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>
42 #include <linux/iversion.h>
44 #include "ext4_jbd2.h"
49 #include <trace/events/ext4.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
54 struct ext4_inode_info *ei)
56 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
59 int offset = offsetof(struct ext4_inode, i_checksum_lo);
60 unsigned int csum_size = sizeof(dummy_csum);
62 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
63 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
65 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
66 EXT4_GOOD_OLD_INODE_SIZE - offset);
68 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
69 offset = offsetof(struct ext4_inode, i_checksum_hi);
70 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
71 EXT4_GOOD_OLD_INODE_SIZE,
72 offset - EXT4_GOOD_OLD_INODE_SIZE);
73 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
74 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
78 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
79 EXT4_INODE_SIZE(inode->i_sb) - offset);
85 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
86 struct ext4_inode_info *ei)
88 __u32 provided, calculated;
90 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
91 cpu_to_le32(EXT4_OS_LINUX) ||
92 !ext4_has_metadata_csum(inode->i_sb))
95 provided = le16_to_cpu(raw->i_checksum_lo);
96 calculated = ext4_inode_csum(inode, raw, ei);
97 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
98 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
99 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
101 calculated &= 0xFFFF;
103 return provided == calculated;
106 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
107 struct ext4_inode_info *ei)
111 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
112 cpu_to_le32(EXT4_OS_LINUX) ||
113 !ext4_has_metadata_csum(inode->i_sb))
116 csum = ext4_inode_csum(inode, raw, ei);
117 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
118 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
119 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
120 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
123 static inline int ext4_begin_ordered_truncate(struct inode *inode,
126 trace_ext4_begin_ordered_truncate(inode, new_size);
128 * If jinode is zero, then we never opened the file for
129 * writing, so there's no need to call
130 * jbd2_journal_begin_ordered_truncate() since there's no
131 * outstanding writes we need to flush.
133 if (!EXT4_I(inode)->jinode)
135 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
136 EXT4_I(inode)->jinode,
140 static void ext4_invalidatepage(struct page *page, unsigned int offset,
141 unsigned int length);
142 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
143 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
144 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
148 * Test whether an inode is a fast symlink.
149 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
151 int ext4_inode_is_fast_symlink(struct inode *inode)
153 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
154 int ea_blocks = EXT4_I(inode)->i_file_acl ?
155 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
157 if (ext4_has_inline_data(inode))
160 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
162 return S_ISLNK(inode->i_mode) && inode->i_size &&
163 (inode->i_size < EXT4_N_BLOCKS * 4);
167 * Restart the transaction associated with *handle. This does a commit,
168 * so before we call here everything must be consistently dirtied against
171 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
177 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
178 * moment, get_block can be called only for blocks inside i_size since
179 * page cache has been already dropped and writes are blocked by
180 * i_mutex. So we can safely drop the i_data_sem here.
182 BUG_ON(EXT4_JOURNAL(inode) == NULL);
183 jbd_debug(2, "restarting handle %p\n", handle);
184 up_write(&EXT4_I(inode)->i_data_sem);
185 ret = ext4_journal_restart(handle, nblocks);
186 down_write(&EXT4_I(inode)->i_data_sem);
187 ext4_discard_preallocations(inode);
193 * Called at the last iput() if i_nlink is zero.
195 void ext4_evict_inode(struct inode *inode)
200 * Credits for final inode cleanup and freeing:
201 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
202 * (xattr block freeing), bitmap, group descriptor (inode freeing)
204 int extra_credits = 6;
205 struct ext4_xattr_inode_array *ea_inode_array = NULL;
206 bool freeze_protected = false;
208 trace_ext4_evict_inode(inode);
210 if (inode->i_nlink) {
212 * When journalling data dirty buffers are tracked only in the
213 * journal. So although mm thinks everything is clean and
214 * ready for reaping the inode might still have some pages to
215 * write in the running transaction or waiting to be
216 * checkpointed. Thus calling jbd2_journal_invalidatepage()
217 * (via truncate_inode_pages()) to discard these buffers can
218 * cause data loss. Also even if we did not discard these
219 * buffers, we would have no way to find them after the inode
220 * is reaped and thus user could see stale data if he tries to
221 * read them before the transaction is checkpointed. So be
222 * careful and force everything to disk here... We use
223 * ei->i_datasync_tid to store the newest transaction
224 * containing inode's data.
226 * Note that directories do not have this problem because they
227 * don't use page cache.
229 if (inode->i_ino != EXT4_JOURNAL_INO &&
230 ext4_should_journal_data(inode) &&
231 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
232 inode->i_data.nrpages) {
233 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
234 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
236 jbd2_complete_transaction(journal, commit_tid);
237 filemap_write_and_wait(&inode->i_data);
239 truncate_inode_pages_final(&inode->i_data);
244 if (is_bad_inode(inode))
246 dquot_initialize(inode);
248 if (ext4_should_order_data(inode))
249 ext4_begin_ordered_truncate(inode, 0);
250 truncate_inode_pages_final(&inode->i_data);
253 * Protect us against freezing - iput() caller didn't have to have any
254 * protection against it. When we are in a running transaction though,
255 * we are already protected against freezing and we cannot grab further
256 * protection due to lock ordering constraints.
258 if (!ext4_journal_current_handle()) {
259 sb_start_intwrite(inode->i_sb);
260 freeze_protected = true;
263 if (!IS_NOQUOTA(inode))
264 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
267 * Block bitmap, group descriptor, and inode are accounted in both
268 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
270 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
271 ext4_blocks_for_truncate(inode) + extra_credits - 3);
272 if (IS_ERR(handle)) {
273 ext4_std_error(inode->i_sb, PTR_ERR(handle));
275 * If we're going to skip the normal cleanup, we still need to
276 * make sure that the in-core orphan linked list is properly
279 ext4_orphan_del(NULL, inode);
280 if (freeze_protected)
281 sb_end_intwrite(inode->i_sb);
286 ext4_handle_sync(handle);
289 * Set inode->i_size to 0 before calling ext4_truncate(). We need
290 * special handling of symlinks here because i_size is used to
291 * determine whether ext4_inode_info->i_data contains symlink data or
292 * block mappings. Setting i_size to 0 will remove its fast symlink
293 * status. Erase i_data so that it becomes a valid empty block map.
295 if (ext4_inode_is_fast_symlink(inode))
296 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
298 err = ext4_mark_inode_dirty(handle, inode);
300 ext4_warning(inode->i_sb,
301 "couldn't mark inode dirty (err %d)", err);
304 if (inode->i_blocks) {
305 err = ext4_truncate(inode);
307 ext4_error(inode->i_sb,
308 "couldn't truncate inode %lu (err %d)",
314 /* Remove xattr references. */
315 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
318 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
320 ext4_journal_stop(handle);
321 ext4_orphan_del(NULL, inode);
322 if (freeze_protected)
323 sb_end_intwrite(inode->i_sb);
324 ext4_xattr_inode_array_free(ea_inode_array);
329 * Kill off the orphan record which ext4_truncate created.
330 * AKPM: I think this can be inside the above `if'.
331 * Note that ext4_orphan_del() has to be able to cope with the
332 * deletion of a non-existent orphan - this is because we don't
333 * know if ext4_truncate() actually created an orphan record.
334 * (Well, we could do this if we need to, but heck - it works)
336 ext4_orphan_del(handle, inode);
337 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
340 * One subtle ordering requirement: if anything has gone wrong
341 * (transaction abort, IO errors, whatever), then we can still
342 * do these next steps (the fs will already have been marked as
343 * having errors), but we can't free the inode if the mark_dirty
346 if (ext4_mark_inode_dirty(handle, inode))
347 /* If that failed, just do the required in-core inode clear. */
348 ext4_clear_inode(inode);
350 ext4_free_inode(handle, inode);
351 ext4_journal_stop(handle);
352 if (freeze_protected)
353 sb_end_intwrite(inode->i_sb);
354 ext4_xattr_inode_array_free(ea_inode_array);
357 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
361 qsize_t *ext4_get_reserved_space(struct inode *inode)
363 return &EXT4_I(inode)->i_reserved_quota;
368 * Called with i_data_sem down, which is important since we can call
369 * ext4_discard_preallocations() from here.
371 void ext4_da_update_reserve_space(struct inode *inode,
372 int used, int quota_claim)
374 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
375 struct ext4_inode_info *ei = EXT4_I(inode);
377 spin_lock(&ei->i_block_reservation_lock);
378 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
379 if (unlikely(used > ei->i_reserved_data_blocks)) {
380 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
381 "with only %d reserved data blocks",
382 __func__, inode->i_ino, used,
383 ei->i_reserved_data_blocks);
385 used = ei->i_reserved_data_blocks;
388 /* Update per-inode reservations */
389 ei->i_reserved_data_blocks -= used;
390 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
392 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
394 /* Update quota subsystem for data blocks */
396 dquot_claim_block(inode, EXT4_C2B(sbi, used));
399 * We did fallocate with an offset that is already delayed
400 * allocated. So on delayed allocated writeback we should
401 * not re-claim the quota for fallocated blocks.
403 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
407 * If we have done all the pending block allocations and if
408 * there aren't any writers on the inode, we can discard the
409 * inode's preallocations.
411 if ((ei->i_reserved_data_blocks == 0) &&
412 (atomic_read(&inode->i_writecount) == 0))
413 ext4_discard_preallocations(inode);
416 static int __check_block_validity(struct inode *inode, const char *func,
418 struct ext4_map_blocks *map)
420 if (ext4_has_feature_journal(inode->i_sb) &&
422 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
424 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
425 ext4_error_inode(inode, func, line, map->m_pblk,
426 "lblock %lu mapped to illegal pblock %llu "
427 "(length %d)", (unsigned long) map->m_lblk,
428 map->m_pblk, map->m_len);
429 return -EFSCORRUPTED;
434 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
439 if (ext4_encrypted_inode(inode))
440 return fscrypt_zeroout_range(inode, lblk, pblk, len);
442 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
449 #define check_block_validity(inode, map) \
450 __check_block_validity((inode), __func__, __LINE__, (map))
452 #ifdef ES_AGGRESSIVE_TEST
453 static void ext4_map_blocks_es_recheck(handle_t *handle,
455 struct ext4_map_blocks *es_map,
456 struct ext4_map_blocks *map,
463 * There is a race window that the result is not the same.
464 * e.g. xfstests #223 when dioread_nolock enables. The reason
465 * is that we lookup a block mapping in extent status tree with
466 * out taking i_data_sem. So at the time the unwritten extent
467 * could be converted.
469 down_read(&EXT4_I(inode)->i_data_sem);
470 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
471 retval = ext4_ext_map_blocks(handle, inode, map, flags &
472 EXT4_GET_BLOCKS_KEEP_SIZE);
474 retval = ext4_ind_map_blocks(handle, inode, map, flags &
475 EXT4_GET_BLOCKS_KEEP_SIZE);
477 up_read((&EXT4_I(inode)->i_data_sem));
480 * We don't check m_len because extent will be collpased in status
481 * tree. So the m_len might not equal.
483 if (es_map->m_lblk != map->m_lblk ||
484 es_map->m_flags != map->m_flags ||
485 es_map->m_pblk != map->m_pblk) {
486 printk("ES cache assertion failed for inode: %lu "
487 "es_cached ex [%d/%d/%llu/%x] != "
488 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
489 inode->i_ino, es_map->m_lblk, es_map->m_len,
490 es_map->m_pblk, es_map->m_flags, map->m_lblk,
491 map->m_len, map->m_pblk, map->m_flags,
495 #endif /* ES_AGGRESSIVE_TEST */
498 * The ext4_map_blocks() function tries to look up the requested blocks,
499 * and returns if the blocks are already mapped.
501 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
502 * and store the allocated blocks in the result buffer head and mark it
505 * If file type is extents based, it will call ext4_ext_map_blocks(),
506 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
509 * On success, it returns the number of blocks being mapped or allocated. if
510 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
511 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
513 * It returns 0 if plain look up failed (blocks have not been allocated), in
514 * that case, @map is returned as unmapped but we still do fill map->m_len to
515 * indicate the length of a hole starting at map->m_lblk.
517 * It returns the error in case of allocation failure.
519 int ext4_map_blocks(handle_t *handle, struct inode *inode,
520 struct ext4_map_blocks *map, int flags)
522 struct extent_status es;
525 #ifdef ES_AGGRESSIVE_TEST
526 struct ext4_map_blocks orig_map;
528 memcpy(&orig_map, map, sizeof(*map));
532 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
533 "logical block %lu\n", inode->i_ino, flags, map->m_len,
534 (unsigned long) map->m_lblk);
537 * ext4_map_blocks returns an int, and m_len is an unsigned int
539 if (unlikely(map->m_len > INT_MAX))
540 map->m_len = INT_MAX;
542 /* We can handle the block number less than EXT_MAX_BLOCKS */
543 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
544 return -EFSCORRUPTED;
546 /* Lookup extent status tree firstly */
547 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
548 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
549 map->m_pblk = ext4_es_pblock(&es) +
550 map->m_lblk - es.es_lblk;
551 map->m_flags |= ext4_es_is_written(&es) ?
552 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
553 retval = es.es_len - (map->m_lblk - es.es_lblk);
554 if (retval > map->m_len)
557 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
559 retval = es.es_len - (map->m_lblk - es.es_lblk);
560 if (retval > map->m_len)
567 #ifdef ES_AGGRESSIVE_TEST
568 ext4_map_blocks_es_recheck(handle, inode, map,
575 * Try to see if we can get the block without requesting a new
578 down_read(&EXT4_I(inode)->i_data_sem);
579 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
580 retval = ext4_ext_map_blocks(handle, inode, map, flags &
581 EXT4_GET_BLOCKS_KEEP_SIZE);
583 retval = ext4_ind_map_blocks(handle, inode, map, flags &
584 EXT4_GET_BLOCKS_KEEP_SIZE);
589 if (unlikely(retval != map->m_len)) {
590 ext4_warning(inode->i_sb,
591 "ES len assertion failed for inode "
592 "%lu: retval %d != map->m_len %d",
593 inode->i_ino, retval, map->m_len);
597 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
598 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
599 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
600 !(status & EXTENT_STATUS_WRITTEN) &&
601 ext4_find_delalloc_range(inode, map->m_lblk,
602 map->m_lblk + map->m_len - 1))
603 status |= EXTENT_STATUS_DELAYED;
604 ret = ext4_es_insert_extent(inode, map->m_lblk,
605 map->m_len, map->m_pblk, status);
609 up_read((&EXT4_I(inode)->i_data_sem));
612 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
613 ret = check_block_validity(inode, map);
618 /* If it is only a block(s) look up */
619 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
623 * Returns if the blocks have already allocated
625 * Note that if blocks have been preallocated
626 * ext4_ext_get_block() returns the create = 0
627 * with buffer head unmapped.
629 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
631 * If we need to convert extent to unwritten
632 * we continue and do the actual work in
633 * ext4_ext_map_blocks()
635 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
639 * Here we clear m_flags because after allocating an new extent,
640 * it will be set again.
642 map->m_flags &= ~EXT4_MAP_FLAGS;
645 * New blocks allocate and/or writing to unwritten extent
646 * will possibly result in updating i_data, so we take
647 * the write lock of i_data_sem, and call get_block()
648 * with create == 1 flag.
650 down_write(&EXT4_I(inode)->i_data_sem);
653 * We need to check for EXT4 here because migrate
654 * could have changed the inode type in between
656 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
657 retval = ext4_ext_map_blocks(handle, inode, map, flags);
659 retval = ext4_ind_map_blocks(handle, inode, map, flags);
661 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
663 * We allocated new blocks which will result in
664 * i_data's format changing. Force the migrate
665 * to fail by clearing migrate flags
667 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
671 * Update reserved blocks/metadata blocks after successful
672 * block allocation which had been deferred till now. We don't
673 * support fallocate for non extent files. So we can update
674 * reserve space here.
677 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
678 ext4_da_update_reserve_space(inode, retval, 1);
684 if (unlikely(retval != map->m_len)) {
685 ext4_warning(inode->i_sb,
686 "ES len assertion failed for inode "
687 "%lu: retval %d != map->m_len %d",
688 inode->i_ino, retval, map->m_len);
693 * We have to zeroout blocks before inserting them into extent
694 * status tree. Otherwise someone could look them up there and
695 * use them before they are really zeroed. We also have to
696 * unmap metadata before zeroing as otherwise writeback can
697 * overwrite zeros with stale data from block device.
699 if (flags & EXT4_GET_BLOCKS_ZERO &&
700 map->m_flags & EXT4_MAP_MAPPED &&
701 map->m_flags & EXT4_MAP_NEW) {
702 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
704 ret = ext4_issue_zeroout(inode, map->m_lblk,
705 map->m_pblk, map->m_len);
713 * If the extent has been zeroed out, we don't need to update
714 * extent status tree.
716 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
717 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
718 if (ext4_es_is_written(&es))
721 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
722 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
723 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
724 !(status & EXTENT_STATUS_WRITTEN) &&
725 ext4_find_delalloc_range(inode, map->m_lblk,
726 map->m_lblk + map->m_len - 1))
727 status |= EXTENT_STATUS_DELAYED;
728 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
729 map->m_pblk, status);
737 up_write((&EXT4_I(inode)->i_data_sem));
738 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
739 ret = check_block_validity(inode, map);
744 * Inodes with freshly allocated blocks where contents will be
745 * visible after transaction commit must be on transaction's
748 if (map->m_flags & EXT4_MAP_NEW &&
749 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
750 !(flags & EXT4_GET_BLOCKS_ZERO) &&
751 !ext4_is_quota_file(inode) &&
752 ext4_should_order_data(inode)) {
754 (loff_t)map->m_lblk << inode->i_blkbits;
755 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
757 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
758 ret = ext4_jbd2_inode_add_wait(handle, inode,
761 ret = ext4_jbd2_inode_add_write(handle, inode,
771 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
772 * we have to be careful as someone else may be manipulating b_state as well.
774 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
776 unsigned long old_state;
777 unsigned long new_state;
779 flags &= EXT4_MAP_FLAGS;
781 /* Dummy buffer_head? Set non-atomically. */
783 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
787 * Someone else may be modifying b_state. Be careful! This is ugly but
788 * once we get rid of using bh as a container for mapping information
789 * to pass to / from get_block functions, this can go away.
792 old_state = READ_ONCE(bh->b_state);
793 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
795 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
798 static int _ext4_get_block(struct inode *inode, sector_t iblock,
799 struct buffer_head *bh, int flags)
801 struct ext4_map_blocks map;
804 if (ext4_has_inline_data(inode))
808 map.m_len = bh->b_size >> inode->i_blkbits;
810 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
813 map_bh(bh, inode->i_sb, map.m_pblk);
814 ext4_update_bh_state(bh, map.m_flags);
815 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
817 } else if (ret == 0) {
818 /* hole case, need to fill in bh->b_size */
819 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
824 int ext4_get_block(struct inode *inode, sector_t iblock,
825 struct buffer_head *bh, int create)
827 return _ext4_get_block(inode, iblock, bh,
828 create ? EXT4_GET_BLOCKS_CREATE : 0);
832 * Get block function used when preparing for buffered write if we require
833 * creating an unwritten extent if blocks haven't been allocated. The extent
834 * will be converted to written after the IO is complete.
