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/mount.h>
24 #include <linux/time.h>
25 #include <linux/highuid.h>
26 #include <linux/pagemap.h>
27 #include <linux/dax.h>
28 #include <linux/quotaops.h>
29 #include <linux/string.h>
30 #include <linux/buffer_head.h>
31 #include <linux/writeback.h>
32 #include <linux/pagevec.h>
33 #include <linux/mpage.h>
34 #include <linux/namei.h>
35 #include <linux/uio.h>
36 #include <linux/bio.h>
37 #include <linux/workqueue.h>
38 #include <linux/kernel.h>
39 #include <linux/printk.h>
40 #include <linux/slab.h>
41 #include <linux/bitops.h>
42 #include <linux/iomap.h>
43 #include <linux/iversion.h>
45 #include "ext4_jbd2.h"
50 #include <trace/events/ext4.h>
52 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
53 struct ext4_inode_info *ei)
55 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 int offset = offsetof(struct ext4_inode, i_checksum_lo);
59 unsigned int csum_size = sizeof(dummy_csum);
61 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
62 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
64 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
65 EXT4_GOOD_OLD_INODE_SIZE - offset);
67 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
68 offset = offsetof(struct ext4_inode, i_checksum_hi);
69 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
70 EXT4_GOOD_OLD_INODE_SIZE,
71 offset - EXT4_GOOD_OLD_INODE_SIZE);
72 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
73 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
77 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
78 EXT4_INODE_SIZE(inode->i_sb) - offset);
84 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
85 struct ext4_inode_info *ei)
87 __u32 provided, calculated;
89 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
90 cpu_to_le32(EXT4_OS_LINUX) ||
91 !ext4_has_metadata_csum(inode->i_sb))
94 provided = le16_to_cpu(raw->i_checksum_lo);
95 calculated = ext4_inode_csum(inode, raw, ei);
96 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
97 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
98 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
100 calculated &= 0xFFFF;
102 return provided == calculated;
105 void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
106 struct ext4_inode_info *ei)
110 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
111 cpu_to_le32(EXT4_OS_LINUX) ||
112 !ext4_has_metadata_csum(inode->i_sb))
115 csum = ext4_inode_csum(inode, raw, ei);
116 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
117 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
118 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
119 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
122 static inline int ext4_begin_ordered_truncate(struct inode *inode,
125 trace_ext4_begin_ordered_truncate(inode, new_size);
127 * If jinode is zero, then we never opened the file for
128 * writing, so there's no need to call
129 * jbd2_journal_begin_ordered_truncate() since there's no
130 * outstanding writes we need to flush.
132 if (!EXT4_I(inode)->jinode)
134 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
135 EXT4_I(inode)->jinode,
139 static void ext4_invalidatepage(struct page *page, unsigned int offset,
140 unsigned int length);
141 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
142 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
146 * Test whether an inode is a fast symlink.
147 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
149 int ext4_inode_is_fast_symlink(struct inode *inode)
151 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
152 int ea_blocks = EXT4_I(inode)->i_file_acl ?
153 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
155 if (ext4_has_inline_data(inode))
158 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
160 return S_ISLNK(inode->i_mode) && inode->i_size &&
161 (inode->i_size < EXT4_N_BLOCKS * 4);
165 * Called at the last iput() if i_nlink is zero.
167 void ext4_evict_inode(struct inode *inode)
172 * Credits for final inode cleanup and freeing:
173 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
174 * (xattr block freeing), bitmap, group descriptor (inode freeing)
176 int extra_credits = 6;
177 struct ext4_xattr_inode_array *ea_inode_array = NULL;
178 bool freeze_protected = false;
180 trace_ext4_evict_inode(inode);
182 if (inode->i_nlink) {
184 * When journalling data dirty buffers are tracked only in the
185 * journal. So although mm thinks everything is clean and
186 * ready for reaping the inode might still have some pages to
187 * write in the running transaction or waiting to be
188 * checkpointed. Thus calling jbd2_journal_invalidatepage()
189 * (via truncate_inode_pages()) to discard these buffers can
190 * cause data loss. Also even if we did not discard these
191 * buffers, we would have no way to find them after the inode
192 * is reaped and thus user could see stale data if he tries to
193 * read them before the transaction is checkpointed. So be
194 * careful and force everything to disk here... We use
195 * ei->i_datasync_tid to store the newest transaction
196 * containing inode's data.
198 * Note that directories do not have this problem because they
199 * don't use page cache.
201 if (inode->i_ino != EXT4_JOURNAL_INO &&
202 ext4_should_journal_data(inode) &&
203 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
204 inode->i_data.nrpages) {
205 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
206 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
208 jbd2_complete_transaction(journal, commit_tid);
209 filemap_write_and_wait(&inode->i_data);
211 truncate_inode_pages_final(&inode->i_data);
216 if (is_bad_inode(inode))
218 dquot_initialize(inode);
220 if (ext4_should_order_data(inode))
221 ext4_begin_ordered_truncate(inode, 0);
222 truncate_inode_pages_final(&inode->i_data);
225 * For inodes with journalled data, transaction commit could have
226 * dirtied the inode. Flush worker is ignoring it because of I_FREEING
227 * flag but we still need to remove the inode from the writeback lists.
229 if (!list_empty_careful(&inode->i_io_list)) {
230 WARN_ON_ONCE(!ext4_should_journal_data(inode));
231 inode_io_list_del(inode);
235 * Protect us against freezing - iput() caller didn't have to have any
236 * protection against it. When we are in a running transaction though,
237 * we are already protected against freezing and we cannot grab further
238 * protection due to lock ordering constraints.
240 if (!ext4_journal_current_handle()) {
241 sb_start_intwrite(inode->i_sb);
242 freeze_protected = true;
245 if (!IS_NOQUOTA(inode))
246 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
249 * Block bitmap, group descriptor, and inode are accounted in both
250 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
252 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
253 ext4_blocks_for_truncate(inode) + extra_credits - 3);
254 if (IS_ERR(handle)) {
255 ext4_std_error(inode->i_sb, PTR_ERR(handle));
257 * If we're going to skip the normal cleanup, we still need to
258 * make sure that the in-core orphan linked list is properly
261 ext4_orphan_del(NULL, inode);
262 if (freeze_protected)
263 sb_end_intwrite(inode->i_sb);
268 ext4_handle_sync(handle);
271 * Set inode->i_size to 0 before calling ext4_truncate(). We need
272 * special handling of symlinks here because i_size is used to
273 * determine whether ext4_inode_info->i_data contains symlink data or
274 * block mappings. Setting i_size to 0 will remove its fast symlink
275 * status. Erase i_data so that it becomes a valid empty block map.
277 if (ext4_inode_is_fast_symlink(inode))
278 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
280 err = ext4_mark_inode_dirty(handle, inode);
282 ext4_warning(inode->i_sb,
283 "couldn't mark inode dirty (err %d)", err);
286 if (inode->i_blocks) {
287 err = ext4_truncate(inode);
289 ext4_error_err(inode->i_sb, -err,
290 "couldn't truncate inode %lu (err %d)",
296 /* Remove xattr references. */
297 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
300 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
302 ext4_journal_stop(handle);
303 ext4_orphan_del(NULL, inode);
304 if (freeze_protected)
305 sb_end_intwrite(inode->i_sb);
306 ext4_xattr_inode_array_free(ea_inode_array);
311 * Kill off the orphan record which ext4_truncate created.
312 * AKPM: I think this can be inside the above `if'.
313 * Note that ext4_orphan_del() has to be able to cope with the
314 * deletion of a non-existent orphan - this is because we don't
315 * know if ext4_truncate() actually created an orphan record.
316 * (Well, we could do this if we need to, but heck - it works)
318 ext4_orphan_del(handle, inode);
319 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
322 * One subtle ordering requirement: if anything has gone wrong
323 * (transaction abort, IO errors, whatever), then we can still
324 * do these next steps (the fs will already have been marked as
325 * having errors), but we can't free the inode if the mark_dirty
328 if (ext4_mark_inode_dirty(handle, inode))
329 /* If that failed, just do the required in-core inode clear. */
330 ext4_clear_inode(inode);
332 ext4_free_inode(handle, inode);
333 ext4_journal_stop(handle);
334 if (freeze_protected)
335 sb_end_intwrite(inode->i_sb);
336 ext4_xattr_inode_array_free(ea_inode_array);
339 if (!list_empty(&EXT4_I(inode)->i_fc_list))
340 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM);
341 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
345 qsize_t *ext4_get_reserved_space(struct inode *inode)
347 return &EXT4_I(inode)->i_reserved_quota;
352 * Called with i_data_sem down, which is important since we can call
353 * ext4_discard_preallocations() from here.
355 void ext4_da_update_reserve_space(struct inode *inode,
356 int used, int quota_claim)
358 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
359 struct ext4_inode_info *ei = EXT4_I(inode);
361 spin_lock(&ei->i_block_reservation_lock);
362 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
363 if (unlikely(used > ei->i_reserved_data_blocks)) {
364 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
365 "with only %d reserved data blocks",
366 __func__, inode->i_ino, used,
367 ei->i_reserved_data_blocks);
369 used = ei->i_reserved_data_blocks;
372 /* Update per-inode reservations */
373 ei->i_reserved_data_blocks -= used;
374 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
376 spin_unlock(&ei->i_block_reservation_lock);
378 /* Update quota subsystem for data blocks */
380 dquot_claim_block(inode, EXT4_C2B(sbi, used));
383 * We did fallocate with an offset that is already delayed
384 * allocated. So on delayed allocated writeback we should
385 * not re-claim the quota for fallocated blocks.
387 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
391 * If we have done all the pending block allocations and if
392 * there aren't any writers on the inode, we can discard the
393 * inode's preallocations.
395 if ((ei->i_reserved_data_blocks == 0) &&
396 !inode_is_open_for_write(inode))
397 ext4_discard_preallocations(inode, 0);
400 static int __check_block_validity(struct inode *inode, const char *func,
402 struct ext4_map_blocks *map)
404 if (ext4_has_feature_journal(inode->i_sb) &&
406 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
408 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
409 ext4_error_inode(inode, func, line, map->m_pblk,
410 "lblock %lu mapped to illegal pblock %llu "
411 "(length %d)", (unsigned long) map->m_lblk,
412 map->m_pblk, map->m_len);
413 return -EFSCORRUPTED;
418 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
423 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
424 return fscrypt_zeroout_range(inode, lblk, pblk, len);
426 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
433 #define check_block_validity(inode, map) \
434 __check_block_validity((inode), __func__, __LINE__, (map))
436 #ifdef ES_AGGRESSIVE_TEST
437 static void ext4_map_blocks_es_recheck(handle_t *handle,
439 struct ext4_map_blocks *es_map,
440 struct ext4_map_blocks *map,
447 * There is a race window that the result is not the same.
448 * e.g. xfstests #223 when dioread_nolock enables. The reason
449 * is that we lookup a block mapping in extent status tree with
450 * out taking i_data_sem. So at the time the unwritten extent
451 * could be converted.
453 down_read(&EXT4_I(inode)->i_data_sem);
454 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
455 retval = ext4_ext_map_blocks(handle, inode, map, 0);
457 retval = ext4_ind_map_blocks(handle, inode, map, 0);
459 up_read((&EXT4_I(inode)->i_data_sem));
462 * We don't check m_len because extent will be collpased in status
463 * tree. So the m_len might not equal.
465 if (es_map->m_lblk != map->m_lblk ||
466 es_map->m_flags != map->m_flags ||
467 es_map->m_pblk != map->m_pblk) {
468 printk("ES cache assertion failed for inode: %lu "
469 "es_cached ex [%d/%d/%llu/%x] != "
470 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
471 inode->i_ino, es_map->m_lblk, es_map->m_len,
472 es_map->m_pblk, es_map->m_flags, map->m_lblk,
473 map->m_len, map->m_pblk, map->m_flags,
477 #endif /* ES_AGGRESSIVE_TEST */
480 * The ext4_map_blocks() function tries to look up the requested blocks,
481 * and returns if the blocks are already mapped.
483 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
484 * and store the allocated blocks in the result buffer head and mark it
487 * If file type is extents based, it will call ext4_ext_map_blocks(),
488 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
491 * On success, it returns the number of blocks being mapped or allocated. if
492 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
493 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
495 * It returns 0 if plain look up failed (blocks have not been allocated), in
496 * that case, @map is returned as unmapped but we still do fill map->m_len to
497 * indicate the length of a hole starting at map->m_lblk.
499 * It returns the error in case of allocation failure.
501 int ext4_map_blocks(handle_t *handle, struct inode *inode,
502 struct ext4_map_blocks *map, int flags)
504 struct extent_status es;
507 #ifdef ES_AGGRESSIVE_TEST
508 struct ext4_map_blocks orig_map;
510 memcpy(&orig_map, map, sizeof(*map));
514 ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
515 flags, map->m_len, (unsigned long) map->m_lblk);
518 * ext4_map_blocks returns an int, and m_len is an unsigned int
520 if (unlikely(map->m_len > INT_MAX))
521 map->m_len = INT_MAX;
523 /* We can handle the block number less than EXT_MAX_BLOCKS */
524 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
525 return -EFSCORRUPTED;
527 /* Lookup extent status tree firstly */
528 if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
529 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
530 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
531 map->m_pblk = ext4_es_pblock(&es) +
532 map->m_lblk - es.es_lblk;
533 map->m_flags |= ext4_es_is_written(&es) ?
534 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
535 retval = es.es_len - (map->m_lblk - es.es_lblk);
536 if (retval > map->m_len)
539 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
541 retval = es.es_len - (map->m_lblk - es.es_lblk);
542 if (retval > map->m_len)
549 #ifdef ES_AGGRESSIVE_TEST
550 ext4_map_blocks_es_recheck(handle, inode, map,
557 * Try to see if we can get the block without requesting a new
560 down_read(&EXT4_I(inode)->i_data_sem);
561 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
562 retval = ext4_ext_map_blocks(handle, inode, map, 0);
564 retval = ext4_ind_map_blocks(handle, inode, map, 0);
569 if (unlikely(retval != map->m_len)) {
570 ext4_warning(inode->i_sb,
571 "ES len assertion failed for inode "
572 "%lu: retval %d != map->m_len %d",
573 inode->i_ino, retval, map->m_len);
577 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
578 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
579 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
580 !(status & EXTENT_STATUS_WRITTEN) &&
581 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
582 map->m_lblk + map->m_len - 1))
583 status |= EXTENT_STATUS_DELAYED;
584 ret = ext4_es_insert_extent(inode, map->m_lblk,
585 map->m_len, map->m_pblk, status);
589 up_read((&EXT4_I(inode)->i_data_sem));
592 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
593 ret = check_block_validity(inode, map);
598 /* If it is only a block(s) look up */
599 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
603 * Returns if the blocks have already allocated
605 * Note that if blocks have been preallocated
606 * ext4_ext_get_block() returns the create = 0
607 * with buffer head unmapped.
609 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
611 * If we need to convert extent to unwritten
612 * we continue and do the actual work in
613 * ext4_ext_map_blocks()
615 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
619 * Here we clear m_flags because after allocating an new extent,
620 * it will be set again.
622 map->m_flags &= ~EXT4_MAP_FLAGS;
625 * New blocks allocate and/or writing to unwritten extent
626 * will possibly result in updating i_data, so we take
627 * the write lock of i_data_sem, and call get_block()
628 * with create == 1 flag.
630 down_write(&EXT4_I(inode)->i_data_sem);
633 * We need to check for EXT4 here because migrate
634 * could have changed the inode type in between
636 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
637 retval = ext4_ext_map_blocks(handle, inode, map, flags);
639 retval = ext4_ind_map_blocks(handle, inode, map, flags);
641 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
643 * We allocated new blocks which will result in
644 * i_data's format changing. Force the migrate
645 * to fail by clearing migrate flags
647 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
651 * Update reserved blocks/metadata blocks after successful
652 * block allocation which had been deferred till now. We don't
653 * support fallocate for non extent files. So we can update
654 * reserve space here.
657 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
658 ext4_da_update_reserve_space(inode, retval, 1);
664 if (unlikely(retval != map->m_len)) {
665 ext4_warning(inode->i_sb,
666 "ES len assertion failed for inode "
667 "%lu: retval %d != map->m_len %d",
668 inode->i_ino, retval, map->m_len);
673 * We have to zeroout blocks before inserting them into extent
674 * status tree. Otherwise someone could look them up there and
675 * use them before they are really zeroed. We also have to
676 * unmap metadata before zeroing as otherwise writeback can
677 * overwrite zeros with stale data from block device.
679 if (flags & EXT4_GET_BLOCKS_ZERO &&
680 map->m_flags & EXT4_MAP_MAPPED &&
681 map->m_flags & EXT4_MAP_NEW) {
682 ret = ext4_issue_zeroout(inode, map->m_lblk,
683 map->m_pblk, map->m_len);
691 * If the extent has been zeroed out, we don't need to update
692 * extent status tree.
694 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
695 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
696 if (ext4_es_is_written(&es))
699 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
700 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
701 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
702 !(status & EXTENT_STATUS_WRITTEN) &&
703 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
704 map->m_lblk + map->m_len - 1))
705 status |= EXTENT_STATUS_DELAYED;
706 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
707 map->m_pblk, status);
715 up_write((&EXT4_I(inode)->i_data_sem));
716 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
717 ret = check_block_validity(inode, map);
722 * Inodes with freshly allocated blocks where contents will be
723 * visible after transaction commit must be on transaction's
726 if (map->m_flags & EXT4_MAP_NEW &&
727 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
728 !(flags & EXT4_GET_BLOCKS_ZERO) &&
729 !ext4_is_quota_file(inode) &&
730 ext4_should_order_data(inode)) {
732 (loff_t)map->m_lblk << inode->i_blkbits;
733 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
735 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
736 ret = ext4_jbd2_inode_add_wait(handle, inode,
739 ret = ext4_jbd2_inode_add_write(handle, inode,
744 ext4_fc_track_range(handle, inode, map->m_lblk,
745 map->m_lblk + map->m_len - 1);
749 ext_debug(inode, "failed with err %d\n", retval);
754 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
755 * we have to be careful as someone else may be manipulating b_state as well.
