buffer_head *bh_result, int create)
903 /* We don't expect handle for direct IO */
904 WARN_ON_ONCE(ext4_journal_current_handle());
906 ret = ext4_get_block_trans(inode, iblock, bh_result,
907 EXT4_GET_BLOCKS_IO_CREATE_EXT);
910 * Mark inode as having pending DIO writes to unwritten extents.
911 * ext4_direct_IO_write() checks this flag and converts extents to
914 if (!ret && buffer_unwritten(bh_result))
915 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
920 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
921 struct buffer_head *bh_result, int create)
925 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
926 inode->i_ino, create);
927 /* We don't expect handle for direct IO */
928 WARN_ON_ONCE(ext4_journal_current_handle());
930 ret = _ext4_get_block(inode, iblock, bh_result, 0);
932 * Blocks should have been preallocated! ext4_file_write_iter() checks
935 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
942 * `handle' can be NULL if create is zero
944 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
945 ext4_lblk_t block, int map_flags)
947 struct ext4_map_blocks map;
948 struct buffer_head *bh;
949 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
952 J_ASSERT(handle != NULL || create == 0);
956 err = ext4_map_blocks(handle, inode, &map, map_flags);
959 return create ? ERR_PTR(-ENOSPC) : NULL;
963 bh = sb_getblk(inode->i_sb, map.m_pblk);
965 return ERR_PTR(-ENOMEM);
966 if (map.m_flags & EXT4_MAP_NEW) {
967 J_ASSERT(create != 0);
968 J_ASSERT(handle != NULL);
971 * Now that we do not always journal data, we should
972 * keep in mind whether this should always journal the
973 * new buffer as metadata. For now, regular file
974 * writes use ext4_get_block instead, so it's not a
978 BUFFER_TRACE(bh, "call get_create_access");
979 err = ext4_journal_get_create_access(handle, bh);
984 if (!buffer_uptodate(bh)) {
985 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
986 set_buffer_uptodate(bh);
989 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
990 err = ext4_handle_dirty_metadata(handle, inode, bh);
994 BUFFER_TRACE(bh, "not a new buffer");
1001 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1002 ext4_lblk_t block, int map_flags)
1004 struct buffer_head *bh;
1006 bh = ext4_getblk(handle, inode, block, map_flags);
1009 if (!bh || buffer_uptodate(bh))
1011 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1013 if (buffer_uptodate(bh))
1016 return ERR_PTR(-EIO);
1019 /* Read a contiguous batch of blocks. */
1020 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1021 bool wait, struct buffer_head **bhs)
1025 for (i = 0; i < bh_count; i++) {
1026 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1027 if (IS_ERR(bhs[i])) {
1028 err = PTR_ERR(bhs[i]);
1034 for (i = 0; i < bh_count; i++)
1035 /* Note that NULL bhs[i] is valid because of holes. */
1036 if (bhs[i] && !buffer_uptodate(bhs[i]))
1037 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1043 for (i = 0; i < bh_count; i++)
1045 wait_on_buffer(bhs[i]);
1047 for (i = 0; i < bh_count; i++) {
1048 if (bhs[i] && !buffer_uptodate(bhs[i])) {
1056 for (i = 0; i < bh_count; i++) {
1063 int ext4_walk_page_buffers(handle_t *handle,
1064 struct buffer_head *head,
1068 int (*fn)(handle_t *handle,
1069 struct buffer_head *bh))
1071 struct buffer_head *bh;
1072 unsigned block_start, block_end;
1073 unsigned blocksize = head->b_size;
1075 struct buffer_head *next;
1077 for (bh = head, block_start = 0;
1078 ret == 0 && (bh != head || !block_start);
1079 block_start = block_end, bh = next) {
1080 next = bh->b_this_page;
1081 block_end = block_start + blocksize;
1082 if (block_end <= from || block_start >= to) {
1083 if (partial && !buffer_uptodate(bh))
1087 err = (*fn)(handle, bh);
1095 * To preserve ordering, it is essential that the hole instantiation and
1096 * the data write be encapsulated in a single transaction. We cannot
1097 * close off a transaction and start a new one between the ext4_get_block()
1098 * and the commit_write(). So doing the jbd2_journal_start at the start of
1099 * prepare_write() is the right place.
1101 * Also, this function can nest inside ext4_writepage(). In that case, we
1102 * *know* that ext4_writepage() has generated enough buffer credits to do the
1103 * whole page. So we won't block on the journal in that case, which is good,
1104 * because the caller may be PF_MEMALLOC.
1106 * By accident, ext4 can be reentered when a transaction is open via
1107 * quota file writes. If we were to commit the transaction while thus
1108 * reentered, there can be a deadlock - we would be holding a quota
1109 * lock, and the commit would never complete if another thread had a
1110 * transaction open and was blocking on the quota lock - a ranking
1113 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1114 * will _not_ run commit under these circumstances because handle->h_ref
1115 * is elevated. We'll still have enough credits for the tiny quotafile
1118 int do_journal_get_write_access(handle_t *handle,
1119 struct buffer_head *bh)
1121 int dirty = buffer_dirty(bh);
1124 if (!buffer_mapped(bh) || buffer_freed(bh))
1127 * __block_write_begin() could have dirtied some buffers. Clean
1128 * the dirty bit as jbd2_journal_get_write_access() could complain
1129 * otherwise about fs integrity issues. Setting of the dirty bit
1130 * by __block_write_begin() isn't a real problem here as we clear
1131 * the bit before releasing a page lock and thus writeback cannot
1132 * ever write the buffer.
1135 clear_buffer_dirty(bh);
1136 BUFFER_TRACE(bh, "get write access");
1137 ret = ext4_journal_get_write_access(handle, bh);
1139 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1143 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1144 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1145 get_block_t *get_block)
1147 unsigned from = pos & (PAGE_SIZE - 1);
1148 unsigned to = from + len;
1149 struct inode *inode = page->mapping->host;
1150 unsigned block_start, block_end;
1153 unsigned blocksize = inode->i_sb->s_blocksize;
1155 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1156 bool decrypt = false;
1158 BUG_ON(!PageLocked(page));
1159 BUG_ON(from > PAGE_SIZE);
1160 BUG_ON(to > PAGE_SIZE);
1163 if (!page_has_buffers(page))
1164 create_empty_buffers(page, blocksize, 0);
1165 head = page_buffers(page);
1166 bbits = ilog2(blocksize);
1167 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1169 for (bh = head, block_start = 0; bh != head || !block_start;
1170 block++, block_start = block_end, bh = bh->b_this_page) {
1171 block_end = block_start + blocksize;
1172 if (block_end <= from || block_start >= to) {
1173 if (PageUptodate(page)) {
1174 if (!buffer_uptodate(bh))
1175 set_buffer_uptodate(bh);
1180 clear_buffer_new(bh);
1181 if (!buffer_mapped(bh)) {
1182 WARN_ON(bh->b_size != blocksize);
1183 err = get_block(inode, block, bh, 1);
1186 if (buffer_new(bh)) {
1187 clean_bdev_bh_alias(bh);
1188 if (PageUptodate(page)) {
1189 clear_buffer_new(bh);
1190 set_buffer_uptodate(bh);
1191 mark_buffer_dirty(bh);
1194 if (block_end > to || block_start < from)
1195 zero_user_segments(page, to, block_end,
1200 if (PageUptodate(page)) {
1201 if (!buffer_uptodate(bh))
1202 set_buffer_uptodate(bh);
1205 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1206 !buffer_unwritten(bh) &&
1207 (block_start < from || block_end > to)) {
1208 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1210 decrypt = ext4_encrypted_inode(inode) &&
1211 S_ISREG(inode->i_mode);
1215 * If we issued read requests, let them complete.
1217 while (wait_bh > wait) {
1218 wait_on_buffer(*--wait_bh);
1219 if (!buffer_uptodate(*wait_bh))
1223 page_zero_new_buffers(page, from, to);
1225 err = fscrypt_decrypt_page(page->mapping->host, page,
1226 PAGE_SIZE, 0, page->index);
1231 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1232 loff_t pos, unsigned len, unsigned flags,
1233 struct page **pagep, void **fsdata)
1235 struct inode *inode = mapping->host;
1236 int ret, needed_blocks;
1243 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1246 trace_ext4_write_begin(inode, pos, len, flags);
1248 * Reserve one block more for addition to orphan list in case
1249 * we allocate blocks but write fails for some reason
1251 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1252 index = pos >> PAGE_SHIFT;
1253 from = pos & (PAGE_SIZE - 1);
1256 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1257 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1266 * grab_cache_page_write_begin() can take a long time if the
1267 * system is thrashing due to memory pressure, or if the page
1268 * is being written back. So grab it first before we start
1269 * the transaction handle. This also allows us to allocate
1270 * the page (if needed) without using GFP_NOFS.
1273 page = grab_cache_page_write_begin(mapping, index, flags);
1279 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1280 if (IS_ERR(handle)) {
1282 return PTR_ERR(handle);
1286 if (page->mapping != mapping) {
1287 /* The page got truncated from under us */
1290 ext4_journal_stop(handle);
1293 /* In case writeback began while the page was unlocked */
1294 wait_for_stable_page(page);
1296 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1297 if (ext4_should_dioread_nolock(inode))
1298 ret = ext4_block_write_begin(page, pos, len,
1299 ext4_get_block_unwritten);
1301 ret = ext4_block_write_begin(page, pos, len,
1304 if (ext4_should_dioread_nolock(inode))
1305 ret = __block_write_begin(page, pos, len,
1306 ext4_get_block_unwritten);
1308 ret = __block_write_begin(page, pos, len, ext4_get_block);
1310 if (!ret && ext4_should_journal_data(inode)) {
1311 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1313 do_journal_get_write_access);
1319 * __block_write_begin may have instantiated a few blocks
1320 * outside i_size. Trim these off again. Don't need
1321 * i_size_read because we hold i_mutex.
1323 * Add inode to orphan list in case we crash before
1326 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1327 ext4_orphan_add(handle, inode);
1329 ext4_journal_stop(handle);
1330 if (pos + len > inode->i_size) {
1331 ext4_truncate_failed_write(inode);
1333 * If truncate failed early the inode might
1334 * still be on the orphan list; we need to
1335 * make sure the inode is removed from the
1336 * orphan list in that case.
1339 ext4_orphan_del(NULL, inode);
1342 if (ret == -ENOSPC &&
1343 ext4_should_retry_alloc(inode->i_sb, &retries))
1352 /* For write_end() in data=journal mode */
1353 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1356 if (!buffer_mapped(bh) || buffer_freed(bh))
1358 set_buffer_uptodate(bh);
1359 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1360 clear_buffer_meta(bh);
1361 clear_buffer_prio(bh);
1366 * We need to pick up the new inode size which generic_commit_write gave us
1367 * `file' can be NULL - eg, when called from page_symlink().
1369 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1370 * buffers are managed internally.
1372 static int ext4_write_end(struct file *file,
1373 struct address_space *mapping,
1374 loff_t pos, unsigned len, unsigned copied,
1375 struct page *page, void *fsdata)
1377 handle_t *handle = ext4_journal_current_handle();
1378 struct inode *inode = mapping->host;
1379 loff_t old_size = inode->i_size;
1381 int i_size_changed = 0;
1383 trace_ext4_write_end(inode, pos, len, copied);
1384 if (ext4_has_inline_data(inode)) {
1385 ret = ext4_write_inline_data_end(inode, pos, len,
1394 copied = block_write_end(file, mapping, pos,
1395 len, copied, page, fsdata);
1397 * it's important to update i_size while still holding page lock:
1398 * page writeout could otherwise come in and zero beyond i_size.
1400 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1405 pagecache_isize_extended(inode, old_size, pos);
1407 * Don't mark the inode dirty under page lock. First, it unnecessarily
1408 * makes the holding time of page lock longer. Second, it forces lock
1409 * ordering of page lock and transaction start for journaling
1413 ext4_mark_inode_dirty(handle, inode);
1415 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1416 /* if we have allocated more blocks and copied
1417 * less. We will have blocks allocated outside
1418 * inode->i_size. So truncate them
1420 ext4_orphan_add(handle, inode);
1422 ret2 = ext4_journal_stop(handle);
1426 if (pos + len > inode->i_size) {
1427 ext4_truncate_failed_write(inode);
1429 * If truncate failed early the inode might still be
1430 * on the orphan list; we need to make sure the inode
1431 * is removed from the orphan list in that case.
1434 ext4_orphan_del(NULL, inode);
1437 return ret ? ret : copied;
1441 * This is a private version of page_zero_new_buffers() which doesn't
1442 * set the buffer to be dirty, since in data=journalled mode we need
1443 * to call ext4_handle_dirty_metadata() instead.
1445 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1447 unsigned from, unsigned to)
1449 unsigned int block_start = 0, block_end;
1450 struct buffer_head *head, *bh;
1452 bh = head = page_buffers(page);
1454 block_end = block_start + bh->b_size;
1455 if (buffer_new(bh)) {
1456 if (block_end > from && block_start < to) {
1457 if (!PageUptodate(page)) {
1458 unsigned start, size;
1460 start = max(from, block_start);
1461 size = min(to, block_end) - start;
1463 zero_user(page, start, size);
1464 write_end_fn(handle, bh);
1466 clear_buffer_new(bh);
1469 block_start = block_end;
1470 bh = bh->b_this_page;
1471 } while (bh != head);
1474 static int ext4_journalled_write_end(struct file *file,
1475 struct address_space *mapping,
1476 loff_t pos, unsigned len, unsigned copied,
1477 struct page *page, void *fsdata)
1479 handle_t *handle = ext4_journal_current_handle();
1480 struct inode *inode = mapping->host;
1481 loff_t old_size = inode->i_size;
1485 int size_changed = 0;
1487 trace_ext4_journalled_write_end(inode, pos, len, copied);
1488 from = pos & (PAGE_SIZE - 1);
1491 BUG_ON(!ext4_handle_valid(handle));
1493 if (ext4_has_inline_data(inode)) {
1494 ret = ext4_write_inline_data_end(inode, pos, len,
1502 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1504 ext4_journalled_zero_new_buffers(handle, page, from, to);
1506 if (unlikely(copied < len))
1507 ext4_journalled_zero_new_buffers(handle, page,
1509 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1510 from + copied, &partial,
1513 SetPageUptodate(page);
1515 size_changed = ext4_update_inode_size(inode, pos + copied);
1516 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1517 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1522 pagecache_isize_extended(inode, old_size, pos);
1525 ret2 = ext4_mark_inode_dirty(handle, inode);
1530 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1531 /* if we have allocated more blocks and copied
1532 * less. We will have blocks allocated outside
1533 * inode->i_size. So truncate them
1535 ext4_orphan_add(handle, inode);
1538 ret2 = ext4_journal_stop(handle);
1541 if (pos + len > inode->i_size) {
1542 ext4_truncate_failed_write(inode);
1544 * If truncate failed early the inode might still be
1545 * on the orphan list; we need to make sure the inode
1546 * is removed from the orphan list in that case.
1549 ext4_orphan_del(NULL, inode);
1552 return ret ? ret : copied;
1556 * Reserve space for a single cluster
1558 static int ext4_da_reserve_space(struct inode *inode)
1560 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1561 struct ext4_inode_info *ei = EXT4_I(inode);
1565 * We will charge metadata quota at writeout time; this saves
1566 * us from metadata over-estimation, though we may go over by
1567 * a small amount in the end. Here we just reserve for data.
1569 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1573 spin_lock(&ei->i_block_reservation_lock);
1574 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1575 spin_unlock(&ei->i_block_reservation_lock);
1576 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1579 ei->i_reserved_data_blocks++;
1580 trace_ext4_da_reserve_space(inode);
1581 spin_unlock(&ei->i_block_reservation_lock);
1583 return 0; /* success */
1586 static void ext4_da_release_space(struct inode *inode, int to_free)
1588 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1589 struct ext4_inode_info *ei = EXT4_I(inode);
1592 return; /* Nothing to release, exit */
1594 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1596 trace_ext4_da_release_space(inode, to_free);
1597 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1599 * if there aren't enough reserved blocks, then the
1600 * counter is messed up somewhere. Since this
1601 * function is called from invalidate page, it's
1602 * harmless to return without any action.