836 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
837 struct buffer_head *bh_result, int create)
839 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
840 inode->i_ino, create);
841 return _ext4_get_block(inode, iblock, bh_result,
842 EXT4_GET_BLOCKS_IO_CREATE_EXT);
845 /* Maximum number of blocks we map for direct IO at once. */
846 #define DIO_MAX_BLOCKS 4096
849 * Get blocks function for the cases that need to start a transaction -
850 * generally difference cases of direct IO and DAX IO. It also handles retries
853 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
854 struct buffer_head *bh_result, int flags)
861 /* Trim mapping request to maximum we can map at once for DIO */
862 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
863 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
864 dio_credits = ext4_chunk_trans_blocks(inode,
865 bh_result->b_size >> inode->i_blkbits);
867 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
869 return PTR_ERR(handle);
871 ret = _ext4_get_block(inode, iblock, bh_result, flags);
872 ext4_journal_stop(handle);
874 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
879 /* Get block function for DIO reads and writes to inodes without extents */
880 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
881 struct buffer_head *bh, int create)
883 /* We don't expect handle for direct IO */
884 WARN_ON_ONCE(ext4_journal_current_handle());
887 return _ext4_get_block(inode, iblock, bh, 0);
888 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
892 * Get block function for AIO DIO writes when we create unwritten extent if
893 * blocks are not allocated yet. The extent will be converted to written
894 * after IO is complete.
896 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
897 sector_t iblock, struct buffer_head *bh_result, int create)
901 /* We don't expect handle for direct IO */
902 WARN_ON_ONCE(ext4_journal_current_handle());
904 ret = ext4_get_block_trans(inode, iblock, bh_result,
905 EXT4_GET_BLOCKS_IO_CREATE_EXT);
908 * When doing DIO using unwritten extents, we need io_end to convert
909 * unwritten extents to written on IO completion. We allocate io_end
910 * once we spot unwritten extent and store it in b_private. Generic
911 * DIO code keeps b_private set and furthermore passes the value to
912 * our completion callback in 'private' argument.
914 if (!ret && buffer_unwritten(bh_result)) {
915 if (!bh_result->b_private) {
916 ext4_io_end_t *io_end;
918 io_end = ext4_init_io_end(inode, GFP_KERNEL);
921 bh_result->b_private = io_end;
922 ext4_set_io_unwritten_flag(inode, io_end);
924 set_buffer_defer_completion(bh_result);
931 * Get block function for non-AIO DIO writes when we create unwritten extent if
932 * blocks are not allocated yet. The extent will be converted to written
933 * after IO is complete by ext4_direct_IO_write().
935 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
936 sector_t iblock, struct buffer_head *bh_result, int create)
940 /* We don't expect handle for direct IO */
941 WARN_ON_ONCE(ext4_journal_current_handle());
943 ret = ext4_get_block_trans(inode, iblock, bh_result,
944 EXT4_GET_BLOCKS_IO_CREATE_EXT);
947 * Mark inode as having pending DIO writes to unwritten extents.
948 * ext4_direct_IO_write() checks this flag and converts extents to
951 if (!ret && buffer_unwritten(bh_result))
952 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
957 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
958 struct buffer_head *bh_result, int create)
962 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
963 inode->i_ino, create);
964 /* We don't expect handle for direct IO */
965 WARN_ON_ONCE(ext4_journal_current_handle());
967 ret = _ext4_get_block(inode, iblock, bh_result, 0);
969 * Blocks should have been preallocated! ext4_file_write_iter() checks
972 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
979 * `handle' can be NULL if create is zero
981 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
982 ext4_lblk_t block, int map_flags)
984 struct ext4_map_blocks map;
985 struct buffer_head *bh;
986 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
989 J_ASSERT(handle != NULL || create == 0);
993 err = ext4_map_blocks(handle, inode, &map, map_flags);
996 return create ? ERR_PTR(-ENOSPC) : NULL;
1000 bh = sb_getblk(inode->i_sb, map.m_pblk);
1002 return ERR_PTR(-ENOMEM);
1003 if (map.m_flags & EXT4_MAP_NEW) {
1004 J_ASSERT(create != 0);
1005 J_ASSERT(handle != NULL);
1008 * Now that we do not always journal data, we should
1009 * keep in mind whether this should always journal the
1010 * new buffer as metadata. For now, regular file
1011 * writes use ext4_get_block instead, so it's not a
1015 BUFFER_TRACE(bh, "call get_create_access");
1016 err = ext4_journal_get_create_access(handle, bh);
1017 if (unlikely(err)) {
1021 if (!buffer_uptodate(bh)) {
1022 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1023 set_buffer_uptodate(bh);
1026 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1027 err = ext4_handle_dirty_metadata(handle, inode, bh);
1031 BUFFER_TRACE(bh, "not a new buffer");
1035 return ERR_PTR(err);
1038 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1039 ext4_lblk_t block, int map_flags)
1041 struct buffer_head *bh;
1043 bh = ext4_getblk(handle, inode, block, map_flags);
1046 if (!bh || buffer_uptodate(bh))
1048 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1050 if (buffer_uptodate(bh))
1053 return ERR_PTR(-EIO);
1056 /* Read a contiguous batch of blocks. */
1057 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1058 bool wait, struct buffer_head **bhs)
1062 for (i = 0; i < bh_count; i++) {
1063 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1064 if (IS_ERR(bhs[i])) {
1065 err = PTR_ERR(bhs[i]);
1071 for (i = 0; i < bh_count; i++)
1072 /* Note that NULL bhs[i] is valid because of holes. */
1073 if (bhs[i] && !buffer_uptodate(bhs[i]))
1074 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1080 for (i = 0; i < bh_count; i++)
1082 wait_on_buffer(bhs[i]);
1084 for (i = 0; i < bh_count; i++) {
1085 if (bhs[i] && !buffer_uptodate(bhs[i])) {
1093 for (i = 0; i < bh_count; i++) {
1100 int ext4_walk_page_buffers(handle_t *handle,
1101 struct buffer_head *head,
1105 int (*fn)(handle_t *handle,
1106 struct buffer_head *bh))
1108 struct buffer_head *bh;
1109 unsigned block_start, block_end;
1110 unsigned blocksize = head->b_size;
1112 struct buffer_head *next;
1114 for (bh = head, block_start = 0;
1115 ret == 0 && (bh != head || !block_start);
1116 block_start = block_end, bh = next) {
1117 next = bh->b_this_page;
1118 block_end = block_start + blocksize;
1119 if (block_end <= from || block_start >= to) {
1120 if (partial && !buffer_uptodate(bh))
1124 err = (*fn)(handle, bh);
1132 * To preserve ordering, it is essential that the hole instantiation and
1133 * the data write be encapsulated in a single transaction. We cannot
1134 * close off a transaction and start a new one between the ext4_get_block()
1135 * and the commit_write(). So doing the jbd2_journal_start at the start of
1136 * prepare_write() is the right place.
1138 * Also, this function can nest inside ext4_writepage(). In that case, we
1139 * *know* that ext4_writepage() has generated enough buffer credits to do the
1140 * whole page. So we won't block on the journal in that case, which is good,
1141 * because the caller may be PF_MEMALLOC.
1143 * By accident, ext4 can be reentered when a transaction is open via
1144 * quota file writes. If we were to commit the transaction while thus
1145 * reentered, there can be a deadlock - we would be holding a quota
1146 * lock, and the commit would never complete if another thread had a
1147 * transaction open and was blocking on the quota lock - a ranking
1150 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1151 * will _not_ run commit under these circumstances because handle->h_ref
1152 * is elevated. We'll still have enough credits for the tiny quotafile
1155 int do_journal_get_write_access(handle_t *handle,
1156 struct buffer_head *bh)
1158 int dirty = buffer_dirty(bh);
1161 if (!buffer_mapped(bh) || buffer_freed(bh))
1164 * __block_write_begin() could have dirtied some buffers. Clean
1165 * the dirty bit as jbd2_journal_get_write_access() could complain
1166 * otherwise about fs integrity issues. Setting of the dirty bit
1167 * by __block_write_begin() isn't a real problem here as we clear
1168 * the bit before releasing a page lock and thus writeback cannot
1169 * ever write the buffer.
1172 clear_buffer_dirty(bh);
1173 BUFFER_TRACE(bh, "get write access");
1174 ret = ext4_journal_get_write_access(handle, bh);
1176 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1180 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1181 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1182 get_block_t *get_block)
1184 unsigned from = pos & (PAGE_SIZE - 1);
1185 unsigned to = from + len;
1186 struct inode *inode = page->mapping->host;
1187 unsigned block_start, block_end;
1190 unsigned blocksize = inode->i_sb->s_blocksize;
1192 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1193 bool decrypt = false;
1195 BUG_ON(!PageLocked(page));
1196 BUG_ON(from > PAGE_SIZE);
1197 BUG_ON(to > PAGE_SIZE);
1200 if (!page_has_buffers(page))
1201 create_empty_buffers(page, blocksize, 0);
1202 head = page_buffers(page);
1203 bbits = ilog2(blocksize);
1204 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1206 for (bh = head, block_start = 0; bh != head || !block_start;
1207 block++, block_start = block_end, bh = bh->b_this_page) {
1208 block_end = block_start + blocksize;
1209 if (block_end <= from || block_start >= to) {
1210 if (PageUptodate(page)) {
1211 if (!buffer_uptodate(bh))
1212 set_buffer_uptodate(bh);
1217 clear_buffer_new(bh);
1218 if (!buffer_mapped(bh)) {
1219 WARN_ON(bh->b_size != blocksize);
1220 err = get_block(inode, block, bh, 1);
1223 if (buffer_new(bh)) {
1224 clean_bdev_bh_alias(bh);
1225 if (PageUptodate(page)) {
1226 clear_buffer_new(bh);
1227 set_buffer_uptodate(bh);
1228 mark_buffer_dirty(bh);
1231 if (block_end > to || block_start < from)
1232 zero_user_segments(page, to, block_end,
1237 if (PageUptodate(page)) {
1238 if (!buffer_uptodate(bh))
1239 set_buffer_uptodate(bh);
1242 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1243 !buffer_unwritten(bh) &&
1244 (block_start < from || block_end > to)) {
1245 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1247 decrypt = ext4_encrypted_inode(inode) &&
1248 S_ISREG(inode->i_mode);
1252 * If we issued read requests, let them complete.
1254 while (wait_bh > wait) {
1255 wait_on_buffer(*--wait_bh);
1256 if (!buffer_uptodate(*wait_bh))
1260 page_zero_new_buffers(page, from, to);
1262 err = fscrypt_decrypt_page(page->mapping->host, page,
1263 PAGE_SIZE, 0, page->index);
1268 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1269 loff_t pos, unsigned len, unsigned flags,
1270 struct page **pagep, void **fsdata)
1272 struct inode *inode = mapping->host;
1273 int ret, needed_blocks;
1280 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1283 trace_ext4_write_begin(inode, pos, len, flags);
1285 * Reserve one block more for addition to orphan list in case
1286 * we allocate blocks but write fails for some reason
1288 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1289 index = pos >> PAGE_SHIFT;
1290 from = pos & (PAGE_SIZE - 1);
1293 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1294 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1303 * grab_cache_page_write_begin() can take a long time if the
1304 * system is thrashing due to memory pressure, or if the page
1305 * is being written back. So grab it first before we start
1306 * the transaction handle. This also allows us to allocate
1307 * the page (if needed) without using GFP_NOFS.
1310 page = grab_cache_page_write_begin(mapping, index, flags);
1316 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1317 if (IS_ERR(handle)) {
1319 return PTR_ERR(handle);
1323 if (page->mapping != mapping) {
1324 /* The page got truncated from under us */
1327 ext4_journal_stop(handle);
1330 /* In case writeback began while the page was unlocked */
1331 wait_for_stable_page(page);
1333 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1334 if (ext4_should_dioread_nolock(inode))
1335 ret = ext4_block_write_begin(page, pos, len,
1336 ext4_get_block_unwritten);
1338 ret = ext4_block_write_begin(page, pos, len,
1341 if (ext4_should_dioread_nolock(inode))
1342 ret = __block_write_begin(page, pos, len,
1343 ext4_get_block_unwritten);
1345 ret = __block_write_begin(page, pos, len, ext4_get_block);
1347 if (!ret && ext4_should_journal_data(inode)) {
1348 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1350 do_journal_get_write_access);
1356 * __block_write_begin may have instantiated a few blocks
1357 * outside i_size. Trim these off again. Don't need
1358 * i_size_read because we hold i_mutex.
1360 * Add inode to orphan list in case we crash before
1363 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1364 ext4_orphan_add(handle, inode);
1366 ext4_journal_stop(handle);
1367 if (pos + len > inode->i_size) {
1368 ext4_truncate_failed_write(inode);
1370 * If truncate failed early the inode might
1371 * still be on the orphan list; we need to
1372 * make sure the inode is removed from the
1373 * orphan list in that case.
1376 ext4_orphan_del(NULL, inode);
1379 if (ret == -ENOSPC &&
1380 ext4_should_retry_alloc(inode->i_sb, &retries))
1389 /* For write_end() in data=journal mode */
1390 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1393 if (!buffer_mapped(bh) || buffer_freed(bh))
1395 set_buffer_uptodate(bh);
1396 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1397 clear_buffer_meta(bh);
1398 clear_buffer_prio(bh);
1403 * We need to pick up the new inode size which generic_commit_write gave us
1404 * `file' can be NULL - eg, when called from page_symlink().
1406 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1407 * buffers are managed internally.
1409 static int ext4_write_end(struct file *file,
1410 struct address_space *mapping,
1411 loff_t pos, unsigned len, unsigned copied,
1412 struct page *page, void *fsdata)
1414 handle_t *handle = ext4_journal_current_handle();
1415 struct inode *inode = mapping->host;
1416 loff_t old_size = inode->i_size;
1418 int i_size_changed = 0;
1419 int inline_data = ext4_has_inline_data(inode);
1421 trace_ext4_write_end(inode, pos, len, copied);
1423 ret = ext4_write_inline_data_end(inode, pos, len,
1432 copied = block_write_end(file, mapping, pos,
1433 len, copied, page, fsdata);
1435 * it's important to update i_size while still holding page lock:
1436 * page writeout could otherwise come in and zero beyond i_size.
1438 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1443 pagecache_isize_extended(inode, old_size, pos);
1445 * Don't mark the inode dirty under page lock. First, it unnecessarily
1446 * makes the holding time of page lock longer. Second, it forces lock
1447 * ordering of page lock and transaction start for journaling
1450 if (i_size_changed || inline_data)
1451 ext4_mark_inode_dirty(handle, inode);
1453 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1454 /* if we have allocated more blocks and copied
1455 * less. We will have blocks allocated outside
1456 * inode->i_size. So truncate them
1458 ext4_orphan_add(handle, inode);
1460 ret2 = ext4_journal_stop(handle);
1464 if (pos + len > inode->i_size) {
1465 ext4_truncate_failed_write(inode);
1467 * If truncate failed early the inode might still be
1468 * on the orphan list; we need to make sure the inode
1469 * is removed from the orphan list in that case.
1472 ext4_orphan_del(NULL, inode);
1475 return ret ? ret : copied;
1479 * This is a private version of page_zero_new_buffers() which doesn't
1480 * set the buffer to be dirty, since in data=journalled mode we need
1481 * to call ext4_handle_dirty_metadata() instead.
1483 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1485 unsigned from, unsigned to)
1487 unsigned int block_start = 0, block_end;
1488 struct buffer_head *head, *bh;
1490 bh = head = page_buffers(page);
1492 block_end = block_start + bh->b_size;
1493 if (buffer_new(bh)) {
1494 if (block_end > from && block_start < to) {
1495 if (!PageUptodate(page)) {
1496 unsigned start, size;
1498 start = max(from, block_start);
1499 size = min(to, block_end) - start;
1501 zero_user(page, start, size);
1502 write_end_fn(handle, bh);
1504 clear_buffer_new(bh);
1507 block_start = block_end;
1508 bh = bh->b_this_page;
1509 } while (bh != head);
1512 static int ext4_journalled_write_end(struct file *file,
1513 struct address_space *mapping,
1514 loff_t pos, unsigned len, unsigned copied,
1515 struct page *page, void *fsdata)
1517 handle_t *handle = ext4_journal_current_handle();
1518 struct inode *inode = mapping->host;
1519 loff_t old_size = inode->i_size;
1523 int size_changed = 0;
1524 int inline_data = ext4_has_inline_data(inode);
1526 trace_ext4_journalled_write_end(inode, pos, len, copied);
1527 from = pos & (PAGE_SIZE - 1);
1530 BUG_ON(!ext4_handle_valid(handle));
1533 ret = ext4_write_inline_data_end(inode, pos, len,
1541 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1543 ext4_journalled_zero_new_buffers(handle, page, from, to);
1545 if (unlikely(copied < len))
1546 ext4_journalled_zero_new_buffers(handle, page,
1548 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1549 from + copied, &partial,
1552 SetPageUptodate(page);
1554 size_changed = ext4_update_inode_size(inode, pos + copied);
1555 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1556 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1561 pagecache_isize_extended(inode, old_size, pos);
1563 if (size_changed || inline_data) {
1564 ret2 = ext4_mark_inode_dirty(handle, inode);
1569 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1570 /* if we have allocated more blocks and copied
1571 * less. We will have blocks allocated outside
1572 * inode->i_size. So truncate them
1574 ext4_orphan_add(handle, inode);
1577 ret2 = ext4_journal_stop(handle);
1580 if (pos + len > inode->i_size) {
1581 ext4_truncate_failed_write(inode);
1583 * If truncate failed early the inode might still be
1584 * on the orphan list; we need to make sure the inode
1585 * is removed from the orphan list in that case.
1588 ext4_orphan_del(NULL, inode);
1591 return ret ? ret : copied;
1595 * Reserve space for a single cluster
1597 static int ext4_da_reserve_space(struct inode *inode)
1599 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1600 struct ext4_inode_info *ei = EXT4_I(inode);
1604 * We will charge metadata quota at writeout time; this saves
1605 * us from metadata over-estimation, though we may go over by
1606 * a small amount in the end. Here we just reserve for data.
1608 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1612 spin_lock(&ei->i_block_reservation_lock);
1613 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1614 spin_unlock(&ei->i_block_reservation_lock);
1615 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1618 ei->i_reserved_data_blocks++;
1619 trace_ext4_da_reserve_space(inode);
1620 spin_unlock(&ei->i_block_reservation_lock);
1622 return 0; /* success */
1625 static void ext4_da_release_space(struct inode *inode, int to_free)
1627 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1628 struct ext4_inode_info *ei = EXT4_I(inode);
1631 return; /* Nothing to release, exit */
1633 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1635 trace_ext4_da_release_space(inode, to_free);
1636 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1638 * if there aren't enough reserved blocks, then the
1639 * counter is messed up somewhere. Since this
1640 * function is called from invalidate page, it's
1641 * harmless to return without any action.