757 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
759 unsigned long old_state;
760 unsigned long new_state;
762 flags &= EXT4_MAP_FLAGS;
764 /* Dummy buffer_head? Set non-atomically. */
766 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
770 * Someone else may be modifying b_state. Be careful! This is ugly but
771 * once we get rid of using bh as a container for mapping information
772 * to pass to / from get_block functions, this can go away.
775 old_state = READ_ONCE(bh->b_state);
776 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
778 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
781 static int _ext4_get_block(struct inode *inode, sector_t iblock,
782 struct buffer_head *bh, int flags)
784 struct ext4_map_blocks map;
787 if (ext4_has_inline_data(inode))
791 map.m_len = bh->b_size >> inode->i_blkbits;
793 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
796 map_bh(bh, inode->i_sb, map.m_pblk);
797 ext4_update_bh_state(bh, map.m_flags);
798 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
800 } else if (ret == 0) {
801 /* hole case, need to fill in bh->b_size */
802 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
807 int ext4_get_block(struct inode *inode, sector_t iblock,
808 struct buffer_head *bh, int create)
810 return _ext4_get_block(inode, iblock, bh,
811 create ? EXT4_GET_BLOCKS_CREATE : 0);
815 * Get block function used when preparing for buffered write if we require
816 * creating an unwritten extent if blocks haven't been allocated. The extent
817 * will be converted to written after the IO is complete.
819 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
820 struct buffer_head *bh_result, int create)
822 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
823 inode->i_ino, create);
824 return _ext4_get_block(inode, iblock, bh_result,
825 EXT4_GET_BLOCKS_IO_CREATE_EXT);
828 /* Maximum number of blocks we map for direct IO at once. */
829 #define DIO_MAX_BLOCKS 4096
832 * `handle' can be NULL if create is zero
834 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
835 ext4_lblk_t block, int map_flags)
837 struct ext4_map_blocks map;
838 struct buffer_head *bh;
839 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
842 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
843 || handle != NULL || create == 0);
847 err = ext4_map_blocks(handle, inode, &map, map_flags);
850 return create ? ERR_PTR(-ENOSPC) : NULL;
854 bh = sb_getblk(inode->i_sb, map.m_pblk);
856 return ERR_PTR(-ENOMEM);
857 if (map.m_flags & EXT4_MAP_NEW) {
859 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
860 || (handle != NULL));
863 * Now that we do not always journal data, we should
864 * keep in mind whether this should always journal the
865 * new buffer as metadata. For now, regular file
866 * writes use ext4_get_block instead, so it's not a
870 BUFFER_TRACE(bh, "call get_create_access");
871 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
877 if (!buffer_uptodate(bh)) {
878 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
879 set_buffer_uptodate(bh);
882 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
883 err = ext4_handle_dirty_metadata(handle, inode, bh);
887 BUFFER_TRACE(bh, "not a new buffer");
894 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
895 ext4_lblk_t block, int map_flags)
897 struct buffer_head *bh;
900 bh = ext4_getblk(handle, inode, block, map_flags);
903 if (!bh || ext4_buffer_uptodate(bh))
906 ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
914 /* Read a contiguous batch of blocks. */
915 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
916 bool wait, struct buffer_head **bhs)
920 for (i = 0; i < bh_count; i++) {
921 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
922 if (IS_ERR(bhs[i])) {
923 err = PTR_ERR(bhs[i]);
929 for (i = 0; i < bh_count; i++)
930 /* Note that NULL bhs[i] is valid because of holes. */
931 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
932 ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
937 for (i = 0; i < bh_count; i++)
939 wait_on_buffer(bhs[i]);
941 for (i = 0; i < bh_count; i++) {
942 if (bhs[i] && !buffer_uptodate(bhs[i])) {
950 for (i = 0; i < bh_count; i++) {
957 int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
958 struct buffer_head *head,
962 int (*fn)(handle_t *handle, struct inode *inode,
963 struct buffer_head *bh))
965 struct buffer_head *bh;
966 unsigned block_start, block_end;
967 unsigned blocksize = head->b_size;
969 struct buffer_head *next;
971 for (bh = head, block_start = 0;
972 ret == 0 && (bh != head || !block_start);
973 block_start = block_end, bh = next) {
974 next = bh->b_this_page;
975 block_end = block_start + blocksize;
976 if (block_end <= from || block_start >= to) {
977 if (partial && !buffer_uptodate(bh))
981 err = (*fn)(handle, inode, bh);
989 * To preserve ordering, it is essential that the hole instantiation and
990 * the data write be encapsulated in a single transaction. We cannot
991 * close off a transaction and start a new one between the ext4_get_block()
992 * and the commit_write(). So doing the jbd2_journal_start at the start of
993 * prepare_write() is the right place.
995 * Also, this function can nest inside ext4_writepage(). In that case, we
996 * *know* that ext4_writepage() has generated enough buffer credits to do the
997 * whole page. So we won't block on the journal in that case, which is good,
998 * because the caller may be PF_MEMALLOC.
1000 * By accident, ext4 can be reentered when a transaction is open via
1001 * quota file writes. If we were to commit the transaction while thus
1002 * reentered, there can be a deadlock - we would be holding a quota
1003 * lock, and the commit would never complete if another thread had a
1004 * transaction open and was blocking on the quota lock - a ranking
1007 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1008 * will _not_ run commit under these circumstances because handle->h_ref
1009 * is elevated. We'll still have enough credits for the tiny quotafile
1012 int do_journal_get_write_access(handle_t *handle, struct inode *inode,
1013 struct buffer_head *bh)
1015 int dirty = buffer_dirty(bh);
1018 if (!buffer_mapped(bh) || buffer_freed(bh))
1021 * __block_write_begin() could have dirtied some buffers. Clean
1022 * the dirty bit as jbd2_journal_get_write_access() could complain
1023 * otherwise about fs integrity issues. Setting of the dirty bit
1024 * by __block_write_begin() isn't a real problem here as we clear
1025 * the bit before releasing a page lock and thus writeback cannot
1026 * ever write the buffer.
1029 clear_buffer_dirty(bh);
1030 BUFFER_TRACE(bh, "get write access");
1031 ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1034 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1038 #ifdef CONFIG_FS_ENCRYPTION
1039 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1040 get_block_t *get_block)
1042 unsigned from = pos & (PAGE_SIZE - 1);
1043 unsigned to = from + len;
1044 struct inode *inode = page->mapping->host;
1045 unsigned block_start, block_end;
1048 unsigned blocksize = inode->i_sb->s_blocksize;
1050 struct buffer_head *bh, *head, *wait[2];
1054 BUG_ON(!PageLocked(page));
1055 BUG_ON(from > PAGE_SIZE);
1056 BUG_ON(to > PAGE_SIZE);
1059 if (!page_has_buffers(page))
1060 create_empty_buffers(page, blocksize, 0);
1061 head = page_buffers(page);
1062 bbits = ilog2(blocksize);
1063 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1065 for (bh = head, block_start = 0; bh != head || !block_start;
1066 block++, block_start = block_end, bh = bh->b_this_page) {
1067 block_end = block_start + blocksize;
1068 if (block_end <= from || block_start >= to) {
1069 if (PageUptodate(page)) {
1070 set_buffer_uptodate(bh);
1075 clear_buffer_new(bh);
1076 if (!buffer_mapped(bh)) {
1077 WARN_ON(bh->b_size != blocksize);
1078 err = get_block(inode, block, bh, 1);
1081 if (buffer_new(bh)) {
1082 if (PageUptodate(page)) {
1083 clear_buffer_new(bh);
1084 set_buffer_uptodate(bh);
1085 mark_buffer_dirty(bh);
1088 if (block_end > to || block_start < from)
1089 zero_user_segments(page, to, block_end,
1094 if (PageUptodate(page)) {
1095 set_buffer_uptodate(bh);
1098 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1099 !buffer_unwritten(bh) &&
1100 (block_start < from || block_end > to)) {
1101 ext4_read_bh_lock(bh, 0, false);
1102 wait[nr_wait++] = bh;
1106 * If we issued read requests, let them complete.
1108 for (i = 0; i < nr_wait; i++) {
1109 wait_on_buffer(wait[i]);
1110 if (!buffer_uptodate(wait[i]))
1113 if (unlikely(err)) {
1114 page_zero_new_buffers(page, from, to);
1115 } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1116 for (i = 0; i < nr_wait; i++) {
1119 err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1120 bh_offset(wait[i]));
1122 clear_buffer_uptodate(wait[i]);
1132 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1133 loff_t pos, unsigned len, unsigned flags,
1134 struct page **pagep, void **fsdata)
1136 struct inode *inode = mapping->host;
1137 int ret, needed_blocks;
1144 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1147 trace_ext4_write_begin(inode, pos, len, flags);
1149 * Reserve one block more for addition to orphan list in case
1150 * we allocate blocks but write fails for some reason
1152 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1153 index = pos >> PAGE_SHIFT;
1154 from = pos & (PAGE_SIZE - 1);
1157 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1158 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1167 * grab_cache_page_write_begin() can take a long time if the
1168 * system is thrashing due to memory pressure, or if the page
1169 * is being written back. So grab it first before we start
1170 * the transaction handle. This also allows us to allocate
1171 * the page (if needed) without using GFP_NOFS.
1174 page = grab_cache_page_write_begin(mapping, index, flags);
1180 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1181 if (IS_ERR(handle)) {
1183 return PTR_ERR(handle);
1187 if (page->mapping != mapping) {
1188 /* The page got truncated from under us */
1191 ext4_journal_stop(handle);
1194 /* In case writeback began while the page was unlocked */
1195 wait_for_stable_page(page);
1197 #ifdef CONFIG_FS_ENCRYPTION
1198 if (ext4_should_dioread_nolock(inode))
1199 ret = ext4_block_write_begin(page, pos, len,
1200 ext4_get_block_unwritten);
1202 ret = ext4_block_write_begin(page, pos, len,
1205 if (ext4_should_dioread_nolock(inode))
1206 ret = __block_write_begin(page, pos, len,
1207 ext4_get_block_unwritten);
1209 ret = __block_write_begin(page, pos, len, ext4_get_block);
1211 if (!ret && ext4_should_journal_data(inode)) {
1212 ret = ext4_walk_page_buffers(handle, inode,
1213 page_buffers(page), from, to, NULL,
1214 do_journal_get_write_access);
1218 bool extended = (pos + len > inode->i_size) &&
1219 !ext4_verity_in_progress(inode);
1223 * __block_write_begin may have instantiated a few blocks
1224 * outside i_size. Trim these off again. Don't need
1225 * i_size_read because we hold i_mutex.
1227 * Add inode to orphan list in case we crash before
1230 if (extended && ext4_can_truncate(inode))
1231 ext4_orphan_add(handle, inode);
1233 ext4_journal_stop(handle);
1235 ext4_truncate_failed_write(inode);
1237 * If truncate failed early the inode might
1238 * still be on the orphan list; we need to
1239 * make sure the inode is removed from the
1240 * orphan list in that case.
1243 ext4_orphan_del(NULL, inode);
1246 if (ret == -ENOSPC &&
1247 ext4_should_retry_alloc(inode->i_sb, &retries))
1256 /* For write_end() in data=journal mode */
1257 static int write_end_fn(handle_t *handle, struct inode *inode,
1258 struct buffer_head *bh)
1261 if (!buffer_mapped(bh) || buffer_freed(bh))
1263 set_buffer_uptodate(bh);
1264 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1265 clear_buffer_meta(bh);
1266 clear_buffer_prio(bh);
1271 * We need to pick up the new inode size which generic_commit_write gave us
1272 * `file' can be NULL - eg, when called from page_symlink().
1274 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1275 * buffers are managed internally.
1277 static int ext4_write_end(struct file *file,
1278 struct address_space *mapping,
1279 loff_t pos, unsigned len, unsigned copied,
1280 struct page *page, void *fsdata)
1282 handle_t *handle = ext4_journal_current_handle();
1283 struct inode *inode = mapping->host;
1284 loff_t old_size = inode->i_size;
1286 int i_size_changed = 0;
1287 bool verity = ext4_verity_in_progress(inode);
1289 trace_ext4_write_end(inode, pos, len, copied);
1291 if (ext4_has_inline_data(inode))
1292 return ext4_write_inline_data_end(inode, pos, len, copied, page);
1294 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1296 * it's important to update i_size while still holding page lock:
1297 * page writeout could otherwise come in and zero beyond i_size.
1299 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1300 * blocks are being written past EOF, so skip the i_size update.
1303 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1307 if (old_size < pos && !verity)
1308 pagecache_isize_extended(inode, old_size, pos);
1310 * Don't mark the inode dirty under page lock. First, it unnecessarily
1311 * makes the holding time of page lock longer. Second, it forces lock
1312 * ordering of page lock and transaction start for journaling
1316 ret = ext4_mark_inode_dirty(handle, inode);
1318 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1319 /* if we have allocated more blocks and copied
1320 * less. We will have blocks allocated outside
1321 * inode->i_size. So truncate them
1323 ext4_orphan_add(handle, inode);
1325 ret2 = ext4_journal_stop(handle);
1329 if (pos + len > inode->i_size && !verity) {
1330 ext4_truncate_failed_write(inode);
1332 * If truncate failed early the inode might still be
1333 * on the orphan list; we need to make sure the inode
1334 * is removed from the orphan list in that case.
1337 ext4_orphan_del(NULL, inode);
1340 return ret ? ret : copied;
1344 * This is a private version of page_zero_new_buffers() which doesn't
1345 * set the buffer to be dirty, since in data=journalled mode we need
1346 * to call ext4_handle_dirty_metadata() instead.
1348 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1349 struct inode *inode,
1351 unsigned from, unsigned to)
1353 unsigned int block_start = 0, block_end;
1354 struct buffer_head *head, *bh;
1356 bh = head = page_buffers(page);
1358 block_end = block_start + bh->b_size;
1359 if (buffer_new(bh)) {
1360 if (block_end > from && block_start < to) {
1361 if (!PageUptodate(page)) {
1362 unsigned start, size;
1364 start = max(from, block_start);
1365 size = min(to, block_end) - start;
1367 zero_user(page, start, size);
1368 write_end_fn(handle, inode, bh);
1370 clear_buffer_new(bh);
1373 block_start = block_end;
1374 bh = bh->b_this_page;
1375 } while (bh != head);
1378 static int ext4_journalled_write_end(struct file *file,
1379 struct address_space *mapping,
1380 loff_t pos, unsigned len, unsigned copied,
1381 struct page *page, void *fsdata)
1383 handle_t *handle = ext4_journal_current_handle();
1384 struct inode *inode = mapping->host;
1385 loff_t old_size = inode->i_size;
1389 int size_changed = 0;
1390 bool verity = ext4_verity_in_progress(inode);
1392 trace_ext4_journalled_write_end(inode, pos, len, copied);
1393 from = pos & (PAGE_SIZE - 1);
1396 BUG_ON(!ext4_handle_valid(handle));
1398 if (ext4_has_inline_data(inode))
1399 return ext4_write_inline_data_end(inode, pos, len, copied, page);
1401 if (unlikely(copied < len) && !PageUptodate(page)) {
1403 ext4_journalled_zero_new_buffers(handle, inode, page, from, to);
1405 if (unlikely(copied < len))
1406 ext4_journalled_zero_new_buffers(handle, inode, page,
1408 ret = ext4_walk_page_buffers(handle, inode, page_buffers(page),
1409 from, from + copied, &partial,
1412 SetPageUptodate(page);
1415 size_changed = ext4_update_inode_size(inode, pos + copied);
1416 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1417 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1421 if (old_size < pos && !verity)
1422 pagecache_isize_extended(inode, old_size, pos);
1425 ret2 = ext4_mark_inode_dirty(handle, inode);
1430 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1431 /* if we have allocated more blocks and copied
1432 * less. We will have blocks allocated outside
1433 * inode->i_size. So truncate them
1435 ext4_orphan_add(handle, inode);
1437 ret2 = ext4_journal_stop(handle);
1440 if (pos + len > inode->i_size && !verity) {
1441 ext4_truncate_failed_write(inode);
1443 * If truncate failed early the inode might still be
1444 * on the orphan list; we need to make sure the inode
1445 * is removed from the orphan list in that case.
1448 ext4_orphan_del(NULL, inode);
1451 return ret ? ret : copied;
1455 * Reserve space for a single cluster
1457 static int ext4_da_reserve_space(struct inode *inode)
1459 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1460 struct ext4_inode_info *ei = EXT4_I(inode);
1464 * We will charge metadata quota at writeout time; this saves
1465 * us from metadata over-estimation, though we may go over by
1466 * a small amount in the end. Here we just reserve for data.
1468 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1472 spin_lock(&ei->i_block_reservation_lock);
1473 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1474 spin_unlock(&ei->i_block_reservation_lock);
1475 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1478 ei->i_reserved_data_blocks++;
1479 trace_ext4_da_reserve_space(inode);
1480 spin_unlock(&ei->i_block_reservation_lock);
1482 return 0; /* success */
1485 void ext4_da_release_space(struct inode *inode, int to_free)
1487 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1488 struct ext4_inode_info *ei = EXT4_I(inode);
1491 return; /* Nothing to release, exit */
1493 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1495 trace_ext4_da_release_space(inode, to_free);
1496 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1498 * if there aren't enough reserved blocks, then the
1499 * counter is messed up somewhere. Since this
1500 * function is called from invalidate page, it's
1501 * harmless to return without any action.