1604 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1605 "ino %lu, to_free %d with only %d reserved "
1606 "data blocks", inode->i_ino, to_free,
1607 ei->i_reserved_data_blocks);
1609 to_free = ei->i_reserved_data_blocks;
1611 ei->i_reserved_data_blocks -= to_free;
1613 /* update fs dirty data blocks counter */
1614 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1616 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1618 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1621 static void ext4_da_page_release_reservation(struct page *page,
1622 unsigned int offset,
1623 unsigned int length)
1625 int to_release = 0, contiguous_blks = 0;
1626 struct buffer_head *head, *bh;
1627 unsigned int curr_off = 0;
1628 struct inode *inode = page->mapping->host;
1629 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1630 unsigned int stop = offset + length;
1634 BUG_ON(stop > PAGE_SIZE || stop < length);
1636 head = page_buffers(page);
1639 unsigned int next_off = curr_off + bh->b_size;
1641 if (next_off > stop)
1644 if ((offset <= curr_off) && (buffer_delay(bh))) {
1647 clear_buffer_delay(bh);
1648 } else if (contiguous_blks) {
1649 lblk = page->index <<
1650 (PAGE_SHIFT - inode->i_blkbits);
1651 lblk += (curr_off >> inode->i_blkbits) -
1653 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1654 contiguous_blks = 0;
1656 curr_off = next_off;
1657 } while ((bh = bh->b_this_page) != head);
1659 if (contiguous_blks) {
1660 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1661 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1662 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1665 /* If we have released all the blocks belonging to a cluster, then we
1666 * need to release the reserved space for that cluster. */
1667 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1668 while (num_clusters > 0) {
1669 lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1670 ((num_clusters - 1) << sbi->s_cluster_bits);
1671 if (sbi->s_cluster_ratio == 1 ||
1672 !ext4_find_delalloc_cluster(inode, lblk))
1673 ext4_da_release_space(inode, 1);
1680 * Delayed allocation stuff
1683 struct mpage_da_data {
1684 struct inode *inode;
1685 struct writeback_control *wbc;
1687 pgoff_t first_page; /* The first page to write */
1688 pgoff_t next_page; /* Current page to examine */
1689 pgoff_t last_page; /* Last page to examine */
1691 * Extent to map - this can be after first_page because that can be
1692 * fully mapped. We somewhat abuse m_flags to store whether the extent
1693 * is delalloc or unwritten.
1695 struct ext4_map_blocks map;
1696 struct ext4_io_submit io_submit; /* IO submission data */
1697 unsigned int do_map:1;
1700 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1705 struct pagevec pvec;
1706 struct inode *inode = mpd->inode;
1707 struct address_space *mapping = inode->i_mapping;
1709 /* This is necessary when next_page == 0. */
1710 if (mpd->first_page >= mpd->next_page)
1713 index = mpd->first_page;
1714 end = mpd->next_page - 1;
1716 ext4_lblk_t start, last;
1717 start = index << (PAGE_SHIFT - inode->i_blkbits);
1718 last = end << (PAGE_SHIFT - inode->i_blkbits);
1719 ext4_es_remove_extent(inode, start, last - start + 1);
1722 pagevec_init(&pvec, 0);
1723 while (index <= end) {
1724 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1727 for (i = 0; i < nr_pages; i++) {
1728 struct page *page = pvec.pages[i];
1730 BUG_ON(!PageLocked(page));
1731 BUG_ON(PageWriteback(page));
1733 if (page_mapped(page))
1734 clear_page_dirty_for_io(page);
1735 block_invalidatepage(page, 0, PAGE_SIZE);
1736 ClearPageUptodate(page);
1740 pagevec_release(&pvec);
1744 static void ext4_print_free_blocks(struct inode *inode)
1746 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1747 struct super_block *sb = inode->i_sb;
1748 struct ext4_inode_info *ei = EXT4_I(inode);
1750 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1751 EXT4_C2B(EXT4_SB(inode->i_sb),
1752 ext4_count_free_clusters(sb)));
1753 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1754 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1755 (long long) EXT4_C2B(EXT4_SB(sb),
1756 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1757 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1758 (long long) EXT4_C2B(EXT4_SB(sb),
1759 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1760 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1761 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1762 ei->i_reserved_data_blocks);
1766 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1768 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1772 * This function is grabs code from the very beginning of
1773 * ext4_map_blocks, but assumes that the caller is from delayed write
1774 * time. This function looks up the requested blocks and sets the
1775 * buffer delay bit under the protection of i_data_sem.
1777 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1778 struct ext4_map_blocks *map,
1779 struct buffer_head *bh)
1781 struct extent_status es;
1783 sector_t invalid_block = ~((sector_t) 0xffff);
1784 #ifdef ES_AGGRESSIVE_TEST
1785 struct ext4_map_blocks orig_map;
1787 memcpy(&orig_map, map, sizeof(*map));
1790 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1794 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1795 "logical block %lu\n", inode->i_ino, map->m_len,
1796 (unsigned long) map->m_lblk);
1798 /* Lookup extent status tree firstly */
1799 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1800 if (ext4_es_is_hole(&es)) {
1802 down_read(&EXT4_I(inode)->i_data_sem);
1807 * Delayed extent could be allocated by fallocate.
1808 * So we need to check it.
1810 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1811 map_bh(bh, inode->i_sb, invalid_block);
1813 set_buffer_delay(bh);
1817 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1818 retval = es.es_len - (iblock - es.es_lblk);
1819 if (retval > map->m_len)
1820 retval = map->m_len;
1821 map->m_len = retval;
1822 if (ext4_es_is_written(&es))
1823 map->m_flags |= EXT4_MAP_MAPPED;
1824 else if (ext4_es_is_unwritten(&es))
1825 map->m_flags |= EXT4_MAP_UNWRITTEN;
1829 #ifdef ES_AGGRESSIVE_TEST
1830 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1836 * Try to see if we can get the block without requesting a new
1837 * file system block.
1839 down_read(&EXT4_I(inode)->i_data_sem);
1840 if (ext4_has_inline_data(inode))
1842 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1843 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1845 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1851 * XXX: __block_prepare_write() unmaps passed block,
1855 * If the block was allocated from previously allocated cluster,
1856 * then we don't need to reserve it again. However we still need
1857 * to reserve metadata for every block we're going to write.
1859 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1860 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1861 ret = ext4_da_reserve_space(inode);
1863 /* not enough space to reserve */
1869 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1870 ~0, EXTENT_STATUS_DELAYED);
1876 map_bh(bh, inode->i_sb, invalid_block);
1878 set_buffer_delay(bh);
1879 } else if (retval > 0) {
1881 unsigned int status;
1883 if (unlikely(retval != map->m_len)) {
1884 ext4_warning(inode->i_sb,
1885 "ES len assertion failed for inode "
1886 "%lu: retval %d != map->m_len %d",
1887 inode->i_ino, retval, map->m_len);
1891 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1892 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1893 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1894 map->m_pblk, status);
1900 up_read((&EXT4_I(inode)->i_data_sem));
1906 * This is a special get_block_t callback which is used by
1907 * ext4_da_write_begin(). It will either return mapped block or
1908 * reserve space for a single block.
1910 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1911 * We also have b_blocknr = -1 and b_bdev initialized properly
1913 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1914 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1915 * initialized properly.
1917 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1918 struct buffer_head *bh, int create)
1920 struct ext4_map_blocks map;
1923 BUG_ON(create == 0);
1924 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1926 map.m_lblk = iblock;
1930 * first, we need to know whether the block is allocated already
1931 * preallocated blocks are unmapped but should treated
1932 * the same as allocated blocks.
1934 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1938 map_bh(bh, inode->i_sb, map.m_pblk);
1939 ext4_update_bh_state(bh, map.m_flags);
1941 if (buffer_unwritten(bh)) {
1942 /* A delayed write to unwritten bh should be marked
1943 * new and mapped. Mapped ensures that we don't do
1944 * get_block multiple times when we write to the same
1945 * offset and new ensures that we do proper zero out
1946 * for partial write.
1949 set_buffer_mapped(bh);
1954 static int bget_one(handle_t *handle, struct buffer_head *bh)
1960 static int bput_one(handle_t *handle, struct buffer_head *bh)
1966 static int __ext4_journalled_writepage(struct page *page,
1969 struct address_space *mapping = page->mapping;
1970 struct inode *inode = mapping->host;
1971 struct buffer_head *page_bufs = NULL;
1972 handle_t *handle = NULL;
1973 int ret = 0, err = 0;
1974 int inline_data = ext4_has_inline_data(inode);
1975 struct buffer_head *inode_bh = NULL;
1977 ClearPageChecked(page);
1980 BUG_ON(page->index != 0);
1981 BUG_ON(len > ext4_get_max_inline_size(inode));
1982 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1983 if (inode_bh == NULL)
1986 page_bufs = page_buffers(page);
1991 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1995 * We need to release the page lock before we start the
1996 * journal, so grab a reference so the page won't disappear
1997 * out from under us.
2002 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2003 ext4_writepage_trans_blocks(inode));
2004 if (IS_ERR(handle)) {
2005 ret = PTR_ERR(handle);
2007 goto out_no_pagelock;
2009 BUG_ON(!ext4_handle_valid(handle));
2013 if (page->mapping != mapping) {
2014 /* The page got truncated from under us */
2015 ext4_journal_stop(handle);
2021 BUFFER_TRACE(inode_bh, "get write access");
2022 ret = ext4_journal_get_write_access(handle, inode_bh);
2024 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2027 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2028 do_journal_get_write_access);
2030 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2035 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2036 err = ext4_journal_stop(handle);
2040 if (!ext4_has_inline_data(inode))
2041 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2043 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2052 * Note that we don't need to start a transaction unless we're journaling data
2053 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2054 * need to file the inode to the transaction's list in ordered mode because if
2055 * we are writing back data added by write(), the inode is already there and if
2056 * we are writing back data modified via mmap(), no one guarantees in which
2057 * transaction the data will hit the disk. In case we are journaling data, we
2058 * cannot start transaction directly because transaction start ranks above page
2059 * lock so we have to do some magic.
2061 * This function can get called via...
2062 * - ext4_writepages after taking page lock (have journal handle)
2063 * - journal_submit_inode_data_buffers (no journal handle)
2064 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2065 * - grab_page_cache when doing write_begin (have journal handle)
2067 * We don't do any block allocation in this function. If we have page with
2068 * multiple blocks we need to write those buffer_heads that are mapped. This
2069 * is important for mmaped based write. So if we do with blocksize 1K
2070 * truncate(f, 1024);
2071 * a = mmap(f, 0, 4096);
2073 * truncate(f, 4096);
2074 * we have in the page first buffer_head mapped via page_mkwrite call back
2075 * but other buffer_heads would be unmapped but dirty (dirty done via the
2076 * do_wp_page). So writepage should write the first block. If we modify
2077 * the mmap area beyond 1024 we will again get a page_fault and the
2078 * page_mkwrite callback will do the block allocation and mark the
2079 * buffer_heads mapped.
2081 * We redirty the page if we have any buffer_heads that is either delay or
2082 * unwritten in the page.
2084 * We can get recursively called as show below.
2086 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2089 * But since we don't do any block allocation we should not deadlock.
2090 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2092 static int ext4_writepage(struct page *page,
2093 struct writeback_control *wbc)
2098 struct buffer_head *page_bufs = NULL;
2099 struct inode *inode = page->mapping->host;
2100 struct ext4_io_submit io_submit;
2101 bool keep_towrite = false;
2103 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2104 ext4_invalidatepage(page, 0, PAGE_SIZE);
2109 trace_ext4_writepage(page);
2110 size = i_size_read(inode);
2111 if (page->index == size >> PAGE_SHIFT)
2112 len = size & ~PAGE_MASK;
2116 page_bufs = page_buffers(page);
2118 * We cannot do block allocation or other extent handling in this
2119 * function. If there are buffers needing that, we have to redirty
2120 * the page. But we may reach here when we do a journal commit via
2121 * journal_submit_inode_data_buffers() and in that case we must write
2122 * allocated buffers to achieve data=ordered mode guarantees.
2124 * Also, if there is only one buffer per page (the fs block
2125 * size == the page size), if one buffer needs block
2126 * allocation or needs to modify the extent tree to clear the
2127 * unwritten flag, we know that the page can't be written at
2128 * all, so we might as well refuse the write immediately.
2129 * Unfortunately if the block size != page size, we can't as
2130 * easily detect this case using ext4_walk_page_buffers(), but
2131 * for the extremely common case, this is an optimization that
2132 * skips a useless round trip through ext4_bio_write_page().
2134 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2135 ext4_bh_delay_or_unwritten)) {
2136 redirty_page_for_writepage(wbc, page);
2137 if ((current->flags & PF_MEMALLOC) ||
2138 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2140 * For memory cleaning there's no point in writing only
2141 * some buffers. So just bail out. Warn if we came here
2142 * from direct reclaim.
2144 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2149 keep_towrite = true;
2152 if (PageChecked(page) && ext4_should_journal_data(inode))
2154 * It's mmapped pagecache. Add buffers and journal it. There
2155 * doesn't seem much point in redirtying the page here.
2157 return __ext4_journalled_writepage(page, len);
2159 ext4_io_submit_init(&io_submit, wbc);
2160 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2161 if (!io_submit.io_end) {
2162 redirty_page_for_writepage(wbc, page);
2166 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2167 ext4_io_submit(&io_submit);
2168 /* Drop io_end reference we got from init */
2169 ext4_put_io_end_defer(io_submit.io_end);
2173 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2179 BUG_ON(page->index != mpd->first_page);
2180 clear_page_dirty_for_io(page);
2182 * We have to be very careful here! Nothing protects writeback path
2183 * against i_size changes and the page can be writeably mapped into
2184 * page tables. So an application can be growing i_size and writing
2185 * data through mmap while writeback runs. clear_page_dirty_for_io()
2186 * write-protects our page in page tables and the page cannot get
2187 * written to again until we release page lock. So only after
2188 * clear_page_dirty_for_io() we are safe to sample i_size for
2189 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2190 * on the barrier provided by TestClearPageDirty in
2191 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2192 * after page tables are updated.
2194 size = i_size_read(mpd->inode);
2195 if (page->index == size >> PAGE_SHIFT)
2196 len = size & ~PAGE_MASK;
2199 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2201 mpd->wbc->nr_to_write--;
2207 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2210 * mballoc gives us at most this number of blocks...
2211 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2212 * The rest of mballoc seems to handle chunks up to full group size.
2214 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2217 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2219 * @mpd - extent of blocks
2220 * @lblk - logical number of the block in the file
2221 * @bh - buffer head we want to add to the extent
2223 * The function is used to collect contig. blocks in the same state. If the
2224 * buffer doesn't require mapping for writeback and we haven't started the
2225 * extent of buffers to map yet, the function returns 'true' immediately - the
2226 * caller can write the buffer right away. Otherwise the function returns true
2227 * if the block has been added to the extent, false if the block couldn't be
2230 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2231 struct buffer_head *bh)
2233 struct ext4_map_blocks *map = &mpd->map;
2235 /* Buffer that doesn't need mapping for writeback? */
2236 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2237 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2238 /* So far no extent to map => we write the buffer right away */
2239 if (map->m_len == 0)
2244 /* First block in the extent? */
2245 if (map->m_len == 0) {
2246 /* We cannot map unless handle is started... */
2251 map->m_flags = bh->b_state & BH_FLAGS;
2255 /* Don't go larger than mballoc is willing to allocate */
2256 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2259 /* Can we merge the block to our big extent? */
2260 if (lblk == map->m_lblk + map->m_len &&
2261 (bh->b_state & BH_FLAGS) == map->m_flags) {
2269 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2271 * @mpd - extent of blocks for mapping
2272 * @head - the first buffer in the page
2273 * @bh - buffer we should start processing from
2274 * @lblk - logical number of the block in the file corresponding to @bh
2276 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2277 * the page for IO if all buffers in this page were mapped and there's no
2278 * accumulated extent of buffers to map or add buffers in the page to the
2279 * extent of buffers to map. The function returns 1 if the caller can continue
2280 * by processing the next page, 0 if it should stop adding buffers to the
2281 * extent to map because we cannot extend it anymore. It can also return value
2282 * < 0 in case of error during IO submission.