1643 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1644 "ino %lu, to_free %d with only %d reserved "
1645 "data blocks", inode->i_ino, to_free,
1646 ei->i_reserved_data_blocks);
1648 to_free = ei->i_reserved_data_blocks;
1650 ei->i_reserved_data_blocks -= to_free;
1652 /* update fs dirty data blocks counter */
1653 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1655 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1657 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1660 static void ext4_da_page_release_reservation(struct page *page,
1661 unsigned int offset,
1662 unsigned int length)
1664 int to_release = 0, contiguous_blks = 0;
1665 struct buffer_head *head, *bh;
1666 unsigned int curr_off = 0;
1667 struct inode *inode = page->mapping->host;
1668 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1669 unsigned int stop = offset + length;
1673 BUG_ON(stop > PAGE_SIZE || stop < length);
1675 head = page_buffers(page);
1678 unsigned int next_off = curr_off + bh->b_size;
1680 if (next_off > stop)
1683 if ((offset <= curr_off) && (buffer_delay(bh))) {
1686 clear_buffer_delay(bh);
1687 } else if (contiguous_blks) {
1688 lblk = page->index <<
1689 (PAGE_SHIFT - inode->i_blkbits);
1690 lblk += (curr_off >> inode->i_blkbits) -
1692 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1693 contiguous_blks = 0;
1695 curr_off = next_off;
1696 } while ((bh = bh->b_this_page) != head);
1698 if (contiguous_blks) {
1699 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1700 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1701 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1704 /* If we have released all the blocks belonging to a cluster, then we
1705 * need to release the reserved space for that cluster. */
1706 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1707 while (num_clusters > 0) {
1708 lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1709 ((num_clusters - 1) << sbi->s_cluster_bits);
1710 if (sbi->s_cluster_ratio == 1 ||
1711 !ext4_find_delalloc_cluster(inode, lblk))
1712 ext4_da_release_space(inode, 1);
1719 * Delayed allocation stuff
1722 struct mpage_da_data {
1723 struct inode *inode;
1724 struct writeback_control *wbc;
1726 pgoff_t first_page; /* The first page to write */
1727 pgoff_t next_page; /* Current page to examine */
1728 pgoff_t last_page; /* Last page to examine */
1730 * Extent to map - this can be after first_page because that can be
1731 * fully mapped. We somewhat abuse m_flags to store whether the extent
1732 * is delalloc or unwritten.
1734 struct ext4_map_blocks map;
1735 struct ext4_io_submit io_submit; /* IO submission data */
1736 unsigned int do_map:1;
1739 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1744 struct pagevec pvec;
1745 struct inode *inode = mpd->inode;
1746 struct address_space *mapping = inode->i_mapping;
1748 /* This is necessary when next_page == 0. */
1749 if (mpd->first_page >= mpd->next_page)
1752 index = mpd->first_page;
1753 end = mpd->next_page - 1;
1755 ext4_lblk_t start, last;
1756 start = index << (PAGE_SHIFT - inode->i_blkbits);
1757 last = end << (PAGE_SHIFT - inode->i_blkbits);
1758 ext4_es_remove_extent(inode, start, last - start + 1);
1761 pagevec_init(&pvec);
1762 while (index <= end) {
1763 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1766 for (i = 0; i < nr_pages; i++) {
1767 struct page *page = pvec.pages[i];
1769 BUG_ON(!PageLocked(page));
1770 BUG_ON(PageWriteback(page));
1772 if (page_mapped(page))
1773 clear_page_dirty_for_io(page);
1774 block_invalidatepage(page, 0, PAGE_SIZE);
1775 ClearPageUptodate(page);
1779 pagevec_release(&pvec);
1783 static void ext4_print_free_blocks(struct inode *inode)
1785 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1786 struct super_block *sb = inode->i_sb;
1787 struct ext4_inode_info *ei = EXT4_I(inode);
1789 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1790 EXT4_C2B(EXT4_SB(inode->i_sb),
1791 ext4_count_free_clusters(sb)));
1792 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1793 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1794 (long long) EXT4_C2B(EXT4_SB(sb),
1795 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1796 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1797 (long long) EXT4_C2B(EXT4_SB(sb),
1798 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1799 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1800 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1801 ei->i_reserved_data_blocks);
1805 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1807 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1811 * This function is grabs code from the very beginning of
1812 * ext4_map_blocks, but assumes that the caller is from delayed write
1813 * time. This function looks up the requested blocks and sets the
1814 * buffer delay bit under the protection of i_data_sem.
1816 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1817 struct ext4_map_blocks *map,
1818 struct buffer_head *bh)
1820 struct extent_status es;
1822 sector_t invalid_block = ~((sector_t) 0xffff);
1823 #ifdef ES_AGGRESSIVE_TEST
1824 struct ext4_map_blocks orig_map;
1826 memcpy(&orig_map, map, sizeof(*map));
1829 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1833 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1834 "logical block %lu\n", inode->i_ino, map->m_len,
1835 (unsigned long) map->m_lblk);
1837 /* Lookup extent status tree firstly */
1838 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1839 if (ext4_es_is_hole(&es)) {
1841 down_read(&EXT4_I(inode)->i_data_sem);
1846 * Delayed extent could be allocated by fallocate.
1847 * So we need to check it.
1849 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1850 map_bh(bh, inode->i_sb, invalid_block);
1852 set_buffer_delay(bh);
1856 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1857 retval = es.es_len - (iblock - es.es_lblk);
1858 if (retval > map->m_len)
1859 retval = map->m_len;
1860 map->m_len = retval;
1861 if (ext4_es_is_written(&es))
1862 map->m_flags |= EXT4_MAP_MAPPED;
1863 else if (ext4_es_is_unwritten(&es))
1864 map->m_flags |= EXT4_MAP_UNWRITTEN;
1868 #ifdef ES_AGGRESSIVE_TEST
1869 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1875 * Try to see if we can get the block without requesting a new
1876 * file system block.
1878 down_read(&EXT4_I(inode)->i_data_sem);
1879 if (ext4_has_inline_data(inode))
1881 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1882 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1884 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1890 * XXX: __block_prepare_write() unmaps passed block,
1894 * If the block was allocated from previously allocated cluster,
1895 * then we don't need to reserve it again. However we still need
1896 * to reserve metadata for every block we're going to write.
1898 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1899 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1900 ret = ext4_da_reserve_space(inode);
1902 /* not enough space to reserve */
1908 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1909 ~0, EXTENT_STATUS_DELAYED);
1915 map_bh(bh, inode->i_sb, invalid_block);
1917 set_buffer_delay(bh);
1918 } else if (retval > 0) {
1920 unsigned int status;
1922 if (unlikely(retval != map->m_len)) {
1923 ext4_warning(inode->i_sb,
1924 "ES len assertion failed for inode "
1925 "%lu: retval %d != map->m_len %d",
1926 inode->i_ino, retval, map->m_len);
1930 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1931 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1932 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1933 map->m_pblk, status);
1939 up_read((&EXT4_I(inode)->i_data_sem));
1945 * This is a special get_block_t callback which is used by
1946 * ext4_da_write_begin(). It will either return mapped block or
1947 * reserve space for a single block.
1949 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1950 * We also have b_blocknr = -1 and b_bdev initialized properly
1952 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1953 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1954 * initialized properly.
1956 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1957 struct buffer_head *bh, int create)
1959 struct ext4_map_blocks map;
1962 BUG_ON(create == 0);
1963 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1965 map.m_lblk = iblock;
1969 * first, we need to know whether the block is allocated already
1970 * preallocated blocks are unmapped but should treated
1971 * the same as allocated blocks.
1973 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1977 map_bh(bh, inode->i_sb, map.m_pblk);
1978 ext4_update_bh_state(bh, map.m_flags);
1980 if (buffer_unwritten(bh)) {
1981 /* A delayed write to unwritten bh should be marked
1982 * new and mapped. Mapped ensures that we don't do
1983 * get_block multiple times when we write to the same
1984 * offset and new ensures that we do proper zero out
1985 * for partial write.
1988 set_buffer_mapped(bh);
1993 static int bget_one(handle_t *handle, struct buffer_head *bh)
1999 static int bput_one(handle_t *handle, struct buffer_head *bh)
2005 static int __ext4_journalled_writepage(struct page *page,
2008 struct address_space *mapping = page->mapping;
2009 struct inode *inode = mapping->host;
2010 struct buffer_head *page_bufs = NULL;
2011 handle_t *handle = NULL;
2012 int ret = 0, err = 0;
2013 int inline_data = ext4_has_inline_data(inode);
2014 struct buffer_head *inode_bh = NULL;
2016 ClearPageChecked(page);
2019 BUG_ON(page->index != 0);
2020 BUG_ON(len > ext4_get_max_inline_size(inode));
2021 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2022 if (inode_bh == NULL)
2025 page_bufs = page_buffers(page);
2030 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2034 * We need to release the page lock before we start the
2035 * journal, so grab a reference so the page won't disappear
2036 * out from under us.
2041 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2042 ext4_writepage_trans_blocks(inode));
2043 if (IS_ERR(handle)) {
2044 ret = PTR_ERR(handle);
2046 goto out_no_pagelock;
2048 BUG_ON(!ext4_handle_valid(handle));
2052 if (page->mapping != mapping) {
2053 /* The page got truncated from under us */
2054 ext4_journal_stop(handle);
2060 ret = ext4_mark_inode_dirty(handle, inode);
2062 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2063 do_journal_get_write_access);
2065 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2070 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2071 err = ext4_journal_stop(handle);
2075 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2079 if (!inline_data && page_bufs)
2080 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2087 * Note that we don't need to start a transaction unless we're journaling data
2088 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2089 * need to file the inode to the transaction's list in ordered mode because if
2090 * we are writing back data added by write(), the inode is already there and if
2091 * we are writing back data modified via mmap(), no one guarantees in which
2092 * transaction the data will hit the disk. In case we are journaling data, we
2093 * cannot start transaction directly because transaction start ranks above page
2094 * lock so we have to do some magic.
2096 * This function can get called via...
2097 * - ext4_writepages after taking page lock (have journal handle)
2098 * - journal_submit_inode_data_buffers (no journal handle)
2099 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2100 * - grab_page_cache when doing write_begin (have journal handle)
2102 * We don't do any block allocation in this function. If we have page with
2103 * multiple blocks we need to write those buffer_heads that are mapped. This
2104 * is important for mmaped based write. So if we do with blocksize 1K
2105 * truncate(f, 1024);
2106 * a = mmap(f, 0, 4096);
2108 * truncate(f, 4096);
2109 * we have in the page first buffer_head mapped via page_mkwrite call back
2110 * but other buffer_heads would be unmapped but dirty (dirty done via the
2111 * do_wp_page). So writepage should write the first block. If we modify
2112 * the mmap area beyond 1024 we will again get a page_fault and the
2113 * page_mkwrite callback will do the block allocation and mark the
2114 * buffer_heads mapped.
2116 * We redirty the page if we have any buffer_heads that is either delay or
2117 * unwritten in the page.
2119 * We can get recursively called as show below.
2121 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2124 * But since we don't do any block allocation we should not deadlock.
2125 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2127 static int ext4_writepage(struct page *page,
2128 struct writeback_control *wbc)
2133 struct buffer_head *page_bufs = NULL;
2134 struct inode *inode = page->mapping->host;
2135 struct ext4_io_submit io_submit;
2136 bool keep_towrite = false;
2138 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2139 inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
2144 trace_ext4_writepage(page);
2145 size = i_size_read(inode);
2146 if (page->index == size >> PAGE_SHIFT)
2147 len = size & ~PAGE_MASK;
2151 page_bufs = page_buffers(page);
2153 * We cannot do block allocation or other extent handling in this
2154 * function. If there are buffers needing that, we have to redirty
2155 * the page. But we may reach here when we do a journal commit via
2156 * journal_submit_inode_data_buffers() and in that case we must write
2157 * allocated buffers to achieve data=ordered mode guarantees.
2159 * Also, if there is only one buffer per page (the fs block
2160 * size == the page size), if one buffer needs block
2161 * allocation or needs to modify the extent tree to clear the
2162 * unwritten flag, we know that the page can't be written at
2163 * all, so we might as well refuse the write immediately.
2164 * Unfortunately if the block size != page size, we can't as
2165 * easily detect this case using ext4_walk_page_buffers(), but
2166 * for the extremely common case, this is an optimization that
2167 * skips a useless round trip through ext4_bio_write_page().
2169 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2170 ext4_bh_delay_or_unwritten)) {
2171 redirty_page_for_writepage(wbc, page);
2172 if ((current->flags & PF_MEMALLOC) ||
2173 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2175 * For memory cleaning there's no point in writing only
2176 * some buffers. So just bail out. Warn if we came here
2177 * from direct reclaim.
2179 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2184 keep_towrite = true;
2187 if (PageChecked(page) && ext4_should_journal_data(inode))
2189 * It's mmapped pagecache. Add buffers and journal it. There
2190 * doesn't seem much point in redirtying the page here.
2192 return __ext4_journalled_writepage(page, len);
2194 ext4_io_submit_init(&io_submit, wbc);
2195 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2196 if (!io_submit.io_end) {
2197 redirty_page_for_writepage(wbc, page);
2201 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2202 ext4_io_submit(&io_submit);
2203 /* Drop io_end reference we got from init */
2204 ext4_put_io_end_defer(io_submit.io_end);
2208 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2214 BUG_ON(page->index != mpd->first_page);
2215 clear_page_dirty_for_io(page);
2217 * We have to be very careful here! Nothing protects writeback path
2218 * against i_size changes and the page can be writeably mapped into
2219 * page tables. So an application can be growing i_size and writing
2220 * data through mmap while writeback runs. clear_page_dirty_for_io()
2221 * write-protects our page in page tables and the page cannot get
2222 * written to again until we release page lock. So only after
2223 * clear_page_dirty_for_io() we are safe to sample i_size for
2224 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2225 * on the barrier provided by TestClearPageDirty in
2226 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2227 * after page tables are updated.
2229 size = i_size_read(mpd->inode);
2230 if (page->index == size >> PAGE_SHIFT)
2231 len = size & ~PAGE_MASK;
2234 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2236 mpd->wbc->nr_to_write--;
2242 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2245 * mballoc gives us at most this number of blocks...
2246 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2247 * The rest of mballoc seems to handle chunks up to full group size.
2249 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2252 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2254 * @mpd - extent of blocks
2255 * @lblk - logical number of the block in the file
2256 * @bh - buffer head we want to add to the extent
2258 * The function is used to collect contig. blocks in the same state. If the
2259 * buffer doesn't require mapping for writeback and we haven't started the
2260 * extent of buffers to map yet, the function returns 'true' immediately - the
2261 * caller can write the buffer right away. Otherwise the function returns true
2262 * if the block has been added to the extent, false if the block couldn't be
2265 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2266 struct buffer_head *bh)
2268 struct ext4_map_blocks *map = &mpd->map;
2270 /* Buffer that doesn't need mapping for writeback? */
2271 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2272 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2273 /* So far no extent to map => we write the buffer right away */
2274 if (map->m_len == 0)
2279 /* First block in the extent? */
2280 if (map->m_len == 0) {
2281 /* We cannot map unless handle is started... */
2286 map->m_flags = bh->b_state & BH_FLAGS;
2290 /* Don't go larger than mballoc is willing to allocate */
2291 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2294 /* Can we merge the block to our big extent? */
2295 if (lblk == map->m_lblk + map->m_len &&
2296 (bh->b_state & BH_FLAGS) == map->m_flags) {
2304 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2306 * @mpd - extent of blocks for mapping
2307 * @head - the first buffer in the page
2308 * @bh - buffer we should start processing from
2309 * @lblk - logical number of the block in the file corresponding to @bh
2311 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2312 * the page for IO if all buffers in this page were mapped and there's no
2313 * accumulated extent of buffers to map or add buffers in the page to the
2314 * extent of buffers to map. The function returns 1 if the caller can continue
2315 * by processing the next page, 0 if it should stop adding buffers to the
2316 * extent to map because we cannot extend it anymore. It can also return value
2317 * < 0 in case of error during IO submission.
2319 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2320 struct buffer_head *head,
2321 struct buffer_head *bh,
2324 struct inode *inode = mpd->inode;
2326 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2327 >> inode->i_blkbits;
2330 BUG_ON(buffer_locked(bh));
2332 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2333 /* Found extent to map? */
2336 /* Buffer needs mapping and handle is not started? */
2339 /* Everything mapped so far and we hit EOF */
2342 } while (lblk++, (bh = bh->b_this_page) != head);
2343 /* So far everything mapped? Submit the page for IO. */
2344 if (mpd->map.m_len == 0) {
2345 err = mpage_submit_page(mpd, head->b_page);
2349 return lblk < blocks;
2353 * mpage_map_buffers - update buffers corresponding to changed extent and
2354 * submit fully mapped pages for IO
2356 * @mpd - description of extent to map, on return next extent to map
2358 * Scan buffers corresponding to changed extent (we expect corresponding pages
2359 * to be already locked) and update buffer state according to new extent state.
2360 * We map delalloc buffers to their physical location, clear unwritten bits,
2361 * and mark buffers as uninit when we perform writes to unwritten extents
2362 * and do extent conversion after IO is finished. If the last page is not fully
2363 * mapped, we update @map to the next extent in the last page that needs
2364 * mapping. Otherwise we submit the page for IO.
2366 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2368 struct pagevec pvec;
2370 struct inode *inode = mpd->inode;
2371 struct buffer_head *head, *bh;
2372 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2378 start = mpd->map.m_lblk >> bpp_bits;
2379 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2380 lblk = start << bpp_bits;
2381 pblock = mpd->map.m_pblk;
2383 pagevec_init(&pvec);
2384 while (start <= end) {
2385 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2389 for (i = 0; i < nr_pages; i++) {
2390 struct page *page = pvec.pages[i];
2392 bh = head = page_buffers(page);
2394 if (lblk < mpd->map.m_lblk)
2396 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2398 * Buffer after end of mapped extent.
2399 * Find next buffer in the page to map.
2402 mpd->map.m_flags = 0;
2404 * FIXME: If dioread_nolock supports
2405 * blocksize < pagesize, we need to make
2406 * sure we add size mapped so far to
2407 * io_end->size as the following call
2408 * can submit the page for IO.
2410 err = mpage_process_page_bufs(mpd, head,
2412 pagevec_release(&pvec);
2417 if (buffer_delay(bh)) {
2418 clear_buffer_delay(bh);
2419 bh->b_blocknr = pblock++;
2421 clear_buffer_unwritten(bh);
2422 } while (lblk++, (bh = bh->b_this_page) != head);
2425 * FIXME: This is going to break if dioread_nolock
2426 * supports blocksize < pagesize as we will try to
2427 * convert potentially unmapped parts of inode.
2429 mpd->io_submit.io_end->size += PAGE_SIZE;
2430 /* Page fully mapped - let IO run! */
2431 err = mpage_submit_page(mpd, page);
2433 pagevec_release(&pvec);
2437 pagevec_release(&pvec);
2439 /* Extent fully mapped and matches with page boundary. We are done. */
2441 mpd->map.m_flags = 0;
2445 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2447 struct inode *inode = mpd->inode;
2448 struct ext4_map_blocks *map = &mpd->map;
2449 int get_blocks_flags;
2450 int err, dioread_nolock;
2452 trace_ext4_da_write_pages_extent(inode, map);
2454 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2455 * to convert an unwritten extent to be initialized (in the case
2456 * where we have written into one or more preallocated blocks). It is
2457 * possible that we're going to need more metadata blocks than
2458 * previously reserved. However we must not fail because we're in
2459 * writeback and there is nothing we can do about it so it might result
2460 * in data loss. So use reserved blocks to allocate metadata if
2463 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2464 * the blocks in question are delalloc blocks. This indicates
2465 * that the blocks and quotas has already been checked when
2466 * the data was copied into the page cache.
2468 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2469 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2470 EXT4_GET_BLOCKS_IO_SUBMIT;
2471 dioread_nolock = ext4_should_dioread_nolock(inode);
2473 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2474 if (map->m_flags & (1 << BH_Delay))
2475 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2477 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2480 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2481 if (!mpd->io_submit.io_end->handle &&
2482 ext4_handle_valid(handle)) {
2483 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2484 handle->h_rsv_handle = NULL;
2486 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2489 BUG_ON(map->m_len == 0);
2490 if (map->m_flags & EXT4_MAP_NEW) {
2491 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
2498 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2499 * mpd->len and submit pages underlying it for IO
2501 * @handle - handle for journal operations
2502 * @mpd - extent to map
2503 * @give_up_on_write - we set this to true iff there is a fatal error and there
2504 * is no hope of writing the data. The caller should discard
2505 * dirty pages to avoid infinite loops.