1503 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1504 "ino %lu, to_free %d with only %d reserved "
1505 "data blocks", inode->i_ino, to_free,
1506 ei->i_reserved_data_blocks);
1508 to_free = ei->i_reserved_data_blocks;
1510 ei->i_reserved_data_blocks -= to_free;
1512 /* update fs dirty data blocks counter */
1513 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1515 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1517 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1521 * Delayed allocation stuff
1524 struct mpage_da_data {
1525 struct inode *inode;
1526 struct writeback_control *wbc;
1528 pgoff_t first_page; /* The first page to write */
1529 pgoff_t next_page; /* Current page to examine */
1530 pgoff_t last_page; /* Last page to examine */
1532 * Extent to map - this can be after first_page because that can be
1533 * fully mapped. We somewhat abuse m_flags to store whether the extent
1534 * is delalloc or unwritten.
1536 struct ext4_map_blocks map;
1537 struct ext4_io_submit io_submit; /* IO submission data */
1538 unsigned int do_map:1;
1539 unsigned int scanned_until_end:1;
1542 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1547 struct pagevec pvec;
1548 struct inode *inode = mpd->inode;
1549 struct address_space *mapping = inode->i_mapping;
1551 /* This is necessary when next_page == 0. */
1552 if (mpd->first_page >= mpd->next_page)
1555 mpd->scanned_until_end = 0;
1556 index = mpd->first_page;
1557 end = mpd->next_page - 1;
1559 ext4_lblk_t start, last;
1560 start = index << (PAGE_SHIFT - inode->i_blkbits);
1561 last = end << (PAGE_SHIFT - inode->i_blkbits);
1562 ext4_es_remove_extent(inode, start, last - start + 1);
1565 pagevec_init(&pvec);
1566 while (index <= end) {
1567 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1570 for (i = 0; i < nr_pages; i++) {
1571 struct page *page = pvec.pages[i];
1573 BUG_ON(!PageLocked(page));
1574 BUG_ON(PageWriteback(page));
1576 if (page_mapped(page))
1577 clear_page_dirty_for_io(page);
1578 block_invalidatepage(page, 0, PAGE_SIZE);
1579 ClearPageUptodate(page);
1583 pagevec_release(&pvec);
1587 static void ext4_print_free_blocks(struct inode *inode)
1589 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1590 struct super_block *sb = inode->i_sb;
1591 struct ext4_inode_info *ei = EXT4_I(inode);
1593 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1594 EXT4_C2B(EXT4_SB(inode->i_sb),
1595 ext4_count_free_clusters(sb)));
1596 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1597 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1598 (long long) EXT4_C2B(EXT4_SB(sb),
1599 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1600 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1601 (long long) EXT4_C2B(EXT4_SB(sb),
1602 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1603 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1604 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1605 ei->i_reserved_data_blocks);
1609 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct inode *inode,
1610 struct buffer_head *bh)
1612 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1616 * ext4_insert_delayed_block - adds a delayed block to the extents status
1617 * tree, incrementing the reserved cluster/block
1618 * count or making a pending reservation
1621 * @inode - file containing the newly added block
1622 * @lblk - logical block to be added
1624 * Returns 0 on success, negative error code on failure.
1626 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1628 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1630 bool allocated = false;
1631 bool reserved = false;
1634 * If the cluster containing lblk is shared with a delayed,
1635 * written, or unwritten extent in a bigalloc file system, it's
1636 * already been accounted for and does not need to be reserved.
1637 * A pending reservation must be made for the cluster if it's
1638 * shared with a written or unwritten extent and doesn't already
1639 * have one. Written and unwritten extents can be purged from the
1640 * extents status tree if the system is under memory pressure, so
1641 * it's necessary to examine the extent tree if a search of the
1642 * extents status tree doesn't get a match.
1644 if (sbi->s_cluster_ratio == 1) {
1645 ret = ext4_da_reserve_space(inode);
1646 if (ret != 0) /* ENOSPC */
1649 } else { /* bigalloc */
1650 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1651 if (!ext4_es_scan_clu(inode,
1652 &ext4_es_is_mapped, lblk)) {
1653 ret = ext4_clu_mapped(inode,
1654 EXT4_B2C(sbi, lblk));
1658 ret = ext4_da_reserve_space(inode);
1659 if (ret != 0) /* ENOSPC */
1671 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1672 if (ret && reserved)
1673 ext4_da_release_space(inode, 1);
1680 * This function is grabs code from the very beginning of
1681 * ext4_map_blocks, but assumes that the caller is from delayed write
1682 * time. This function looks up the requested blocks and sets the
1683 * buffer delay bit under the protection of i_data_sem.
1685 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1686 struct ext4_map_blocks *map,
1687 struct buffer_head *bh)
1689 struct extent_status es;
1691 sector_t invalid_block = ~((sector_t) 0xffff);
1692 #ifdef ES_AGGRESSIVE_TEST
1693 struct ext4_map_blocks orig_map;
1695 memcpy(&orig_map, map, sizeof(*map));
1698 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1702 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1703 (unsigned long) map->m_lblk);
1705 /* Lookup extent status tree firstly */
1706 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1707 if (ext4_es_is_hole(&es)) {
1709 down_read(&EXT4_I(inode)->i_data_sem);
1714 * the buffer head associated with a delayed and not unwritten
1715 * block found in the extent status cache must contain an
1716 * invalid block number and have its BH_New and BH_Delay bits
1717 * set, reflecting the state assigned when the block was
1718 * initially delayed allocated
1720 if (ext4_es_is_delonly(&es)) {
1721 BUG_ON(bh->b_blocknr != invalid_block);
1722 BUG_ON(!buffer_new(bh));
1723 BUG_ON(!buffer_delay(bh));
1727 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1728 retval = es.es_len - (iblock - es.es_lblk);
1729 if (retval > map->m_len)
1730 retval = map->m_len;
1731 map->m_len = retval;
1732 if (ext4_es_is_written(&es))
1733 map->m_flags |= EXT4_MAP_MAPPED;
1734 else if (ext4_es_is_unwritten(&es))
1735 map->m_flags |= EXT4_MAP_UNWRITTEN;
1739 #ifdef ES_AGGRESSIVE_TEST
1740 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1746 * Try to see if we can get the block without requesting a new
1747 * file system block.
1749 down_read(&EXT4_I(inode)->i_data_sem);
1750 if (ext4_has_inline_data(inode))
1752 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1753 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1755 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1762 * XXX: __block_prepare_write() unmaps passed block,
1766 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1772 map_bh(bh, inode->i_sb, invalid_block);
1774 set_buffer_delay(bh);
1775 } else if (retval > 0) {
1777 unsigned int status;
1779 if (unlikely(retval != map->m_len)) {
1780 ext4_warning(inode->i_sb,
1781 "ES len assertion failed for inode "
1782 "%lu: retval %d != map->m_len %d",
1783 inode->i_ino, retval, map->m_len);
1787 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1788 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1789 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1790 map->m_pblk, status);
1796 up_read((&EXT4_I(inode)->i_data_sem));
1802 * This is a special get_block_t callback which is used by
1803 * ext4_da_write_begin(). It will either return mapped block or
1804 * reserve space for a single block.
1806 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1807 * We also have b_blocknr = -1 and b_bdev initialized properly
1809 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1810 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1811 * initialized properly.
1813 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1814 struct buffer_head *bh, int create)
1816 struct ext4_map_blocks map;
1819 BUG_ON(create == 0);
1820 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1822 map.m_lblk = iblock;
1826 * first, we need to know whether the block is allocated already
1827 * preallocated blocks are unmapped but should treated
1828 * the same as allocated blocks.
1830 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1834 map_bh(bh, inode->i_sb, map.m_pblk);
1835 ext4_update_bh_state(bh, map.m_flags);
1837 if (buffer_unwritten(bh)) {
1838 /* A delayed write to unwritten bh should be marked
1839 * new and mapped. Mapped ensures that we don't do
1840 * get_block multiple times when we write to the same
1841 * offset and new ensures that we do proper zero out
1842 * for partial write.
1845 set_buffer_mapped(bh);
1850 static int bget_one(handle_t *handle, struct inode *inode,
1851 struct buffer_head *bh)
1857 static int bput_one(handle_t *handle, struct inode *inode,
1858 struct buffer_head *bh)
1864 static int __ext4_journalled_writepage(struct page *page,
1867 struct address_space *mapping = page->mapping;
1868 struct inode *inode = mapping->host;
1869 struct buffer_head *page_bufs = NULL;
1870 handle_t *handle = NULL;
1871 int ret = 0, err = 0;
1872 int inline_data = ext4_has_inline_data(inode);
1873 struct buffer_head *inode_bh = NULL;
1875 ClearPageChecked(page);
1878 BUG_ON(page->index != 0);
1879 BUG_ON(len > ext4_get_max_inline_size(inode));
1880 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1881 if (inode_bh == NULL)
1884 page_bufs = page_buffers(page);
1889 ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1893 * We need to release the page lock before we start the
1894 * journal, so grab a reference so the page won't disappear
1895 * out from under us.
1900 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1901 ext4_writepage_trans_blocks(inode));
1902 if (IS_ERR(handle)) {
1903 ret = PTR_ERR(handle);
1905 goto out_no_pagelock;
1907 BUG_ON(!ext4_handle_valid(handle));
1911 if (page->mapping != mapping) {
1912 /* The page got truncated from under us */
1913 ext4_journal_stop(handle);
1919 ret = ext4_mark_inode_dirty(handle, inode);
1921 ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1922 NULL, do_journal_get_write_access);
1924 err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1925 NULL, write_end_fn);
1929 err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len);
1932 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1933 err = ext4_journal_stop(handle);
1937 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1941 if (!inline_data && page_bufs)
1942 ext4_walk_page_buffers(NULL, inode, page_bufs, 0, len,
1949 * Note that we don't need to start a transaction unless we're journaling data
1950 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1951 * need to file the inode to the transaction's list in ordered mode because if
1952 * we are writing back data added by write(), the inode is already there and if
1953 * we are writing back data modified via mmap(), no one guarantees in which
1954 * transaction the data will hit the disk. In case we are journaling data, we
1955 * cannot start transaction directly because transaction start ranks above page
1956 * lock so we have to do some magic.
1958 * This function can get called via...
1959 * - ext4_writepages after taking page lock (have journal handle)
1960 * - journal_submit_inode_data_buffers (no journal handle)
1961 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1962 * - grab_page_cache when doing write_begin (have journal handle)
1964 * We don't do any block allocation in this function. If we have page with
1965 * multiple blocks we need to write those buffer_heads that are mapped. This
1966 * is important for mmaped based write. So if we do with blocksize 1K
1967 * truncate(f, 1024);
1968 * a = mmap(f, 0, 4096);
1970 * truncate(f, 4096);
1971 * we have in the page first buffer_head mapped via page_mkwrite call back
1972 * but other buffer_heads would be unmapped but dirty (dirty done via the
1973 * do_wp_page). So writepage should write the first block. If we modify
1974 * the mmap area beyond 1024 we will again get a page_fault and the
1975 * page_mkwrite callback will do the block allocation and mark the
1976 * buffer_heads mapped.
1978 * We redirty the page if we have any buffer_heads that is either delay or
1979 * unwritten in the page.
1981 * We can get recursively called as show below.
1983 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1986 * But since we don't do any block allocation we should not deadlock.
1987 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1989 static int ext4_writepage(struct page *page,
1990 struct writeback_control *wbc)
1995 struct buffer_head *page_bufs = NULL;
1996 struct inode *inode = page->mapping->host;
1997 struct ext4_io_submit io_submit;
1998 bool keep_towrite = false;
2000 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2001 inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
2006 trace_ext4_writepage(page);
2007 size = i_size_read(inode);
2008 if (page->index == size >> PAGE_SHIFT &&
2009 !ext4_verity_in_progress(inode))
2010 len = size & ~PAGE_MASK;
2014 page_bufs = page_buffers(page);
2016 * We cannot do block allocation or other extent handling in this
2017 * function. If there are buffers needing that, we have to redirty
2018 * the page. But we may reach here when we do a journal commit via
2019 * journal_submit_inode_data_buffers() and in that case we must write
2020 * allocated buffers to achieve data=ordered mode guarantees.
2022 * Also, if there is only one buffer per page (the fs block
2023 * size == the page size), if one buffer needs block
2024 * allocation or needs to modify the extent tree to clear the
2025 * unwritten flag, we know that the page can't be written at
2026 * all, so we might as well refuse the write immediately.
2027 * Unfortunately if the block size != page size, we can't as
2028 * easily detect this case using ext4_walk_page_buffers(), but
2029 * for the extremely common case, this is an optimization that
2030 * skips a useless round trip through ext4_bio_write_page().
2032 if (ext4_walk_page_buffers(NULL, inode, page_bufs, 0, len, NULL,
2033 ext4_bh_delay_or_unwritten)) {
2034 redirty_page_for_writepage(wbc, page);
2035 if ((current->flags & PF_MEMALLOC) ||
2036 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2038 * For memory cleaning there's no point in writing only
2039 * some buffers. So just bail out. Warn if we came here
2040 * from direct reclaim.
2042 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2047 keep_towrite = true;
2050 if (PageChecked(page) && ext4_should_journal_data(inode))
2052 * It's mmapped pagecache. Add buffers and journal it. There
2053 * doesn't seem much point in redirtying the page here.
2055 return __ext4_journalled_writepage(page, len);
2057 ext4_io_submit_init(&io_submit, wbc);
2058 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2059 if (!io_submit.io_end) {
2060 redirty_page_for_writepage(wbc, page);
2064 ret = ext4_bio_write_page(&io_submit, page, len, keep_towrite);
2065 ext4_io_submit(&io_submit);
2066 /* Drop io_end reference we got from init */
2067 ext4_put_io_end_defer(io_submit.io_end);
2071 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2077 BUG_ON(page->index != mpd->first_page);
2078 clear_page_dirty_for_io(page);
2080 * We have to be very careful here! Nothing protects writeback path
2081 * against i_size changes and the page can be writeably mapped into
2082 * page tables. So an application can be growing i_size and writing
2083 * data through mmap while writeback runs. clear_page_dirty_for_io()
2084 * write-protects our page in page tables and the page cannot get
2085 * written to again until we release page lock. So only after
2086 * clear_page_dirty_for_io() we are safe to sample i_size for
2087 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2088 * on the barrier provided by TestClearPageDirty in
2089 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2090 * after page tables are updated.
2092 size = i_size_read(mpd->inode);
2093 if (page->index == size >> PAGE_SHIFT &&
2094 !ext4_verity_in_progress(mpd->inode))
2095 len = size & ~PAGE_MASK;
2098 err = ext4_bio_write_page(&mpd->io_submit, page, len, false);
2100 mpd->wbc->nr_to_write--;
2106 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2109 * mballoc gives us at most this number of blocks...
2110 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2111 * The rest of mballoc seems to handle chunks up to full group size.
2113 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2116 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2118 * @mpd - extent of blocks
2119 * @lblk - logical number of the block in the file
2120 * @bh - buffer head we want to add to the extent
2122 * The function is used to collect contig. blocks in the same state. If the
2123 * buffer doesn't require mapping for writeback and we haven't started the
2124 * extent of buffers to map yet, the function returns 'true' immediately - the
2125 * caller can write the buffer right away. Otherwise the function returns true
2126 * if the block has been added to the extent, false if the block couldn't be
2129 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2130 struct buffer_head *bh)
2132 struct ext4_map_blocks *map = &mpd->map;
2134 /* Buffer that doesn't need mapping for writeback? */
2135 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2136 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2137 /* So far no extent to map => we write the buffer right away */
2138 if (map->m_len == 0)
2143 /* First block in the extent? */
2144 if (map->m_len == 0) {
2145 /* We cannot map unless handle is started... */
2150 map->m_flags = bh->b_state & BH_FLAGS;
2154 /* Don't go larger than mballoc is willing to allocate */
2155 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2158 /* Can we merge the block to our big extent? */
2159 if (lblk == map->m_lblk + map->m_len &&
2160 (bh->b_state & BH_FLAGS) == map->m_flags) {
2168 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2170 * @mpd - extent of blocks for mapping
2171 * @head - the first buffer in the page
2172 * @bh - buffer we should start processing from
2173 * @lblk - logical number of the block in the file corresponding to @bh
2175 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2176 * the page for IO if all buffers in this page were mapped and there's no
2177 * accumulated extent of buffers to map or add buffers in the page to the
2178 * extent of buffers to map. The function returns 1 if the caller can continue
2179 * by processing the next page, 0 if it should stop adding buffers to the
2180 * extent to map because we cannot extend it anymore. It can also return value
2181 * < 0 in case of error during IO submission.
2183 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2184 struct buffer_head *head,
2185 struct buffer_head *bh,
2188 struct inode *inode = mpd->inode;
2190 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2191 >> inode->i_blkbits;
2193 if (ext4_verity_in_progress(inode))
2194 blocks = EXT_MAX_BLOCKS;
2197 BUG_ON(buffer_locked(bh));
2199 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2200 /* Found extent to map? */
2203 /* Buffer needs mapping and handle is not started? */
2206 /* Everything mapped so far and we hit EOF */
2209 } while (lblk++, (bh = bh->b_this_page) != head);
2210 /* So far everything mapped? Submit the page for IO. */
2211 if (mpd->map.m_len == 0) {
2212 err = mpage_submit_page(mpd, head->b_page);
2216 if (lblk >= blocks) {
2217 mpd->scanned_until_end = 1;
2224 * mpage_process_page - update page buffers corresponding to changed extent and
2225 * may submit fully mapped page for IO
2227 * @mpd - description of extent to map, on return next extent to map
2228 * @m_lblk - logical block mapping.
2229 * @m_pblk - corresponding physical mapping.
2230 * @map_bh - determines on return whether this page requires any further
2232 * Scan given page buffers corresponding to changed extent and update buffer
2233 * state according to new extent state.
2234 * We map delalloc buffers to their physical location, clear unwritten bits.
2235 * If the given page is not fully mapped, we update @map to the next extent in
2236 * the given page that needs mapping & return @map_bh as true.