2284 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2285 struct buffer_head *head,
2286 struct buffer_head *bh,
2289 struct inode *inode = mpd->inode;
2291 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2292 >> inode->i_blkbits;
2295 BUG_ON(buffer_locked(bh));
2297 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2298 /* Found extent to map? */
2301 /* Buffer needs mapping and handle is not started? */
2304 /* Everything mapped so far and we hit EOF */
2307 } while (lblk++, (bh = bh->b_this_page) != head);
2308 /* So far everything mapped? Submit the page for IO. */
2309 if (mpd->map.m_len == 0) {
2310 err = mpage_submit_page(mpd, head->b_page);
2314 return lblk < blocks;
2318 * mpage_map_buffers - update buffers corresponding to changed extent and
2319 * submit fully mapped pages for IO
2321 * @mpd - description of extent to map, on return next extent to map
2323 * Scan buffers corresponding to changed extent (we expect corresponding pages
2324 * to be already locked) and update buffer state according to new extent state.
2325 * We map delalloc buffers to their physical location, clear unwritten bits,
2326 * and mark buffers as uninit when we perform writes to unwritten extents
2327 * and do extent conversion after IO is finished. If the last page is not fully
2328 * mapped, we update @map to the next extent in the last page that needs
2329 * mapping. Otherwise we submit the page for IO.
2331 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2333 struct pagevec pvec;
2335 struct inode *inode = mpd->inode;
2336 struct buffer_head *head, *bh;
2337 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2343 start = mpd->map.m_lblk >> bpp_bits;
2344 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2345 lblk = start << bpp_bits;
2346 pblock = mpd->map.m_pblk;
2348 pagevec_init(&pvec, 0);
2349 while (start <= end) {
2350 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2354 for (i = 0; i < nr_pages; i++) {
2355 struct page *page = pvec.pages[i];
2357 bh = head = page_buffers(page);
2359 if (lblk < mpd->map.m_lblk)
2361 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2363 * Buffer after end of mapped extent.
2364 * Find next buffer in the page to map.
2367 mpd->map.m_flags = 0;
2369 * FIXME: If dioread_nolock supports
2370 * blocksize < pagesize, we need to make
2371 * sure we add size mapped so far to
2372 * io_end->size as the following call
2373 * can submit the page for IO.
2375 err = mpage_process_page_bufs(mpd, head,
2377 pagevec_release(&pvec);
2382 if (buffer_delay(bh)) {
2383 clear_buffer_delay(bh);
2384 bh->b_blocknr = pblock++;
2386 clear_buffer_unwritten(bh);
2387 } while (lblk++, (bh = bh->b_this_page) != head);
2390 * FIXME: This is going to break if dioread_nolock
2391 * supports blocksize < pagesize as we will try to
2392 * convert potentially unmapped parts of inode.
2394 mpd->io_submit.io_end->size += PAGE_SIZE;
2395 /* Page fully mapped - let IO run! */
2396 err = mpage_submit_page(mpd, page);
2398 pagevec_release(&pvec);
2402 pagevec_release(&pvec);
2404 /* Extent fully mapped and matches with page boundary. We are done. */
2406 mpd->map.m_flags = 0;
2410 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2412 struct inode *inode = mpd->inode;
2413 struct ext4_map_blocks *map = &mpd->map;
2414 int get_blocks_flags;
2415 int err, dioread_nolock;
2417 trace_ext4_da_write_pages_extent(inode, map);
2419 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2420 * to convert an unwritten extent to be initialized (in the case
2421 * where we have written into one or more preallocated blocks). It is
2422 * possible that we're going to need more metadata blocks than
2423 * previously reserved. However we must not fail because we're in
2424 * writeback and there is nothing we can do about it so it might result
2425 * in data loss. So use reserved blocks to allocate metadata if
2428 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2429 * the blocks in question are delalloc blocks. This indicates
2430 * that the blocks and quotas has already been checked when
2431 * the data was copied into the page cache.
2433 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2434 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2435 EXT4_GET_BLOCKS_IO_SUBMIT;
2436 dioread_nolock = ext4_should_dioread_nolock(inode);
2438 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2439 if (map->m_flags & (1 << BH_Delay))
2440 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2442 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2445 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2446 if (!mpd->io_submit.io_end->handle &&
2447 ext4_handle_valid(handle)) {
2448 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2449 handle->h_rsv_handle = NULL;
2451 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2454 BUG_ON(map->m_len == 0);
2455 if (map->m_flags & EXT4_MAP_NEW) {
2456 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
2463 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2464 * mpd->len and submit pages underlying it for IO
2466 * @handle - handle for journal operations
2467 * @mpd - extent to map
2468 * @give_up_on_write - we set this to true iff there is a fatal error and there
2469 * is no hope of writing the data. The caller should discard
2470 * dirty pages to avoid infinite loops.
2472 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2473 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2474 * them to initialized or split the described range from larger unwritten
2475 * extent. Note that we need not map all the described range since allocation
2476 * can return less blocks or the range is covered by more unwritten extents. We
2477 * cannot map more because we are limited by reserved transaction credits. On
2478 * the other hand we always make sure that the last touched page is fully
2479 * mapped so that it can be written out (and thus forward progress is
2480 * guaranteed). After mapping we submit all mapped pages for IO.
2482 static int mpage_map_and_submit_extent(handle_t *handle,
2483 struct mpage_da_data *mpd,
2484 bool *give_up_on_write)
2486 struct inode *inode = mpd->inode;
2487 struct ext4_map_blocks *map = &mpd->map;
2492 mpd->io_submit.io_end->offset =
2493 ((loff_t)map->m_lblk) << inode->i_blkbits;
2495 err = mpage_map_one_extent(handle, mpd);
2497 struct super_block *sb = inode->i_sb;
2499 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2500 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2501 goto invalidate_dirty_pages;
2503 * Let the uper layers retry transient errors.
2504 * In the case of ENOSPC, if ext4_count_free_blocks()
2505 * is non-zero, a commit should free up blocks.
2507 if ((err == -ENOMEM) ||
2508 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2510 goto update_disksize;
2513 ext4_msg(sb, KERN_CRIT,
2514 "Delayed block allocation failed for "
2515 "inode %lu at logical offset %llu with"
2516 " max blocks %u with error %d",
2518 (unsigned long long)map->m_lblk,
2519 (unsigned)map->m_len, -err);
2520 ext4_msg(sb, KERN_CRIT,
2521 "This should not happen!! Data will "
2524 ext4_print_free_blocks(inode);
2525 invalidate_dirty_pages:
2526 *give_up_on_write = true;
2531 * Update buffer state, submit mapped pages, and get us new
2534 err = mpage_map_and_submit_buffers(mpd);
2536 goto update_disksize;
2537 } while (map->m_len);
2541 * Update on-disk size after IO is submitted. Races with
2542 * truncate are avoided by checking i_size under i_data_sem.
2544 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2545 if (disksize > EXT4_I(inode)->i_disksize) {
2549 down_write(&EXT4_I(inode)->i_data_sem);
2550 i_size = i_size_read(inode);
2551 if (disksize > i_size)
2553 if (disksize > EXT4_I(inode)->i_disksize)
2554 EXT4_I(inode)->i_disksize = disksize;
2555 up_write(&EXT4_I(inode)->i_data_sem);
2556 err2 = ext4_mark_inode_dirty(handle, inode);
2558 ext4_error(inode->i_sb,
2559 "Failed to mark inode %lu dirty",
2568 * Calculate the total number of credits to reserve for one writepages
2569 * iteration. This is called from ext4_writepages(). We map an extent of
2570 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2571 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2572 * bpp - 1 blocks in bpp different extents.
2574 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2576 int bpp = ext4_journal_blocks_per_page(inode);
2578 return ext4_meta_trans_blocks(inode,
2579 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2583 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2584 * and underlying extent to map
2586 * @mpd - where to look for pages
2588 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2589 * IO immediately. When we find a page which isn't mapped we start accumulating
2590 * extent of buffers underlying these pages that needs mapping (formed by
2591 * either delayed or unwritten buffers). We also lock the pages containing
2592 * these buffers. The extent found is returned in @mpd structure (starting at
2593 * mpd->lblk with length mpd->len blocks).
2595 * Note that this function can attach bios to one io_end structure which are
2596 * neither logically nor physically contiguous. Although it may seem as an
2597 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2598 * case as we need to track IO to all buffers underlying a page in one io_end.
2600 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2602 struct address_space *mapping = mpd->inode->i_mapping;
2603 struct pagevec pvec;
2604 unsigned int nr_pages;
2605 long left = mpd->wbc->nr_to_write;
2606 pgoff_t index = mpd->first_page;
2607 pgoff_t end = mpd->last_page;
2610 int blkbits = mpd->inode->i_blkbits;
2612 struct buffer_head *head;
2614 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2615 tag = PAGECACHE_TAG_TOWRITE;
2617 tag = PAGECACHE_TAG_DIRTY;
2619 pagevec_init(&pvec, 0);
2621 mpd->next_page = index;
2622 while (index <= end) {
2623 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2624 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2628 for (i = 0; i < nr_pages; i++) {
2629 struct page *page = pvec.pages[i];
2632 * At this point, the page may be truncated or
2633 * invalidated (changing page->mapping to NULL), or
2634 * even swizzled back from swapper_space to tmpfs file
2635 * mapping. However, page->index will not change
2636 * because we have a reference on the page.
2638 if (page->index > end)
2642 * Accumulated enough dirty pages? This doesn't apply
2643 * to WB_SYNC_ALL mode. For integrity sync we have to
2644 * keep going because someone may be concurrently
2645 * dirtying pages, and we might have synced a lot of
2646 * newly appeared dirty pages, but have not synced all
2647 * of the old dirty pages.
2649 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2652 /* If we can't merge this page, we are done. */
2653 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2658 * If the page is no longer dirty, or its mapping no
2659 * longer corresponds to inode we are writing (which
2660 * means it has been truncated or invalidated), or the
2661 * page is already under writeback and we are not doing
2662 * a data integrity writeback, skip the page
2664 if (!PageDirty(page) ||
2665 (PageWriteback(page) &&
2666 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2667 unlikely(page->mapping != mapping)) {
2672 wait_on_page_writeback(page);
2673 BUG_ON(PageWriteback(page));
2675 if (mpd->map.m_len == 0)
2676 mpd->first_page = page->index;
2677 mpd->next_page = page->index + 1;
2678 /* Add all dirty buffers to mpd */
2679 lblk = ((ext4_lblk_t)page->index) <<
2680 (PAGE_SHIFT - blkbits);
2681 head = page_buffers(page);
2682 err = mpage_process_page_bufs(mpd, head, head, lblk);
2688 pagevec_release(&pvec);
2693 pagevec_release(&pvec);
2697 static int __writepage(struct page *page, struct writeback_control *wbc,
2700 struct address_space *mapping = data;
2701 int ret = ext4_writepage(page, wbc);
2702 mapping_set_error(mapping, ret);
2706 static int ext4_writepages(struct address_space *mapping,
2707 struct writeback_control *wbc)
2709 pgoff_t writeback_index = 0;
2710 long nr_to_write = wbc->nr_to_write;
2711 int range_whole = 0;
2713 handle_t *handle = NULL;
2714 struct mpage_da_data mpd;
2715 struct inode *inode = mapping->host;
2716 int needed_blocks, rsv_blocks = 0, ret = 0;
2717 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2719 struct blk_plug plug;
2720 bool give_up_on_write = false;
2722 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2725 percpu_down_read(&sbi->s_journal_flag_rwsem);
2726 trace_ext4_writepages(inode, wbc);
2728 if (dax_mapping(mapping)) {
2729 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev,
2731 goto out_writepages;
2735 * No pages to write? This is mainly a kludge to avoid starting
2736 * a transaction for special inodes like journal inode on last iput()
2737 * because that could violate lock ordering on umount
2739 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2740 goto out_writepages;
2742 if (ext4_should_journal_data(inode)) {
2743 struct blk_plug plug;
2745 blk_start_plug(&plug);
2746 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2747 blk_finish_plug(&plug);
2748 goto out_writepages;
2752 * If the filesystem has aborted, it is read-only, so return
2753 * right away instead of dumping stack traces later on that
2754 * will obscure the real source of the problem. We test
2755 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2756 * the latter could be true if the filesystem is mounted
2757 * read-only, and in that case, ext4_writepages should
2758 * *never* be called, so if that ever happens, we would want
2761 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2762 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2764 goto out_writepages;
2767 if (ext4_should_dioread_nolock(inode)) {
2769 * We may need to convert up to one extent per block in
2770 * the page and we may dirty the inode.
2772 rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits);
2776 * If we have inline data and arrive here, it means that
2777 * we will soon create the block for the 1st page, so
2778 * we'd better clear the inline data here.
2780 if (ext4_has_inline_data(inode)) {
2781 /* Just inode will be modified... */
2782 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2783 if (IS_ERR(handle)) {
2784 ret = PTR_ERR(handle);
2785 goto out_writepages;
2787 BUG_ON(ext4_test_inode_state(inode,
2788 EXT4_STATE_MAY_INLINE_DATA));
2789 ext4_destroy_inline_data(handle, inode);
2790 ext4_journal_stop(handle);
2793 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2796 if (wbc->range_cyclic) {
2797 writeback_index = mapping->writeback_index;
2798 if (writeback_index)
2800 mpd.first_page = writeback_index;
2803 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2804 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2809 ext4_io_submit_init(&mpd.io_submit, wbc);
2811 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2812 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2814 blk_start_plug(&plug);
2817 * First writeback pages that don't need mapping - we can avoid
2818 * starting a transaction unnecessarily and also avoid being blocked
2819 * in the block layer on device congestion while having transaction
2823 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2824 if (!mpd.io_submit.io_end) {
2828 ret = mpage_prepare_extent_to_map(&mpd);
2829 /* Submit prepared bio */
2830 ext4_io_submit(&mpd.io_submit);
2831 ext4_put_io_end_defer(mpd.io_submit.io_end);
2832 mpd.io_submit.io_end = NULL;
2833 /* Unlock pages we didn't use */
2834 mpage_release_unused_pages(&mpd, false);
2838 while (!done && mpd.first_page <= mpd.last_page) {
2839 /* For each extent of pages we use new io_end */
2840 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2841 if (!mpd.io_submit.io_end) {
2847 * We have two constraints: We find one extent to map and we
2848 * must always write out whole page (makes a difference when
2849 * blocksize < pagesize) so that we don't block on IO when we
2850 * try to write out the rest of the page. Journalled mode is
2851 * not supported by delalloc.
2853 BUG_ON(ext4_should_journal_data(inode));
2854 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2856 /* start a new transaction */
2857 handle = ext4_journal_start_with_reserve(inode,
2858 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2859 if (IS_ERR(handle)) {
2860 ret = PTR_ERR(handle);
2861 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2862 "%ld pages, ino %lu; err %d", __func__,
2863 wbc->nr_to_write, inode->i_ino, ret);
2864 /* Release allocated io_end */
2865 ext4_put_io_end(mpd.io_submit.io_end);
2866 mpd.io_submit.io_end = NULL;
2871 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2872 ret = mpage_prepare_extent_to_map(&mpd);
2875 ret = mpage_map_and_submit_extent(handle, &mpd,
2879 * We scanned the whole range (or exhausted
2880 * nr_to_write), submitted what was mapped and
2881 * didn't find anything needing mapping. We are
2888 * Caution: If the handle is synchronous,
2889 * ext4_journal_stop() can wait for transaction commit
2890 * to finish which may depend on writeback of pages to
2891 * complete or on page lock to be released. In that
2892 * case, we have to wait until after after we have
2893 * submitted all the IO, released page locks we hold,
2894 * and dropped io_end reference (for extent conversion
2895 * to be able to complete) before stopping the handle.
2897 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2898 ext4_journal_stop(handle);
2902 /* Submit prepared bio */
2903 ext4_io_submit(&mpd.io_submit);
2904 /* Unlock pages we didn't use */
2905 mpage_release_unused_pages(&mpd, give_up_on_write);
2907 * Drop our io_end reference we got from init. We have
2908 * to be careful and use deferred io_end finishing if
2909 * we are still holding the transaction as we can
2910 * release the last reference to io_end which may end
2911 * up doing unwritten extent conversion.