2507 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2508 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2509 * them to initialized or split the described range from larger unwritten
2510 * extent. Note that we need not map all the described range since allocation
2511 * can return less blocks or the range is covered by more unwritten extents. We
2512 * cannot map more because we are limited by reserved transaction credits. On
2513 * the other hand we always make sure that the last touched page is fully
2514 * mapped so that it can be written out (and thus forward progress is
2515 * guaranteed). After mapping we submit all mapped pages for IO.
2517 static int mpage_map_and_submit_extent(handle_t *handle,
2518 struct mpage_da_data *mpd,
2519 bool *give_up_on_write)
2521 struct inode *inode = mpd->inode;
2522 struct ext4_map_blocks *map = &mpd->map;
2527 mpd->io_submit.io_end->offset =
2528 ((loff_t)map->m_lblk) << inode->i_blkbits;
2530 err = mpage_map_one_extent(handle, mpd);
2532 struct super_block *sb = inode->i_sb;
2534 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2535 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2536 goto invalidate_dirty_pages;
2538 * Let the uper layers retry transient errors.
2539 * In the case of ENOSPC, if ext4_count_free_blocks()
2540 * is non-zero, a commit should free up blocks.
2542 if ((err == -ENOMEM) ||
2543 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2545 goto update_disksize;
2548 ext4_msg(sb, KERN_CRIT,
2549 "Delayed block allocation failed for "
2550 "inode %lu at logical offset %llu with"
2551 " max blocks %u with error %d",
2553 (unsigned long long)map->m_lblk,
2554 (unsigned)map->m_len, -err);
2555 ext4_msg(sb, KERN_CRIT,
2556 "This should not happen!! Data will "
2559 ext4_print_free_blocks(inode);
2560 invalidate_dirty_pages:
2561 *give_up_on_write = true;
2566 * Update buffer state, submit mapped pages, and get us new
2569 err = mpage_map_and_submit_buffers(mpd);
2571 goto update_disksize;
2572 } while (map->m_len);
2576 * Update on-disk size after IO is submitted. Races with
2577 * truncate are avoided by checking i_size under i_data_sem.
2579 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2580 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2584 down_write(&EXT4_I(inode)->i_data_sem);
2585 i_size = i_size_read(inode);
2586 if (disksize > i_size)
2588 if (disksize > EXT4_I(inode)->i_disksize)
2589 EXT4_I(inode)->i_disksize = disksize;
2590 up_write(&EXT4_I(inode)->i_data_sem);
2591 err2 = ext4_mark_inode_dirty(handle, inode);
2593 ext4_error(inode->i_sb,
2594 "Failed to mark inode %lu dirty",
2603 * Calculate the total number of credits to reserve for one writepages
2604 * iteration. This is called from ext4_writepages(). We map an extent of
2605 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2606 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2607 * bpp - 1 blocks in bpp different extents.
2609 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2611 int bpp = ext4_journal_blocks_per_page(inode);
2613 return ext4_meta_trans_blocks(inode,
2614 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2618 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2619 * and underlying extent to map
2621 * @mpd - where to look for pages
2623 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2624 * IO immediately. When we find a page which isn't mapped we start accumulating
2625 * extent of buffers underlying these pages that needs mapping (formed by
2626 * either delayed or unwritten buffers). We also lock the pages containing
2627 * these buffers. The extent found is returned in @mpd structure (starting at
2628 * mpd->lblk with length mpd->len blocks).
2630 * Note that this function can attach bios to one io_end structure which are
2631 * neither logically nor physically contiguous. Although it may seem as an
2632 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2633 * case as we need to track IO to all buffers underlying a page in one io_end.
2635 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2637 struct address_space *mapping = mpd->inode->i_mapping;
2638 struct pagevec pvec;
2639 unsigned int nr_pages;
2640 long left = mpd->wbc->nr_to_write;
2641 pgoff_t index = mpd->first_page;
2642 pgoff_t end = mpd->last_page;
2645 int blkbits = mpd->inode->i_blkbits;
2647 struct buffer_head *head;
2649 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2650 tag = PAGECACHE_TAG_TOWRITE;
2652 tag = PAGECACHE_TAG_DIRTY;
2654 pagevec_init(&pvec);
2656 mpd->next_page = index;
2657 while (index <= end) {
2658 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2663 for (i = 0; i < nr_pages; i++) {
2664 struct page *page = pvec.pages[i];
2667 * Accumulated enough dirty pages? This doesn't apply
2668 * to WB_SYNC_ALL mode. For integrity sync we have to
2669 * keep going because someone may be concurrently
2670 * dirtying pages, and we might have synced a lot of
2671 * newly appeared dirty pages, but have not synced all
2672 * of the old dirty pages.
2674 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2677 /* If we can't merge this page, we are done. */
2678 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2683 * If the page is no longer dirty, or its mapping no
2684 * longer corresponds to inode we are writing (which
2685 * means it has been truncated or invalidated), or the
2686 * page is already under writeback and we are not doing
2687 * a data integrity writeback, skip the page
2689 if (!PageDirty(page) ||
2690 (PageWriteback(page) &&
2691 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2692 unlikely(page->mapping != mapping)) {
2697 wait_on_page_writeback(page);
2698 BUG_ON(PageWriteback(page));
2700 if (mpd->map.m_len == 0)
2701 mpd->first_page = page->index;
2702 mpd->next_page = page->index + 1;
2703 /* Add all dirty buffers to mpd */
2704 lblk = ((ext4_lblk_t)page->index) <<
2705 (PAGE_SHIFT - blkbits);
2706 head = page_buffers(page);
2707 err = mpage_process_page_bufs(mpd, head, head, lblk);
2713 pagevec_release(&pvec);
2718 pagevec_release(&pvec);
2722 static int ext4_writepages(struct address_space *mapping,
2723 struct writeback_control *wbc)
2725 pgoff_t writeback_index = 0;
2726 long nr_to_write = wbc->nr_to_write;
2727 int range_whole = 0;
2729 handle_t *handle = NULL;
2730 struct mpage_da_data mpd;
2731 struct inode *inode = mapping->host;
2732 int needed_blocks, rsv_blocks = 0, ret = 0;
2733 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2735 struct blk_plug plug;
2736 bool give_up_on_write = false;
2738 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2741 percpu_down_read(&sbi->s_writepages_rwsem);
2742 trace_ext4_writepages(inode, wbc);
2745 * No pages to write? This is mainly a kludge to avoid starting
2746 * a transaction for special inodes like journal inode on last iput()
2747 * because that could violate lock ordering on umount
2749 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2750 goto out_writepages;
2752 if (ext4_should_journal_data(inode)) {
2753 ret = generic_writepages(mapping, wbc);
2754 goto out_writepages;
2758 * If the filesystem has aborted, it is read-only, so return
2759 * right away instead of dumping stack traces later on that
2760 * will obscure the real source of the problem. We test
2761 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2762 * the latter could be true if the filesystem is mounted
2763 * read-only, and in that case, ext4_writepages should
2764 * *never* be called, so if that ever happens, we would want
2767 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2768 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2770 goto out_writepages;
2773 if (ext4_should_dioread_nolock(inode)) {
2775 * We may need to convert up to one extent per block in
2776 * the page and we may dirty the inode.
2778 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2779 PAGE_SIZE >> inode->i_blkbits);
2783 * If we have inline data and arrive here, it means that
2784 * we will soon create the block for the 1st page, so
2785 * we'd better clear the inline data here.
2787 if (ext4_has_inline_data(inode)) {
2788 /* Just inode will be modified... */
2789 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2790 if (IS_ERR(handle)) {
2791 ret = PTR_ERR(handle);
2792 goto out_writepages;
2794 BUG_ON(ext4_test_inode_state(inode,
2795 EXT4_STATE_MAY_INLINE_DATA));
2796 ext4_destroy_inline_data(handle, inode);
2797 ext4_journal_stop(handle);
2800 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2803 if (wbc->range_cyclic) {
2804 writeback_index = mapping->writeback_index;
2805 if (writeback_index)
2807 mpd.first_page = writeback_index;
2810 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2811 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2816 ext4_io_submit_init(&mpd.io_submit, wbc);
2818 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2819 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2821 blk_start_plug(&plug);
2824 * First writeback pages that don't need mapping - we can avoid
2825 * starting a transaction unnecessarily and also avoid being blocked
2826 * in the block layer on device congestion while having transaction
2830 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2831 if (!mpd.io_submit.io_end) {
2835 ret = mpage_prepare_extent_to_map(&mpd);
2836 /* Submit prepared bio */
2837 ext4_io_submit(&mpd.io_submit);
2838 ext4_put_io_end_defer(mpd.io_submit.io_end);
2839 mpd.io_submit.io_end = NULL;
2840 /* Unlock pages we didn't use */
2841 mpage_release_unused_pages(&mpd, false);
2845 while (!done && mpd.first_page <= mpd.last_page) {
2846 /* For each extent of pages we use new io_end */
2847 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2848 if (!mpd.io_submit.io_end) {
2854 * We have two constraints: We find one extent to map and we
2855 * must always write out whole page (makes a difference when
2856 * blocksize < pagesize) so that we don't block on IO when we
2857 * try to write out the rest of the page. Journalled mode is
2858 * not supported by delalloc.
2860 BUG_ON(ext4_should_journal_data(inode));
2861 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2863 /* start a new transaction */
2864 handle = ext4_journal_start_with_reserve(inode,
2865 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2866 if (IS_ERR(handle)) {
2867 ret = PTR_ERR(handle);
2868 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2869 "%ld pages, ino %lu; err %d", __func__,
2870 wbc->nr_to_write, inode->i_ino, ret);
2871 /* Release allocated io_end */
2872 ext4_put_io_end(mpd.io_submit.io_end);
2873 mpd.io_submit.io_end = NULL;
2878 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2879 ret = mpage_prepare_extent_to_map(&mpd);
2882 ret = mpage_map_and_submit_extent(handle, &mpd,
2886 * We scanned the whole range (or exhausted
2887 * nr_to_write), submitted what was mapped and
2888 * didn't find anything needing mapping. We are
2895 * Caution: If the handle is synchronous,
2896 * ext4_journal_stop() can wait for transaction commit
2897 * to finish which may depend on writeback of pages to
2898 * complete or on page lock to be released. In that
2899 * case, we have to wait until after after we have
2900 * submitted all the IO, released page locks we hold,
2901 * and dropped io_end reference (for extent conversion
2902 * to be able to complete) before stopping the handle.
2904 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2905 ext4_journal_stop(handle);
2909 /* Submit prepared bio */
2910 ext4_io_submit(&mpd.io_submit);
2911 /* Unlock pages we didn't use */
2912 mpage_release_unused_pages(&mpd, give_up_on_write);
2914 * Drop our io_end reference we got from init. We have
2915 * to be careful and use deferred io_end finishing if
2916 * we are still holding the transaction as we can
2917 * release the last reference to io_end which may end
2918 * up doing unwritten extent conversion.
2921 ext4_put_io_end_defer(mpd.io_submit.io_end);
2922 ext4_journal_stop(handle);
2924 ext4_put_io_end(mpd.io_submit.io_end);
2925 mpd.io_submit.io_end = NULL;
2927 if (ret == -ENOSPC && sbi->s_journal) {
2929 * Commit the transaction which would
2930 * free blocks released in the transaction
2933 jbd2_journal_force_commit_nested(sbi->s_journal);
2937 /* Fatal error - ENOMEM, EIO... */
2942 blk_finish_plug(&plug);
2943 if (!ret && !cycled && wbc->nr_to_write > 0) {
2945 mpd.last_page = writeback_index - 1;
2951 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2953 * Set the writeback_index so that range_cyclic
2954 * mode will write it back later
2956 mapping->writeback_index = mpd.first_page;
2959 trace_ext4_writepages_result(inode, wbc, ret,
2960 nr_to_write - wbc->nr_to_write);
2961 percpu_up_read(&sbi->s_writepages_rwsem);
2965 static int ext4_dax_writepages(struct address_space *mapping,
2966 struct writeback_control *wbc)
2969 long nr_to_write = wbc->nr_to_write;
2970 struct inode *inode = mapping->host;
2971 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2973 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2976 percpu_down_read(&sbi->s_writepages_rwsem);
2977 trace_ext4_writepages(inode, wbc);
2979 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, wbc);
2980 trace_ext4_writepages_result(inode, wbc, ret,
2981 nr_to_write - wbc->nr_to_write);
2982 percpu_up_read(&sbi->s_writepages_rwsem);
2986 static int ext4_nonda_switch(struct super_block *sb)
2988 s64 free_clusters, dirty_clusters;
2989 struct ext4_sb_info *sbi = EXT4_SB(sb);
2992 * switch to non delalloc mode if we are running low
2993 * on free block. The free block accounting via percpu
2994 * counters can get slightly wrong with percpu_counter_batch getting
2995 * accumulated on each CPU without updating global counters
2996 * Delalloc need an accurate free block accounting. So switch
2997 * to non delalloc when we are near to error range.
3000 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
3002 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
3004 * Start pushing delalloc when 1/2 of free blocks are dirty.
3006 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
3007 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
3009 if (2 * free_clusters < 3 * dirty_clusters ||
3010 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
3012 * free block count is less than 150% of dirty blocks
3013 * or free blocks is less than watermark
3020 /* We always reserve for an inode update; the superblock could be there too */
3021 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
3023 if (likely(ext4_has_feature_large_file(inode->i_sb)))
3026 if (pos + len <= 0x7fffffffULL)
3029 /* We might need to update the superblock to set LARGE_FILE */
3033 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3034 loff_t pos, unsigned len, unsigned flags,
3035 struct page **pagep, void **fsdata)
3037 int ret, retries = 0;
3040 struct inode *inode = mapping->host;
3043 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3046 index = pos >> PAGE_SHIFT;
3048 if (ext4_nonda_switch(inode->i_sb) ||
3049 S_ISLNK(inode->i_mode)) {
3050 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3051 return ext4_write_begin(file, mapping, pos,
3052 len, flags, pagep, fsdata);
3054 *fsdata = (void *)0;
3055 trace_ext4_da_write_begin(inode, pos, len, flags);
3057 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3058 ret = ext4_da_write_inline_data_begin(mapping, inode,
3068 * grab_cache_page_write_begin() can take a long time if the
3069 * system is thrashing due to memory pressure, or if the page
3070 * is being written back. So grab it first before we start
3071 * the transaction handle. This also allows us to allocate
3072 * the page (if needed) without using GFP_NOFS.
3075 page = grab_cache_page_write_begin(mapping, index, flags);
3081 * With delayed allocation, we don't log the i_disksize update
3082 * if there is delayed block allocation. But we still need
3083 * to journalling the i_disksize update if writes to the end
3084 * of file which has an already mapped buffer.
3087 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3088 ext4_da_write_credits(inode, pos, len));
3089 if (IS_ERR(handle)) {
3091 return PTR_ERR(handle);
3095 if (page->mapping != mapping) {
3096 /* The page got truncated from under us */
3099 ext4_journal_stop(handle);
3102 /* In case writeback began while the page was unlocked */
3103 wait_for_stable_page(page);
3105 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3106 ret = ext4_block_write_begin(page, pos, len,
3107 ext4_da_get_block_prep);
3109 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3113 ext4_journal_stop(handle);
3115 * block_write_begin may have instantiated a few blocks
3116 * outside i_size. Trim these off again. Don't need
3117 * i_size_read because we hold i_mutex.
3119 if (pos + len > inode->i_size)
3120 ext4_truncate_failed_write(inode);
3122 if (ret == -ENOSPC &&
3123 ext4_should_retry_alloc(inode->i_sb, &retries))
3135 * Check if we should update i_disksize
3136 * when write to the end of file but not require block allocation
3138 static int ext4_da_should_update_i_disksize(struct page *page,
3139 unsigned long offset)
3141 struct buffer_head *bh;
3142 struct inode *inode = page->mapping->host;
3146 bh = page_buffers(page);
3147 idx = offset >> inode->i_blkbits;
3149 for (i = 0; i < idx; i++)
3150 bh = bh->b_this_page;
3152 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3157 static int ext4_da_write_end(struct file *file,
3158 struct address_space *mapping,
3159 loff_t pos, unsigned len, unsigned copied,
3160 struct page *page, void *fsdata)
3162 struct inode *inode = mapping->host;
3164 handle_t *handle = ext4_journal_current_handle();
3166 unsigned long start, end;
3167 int write_mode = (int)(unsigned long)fsdata;
3169 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3170 return ext4_write_end(file, mapping, pos,
3171 len, copied, page, fsdata);
3173 trace_ext4_da_write_end(inode, pos, len, copied);
3174 start = pos & (PAGE_SIZE - 1);
3175 end = start + copied - 1;
3178 * generic_write_end() will run mark_inode_dirty() if i_size
3179 * changes. So let's piggyback the i_disksize mark_inode_dirty
3182 new_i_size = pos + copied;
3183 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3184 if (ext4_has_inline_data(inode) ||
3185 ext4_da_should_update_i_disksize(page, end)) {
3186 ext4_update_i_disksize(inode, new_i_size);
3187 /* We need to mark inode dirty even if
3188 * new_i_size is less that inode->i_size
3189 * bu greater than i_disksize.(hint delalloc)
3191 ext4_mark_inode_dirty(handle, inode);
3195 if (write_mode != CONVERT_INLINE_DATA &&
3196 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3197 ext4_has_inline_data(inode))
3198 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3201 ret2 = generic_write_end(file, mapping, pos, len, copied,
3207 ret2 = ext4_journal_stop(handle);
3211 return ret ? ret : copied;
3214 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3215 unsigned int length)
3218 * Drop reserved blocks
3220 BUG_ON(!PageLocked(page));
3221 if (!page_has_buffers(page))
3224 ext4_da_page_release_reservation(page, offset, length);
3227 ext4_invalidatepage(page, offset, length);
3233 * Force all delayed allocation blocks to be allocated for a given inode.
3235 int ext4_alloc_da_blocks(struct inode *inode)
3237 trace_ext4_alloc_da_blocks(inode);
3239 if (!EXT4_I(inode)->i_reserved_data_blocks)
3243 * We do something simple for now. The filemap_flush() will
3244 * also start triggering a write of the data blocks, which is
3245 * not strictly speaking necessary (and for users of
3246 * laptop_mode, not even desirable). However, to do otherwise
3247 * would require replicating code paths in:
3249 * ext4_writepages() ->
3250 * write_cache_pages() ---> (via passed in callback function)
3251 * __mpage_da_writepage() -->
3252 * mpage_add_bh_to_extent()
3253 * mpage_da_map_blocks()
3255 * The problem is that write_cache_pages(), located in
3256 * mm/page-writeback.c, marks pages clean in preparation for
3257 * doing I/O, which is not desirable if we're not planning on
3260 * We could call write_cache_pages(), and then redirty all of
3261 * the pages by calling redirty_page_for_writepage() but that
3262 * would be ugly in the extreme. So instead we would need to
3263 * replicate parts of the code in the above functions,
3264 * simplifying them because we wouldn't actually intend to
3265 * write out the pages, but rather only collect contiguous
3266 * logical block extents, call the multi-block allocator, and
3267 * then update the buffer heads with the block allocations.
3269 * For now, though, we'll cheat by calling filemap_flush(),
3270 * which will map the blocks, and start the I/O, but not
3271 * actually wait for the I/O to complete.