2238 static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2239 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2242 struct buffer_head *head, *bh;
2243 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2244 ext4_lblk_t lblk = *m_lblk;
2245 ext4_fsblk_t pblock = *m_pblk;
2247 int blkbits = mpd->inode->i_blkbits;
2248 ssize_t io_end_size = 0;
2249 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2251 bh = head = page_buffers(page);
2253 if (lblk < mpd->map.m_lblk)
2255 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2257 * Buffer after end of mapped extent.
2258 * Find next buffer in the page to map.
2261 mpd->map.m_flags = 0;
2262 io_end_vec->size += io_end_size;
2265 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2268 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2269 io_end_vec = ext4_alloc_io_end_vec(io_end);
2270 if (IS_ERR(io_end_vec)) {
2271 err = PTR_ERR(io_end_vec);
2274 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2279 if (buffer_delay(bh)) {
2280 clear_buffer_delay(bh);
2281 bh->b_blocknr = pblock++;
2283 clear_buffer_unwritten(bh);
2284 io_end_size += (1 << blkbits);
2285 } while (lblk++, (bh = bh->b_this_page) != head);
2287 io_end_vec->size += io_end_size;
2297 * mpage_map_buffers - update buffers corresponding to changed extent and
2298 * submit fully mapped pages for IO
2300 * @mpd - description of extent to map, on return next extent to map
2302 * Scan buffers corresponding to changed extent (we expect corresponding pages
2303 * to be already locked) and update buffer state according to new extent state.
2304 * We map delalloc buffers to their physical location, clear unwritten bits,
2305 * and mark buffers as uninit when we perform writes to unwritten extents
2306 * and do extent conversion after IO is finished. If the last page is not fully
2307 * mapped, we update @map to the next extent in the last page that needs
2308 * mapping. Otherwise we submit the page for IO.
2310 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2312 struct pagevec pvec;
2314 struct inode *inode = mpd->inode;
2315 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2318 ext4_fsblk_t pblock;
2320 bool map_bh = false;
2322 start = mpd->map.m_lblk >> bpp_bits;
2323 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2324 lblk = start << bpp_bits;
2325 pblock = mpd->map.m_pblk;
2327 pagevec_init(&pvec);
2328 while (start <= end) {
2329 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2333 for (i = 0; i < nr_pages; i++) {
2334 struct page *page = pvec.pages[i];
2336 err = mpage_process_page(mpd, page, &lblk, &pblock,
2339 * If map_bh is true, means page may require further bh
2340 * mapping, or maybe the page was submitted for IO.
2341 * So we return to call further extent mapping.
2343 if (err < 0 || map_bh)
2345 /* Page fully mapped - let IO run! */
2346 err = mpage_submit_page(mpd, page);
2350 pagevec_release(&pvec);
2352 /* Extent fully mapped and matches with page boundary. We are done. */
2354 mpd->map.m_flags = 0;
2357 pagevec_release(&pvec);
2361 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2363 struct inode *inode = mpd->inode;
2364 struct ext4_map_blocks *map = &mpd->map;
2365 int get_blocks_flags;
2366 int err, dioread_nolock;
2368 trace_ext4_da_write_pages_extent(inode, map);
2370 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2371 * to convert an unwritten extent to be initialized (in the case
2372 * where we have written into one or more preallocated blocks). It is
2373 * possible that we're going to need more metadata blocks than
2374 * previously reserved. However we must not fail because we're in
2375 * writeback and there is nothing we can do about it so it might result
2376 * in data loss. So use reserved blocks to allocate metadata if
2379 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2380 * the blocks in question are delalloc blocks. This indicates
2381 * that the blocks and quotas has already been checked when
2382 * the data was copied into the page cache.
2384 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2385 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2386 EXT4_GET_BLOCKS_IO_SUBMIT;
2387 dioread_nolock = ext4_should_dioread_nolock(inode);
2389 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2390 if (map->m_flags & BIT(BH_Delay))
2391 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2393 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2396 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2397 if (!mpd->io_submit.io_end->handle &&
2398 ext4_handle_valid(handle)) {
2399 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2400 handle->h_rsv_handle = NULL;
2402 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2405 BUG_ON(map->m_len == 0);
2410 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2411 * mpd->len and submit pages underlying it for IO
2413 * @handle - handle for journal operations
2414 * @mpd - extent to map
2415 * @give_up_on_write - we set this to true iff there is a fatal error and there
2416 * is no hope of writing the data. The caller should discard
2417 * dirty pages to avoid infinite loops.
2419 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2420 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2421 * them to initialized or split the described range from larger unwritten
2422 * extent. Note that we need not map all the described range since allocation
2423 * can return less blocks or the range is covered by more unwritten extents. We
2424 * cannot map more because we are limited by reserved transaction credits. On
2425 * the other hand we always make sure that the last touched page is fully
2426 * mapped so that it can be written out (and thus forward progress is
2427 * guaranteed). After mapping we submit all mapped pages for IO.
2429 static int mpage_map_and_submit_extent(handle_t *handle,
2430 struct mpage_da_data *mpd,
2431 bool *give_up_on_write)
2433 struct inode *inode = mpd->inode;
2434 struct ext4_map_blocks *map = &mpd->map;
2438 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2439 struct ext4_io_end_vec *io_end_vec;
2441 io_end_vec = ext4_alloc_io_end_vec(io_end);
2442 if (IS_ERR(io_end_vec))
2443 return PTR_ERR(io_end_vec);
2444 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2446 err = mpage_map_one_extent(handle, mpd);
2448 struct super_block *sb = inode->i_sb;
2450 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2451 ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
2452 goto invalidate_dirty_pages;
2454 * Let the uper layers retry transient errors.
2455 * In the case of ENOSPC, if ext4_count_free_blocks()
2456 * is non-zero, a commit should free up blocks.
2458 if ((err == -ENOMEM) ||
2459 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2461 goto update_disksize;
2464 ext4_msg(sb, KERN_CRIT,
2465 "Delayed block allocation failed for "
2466 "inode %lu at logical offset %llu with"
2467 " max blocks %u with error %d",
2469 (unsigned long long)map->m_lblk,
2470 (unsigned)map->m_len, -err);
2471 ext4_msg(sb, KERN_CRIT,
2472 "This should not happen!! Data will "
2475 ext4_print_free_blocks(inode);
2476 invalidate_dirty_pages:
2477 *give_up_on_write = true;
2482 * Update buffer state, submit mapped pages, and get us new
2485 err = mpage_map_and_submit_buffers(mpd);
2487 goto update_disksize;
2488 } while (map->m_len);
2492 * Update on-disk size after IO is submitted. Races with
2493 * truncate are avoided by checking i_size under i_data_sem.
2495 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2496 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2500 down_write(&EXT4_I(inode)->i_data_sem);
2501 i_size = i_size_read(inode);
2502 if (disksize > i_size)
2504 if (disksize > EXT4_I(inode)->i_disksize)
2505 EXT4_I(inode)->i_disksize = disksize;
2506 up_write(&EXT4_I(inode)->i_data_sem);
2507 err2 = ext4_mark_inode_dirty(handle, inode);
2509 ext4_error_err(inode->i_sb, -err2,
2510 "Failed to mark inode %lu dirty",
2520 * Calculate the total number of credits to reserve for one writepages
2521 * iteration. This is called from ext4_writepages(). We map an extent of
2522 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2523 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2524 * bpp - 1 blocks in bpp different extents.
2526 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2528 int bpp = ext4_journal_blocks_per_page(inode);
2530 return ext4_meta_trans_blocks(inode,
2531 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2535 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2536 * and underlying extent to map
2538 * @mpd - where to look for pages
2540 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2541 * IO immediately. When we find a page which isn't mapped we start accumulating
2542 * extent of buffers underlying these pages that needs mapping (formed by
2543 * either delayed or unwritten buffers). We also lock the pages containing
2544 * these buffers. The extent found is returned in @mpd structure (starting at
2545 * mpd->lblk with length mpd->len blocks).
2547 * Note that this function can attach bios to one io_end structure which are
2548 * neither logically nor physically contiguous. Although it may seem as an
2549 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2550 * case as we need to track IO to all buffers underlying a page in one io_end.
2552 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2554 struct address_space *mapping = mpd->inode->i_mapping;
2555 struct pagevec pvec;
2556 unsigned int nr_pages;
2557 long left = mpd->wbc->nr_to_write;
2558 pgoff_t index = mpd->first_page;
2559 pgoff_t end = mpd->last_page;
2562 int blkbits = mpd->inode->i_blkbits;
2564 struct buffer_head *head;
2566 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2567 tag = PAGECACHE_TAG_TOWRITE;
2569 tag = PAGECACHE_TAG_DIRTY;
2571 pagevec_init(&pvec);
2573 mpd->next_page = index;
2574 while (index <= end) {
2575 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2580 for (i = 0; i < nr_pages; i++) {
2581 struct page *page = pvec.pages[i];
2584 * Accumulated enough dirty pages? This doesn't apply
2585 * to WB_SYNC_ALL mode. For integrity sync we have to
2586 * keep going because someone may be concurrently
2587 * dirtying pages, and we might have synced a lot of
2588 * newly appeared dirty pages, but have not synced all
2589 * of the old dirty pages.
2591 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2594 /* If we can't merge this page, we are done. */
2595 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2600 * If the page is no longer dirty, or its mapping no
2601 * longer corresponds to inode we are writing (which
2602 * means it has been truncated or invalidated), or the
2603 * page is already under writeback and we are not doing
2604 * a data integrity writeback, skip the page
2606 if (!PageDirty(page) ||
2607 (PageWriteback(page) &&
2608 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2609 unlikely(page->mapping != mapping)) {
2614 wait_on_page_writeback(page);
2615 BUG_ON(PageWriteback(page));
2617 if (mpd->map.m_len == 0)
2618 mpd->first_page = page->index;
2619 mpd->next_page = page->index + 1;
2620 /* Add all dirty buffers to mpd */
2621 lblk = ((ext4_lblk_t)page->index) <<
2622 (PAGE_SHIFT - blkbits);
2623 head = page_buffers(page);
2624 err = mpage_process_page_bufs(mpd, head, head, lblk);
2630 pagevec_release(&pvec);
2633 mpd->scanned_until_end = 1;
2636 pagevec_release(&pvec);
2640 static int ext4_writepages(struct address_space *mapping,
2641 struct writeback_control *wbc)
2643 pgoff_t writeback_index = 0;
2644 long nr_to_write = wbc->nr_to_write;
2645 int range_whole = 0;
2647 handle_t *handle = NULL;
2648 struct mpage_da_data mpd;
2649 struct inode *inode = mapping->host;
2650 int needed_blocks, rsv_blocks = 0, ret = 0;
2651 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2652 struct blk_plug plug;
2653 bool give_up_on_write = false;
2655 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2658 percpu_down_read(&sbi->s_writepages_rwsem);
2659 trace_ext4_writepages(inode, wbc);
2662 * No pages to write? This is mainly a kludge to avoid starting
2663 * a transaction for special inodes like journal inode on last iput()
2664 * because that could violate lock ordering on umount
2666 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2667 goto out_writepages;
2669 if (ext4_should_journal_data(inode)) {
2670 ret = generic_writepages(mapping, wbc);
2671 goto out_writepages;
2675 * If the filesystem has aborted, it is read-only, so return
2676 * right away instead of dumping stack traces later on that
2677 * will obscure the real source of the problem. We test
2678 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2679 * the latter could be true if the filesystem is mounted
2680 * read-only, and in that case, ext4_writepages should
2681 * *never* be called, so if that ever happens, we would want
2684 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2685 ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) {
2687 goto out_writepages;
2691 * If we have inline data and arrive here, it means that
2692 * we will soon create the block for the 1st page, so
2693 * we'd better clear the inline data here.
2695 if (ext4_has_inline_data(inode)) {
2696 /* Just inode will be modified... */
2697 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2698 if (IS_ERR(handle)) {
2699 ret = PTR_ERR(handle);
2700 goto out_writepages;
2702 BUG_ON(ext4_test_inode_state(inode,
2703 EXT4_STATE_MAY_INLINE_DATA));
2704 ext4_destroy_inline_data(handle, inode);
2705 ext4_journal_stop(handle);
2708 if (ext4_should_dioread_nolock(inode)) {
2710 * We may need to convert up to one extent per block in
2711 * the page and we may dirty the inode.
2713 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2714 PAGE_SIZE >> inode->i_blkbits);
2717 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2720 if (wbc->range_cyclic) {
2721 writeback_index = mapping->writeback_index;
2722 if (writeback_index)
2724 mpd.first_page = writeback_index;
2727 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2728 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2733 ext4_io_submit_init(&mpd.io_submit, wbc);
2735 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2736 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2737 blk_start_plug(&plug);
2740 * First writeback pages that don't need mapping - we can avoid
2741 * starting a transaction unnecessarily and also avoid being blocked
2742 * in the block layer on device congestion while having transaction
2746 mpd.scanned_until_end = 0;
2747 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2748 if (!mpd.io_submit.io_end) {
2752 ret = mpage_prepare_extent_to_map(&mpd);
2753 /* Unlock pages we didn't use */
2754 mpage_release_unused_pages(&mpd, false);
2755 /* Submit prepared bio */
2756 ext4_io_submit(&mpd.io_submit);
2757 ext4_put_io_end_defer(mpd.io_submit.io_end);
2758 mpd.io_submit.io_end = NULL;
2762 while (!mpd.scanned_until_end && wbc->nr_to_write > 0) {
2763 /* For each extent of pages we use new io_end */
2764 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2765 if (!mpd.io_submit.io_end) {
2771 * We have two constraints: We find one extent to map and we
2772 * must always write out whole page (makes a difference when
2773 * blocksize < pagesize) so that we don't block on IO when we
2774 * try to write out the rest of the page. Journalled mode is
2775 * not supported by delalloc.
2777 BUG_ON(ext4_should_journal_data(inode));
2778 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2780 /* start a new transaction */
2781 handle = ext4_journal_start_with_reserve(inode,
2782 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2783 if (IS_ERR(handle)) {
2784 ret = PTR_ERR(handle);
2785 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2786 "%ld pages, ino %lu; err %d", __func__,
2787 wbc->nr_to_write, inode->i_ino, ret);
2788 /* Release allocated io_end */
2789 ext4_put_io_end(mpd.io_submit.io_end);
2790 mpd.io_submit.io_end = NULL;
2795 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2796 ret = mpage_prepare_extent_to_map(&mpd);
2797 if (!ret && mpd.map.m_len)
2798 ret = mpage_map_and_submit_extent(handle, &mpd,
2801 * Caution: If the handle is synchronous,
2802 * ext4_journal_stop() can wait for transaction commit
2803 * to finish which may depend on writeback of pages to
2804 * complete or on page lock to be released. In that
2805 * case, we have to wait until after we have
2806 * submitted all the IO, released page locks we hold,
2807 * and dropped io_end reference (for extent conversion
2808 * to be able to complete) before stopping the handle.
2810 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2811 ext4_journal_stop(handle);
2815 /* Unlock pages we didn't use */
2816 mpage_release_unused_pages(&mpd, give_up_on_write);
2817 /* Submit prepared bio */
2818 ext4_io_submit(&mpd.io_submit);
2821 * Drop our io_end reference we got from init. We have
2822 * to be careful and use deferred io_end finishing if
2823 * we are still holding the transaction as we can
2824 * release the last reference to io_end which may end
2825 * up doing unwritten extent conversion.
2828 ext4_put_io_end_defer(mpd.io_submit.io_end);
2829 ext4_journal_stop(handle);
2831 ext4_put_io_end(mpd.io_submit.io_end);
2832 mpd.io_submit.io_end = NULL;
2834 if (ret == -ENOSPC && sbi->s_journal) {
2836 * Commit the transaction which would
2837 * free blocks released in the transaction
2840 jbd2_journal_force_commit_nested(sbi->s_journal);
2844 /* Fatal error - ENOMEM, EIO... */
2849 blk_finish_plug(&plug);
2850 if (!ret && !cycled && wbc->nr_to_write > 0) {
2852 mpd.last_page = writeback_index - 1;
2858 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2860 * Set the writeback_index so that range_cyclic
2861 * mode will write it back later
2863 mapping->writeback_index = mpd.first_page;
2866 trace_ext4_writepages_result(inode, wbc, ret,
2867 nr_to_write - wbc->nr_to_write);
2868 percpu_up_read(&sbi->s_writepages_rwsem);
2872 static int ext4_dax_writepages(struct address_space *mapping,
2873 struct writeback_control *wbc)
2876 long nr_to_write = wbc->nr_to_write;
2877 struct inode *inode = mapping->host;
2878 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2880 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2883 percpu_down_read(&sbi->s_writepages_rwsem);
2884 trace_ext4_writepages(inode, wbc);
2886 ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
2887 trace_ext4_writepages_result(inode, wbc, ret,
2888 nr_to_write - wbc->nr_to_write);
2889 percpu_up_read(&sbi->s_writepages_rwsem);
2893 static int ext4_nonda_switch(struct super_block *sb)
2895 s64 free_clusters, dirty_clusters;
2896 struct ext4_sb_info *sbi = EXT4_SB(sb);
2899 * switch to non delalloc mode if we are running low
2900 * on free block. The free block accounting via percpu
2901 * counters can get slightly wrong with percpu_counter_batch getting
2902 * accumulated on each CPU without updating global counters
2903 * Delalloc need an accurate free block accounting. So switch
2904 * to non delalloc when we are near to error range.
2907 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2909 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2911 * Start pushing delalloc when 1/2 of free blocks are dirty.