2914 ext4_put_io_end_defer(mpd.io_submit.io_end);
2915 ext4_journal_stop(handle);
2917 ext4_put_io_end(mpd.io_submit.io_end);
2918 mpd.io_submit.io_end = NULL;
2920 if (ret == -ENOSPC && sbi->s_journal) {
2922 * Commit the transaction which would
2923 * free blocks released in the transaction
2926 jbd2_journal_force_commit_nested(sbi->s_journal);
2930 /* Fatal error - ENOMEM, EIO... */
2935 blk_finish_plug(&plug);
2936 if (!ret && !cycled && wbc->nr_to_write > 0) {
2938 mpd.last_page = writeback_index - 1;
2944 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2946 * Set the writeback_index so that range_cyclic
2947 * mode will write it back later
2949 mapping->writeback_index = mpd.first_page;
2952 trace_ext4_writepages_result(inode, wbc, ret,
2953 nr_to_write - wbc->nr_to_write);
2954 percpu_up_read(&sbi->s_journal_flag_rwsem);
2958 static int ext4_nonda_switch(struct super_block *sb)
2960 s64 free_clusters, dirty_clusters;
2961 struct ext4_sb_info *sbi = EXT4_SB(sb);
2964 * switch to non delalloc mode if we are running low
2965 * on free block. The free block accounting via percpu
2966 * counters can get slightly wrong with percpu_counter_batch getting
2967 * accumulated on each CPU without updating global counters
2968 * Delalloc need an accurate free block accounting. So switch
2969 * to non delalloc when we are near to error range.
2972 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2974 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2976 * Start pushing delalloc when 1/2 of free blocks are dirty.
2978 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2979 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2981 if (2 * free_clusters < 3 * dirty_clusters ||
2982 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2984 * free block count is less than 150% of dirty blocks
2985 * or free blocks is less than watermark
2992 /* We always reserve for an inode update; the superblock could be there too */
2993 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2995 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2998 if (pos + len <= 0x7fffffffULL)
3001 /* We might need to update the superblock to set LARGE_FILE */
3005 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3006 loff_t pos, unsigned len, unsigned flags,
3007 struct page **pagep, void **fsdata)
3009 int ret, retries = 0;
3012 struct inode *inode = mapping->host;
3015 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3018 index = pos >> PAGE_SHIFT;
3020 if (ext4_nonda_switch(inode->i_sb) ||
3021 S_ISLNK(inode->i_mode)) {
3022 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3023 return ext4_write_begin(file, mapping, pos,
3024 len, flags, pagep, fsdata);
3026 *fsdata = (void *)0;
3027 trace_ext4_da_write_begin(inode, pos, len, flags);
3029 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3030 ret = ext4_da_write_inline_data_begin(mapping, inode,
3040 * grab_cache_page_write_begin() can take a long time if the
3041 * system is thrashing due to memory pressure, or if the page
3042 * is being written back. So grab it first before we start
3043 * the transaction handle. This also allows us to allocate
3044 * the page (if needed) without using GFP_NOFS.
3047 page = grab_cache_page_write_begin(mapping, index, flags);
3053 * With delayed allocation, we don't log the i_disksize update
3054 * if there is delayed block allocation. But we still need
3055 * to journalling the i_disksize update if writes to the end
3056 * of file which has an already mapped buffer.
3059 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3060 ext4_da_write_credits(inode, pos, len));
3061 if (IS_ERR(handle)) {
3063 return PTR_ERR(handle);
3067 if (page->mapping != mapping) {
3068 /* The page got truncated from under us */
3071 ext4_journal_stop(handle);
3074 /* In case writeback began while the page was unlocked */
3075 wait_for_stable_page(page);
3077 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3078 ret = ext4_block_write_begin(page, pos, len,
3079 ext4_da_get_block_prep);
3081 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3085 ext4_journal_stop(handle);
3087 * block_write_begin may have instantiated a few blocks
3088 * outside i_size. Trim these off again. Don't need
3089 * i_size_read because we hold i_mutex.
3091 if (pos + len > inode->i_size)
3092 ext4_truncate_failed_write(inode);
3094 if (ret == -ENOSPC &&
3095 ext4_should_retry_alloc(inode->i_sb, &retries))
3107 * Check if we should update i_disksize
3108 * when write to the end of file but not require block allocation
3110 static int ext4_da_should_update_i_disksize(struct page *page,
3111 unsigned long offset)
3113 struct buffer_head *bh;
3114 struct inode *inode = page->mapping->host;
3118 bh = page_buffers(page);
3119 idx = offset >> inode->i_blkbits;
3121 for (i = 0; i < idx; i++)
3122 bh = bh->b_this_page;
3124 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3129 static int ext4_da_write_end(struct file *file,
3130 struct address_space *mapping,
3131 loff_t pos, unsigned len, unsigned copied,
3132 struct page *page, void *fsdata)
3134 struct inode *inode = mapping->host;
3136 handle_t *handle = ext4_journal_current_handle();
3138 unsigned long start, end;
3139 int write_mode = (int)(unsigned long)fsdata;
3141 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3142 return ext4_write_end(file, mapping, pos,
3143 len, copied, page, fsdata);
3145 trace_ext4_da_write_end(inode, pos, len, copied);
3146 start = pos & (PAGE_SIZE - 1);
3147 end = start + copied - 1;
3150 * generic_write_end() will run mark_inode_dirty() if i_size
3151 * changes. So let's piggyback the i_disksize mark_inode_dirty
3154 new_i_size = pos + copied;
3155 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3156 if (ext4_has_inline_data(inode) ||
3157 ext4_da_should_update_i_disksize(page, end)) {
3158 ext4_update_i_disksize(inode, new_i_size);
3159 /* We need to mark inode dirty even if
3160 * new_i_size is less that inode->i_size
3161 * bu greater than i_disksize.(hint delalloc)
3163 ext4_mark_inode_dirty(handle, inode);
3167 if (write_mode != CONVERT_INLINE_DATA &&
3168 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3169 ext4_has_inline_data(inode))
3170 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3173 ret2 = generic_write_end(file, mapping, pos, len, copied,
3179 ret2 = ext4_journal_stop(handle);
3183 return ret ? ret : copied;
3186 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3187 unsigned int length)
3190 * Drop reserved blocks
3192 BUG_ON(!PageLocked(page));
3193 if (!page_has_buffers(page))
3196 ext4_da_page_release_reservation(page, offset, length);
3199 ext4_invalidatepage(page, offset, length);
3205 * Force all delayed allocation blocks to be allocated for a given inode.
3207 int ext4_alloc_da_blocks(struct inode *inode)
3209 trace_ext4_alloc_da_blocks(inode);
3211 if (!EXT4_I(inode)->i_reserved_data_blocks)
3215 * We do something simple for now. The filemap_flush() will
3216 * also start triggering a write of the data blocks, which is
3217 * not strictly speaking necessary (and for users of
3218 * laptop_mode, not even desirable). However, to do otherwise
3219 * would require replicating code paths in:
3221 * ext4_writepages() ->
3222 * write_cache_pages() ---> (via passed in callback function)
3223 * __mpage_da_writepage() -->
3224 * mpage_add_bh_to_extent()
3225 * mpage_da_map_blocks()
3227 * The problem is that write_cache_pages(), located in
3228 * mm/page-writeback.c, marks pages clean in preparation for
3229 * doing I/O, which is not desirable if we're not planning on
3232 * We could call write_cache_pages(), and then redirty all of
3233 * the pages by calling redirty_page_for_writepage() but that
3234 * would be ugly in the extreme. So instead we would need to
3235 * replicate parts of the code in the above functions,
3236 * simplifying them because we wouldn't actually intend to
3237 * write out the pages, but rather only collect contiguous
3238 * logical block extents, call the multi-block allocator, and
3239 * then update the buffer heads with the block allocations.
3241 * For now, though, we'll cheat by calling filemap_flush(),
3242 * which will map the blocks, and start the I/O, but not
3243 * actually wait for the I/O to complete.
3245 return filemap_flush(inode->i_mapping);
3249 * bmap() is special. It gets used by applications such as lilo and by
3250 * the swapper to find the on-disk block of a specific piece of data.
3252 * Naturally, this is dangerous if the block concerned is still in the
3253 * journal. If somebody makes a swapfile on an ext4 data-journaling
3254 * filesystem and enables swap, then they may get a nasty shock when the
3255 * data getting swapped to that swapfile suddenly gets overwritten by
3256 * the original zero's written out previously to the journal and
3257 * awaiting writeback in the kernel's buffer cache.
3259 * So, if we see any bmap calls here on a modified, data-journaled file,
3260 * take extra steps to flush any blocks which might be in the cache.
3262 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3264 struct inode *inode = mapping->host;
3269 * We can get here for an inline file via the FIBMAP ioctl
3271 if (ext4_has_inline_data(inode))
3274 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3275 test_opt(inode->i_sb, DELALLOC)) {
3277 * With delalloc we want to sync the file
3278 * so that we can make sure we allocate
3281 filemap_write_and_wait(mapping);
3284 if (EXT4_JOURNAL(inode) &&
3285 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3287 * This is a REALLY heavyweight approach, but the use of
3288 * bmap on dirty files is expected to be extremely rare:
3289 * only if we run lilo or swapon on a freshly made file
3290 * do we expect this to happen.
3292 * (bmap requires CAP_SYS_RAWIO so this does not
3293 * represent an unprivileged user DOS attack --- we'd be
3294 * in trouble if mortal users could trigger this path at
3297 * NB. EXT4_STATE_JDATA is not set on files other than
3298 * regular files. If somebody wants to bmap a directory
3299 * or symlink and gets confused because the buffer
3300 * hasn't yet been flushed to disk, they deserve
3301 * everything they get.
3304 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3305 journal = EXT4_JOURNAL(inode);
3306 jbd2_journal_lock_updates(journal);
3307 err = jbd2_journal_flush(journal);
3308 jbd2_journal_unlock_updates(journal);
3314 return generic_block_bmap(mapping, block, ext4_get_block);
3317 static int ext4_readpage(struct file *file, struct page *page)
3320 struct inode *inode = page->mapping->host;
3322 trace_ext4_readpage(page);
3324 if (ext4_has_inline_data(inode))
3325 ret = ext4_readpage_inline(inode, page);
3328 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3334 ext4_readpages(struct file *file, struct address_space *mapping,
3335 struct list_head *pages, unsigned nr_pages)
3337 struct inode *inode = mapping->host;
3339 /* If the file has inline data, no need to do readpages. */
3340 if (ext4_has_inline_data(inode))
3343 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3346 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3347 unsigned int length)
3349 trace_ext4_invalidatepage(page, offset, length);
3351 /* No journalling happens on data buffers when this function is used */
3352 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3354 block_invalidatepage(page, offset, length);
3357 static int __ext4_journalled_invalidatepage(struct page *page,
3358 unsigned int offset,
3359 unsigned int length)
3361 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3363 trace_ext4_journalled_invalidatepage(page, offset, length);
3366 * If it's a full truncate we just forget about the pending dirtying
3368 if (offset == 0 && length == PAGE_SIZE)
3369 ClearPageChecked(page);
3371 return jbd2_journal_invalidatepage(journal, page, offset, length);
3374 /* Wrapper for aops... */
3375 static void ext4_journalled_invalidatepage(struct page *page,
3376 unsigned int offset,
3377 unsigned int length)
3379 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3382 static int ext4_releasepage(struct page *page, gfp_t wait)
3384 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3386 trace_ext4_releasepage(page);
3388 /* Page has dirty journalled data -> cannot release */
3389 if (PageChecked(page))
3392 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3394 return try_to_free_buffers(page);
3397 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3398 unsigned flags, struct iomap *iomap)
3400 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3401 unsigned int blkbits = inode->i_blkbits;
3402 unsigned long first_block = offset >> blkbits;
3403 unsigned long last_block = (offset + length - 1) >> blkbits;
3404 struct ext4_map_blocks map;
3405 bool delalloc = false;
3409 if (flags & IOMAP_REPORT) {
3410 if (ext4_has_inline_data(inode)) {
3411 ret = ext4_inline_data_iomap(inode, iomap);
3412 if (ret != -EAGAIN) {
3413 if (ret == 0 && offset >= iomap->length)
3419 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3423 map.m_lblk = first_block;
3424 map.m_len = last_block - first_block + 1;
3426 if (flags & IOMAP_REPORT) {
3427 ret = ext4_map_blocks(NULL, inode, &map, 0);
3432 ext4_lblk_t end = map.m_lblk + map.m_len - 1;
3433 struct extent_status es;
3435 ext4_es_find_delayed_extent_range(inode, map.m_lblk, end, &es);
3437 if (!es.es_len || es.es_lblk > end) {
3438 /* entire range is a hole */
3439 } else if (es.es_lblk > map.m_lblk) {
3440 /* range starts with a hole */
3441 map.m_len = es.es_lblk - map.m_lblk;
3443 ext4_lblk_t offs = 0;
3445 if (es.es_lblk < map.m_lblk)
3446 offs = map.m_lblk - es.es_lblk;
3447 map.m_lblk = es.es_lblk + offs;
3448 map.m_len = es.es_len - offs;
3452 } else if (flags & IOMAP_WRITE) {
3457 /* Trim mapping request to maximum we can map at once for DIO */
3458 if (map.m_len > DIO_MAX_BLOCKS)
3459 map.m_len = DIO_MAX_BLOCKS;
3460 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3463 * Either we allocate blocks and then we don't get unwritten
3464 * extent so we have reserved enough credits, or the blocks
3465 * are already allocated and unwritten and in that case
3466 * extent conversion fits in the credits as well.
3468 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3471 return PTR_ERR(handle);
3473 ret = ext4_map_blocks(handle, inode, &map,
3474 EXT4_GET_BLOCKS_CREATE_ZERO);
3476 ext4_journal_stop(handle);
3477 if (ret == -ENOSPC &&
3478 ext4_should_retry_alloc(inode->i_sb, &retries))
3484 * If we added blocks beyond i_size, we need to make sure they
3485 * will get truncated if we crash before updating i_size in
3486 * ext4_iomap_end(). For faults we don't need to do that (and
3487 * even cannot because for orphan list operations inode_lock is
3488 * required) - if we happen to instantiate block beyond i_size,
3489 * it is because we race with truncate which has already added
3490 * the inode to the orphan list.
3492 if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3493 (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3496 err = ext4_orphan_add(handle, inode);
3498 ext4_journal_stop(handle);
3502 ext4_journal_stop(handle);
3504 ret = ext4_map_blocks(NULL, inode, &map, 0);
3510 iomap->bdev = inode->i_sb->s_bdev;
3511 iomap->dax_dev = sbi->s_daxdev;
3512 iomap->offset = first_block << blkbits;
3513 iomap->length = (u64)map.m_len << blkbits;
3516 iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE;
3517 iomap->addr = IOMAP_NULL_ADDR;
3519 if (map.m_flags & EXT4_MAP_MAPPED) {
3520 iomap->type = IOMAP_MAPPED;
3521 } else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3522 iomap->type = IOMAP_UNWRITTEN;
3527 iomap->addr = (u64)map.m_pblk << blkbits;
3530 if (map.m_flags & EXT4_MAP_NEW)
3531 iomap->flags |= IOMAP_F_NEW;
3536 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3537 ssize_t written, unsigned flags, struct iomap *iomap)
3541 int blkbits = inode->i_blkbits;
3542 bool truncate = false;
3544 if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3547 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3548 if (IS_ERR(handle)) {
3549 ret = PTR_ERR(handle);
3552 if (ext4_update_inode_size(inode, offset + written))
3553 ext4_mark_inode_dirty(handle, inode);
3555 * We may need to truncate allocated but not written blocks beyond EOF.
3557 if (iomap->offset + iomap->length >
3558 ALIGN(inode->i_size, 1 << blkbits)) {
3559 ext4_lblk_t written_blk, end_blk;
3561 written_blk = (offset + written) >> blkbits;
3562 end_blk = (offset + length) >> blkbits;
3563 if (written_blk < end_blk && ext4_can_truncate(inode))
3567 * Remove inode from orphan list if we were extending a inode and
3568 * everything went fine.