3273 return filemap_flush(inode->i_mapping);
3277 * bmap() is special. It gets used by applications such as lilo and by
3278 * the swapper to find the on-disk block of a specific piece of data.
3280 * Naturally, this is dangerous if the block concerned is still in the
3281 * journal. If somebody makes a swapfile on an ext4 data-journaling
3282 * filesystem and enables swap, then they may get a nasty shock when the
3283 * data getting swapped to that swapfile suddenly gets overwritten by
3284 * the original zero's written out previously to the journal and
3285 * awaiting writeback in the kernel's buffer cache.
3287 * So, if we see any bmap calls here on a modified, data-journaled file,
3288 * take extra steps to flush any blocks which might be in the cache.
3290 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3292 struct inode *inode = mapping->host;
3297 * We can get here for an inline file via the FIBMAP ioctl
3299 if (ext4_has_inline_data(inode))
3302 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3303 test_opt(inode->i_sb, DELALLOC)) {
3305 * With delalloc we want to sync the file
3306 * so that we can make sure we allocate
3309 filemap_write_and_wait(mapping);
3312 if (EXT4_JOURNAL(inode) &&
3313 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3315 * This is a REALLY heavyweight approach, but the use of
3316 * bmap on dirty files is expected to be extremely rare:
3317 * only if we run lilo or swapon on a freshly made file
3318 * do we expect this to happen.
3320 * (bmap requires CAP_SYS_RAWIO so this does not
3321 * represent an unprivileged user DOS attack --- we'd be
3322 * in trouble if mortal users could trigger this path at
3325 * NB. EXT4_STATE_JDATA is not set on files other than
3326 * regular files. If somebody wants to bmap a directory
3327 * or symlink and gets confused because the buffer
3328 * hasn't yet been flushed to disk, they deserve
3329 * everything they get.
3332 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3333 journal = EXT4_JOURNAL(inode);
3334 jbd2_journal_lock_updates(journal);
3335 err = jbd2_journal_flush(journal);
3336 jbd2_journal_unlock_updates(journal);
3342 return generic_block_bmap(mapping, block, ext4_get_block);
3345 static int ext4_readpage(struct file *file, struct page *page)
3348 struct inode *inode = page->mapping->host;
3350 trace_ext4_readpage(page);
3352 if (ext4_has_inline_data(inode))
3353 ret = ext4_readpage_inline(inode, page);
3356 return ext4_mpage_readpages(page->mapping, NULL, page, 1,
3363 ext4_readpages(struct file *file, struct address_space *mapping,
3364 struct list_head *pages, unsigned nr_pages)
3366 struct inode *inode = mapping->host;
3368 /* If the file has inline data, no need to do readpages. */
3369 if (ext4_has_inline_data(inode))
3372 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages, true);
3375 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3376 unsigned int length)
3378 trace_ext4_invalidatepage(page, offset, length);
3380 /* No journalling happens on data buffers when this function is used */
3381 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3383 block_invalidatepage(page, offset, length);
3386 static int __ext4_journalled_invalidatepage(struct page *page,
3387 unsigned int offset,
3388 unsigned int length)
3390 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3392 trace_ext4_journalled_invalidatepage(page, offset, length);
3395 * If it's a full truncate we just forget about the pending dirtying
3397 if (offset == 0 && length == PAGE_SIZE)
3398 ClearPageChecked(page);
3400 return jbd2_journal_invalidatepage(journal, page, offset, length);
3403 /* Wrapper for aops... */
3404 static void ext4_journalled_invalidatepage(struct page *page,
3405 unsigned int offset,
3406 unsigned int length)
3408 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3411 static int ext4_releasepage(struct page *page, gfp_t wait)
3413 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3415 trace_ext4_releasepage(page);
3417 /* Page has dirty journalled data -> cannot release */
3418 if (PageChecked(page))
3421 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3423 return try_to_free_buffers(page);
3426 static bool ext4_inode_datasync_dirty(struct inode *inode)
3428 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3431 return !jbd2_transaction_committed(journal,
3432 EXT4_I(inode)->i_datasync_tid);
3433 /* Any metadata buffers to write? */
3434 if (!list_empty(&inode->i_mapping->private_list))
3436 return inode->i_state & I_DIRTY_DATASYNC;
3439 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3440 unsigned flags, struct iomap *iomap)
3442 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3443 unsigned int blkbits = inode->i_blkbits;
3444 unsigned long first_block, last_block;
3445 struct ext4_map_blocks map;
3446 bool delalloc = false;
3449 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3451 first_block = offset >> blkbits;
3452 last_block = min_t(loff_t, (offset + length - 1) >> blkbits,
3453 EXT4_MAX_LOGICAL_BLOCK);
3455 if (flags & IOMAP_REPORT) {
3456 if (ext4_has_inline_data(inode)) {
3457 ret = ext4_inline_data_iomap(inode, iomap);
3458 if (ret != -EAGAIN) {
3459 if (ret == 0 && offset >= iomap->length)
3465 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3469 map.m_lblk = first_block;
3470 map.m_len = last_block - first_block + 1;
3472 if (flags & IOMAP_REPORT) {
3473 ret = ext4_map_blocks(NULL, inode, &map, 0);
3478 ext4_lblk_t end = map.m_lblk + map.m_len - 1;
3479 struct extent_status es;
3481 ext4_es_find_delayed_extent_range(inode, map.m_lblk, end, &es);
3483 if (!es.es_len || es.es_lblk > end) {
3484 /* entire range is a hole */
3485 } else if (es.es_lblk > map.m_lblk) {
3486 /* range starts with a hole */
3487 map.m_len = es.es_lblk - map.m_lblk;
3489 ext4_lblk_t offs = 0;
3491 if (es.es_lblk < map.m_lblk)
3492 offs = map.m_lblk - es.es_lblk;
3493 map.m_lblk = es.es_lblk + offs;
3494 map.m_len = es.es_len - offs;
3498 } else if (flags & IOMAP_WRITE) {
3503 /* Trim mapping request to maximum we can map at once for DIO */
3504 if (map.m_len > DIO_MAX_BLOCKS)
3505 map.m_len = DIO_MAX_BLOCKS;
3506 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3509 * Either we allocate blocks and then we don't get unwritten
3510 * extent so we have reserved enough credits, or the blocks
3511 * are already allocated and unwritten and in that case
3512 * extent conversion fits in the credits as well.
3514 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3517 return PTR_ERR(handle);
3519 ret = ext4_map_blocks(handle, inode, &map,
3520 EXT4_GET_BLOCKS_CREATE_ZERO);
3522 ext4_journal_stop(handle);
3523 if (ret == -ENOSPC &&
3524 ext4_should_retry_alloc(inode->i_sb, &retries))
3530 * If we added blocks beyond i_size, we need to make sure they
3531 * will get truncated if we crash before updating i_size in
3532 * ext4_iomap_end(). For faults we don't need to do that (and
3533 * even cannot because for orphan list operations inode_lock is
3534 * required) - if we happen to instantiate block beyond i_size,
3535 * it is because we race with truncate which has already added
3536 * the inode to the orphan list.
3538 if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3539 (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3542 err = ext4_orphan_add(handle, inode);
3544 ext4_journal_stop(handle);
3548 ext4_journal_stop(handle);
3550 ret = ext4_map_blocks(NULL, inode, &map, 0);
3556 * Writes that span EOF might trigger an I/O size update on completion,
3557 * so consider them to be dirty for the purposes of O_DSYNC, even if
3558 * there is no other metadata changes being made or are pending here.
3561 if (ext4_inode_datasync_dirty(inode) ||
3562 offset + length > i_size_read(inode))
3563 iomap->flags |= IOMAP_F_DIRTY;
3564 iomap->bdev = inode->i_sb->s_bdev;
3565 iomap->dax_dev = sbi->s_daxdev;
3566 iomap->offset = (u64)first_block << blkbits;
3567 iomap->length = (u64)map.m_len << blkbits;
3570 iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE;
3571 iomap->addr = IOMAP_NULL_ADDR;
3573 if (map.m_flags & EXT4_MAP_MAPPED) {
3574 iomap->type = IOMAP_MAPPED;
3575 } else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3576 iomap->type = IOMAP_UNWRITTEN;
3581 iomap->addr = (u64)map.m_pblk << blkbits;
3584 if (map.m_flags & EXT4_MAP_NEW)
3585 iomap->flags |= IOMAP_F_NEW;
3590 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3591 ssize_t written, unsigned flags, struct iomap *iomap)
3595 int blkbits = inode->i_blkbits;
3596 bool truncate = false;
3598 if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3601 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3602 if (IS_ERR(handle)) {
3603 ret = PTR_ERR(handle);
3606 if (ext4_update_inode_size(inode, offset + written))
3607 ext4_mark_inode_dirty(handle, inode);
3609 * We may need to truncate allocated but not written blocks beyond EOF.
3611 if (iomap->offset + iomap->length >
3612 ALIGN(inode->i_size, 1 << blkbits)) {
3613 ext4_lblk_t written_blk, end_blk;
3615 written_blk = (offset + written) >> blkbits;
3616 end_blk = (offset + length) >> blkbits;
3617 if (written_blk < end_blk && ext4_can_truncate(inode))
3621 * Remove inode from orphan list if we were extending a inode and
3622 * everything went fine.
3624 if (!truncate && inode->i_nlink &&
3625 !list_empty(&EXT4_I(inode)->i_orphan))
3626 ext4_orphan_del(handle, inode);
3627 ext4_journal_stop(handle);
3629 ext4_truncate_failed_write(inode);
3632 * If truncate failed early the inode might still be on the
3633 * orphan list; we need to make sure the inode is removed from
3634 * the orphan list in that case.
3637 ext4_orphan_del(NULL, inode);
3642 const struct iomap_ops ext4_iomap_ops = {
3643 .iomap_begin = ext4_iomap_begin,
3644 .iomap_end = ext4_iomap_end,
3647 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3648 ssize_t size, void *private)
3650 ext4_io_end_t *io_end = private;
3652 /* if not async direct IO just return */
3656 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3657 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3658 io_end, io_end->inode->i_ino, iocb, offset, size);
3661 * Error during AIO DIO. We cannot convert unwritten extents as the
3662 * data was not written. Just clear the unwritten flag and drop io_end.
3665 ext4_clear_io_unwritten_flag(io_end);
3668 io_end->offset = offset;
3669 io_end->size = size;
3670 ext4_put_io_end(io_end);
3676 * Handling of direct IO writes.
3678 * For ext4 extent files, ext4 will do direct-io write even to holes,
3679 * preallocated extents, and those write extend the file, no need to
3680 * fall back to buffered IO.
3682 * For holes, we fallocate those blocks, mark them as unwritten
3683 * If those blocks were preallocated, we mark sure they are split, but
3684 * still keep the range to write as unwritten.
3686 * The unwritten extents will be converted to written when DIO is completed.
3687 * For async direct IO, since the IO may still pending when return, we
3688 * set up an end_io call back function, which will do the conversion
3689 * when async direct IO completed.
3691 * If the O_DIRECT write will extend the file then add this inode to the
3692 * orphan list. So recovery will truncate it back to the original size
3693 * if the machine crashes during the write.
3696 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3698 struct file *file = iocb->ki_filp;
3699 struct inode *inode = file->f_mapping->host;
3700 struct ext4_inode_info *ei = EXT4_I(inode);
3702 loff_t offset = iocb->ki_pos;
3703 size_t count = iov_iter_count(iter);
3705 get_block_t *get_block_func = NULL;
3707 loff_t final_size = offset + count;
3711 if (final_size > inode->i_size || final_size > ei->i_disksize) {
3712 /* Credits for sb + inode write */
3713 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3714 if (IS_ERR(handle)) {
3715 ret = PTR_ERR(handle);
3718 ret = ext4_orphan_add(handle, inode);
3720 ext4_journal_stop(handle);
3724 ext4_update_i_disksize(inode, inode->i_size);
3725 ext4_journal_stop(handle);
3728 BUG_ON(iocb->private == NULL);
3731 * Make all waiters for direct IO properly wait also for extent
3732 * conversion. This also disallows race between truncate() and
3733 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3735 inode_dio_begin(inode);
3737 /* If we do a overwrite dio, i_mutex locking can be released */
3738 overwrite = *((int *)iocb->private);
3741 inode_unlock(inode);
3744 * For extent mapped files we could direct write to holes and fallocate.
3746 * Allocated blocks to fill the hole are marked as unwritten to prevent
3747 * parallel buffered read to expose the stale data before DIO complete
3750 * As to previously fallocated extents, ext4 get_block will just simply
3751 * mark the buffer mapped but still keep the extents unwritten.
3753 * For non AIO case, we will convert those unwritten extents to written
3754 * after return back from blockdev_direct_IO. That way we save us from
3755 * allocating io_end structure and also the overhead of offloading
3756 * the extent convertion to a workqueue.
3758 * For async DIO, the conversion needs to be deferred when the
3759 * IO is completed. The ext4 end_io callback function will be
3760 * called to take care of the conversion work. Here for async
3761 * case, we allocate an io_end structure to hook to the iocb.
3763 iocb->private = NULL;
3765 get_block_func = ext4_dio_get_block_overwrite;
3766 else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3767 round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3768 get_block_func = ext4_dio_get_block;
3769 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3770 } else if (is_sync_kiocb(iocb)) {
3771 get_block_func = ext4_dio_get_block_unwritten_sync;
3772 dio_flags = DIO_LOCKING;
3774 get_block_func = ext4_dio_get_block_unwritten_async;
3775 dio_flags = DIO_LOCKING;
3777 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3778 get_block_func, ext4_end_io_dio, NULL,
3781 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3782 EXT4_STATE_DIO_UNWRITTEN)) {
3785 * for non AIO case, since the IO is already
3786 * completed, we could do the conversion right here
3788 err = ext4_convert_unwritten_extents(NULL, inode,
3792 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3795 inode_dio_end(inode);
3796 /* take i_mutex locking again if we do a ovewrite dio */
3800 if (ret < 0 && final_size > inode->i_size)
3801 ext4_truncate_failed_write(inode);
3803 /* Handle extending of i_size after direct IO write */
3807 /* Credits for sb + inode write */
3808 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3809 if (IS_ERR(handle)) {
3811 * We wrote the data but cannot extend
3812 * i_size. Bail out. In async io case, we do
3813 * not return error here because we have
3814 * already submmitted the corresponding
3815 * bio. Returning error here makes the caller
3816 * think that this IO is done and failed
3817 * resulting in race with bio's completion
3821 ret = PTR_ERR(handle);
3823 ext4_orphan_del(NULL, inode);
3828 ext4_orphan_del(handle, inode);
3830 loff_t end = offset + ret;
3831 if (end > inode->i_size || end > ei->i_disksize) {
3832 ext4_update_i_disksize(inode, end);
3833 if (end > inode->i_size)
3834 i_size_write(inode, end);
3836 * We're going to return a positive `ret'
3837 * here due to non-zero-length I/O, so there's
3838 * no way of reporting error returns from
3839 * ext4_mark_inode_dirty() to userspace. So
3842 ext4_mark_inode_dirty(handle, inode);
3845 err = ext4_journal_stop(handle);
3853 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3855 struct address_space *mapping = iocb->ki_filp->f_mapping;
3856 struct inode *inode = mapping->host;
3857 size_t count = iov_iter_count(iter);
3859 loff_t offset = iocb->ki_pos;
3860 loff_t size = i_size_read(inode);
3866 * Shared inode_lock is enough for us - it protects against concurrent
3867 * writes & truncates and since we take care of writing back page cache,
3868 * we are protected against page writeback as well.
3870 if (iocb->ki_flags & IOCB_NOWAIT) {
3871 if (!inode_trylock_shared(inode))
3874 inode_lock_shared(inode);
3877 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3878 iocb->ki_pos + count - 1);
3881 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3882 iter, ext4_dio_get_block, NULL, NULL, 0);
3884 inode_unlock_shared(inode);
3888 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3890 struct file *file = iocb->ki_filp;
3891 struct inode *inode = file->f_mapping->host;
3892 size_t count = iov_iter_count(iter);
3893 loff_t offset = iocb->ki_pos;
3896 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3897 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3902 * If we are doing data journalling we don't support O_DIRECT
3904 if (ext4_should_journal_data(inode))
3907 /* Let buffer I/O handle the inline data case. */
3908 if (ext4_has_inline_data(inode))
3911 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3912 if (iov_iter_rw(iter) == READ)
3913 ret = ext4_direct_IO_read(iocb, iter);
3915 ret = ext4_direct_IO_write(iocb, iter);
3916 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3921 * Pages can be marked dirty completely asynchronously from ext4's journalling
3922 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3923 * much here because ->set_page_dirty is called under VFS locks. The page is
3924 * not necessarily locked.
3926 * We cannot just dirty the page and leave attached buffers clean, because the
3927 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3928 * or jbddirty because all the journalling code will explode.
3930 * So what we do is to mark the page "pending dirty" and next time writepage
3931 * is called, propagate that into the buffers appropriately.