2913 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2914 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2916 if (2 * free_clusters < 3 * dirty_clusters ||
2917 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2919 * free block count is less than 150% of dirty blocks
2920 * or free blocks is less than watermark
2927 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2928 loff_t pos, unsigned len, unsigned flags,
2929 struct page **pagep, void **fsdata)
2931 int ret, retries = 0;
2934 struct inode *inode = mapping->host;
2936 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2939 index = pos >> PAGE_SHIFT;
2941 if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) ||
2942 ext4_verity_in_progress(inode)) {
2943 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2944 return ext4_write_begin(file, mapping, pos,
2945 len, flags, pagep, fsdata);
2947 *fsdata = (void *)0;
2948 trace_ext4_da_write_begin(inode, pos, len, flags);
2950 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2951 ret = ext4_da_write_inline_data_begin(mapping, inode,
2961 page = grab_cache_page_write_begin(mapping, index, flags);
2965 /* In case writeback began while the page was unlocked */
2966 wait_for_stable_page(page);
2968 #ifdef CONFIG_FS_ENCRYPTION
2969 ret = ext4_block_write_begin(page, pos, len,
2970 ext4_da_get_block_prep);
2972 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2978 * block_write_begin may have instantiated a few blocks
2979 * outside i_size. Trim these off again. Don't need
2980 * i_size_read because we hold inode lock.
2982 if (pos + len > inode->i_size)
2983 ext4_truncate_failed_write(inode);
2985 if (ret == -ENOSPC &&
2986 ext4_should_retry_alloc(inode->i_sb, &retries))
2996 * Check if we should update i_disksize
2997 * when write to the end of file but not require block allocation
2999 static int ext4_da_should_update_i_disksize(struct page *page,
3000 unsigned long offset)
3002 struct buffer_head *bh;
3003 struct inode *inode = page->mapping->host;
3007 bh = page_buffers(page);
3008 idx = offset >> inode->i_blkbits;
3010 for (i = 0; i < idx; i++)
3011 bh = bh->b_this_page;
3013 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3018 static int ext4_da_write_end(struct file *file,
3019 struct address_space *mapping,
3020 loff_t pos, unsigned len, unsigned copied,
3021 struct page *page, void *fsdata)
3023 struct inode *inode = mapping->host;
3025 unsigned long start, end;
3026 int write_mode = (int)(unsigned long)fsdata;
3028 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3029 return ext4_write_end(file, mapping, pos,
3030 len, copied, page, fsdata);
3032 trace_ext4_da_write_end(inode, pos, len, copied);
3034 if (write_mode != CONVERT_INLINE_DATA &&
3035 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3036 ext4_has_inline_data(inode))
3037 return ext4_write_inline_data_end(inode, pos, len, copied, page);
3039 start = pos & (PAGE_SIZE - 1);
3040 end = start + copied - 1;
3043 * Since we are holding inode lock, we are sure i_disksize <=
3044 * i_size. We also know that if i_disksize < i_size, there are
3045 * delalloc writes pending in the range upto i_size. If the end of
3046 * the current write is <= i_size, there's no need to touch
3047 * i_disksize since writeback will push i_disksize upto i_size
3048 * eventually. If the end of the current write is > i_size and
3049 * inside an allocated block (ext4_da_should_update_i_disksize()
3050 * check), we need to update i_disksize here as neither
3051 * ext4_writepage() nor certain ext4_writepages() paths not
3052 * allocating blocks update i_disksize.
3054 * Note that we defer inode dirtying to generic_write_end() /
3055 * ext4_da_write_inline_data_end().
3057 new_i_size = pos + copied;
3058 if (copied && new_i_size > inode->i_size &&
3059 ext4_da_should_update_i_disksize(page, end))
3060 ext4_update_i_disksize(inode, new_i_size);
3062 return generic_write_end(file, mapping, pos, len, copied, page, fsdata);
3066 * Force all delayed allocation blocks to be allocated for a given inode.
3068 int ext4_alloc_da_blocks(struct inode *inode)
3070 trace_ext4_alloc_da_blocks(inode);
3072 if (!EXT4_I(inode)->i_reserved_data_blocks)
3076 * We do something simple for now. The filemap_flush() will
3077 * also start triggering a write of the data blocks, which is
3078 * not strictly speaking necessary (and for users of
3079 * laptop_mode, not even desirable). However, to do otherwise
3080 * would require replicating code paths in:
3082 * ext4_writepages() ->
3083 * write_cache_pages() ---> (via passed in callback function)
3084 * __mpage_da_writepage() -->
3085 * mpage_add_bh_to_extent()
3086 * mpage_da_map_blocks()
3088 * The problem is that write_cache_pages(), located in
3089 * mm/page-writeback.c, marks pages clean in preparation for
3090 * doing I/O, which is not desirable if we're not planning on
3093 * We could call write_cache_pages(), and then redirty all of
3094 * the pages by calling redirty_page_for_writepage() but that
3095 * would be ugly in the extreme. So instead we would need to
3096 * replicate parts of the code in the above functions,
3097 * simplifying them because we wouldn't actually intend to
3098 * write out the pages, but rather only collect contiguous
3099 * logical block extents, call the multi-block allocator, and
3100 * then update the buffer heads with the block allocations.
3102 * For now, though, we'll cheat by calling filemap_flush(),
3103 * which will map the blocks, and start the I/O, but not
3104 * actually wait for the I/O to complete.
3106 return filemap_flush(inode->i_mapping);
3110 * bmap() is special. It gets used by applications such as lilo and by
3111 * the swapper to find the on-disk block of a specific piece of data.
3113 * Naturally, this is dangerous if the block concerned is still in the
3114 * journal. If somebody makes a swapfile on an ext4 data-journaling
3115 * filesystem and enables swap, then they may get a nasty shock when the
3116 * data getting swapped to that swapfile suddenly gets overwritten by
3117 * the original zero's written out previously to the journal and
3118 * awaiting writeback in the kernel's buffer cache.
3120 * So, if we see any bmap calls here on a modified, data-journaled file,
3121 * take extra steps to flush any blocks which might be in the cache.
3123 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3125 struct inode *inode = mapping->host;
3130 * We can get here for an inline file via the FIBMAP ioctl
3132 if (ext4_has_inline_data(inode))
3135 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3136 test_opt(inode->i_sb, DELALLOC)) {
3138 * With delalloc we want to sync the file
3139 * so that we can make sure we allocate
3142 filemap_write_and_wait(mapping);
3145 if (EXT4_JOURNAL(inode) &&
3146 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3148 * This is a REALLY heavyweight approach, but the use of
3149 * bmap on dirty files is expected to be extremely rare:
3150 * only if we run lilo or swapon on a freshly made file
3151 * do we expect this to happen.
3153 * (bmap requires CAP_SYS_RAWIO so this does not
3154 * represent an unprivileged user DOS attack --- we'd be
3155 * in trouble if mortal users could trigger this path at
3158 * NB. EXT4_STATE_JDATA is not set on files other than
3159 * regular files. If somebody wants to bmap a directory
3160 * or symlink and gets confused because the buffer
3161 * hasn't yet been flushed to disk, they deserve
3162 * everything they get.
3165 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3166 journal = EXT4_JOURNAL(inode);
3167 jbd2_journal_lock_updates(journal);
3168 err = jbd2_journal_flush(journal, 0);
3169 jbd2_journal_unlock_updates(journal);
3175 return iomap_bmap(mapping, block, &ext4_iomap_ops);
3178 static int ext4_readpage(struct file *file, struct page *page)
3181 struct inode *inode = page->mapping->host;
3183 trace_ext4_readpage(page);
3185 if (ext4_has_inline_data(inode))
3186 ret = ext4_readpage_inline(inode, page);
3189 return ext4_mpage_readpages(inode, NULL, page);
3194 static void ext4_readahead(struct readahead_control *rac)
3196 struct inode *inode = rac->mapping->host;
3198 /* If the file has inline data, no need to do readahead. */
3199 if (ext4_has_inline_data(inode))
3202 ext4_mpage_readpages(inode, rac, NULL);
3205 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3206 unsigned int length)
3208 trace_ext4_invalidatepage(page, offset, length);
3210 /* No journalling happens on data buffers when this function is used */
3211 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3213 block_invalidatepage(page, offset, length);
3216 static int __ext4_journalled_invalidatepage(struct page *page,
3217 unsigned int offset,
3218 unsigned int length)
3220 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3222 trace_ext4_journalled_invalidatepage(page, offset, length);
3225 * If it's a full truncate we just forget about the pending dirtying
3227 if (offset == 0 && length == PAGE_SIZE)
3228 ClearPageChecked(page);
3230 return jbd2_journal_invalidatepage(journal, page, offset, length);
3233 /* Wrapper for aops... */
3234 static void ext4_journalled_invalidatepage(struct page *page,
3235 unsigned int offset,
3236 unsigned int length)
3238 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3241 static int ext4_releasepage(struct page *page, gfp_t wait)
3243 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3245 trace_ext4_releasepage(page);
3247 /* Page has dirty journalled data -> cannot release */
3248 if (PageChecked(page))
3251 return jbd2_journal_try_to_free_buffers(journal, page);
3253 return try_to_free_buffers(page);
3256 static bool ext4_inode_datasync_dirty(struct inode *inode)
3258 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3261 if (jbd2_transaction_committed(journal,
3262 EXT4_I(inode)->i_datasync_tid))
3264 if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3265 return !list_empty(&EXT4_I(inode)->i_fc_list);
3269 /* Any metadata buffers to write? */
3270 if (!list_empty(&inode->i_mapping->private_list))
3272 return inode->i_state & I_DIRTY_DATASYNC;
3275 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3276 struct ext4_map_blocks *map, loff_t offset,
3279 u8 blkbits = inode->i_blkbits;
3282 * Writes that span EOF might trigger an I/O size update on completion,
3283 * so consider them to be dirty for the purpose of O_DSYNC, even if
3284 * there is no other metadata changes being made or are pending.
3287 if (ext4_inode_datasync_dirty(inode) ||
3288 offset + length > i_size_read(inode))
3289 iomap->flags |= IOMAP_F_DIRTY;
3291 if (map->m_flags & EXT4_MAP_NEW)
3292 iomap->flags |= IOMAP_F_NEW;
3294 iomap->bdev = inode->i_sb->s_bdev;
3295 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3296 iomap->offset = (u64) map->m_lblk << blkbits;
3297 iomap->length = (u64) map->m_len << blkbits;
3299 if ((map->m_flags & EXT4_MAP_MAPPED) &&
3300 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3301 iomap->flags |= IOMAP_F_MERGED;
3304 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3305 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3306 * set. In order for any allocated unwritten extents to be converted
3307 * into written extents correctly within the ->end_io() handler, we
3308 * need to ensure that the iomap->type is set appropriately. Hence, the
3309 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3312 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3313 iomap->type = IOMAP_UNWRITTEN;
3314 iomap->addr = (u64) map->m_pblk << blkbits;
3315 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3316 iomap->type = IOMAP_MAPPED;
3317 iomap->addr = (u64) map->m_pblk << blkbits;
3319 iomap->type = IOMAP_HOLE;
3320 iomap->addr = IOMAP_NULL_ADDR;
3324 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3328 u8 blkbits = inode->i_blkbits;
3329 int ret, dio_credits, m_flags = 0, retries = 0;
3332 * Trim the mapping request to the maximum value that we can map at
3333 * once for direct I/O.
3335 if (map->m_len > DIO_MAX_BLOCKS)
3336 map->m_len = DIO_MAX_BLOCKS;
3337 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3341 * Either we allocate blocks and then don't get an unwritten extent, so
3342 * in that case we have reserved enough credits. Or, the blocks are
3343 * already allocated and unwritten. In that case, the extent conversion
3344 * fits into the credits as well.
3346 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3348 return PTR_ERR(handle);
3351 * DAX and direct I/O are the only two operations that are currently
3352 * supported with IOMAP_WRITE.
3354 WARN_ON(!IS_DAX(inode) && !(flags & IOMAP_DIRECT));
3356 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3358 * We use i_size instead of i_disksize here because delalloc writeback
3359 * can complete at any point during the I/O and subsequently push the
3360 * i_disksize out to i_size. This could be beyond where direct I/O is
3361 * happening and thus expose allocated blocks to direct I/O reads.
3363 else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3364 m_flags = EXT4_GET_BLOCKS_CREATE;
3365 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3366 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3368 ret = ext4_map_blocks(handle, inode, map, m_flags);
3371 * We cannot fill holes in indirect tree based inodes as that could
3372 * expose stale data in the case of a crash. Use the magic error code
3373 * to fallback to buffered I/O.
3375 if (!m_flags && !ret)
3378 ext4_journal_stop(handle);
3379 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3386 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3387 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3390 struct ext4_map_blocks map;
3391 u8 blkbits = inode->i_blkbits;
3393 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3396 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3400 * Calculate the first and last logical blocks respectively.
3402 map.m_lblk = offset >> blkbits;
3403 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3404 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3406 if (flags & IOMAP_WRITE) {
3408 * We check here if the blocks are already allocated, then we
3409 * don't need to start a journal txn and we can directly return
3410 * the mapping information. This could boost performance
3411 * especially in multi-threaded overwrite requests.
3413 if (offset + length <= i_size_read(inode)) {
3414 ret = ext4_map_blocks(NULL, inode, &map, 0);
3415 if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3418 ret = ext4_iomap_alloc(inode, &map, flags);
3420 ret = ext4_map_blocks(NULL, inode, &map, 0);
3426 ext4_set_iomap(inode, iomap, &map, offset, length);
3431 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3432 loff_t length, unsigned flags, struct iomap *iomap,
3433 struct iomap *srcmap)
3438 * Even for writes we don't need to allocate blocks, so just pretend
3439 * we are reading to save overhead of starting a transaction.
3441 flags &= ~IOMAP_WRITE;
3442 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3443 WARN_ON_ONCE(iomap->type != IOMAP_MAPPED);
3447 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3448 ssize_t written, unsigned flags, struct iomap *iomap)
3451 * Check to see whether an error occurred while writing out the data to
3452 * the allocated blocks. If so, return the magic error code so that we
3453 * fallback to buffered I/O and attempt to complete the remainder of
3454 * the I/O. Any blocks that may have been allocated in preparation for
3455 * the direct I/O will be reused during buffered I/O.
3457 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3463 const struct iomap_ops ext4_iomap_ops = {
3464 .iomap_begin = ext4_iomap_begin,
3465 .iomap_end = ext4_iomap_end,
3468 const struct iomap_ops ext4_iomap_overwrite_ops = {
3469 .iomap_begin = ext4_iomap_overwrite_begin,
3470 .iomap_end = ext4_iomap_end,
3473 static bool ext4_iomap_is_delalloc(struct inode *inode,
3474 struct ext4_map_blocks *map)
3476 struct extent_status es;
3477 ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3479 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3480 map->m_lblk, end, &es);
3482 if (!es.es_len || es.es_lblk > end)
3485 if (es.es_lblk > map->m_lblk) {
3486 map->m_len = es.es_lblk - map->m_lblk;
3490 offset = map->m_lblk - es.es_lblk;
3491 map->m_len = es.es_len - offset;
3496 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3497 loff_t length, unsigned int flags,
3498 struct iomap *iomap, struct iomap *srcmap)
3501 bool delalloc = false;
3502 struct ext4_map_blocks map;
3503 u8 blkbits = inode->i_blkbits;
3505 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3508 if (ext4_has_inline_data(inode)) {
3509 ret = ext4_inline_data_iomap(inode, iomap);
3510 if (ret != -EAGAIN) {
3511 if (ret == 0 && offset >= iomap->length)
3518 * Calculate the first and last logical block respectively.
3520 map.m_lblk = offset >> blkbits;
3521 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3522 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3525 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3526 * So handle it here itself instead of querying ext4_map_blocks().
3527 * Since ext4_map_blocks() will warn about it and will return
3530 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3531 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3533 if (offset >= sbi->s_bitmap_maxbytes) {
3539 ret = ext4_map_blocks(NULL, inode, &map, 0);
3543 delalloc = ext4_iomap_is_delalloc(inode, &map);
3546 ext4_set_iomap(inode, iomap, &map, offset, length);
3547 if (delalloc && iomap->type == IOMAP_HOLE)
3548 iomap->type = IOMAP_DELALLOC;
3553 const struct iomap_ops ext4_iomap_report_ops = {
3554 .iomap_begin = ext4_iomap_begin_report,
3558 * Pages can be marked dirty completely asynchronously from ext4's journalling
3559 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3560 * much here because ->set_page_dirty is called under VFS locks. The page is
3561 * not necessarily locked.
3563 * We cannot just dirty the page and leave attached buffers clean, because the
3564 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3565 * or jbddirty because all the journalling code will explode.
3567 * So what we do is to mark the page "pending dirty" and next time writepage
3568 * is called, propagate that into the buffers appropriately.