3570 if (!truncate && inode->i_nlink &&
3571 !list_empty(&EXT4_I(inode)->i_orphan))
3572 ext4_orphan_del(handle, inode);
3573 ext4_journal_stop(handle);
3575 ext4_truncate_failed_write(inode);
3578 * If truncate failed early the inode might still be on the
3579 * orphan list; we need to make sure the inode is removed from
3580 * the orphan list in that case.
3583 ext4_orphan_del(NULL, inode);
3588 const struct iomap_ops ext4_iomap_ops = {
3589 .iomap_begin = ext4_iomap_begin,
3590 .iomap_end = ext4_iomap_end,
3593 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3594 ssize_t size, void *private)
3596 ext4_io_end_t *io_end = private;
3598 /* if not async direct IO just return */
3602 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3603 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3604 io_end, io_end->inode->i_ino, iocb, offset, size);
3607 * Error during AIO DIO. We cannot convert unwritten extents as the
3608 * data was not written. Just clear the unwritten flag and drop io_end.
3611 ext4_clear_io_unwritten_flag(io_end);
3614 io_end->offset = offset;
3615 io_end->size = size;
3616 ext4_put_io_end(io_end);
3622 * Handling of direct IO writes.
3624 * For ext4 extent files, ext4 will do direct-io write even to holes,
3625 * preallocated extents, and those write extend the file, no need to
3626 * fall back to buffered IO.
3628 * For holes, we fallocate those blocks, mark them as unwritten
3629 * If those blocks were preallocated, we mark sure they are split, but
3630 * still keep the range to write as unwritten.
3632 * The unwritten extents will be converted to written when DIO is completed.
3633 * For async direct IO, since the IO may still pending when return, we
3634 * set up an end_io call back function, which will do the conversion
3635 * when async direct IO completed.
3637 * If the O_DIRECT write will extend the file then add this inode to the
3638 * orphan list. So recovery will truncate it back to the original size
3639 * if the machine crashes during the write.
3642 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3644 struct file *file = iocb->ki_filp;
3645 struct inode *inode = file->f_mapping->host;
3646 struct ext4_inode_info *ei = EXT4_I(inode);
3648 loff_t offset = iocb->ki_pos;
3649 size_t count = iov_iter_count(iter);
3651 get_block_t *get_block_func = NULL;
3653 loff_t final_size = offset + count;
3657 if (final_size > inode->i_size) {
3658 /* Credits for sb + inode write */
3659 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3660 if (IS_ERR(handle)) {
3661 ret = PTR_ERR(handle);
3664 ret = ext4_orphan_add(handle, inode);
3666 ext4_journal_stop(handle);
3670 ei->i_disksize = inode->i_size;
3671 ext4_journal_stop(handle);
3674 BUG_ON(iocb->private == NULL);
3677 * Make all waiters for direct IO properly wait also for extent
3678 * conversion. This also disallows race between truncate() and
3679 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3681 inode_dio_begin(inode);
3683 /* If we do a overwrite dio, i_mutex locking can be released */
3684 overwrite = *((int *)iocb->private);
3687 inode_unlock(inode);
3690 * For extent mapped files we could direct write to holes and fallocate.
3692 * Allocated blocks to fill the hole are marked as unwritten to prevent
3693 * parallel buffered read to expose the stale data before DIO complete
3696 * As to previously fallocated extents, ext4 get_block will just simply
3697 * mark the buffer mapped but still keep the extents unwritten.
3699 * For non AIO case, we will convert those unwritten extents to written
3700 * after return back from blockdev_direct_IO. That way we save us from
3701 * allocating io_end structure and also the overhead of offloading
3702 * the extent convertion to a workqueue.
3704 * For async DIO, the conversion needs to be deferred when the
3705 * IO is completed. The ext4 end_io callback function will be
3706 * called to take care of the conversion work. Here for async
3707 * case, we allocate an io_end structure to hook to the iocb.
3709 iocb->private = NULL;
3711 get_block_func = ext4_dio_get_block_overwrite;
3712 else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3713 round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3714 get_block_func = ext4_dio_get_block;
3715 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3716 } else if (is_sync_kiocb(iocb)) {
3717 get_block_func = ext4_dio_get_block_unwritten_sync;
3718 dio_flags = DIO_LOCKING;
3720 get_block_func = ext4_dio_get_block_unwritten_async;
3721 dio_flags = DIO_LOCKING;
3723 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3724 get_block_func, ext4_end_io_dio, NULL,
3727 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3728 EXT4_STATE_DIO_UNWRITTEN)) {
3731 * for non AIO case, since the IO is already
3732 * completed, we could do the conversion right here
3734 err = ext4_convert_unwritten_extents(NULL, inode,
3738 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3741 inode_dio_end(inode);
3742 /* take i_mutex locking again if we do a ovewrite dio */
3746 if (ret < 0 && final_size > inode->i_size)
3747 ext4_truncate_failed_write(inode);
3749 /* Handle extending of i_size after direct IO write */
3753 /* Credits for sb + inode write */
3754 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3755 if (IS_ERR(handle)) {
3756 /* This is really bad luck. We've written the data
3757 * but cannot extend i_size. Bail out and pretend
3758 * the write failed... */
3759 ret = PTR_ERR(handle);
3761 ext4_orphan_del(NULL, inode);
3766 ext4_orphan_del(handle, inode);
3768 loff_t end = offset + ret;
3769 if (end > inode->i_size) {
3770 ei->i_disksize = end;
3771 i_size_write(inode, end);
3773 * We're going to return a positive `ret'
3774 * here due to non-zero-length I/O, so there's
3775 * no way of reporting error returns from
3776 * ext4_mark_inode_dirty() to userspace. So
3779 ext4_mark_inode_dirty(handle, inode);
3782 err = ext4_journal_stop(handle);
3790 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3792 struct address_space *mapping = iocb->ki_filp->f_mapping;
3793 struct inode *inode = mapping->host;
3794 size_t count = iov_iter_count(iter);
3798 * Shared inode_lock is enough for us - it protects against concurrent
3799 * writes & truncates and since we take care of writing back page cache,
3800 * we are protected against page writeback as well.
3802 inode_lock_shared(inode);
3803 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3804 iocb->ki_pos + count - 1);
3807 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3808 iter, ext4_dio_get_block, NULL, NULL, 0);
3810 inode_unlock_shared(inode);
3814 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3816 struct file *file = iocb->ki_filp;
3817 struct inode *inode = file->f_mapping->host;
3818 size_t count = iov_iter_count(iter);
3819 loff_t offset = iocb->ki_pos;
3822 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3823 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3828 * If we are doing data journalling we don't support O_DIRECT
3830 if (ext4_should_journal_data(inode))
3833 /* Let buffer I/O handle the inline data case. */
3834 if (ext4_has_inline_data(inode))
3837 /* DAX uses iomap path now */
3838 if (WARN_ON_ONCE(IS_DAX(inode)))
3841 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3842 if (iov_iter_rw(iter) == READ)
3843 ret = ext4_direct_IO_read(iocb, iter);
3845 ret = ext4_direct_IO_write(iocb, iter);
3846 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3851 * Pages can be marked dirty completely asynchronously from ext4's journalling
3852 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3853 * much here because ->set_page_dirty is called under VFS locks. The page is
3854 * not necessarily locked.
3856 * We cannot just dirty the page and leave attached buffers clean, because the
3857 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3858 * or jbddirty because all the journalling code will explode.
3860 * So what we do is to mark the page "pending dirty" and next time writepage
3861 * is called, propagate that into the buffers appropriately.
3863 static int ext4_journalled_set_page_dirty(struct page *page)
3865 SetPageChecked(page);
3866 return __set_page_dirty_nobuffers(page);
3869 static int ext4_set_page_dirty(struct page *page)
3871 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3872 WARN_ON_ONCE(!page_has_buffers(page));
3873 return __set_page_dirty_buffers(page);
3876 static const struct address_space_operations ext4_aops = {
3877 .readpage = ext4_readpage,
3878 .readpages = ext4_readpages,
3879 .writepage = ext4_writepage,
3880 .writepages = ext4_writepages,
3881 .write_begin = ext4_write_begin,
3882 .write_end = ext4_write_end,
3883 .set_page_dirty = ext4_set_page_dirty,
3885 .invalidatepage = ext4_invalidatepage,
3886 .releasepage = ext4_releasepage,
3887 .direct_IO = ext4_direct_IO,
3888 .migratepage = buffer_migrate_page,
3889 .is_partially_uptodate = block_is_partially_uptodate,
3890 .error_remove_page = generic_error_remove_page,
3893 static const struct address_space_operations ext4_journalled_aops = {
3894 .readpage = ext4_readpage,
3895 .readpages = ext4_readpages,
3896 .writepage = ext4_writepage,
3897 .writepages = ext4_writepages,
3898 .write_begin = ext4_write_begin,
3899 .write_end = ext4_journalled_write_end,
3900 .set_page_dirty = ext4_journalled_set_page_dirty,
3902 .invalidatepage = ext4_journalled_invalidatepage,
3903 .releasepage = ext4_releasepage,
3904 .direct_IO = ext4_direct_IO,
3905 .is_partially_uptodate = block_is_partially_uptodate,
3906 .error_remove_page = generic_error_remove_page,
3909 static const struct address_space_operations ext4_da_aops = {
3910 .readpage = ext4_readpage,
3911 .readpages = ext4_readpages,
3912 .writepage = ext4_writepage,
3913 .writepages = ext4_writepages,
3914 .write_begin = ext4_da_write_begin,
3915 .write_end = ext4_da_write_end,
3916 .set_page_dirty = ext4_set_page_dirty,
3918 .invalidatepage = ext4_da_invalidatepage,
3919 .releasepage = ext4_releasepage,
3920 .direct_IO = ext4_direct_IO,
3921 .migratepage = buffer_migrate_page,
3922 .is_partially_uptodate = block_is_partially_uptodate,
3923 .error_remove_page = generic_error_remove_page,
3926 void ext4_set_aops(struct inode *inode)
3928 switch (ext4_inode_journal_mode(inode)) {
3929 case EXT4_INODE_ORDERED_DATA_MODE:
3930 case EXT4_INODE_WRITEBACK_DATA_MODE:
3932 case EXT4_INODE_JOURNAL_DATA_MODE:
3933 inode->i_mapping->a_ops = &ext4_journalled_aops;
3938 if (test_opt(inode->i_sb, DELALLOC))
3939 inode->i_mapping->a_ops = &ext4_da_aops;
3941 inode->i_mapping->a_ops = &ext4_aops;
3944 static int __ext4_block_zero_page_range(handle_t *handle,
3945 struct address_space *mapping, loff_t from, loff_t length)
3947 ext4_fsblk_t index = from >> PAGE_SHIFT;
3948 unsigned offset = from & (PAGE_SIZE-1);
3949 unsigned blocksize, pos;
3951 struct inode *inode = mapping->host;
3952 struct buffer_head *bh;
3956 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3957 mapping_gfp_constraint(mapping, ~__GFP_FS));
3961 blocksize = inode->i_sb->s_blocksize;
3963 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3965 if (!page_has_buffers(page))
3966 create_empty_buffers(page, blocksize, 0);
3968 /* Find the buffer that contains "offset" */
3969 bh = page_buffers(page);
3971 while (offset >= pos) {
3972 bh = bh->b_this_page;
3976 if (buffer_freed(bh)) {
3977 BUFFER_TRACE(bh, "freed: skip");
3980 if (!buffer_mapped(bh)) {
3981 BUFFER_TRACE(bh, "unmapped");
3982 ext4_get_block(inode, iblock, bh, 0);
3983 /* unmapped? It's a hole - nothing to do */
3984 if (!buffer_mapped(bh)) {
3985 BUFFER_TRACE(bh, "still unmapped");
3990 /* Ok, it's mapped. Make sure it's up-to-date */
3991 if (PageUptodate(page))
3992 set_buffer_uptodate(bh);
3994 if (!buffer_uptodate(bh)) {
3996 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
3998 /* Uhhuh. Read error. Complain and punt. */
3999 if (!buffer_uptodate(bh))
4001 if (S_ISREG(inode->i_mode) &&
4002 ext4_encrypted_inode(inode)) {
4003 /* We expect the key to be set. */
4004 BUG_ON(!fscrypt_has_encryption_key(inode));
4005 BUG_ON(blocksize != PAGE_SIZE);
4006 WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host,
4007 page, PAGE_SIZE, 0, page->index));
4010 if (ext4_should_journal_data(inode)) {
4011 BUFFER_TRACE(bh, "get write access");
4012 err = ext4_journal_get_write_access(handle, bh);
4016 zero_user(page, offset, length);
4017 BUFFER_TRACE(bh, "zeroed end of block");
4019 if (ext4_should_journal_data(inode)) {
4020 err = ext4_handle_dirty_metadata(handle, inode, bh);
4023 mark_buffer_dirty(bh);
4024 if (ext4_should_order_data(inode))
4025 err = ext4_jbd2_inode_add_write(handle, inode);
4035 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4036 * starting from file offset 'from'. The range to be zero'd must
4037 * be contained with in one block. If the specified range exceeds
4038 * the end of the block it will be shortened to end of the block
4039 * that cooresponds to 'from'
4041 static int ext4_block_zero_page_range(handle_t *handle,
4042 struct address_space *mapping, loff_t from, loff_t length)
4044 struct inode *inode = mapping->host;
4045 unsigned offset = from & (PAGE_SIZE-1);
4046 unsigned blocksize = inode->i_sb->s_blocksize;
4047 unsigned max = blocksize - (offset & (blocksize - 1));
4050 * correct length if it does not fall between
4051 * 'from' and the end of the block
4053 if (length > max || length < 0)
4056 if (IS_DAX(inode)) {
4057 return iomap_zero_range(inode, from, length, NULL,
4060 return __ext4_block_zero_page_range(handle, mapping, from, length);
4064 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4065 * up to the end of the block which corresponds to `from'.
4066 * This required during truncate. We need to physically zero the tail end
4067 * of that block so it doesn't yield old data if the file is later grown.
4069 static int ext4_block_truncate_page(handle_t *handle,
4070 struct address_space *mapping, loff_t from)
4072 unsigned offset = from & (PAGE_SIZE-1);
4075 struct inode *inode = mapping->host;
4077 /* If we are processing an encrypted inode during orphan list handling */
4078 if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode))
4081 blocksize = inode->i_sb->s_blocksize;
4082 length = blocksize - (offset & (blocksize - 1));
4084 return ext4_block_zero_page_range(handle, mapping, from, length);
4087 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4088 loff_t lstart, loff_t length)
4090 struct super_block *sb = inode->i_sb;
4091 struct address_space *mapping = inode->i_mapping;
4092 unsigned partial_start, partial_end;
4093 ext4_fsblk_t start, end;
4094 loff_t byte_end = (lstart + length - 1);
4097 partial_start = lstart & (sb->s_blocksize - 1);
4098 partial_end = byte_end & (sb->s_blocksize - 1);
4100 start = lstart >> sb->s_blocksize_bits;
4101 end = byte_end >> sb->s_blocksize_bits;
4103 /* Handle partial zero within the single block */
4105 (partial_start || (partial_end != sb->s_blocksize - 1))) {
4106 err = ext4_block_zero_page_range(handle, mapping,
4110 /* Handle partial zero out on the start of the range */
4111 if (partial_start) {
4112 err = ext4_block_zero_page_range(handle, mapping,
4113 lstart, sb->s_blocksize);
4117 /* Handle partial zero out on the end of the range */
4118 if (partial_end != sb->s_blocksize - 1)
4119 err = ext4_block_zero_page_range(handle, mapping,
4120 byte_end - partial_end,
4125 int ext4_can_truncate(struct inode *inode)
4127 if (S_ISREG(inode->i_mode))
4129 if (S_ISDIR(inode->i_mode))
4131 if (S_ISLNK(inode->i_mode))
4132 return !ext4_inode_is_fast_symlink(inode);
4137 * We have to make sure i_disksize gets properly updated before we truncate
4138 * page cache due to hole punching or zero range. Otherwise i_disksize update
4139 * can get lost as it may have been postponed to submission of writeback but
4140 * that will never happen after we truncate page cache.