3933 static int ext4_journalled_set_page_dirty(struct page *page)
3935 SetPageChecked(page);
3936 return __set_page_dirty_nobuffers(page);
3939 static int ext4_set_page_dirty(struct page *page)
3941 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3942 WARN_ON_ONCE(!page_has_buffers(page));
3943 return __set_page_dirty_buffers(page);
3946 static const struct address_space_operations ext4_aops = {
3947 .readpage = ext4_readpage,
3948 .readpages = ext4_readpages,
3949 .writepage = ext4_writepage,
3950 .writepages = ext4_writepages,
3951 .write_begin = ext4_write_begin,
3952 .write_end = ext4_write_end,
3953 .set_page_dirty = ext4_set_page_dirty,
3955 .invalidatepage = ext4_invalidatepage,
3956 .releasepage = ext4_releasepage,
3957 .direct_IO = ext4_direct_IO,
3958 .migratepage = buffer_migrate_page,
3959 .is_partially_uptodate = block_is_partially_uptodate,
3960 .error_remove_page = generic_error_remove_page,
3963 static const struct address_space_operations ext4_journalled_aops = {
3964 .readpage = ext4_readpage,
3965 .readpages = ext4_readpages,
3966 .writepage = ext4_writepage,
3967 .writepages = ext4_writepages,
3968 .write_begin = ext4_write_begin,
3969 .write_end = ext4_journalled_write_end,
3970 .set_page_dirty = ext4_journalled_set_page_dirty,
3972 .invalidatepage = ext4_journalled_invalidatepage,
3973 .releasepage = ext4_releasepage,
3974 .direct_IO = ext4_direct_IO,
3975 .is_partially_uptodate = block_is_partially_uptodate,
3976 .error_remove_page = generic_error_remove_page,
3979 static const struct address_space_operations ext4_da_aops = {
3980 .readpage = ext4_readpage,
3981 .readpages = ext4_readpages,
3982 .writepage = ext4_writepage,
3983 .writepages = ext4_writepages,
3984 .write_begin = ext4_da_write_begin,
3985 .write_end = ext4_da_write_end,
3986 .set_page_dirty = ext4_set_page_dirty,
3988 .invalidatepage = ext4_da_invalidatepage,
3989 .releasepage = ext4_releasepage,
3990 .direct_IO = ext4_direct_IO,
3991 .migratepage = buffer_migrate_page,
3992 .is_partially_uptodate = block_is_partially_uptodate,
3993 .error_remove_page = generic_error_remove_page,
3996 static const struct address_space_operations ext4_dax_aops = {
3997 .writepages = ext4_dax_writepages,
3998 .direct_IO = noop_direct_IO,
3999 .set_page_dirty = noop_set_page_dirty,
4001 .invalidatepage = noop_invalidatepage,
4004 void ext4_set_aops(struct inode *inode)
4006 switch (ext4_inode_journal_mode(inode)) {
4007 case EXT4_INODE_ORDERED_DATA_MODE:
4008 case EXT4_INODE_WRITEBACK_DATA_MODE:
4010 case EXT4_INODE_JOURNAL_DATA_MODE:
4011 inode->i_mapping->a_ops = &ext4_journalled_aops;
4017 inode->i_mapping->a_ops = &ext4_dax_aops;
4018 else if (test_opt(inode->i_sb, DELALLOC))
4019 inode->i_mapping->a_ops = &ext4_da_aops;
4021 inode->i_mapping->a_ops = &ext4_aops;
4024 static int __ext4_block_zero_page_range(handle_t *handle,
4025 struct address_space *mapping, loff_t from, loff_t length)
4027 ext4_fsblk_t index = from >> PAGE_SHIFT;
4028 unsigned offset = from & (PAGE_SIZE-1);
4029 unsigned blocksize, pos;
4031 struct inode *inode = mapping->host;
4032 struct buffer_head *bh;
4036 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
4037 mapping_gfp_constraint(mapping, ~__GFP_FS));
4041 blocksize = inode->i_sb->s_blocksize;
4043 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
4045 if (!page_has_buffers(page))
4046 create_empty_buffers(page, blocksize, 0);
4048 /* Find the buffer that contains "offset" */
4049 bh = page_buffers(page);
4051 while (offset >= pos) {
4052 bh = bh->b_this_page;
4056 if (buffer_freed(bh)) {
4057 BUFFER_TRACE(bh, "freed: skip");
4060 if (!buffer_mapped(bh)) {
4061 BUFFER_TRACE(bh, "unmapped");
4062 ext4_get_block(inode, iblock, bh, 0);
4063 /* unmapped? It's a hole - nothing to do */
4064 if (!buffer_mapped(bh)) {
4065 BUFFER_TRACE(bh, "still unmapped");
4070 /* Ok, it's mapped. Make sure it's up-to-date */
4071 if (PageUptodate(page))
4072 set_buffer_uptodate(bh);
4074 if (!buffer_uptodate(bh)) {
4076 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
4078 /* Uhhuh. Read error. Complain and punt. */
4079 if (!buffer_uptodate(bh))
4081 if (S_ISREG(inode->i_mode) &&
4082 ext4_encrypted_inode(inode)) {
4083 /* We expect the key to be set. */
4084 BUG_ON(!fscrypt_has_encryption_key(inode));
4085 BUG_ON(blocksize != PAGE_SIZE);
4086 WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host,
4087 page, PAGE_SIZE, 0, page->index));
4090 if (ext4_should_journal_data(inode)) {
4091 BUFFER_TRACE(bh, "get write access");
4092 err = ext4_journal_get_write_access(handle, bh);
4096 zero_user(page, offset, length);
4097 BUFFER_TRACE(bh, "zeroed end of block");
4099 if (ext4_should_journal_data(inode)) {
4100 err = ext4_handle_dirty_metadata(handle, inode, bh);
4103 mark_buffer_dirty(bh);
4104 if (ext4_should_order_data(inode))
4105 err = ext4_jbd2_inode_add_write(handle, inode, from,
4116 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4117 * starting from file offset 'from'. The range to be zero'd must
4118 * be contained with in one block. If the specified range exceeds
4119 * the end of the block it will be shortened to end of the block
4120 * that cooresponds to 'from'
4122 static int ext4_block_zero_page_range(handle_t *handle,
4123 struct address_space *mapping, loff_t from, loff_t length)
4125 struct inode *inode = mapping->host;
4126 unsigned offset = from & (PAGE_SIZE-1);
4127 unsigned blocksize = inode->i_sb->s_blocksize;
4128 unsigned max = blocksize - (offset & (blocksize - 1));
4131 * correct length if it does not fall between
4132 * 'from' and the end of the block
4134 if (length > max || length < 0)
4137 if (IS_DAX(inode)) {
4138 return iomap_zero_range(inode, from, length, NULL,
4141 return __ext4_block_zero_page_range(handle, mapping, from, length);
4145 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4146 * up to the end of the block which corresponds to `from'.
4147 * This required during truncate. We need to physically zero the tail end
4148 * of that block so it doesn't yield old data if the file is later grown.
4150 static int ext4_block_truncate_page(handle_t *handle,
4151 struct address_space *mapping, loff_t from)
4153 unsigned offset = from & (PAGE_SIZE-1);
4156 struct inode *inode = mapping->host;
4158 /* If we are processing an encrypted inode during orphan list handling */
4159 if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode))
4162 blocksize = inode->i_sb->s_blocksize;
4163 length = blocksize - (offset & (blocksize - 1));
4165 return ext4_block_zero_page_range(handle, mapping, from, length);
4168 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4169 loff_t lstart, loff_t length)
4171 struct super_block *sb = inode->i_sb;
4172 struct address_space *mapping = inode->i_mapping;
4173 unsigned partial_start, partial_end;
4174 ext4_fsblk_t start, end;
4175 loff_t byte_end = (lstart + length - 1);
4178 partial_start = lstart & (sb->s_blocksize - 1);
4179 partial_end = byte_end & (sb->s_blocksize - 1);
4181 start = lstart >> sb->s_blocksize_bits;
4182 end = byte_end >> sb->s_blocksize_bits;
4184 /* Handle partial zero within the single block */
4186 (partial_start || (partial_end != sb->s_blocksize - 1))) {
4187 err = ext4_block_zero_page_range(handle, mapping,
4191 /* Handle partial zero out on the start of the range */
4192 if (partial_start) {
4193 err = ext4_block_zero_page_range(handle, mapping,
4194 lstart, sb->s_blocksize);
4198 /* Handle partial zero out on the end of the range */
4199 if (partial_end != sb->s_blocksize - 1)
4200 err = ext4_block_zero_page_range(handle, mapping,
4201 byte_end - partial_end,
4206 int ext4_can_truncate(struct inode *inode)
4208 if (S_ISREG(inode->i_mode))
4210 if (S_ISDIR(inode->i_mode))
4212 if (S_ISLNK(inode->i_mode))
4213 return !ext4_inode_is_fast_symlink(inode);
4218 * We have to make sure i_disksize gets properly updated before we truncate
4219 * page cache due to hole punching or zero range. Otherwise i_disksize update
4220 * can get lost as it may have been postponed to submission of writeback but
4221 * that will never happen after we truncate page cache.
4223 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4227 loff_t size = i_size_read(inode);
4229 WARN_ON(!inode_is_locked(inode));
4230 if (offset > size || offset + len < size)
4233 if (EXT4_I(inode)->i_disksize >= size)
4236 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4238 return PTR_ERR(handle);
4239 ext4_update_i_disksize(inode, size);
4240 ext4_mark_inode_dirty(handle, inode);
4241 ext4_journal_stop(handle);
4246 static void ext4_wait_dax_page(struct ext4_inode_info *ei)
4248 up_write(&ei->i_mmap_sem);
4250 down_write(&ei->i_mmap_sem);
4253 int ext4_break_layouts(struct inode *inode)
4255 struct ext4_inode_info *ei = EXT4_I(inode);
4259 if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
4263 page = dax_layout_busy_page(inode->i_mapping);
4267 error = ___wait_var_event(&page->_refcount,
4268 atomic_read(&page->_refcount) == 1,
4269 TASK_INTERRUPTIBLE, 0, 0,
4270 ext4_wait_dax_page(ei));
4271 } while (error == 0);
4277 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4278 * associated with the given offset and length
4280 * @inode: File inode
4281 * @offset: The offset where the hole will begin
4282 * @len: The length of the hole
4284 * Returns: 0 on success or negative on failure
4287 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4289 struct super_block *sb = inode->i_sb;
4290 ext4_lblk_t first_block, stop_block;
4291 struct address_space *mapping = inode->i_mapping;
4292 loff_t first_block_offset, last_block_offset;
4294 unsigned int credits;
4297 if (!S_ISREG(inode->i_mode))
4300 trace_ext4_punch_hole(inode, offset, length, 0);
4302 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4303 if (ext4_has_inline_data(inode)) {
4304 down_write(&EXT4_I(inode)->i_mmap_sem);
4305 ret = ext4_convert_inline_data(inode);
4306 up_write(&EXT4_I(inode)->i_mmap_sem);
4312 * Write out all dirty pages to avoid race conditions
4313 * Then release them.
4315 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4316 ret = filemap_write_and_wait_range(mapping, offset,
4317 offset + length - 1);
4324 /* No need to punch hole beyond i_size */
4325 if (offset >= inode->i_size)
4329 * If the hole extends beyond i_size, set the hole
4330 * to end after the page that contains i_size
4332 if (offset + length > inode->i_size) {
4333 length = inode->i_size +
4334 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4338 if (offset & (sb->s_blocksize - 1) ||
4339 (offset + length) & (sb->s_blocksize - 1)) {
4341 * Attach jinode to inode for jbd2 if we do any zeroing of
4344 ret = ext4_inode_attach_jinode(inode);
4350 /* Wait all existing dio workers, newcomers will block on i_mutex */
4351 inode_dio_wait(inode);
4354 * Prevent page faults from reinstantiating pages we have released from
4357 down_write(&EXT4_I(inode)->i_mmap_sem);
4359 ret = ext4_break_layouts(inode);
4363 first_block_offset = round_up(offset, sb->s_blocksize);
4364 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4366 /* Now release the pages and zero block aligned part of pages*/
4367 if (last_block_offset > first_block_offset) {
4368 ret = ext4_update_disksize_before_punch(inode, offset, length);
4371 truncate_pagecache_range(inode, first_block_offset,
4375 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4376 credits = ext4_writepage_trans_blocks(inode);
4378 credits = ext4_blocks_for_truncate(inode);
4379 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4380 if (IS_ERR(handle)) {
4381 ret = PTR_ERR(handle);
4382 ext4_std_error(sb, ret);
4386 ret = ext4_zero_partial_blocks(handle, inode, offset,
4391 first_block = (offset + sb->s_blocksize - 1) >>
4392 EXT4_BLOCK_SIZE_BITS(sb);
4393 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4395 /* If there are blocks to remove, do it */
4396 if (stop_block > first_block) {
4398 down_write(&EXT4_I(inode)->i_data_sem);
4399 ext4_discard_preallocations(inode);
4401 ret = ext4_es_remove_extent(inode, first_block,
4402 stop_block - first_block);
4404 up_write(&EXT4_I(inode)->i_data_sem);
4408 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4409 ret = ext4_ext_remove_space(inode, first_block,
4412 ret = ext4_ind_remove_space(handle, inode, first_block,
4415 up_write(&EXT4_I(inode)->i_data_sem);
4418 ext4_handle_sync(handle);
4420 inode->i_mtime = inode->i_ctime = current_time(inode);
4421 ext4_mark_inode_dirty(handle, inode);
4423 ext4_update_inode_fsync_trans(handle, inode, 1);
4425 ext4_journal_stop(handle);
4427 up_write(&EXT4_I(inode)->i_mmap_sem);
4429 inode_unlock(inode);
4433 int ext4_inode_attach_jinode(struct inode *inode)
4435 struct ext4_inode_info *ei = EXT4_I(inode);
4436 struct jbd2_inode *jinode;
4438 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4441 jinode = jbd2_alloc_inode(GFP_KERNEL);
4442 spin_lock(&inode->i_lock);
4445 spin_unlock(&inode->i_lock);
4448 ei->jinode = jinode;
4449 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4452 spin_unlock(&inode->i_lock);
4453 if (unlikely(jinode != NULL))
4454 jbd2_free_inode(jinode);
4461 * We block out ext4_get_block() block instantiations across the entire
4462 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4463 * simultaneously on behalf of the same inode.
4465 * As we work through the truncate and commit bits of it to the journal there
4466 * is one core, guiding principle: the file's tree must always be consistent on
4467 * disk. We must be able to restart the truncate after a crash.
4469 * The file's tree may be transiently inconsistent in memory (although it
4470 * probably isn't), but whenever we close off and commit a journal transaction,
4471 * the contents of (the filesystem + the journal) must be consistent and
4472 * restartable. It's pretty simple, really: bottom up, right to left (although
4473 * left-to-right works OK too).
4475 * Note that at recovery time, journal replay occurs *before* the restart of
4476 * truncate against the orphan inode list.
4478 * The committed inode has the new, desired i_size (which is the same as
4479 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4480 * that this inode's truncate did not complete and it will again call
4481 * ext4_truncate() to have another go. So there will be instantiated blocks
4482 * to the right of the truncation point in a crashed ext4 filesystem. But
4483 * that's fine - as long as they are linked from the inode, the post-crash
4484 * ext4_truncate() run will find them and release them.
4486 int ext4_truncate(struct inode *inode)
4488 struct ext4_inode_info *ei = EXT4_I(inode);
4489 unsigned int credits;
4492 struct address_space *mapping = inode->i_mapping;
4495 * There is a possibility that we're either freeing the inode
4496 * or it's a completely new inode. In those cases we might not
4497 * have i_mutex locked because it's not necessary.
4499 if (!(inode->i_state & (I_NEW|I_FREEING)))
4500 WARN_ON(!inode_is_locked(inode));
4501 trace_ext4_truncate_enter(inode);
4503 if (!ext4_can_truncate(inode))
4506 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4508 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4509 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4511 if (ext4_has_inline_data(inode)) {
4514 err = ext4_inline_data_truncate(inode, &has_inline);
4521 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4522 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4523 if (ext4_inode_attach_jinode(inode) < 0)
4527 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4528 credits = ext4_writepage_trans_blocks(inode);
4530 credits = ext4_blocks_for_truncate(inode);
4532 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4534 return PTR_ERR(handle);
4536 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4537 ext4_block_truncate_page(handle, mapping, inode->i_size);
4540 * We add the inode to the orphan list, so that if this
4541 * truncate spans multiple transactions, and we crash, we will
4542 * resume the truncate when the filesystem recovers. It also
4543 * marks the inode dirty, to catch the new size.
4545 * Implication: the file must always be in a sane, consistent
4546 * truncatable state while each transaction commits.
4548 err = ext4_orphan_add(handle, inode);
4552 down_write(&EXT4_I(inode)->i_data_sem);
4554 ext4_discard_preallocations(inode);
4556 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4557 err = ext4_ext_truncate(handle, inode);
4559 ext4_ind_truncate(handle, inode);
4561 up_write(&ei->i_data_sem);
4566 ext4_handle_sync(handle);
4570 * If this was a simple ftruncate() and the file will remain alive,
4571 * then we need to clear up the orphan record which we created above.
4572 * However, if this was a real unlink then we were called by
4573 * ext4_evict_inode(), and we allow that function to clean up the
4574 * orphan info for us.
4577 ext4_orphan_del(handle, inode);
4579 inode->i_mtime = inode->i_ctime = current_time(inode);
4580 ext4_mark_inode_dirty(handle, inode);
4581 ext4_journal_stop(handle);
4583 trace_ext4_truncate_exit(inode);
4588 * ext4_get_inode_loc returns with an extra refcount against the inode's
4589 * underlying buffer_head on success. If 'in_mem' is true, we have all
4590 * data in memory that is needed to recreate the on-disk version of this
4593 static int __ext4_get_inode_loc(struct inode *inode,
4594 struct ext4_iloc *iloc, int in_mem)
4596 struct ext4_group_desc *gdp;
4597 struct buffer_head *bh;
4598 struct super_block *sb = inode->i_sb;
4600 int inodes_per_block, inode_offset;
4603 if (inode->i_ino < EXT4_ROOT_INO ||
4604 inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4605 return -EFSCORRUPTED;
4607 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4608 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4613 * Figure out the offset within the block group inode table
4615 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4616 inode_offset = ((inode->i_ino - 1) %
4617 EXT4_INODES_PER_GROUP(sb));
4618 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4619 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4621 bh = sb_getblk(sb, block);
4624 if (!buffer_uptodate(bh)) {
4628 * If the buffer has the write error flag, we have failed
4629 * to write out another inode in the same block. In this
4630 * case, we don't have to read the block because we may
4631 * read the old inode data successfully.
4633 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4634 set_buffer_uptodate(bh);
4636 if (buffer_uptodate(bh)) {
4637 /* someone brought it uptodate while we waited */
4643 * If we have all information of the inode in memory and this
4644 * is the only valid inode in the block, we need not read the
4648 struct buffer_head *bitmap_bh;
4651 start = inode_offset & ~(inodes_per_block - 1);
4653 /* Is the inode bitmap in cache? */
4654 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4655 if (unlikely(!bitmap_bh))
4659 * If the inode bitmap isn't in cache then the
4660 * optimisation may end up performing two reads instead
4661 * of one, so skip it.
4663 if (!buffer_uptodate(bitmap_bh)) {
4667 for (i = start; i < start + inodes_per_block; i++) {
4668 if (i == inode_offset)
4670 if (ext4_test_bit(i, bitmap_bh->b_data))
4674 if (i == start + inodes_per_block) {
4675 /* all other inodes are free, so skip I/O */
4676 memset(bh->b_data, 0, bh->b_size);
4677 set_buffer_uptodate(bh);
4685 * If we need to do any I/O, try to pre-readahead extra
4686 * blocks from the inode table.
4688 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4689 ext4_fsblk_t b, end, table;
4691 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4693 table = ext4_inode_table(sb, gdp);
4694 /* s_inode_readahead_blks is always a power of 2 */
4695 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4699 num = EXT4_INODES_PER_GROUP(sb);
4700 if (ext4_has_group_desc_csum(sb))
4701 num -= ext4_itable_unused_count(sb, gdp);
4702 table += num / inodes_per_block;
4706 sb_breadahead_unmovable(sb, b++);
4710 * There are other valid inodes in the buffer, this inode
4711 * has in-inode xattrs, or we don't have this inode in memory.
4712 * Read the block from disk.