3570 static int ext4_journalled_set_page_dirty(struct page *page)
3572 SetPageChecked(page);
3573 return __set_page_dirty_nobuffers(page);
3576 static int ext4_set_page_dirty(struct page *page)
3578 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3579 WARN_ON_ONCE(!page_has_buffers(page));
3580 return __set_page_dirty_buffers(page);
3583 static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3584 struct file *file, sector_t *span)
3586 return iomap_swapfile_activate(sis, file, span,
3587 &ext4_iomap_report_ops);
3590 static const struct address_space_operations ext4_aops = {
3591 .readpage = ext4_readpage,
3592 .readahead = ext4_readahead,
3593 .writepage = ext4_writepage,
3594 .writepages = ext4_writepages,
3595 .write_begin = ext4_write_begin,
3596 .write_end = ext4_write_end,
3597 .set_page_dirty = ext4_set_page_dirty,
3599 .invalidatepage = ext4_invalidatepage,
3600 .releasepage = ext4_releasepage,
3601 .direct_IO = noop_direct_IO,
3602 .migratepage = buffer_migrate_page,
3603 .is_partially_uptodate = block_is_partially_uptodate,
3604 .error_remove_page = generic_error_remove_page,
3605 .swap_activate = ext4_iomap_swap_activate,
3608 static const struct address_space_operations ext4_journalled_aops = {
3609 .readpage = ext4_readpage,
3610 .readahead = ext4_readahead,
3611 .writepage = ext4_writepage,
3612 .writepages = ext4_writepages,
3613 .write_begin = ext4_write_begin,
3614 .write_end = ext4_journalled_write_end,
3615 .set_page_dirty = ext4_journalled_set_page_dirty,
3617 .invalidatepage = ext4_journalled_invalidatepage,
3618 .releasepage = ext4_releasepage,
3619 .direct_IO = noop_direct_IO,
3620 .is_partially_uptodate = block_is_partially_uptodate,
3621 .error_remove_page = generic_error_remove_page,
3622 .swap_activate = ext4_iomap_swap_activate,
3625 static const struct address_space_operations ext4_da_aops = {
3626 .readpage = ext4_readpage,
3627 .readahead = ext4_readahead,
3628 .writepage = ext4_writepage,
3629 .writepages = ext4_writepages,
3630 .write_begin = ext4_da_write_begin,
3631 .write_end = ext4_da_write_end,
3632 .set_page_dirty = ext4_set_page_dirty,
3634 .invalidatepage = ext4_invalidatepage,
3635 .releasepage = ext4_releasepage,
3636 .direct_IO = noop_direct_IO,
3637 .migratepage = buffer_migrate_page,
3638 .is_partially_uptodate = block_is_partially_uptodate,
3639 .error_remove_page = generic_error_remove_page,
3640 .swap_activate = ext4_iomap_swap_activate,
3643 static const struct address_space_operations ext4_dax_aops = {
3644 .writepages = ext4_dax_writepages,
3645 .direct_IO = noop_direct_IO,
3646 .set_page_dirty = __set_page_dirty_no_writeback,
3648 .invalidatepage = noop_invalidatepage,
3649 .swap_activate = ext4_iomap_swap_activate,
3652 void ext4_set_aops(struct inode *inode)
3654 switch (ext4_inode_journal_mode(inode)) {
3655 case EXT4_INODE_ORDERED_DATA_MODE:
3656 case EXT4_INODE_WRITEBACK_DATA_MODE:
3658 case EXT4_INODE_JOURNAL_DATA_MODE:
3659 inode->i_mapping->a_ops = &ext4_journalled_aops;
3665 inode->i_mapping->a_ops = &ext4_dax_aops;
3666 else if (test_opt(inode->i_sb, DELALLOC))
3667 inode->i_mapping->a_ops = &ext4_da_aops;
3669 inode->i_mapping->a_ops = &ext4_aops;
3672 static int __ext4_block_zero_page_range(handle_t *handle,
3673 struct address_space *mapping, loff_t from, loff_t length)
3675 ext4_fsblk_t index = from >> PAGE_SHIFT;
3676 unsigned offset = from & (PAGE_SIZE-1);
3677 unsigned blocksize, pos;
3679 struct inode *inode = mapping->host;
3680 struct buffer_head *bh;
3684 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3685 mapping_gfp_constraint(mapping, ~__GFP_FS));
3689 blocksize = inode->i_sb->s_blocksize;
3691 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3693 if (!page_has_buffers(page))
3694 create_empty_buffers(page, blocksize, 0);
3696 /* Find the buffer that contains "offset" */
3697 bh = page_buffers(page);
3699 while (offset >= pos) {
3700 bh = bh->b_this_page;
3704 if (buffer_freed(bh)) {
3705 BUFFER_TRACE(bh, "freed: skip");
3708 if (!buffer_mapped(bh)) {
3709 BUFFER_TRACE(bh, "unmapped");
3710 ext4_get_block(inode, iblock, bh, 0);
3711 /* unmapped? It's a hole - nothing to do */
3712 if (!buffer_mapped(bh)) {
3713 BUFFER_TRACE(bh, "still unmapped");
3718 /* Ok, it's mapped. Make sure it's up-to-date */
3719 if (PageUptodate(page))
3720 set_buffer_uptodate(bh);
3722 if (!buffer_uptodate(bh)) {
3723 err = ext4_read_bh_lock(bh, 0, true);
3726 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3727 /* We expect the key to be set. */
3728 BUG_ON(!fscrypt_has_encryption_key(inode));
3729 err = fscrypt_decrypt_pagecache_blocks(page, blocksize,
3732 clear_buffer_uptodate(bh);
3737 if (ext4_should_journal_data(inode)) {
3738 BUFFER_TRACE(bh, "get write access");
3739 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
3744 zero_user(page, offset, length);
3745 BUFFER_TRACE(bh, "zeroed end of block");
3747 if (ext4_should_journal_data(inode)) {
3748 err = ext4_handle_dirty_metadata(handle, inode, bh);
3751 mark_buffer_dirty(bh);
3752 if (ext4_should_order_data(inode))
3753 err = ext4_jbd2_inode_add_write(handle, inode, from,
3764 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3765 * starting from file offset 'from'. The range to be zero'd must
3766 * be contained with in one block. If the specified range exceeds
3767 * the end of the block it will be shortened to end of the block
3768 * that corresponds to 'from'
3770 static int ext4_block_zero_page_range(handle_t *handle,
3771 struct address_space *mapping, loff_t from, loff_t length)
3773 struct inode *inode = mapping->host;
3774 unsigned offset = from & (PAGE_SIZE-1);
3775 unsigned blocksize = inode->i_sb->s_blocksize;
3776 unsigned max = blocksize - (offset & (blocksize - 1));
3779 * correct length if it does not fall between
3780 * 'from' and the end of the block
3782 if (length > max || length < 0)
3785 if (IS_DAX(inode)) {
3786 return iomap_zero_range(inode, from, length, NULL,
3789 return __ext4_block_zero_page_range(handle, mapping, from, length);
3793 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3794 * up to the end of the block which corresponds to `from'.
3795 * This required during truncate. We need to physically zero the tail end
3796 * of that block so it doesn't yield old data if the file is later grown.
3798 static int ext4_block_truncate_page(handle_t *handle,
3799 struct address_space *mapping, loff_t from)
3801 unsigned offset = from & (PAGE_SIZE-1);
3804 struct inode *inode = mapping->host;
3806 /* If we are processing an encrypted inode during orphan list handling */
3807 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3810 blocksize = inode->i_sb->s_blocksize;
3811 length = blocksize - (offset & (blocksize - 1));
3813 return ext4_block_zero_page_range(handle, mapping, from, length);
3816 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3817 loff_t lstart, loff_t length)
3819 struct super_block *sb = inode->i_sb;
3820 struct address_space *mapping = inode->i_mapping;
3821 unsigned partial_start, partial_end;
3822 ext4_fsblk_t start, end;
3823 loff_t byte_end = (lstart + length - 1);
3826 partial_start = lstart & (sb->s_blocksize - 1);
3827 partial_end = byte_end & (sb->s_blocksize - 1);
3829 start = lstart >> sb->s_blocksize_bits;
3830 end = byte_end >> sb->s_blocksize_bits;
3832 /* Handle partial zero within the single block */
3834 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3835 err = ext4_block_zero_page_range(handle, mapping,
3839 /* Handle partial zero out on the start of the range */
3840 if (partial_start) {
3841 err = ext4_block_zero_page_range(handle, mapping,
3842 lstart, sb->s_blocksize);
3846 /* Handle partial zero out on the end of the range */
3847 if (partial_end != sb->s_blocksize - 1)
3848 err = ext4_block_zero_page_range(handle, mapping,
3849 byte_end - partial_end,
3854 int ext4_can_truncate(struct inode *inode)
3856 if (S_ISREG(inode->i_mode))
3858 if (S_ISDIR(inode->i_mode))
3860 if (S_ISLNK(inode->i_mode))
3861 return !ext4_inode_is_fast_symlink(inode);
3866 * We have to make sure i_disksize gets properly updated before we truncate
3867 * page cache due to hole punching or zero range. Otherwise i_disksize update
3868 * can get lost as it may have been postponed to submission of writeback but
3869 * that will never happen after we truncate page cache.
3871 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3877 loff_t size = i_size_read(inode);
3879 WARN_ON(!inode_is_locked(inode));
3880 if (offset > size || offset + len < size)
3883 if (EXT4_I(inode)->i_disksize >= size)
3886 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3888 return PTR_ERR(handle);
3889 ext4_update_i_disksize(inode, size);
3890 ret = ext4_mark_inode_dirty(handle, inode);
3891 ext4_journal_stop(handle);
3896 static void ext4_wait_dax_page(struct inode *inode)
3898 filemap_invalidate_unlock(inode->i_mapping);
3900 filemap_invalidate_lock(inode->i_mapping);
3903 int ext4_break_layouts(struct inode *inode)
3908 if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
3912 page = dax_layout_busy_page(inode->i_mapping);
3916 error = ___wait_var_event(&page->_refcount,
3917 atomic_read(&page->_refcount) == 1,
3918 TASK_INTERRUPTIBLE, 0, 0,
3919 ext4_wait_dax_page(inode));
3920 } while (error == 0);
3926 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3927 * associated with the given offset and length
3929 * @inode: File inode
3930 * @offset: The offset where the hole will begin
3931 * @len: The length of the hole
3933 * Returns: 0 on success or negative on failure
3936 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3938 struct super_block *sb = inode->i_sb;
3939 ext4_lblk_t first_block, stop_block;
3940 struct address_space *mapping = inode->i_mapping;
3941 loff_t first_block_offset, last_block_offset;
3943 unsigned int credits;
3944 int ret = 0, ret2 = 0;
3946 trace_ext4_punch_hole(inode, offset, length, 0);
3948 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
3949 if (ext4_has_inline_data(inode)) {
3950 filemap_invalidate_lock(mapping);
3951 ret = ext4_convert_inline_data(inode);
3952 filemap_invalidate_unlock(mapping);
3958 * Write out all dirty pages to avoid race conditions
3959 * Then release them.
3961 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3962 ret = filemap_write_and_wait_range(mapping, offset,
3963 offset + length - 1);
3970 /* No need to punch hole beyond i_size */
3971 if (offset >= inode->i_size)
3975 * If the hole extends beyond i_size, set the hole
3976 * to end after the page that contains i_size
3978 if (offset + length > inode->i_size) {
3979 length = inode->i_size +
3980 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3984 if (offset & (sb->s_blocksize - 1) ||
3985 (offset + length) & (sb->s_blocksize - 1)) {
3987 * Attach jinode to inode for jbd2 if we do any zeroing of
3990 ret = ext4_inode_attach_jinode(inode);
3996 /* Wait all existing dio workers, newcomers will block on i_mutex */
3997 inode_dio_wait(inode);
4000 * Prevent page faults from reinstantiating pages we have released from
4003 filemap_invalidate_lock(mapping);
4005 ret = ext4_break_layouts(inode);
4009 first_block_offset = round_up(offset, sb->s_blocksize);
4010 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4012 /* Now release the pages and zero block aligned part of pages*/
4013 if (last_block_offset > first_block_offset) {
4014 ret = ext4_update_disksize_before_punch(inode, offset, length);
4017 truncate_pagecache_range(inode, first_block_offset,
4021 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4022 credits = ext4_writepage_trans_blocks(inode);
4024 credits = ext4_blocks_for_truncate(inode);
4025 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4026 if (IS_ERR(handle)) {
4027 ret = PTR_ERR(handle);
4028 ext4_std_error(sb, ret);
4032 ret = ext4_zero_partial_blocks(handle, inode, offset,
4037 first_block = (offset + sb->s_blocksize - 1) >>
4038 EXT4_BLOCK_SIZE_BITS(sb);
4039 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4041 /* If there are blocks to remove, do it */
4042 if (stop_block > first_block) {
4044 down_write(&EXT4_I(inode)->i_data_sem);
4045 ext4_discard_preallocations(inode, 0);
4047 ret = ext4_es_remove_extent(inode, first_block,
4048 stop_block - first_block);
4050 up_write(&EXT4_I(inode)->i_data_sem);
4054 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4055 ret = ext4_ext_remove_space(inode, first_block,
4058 ret = ext4_ind_remove_space(handle, inode, first_block,
4061 up_write(&EXT4_I(inode)->i_data_sem);
4063 ext4_fc_track_range(handle, inode, first_block, stop_block);
4065 ext4_handle_sync(handle);
4067 inode->i_mtime = inode->i_ctime = current_time(inode);
4068 ret2 = ext4_mark_inode_dirty(handle, inode);
4072 ext4_update_inode_fsync_trans(handle, inode, 1);
4074 ext4_journal_stop(handle);
4076 filemap_invalidate_unlock(mapping);
4078 inode_unlock(inode);
4082 int ext4_inode_attach_jinode(struct inode *inode)
4084 struct ext4_inode_info *ei = EXT4_I(inode);
4085 struct jbd2_inode *jinode;
4087 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4090 jinode = jbd2_alloc_inode(GFP_KERNEL);
4091 spin_lock(&inode->i_lock);
4094 spin_unlock(&inode->i_lock);
4097 ei->jinode = jinode;
4098 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4101 spin_unlock(&inode->i_lock);
4102 if (unlikely(jinode != NULL))
4103 jbd2_free_inode(jinode);
4110 * We block out ext4_get_block() block instantiations across the entire
4111 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4112 * simultaneously on behalf of the same inode.
4114 * As we work through the truncate and commit bits of it to the journal there
4115 * is one core, guiding principle: the file's tree must always be consistent on
4116 * disk. We must be able to restart the truncate after a crash.
4118 * The file's tree may be transiently inconsistent in memory (although it
4119 * probably isn't), but whenever we close off and commit a journal transaction,
4120 * the contents of (the filesystem + the journal) must be consistent and
4121 * restartable. It's pretty simple, really: bottom up, right to left (although
4122 * left-to-right works OK too).
4124 * Note that at recovery time, journal replay occurs *before* the restart of
4125 * truncate against the orphan inode list.
4127 * The committed inode has the new, desired i_size (which is the same as
4128 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4129 * that this inode's truncate did not complete and it will again call
4130 * ext4_truncate() to have another go. So there will be instantiated blocks
4131 * to the right of the truncation point in a crashed ext4 filesystem. But
4132 * that's fine - as long as they are linked from the inode, the post-crash
4133 * ext4_truncate() run will find them and release them.
4135 int ext4_truncate(struct inode *inode)
4137 struct ext4_inode_info *ei = EXT4_I(inode);
4138 unsigned int credits;
4141 struct address_space *mapping = inode->i_mapping;
4144 * There is a possibility that we're either freeing the inode
4145 * or it's a completely new inode. In those cases we might not
4146 * have i_mutex locked because it's not necessary.
4148 if (!(inode->i_state & (I_NEW|I_FREEING)))
4149 WARN_ON(!inode_is_locked(inode));
4150 trace_ext4_truncate_enter(inode);
4152 if (!ext4_can_truncate(inode))
4155 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4156 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4158 if (ext4_has_inline_data(inode)) {
4161 err = ext4_inline_data_truncate(inode, &has_inline);
4162 if (err || has_inline)
4166 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4167 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4168 if (ext4_inode_attach_jinode(inode) < 0)
4172 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4173 credits = ext4_writepage_trans_blocks(inode);
4175 credits = ext4_blocks_for_truncate(inode);
4177 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4178 if (IS_ERR(handle)) {
4179 err = PTR_ERR(handle);
4183 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4184 ext4_block_truncate_page(handle, mapping, inode->i_size);
4187 * We add the inode to the orphan list, so that if this
4188 * truncate spans multiple transactions, and we crash, we will
4189 * resume the truncate when the filesystem recovers. It also
4190 * marks the inode dirty, to catch the new size.
4192 * Implication: the file must always be in a sane, consistent
4193 * truncatable state while each transaction commits.
4195 err = ext4_orphan_add(handle, inode);
4199 down_write(&EXT4_I(inode)->i_data_sem);
4201 ext4_discard_preallocations(inode, 0);
4203 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4204 err = ext4_ext_truncate(handle, inode);
4206 ext4_ind_truncate(handle, inode);
4208 up_write(&ei->i_data_sem);
4213 ext4_handle_sync(handle);
4217 * If this was a simple ftruncate() and the file will remain alive,
4218 * then we need to clear up the orphan record which we created above.
4219 * However, if this was a real unlink then we were called by
4220 * ext4_evict_inode(), and we allow that function to clean up the
4221 * orphan info for us.
4224 ext4_orphan_del(handle, inode);
4226 inode->i_mtime = inode->i_ctime = current_time(inode);
4227 err2 = ext4_mark_inode_dirty(handle, inode);
4228 if (unlikely(err2 && !err))
4230 ext4_journal_stop(handle);
4233 trace_ext4_truncate_exit(inode);
4238 * ext4_get_inode_loc returns with an extra refcount against the inode's
4239 * underlying buffer_head on success. If 'in_mem' is true, we have all
4240 * data in memory that is needed to recreate the on-disk version of this
4243 static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4244 struct ext4_iloc *iloc, int in_mem,
4245 ext4_fsblk_t *ret_block)
4247 struct ext4_group_desc *gdp;
4248 struct buffer_head *bh;
4250 struct blk_plug plug;
4251 int inodes_per_block, inode_offset;
4254 if (ino < EXT4_ROOT_INO ||
4255 ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4256 return -EFSCORRUPTED;
4258 iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4259 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4264 * Figure out the offset within the block group inode table
4266 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4267 inode_offset = ((ino - 1) %
4268 EXT4_INODES_PER_GROUP(sb));
4269 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4270 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4272 bh = sb_getblk(sb, block);
4275 if (ext4_buffer_uptodate(bh))
4279 if (ext4_buffer_uptodate(bh)) {
4280 /* Someone brought it uptodate while we waited */
4286 * If we have all information of the inode in memory and this
4287 * is the only valid inode in the block, we need not read the
4291 struct buffer_head *bitmap_bh;
4294 start = inode_offset & ~(inodes_per_block - 1);
4296 /* Is the inode bitmap in cache? */
4297 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4298 if (unlikely(!bitmap_bh))
4302 * If the inode bitmap isn't in cache then the
4303 * optimisation may end up performing two reads instead
4304 * of one, so skip it.