4142 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4146 loff_t size = i_size_read(inode);
4148 WARN_ON(!inode_is_locked(inode));
4149 if (offset > size || offset + len < size)
4152 if (EXT4_I(inode)->i_disksize >= size)
4155 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4157 return PTR_ERR(handle);
4158 ext4_update_i_disksize(inode, size);
4159 ext4_mark_inode_dirty(handle, inode);
4160 ext4_journal_stop(handle);
4166 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4167 * associated with the given offset and length
4169 * @inode: File inode
4170 * @offset: The offset where the hole will begin
4171 * @len: The length of the hole
4173 * Returns: 0 on success or negative on failure
4176 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4178 struct super_block *sb = inode->i_sb;
4179 ext4_lblk_t first_block, stop_block;
4180 struct address_space *mapping = inode->i_mapping;
4181 loff_t first_block_offset, last_block_offset;
4183 unsigned int credits;
4186 if (!S_ISREG(inode->i_mode))
4189 trace_ext4_punch_hole(inode, offset, length, 0);
4192 * Write out all dirty pages to avoid race conditions
4193 * Then release them.
4195 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4196 ret = filemap_write_and_wait_range(mapping, offset,
4197 offset + length - 1);
4204 /* No need to punch hole beyond i_size */
4205 if (offset >= inode->i_size)
4209 * If the hole extends beyond i_size, set the hole
4210 * to end after the page that contains i_size
4212 if (offset + length > inode->i_size) {
4213 length = inode->i_size +
4214 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4218 if (offset & (sb->s_blocksize - 1) ||
4219 (offset + length) & (sb->s_blocksize - 1)) {
4221 * Attach jinode to inode for jbd2 if we do any zeroing of
4224 ret = ext4_inode_attach_jinode(inode);
4230 /* Wait all existing dio workers, newcomers will block on i_mutex */
4231 ext4_inode_block_unlocked_dio(inode);
4232 inode_dio_wait(inode);
4235 * Prevent page faults from reinstantiating pages we have released from
4238 down_write(&EXT4_I(inode)->i_mmap_sem);
4239 first_block_offset = round_up(offset, sb->s_blocksize);
4240 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4242 /* Now release the pages and zero block aligned part of pages*/
4243 if (last_block_offset > first_block_offset) {
4244 ret = ext4_update_disksize_before_punch(inode, offset, length);
4247 truncate_pagecache_range(inode, first_block_offset,
4251 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4252 credits = ext4_writepage_trans_blocks(inode);
4254 credits = ext4_blocks_for_truncate(inode);
4255 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4256 if (IS_ERR(handle)) {
4257 ret = PTR_ERR(handle);
4258 ext4_std_error(sb, ret);
4262 ret = ext4_zero_partial_blocks(handle, inode, offset,
4267 first_block = (offset + sb->s_blocksize - 1) >>
4268 EXT4_BLOCK_SIZE_BITS(sb);
4269 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4271 /* If there are no blocks to remove, return now */
4272 if (first_block >= stop_block)
4275 down_write(&EXT4_I(inode)->i_data_sem);
4276 ext4_discard_preallocations(inode);
4278 ret = ext4_es_remove_extent(inode, first_block,
4279 stop_block - first_block);
4281 up_write(&EXT4_I(inode)->i_data_sem);
4285 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4286 ret = ext4_ext_remove_space(inode, first_block,
4289 ret = ext4_ind_remove_space(handle, inode, first_block,
4292 up_write(&EXT4_I(inode)->i_data_sem);
4294 ext4_handle_sync(handle);
4296 inode->i_mtime = inode->i_ctime = current_time(inode);
4297 ext4_mark_inode_dirty(handle, inode);
4299 ext4_update_inode_fsync_trans(handle, inode, 1);
4301 ext4_journal_stop(handle);
4303 up_write(&EXT4_I(inode)->i_mmap_sem);
4304 ext4_inode_resume_unlocked_dio(inode);
4306 inode_unlock(inode);
4310 int ext4_inode_attach_jinode(struct inode *inode)
4312 struct ext4_inode_info *ei = EXT4_I(inode);
4313 struct jbd2_inode *jinode;
4315 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4318 jinode = jbd2_alloc_inode(GFP_KERNEL);
4319 spin_lock(&inode->i_lock);
4322 spin_unlock(&inode->i_lock);
4325 ei->jinode = jinode;
4326 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4329 spin_unlock(&inode->i_lock);
4330 if (unlikely(jinode != NULL))
4331 jbd2_free_inode(jinode);
4338 * We block out ext4_get_block() block instantiations across the entire
4339 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4340 * simultaneously on behalf of the same inode.
4342 * As we work through the truncate and commit bits of it to the journal there
4343 * is one core, guiding principle: the file's tree must always be consistent on
4344 * disk. We must be able to restart the truncate after a crash.
4346 * The file's tree may be transiently inconsistent in memory (although it
4347 * probably isn't), but whenever we close off and commit a journal transaction,
4348 * the contents of (the filesystem + the journal) must be consistent and
4349 * restartable. It's pretty simple, really: bottom up, right to left (although
4350 * left-to-right works OK too).
4352 * Note that at recovery time, journal replay occurs *before* the restart of
4353 * truncate against the orphan inode list.
4355 * The committed inode has the new, desired i_size (which is the same as
4356 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4357 * that this inode's truncate did not complete and it will again call
4358 * ext4_truncate() to have another go. So there will be instantiated blocks
4359 * to the right of the truncation point in a crashed ext4 filesystem. But
4360 * that's fine - as long as they are linked from the inode, the post-crash
4361 * ext4_truncate() run will find them and release them.
4363 int ext4_truncate(struct inode *inode)
4365 struct ext4_inode_info *ei = EXT4_I(inode);
4366 unsigned int credits;
4369 struct address_space *mapping = inode->i_mapping;
4372 * There is a possibility that we're either freeing the inode
4373 * or it's a completely new inode. In those cases we might not
4374 * have i_mutex locked because it's not necessary.
4376 if (!(inode->i_state & (I_NEW|I_FREEING)))
4377 WARN_ON(!inode_is_locked(inode));
4378 trace_ext4_truncate_enter(inode);
4380 if (!ext4_can_truncate(inode))
4383 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4385 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4386 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4388 if (ext4_has_inline_data(inode)) {
4391 err = ext4_inline_data_truncate(inode, &has_inline);
4398 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4399 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4400 if (ext4_inode_attach_jinode(inode) < 0)
4404 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4405 credits = ext4_writepage_trans_blocks(inode);
4407 credits = ext4_blocks_for_truncate(inode);
4409 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4411 return PTR_ERR(handle);
4413 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4414 ext4_block_truncate_page(handle, mapping, inode->i_size);
4417 * We add the inode to the orphan list, so that if this
4418 * truncate spans multiple transactions, and we crash, we will
4419 * resume the truncate when the filesystem recovers. It also
4420 * marks the inode dirty, to catch the new size.
4422 * Implication: the file must always be in a sane, consistent
4423 * truncatable state while each transaction commits.
4425 err = ext4_orphan_add(handle, inode);
4429 down_write(&EXT4_I(inode)->i_data_sem);
4431 ext4_discard_preallocations(inode);
4433 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4434 err = ext4_ext_truncate(handle, inode);
4436 ext4_ind_truncate(handle, inode);
4438 up_write(&ei->i_data_sem);
4443 ext4_handle_sync(handle);
4447 * If this was a simple ftruncate() and the file will remain alive,
4448 * then we need to clear up the orphan record which we created above.
4449 * However, if this was a real unlink then we were called by
4450 * ext4_evict_inode(), and we allow that function to clean up the
4451 * orphan info for us.
4454 ext4_orphan_del(handle, inode);
4456 inode->i_mtime = inode->i_ctime = current_time(inode);
4457 ext4_mark_inode_dirty(handle, inode);
4458 ext4_journal_stop(handle);
4460 trace_ext4_truncate_exit(inode);
4465 * ext4_get_inode_loc returns with an extra refcount against the inode's
4466 * underlying buffer_head on success. If 'in_mem' is true, we have all
4467 * data in memory that is needed to recreate the on-disk version of this
4470 static int __ext4_get_inode_loc(struct inode *inode,
4471 struct ext4_iloc *iloc, int in_mem)
4473 struct ext4_group_desc *gdp;
4474 struct buffer_head *bh;
4475 struct super_block *sb = inode->i_sb;
4477 int inodes_per_block, inode_offset;
4480 if (!ext4_valid_inum(sb, inode->i_ino))
4481 return -EFSCORRUPTED;
4483 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4484 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4489 * Figure out the offset within the block group inode table
4491 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4492 inode_offset = ((inode->i_ino - 1) %
4493 EXT4_INODES_PER_GROUP(sb));
4494 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4495 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4497 bh = sb_getblk(sb, block);
4500 if (!buffer_uptodate(bh)) {
4504 * If the buffer has the write error flag, we have failed
4505 * to write out another inode in the same block. In this
4506 * case, we don't have to read the block because we may
4507 * read the old inode data successfully.
4509 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4510 set_buffer_uptodate(bh);
4512 if (buffer_uptodate(bh)) {
4513 /* someone brought it uptodate while we waited */
4519 * If we have all information of the inode in memory and this
4520 * is the only valid inode in the block, we need not read the
4524 struct buffer_head *bitmap_bh;
4527 start = inode_offset & ~(inodes_per_block - 1);
4529 /* Is the inode bitmap in cache? */
4530 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4531 if (unlikely(!bitmap_bh))
4535 * If the inode bitmap isn't in cache then the
4536 * optimisation may end up performing two reads instead
4537 * of one, so skip it.
4539 if (!buffer_uptodate(bitmap_bh)) {
4543 for (i = start; i < start + inodes_per_block; i++) {
4544 if (i == inode_offset)
4546 if (ext4_test_bit(i, bitmap_bh->b_data))
4550 if (i == start + inodes_per_block) {
4551 /* all other inodes are free, so skip I/O */
4552 memset(bh->b_data, 0, bh->b_size);
4553 set_buffer_uptodate(bh);
4561 * If we need to do any I/O, try to pre-readahead extra
4562 * blocks from the inode table.
4564 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4565 ext4_fsblk_t b, end, table;
4567 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4569 table = ext4_inode_table(sb, gdp);
4570 /* s_inode_readahead_blks is always a power of 2 */
4571 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4575 num = EXT4_INODES_PER_GROUP(sb);
4576 if (ext4_has_group_desc_csum(sb))
4577 num -= ext4_itable_unused_count(sb, gdp);
4578 table += num / inodes_per_block;
4582 sb_breadahead(sb, b++);
4586 * There are other valid inodes in the buffer, this inode
4587 * has in-inode xattrs, or we don't have this inode in memory.
4588 * Read the block from disk.
4590 trace_ext4_load_inode(inode);
4592 bh->b_end_io = end_buffer_read_sync;
4593 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4595 if (!buffer_uptodate(bh)) {
4596 EXT4_ERROR_INODE_BLOCK(inode, block,
4597 "unable to read itable block");
4607 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4609 /* We have all inode data except xattrs in memory here. */
4610 return __ext4_get_inode_loc(inode, iloc,
4611 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4614 static bool ext4_should_use_dax(struct inode *inode)
4616 if (!test_opt(inode->i_sb, DAX))
4618 if (!S_ISREG(inode->i_mode))
4620 if (ext4_should_journal_data(inode))
4622 if (ext4_has_inline_data(inode))
4624 if (ext4_encrypted_inode(inode))
4629 void ext4_set_inode_flags(struct inode *inode)
4631 unsigned int flags = EXT4_I(inode)->i_flags;
4632 unsigned int new_fl = 0;
4634 if (flags & EXT4_SYNC_FL)
4636 if (flags & EXT4_APPEND_FL)
4638 if (flags & EXT4_IMMUTABLE_FL)
4639 new_fl |= S_IMMUTABLE;
4640 if (flags & EXT4_NOATIME_FL)
4641 new_fl |= S_NOATIME;
4642 if (flags & EXT4_DIRSYNC_FL)
4643 new_fl |= S_DIRSYNC;
4644 if (ext4_should_use_dax(inode))
4646 if (flags & EXT4_ENCRYPT_FL)
4647 new_fl |= S_ENCRYPTED;
4648 inode_set_flags(inode, new_fl,
4649 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4653 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4654 struct ext4_inode_info *ei)
4657 struct inode *inode = &(ei->vfs_inode);
4658 struct super_block *sb = inode->i_sb;
4660 if (ext4_has_feature_huge_file(sb)) {
4661 /* we are using combined 48 bit field */
4662 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4663 le32_to_cpu(raw_inode->i_blocks_lo);
4664 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4665 /* i_blocks represent file system block size */
4666 return i_blocks << (inode->i_blkbits - 9);
4671 return le32_to_cpu(raw_inode->i_blocks_lo);
4675 static inline void ext4_iget_extra_inode(struct inode *inode,
4676 struct ext4_inode *raw_inode,
4677 struct ext4_inode_info *ei)
4679 __le32 *magic = (void *)raw_inode +
4680 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4681 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4682 EXT4_INODE_SIZE(inode->i_sb) &&
4683 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4684 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4685 ext4_find_inline_data_nolock(inode);
4687 EXT4_I(inode)->i_inline_off = 0;
4690 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4692 if (!ext4_has_feature_project(inode->i_sb))
4694 *projid = EXT4_I(inode)->i_projid;
4698 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4700 struct ext4_iloc iloc;
4701 struct ext4_inode *raw_inode;
4702 struct ext4_inode_info *ei;
4703 struct inode *inode;
4704 journal_t *journal = EXT4_SB(sb)->s_journal;
4712 inode = iget_locked(sb, ino);
4714 return ERR_PTR(-ENOMEM);
4715 if (!(inode->i_state & I_NEW))
4721 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4724 raw_inode = ext4_raw_inode(&iloc);
4726 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4727 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4728 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4729 EXT4_INODE_SIZE(inode->i_sb) ||
4730 (ei->i_extra_isize & 3)) {
4731 EXT4_ERROR_INODE(inode,
4732 "bad extra_isize %u (inode size %u)",
4734 EXT4_INODE_SIZE(inode->i_sb));
4735 ret = -EFSCORRUPTED;
4739 ei->i_extra_isize = 0;
4741 /* Precompute checksum seed for inode metadata */
4742 if (ext4_has_metadata_csum(sb)) {
4743 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4745 __le32 inum = cpu_to_le32(inode->i_ino);
4746 __le32 gen = raw_inode->i_generation;
4747 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4749 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4753 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4754 EXT4_ERROR_INODE(inode, "checksum invalid");
4759 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4760 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4761 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4762 if (ext4_has_feature_project(sb) &&
4763 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4764 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4765 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4767 i_projid = EXT4_DEF_PROJID;
4769 if (!(test_opt(inode->i_sb, NO_UID32))) {
4770 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4771 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4773 i_uid_write(inode, i_uid);
4774 i_gid_write(inode, i_gid);
4775 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4776 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4778 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4779 ei->i_inline_off = 0;
4780 ei->i_dir_start_lookup = 0;
4781 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4782 /* We now have enough fields to check if the inode was active or not.
4783 * This is needed because nfsd might try to access dead inodes
4784 * the test is that same one that e2fsck uses
4785 * NeilBrown 1999oct15
4787 if (inode->i_nlink == 0) {
4788 if ((inode->i_mode == 0 ||
4789 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4790 ino != EXT4_BOOT_LOADER_INO) {
4791 /* this inode is deleted */
4795 /* The only unlinked inodes we let through here have
4796 * valid i_mode and are being read by the orphan
4797 * recovery code: that's fine, we're about to complete
4798 * the process of deleting those.
4799 * OR it is the EXT4_BOOT_LOADER_INO which is
4800 * not initialized on a new filesystem. */
4802 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4803 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4804 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4805 if (ext4_has_feature_64bit(sb))
4807 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4808 inode->i_size = ext4_isize(sb, raw_inode);
4809 if ((size = i_size_read(inode)) < 0) {
4810 EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size);
4811 ret = -EFSCORRUPTED;
4814 ei->i_disksize = inode->i_size;
4816 ei->i_reserved_quota = 0;
4818 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4819 ei->i_block_group = iloc.block_group;
4820 ei->i_last_alloc_group = ~0;
4822 * NOTE! The in-memory inode i_data array is in little-endian order
4823 * even on big-endian machines: we do NOT byteswap the block numbers!