4714 trace_ext4_load_inode(inode);
4716 bh->b_end_io = end_buffer_read_sync;
4717 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4719 if (!buffer_uptodate(bh)) {
4720 EXT4_ERROR_INODE_BLOCK(inode, block,
4721 "unable to read itable block");
4731 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4733 /* We have all inode data except xattrs in memory here. */
4734 return __ext4_get_inode_loc(inode, iloc,
4735 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4738 static bool ext4_should_use_dax(struct inode *inode)
4740 if (!test_opt(inode->i_sb, DAX))
4742 if (!S_ISREG(inode->i_mode))
4744 if (ext4_should_journal_data(inode))
4746 if (ext4_has_inline_data(inode))
4748 if (ext4_encrypted_inode(inode))
4753 void ext4_set_inode_flags(struct inode *inode)
4755 unsigned int flags = EXT4_I(inode)->i_flags;
4756 unsigned int new_fl = 0;
4758 if (flags & EXT4_SYNC_FL)
4760 if (flags & EXT4_APPEND_FL)
4762 if (flags & EXT4_IMMUTABLE_FL)
4763 new_fl |= S_IMMUTABLE;
4764 if (flags & EXT4_NOATIME_FL)
4765 new_fl |= S_NOATIME;
4766 if (flags & EXT4_DIRSYNC_FL)
4767 new_fl |= S_DIRSYNC;
4768 if (ext4_should_use_dax(inode))
4770 if (flags & EXT4_ENCRYPT_FL)
4771 new_fl |= S_ENCRYPTED;
4772 inode_set_flags(inode, new_fl,
4773 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4777 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4778 struct ext4_inode_info *ei)
4781 struct inode *inode = &(ei->vfs_inode);
4782 struct super_block *sb = inode->i_sb;
4784 if (ext4_has_feature_huge_file(sb)) {
4785 /* we are using combined 48 bit field */
4786 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4787 le32_to_cpu(raw_inode->i_blocks_lo);
4788 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4789 /* i_blocks represent file system block size */
4790 return i_blocks << (inode->i_blkbits - 9);
4795 return le32_to_cpu(raw_inode->i_blocks_lo);
4799 static inline int ext4_iget_extra_inode(struct inode *inode,
4800 struct ext4_inode *raw_inode,
4801 struct ext4_inode_info *ei)
4803 __le32 *magic = (void *)raw_inode +
4804 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4806 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4807 EXT4_INODE_SIZE(inode->i_sb) &&
4808 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4809 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4810 return ext4_find_inline_data_nolock(inode);
4812 EXT4_I(inode)->i_inline_off = 0;
4816 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4818 if (!ext4_has_feature_project(inode->i_sb))
4820 *projid = EXT4_I(inode)->i_projid;
4825 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4826 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4829 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4831 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4832 inode_set_iversion_raw(inode, val);
4834 inode_set_iversion_queried(inode, val);
4836 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4838 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4839 return inode_peek_iversion_raw(inode);
4841 return inode_peek_iversion(inode);
4844 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4845 ext4_iget_flags flags, const char *function,
4848 struct ext4_iloc iloc;
4849 struct ext4_inode *raw_inode;
4850 struct ext4_inode_info *ei;
4851 struct inode *inode;
4852 journal_t *journal = EXT4_SB(sb)->s_journal;
4860 if ((!(flags & EXT4_IGET_SPECIAL) &&
4861 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4862 (ino < EXT4_ROOT_INO) ||
4863 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4864 if (flags & EXT4_IGET_HANDLE)
4865 return ERR_PTR(-ESTALE);
4866 __ext4_error(sb, function, line,
4867 "inode #%lu: comm %s: iget: illegal inode #",
4868 ino, current->comm);
4869 return ERR_PTR(-EFSCORRUPTED);
4872 inode = iget_locked(sb, ino);
4874 return ERR_PTR(-ENOMEM);
4875 if (!(inode->i_state & I_NEW))
4881 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4884 raw_inode = ext4_raw_inode(&iloc);
4886 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4887 ext4_error_inode(inode, function, line, 0,
4888 "iget: root inode unallocated");
4889 ret = -EFSCORRUPTED;
4893 if ((flags & EXT4_IGET_HANDLE) &&
4894 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4899 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4900 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4901 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4902 EXT4_INODE_SIZE(inode->i_sb) ||
4903 (ei->i_extra_isize & 3)) {
4904 ext4_error_inode(inode, function, line, 0,
4905 "iget: bad extra_isize %u "
4908 EXT4_INODE_SIZE(inode->i_sb));
4909 ret = -EFSCORRUPTED;
4913 ei->i_extra_isize = 0;
4915 /* Precompute checksum seed for inode metadata */
4916 if (ext4_has_metadata_csum(sb)) {
4917 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4919 __le32 inum = cpu_to_le32(inode->i_ino);
4920 __le32 gen = raw_inode->i_generation;
4921 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4923 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4927 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4928 ext4_error_inode(inode, function, line, 0,
4929 "iget: checksum invalid");
4934 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4935 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4936 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4937 if (ext4_has_feature_project(sb) &&
4938 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4939 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4940 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4942 i_projid = EXT4_DEF_PROJID;
4944 if (!(test_opt(inode->i_sb, NO_UID32))) {
4945 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4946 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4948 i_uid_write(inode, i_uid);
4949 i_gid_write(inode, i_gid);
4950 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4951 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4953 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4954 ei->i_inline_off = 0;
4955 ei->i_dir_start_lookup = 0;
4956 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4957 /* We now have enough fields to check if the inode was active or not.
4958 * This is needed because nfsd might try to access dead inodes
4959 * the test is that same one that e2fsck uses
4960 * NeilBrown 1999oct15
4962 if (inode->i_nlink == 0) {
4963 if ((inode->i_mode == 0 ||
4964 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4965 ino != EXT4_BOOT_LOADER_INO) {
4966 /* this inode is deleted */
4970 /* The only unlinked inodes we let through here have
4971 * valid i_mode and are being read by the orphan
4972 * recovery code: that's fine, we're about to complete
4973 * the process of deleting those.
4974 * OR it is the EXT4_BOOT_LOADER_INO which is
4975 * not initialized on a new filesystem. */
4977 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4978 ext4_set_inode_flags(inode);
4979 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4980 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4981 if (ext4_has_feature_64bit(sb))
4983 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4984 inode->i_size = ext4_isize(sb, raw_inode);
4985 if ((size = i_size_read(inode)) < 0) {
4986 ext4_error_inode(inode, function, line, 0,
4987 "iget: bad i_size value: %lld", size);
4988 ret = -EFSCORRUPTED;
4992 * If dir_index is not enabled but there's dir with INDEX flag set,
4993 * we'd normally treat htree data as empty space. But with metadata
4994 * checksumming that corrupts checksums so forbid that.
4996 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4997 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4998 ext4_error_inode(inode, function, line, 0,
4999 "iget: Dir with htree data on filesystem without dir_index feature.");
5000 ret = -EFSCORRUPTED;
5003 ei->i_disksize = inode->i_size;
5005 ei->i_reserved_quota = 0;
5007 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
5008 ei->i_block_group = iloc.block_group;
5009 ei->i_last_alloc_group = ~0;
5011 * NOTE! The in-memory inode i_data array is in little-endian order
5012 * even on big-endian machines: we do NOT byteswap the block numbers!
5014 for (block = 0; block < EXT4_N_BLOCKS; block++)
5015 ei->i_data[block] = raw_inode->i_block[block];
5016 INIT_LIST_HEAD(&ei->i_orphan);
5019 * Set transaction id's of transactions that have to be committed
5020 * to finish f[data]sync. We set them to currently running transaction
5021 * as we cannot be sure that the inode or some of its metadata isn't
5022 * part of the transaction - the inode could have been reclaimed and
5023 * now it is reread from disk.
5026 transaction_t *transaction;
5029 read_lock(&journal->j_state_lock);
5030 if (journal->j_running_transaction)
5031 transaction = journal->j_running_transaction;
5033 transaction = journal->j_committing_transaction;
5035 tid = transaction->t_tid;
5037 tid = journal->j_commit_sequence;
5038 read_unlock(&journal->j_state_lock);
5039 ei->i_sync_tid = tid;
5040 ei->i_datasync_tid = tid;
5043 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5044 if (ei->i_extra_isize == 0) {
5045 /* The extra space is currently unused. Use it. */
5046 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
5047 ei->i_extra_isize = sizeof(struct ext4_inode) -
5048 EXT4_GOOD_OLD_INODE_SIZE;
5050 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
5056 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5057 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5058 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5059 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5061 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5062 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
5064 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5065 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5067 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5069 ext4_inode_set_iversion_queried(inode, ivers);
5073 if (ei->i_file_acl &&
5074 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
5075 ext4_error_inode(inode, function, line, 0,
5076 "iget: bad extended attribute block %llu",
5078 ret = -EFSCORRUPTED;
5080 } else if (!ext4_has_inline_data(inode)) {
5081 /* validate the block references in the inode */
5082 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5083 (S_ISLNK(inode->i_mode) &&
5084 !ext4_inode_is_fast_symlink(inode))) {
5085 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5086 ret = ext4_ext_check_inode(inode);
5088 ret = ext4_ind_check_inode(inode);
5094 if (S_ISREG(inode->i_mode)) {
5095 inode->i_op = &ext4_file_inode_operations;
5096 inode->i_fop = &ext4_file_operations;
5097 ext4_set_aops(inode);
5098 } else if (S_ISDIR(inode->i_mode)) {
5099 inode->i_op = &ext4_dir_inode_operations;
5100 inode->i_fop = &ext4_dir_operations;
5101 } else if (S_ISLNK(inode->i_mode)) {
5102 /* VFS does not allow setting these so must be corruption */
5103 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5104 ext4_error_inode(inode, function, line, 0,
5105 "iget: immutable or append flags "
5106 "not allowed on symlinks");
5107 ret = -EFSCORRUPTED;
5110 if (ext4_encrypted_inode(inode)) {
5111 inode->i_op = &ext4_encrypted_symlink_inode_operations;
5112 ext4_set_aops(inode);
5113 } else if (ext4_inode_is_fast_symlink(inode)) {
5114 inode->i_link = (char *)ei->i_data;
5115 inode->i_op = &ext4_fast_symlink_inode_operations;
5116 nd_terminate_link(ei->i_data, inode->i_size,
5117 sizeof(ei->i_data) - 1);
5119 inode->i_op = &ext4_symlink_inode_operations;
5120 ext4_set_aops(inode);
5122 inode_nohighmem(inode);
5123 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5124 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5125 inode->i_op = &ext4_special_inode_operations;
5126 if (raw_inode->i_block[0])
5127 init_special_inode(inode, inode->i_mode,
5128 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5130 init_special_inode(inode, inode->i_mode,
5131 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5132 } else if (ino == EXT4_BOOT_LOADER_INO) {
5133 make_bad_inode(inode);
5135 ret = -EFSCORRUPTED;
5136 ext4_error_inode(inode, function, line, 0,
5137 "iget: bogus i_mode (%o)", inode->i_mode);
5142 unlock_new_inode(inode);
5148 return ERR_PTR(ret);
5151 static int ext4_inode_blocks_set(handle_t *handle,
5152 struct ext4_inode *raw_inode,
5153 struct ext4_inode_info *ei)
5155 struct inode *inode = &(ei->vfs_inode);
5156 u64 i_blocks = READ_ONCE(inode->i_blocks);
5157 struct super_block *sb = inode->i_sb;
5159 if (i_blocks <= ~0U) {
5161 * i_blocks can be represented in a 32 bit variable
5162 * as multiple of 512 bytes
5164 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5165 raw_inode->i_blocks_high = 0;
5166 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5169 if (!ext4_has_feature_huge_file(sb))
5172 if (i_blocks <= 0xffffffffffffULL) {
5174 * i_blocks can be represented in a 48 bit variable
5175 * as multiple of 512 bytes
5177 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5178 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5179 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5181 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5182 /* i_block is stored in file system block size */
5183 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5184 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5185 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5190 struct other_inode {
5191 unsigned long orig_ino;
5192 struct ext4_inode *raw_inode;
5195 static int other_inode_match(struct inode * inode, unsigned long ino,
5198 struct other_inode *oi = (struct other_inode *) data;
5200 if ((inode->i_ino != ino) ||
5201 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5203 ((inode->i_state & I_DIRTY_TIME) == 0))
5205 spin_lock(&inode->i_lock);
5206 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5207 I_DIRTY_INODE)) == 0) &&
5208 (inode->i_state & I_DIRTY_TIME)) {
5209 struct ext4_inode_info *ei = EXT4_I(inode);
5211 inode->i_state &= ~I_DIRTY_TIME;
5212 spin_unlock(&inode->i_lock);
5214 spin_lock(&ei->i_raw_lock);
5215 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
5216 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
5217 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
5218 ext4_inode_csum_set(inode, oi->raw_inode, ei);
5219 spin_unlock(&ei->i_raw_lock);
5220 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
5223 spin_unlock(&inode->i_lock);
5228 * Opportunistically update the other time fields for other inodes in
5229 * the same inode table block.
5231 static void ext4_update_other_inodes_time(struct super_block *sb,
5232 unsigned long orig_ino, char *buf)
5234 struct other_inode oi;
5236 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5237 int inode_size = EXT4_INODE_SIZE(sb);
5239 oi.orig_ino = orig_ino;
5241 * Calculate the first inode in the inode table block. Inode
5242 * numbers are one-based. That is, the first inode in a block
5243 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5245 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5246 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5247 if (ino == orig_ino)
5249 oi.raw_inode = (struct ext4_inode *) buf;
5250 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
5255 * Post the struct inode info into an on-disk inode location in the
5256 * buffer-cache. This gobbles the caller's reference to the
5257 * buffer_head in the inode location struct.
5259 * The caller must have write access to iloc->bh.
5261 static int ext4_do_update_inode(handle_t *handle,
5262 struct inode *inode,
5263 struct ext4_iloc *iloc)
5265 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5266 struct ext4_inode_info *ei = EXT4_I(inode);
5267 struct buffer_head *bh = iloc->bh;
5268 struct super_block *sb = inode->i_sb;
5270 int need_datasync = 0, set_large_file = 0;
5275 spin_lock(&ei->i_raw_lock);
5277 /* For fields not tracked in the in-memory inode,
5278 * initialise them to zero for new inodes. */
5279 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5280 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5282 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5284 spin_unlock(&ei->i_raw_lock);
5288 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5289 i_uid = i_uid_read(inode);
5290 i_gid = i_gid_read(inode);
5291 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5292 if (!(test_opt(inode->i_sb, NO_UID32))) {
5293 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5294 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5296 * Fix up interoperability with old kernels. Otherwise, old inodes get
5297 * re-used with the upper 16 bits of the uid/gid intact
5299 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5300 raw_inode->i_uid_high = 0;
5301 raw_inode->i_gid_high = 0;
5303 raw_inode->i_uid_high =
5304 cpu_to_le16(high_16_bits(i_uid));
5305 raw_inode->i_gid_high =
5306 cpu_to_le16(high_16_bits(i_gid));
5309 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5310 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5311 raw_inode->i_uid_high = 0;
5312 raw_inode->i_gid_high = 0;
5314 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5316 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5317 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5318 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5319 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5321 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5322 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5323 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5324 raw_inode->i_file_acl_high =
5325 cpu_to_le16(ei->i_file_acl >> 32);
5326 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5327 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) {
5328 ext4_isize_set(raw_inode, ei->i_disksize);
5331 if (ei->i_disksize > 0x7fffffffULL) {
5332 if (!ext4_has_feature_large_file(sb) ||
5333 EXT4_SB(sb)->s_es->s_rev_level ==
5334 cpu_to_le32(EXT4_GOOD_OLD_REV))
5337 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5338 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5339 if (old_valid_dev(inode->i_rdev)) {
5340 raw_inode->i_block[0] =
5341 cpu_to_le32(old_encode_dev(inode->i_rdev));
5342 raw_inode->i_block[1] = 0;
5344 raw_inode->i_block[0] = 0;
5345 raw_inode->i_block[1] =
5346 cpu_to_le32(new_encode_dev(inode->i_rdev));
5347 raw_inode->i_block[2] = 0;
5349 } else if (!ext4_has_inline_data(inode)) {
5350 for (block = 0; block < EXT4_N_BLOCKS; block++)
5351 raw_inode->i_block[block] = ei->i_data[block];
5354 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5355 u64 ivers = ext4_inode_peek_iversion(inode);
5357 raw_inode->i_disk_version = cpu_to_le32(ivers);
5358 if (ei->i_extra_isize) {
5359 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5360 raw_inode->i_version_hi =
5361 cpu_to_le32(ivers >> 32);
5362 raw_inode->i_extra_isize =
5363 cpu_to_le16(ei->i_extra_isize);
5367 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5368 i_projid != EXT4_DEF_PROJID);
5370 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5371 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5372 raw_inode->i_projid = cpu_to_le32(i_projid);
5374 ext4_inode_csum_set(inode, raw_inode, ei);
5375 spin_unlock(&ei->i_raw_lock);
5376 if (inode->i_sb->s_flags & SB_LAZYTIME)
5377 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5380 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5381 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5384 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5385 if (set_large_file) {
5386 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5387 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5390 ext4_set_feature_large_file(sb);
5391 ext4_handle_sync(handle);
5392 err = ext4_handle_dirty_super(handle, sb);
5394 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5397 ext4_std_error(inode->i_sb, err);
5402 * ext4_write_inode()
5404 * We are called from a few places:
5406 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5407 * Here, there will be no transaction running. We wait for any running
5408 * transaction to commit.
5410 * - Within flush work (sys_sync(), kupdate and such).
5411 * We wait on commit, if told to.
5413 * - Within iput_final() -> write_inode_now()
5414 * We wait on commit, if told to.
5416 * In all cases it is actually safe for us to return without doing anything,
5417 * because the inode has been copied into a raw inode buffer in
5418 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5421 * Note that we are absolutely dependent upon all inode dirtiers doing the
5422 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5423 * which we are interested.
5425 * It would be a bug for them to not do this. The code:
5427 * mark_inode_dirty(inode)
5429 * inode->i_size = expr;
5431 * is in error because write_inode() could occur while `stuff()' is running,
5432 * and the new i_size will be lost. Plus the inode will no longer be on the
5433 * superblock's dirty inode list.
5435 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5439 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5440 sb_rdonly(inode->i_sb))
5443 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5446 if (EXT4_SB(inode->i_sb)->s_journal) {
5447 if (ext4_journal_current_handle()) {
5448 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5454 * No need to force transaction in WB_SYNC_NONE mode. Also
5455 * ext4_sync_fs() will force the commit after everything is
5458 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5461 err = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5462 EXT4_I(inode)->i_sync_tid);
5464 struct ext4_iloc iloc;
5466 err = __ext4_get_inode_loc(inode, &iloc, 0);
5470 * sync(2) will flush the whole buffer cache. No need to do
5471 * it here separately for each inode.
5473 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5474 sync_dirty_buffer(iloc.bh);
5475 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5476 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5477 "IO error syncing inode");
5486 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5487 * buffers that are attached to a page stradding i_size and are undergoing
5488 * commit. In that case we have to wait for commit to finish and try again.
5490 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5494 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5495 tid_t commit_tid = 0;
5498 offset = inode->i_size & (PAGE_SIZE - 1);
5500 * If the page is fully truncated, we don't need to wait for any commit
5501 * (and we even should not as __ext4_journalled_invalidatepage() may
5502 * strip all buffers from the page but keep the page dirty which can then
5503 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5504 * buffers). Also we don't need to wait for any commit if all buffers in
5505 * the page remain valid. This is most beneficial for the common case of
5506 * blocksize == PAGESIZE.
5508 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5511 page = find_lock_page(inode->i_mapping,
5512 inode->i_size >> PAGE_SHIFT);
5515 ret = __ext4_journalled_invalidatepage(page, offset,
5516 PAGE_SIZE - offset);
5522 read_lock(&journal->j_state_lock);
5523 if (journal->j_committing_transaction)
5524 commit_tid = journal->j_committing_transaction->t_tid;
5525 read_unlock(&journal->j_state_lock);
5527 jbd2_log_wait_commit(journal, commit_tid);
5534 * Called from notify_change.
5536 * We want to trap VFS attempts to truncate the file as soon as
5537 * possible. In particular, we want to make sure that when the VFS
5538 * shrinks i_size, we put the inode on the orphan list and modify
5539 * i_disksize immediately, so that during the subsequent flushing of
5540 * dirty pages and freeing of disk blocks, we can guarantee that any
5541 * commit will leave the blocks being flushed in an unused state on
5542 * disk. (On recovery, the inode will get truncated and the blocks will
5543 * be freed, so we have a strong guarantee that no future commit will
5544 * leave these blocks visible to the user.)
5546 * Another thing we have to assure is that if we are in ordered mode
5547 * and inode is still attached to the committing transaction, we must
5548 * we start writeout of all the dirty pages which are being truncated.
5549 * This way we are sure that all the data written in the previous
5550 * transaction are already on disk (truncate waits for pages under
5553 * Called with inode->i_mutex down.