4306 if (!buffer_uptodate(bitmap_bh)) {
4310 for (i = start; i < start + inodes_per_block; i++) {
4311 if (i == inode_offset)
4313 if (ext4_test_bit(i, bitmap_bh->b_data))
4317 if (i == start + inodes_per_block) {
4318 /* all other inodes are free, so skip I/O */
4319 memset(bh->b_data, 0, bh->b_size);
4320 set_buffer_uptodate(bh);
4328 * If we need to do any I/O, try to pre-readahead extra
4329 * blocks from the inode table.
4331 blk_start_plug(&plug);
4332 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4333 ext4_fsblk_t b, end, table;
4335 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4337 table = ext4_inode_table(sb, gdp);
4338 /* s_inode_readahead_blks is always a power of 2 */
4339 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4343 num = EXT4_INODES_PER_GROUP(sb);
4344 if (ext4_has_group_desc_csum(sb))
4345 num -= ext4_itable_unused_count(sb, gdp);
4346 table += num / inodes_per_block;
4350 ext4_sb_breadahead_unmovable(sb, b++);
4354 * There are other valid inodes in the buffer, this inode
4355 * has in-inode xattrs, or we don't have this inode in memory.
4356 * Read the block from disk.
4358 trace_ext4_load_inode(sb, ino);
4359 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4360 blk_finish_plug(&plug);
4362 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO);
4363 if (!buffer_uptodate(bh)) {
4374 static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4375 struct ext4_iloc *iloc)
4377 ext4_fsblk_t err_blk;
4380 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc, 0,
4384 ext4_error_inode_block(inode, err_blk, EIO,
4385 "unable to read itable block");
4390 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4392 ext4_fsblk_t err_blk;
4395 /* We have all inode data except xattrs in memory here. */
4396 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc,
4397 !ext4_test_inode_state(inode, EXT4_STATE_XATTR), &err_blk);
4400 ext4_error_inode_block(inode, err_blk, EIO,
4401 "unable to read itable block");
4407 int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4408 struct ext4_iloc *iloc)
4410 return __ext4_get_inode_loc(sb, ino, iloc, 0, NULL);
4413 static bool ext4_should_enable_dax(struct inode *inode)
4415 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4417 if (test_opt2(inode->i_sb, DAX_NEVER))
4419 if (!S_ISREG(inode->i_mode))
4421 if (ext4_should_journal_data(inode))
4423 if (ext4_has_inline_data(inode))
4425 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4427 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4429 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4431 if (test_opt(inode->i_sb, DAX_ALWAYS))
4434 return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4437 void ext4_set_inode_flags(struct inode *inode, bool init)
4439 unsigned int flags = EXT4_I(inode)->i_flags;
4440 unsigned int new_fl = 0;
4442 WARN_ON_ONCE(IS_DAX(inode) && init);
4444 if (flags & EXT4_SYNC_FL)
4446 if (flags & EXT4_APPEND_FL)
4448 if (flags & EXT4_IMMUTABLE_FL)
4449 new_fl |= S_IMMUTABLE;
4450 if (flags & EXT4_NOATIME_FL)
4451 new_fl |= S_NOATIME;
4452 if (flags & EXT4_DIRSYNC_FL)
4453 new_fl |= S_DIRSYNC;
4455 /* Because of the way inode_set_flags() works we must preserve S_DAX
4456 * here if already set. */
4457 new_fl |= (inode->i_flags & S_DAX);
4458 if (init && ext4_should_enable_dax(inode))
4461 if (flags & EXT4_ENCRYPT_FL)
4462 new_fl |= S_ENCRYPTED;
4463 if (flags & EXT4_CASEFOLD_FL)
4464 new_fl |= S_CASEFOLD;
4465 if (flags & EXT4_VERITY_FL)
4467 inode_set_flags(inode, new_fl,
4468 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4469 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4472 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4473 struct ext4_inode_info *ei)
4476 struct inode *inode = &(ei->vfs_inode);
4477 struct super_block *sb = inode->i_sb;
4479 if (ext4_has_feature_huge_file(sb)) {
4480 /* we are using combined 48 bit field */
4481 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4482 le32_to_cpu(raw_inode->i_blocks_lo);
4483 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4484 /* i_blocks represent file system block size */
4485 return i_blocks << (inode->i_blkbits - 9);
4490 return le32_to_cpu(raw_inode->i_blocks_lo);
4494 static inline int ext4_iget_extra_inode(struct inode *inode,
4495 struct ext4_inode *raw_inode,
4496 struct ext4_inode_info *ei)
4498 __le32 *magic = (void *)raw_inode +
4499 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4501 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4502 EXT4_INODE_SIZE(inode->i_sb) &&
4503 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4504 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4505 return ext4_find_inline_data_nolock(inode);
4507 EXT4_I(inode)->i_inline_off = 0;
4511 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4513 if (!ext4_has_feature_project(inode->i_sb))
4515 *projid = EXT4_I(inode)->i_projid;
4520 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4521 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4524 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4526 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4527 inode_set_iversion_raw(inode, val);
4529 inode_set_iversion_queried(inode, val);
4531 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4533 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4534 return inode_peek_iversion_raw(inode);
4536 return inode_peek_iversion(inode);
4539 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4540 ext4_iget_flags flags, const char *function,
4543 struct ext4_iloc iloc;
4544 struct ext4_inode *raw_inode;
4545 struct ext4_inode_info *ei;
4546 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4547 struct inode *inode;
4548 journal_t *journal = EXT4_SB(sb)->s_journal;
4556 if ((!(flags & EXT4_IGET_SPECIAL) &&
4557 ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
4558 ino == le32_to_cpu(es->s_usr_quota_inum) ||
4559 ino == le32_to_cpu(es->s_grp_quota_inum) ||
4560 ino == le32_to_cpu(es->s_prj_quota_inum) ||
4561 ino == le32_to_cpu(es->s_orphan_file_inum))) ||
4562 (ino < EXT4_ROOT_INO) ||
4563 (ino > le32_to_cpu(es->s_inodes_count))) {
4564 if (flags & EXT4_IGET_HANDLE)
4565 return ERR_PTR(-ESTALE);
4566 __ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4567 "inode #%lu: comm %s: iget: illegal inode #",
4568 ino, current->comm);
4569 return ERR_PTR(-EFSCORRUPTED);
4572 inode = iget_locked(sb, ino);
4574 return ERR_PTR(-ENOMEM);
4575 if (!(inode->i_state & I_NEW))
4581 ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4584 raw_inode = ext4_raw_inode(&iloc);
4586 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4587 ext4_error_inode(inode, function, line, 0,
4588 "iget: root inode unallocated");
4589 ret = -EFSCORRUPTED;
4593 if ((flags & EXT4_IGET_HANDLE) &&
4594 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4599 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4600 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4601 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4602 EXT4_INODE_SIZE(inode->i_sb) ||
4603 (ei->i_extra_isize & 3)) {
4604 ext4_error_inode(inode, function, line, 0,
4605 "iget: bad extra_isize %u "
4608 EXT4_INODE_SIZE(inode->i_sb));
4609 ret = -EFSCORRUPTED;
4613 ei->i_extra_isize = 0;
4615 /* Precompute checksum seed for inode metadata */
4616 if (ext4_has_metadata_csum(sb)) {
4617 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4619 __le32 inum = cpu_to_le32(inode->i_ino);
4620 __le32 gen = raw_inode->i_generation;
4621 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4623 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4627 if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4628 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4629 (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4630 ext4_error_inode_err(inode, function, line, 0,
4631 EFSBADCRC, "iget: checksum invalid");
4636 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4637 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4638 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4639 if (ext4_has_feature_project(sb) &&
4640 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4641 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4642 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4644 i_projid = EXT4_DEF_PROJID;
4646 if (!(test_opt(inode->i_sb, NO_UID32))) {
4647 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4648 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4650 i_uid_write(inode, i_uid);
4651 i_gid_write(inode, i_gid);
4652 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4653 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4655 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4656 ei->i_inline_off = 0;
4657 ei->i_dir_start_lookup = 0;
4658 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4659 /* We now have enough fields to check if the inode was active or not.
4660 * This is needed because nfsd might try to access dead inodes
4661 * the test is that same one that e2fsck uses
4662 * NeilBrown 1999oct15
4664 if (inode->i_nlink == 0) {
4665 if ((inode->i_mode == 0 ||
4666 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4667 ino != EXT4_BOOT_LOADER_INO) {
4668 /* this inode is deleted */
4672 /* The only unlinked inodes we let through here have
4673 * valid i_mode and are being read by the orphan
4674 * recovery code: that's fine, we're about to complete
4675 * the process of deleting those.
4676 * OR it is the EXT4_BOOT_LOADER_INO which is
4677 * not initialized on a new filesystem. */
4679 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4680 ext4_set_inode_flags(inode, true);
4681 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4682 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4683 if (ext4_has_feature_64bit(sb))
4685 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4686 inode->i_size = ext4_isize(sb, raw_inode);
4687 if ((size = i_size_read(inode)) < 0) {
4688 ext4_error_inode(inode, function, line, 0,
4689 "iget: bad i_size value: %lld", size);
4690 ret = -EFSCORRUPTED;
4694 * If dir_index is not enabled but there's dir with INDEX flag set,
4695 * we'd normally treat htree data as empty space. But with metadata
4696 * checksumming that corrupts checksums so forbid that.
4698 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4699 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4700 ext4_error_inode(inode, function, line, 0,
4701 "iget: Dir with htree data on filesystem without dir_index feature.");
4702 ret = -EFSCORRUPTED;
4705 ei->i_disksize = inode->i_size;
4707 ei->i_reserved_quota = 0;
4709 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4710 ei->i_block_group = iloc.block_group;
4711 ei->i_last_alloc_group = ~0;
4713 * NOTE! The in-memory inode i_data array is in little-endian order
4714 * even on big-endian machines: we do NOT byteswap the block numbers!
4716 for (block = 0; block < EXT4_N_BLOCKS; block++)
4717 ei->i_data[block] = raw_inode->i_block[block];
4718 INIT_LIST_HEAD(&ei->i_orphan);
4719 ext4_fc_init_inode(&ei->vfs_inode);
4722 * Set transaction id's of transactions that have to be committed
4723 * to finish f[data]sync. We set them to currently running transaction
4724 * as we cannot be sure that the inode or some of its metadata isn't
4725 * part of the transaction - the inode could have been reclaimed and
4726 * now it is reread from disk.
4729 transaction_t *transaction;
4732 read_lock(&journal->j_state_lock);
4733 if (journal->j_running_transaction)
4734 transaction = journal->j_running_transaction;
4736 transaction = journal->j_committing_transaction;
4738 tid = transaction->t_tid;
4740 tid = journal->j_commit_sequence;
4741 read_unlock(&journal->j_state_lock);
4742 ei->i_sync_tid = tid;
4743 ei->i_datasync_tid = tid;
4746 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4747 if (ei->i_extra_isize == 0) {
4748 /* The extra space is currently unused. Use it. */
4749 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4750 ei->i_extra_isize = sizeof(struct ext4_inode) -
4751 EXT4_GOOD_OLD_INODE_SIZE;
4753 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4759 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4760 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4761 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4762 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4764 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4765 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4767 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4768 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4770 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4772 ext4_inode_set_iversion_queried(inode, ivers);
4776 if (ei->i_file_acl &&
4777 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4778 ext4_error_inode(inode, function, line, 0,
4779 "iget: bad extended attribute block %llu",
4781 ret = -EFSCORRUPTED;
4783 } else if (!ext4_has_inline_data(inode)) {
4784 /* validate the block references in the inode */
4785 if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4786 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4787 (S_ISLNK(inode->i_mode) &&
4788 !ext4_inode_is_fast_symlink(inode)))) {
4789 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4790 ret = ext4_ext_check_inode(inode);
4792 ret = ext4_ind_check_inode(inode);
4798 if (S_ISREG(inode->i_mode)) {
4799 inode->i_op = &ext4_file_inode_operations;
4800 inode->i_fop = &ext4_file_operations;
4801 ext4_set_aops(inode);
4802 } else if (S_ISDIR(inode->i_mode)) {
4803 inode->i_op = &ext4_dir_inode_operations;
4804 inode->i_fop = &ext4_dir_operations;
4805 } else if (S_ISLNK(inode->i_mode)) {
4806 /* VFS does not allow setting these so must be corruption */
4807 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4808 ext4_error_inode(inode, function, line, 0,
4809 "iget: immutable or append flags "
4810 "not allowed on symlinks");
4811 ret = -EFSCORRUPTED;
4814 if (IS_ENCRYPTED(inode)) {
4815 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4816 ext4_set_aops(inode);
4817 } else if (ext4_inode_is_fast_symlink(inode)) {
4818 inode->i_link = (char *)ei->i_data;
4819 inode->i_op = &ext4_fast_symlink_inode_operations;
4820 nd_terminate_link(ei->i_data, inode->i_size,
4821 sizeof(ei->i_data) - 1);
4823 inode->i_op = &ext4_symlink_inode_operations;
4824 ext4_set_aops(inode);
4826 inode_nohighmem(inode);
4827 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4828 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4829 inode->i_op = &ext4_special_inode_operations;
4830 if (raw_inode->i_block[0])
4831 init_special_inode(inode, inode->i_mode,
4832 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4834 init_special_inode(inode, inode->i_mode,
4835 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4836 } else if (ino == EXT4_BOOT_LOADER_INO) {
4837 make_bad_inode(inode);
4839 ret = -EFSCORRUPTED;
4840 ext4_error_inode(inode, function, line, 0,
4841 "iget: bogus i_mode (%o)", inode->i_mode);
4844 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
4845 ext4_error_inode(inode, function, line, 0,
4846 "casefold flag without casefold feature");
4849 unlock_new_inode(inode);
4855 return ERR_PTR(ret);
4858 static int ext4_inode_blocks_set(handle_t *handle,
4859 struct ext4_inode *raw_inode,
4860 struct ext4_inode_info *ei)
4862 struct inode *inode = &(ei->vfs_inode);
4863 u64 i_blocks = READ_ONCE(inode->i_blocks);
4864 struct super_block *sb = inode->i_sb;
4866 if (i_blocks <= ~0U) {
4868 * i_blocks can be represented in a 32 bit variable
4869 * as multiple of 512 bytes
4871 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4872 raw_inode->i_blocks_high = 0;
4873 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4878 * This should never happen since sb->s_maxbytes should not have
4879 * allowed this, sb->s_maxbytes was set according to the huge_file
4880 * feature in ext4_fill_super().
4882 if (!ext4_has_feature_huge_file(sb))
4883 return -EFSCORRUPTED;
4885 if (i_blocks <= 0xffffffffffffULL) {
4887 * i_blocks can be represented in a 48 bit variable
4888 * as multiple of 512 bytes
4890 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4891 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4892 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4894 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4895 /* i_block is stored in file system block size */
4896 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4897 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4898 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4903 static void __ext4_update_other_inode_time(struct super_block *sb,
4904 unsigned long orig_ino,
4906 struct ext4_inode *raw_inode)
4908 struct inode *inode;
4910 inode = find_inode_by_ino_rcu(sb, ino);
4914 if (!inode_is_dirtytime_only(inode))
4917 spin_lock(&inode->i_lock);
4918 if (inode_is_dirtytime_only(inode)) {
4919 struct ext4_inode_info *ei = EXT4_I(inode);
4921 inode->i_state &= ~I_DIRTY_TIME;
4922 spin_unlock(&inode->i_lock);
4924 spin_lock(&ei->i_raw_lock);
4925 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4926 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4927 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4928 ext4_inode_csum_set(inode, raw_inode, ei);
4929 spin_unlock(&ei->i_raw_lock);
4930 trace_ext4_other_inode_update_time(inode, orig_ino);
4933 spin_unlock(&inode->i_lock);
4937 * Opportunistically update the other time fields for other inodes in
4938 * the same inode table block.
4940 static void ext4_update_other_inodes_time(struct super_block *sb,
4941 unsigned long orig_ino, char *buf)
4944 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4945 int inode_size = EXT4_INODE_SIZE(sb);
4948 * Calculate the first inode in the inode table block. Inode
4949 * numbers are one-based. That is, the first inode in a block
4950 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4952 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4954 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4955 if (ino == orig_ino)
4957 __ext4_update_other_inode_time(sb, orig_ino, ino,
4958 (struct ext4_inode *)buf);
4964 * Post the struct inode info into an on-disk inode location in the
4965 * buffer-cache. This gobbles the caller's reference to the
4966 * buffer_head in the inode location struct.
4968 * The caller must have write access to iloc->bh.
4970 static int ext4_do_update_inode(handle_t *handle,
4971 struct inode *inode,
4972 struct ext4_iloc *iloc)
4974 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4975 struct ext4_inode_info *ei = EXT4_I(inode);
4976 struct buffer_head *bh = iloc->bh;
4977 struct super_block *sb = inode->i_sb;
4979 int need_datasync = 0, set_large_file = 0;
4984 spin_lock(&ei->i_raw_lock);
4987 * For fields not tracked in the in-memory inode, initialise them
4988 * to zero for new inodes.