4825 for (block = 0; block < EXT4_N_BLOCKS; block++)
4826 ei->i_data[block] = raw_inode->i_block[block];
4827 INIT_LIST_HEAD(&ei->i_orphan);
4830 * Set transaction id's of transactions that have to be committed
4831 * to finish f[data]sync. We set them to currently running transaction
4832 * as we cannot be sure that the inode or some of its metadata isn't
4833 * part of the transaction - the inode could have been reclaimed and
4834 * now it is reread from disk.
4837 transaction_t *transaction;
4840 read_lock(&journal->j_state_lock);
4841 if (journal->j_running_transaction)
4842 transaction = journal->j_running_transaction;
4844 transaction = journal->j_committing_transaction;
4846 tid = transaction->t_tid;
4848 tid = journal->j_commit_sequence;
4849 read_unlock(&journal->j_state_lock);
4850 ei->i_sync_tid = tid;
4851 ei->i_datasync_tid = tid;
4854 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4855 if (ei->i_extra_isize == 0) {
4856 /* The extra space is currently unused. Use it. */
4857 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4858 ei->i_extra_isize = sizeof(struct ext4_inode) -
4859 EXT4_GOOD_OLD_INODE_SIZE;
4861 ext4_iget_extra_inode(inode, raw_inode, ei);
4865 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4866 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4867 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4868 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4870 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4871 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4872 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4873 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4875 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4880 if (ei->i_file_acl &&
4881 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4882 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4884 ret = -EFSCORRUPTED;
4886 } else if (!ext4_has_inline_data(inode)) {
4887 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4888 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4889 (S_ISLNK(inode->i_mode) &&
4890 !ext4_inode_is_fast_symlink(inode))))
4891 /* Validate extent which is part of inode */
4892 ret = ext4_ext_check_inode(inode);
4893 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4894 (S_ISLNK(inode->i_mode) &&
4895 !ext4_inode_is_fast_symlink(inode))) {
4896 /* Validate block references which are part of inode */
4897 ret = ext4_ind_check_inode(inode);
4903 if (S_ISREG(inode->i_mode)) {
4904 inode->i_op = &ext4_file_inode_operations;
4905 inode->i_fop = &ext4_file_operations;
4906 ext4_set_aops(inode);
4907 } else if (S_ISDIR(inode->i_mode)) {
4908 inode->i_op = &ext4_dir_inode_operations;
4909 inode->i_fop = &ext4_dir_operations;
4910 } else if (S_ISLNK(inode->i_mode)) {
4911 if (ext4_encrypted_inode(inode)) {
4912 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4913 ext4_set_aops(inode);
4914 } else if (ext4_inode_is_fast_symlink(inode)) {
4915 inode->i_link = (char *)ei->i_data;
4916 inode->i_op = &ext4_fast_symlink_inode_operations;
4917 nd_terminate_link(ei->i_data, inode->i_size,
4918 sizeof(ei->i_data) - 1);
4920 inode->i_op = &ext4_symlink_inode_operations;
4921 ext4_set_aops(inode);
4923 inode_nohighmem(inode);
4924 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4925 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4926 inode->i_op = &ext4_special_inode_operations;
4927 if (raw_inode->i_block[0])
4928 init_special_inode(inode, inode->i_mode,
4929 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4931 init_special_inode(inode, inode->i_mode,
4932 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4933 } else if (ino == EXT4_BOOT_LOADER_INO) {
4934 make_bad_inode(inode);
4936 ret = -EFSCORRUPTED;
4937 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4941 ext4_set_inode_flags(inode);
4943 unlock_new_inode(inode);
4949 return ERR_PTR(ret);
4952 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4954 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4955 return ERR_PTR(-EFSCORRUPTED);
4956 return ext4_iget(sb, ino);
4959 static int ext4_inode_blocks_set(handle_t *handle,
4960 struct ext4_inode *raw_inode,
4961 struct ext4_inode_info *ei)
4963 struct inode *inode = &(ei->vfs_inode);
4964 u64 i_blocks = inode->i_blocks;
4965 struct super_block *sb = inode->i_sb;
4967 if (i_blocks <= ~0U) {
4969 * i_blocks can be represented in a 32 bit variable
4970 * as multiple of 512 bytes
4972 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4973 raw_inode->i_blocks_high = 0;
4974 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4977 if (!ext4_has_feature_huge_file(sb))
4980 if (i_blocks <= 0xffffffffffffULL) {
4982 * i_blocks can be represented in a 48 bit variable
4983 * as multiple of 512 bytes
4985 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4986 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4987 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4989 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4990 /* i_block is stored in file system block size */
4991 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4992 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4993 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4998 struct other_inode {
4999 unsigned long orig_ino;
5000 struct ext4_inode *raw_inode;
5003 static int other_inode_match(struct inode * inode, unsigned long ino,
5006 struct other_inode *oi = (struct other_inode *) data;
5008 if ((inode->i_ino != ino) ||
5009 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5010 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
5011 ((inode->i_state & I_DIRTY_TIME) == 0))
5013 spin_lock(&inode->i_lock);
5014 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5015 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
5016 (inode->i_state & I_DIRTY_TIME)) {
5017 struct ext4_inode_info *ei = EXT4_I(inode);
5019 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
5020 spin_unlock(&inode->i_lock);
5022 spin_lock(&ei->i_raw_lock);
5023 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
5024 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
5025 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
5026 ext4_inode_csum_set(inode, oi->raw_inode, ei);
5027 spin_unlock(&ei->i_raw_lock);
5028 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
5031 spin_unlock(&inode->i_lock);
5036 * Opportunistically update the other time fields for other inodes in
5037 * the same inode table block.
5039 static void ext4_update_other_inodes_time(struct super_block *sb,
5040 unsigned long orig_ino, char *buf)
5042 struct other_inode oi;
5044 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5045 int inode_size = EXT4_INODE_SIZE(sb);
5047 oi.orig_ino = orig_ino;
5049 * Calculate the first inode in the inode table block. Inode
5050 * numbers are one-based. That is, the first inode in a block
5051 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5053 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5054 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5055 if (ino == orig_ino)
5057 oi.raw_inode = (struct ext4_inode *) buf;
5058 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
5063 * Post the struct inode info into an on-disk inode location in the
5064 * buffer-cache. This gobbles the caller's reference to the
5065 * buffer_head in the inode location struct.
5067 * The caller must have write access to iloc->bh.
5069 static int ext4_do_update_inode(handle_t *handle,
5070 struct inode *inode,
5071 struct ext4_iloc *iloc)
5073 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5074 struct ext4_inode_info *ei = EXT4_I(inode);
5075 struct buffer_head *bh = iloc->bh;
5076 struct super_block *sb = inode->i_sb;
5077 int err = 0, rc, block;
5078 int need_datasync = 0, set_large_file = 0;
5083 spin_lock(&ei->i_raw_lock);
5085 /* For fields not tracked in the in-memory inode,
5086 * initialise them to zero for new inodes. */
5087 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5088 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5090 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5091 i_uid = i_uid_read(inode);
5092 i_gid = i_gid_read(inode);
5093 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5094 if (!(test_opt(inode->i_sb, NO_UID32))) {
5095 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5096 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5098 * Fix up interoperability with old kernels. Otherwise, old inodes get
5099 * re-used with the upper 16 bits of the uid/gid intact
5101 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5102 raw_inode->i_uid_high = 0;
5103 raw_inode->i_gid_high = 0;
5105 raw_inode->i_uid_high =
5106 cpu_to_le16(high_16_bits(i_uid));
5107 raw_inode->i_gid_high =
5108 cpu_to_le16(high_16_bits(i_gid));
5111 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5112 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5113 raw_inode->i_uid_high = 0;
5114 raw_inode->i_gid_high = 0;
5116 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5118 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5119 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5120 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5121 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5123 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5125 spin_unlock(&ei->i_raw_lock);
5128 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5129 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5130 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5131 raw_inode->i_file_acl_high =
5132 cpu_to_le16(ei->i_file_acl >> 32);
5133 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5134 if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
5135 ext4_isize_set(raw_inode, ei->i_disksize);
5138 if (ei->i_disksize > 0x7fffffffULL) {
5139 if (!ext4_has_feature_large_file(sb) ||
5140 EXT4_SB(sb)->s_es->s_rev_level ==
5141 cpu_to_le32(EXT4_GOOD_OLD_REV))
5144 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5145 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5146 if (old_valid_dev(inode->i_rdev)) {
5147 raw_inode->i_block[0] =
5148 cpu_to_le32(old_encode_dev(inode->i_rdev));
5149 raw_inode->i_block[1] = 0;
5151 raw_inode->i_block[0] = 0;
5152 raw_inode->i_block[1] =
5153 cpu_to_le32(new_encode_dev(inode->i_rdev));
5154 raw_inode->i_block[2] = 0;
5156 } else if (!ext4_has_inline_data(inode)) {
5157 for (block = 0; block < EXT4_N_BLOCKS; block++)
5158 raw_inode->i_block[block] = ei->i_data[block];
5161 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5162 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5163 if (ei->i_extra_isize) {
5164 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5165 raw_inode->i_version_hi =
5166 cpu_to_le32(inode->i_version >> 32);
5167 raw_inode->i_extra_isize =
5168 cpu_to_le16(ei->i_extra_isize);
5172 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5173 i_projid != EXT4_DEF_PROJID);
5175 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5176 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5177 raw_inode->i_projid = cpu_to_le32(i_projid);
5179 ext4_inode_csum_set(inode, raw_inode, ei);
5180 spin_unlock(&ei->i_raw_lock);
5181 if (inode->i_sb->s_flags & MS_LAZYTIME)
5182 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5185 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5186 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5189 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5190 if (set_large_file) {
5191 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5192 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5195 ext4_update_dynamic_rev(sb);
5196 ext4_set_feature_large_file(sb);
5197 ext4_handle_sync(handle);
5198 err = ext4_handle_dirty_super(handle, sb);
5200 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5203 ext4_std_error(inode->i_sb, err);
5208 * ext4_write_inode()
5210 * We are called from a few places:
5212 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5213 * Here, there will be no transaction running. We wait for any running
5214 * transaction to commit.
5216 * - Within flush work (sys_sync(), kupdate and such).
5217 * We wait on commit, if told to.
5219 * - Within iput_final() -> write_inode_now()
5220 * We wait on commit, if told to.
5222 * In all cases it is actually safe for us to return without doing anything,
5223 * because the inode has been copied into a raw inode buffer in
5224 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5227 * Note that we are absolutely dependent upon all inode dirtiers doing the
5228 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5229 * which we are interested.
5231 * It would be a bug for them to not do this. The code:
5233 * mark_inode_dirty(inode)
5235 * inode->i_size = expr;
5237 * is in error because write_inode() could occur while `stuff()' is running,
5238 * and the new i_size will be lost. Plus the inode will no longer be on the
5239 * superblock's dirty inode list.
5241 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5245 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5248 if (EXT4_SB(inode->i_sb)->s_journal) {
5249 if (ext4_journal_current_handle()) {
5250 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5256 * No need to force transaction in WB_SYNC_NONE mode. Also
5257 * ext4_sync_fs() will force the commit after everything is
5260 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5263 err = ext4_force_commit(inode->i_sb);
5265 struct ext4_iloc iloc;
5267 err = __ext4_get_inode_loc(inode, &iloc, 0);
5271 * sync(2) will flush the whole buffer cache. No need to do
5272 * it here separately for each inode.
5274 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5275 sync_dirty_buffer(iloc.bh);
5276 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5277 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5278 "IO error syncing inode");
5287 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5288 * buffers that are attached to a page stradding i_size and are undergoing
5289 * commit. In that case we have to wait for commit to finish and try again.
5291 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5295 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5296 tid_t commit_tid = 0;
5299 offset = inode->i_size & (PAGE_SIZE - 1);
5301 * All buffers in the last page remain valid? Then there's nothing to
5302 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5305 if (offset > PAGE_SIZE - i_blocksize(inode))
5308 page = find_lock_page(inode->i_mapping,
5309 inode->i_size >> PAGE_SHIFT);
5312 ret = __ext4_journalled_invalidatepage(page, offset,
5313 PAGE_SIZE - offset);
5319 read_lock(&journal->j_state_lock);
5320 if (journal->j_committing_transaction)
5321 commit_tid = journal->j_committing_transaction->t_tid;
5322 read_unlock(&journal->j_state_lock);
5324 jbd2_log_wait_commit(journal, commit_tid);
5331 * Called from notify_change.
5333 * We want to trap VFS attempts to truncate the file as soon as
5334 * possible. In particular, we want to make sure that when the VFS
5335 * shrinks i_size, we put the inode on the orphan list and modify
5336 * i_disksize immediately, so that during the subsequent flushing of
5337 * dirty pages and freeing of disk blocks, we can guarantee that any
5338 * commit will leave the blocks being flushed in an unused state on
5339 * disk. (On recovery, the inode will get truncated and the blocks will
5340 * be freed, so we have a strong guarantee that no future commit will
5341 * leave these blocks visible to the user.)
5343 * Another thing we have to assure is that if we are in ordered mode
5344 * and inode is still attached to the committing transaction, we must
5345 * we start writeout of all the dirty pages which are being truncated.
5346 * This way we are sure that all the data written in the previous
5347 * transaction are already on disk (truncate waits for pages under
5350 * Called with inode->i_mutex down.
5352 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5354 struct inode *inode = d_inode(dentry);
5357 const unsigned int ia_valid = attr->ia_valid;
5359 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5362 error = setattr_prepare(dentry, attr);
5366 error = fscrypt_prepare_setattr(dentry, attr);
5370 if (is_quota_modification(inode, attr)) {
5371 error = dquot_initialize(inode);
5375 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5376 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5379 /* (user+group)*(old+new) structure, inode write (sb,
5380 * inode block, ? - but truncate inode update has it) */
5381 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5382 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5383 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5384 if (IS_ERR(handle)) {
5385 error = PTR_ERR(handle);
5389 /* dquot_transfer() calls back ext4_get_inode_usage() which
5390 * counts xattr inode references.
5392 down_read(&EXT4_I(inode)->xattr_sem);
5393 error = dquot_transfer(inode, attr);
5394 up_read(&EXT4_I(inode)->xattr_sem);
5397 ext4_journal_stop(handle);
5400 /* Update corresponding info in inode so that everything is in
5401 * one transaction */
5402 if (attr->ia_valid & ATTR_UID)
5403 inode->i_uid = attr->ia_uid;
5404 if (attr->ia_valid & ATTR_GID)
5405 inode->i_gid = attr->ia_gid;
5406 error = ext4_mark_inode_dirty(handle, inode);
5407 ext4_journal_stop(handle);
5410 if (attr->ia_valid & ATTR_SIZE) {
5412 loff_t oldsize = inode->i_size;
5413 int shrink = (attr->ia_size <= inode->i_size);
5415 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5416 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5418 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5421 if (!S_ISREG(inode->i_mode))
5424 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5425 inode_inc_iversion(inode);
5427 if (ext4_should_order_data(inode) &&
5428 (attr->ia_size < inode->i_size)) {
5429 error = ext4_begin_ordered_truncate(inode,
5434 if (attr->ia_size != inode->i_size) {
5435 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5436 if (IS_ERR(handle)) {
5437 error = PTR_ERR(handle);
5440 if (ext4_handle_valid(handle) && shrink) {
5441 error = ext4_orphan_add(handle, inode);
5445 * Update c/mtime on truncate up, ext4_truncate() will
5446 * update c/mtime in shrink case below
5449 inode->i_mtime = current_time(inode);
5450 inode->i_ctime = inode->i_mtime;
5452 down_write(&EXT4_I(inode)->i_data_sem);
5453 EXT4_I(inode)->i_disksize = attr->ia_size;
5454 rc = ext4_mark_inode_dirty(handle, inode);
5458 * We have to update i_size under i_data_sem together
5459 * with i_disksize to avoid races with writeback code
5460 * running ext4_wb_update_i_disksize().