5555 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5557 struct inode *inode = d_inode(dentry);
5560 const unsigned int ia_valid = attr->ia_valid;
5562 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5565 if (unlikely(IS_IMMUTABLE(inode)))
5568 if (unlikely(IS_APPEND(inode) &&
5569 (ia_valid & (ATTR_MODE | ATTR_UID |
5570 ATTR_GID | ATTR_TIMES_SET))))
5573 error = setattr_prepare(dentry, attr);
5577 error = fscrypt_prepare_setattr(dentry, attr);
5581 if (is_quota_modification(inode, attr)) {
5582 error = dquot_initialize(inode);
5586 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5587 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5590 /* (user+group)*(old+new) structure, inode write (sb,
5591 * inode block, ? - but truncate inode update has it) */
5592 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5593 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5594 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5595 if (IS_ERR(handle)) {
5596 error = PTR_ERR(handle);
5600 /* dquot_transfer() calls back ext4_get_inode_usage() which
5601 * counts xattr inode references.
5603 down_read(&EXT4_I(inode)->xattr_sem);
5604 error = dquot_transfer(inode, attr);
5605 up_read(&EXT4_I(inode)->xattr_sem);
5608 ext4_journal_stop(handle);
5611 /* Update corresponding info in inode so that everything is in
5612 * one transaction */
5613 if (attr->ia_valid & ATTR_UID)
5614 inode->i_uid = attr->ia_uid;
5615 if (attr->ia_valid & ATTR_GID)
5616 inode->i_gid = attr->ia_gid;
5617 error = ext4_mark_inode_dirty(handle, inode);
5618 ext4_journal_stop(handle);
5621 if (attr->ia_valid & ATTR_SIZE) {
5623 loff_t oldsize = inode->i_size;
5624 int shrink = (attr->ia_size <= inode->i_size);
5626 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5627 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5629 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5632 if (!S_ISREG(inode->i_mode))
5635 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5636 inode_inc_iversion(inode);
5638 if (ext4_should_order_data(inode) &&
5639 (attr->ia_size < inode->i_size)) {
5640 error = ext4_begin_ordered_truncate(inode,
5645 if (attr->ia_size != inode->i_size) {
5646 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5647 if (IS_ERR(handle)) {
5648 error = PTR_ERR(handle);
5651 if (ext4_handle_valid(handle) && shrink) {
5652 error = ext4_orphan_add(handle, inode);
5656 * Update c/mtime on truncate up, ext4_truncate() will
5657 * update c/mtime in shrink case below
5660 inode->i_mtime = current_time(inode);
5661 inode->i_ctime = inode->i_mtime;
5663 down_write(&EXT4_I(inode)->i_data_sem);
5664 EXT4_I(inode)->i_disksize = attr->ia_size;
5665 rc = ext4_mark_inode_dirty(handle, inode);
5669 * We have to update i_size under i_data_sem together
5670 * with i_disksize to avoid races with writeback code
5671 * running ext4_wb_update_i_disksize().
5674 i_size_write(inode, attr->ia_size);
5675 up_write(&EXT4_I(inode)->i_data_sem);
5676 ext4_journal_stop(handle);
5678 if (orphan && inode->i_nlink)
5679 ext4_orphan_del(NULL, inode);
5684 pagecache_isize_extended(inode, oldsize, inode->i_size);
5687 * Blocks are going to be removed from the inode. Wait
5688 * for dio in flight.
5690 inode_dio_wait(inode);
5692 if (orphan && ext4_should_journal_data(inode))
5693 ext4_wait_for_tail_page_commit(inode);
5694 down_write(&EXT4_I(inode)->i_mmap_sem);
5696 rc = ext4_break_layouts(inode);
5698 up_write(&EXT4_I(inode)->i_mmap_sem);
5704 * Truncate pagecache after we've waited for commit
5705 * in data=journal mode to make pages freeable.
5707 truncate_pagecache(inode, inode->i_size);
5709 rc = ext4_truncate(inode);
5713 up_write(&EXT4_I(inode)->i_mmap_sem);
5717 setattr_copy(inode, attr);
5718 mark_inode_dirty(inode);
5722 * If the call to ext4_truncate failed to get a transaction handle at
5723 * all, we need to clean up the in-core orphan list manually.
5725 if (orphan && inode->i_nlink)
5726 ext4_orphan_del(NULL, inode);
5728 if (!error && (ia_valid & ATTR_MODE))
5729 rc = posix_acl_chmod(inode, inode->i_mode);
5732 ext4_std_error(inode->i_sb, error);
5738 int ext4_getattr(const struct path *path, struct kstat *stat,
5739 u32 request_mask, unsigned int query_flags)
5741 struct inode *inode = d_inode(path->dentry);
5742 struct ext4_inode *raw_inode;
5743 struct ext4_inode_info *ei = EXT4_I(inode);
5746 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5747 stat->result_mask |= STATX_BTIME;
5748 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5749 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5752 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5753 if (flags & EXT4_APPEND_FL)
5754 stat->attributes |= STATX_ATTR_APPEND;
5755 if (flags & EXT4_COMPR_FL)
5756 stat->attributes |= STATX_ATTR_COMPRESSED;
5757 if (flags & EXT4_ENCRYPT_FL)
5758 stat->attributes |= STATX_ATTR_ENCRYPTED;
5759 if (flags & EXT4_IMMUTABLE_FL)
5760 stat->attributes |= STATX_ATTR_IMMUTABLE;
5761 if (flags & EXT4_NODUMP_FL)
5762 stat->attributes |= STATX_ATTR_NODUMP;
5764 stat->attributes_mask |= (STATX_ATTR_APPEND |
5765 STATX_ATTR_COMPRESSED |
5766 STATX_ATTR_ENCRYPTED |
5767 STATX_ATTR_IMMUTABLE |
5770 generic_fillattr(inode, stat);
5774 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5775 u32 request_mask, unsigned int query_flags)
5777 struct inode *inode = d_inode(path->dentry);
5778 u64 delalloc_blocks;
5780 ext4_getattr(path, stat, request_mask, query_flags);
5783 * If there is inline data in the inode, the inode will normally not
5784 * have data blocks allocated (it may have an external xattr block).
5785 * Report at least one sector for such files, so tools like tar, rsync,
5786 * others don't incorrectly think the file is completely sparse.
5788 if (unlikely(ext4_has_inline_data(inode)))
5789 stat->blocks += (stat->size + 511) >> 9;
5792 * We can't update i_blocks if the block allocation is delayed
5793 * otherwise in the case of system crash before the real block
5794 * allocation is done, we will have i_blocks inconsistent with
5795 * on-disk file blocks.
5796 * We always keep i_blocks updated together with real
5797 * allocation. But to not confuse with user, stat
5798 * will return the blocks that include the delayed allocation
5799 * blocks for this file.
5801 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5802 EXT4_I(inode)->i_reserved_data_blocks);
5803 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5807 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5810 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5811 return ext4_ind_trans_blocks(inode, lblocks);
5812 return ext4_ext_index_trans_blocks(inode, pextents);
5816 * Account for index blocks, block groups bitmaps and block group
5817 * descriptor blocks if modify datablocks and index blocks
5818 * worse case, the indexs blocks spread over different block groups
5820 * If datablocks are discontiguous, they are possible to spread over
5821 * different block groups too. If they are contiguous, with flexbg,
5822 * they could still across block group boundary.
5824 * Also account for superblock, inode, quota and xattr blocks
5826 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5829 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5835 * How many index blocks need to touch to map @lblocks logical blocks
5836 * to @pextents physical extents?
5838 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5843 * Now let's see how many group bitmaps and group descriptors need
5846 groups = idxblocks + pextents;
5848 if (groups > ngroups)
5850 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5851 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5853 /* bitmaps and block group descriptor blocks */
5854 ret += groups + gdpblocks;
5856 /* Blocks for super block, inode, quota and xattr blocks */
5857 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5863 * Calculate the total number of credits to reserve to fit
5864 * the modification of a single pages into a single transaction,
5865 * which may include multiple chunks of block allocations.
5867 * This could be called via ext4_write_begin()
5869 * We need to consider the worse case, when
5870 * one new block per extent.
5872 int ext4_writepage_trans_blocks(struct inode *inode)
5874 int bpp = ext4_journal_blocks_per_page(inode);
5877 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5879 /* Account for data blocks for journalled mode */
5880 if (ext4_should_journal_data(inode))
5886 * Calculate the journal credits for a chunk of data modification.
5888 * This is called from DIO, fallocate or whoever calling
5889 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5891 * journal buffers for data blocks are not included here, as DIO
5892 * and fallocate do no need to journal data buffers.
5894 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5896 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5900 * The caller must have previously called ext4_reserve_inode_write().
5901 * Give this, we know that the caller already has write access to iloc->bh.
5903 int ext4_mark_iloc_dirty(handle_t *handle,
5904 struct inode *inode, struct ext4_iloc *iloc)
5908 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5912 if (IS_I_VERSION(inode))
5913 inode_inc_iversion(inode);
5915 /* the do_update_inode consumes one bh->b_count */
5918 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5919 err = ext4_do_update_inode(handle, inode, iloc);
5925 * On success, We end up with an outstanding reference count against
5926 * iloc->bh. This _must_ be cleaned up later.
5930 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5931 struct ext4_iloc *iloc)
5935 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5938 err = ext4_get_inode_loc(inode, iloc);
5940 BUFFER_TRACE(iloc->bh, "get_write_access");
5941 err = ext4_journal_get_write_access(handle, iloc->bh);
5947 ext4_std_error(inode->i_sb, err);
5951 static int __ext4_expand_extra_isize(struct inode *inode,
5952 unsigned int new_extra_isize,
5953 struct ext4_iloc *iloc,
5954 handle_t *handle, int *no_expand)
5956 struct ext4_inode *raw_inode;
5957 struct ext4_xattr_ibody_header *header;
5958 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5959 struct ext4_inode_info *ei = EXT4_I(inode);
5962 /* this was checked at iget time, but double check for good measure */
5963 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5964 (ei->i_extra_isize & 3)) {
5965 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5967 EXT4_INODE_SIZE(inode->i_sb));
5968 return -EFSCORRUPTED;
5970 if ((new_extra_isize < ei->i_extra_isize) ||
5971 (new_extra_isize < 4) ||
5972 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5973 return -EINVAL; /* Should never happen */
5975 raw_inode = ext4_raw_inode(iloc);
5977 header = IHDR(inode, raw_inode);
5979 /* No extended attributes present */
5980 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5981 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5982 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5983 EXT4_I(inode)->i_extra_isize, 0,
5984 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5985 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5989 /* try to expand with EAs present */
5990 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5994 * Inode size expansion failed; don't try again
6003 * Expand an inode by new_extra_isize bytes.
6004 * Returns 0 on success or negative error number on failure.
6006 static int ext4_try_to_expand_extra_isize(struct inode *inode,
6007 unsigned int new_extra_isize,
6008 struct ext4_iloc iloc,
6014 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
6018 * In nojournal mode, we can immediately attempt to expand
6019 * the inode. When journaled, we first need to obtain extra
6020 * buffer credits since we may write into the EA block
6021 * with this same handle. If journal_extend fails, then it will
6022 * only result in a minor loss of functionality for that inode.
6023 * If this is felt to be critical, then e2fsck should be run to
6024 * force a large enough s_min_extra_isize.
6026 if (ext4_handle_valid(handle) &&
6027 jbd2_journal_extend(handle,
6028 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0)
6031 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
6034 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
6035 handle, &no_expand);
6036 ext4_write_unlock_xattr(inode, &no_expand);
6041 int ext4_expand_extra_isize(struct inode *inode,
6042 unsigned int new_extra_isize,
6043 struct ext4_iloc *iloc)
6049 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
6054 handle = ext4_journal_start(inode, EXT4_HT_INODE,
6055 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
6056 if (IS_ERR(handle)) {
6057 error = PTR_ERR(handle);
6062 ext4_write_lock_xattr(inode, &no_expand);
6064 BUFFER_TRACE(iloc->bh, "get_write_access");
6065 error = ext4_journal_get_write_access(handle, iloc->bh);
6071 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
6072 handle, &no_expand);
6074 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
6079 ext4_write_unlock_xattr(inode, &no_expand);
6080 ext4_journal_stop(handle);
6085 * What we do here is to mark the in-core inode as clean with respect to inode
6086 * dirtiness (it may still be data-dirty).
6087 * This means that the in-core inode may be reaped by prune_icache
6088 * without having to perform any I/O. This is a very good thing,
6089 * because *any* task may call prune_icache - even ones which
6090 * have a transaction open against a different journal.
6092 * Is this cheating? Not really. Sure, we haven't written the
6093 * inode out, but prune_icache isn't a user-visible syncing function.
6094 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6095 * we start and wait on commits.
6097 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
6099 struct ext4_iloc iloc;
6100 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6104 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6105 err = ext4_reserve_inode_write(handle, inode, &iloc);
6109 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6110 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6113 return ext4_mark_iloc_dirty(handle, inode, &iloc);
6117 * ext4_dirty_inode() is called from __mark_inode_dirty()
6119 * We're really interested in the case where a file is being extended.
6120 * i_size has been changed by generic_commit_write() and we thus need
6121 * to include the updated inode in the current transaction.
6123 * Also, dquot_alloc_block() will always dirty the inode when blocks
6124 * are allocated to the file.
6126 * If the inode is marked synchronous, we don't honour that here - doing
6127 * so would cause a commit on atime updates, which we don't bother doing.
6128 * We handle synchronous inodes at the highest possible level.
6130 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
6131 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
6132 * to copy into the on-disk inode structure are the timestamp files.
6134 void ext4_dirty_inode(struct inode *inode, int flags)
6138 if (flags == I_DIRTY_TIME)
6140 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6144 ext4_mark_inode_dirty(handle, inode);
6146 ext4_journal_stop(handle);
6153 * Bind an inode's backing buffer_head into this transaction, to prevent
6154 * it from being flushed to disk early. Unlike
6155 * ext4_reserve_inode_write, this leaves behind no bh reference and
6156 * returns no iloc structure, so the caller needs to repeat the iloc
6157 * lookup to mark the inode dirty later.
6159 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
6161 struct ext4_iloc iloc;
6165 err = ext4_get_inode_loc(inode, &iloc);
6167 BUFFER_TRACE(iloc.bh, "get_write_access");
6168 err = jbd2_journal_get_write_access(handle, iloc.bh);
6170 err = ext4_handle_dirty_metadata(handle,
6176 ext4_std_error(inode->i_sb, err);
6181 int ext4_change_inode_journal_flag(struct inode *inode, int val)
6186 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6189 * We have to be very careful here: changing a data block's
6190 * journaling status dynamically is dangerous. If we write a
6191 * data block to the journal, change the status and then delete
6192 * that block, we risk forgetting to revoke the old log record
6193 * from the journal and so a subsequent replay can corrupt data.
6194 * So, first we make sure that the journal is empty and that
6195 * nobody is changing anything.
6198 journal = EXT4_JOURNAL(inode);
6201 if (is_journal_aborted(journal))
6204 /* Wait for all existing dio workers */
6205 inode_dio_wait(inode);
6208 * Before flushing the journal and switching inode's aops, we have
6209 * to flush all dirty data the inode has. There can be outstanding
6210 * delayed allocations, there can be unwritten extents created by
6211 * fallocate or buffered writes in dioread_nolock mode covered by
6212 * dirty data which can be converted only after flushing the dirty
6213 * data (and journalled aops don't know how to handle these cases).
6216 down_write(&EXT4_I(inode)->i_mmap_sem);
6217 err = filemap_write_and_wait(inode->i_mapping);
6219 up_write(&EXT4_I(inode)->i_mmap_sem);
6224 percpu_down_write(&sbi->s_writepages_rwsem);
6225 jbd2_journal_lock_updates(journal);
6228 * OK, there are no updates running now, and all cached data is
6229 * synced to disk. We are now in a completely consistent state
6230 * which doesn't have anything in the journal, and we know that
6231 * no filesystem updates are running, so it is safe to modify
6232 * the inode's in-core data-journaling state flag now.
6236 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6238 err = jbd2_journal_flush(journal);
6240 jbd2_journal_unlock_updates(journal);
6241 percpu_up_write(&sbi->s_writepages_rwsem);
6244 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6246 ext4_set_aops(inode);
6248 jbd2_journal_unlock_updates(journal);
6249 percpu_up_write(&sbi->s_writepages_rwsem);
6252 up_write(&EXT4_I(inode)->i_mmap_sem);
6254 /* Finally we can mark the inode as dirty. */
6256 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6258 return PTR_ERR(handle);
6260 err = ext4_mark_inode_dirty(handle, inode);
6261 ext4_handle_sync(handle);
6262 ext4_journal_stop(handle);
6263 ext4_std_error(inode->i_sb, err);
6268 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6270 return !buffer_mapped(bh);
6273 int ext4_page_mkwrite(struct vm_fault *vmf)
6275 struct vm_area_struct *vma = vmf->vma;
6276 struct page *page = vmf->page;
6280 struct file *file = vma->vm_file;
6281 struct inode *inode = file_inode(file);
6282 struct address_space *mapping = inode->i_mapping;
6284 get_block_t *get_block;
6287 if (unlikely(IS_IMMUTABLE(inode)))
6288 return VM_FAULT_SIGBUS;
6290 sb_start_pagefault(inode->i_sb);
6291 file_update_time(vma->vm_file);
6293 down_read(&EXT4_I(inode)->i_mmap_sem);
6295 ret = ext4_convert_inline_data(inode);
6299 /* Delalloc case is easy... */
6300 if (test_opt(inode->i_sb, DELALLOC) &&
6301 !ext4_should_journal_data(inode) &&
6302 !ext4_nonda_switch(inode->i_sb)) {
6304 ret = block_page_mkwrite(vma, vmf,
6305 ext4_da_get_block_prep);
6306 } while (ret == -ENOSPC &&
6307 ext4_should_retry_alloc(inode->i_sb, &retries));
6312 size = i_size_read(inode);
6313 /* Page got truncated from under us? */
6314 if (page->mapping != mapping || page_offset(page) > size) {
6316 ret = VM_FAULT_NOPAGE;
6320 if (page->index == size >> PAGE_SHIFT)
6321 len = size & ~PAGE_MASK;
6325 * Return if we have all the buffers mapped. This avoids the need to do
6326 * journal_start/journal_stop which can block and take a long time
6328 if (page_has_buffers(page)) {
6329 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6331 ext4_bh_unmapped)) {
6332 /* Wait so that we don't change page under IO */
6333 wait_for_stable_page(page);
6334 ret = VM_FAULT_LOCKED;
6339 /* OK, we need to fill the hole... */
6340 if (ext4_should_dioread_nolock(inode))
6341 get_block = ext4_get_block_unwritten;
6343 get_block = ext4_get_block;
6345 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6346 ext4_writepage_trans_blocks(inode));
6347 if (IS_ERR(handle)) {
6348 ret = VM_FAULT_SIGBUS;
6351 ret = block_page_mkwrite(vma, vmf, get_block);
6352 if (!ret && ext4_should_journal_data(inode)) {
6353 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
6354 PAGE_SIZE, NULL, do_journal_get_write_access)) {
6356 ret = VM_FAULT_SIGBUS;
6357 ext4_journal_stop(handle);
6360 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6362 ext4_journal_stop(handle);
6363 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6366 ret = block_page_mkwrite_return(ret);
6368 up_read(&EXT4_I(inode)->i_mmap_sem);
6369 sb_end_pagefault(inode->i_sb);
6373 int ext4_filemap_fault(struct vm_fault *vmf)
6375 struct inode *inode = file_inode(vmf->vma->vm_file);
6378 down_read(&EXT4_I(inode)->i_mmap_sem);
6379 err = filemap_fault(vmf);
6380 up_read(&EXT4_I(inode)->i_mmap_sem);