5463 i_size_write(inode, attr->ia_size);
5464 up_write(&EXT4_I(inode)->i_data_sem);
5465 ext4_journal_stop(handle);
5468 ext4_orphan_del(NULL, inode);
5473 pagecache_isize_extended(inode, oldsize, inode->i_size);
5476 * Blocks are going to be removed from the inode. Wait
5477 * for dio in flight. Temporarily disable
5478 * dioread_nolock to prevent livelock.
5481 if (!ext4_should_journal_data(inode)) {
5482 ext4_inode_block_unlocked_dio(inode);
5483 inode_dio_wait(inode);
5484 ext4_inode_resume_unlocked_dio(inode);
5486 ext4_wait_for_tail_page_commit(inode);
5488 down_write(&EXT4_I(inode)->i_mmap_sem);
5490 * Truncate pagecache after we've waited for commit
5491 * in data=journal mode to make pages freeable.
5493 truncate_pagecache(inode, inode->i_size);
5495 rc = ext4_truncate(inode);
5499 up_write(&EXT4_I(inode)->i_mmap_sem);
5503 setattr_copy(inode, attr);
5504 mark_inode_dirty(inode);
5508 * If the call to ext4_truncate failed to get a transaction handle at
5509 * all, we need to clean up the in-core orphan list manually.
5511 if (orphan && inode->i_nlink)
5512 ext4_orphan_del(NULL, inode);
5514 if (!error && (ia_valid & ATTR_MODE))
5515 rc = posix_acl_chmod(inode, inode->i_mode);
5518 ext4_std_error(inode->i_sb, error);
5524 int ext4_getattr(const struct path *path, struct kstat *stat,
5525 u32 request_mask, unsigned int query_flags)
5527 struct inode *inode = d_inode(path->dentry);
5528 struct ext4_inode *raw_inode;
5529 struct ext4_inode_info *ei = EXT4_I(inode);
5532 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5533 stat->result_mask |= STATX_BTIME;
5534 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5535 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5538 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5539 if (flags & EXT4_APPEND_FL)
5540 stat->attributes |= STATX_ATTR_APPEND;
5541 if (flags & EXT4_COMPR_FL)
5542 stat->attributes |= STATX_ATTR_COMPRESSED;
5543 if (flags & EXT4_ENCRYPT_FL)
5544 stat->attributes |= STATX_ATTR_ENCRYPTED;
5545 if (flags & EXT4_IMMUTABLE_FL)
5546 stat->attributes |= STATX_ATTR_IMMUTABLE;
5547 if (flags & EXT4_NODUMP_FL)
5548 stat->attributes |= STATX_ATTR_NODUMP;
5550 stat->attributes_mask |= (STATX_ATTR_APPEND |
5551 STATX_ATTR_COMPRESSED |
5552 STATX_ATTR_ENCRYPTED |
5553 STATX_ATTR_IMMUTABLE |
5556 generic_fillattr(inode, stat);
5560 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5561 u32 request_mask, unsigned int query_flags)
5563 struct inode *inode = d_inode(path->dentry);
5564 u64 delalloc_blocks;
5566 ext4_getattr(path, stat, request_mask, query_flags);
5569 * If there is inline data in the inode, the inode will normally not
5570 * have data blocks allocated (it may have an external xattr block).
5571 * Report at least one sector for such files, so tools like tar, rsync,
5572 * others don't incorrectly think the file is completely sparse.
5574 if (unlikely(ext4_has_inline_data(inode)))
5575 stat->blocks += (stat->size + 511) >> 9;
5578 * We can't update i_blocks if the block allocation is delayed
5579 * otherwise in the case of system crash before the real block
5580 * allocation is done, we will have i_blocks inconsistent with
5581 * on-disk file blocks.
5582 * We always keep i_blocks updated together with real
5583 * allocation. But to not confuse with user, stat
5584 * will return the blocks that include the delayed allocation
5585 * blocks for this file.
5587 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5588 EXT4_I(inode)->i_reserved_data_blocks);
5589 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5593 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5596 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5597 return ext4_ind_trans_blocks(inode, lblocks);
5598 return ext4_ext_index_trans_blocks(inode, pextents);
5602 * Account for index blocks, block groups bitmaps and block group
5603 * descriptor blocks if modify datablocks and index blocks
5604 * worse case, the indexs blocks spread over different block groups
5606 * If datablocks are discontiguous, they are possible to spread over
5607 * different block groups too. If they are contiguous, with flexbg,
5608 * they could still across block group boundary.
5610 * Also account for superblock, inode, quota and xattr blocks
5612 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5615 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5621 * How many index blocks need to touch to map @lblocks logical blocks
5622 * to @pextents physical extents?
5624 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5629 * Now let's see how many group bitmaps and group descriptors need
5632 groups = idxblocks + pextents;
5634 if (groups > ngroups)
5636 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5637 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5639 /* bitmaps and block group descriptor blocks */
5640 ret += groups + gdpblocks;
5642 /* Blocks for super block, inode, quota and xattr blocks */
5643 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5649 * Calculate the total number of credits to reserve to fit
5650 * the modification of a single pages into a single transaction,
5651 * which may include multiple chunks of block allocations.
5653 * This could be called via ext4_write_begin()
5655 * We need to consider the worse case, when
5656 * one new block per extent.
5658 int ext4_writepage_trans_blocks(struct inode *inode)
5660 int bpp = ext4_journal_blocks_per_page(inode);
5663 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5665 /* Account for data blocks for journalled mode */
5666 if (ext4_should_journal_data(inode))
5672 * Calculate the journal credits for a chunk of data modification.
5674 * This is called from DIO, fallocate or whoever calling
5675 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5677 * journal buffers for data blocks are not included here, as DIO
5678 * and fallocate do no need to journal data buffers.
5680 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5682 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5686 * The caller must have previously called ext4_reserve_inode_write().
5687 * Give this, we know that the caller already has write access to iloc->bh.
5689 int ext4_mark_iloc_dirty(handle_t *handle,
5690 struct inode *inode, struct ext4_iloc *iloc)
5694 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5697 if (IS_I_VERSION(inode))
5698 inode_inc_iversion(inode);
5700 /* the do_update_inode consumes one bh->b_count */
5703 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5704 err = ext4_do_update_inode(handle, inode, iloc);
5710 * On success, We end up with an outstanding reference count against
5711 * iloc->bh. This _must_ be cleaned up later.
5715 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5716 struct ext4_iloc *iloc)
5720 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5723 err = ext4_get_inode_loc(inode, iloc);
5725 BUFFER_TRACE(iloc->bh, "get_write_access");
5726 err = ext4_journal_get_write_access(handle, iloc->bh);
5732 ext4_std_error(inode->i_sb, err);
5736 static int __ext4_expand_extra_isize(struct inode *inode,
5737 unsigned int new_extra_isize,
5738 struct ext4_iloc *iloc,
5739 handle_t *handle, int *no_expand)
5741 struct ext4_inode *raw_inode;
5742 struct ext4_xattr_ibody_header *header;
5745 raw_inode = ext4_raw_inode(iloc);
5747 header = IHDR(inode, raw_inode);
5749 /* No extended attributes present */
5750 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5751 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5752 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5753 EXT4_I(inode)->i_extra_isize, 0,
5754 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5755 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5759 /* try to expand with EAs present */
5760 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5764 * Inode size expansion failed; don't try again
5773 * Expand an inode by new_extra_isize bytes.
5774 * Returns 0 on success or negative error number on failure.
5776 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5777 unsigned int new_extra_isize,
5778 struct ext4_iloc iloc,
5784 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5788 * In nojournal mode, we can immediately attempt to expand
5789 * the inode. When journaled, we first need to obtain extra
5790 * buffer credits since we may write into the EA block
5791 * with this same handle. If journal_extend fails, then it will
5792 * only result in a minor loss of functionality for that inode.
5793 * If this is felt to be critical, then e2fsck should be run to
5794 * force a large enough s_min_extra_isize.
5796 if (ext4_handle_valid(handle) &&
5797 jbd2_journal_extend(handle,
5798 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0)
5801 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5804 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5805 handle, &no_expand);
5806 ext4_write_unlock_xattr(inode, &no_expand);
5811 int ext4_expand_extra_isize(struct inode *inode,
5812 unsigned int new_extra_isize,
5813 struct ext4_iloc *iloc)
5819 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5824 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5825 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5826 if (IS_ERR(handle)) {
5827 error = PTR_ERR(handle);
5832 ext4_write_lock_xattr(inode, &no_expand);
5834 BUFFER_TRACE(iloc.bh, "get_write_access");
5835 error = ext4_journal_get_write_access(handle, iloc->bh);
5841 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5842 handle, &no_expand);
5844 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5848 ext4_write_unlock_xattr(inode, &no_expand);
5850 ext4_journal_stop(handle);
5855 * What we do here is to mark the in-core inode as clean with respect to inode
5856 * dirtiness (it may still be data-dirty).
5857 * This means that the in-core inode may be reaped by prune_icache
5858 * without having to perform any I/O. This is a very good thing,
5859 * because *any* task may call prune_icache - even ones which
5860 * have a transaction open against a different journal.
5862 * Is this cheating? Not really. Sure, we haven't written the
5863 * inode out, but prune_icache isn't a user-visible syncing function.
5864 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5865 * we start and wait on commits.
5867 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5869 struct ext4_iloc iloc;
5870 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5874 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5875 err = ext4_reserve_inode_write(handle, inode, &iloc);
5879 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5880 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5883 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5887 * ext4_dirty_inode() is called from __mark_inode_dirty()
5889 * We're really interested in the case where a file is being extended.
5890 * i_size has been changed by generic_commit_write() and we thus need
5891 * to include the updated inode in the current transaction.
5893 * Also, dquot_alloc_block() will always dirty the inode when blocks
5894 * are allocated to the file.
5896 * If the inode is marked synchronous, we don't honour that here - doing
5897 * so would cause a commit on atime updates, which we don't bother doing.
5898 * We handle synchronous inodes at the highest possible level.
5900 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5901 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5902 * to copy into the on-disk inode structure are the timestamp files.
5904 void ext4_dirty_inode(struct inode *inode, int flags)
5908 if (flags == I_DIRTY_TIME)
5910 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5914 ext4_mark_inode_dirty(handle, inode);
5916 ext4_journal_stop(handle);
5923 * Bind an inode's backing buffer_head into this transaction, to prevent
5924 * it from being flushed to disk early. Unlike
5925 * ext4_reserve_inode_write, this leaves behind no bh reference and
5926 * returns no iloc structure, so the caller needs to repeat the iloc
5927 * lookup to mark the inode dirty later.
5929 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5931 struct ext4_iloc iloc;
5935 err = ext4_get_inode_loc(inode, &iloc);
5937 BUFFER_TRACE(iloc.bh, "get_write_access");
5938 err = jbd2_journal_get_write_access(handle, iloc.bh);
5940 err = ext4_handle_dirty_metadata(handle,
5946 ext4_std_error(inode->i_sb, err);
5951 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5956 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5959 * We have to be very careful here: changing a data block's
5960 * journaling status dynamically is dangerous. If we write a
5961 * data block to the journal, change the status and then delete
5962 * that block, we risk forgetting to revoke the old log record
5963 * from the journal and so a subsequent replay can corrupt data.
5964 * So, first we make sure that the journal is empty and that
5965 * nobody is changing anything.
5968 journal = EXT4_JOURNAL(inode);
5971 if (is_journal_aborted(journal))
5974 /* Wait for all existing dio workers */
5975 ext4_inode_block_unlocked_dio(inode);
5976 inode_dio_wait(inode);
5979 * Before flushing the journal and switching inode's aops, we have
5980 * to flush all dirty data the inode has. There can be outstanding
5981 * delayed allocations, there can be unwritten extents created by
5982 * fallocate or buffered writes in dioread_nolock mode covered by
5983 * dirty data which can be converted only after flushing the dirty
5984 * data (and journalled aops don't know how to handle these cases).
5987 down_write(&EXT4_I(inode)->i_mmap_sem);
5988 err = filemap_write_and_wait(inode->i_mapping);
5990 up_write(&EXT4_I(inode)->i_mmap_sem);
5991 ext4_inode_resume_unlocked_dio(inode);
5996 percpu_down_write(&sbi->s_journal_flag_rwsem);
5997 jbd2_journal_lock_updates(journal);
6000 * OK, there are no updates running now, and all cached data is
6001 * synced to disk. We are now in a completely consistent state
6002 * which doesn't have anything in the journal, and we know that
6003 * no filesystem updates are running, so it is safe to modify
6004 * the inode's in-core data-journaling state flag now.
6008 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6010 err = jbd2_journal_flush(journal);
6012 jbd2_journal_unlock_updates(journal);
6013 percpu_up_write(&sbi->s_journal_flag_rwsem);
6014 ext4_inode_resume_unlocked_dio(inode);
6017 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6019 ext4_set_aops(inode);
6021 jbd2_journal_unlock_updates(journal);
6022 percpu_up_write(&sbi->s_journal_flag_rwsem);
6025 up_write(&EXT4_I(inode)->i_mmap_sem);
6026 ext4_inode_resume_unlocked_dio(inode);
6028 /* Finally we can mark the inode as dirty. */
6030 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6032 return PTR_ERR(handle);
6034 err = ext4_mark_inode_dirty(handle, inode);
6035 ext4_handle_sync(handle);
6036 ext4_journal_stop(handle);
6037 ext4_std_error(inode->i_sb, err);
6042 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6044 return !buffer_mapped(bh);
6047 int ext4_page_mkwrite(struct vm_fault *vmf)
6049 struct vm_area_struct *vma = vmf->vma;
6050 struct page *page = vmf->page;
6054 struct file *file = vma->vm_file;
6055 struct inode *inode = file_inode(file);
6056 struct address_space *mapping = inode->i_mapping;
6058 get_block_t *get_block;
6061 sb_start_pagefault(inode->i_sb);
6062 file_update_time(vma->vm_file);
6064 down_read(&EXT4_I(inode)->i_mmap_sem);
6066 ret = ext4_convert_inline_data(inode);
6070 /* Delalloc case is easy... */
6071 if (test_opt(inode->i_sb, DELALLOC) &&
6072 !ext4_should_journal_data(inode) &&
6073 !ext4_nonda_switch(inode->i_sb)) {
6075 ret = block_page_mkwrite(vma, vmf,
6076 ext4_da_get_block_prep);
6077 } while (ret == -ENOSPC &&
6078 ext4_should_retry_alloc(inode->i_sb, &retries));
6083 size = i_size_read(inode);
6084 /* Page got truncated from under us? */
6085 if (page->mapping != mapping || page_offset(page) > size) {
6087 ret = VM_FAULT_NOPAGE;
6091 if (page->index == size >> PAGE_SHIFT)
6092 len = size & ~PAGE_MASK;
6096 * Return if we have all the buffers mapped. This avoids the need to do
6097 * journal_start/journal_stop which can block and take a long time
6099 if (page_has_buffers(page)) {
6100 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6102 ext4_bh_unmapped)) {
6103 /* Wait so that we don't change page under IO */
6104 wait_for_stable_page(page);
6105 ret = VM_FAULT_LOCKED;
6110 /* OK, we need to fill the hole... */
6111 if (ext4_should_dioread_nolock(inode))
6112 get_block = ext4_get_block_unwritten;
6114 get_block = ext4_get_block;
6116 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6117 ext4_writepage_trans_blocks(inode));
6118 if (IS_ERR(handle)) {
6119 ret = VM_FAULT_SIGBUS;
6122 ret = block_page_mkwrite(vma, vmf, get_block);
6123 if (!ret && ext4_should_journal_data(inode)) {
6124 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
6125 PAGE_SIZE, NULL, do_journal_get_write_access)) {
6127 ret = VM_FAULT_SIGBUS;
6128 ext4_journal_stop(handle);
6131 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6133 ext4_journal_stop(handle);
6134 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6137 ret = block_page_mkwrite_return(ret);
6139 up_read(&EXT4_I(inode)->i_mmap_sem);
6140 sb_end_pagefault(inode->i_sb);
6144 int ext4_filemap_fault(struct vm_fault *vmf)
6146 struct inode *inode = file_inode(vmf->vma->vm_file);
6149 down_read(&EXT4_I(inode)->i_mmap_sem);
6150 err = filemap_fault(vmf);
6151 up_read(&EXT4_I(inode)->i_mmap_sem);