4990 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4991 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4993 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4995 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4996 i_uid = i_uid_read(inode);
4997 i_gid = i_gid_read(inode);
4998 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4999 if (!(test_opt(inode->i_sb, NO_UID32))) {
5000 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5001 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5003 * Fix up interoperability with old kernels. Otherwise,
5004 * old inodes get re-used with the upper 16 bits of the
5007 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5008 raw_inode->i_uid_high = 0;
5009 raw_inode->i_gid_high = 0;
5011 raw_inode->i_uid_high =
5012 cpu_to_le16(high_16_bits(i_uid));
5013 raw_inode->i_gid_high =
5014 cpu_to_le16(high_16_bits(i_gid));
5017 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5018 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5019 raw_inode->i_uid_high = 0;
5020 raw_inode->i_gid_high = 0;
5022 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5024 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5025 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5026 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5027 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5029 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5030 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5031 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5032 raw_inode->i_file_acl_high =
5033 cpu_to_le16(ei->i_file_acl >> 32);
5034 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5035 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) {
5036 ext4_isize_set(raw_inode, ei->i_disksize);
5039 if (ei->i_disksize > 0x7fffffffULL) {
5040 if (!ext4_has_feature_large_file(sb) ||
5041 EXT4_SB(sb)->s_es->s_rev_level ==
5042 cpu_to_le32(EXT4_GOOD_OLD_REV))
5045 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5046 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5047 if (old_valid_dev(inode->i_rdev)) {
5048 raw_inode->i_block[0] =
5049 cpu_to_le32(old_encode_dev(inode->i_rdev));
5050 raw_inode->i_block[1] = 0;
5052 raw_inode->i_block[0] = 0;
5053 raw_inode->i_block[1] =
5054 cpu_to_le32(new_encode_dev(inode->i_rdev));
5055 raw_inode->i_block[2] = 0;
5057 } else if (!ext4_has_inline_data(inode)) {
5058 for (block = 0; block < EXT4_N_BLOCKS; block++)
5059 raw_inode->i_block[block] = ei->i_data[block];
5062 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5063 u64 ivers = ext4_inode_peek_iversion(inode);
5065 raw_inode->i_disk_version = cpu_to_le32(ivers);
5066 if (ei->i_extra_isize) {
5067 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5068 raw_inode->i_version_hi =
5069 cpu_to_le32(ivers >> 32);
5070 raw_inode->i_extra_isize =
5071 cpu_to_le16(ei->i_extra_isize);
5075 if (i_projid != EXT4_DEF_PROJID &&
5076 !ext4_has_feature_project(inode->i_sb))
5077 err = err ?: -EFSCORRUPTED;
5079 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5080 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5081 raw_inode->i_projid = cpu_to_le32(i_projid);
5083 ext4_inode_csum_set(inode, raw_inode, ei);
5084 spin_unlock(&ei->i_raw_lock);
5086 EXT4_ERROR_INODE(inode, "corrupted inode contents");
5090 if (inode->i_sb->s_flags & SB_LAZYTIME)
5091 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5094 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5095 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5098 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5099 if (set_large_file) {
5100 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5101 err = ext4_journal_get_write_access(handle, sb,
5106 lock_buffer(EXT4_SB(sb)->s_sbh);
5107 ext4_set_feature_large_file(sb);
5108 ext4_superblock_csum_set(sb);
5109 unlock_buffer(EXT4_SB(sb)->s_sbh);
5110 ext4_handle_sync(handle);
5111 err = ext4_handle_dirty_metadata(handle, NULL,
5112 EXT4_SB(sb)->s_sbh);
5114 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5116 ext4_std_error(inode->i_sb, err);
5123 * ext4_write_inode()
5125 * We are called from a few places:
5127 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5128 * Here, there will be no transaction running. We wait for any running
5129 * transaction to commit.
5131 * - Within flush work (sys_sync(), kupdate and such).
5132 * We wait on commit, if told to.
5134 * - Within iput_final() -> write_inode_now()
5135 * We wait on commit, if told to.
5137 * In all cases it is actually safe for us to return without doing anything,
5138 * because the inode has been copied into a raw inode buffer in
5139 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5142 * Note that we are absolutely dependent upon all inode dirtiers doing the
5143 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5144 * which we are interested.
5146 * It would be a bug for them to not do this. The code:
5148 * mark_inode_dirty(inode)
5150 * inode->i_size = expr;
5152 * is in error because write_inode() could occur while `stuff()' is running,
5153 * and the new i_size will be lost. Plus the inode will no longer be on the
5154 * superblock's dirty inode list.
5156 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5160 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5161 sb_rdonly(inode->i_sb))
5164 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5167 if (EXT4_SB(inode->i_sb)->s_journal) {
5168 if (ext4_journal_current_handle()) {
5169 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5175 * No need to force transaction in WB_SYNC_NONE mode. Also
5176 * ext4_sync_fs() will force the commit after everything is
5179 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5182 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5183 EXT4_I(inode)->i_sync_tid);
5185 struct ext4_iloc iloc;
5187 err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5191 * sync(2) will flush the whole buffer cache. No need to do
5192 * it here separately for each inode.
5194 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5195 sync_dirty_buffer(iloc.bh);
5196 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5197 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5198 "IO error syncing inode");
5207 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5208 * buffers that are attached to a page stradding i_size and are undergoing
5209 * commit. In that case we have to wait for commit to finish and try again.
5211 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5215 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5216 tid_t commit_tid = 0;
5219 offset = inode->i_size & (PAGE_SIZE - 1);
5221 * If the page is fully truncated, we don't need to wait for any commit
5222 * (and we even should not as __ext4_journalled_invalidatepage() may
5223 * strip all buffers from the page but keep the page dirty which can then
5224 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5225 * buffers). Also we don't need to wait for any commit if all buffers in
5226 * the page remain valid. This is most beneficial for the common case of
5227 * blocksize == PAGESIZE.
5229 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5232 page = find_lock_page(inode->i_mapping,
5233 inode->i_size >> PAGE_SHIFT);
5236 ret = __ext4_journalled_invalidatepage(page, offset,
5237 PAGE_SIZE - offset);
5243 read_lock(&journal->j_state_lock);
5244 if (journal->j_committing_transaction)
5245 commit_tid = journal->j_committing_transaction->t_tid;
5246 read_unlock(&journal->j_state_lock);
5248 jbd2_log_wait_commit(journal, commit_tid);
5255 * Called from notify_change.
5257 * We want to trap VFS attempts to truncate the file as soon as
5258 * possible. In particular, we want to make sure that when the VFS
5259 * shrinks i_size, we put the inode on the orphan list and modify
5260 * i_disksize immediately, so that during the subsequent flushing of
5261 * dirty pages and freeing of disk blocks, we can guarantee that any
5262 * commit will leave the blocks being flushed in an unused state on
5263 * disk. (On recovery, the inode will get truncated and the blocks will
5264 * be freed, so we have a strong guarantee that no future commit will
5265 * leave these blocks visible to the user.)
5267 * Another thing we have to assure is that if we are in ordered mode
5268 * and inode is still attached to the committing transaction, we must
5269 * we start writeout of all the dirty pages which are being truncated.
5270 * This way we are sure that all the data written in the previous
5271 * transaction are already on disk (truncate waits for pages under
5274 * Called with inode->i_mutex down.
5276 int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5279 struct inode *inode = d_inode(dentry);
5282 const unsigned int ia_valid = attr->ia_valid;
5284 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5287 if (unlikely(IS_IMMUTABLE(inode)))
5290 if (unlikely(IS_APPEND(inode) &&
5291 (ia_valid & (ATTR_MODE | ATTR_UID |
5292 ATTR_GID | ATTR_TIMES_SET))))
5295 error = setattr_prepare(mnt_userns, dentry, attr);
5299 error = fscrypt_prepare_setattr(dentry, attr);
5303 error = fsverity_prepare_setattr(dentry, attr);
5307 if (is_quota_modification(inode, attr)) {
5308 error = dquot_initialize(inode);
5312 ext4_fc_start_update(inode);
5313 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5314 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5317 /* (user+group)*(old+new) structure, inode write (sb,
5318 * inode block, ? - but truncate inode update has it) */
5319 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5320 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5321 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5322 if (IS_ERR(handle)) {
5323 error = PTR_ERR(handle);
5327 /* dquot_transfer() calls back ext4_get_inode_usage() which
5328 * counts xattr inode references.
5330 down_read(&EXT4_I(inode)->xattr_sem);
5331 error = dquot_transfer(inode, attr);
5332 up_read(&EXT4_I(inode)->xattr_sem);
5335 ext4_journal_stop(handle);
5336 ext4_fc_stop_update(inode);
5339 /* Update corresponding info in inode so that everything is in
5340 * one transaction */
5341 if (attr->ia_valid & ATTR_UID)
5342 inode->i_uid = attr->ia_uid;
5343 if (attr->ia_valid & ATTR_GID)
5344 inode->i_gid = attr->ia_gid;
5345 error = ext4_mark_inode_dirty(handle, inode);
5346 ext4_journal_stop(handle);
5347 if (unlikely(error)) {
5348 ext4_fc_stop_update(inode);
5353 if (attr->ia_valid & ATTR_SIZE) {
5355 loff_t oldsize = inode->i_size;
5356 int shrink = (attr->ia_size < inode->i_size);
5358 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5359 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5361 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5362 ext4_fc_stop_update(inode);
5366 if (!S_ISREG(inode->i_mode)) {
5367 ext4_fc_stop_update(inode);
5371 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5372 inode_inc_iversion(inode);
5375 if (ext4_should_order_data(inode)) {
5376 error = ext4_begin_ordered_truncate(inode,
5382 * Blocks are going to be removed from the inode. Wait
5383 * for dio in flight.
5385 inode_dio_wait(inode);
5388 filemap_invalidate_lock(inode->i_mapping);
5390 rc = ext4_break_layouts(inode);
5392 filemap_invalidate_unlock(inode->i_mapping);
5396 if (attr->ia_size != inode->i_size) {
5397 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5398 if (IS_ERR(handle)) {
5399 error = PTR_ERR(handle);
5402 if (ext4_handle_valid(handle) && shrink) {
5403 error = ext4_orphan_add(handle, inode);
5407 * Update c/mtime on truncate up, ext4_truncate() will
5408 * update c/mtime in shrink case below
5411 inode->i_mtime = current_time(inode);
5412 inode->i_ctime = inode->i_mtime;
5416 ext4_fc_track_range(handle, inode,
5417 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5418 inode->i_sb->s_blocksize_bits,
5419 (oldsize > 0 ? oldsize - 1 : 0) >>
5420 inode->i_sb->s_blocksize_bits);
5422 ext4_fc_track_range(
5424 (oldsize > 0 ? oldsize - 1 : oldsize) >>
5425 inode->i_sb->s_blocksize_bits,
5426 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5427 inode->i_sb->s_blocksize_bits);
5429 down_write(&EXT4_I(inode)->i_data_sem);
5430 EXT4_I(inode)->i_disksize = attr->ia_size;
5431 rc = ext4_mark_inode_dirty(handle, inode);
5435 * We have to update i_size under i_data_sem together
5436 * with i_disksize to avoid races with writeback code
5437 * running ext4_wb_update_i_disksize().
5440 i_size_write(inode, attr->ia_size);
5441 up_write(&EXT4_I(inode)->i_data_sem);
5442 ext4_journal_stop(handle);
5446 pagecache_isize_extended(inode, oldsize,
5448 } else if (ext4_should_journal_data(inode)) {
5449 ext4_wait_for_tail_page_commit(inode);
5454 * Truncate pagecache after we've waited for commit
5455 * in data=journal mode to make pages freeable.
5457 truncate_pagecache(inode, inode->i_size);
5459 * Call ext4_truncate() even if i_size didn't change to
5460 * truncate possible preallocated blocks.
5462 if (attr->ia_size <= oldsize) {
5463 rc = ext4_truncate(inode);
5468 filemap_invalidate_unlock(inode->i_mapping);
5472 setattr_copy(mnt_userns, inode, attr);
5473 mark_inode_dirty(inode);
5477 * If the call to ext4_truncate failed to get a transaction handle at
5478 * all, we need to clean up the in-core orphan list manually.
5480 if (orphan && inode->i_nlink)
5481 ext4_orphan_del(NULL, inode);
5483 if (!error && (ia_valid & ATTR_MODE))
5484 rc = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5488 ext4_std_error(inode->i_sb, error);
5491 ext4_fc_stop_update(inode);
5495 int ext4_getattr(struct user_namespace *mnt_userns, const struct path *path,
5496 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5498 struct inode *inode = d_inode(path->dentry);
5499 struct ext4_inode *raw_inode;
5500 struct ext4_inode_info *ei = EXT4_I(inode);
5503 if ((request_mask & STATX_BTIME) &&
5504 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5505 stat->result_mask |= STATX_BTIME;
5506 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5507 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5510 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5511 if (flags & EXT4_APPEND_FL)
5512 stat->attributes |= STATX_ATTR_APPEND;
5513 if (flags & EXT4_COMPR_FL)
5514 stat->attributes |= STATX_ATTR_COMPRESSED;
5515 if (flags & EXT4_ENCRYPT_FL)
5516 stat->attributes |= STATX_ATTR_ENCRYPTED;
5517 if (flags & EXT4_IMMUTABLE_FL)
5518 stat->attributes |= STATX_ATTR_IMMUTABLE;
5519 if (flags & EXT4_NODUMP_FL)
5520 stat->attributes |= STATX_ATTR_NODUMP;
5521 if (flags & EXT4_VERITY_FL)
5522 stat->attributes |= STATX_ATTR_VERITY;
5524 stat->attributes_mask |= (STATX_ATTR_APPEND |
5525 STATX_ATTR_COMPRESSED |
5526 STATX_ATTR_ENCRYPTED |
5527 STATX_ATTR_IMMUTABLE |
5531 generic_fillattr(mnt_userns, inode, stat);
5535 int ext4_file_getattr(struct user_namespace *mnt_userns,
5536 const struct path *path, struct kstat *stat,
5537 u32 request_mask, unsigned int query_flags)
5539 struct inode *inode = d_inode(path->dentry);
5540 u64 delalloc_blocks;
5542 ext4_getattr(mnt_userns, path, stat, request_mask, query_flags);
5545 * If there is inline data in the inode, the inode will normally not
5546 * have data blocks allocated (it may have an external xattr block).
5547 * Report at least one sector for such files, so tools like tar, rsync,
5548 * others don't incorrectly think the file is completely sparse.
5550 if (unlikely(ext4_has_inline_data(inode)))
5551 stat->blocks += (stat->size + 511) >> 9;
5554 * We can't update i_blocks if the block allocation is delayed
5555 * otherwise in the case of system crash before the real block
5556 * allocation is done, we will have i_blocks inconsistent with
5557 * on-disk file blocks.
5558 * We always keep i_blocks updated together with real
5559 * allocation. But to not confuse with user, stat
5560 * will return the blocks that include the delayed allocation
5561 * blocks for this file.
5563 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5564 EXT4_I(inode)->i_reserved_data_blocks);
5565 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5569 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5572 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5573 return ext4_ind_trans_blocks(inode, lblocks);
5574 return ext4_ext_index_trans_blocks(inode, pextents);
5578 * Account for index blocks, block groups bitmaps and block group
5579 * descriptor blocks if modify datablocks and index blocks
5580 * worse case, the indexs blocks spread over different block groups
5582 * If datablocks are discontiguous, they are possible to spread over
5583 * different block groups too. If they are contiguous, with flexbg,
5584 * they could still across block group boundary.
5586 * Also account for superblock, inode, quota and xattr blocks
5588 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5591 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5597 * How many index blocks need to touch to map @lblocks logical blocks
5598 * to @pextents physical extents?
5600 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5605 * Now let's see how many group bitmaps and group descriptors need
5608 groups = idxblocks + pextents;
5610 if (groups > ngroups)
5612 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5613 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5615 /* bitmaps and block group descriptor blocks */
5616 ret += groups + gdpblocks;
5618 /* Blocks for super block, inode, quota and xattr blocks */
5619 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5625 * Calculate the total number of credits to reserve to fit
5626 * the modification of a single pages into a single transaction,
5627 * which may include multiple chunks of block allocations.
5629 * This could be called via ext4_write_begin()
5631 * We need to consider the worse case, when
5632 * one new block per extent.
5634 int ext4_writepage_trans_blocks(struct inode *inode)
5636 int bpp = ext4_journal_blocks_per_page(inode);
5639 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5641 /* Account for data blocks for journalled mode */
5642 if (ext4_should_journal_data(inode))
5648 * Calculate the journal credits for a chunk of data modification.
5650 * This is called from DIO, fallocate or whoever calling
5651 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5653 * journal buffers for data blocks are not included here, as DIO
5654 * and fallocate do no need to journal data buffers.
5656 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5658 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5662 * The caller must have previously called ext4_reserve_inode_write().
5663 * Give this, we know that the caller already has write access to iloc->bh.
5665 int ext4_mark_iloc_dirty(handle_t *handle,
5666 struct inode *inode, struct ext4_iloc *iloc)
5670 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5674 ext4_fc_track_inode(handle, inode);
5676 if (IS_I_VERSION(inode))
5677 inode_inc_iversion(inode);
5679 /* the do_update_inode consumes one bh->b_count */
5682 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5683 err = ext4_do_update_inode(handle, inode, iloc);
5689 * On success, We end up with an outstanding reference count against
5690 * iloc->bh. This _must_ be cleaned up later.
5694 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5695 struct ext4_iloc *iloc)
5699 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5702 err = ext4_get_inode_loc(inode, iloc);
5704 BUFFER_TRACE(iloc->bh, "get_write_access");
5705 err = ext4_journal_get_write_access(handle, inode->i_sb,
5706 iloc->bh, EXT4_JTR_NONE);
5712 ext4_std_error(inode->i_sb, err);
5716 static int __ext4_expand_extra_isize(struct inode *inode,
5717 unsigned int new_extra_isize,
5718 struct ext4_iloc *iloc,
5719 handle_t *handle, int *no_expand)
5721 struct ext4_inode *raw_inode;
5722 struct ext4_xattr_ibody_header *header;
5723 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5724 struct ext4_inode_info *ei = EXT4_I(inode);
5727 /* this was checked at iget time, but double check for good measure */
5728 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5729 (ei->i_extra_isize & 3)) {
5730 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5732 EXT4_INODE_SIZE(inode->i_sb));
5733 return -EFSCORRUPTED;
5735 if ((new_extra_isize < ei->i_extra_isize) ||
5736 (new_extra_isize < 4) ||
5737 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5738 return -EINVAL; /* Should never happen */
5740 raw_inode = ext4_raw_inode(iloc);
5742 header = IHDR(inode, raw_inode);
5744 /* No extended attributes present */
5745 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5746 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5747 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5748 EXT4_I(inode)->i_extra_isize, 0,
5749 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5750 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5754 /* try to expand with EAs present */
5755 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5759 * Inode size expansion failed; don't try again
5768 * Expand an inode by new_extra_isize bytes.
5769 * Returns 0 on success or negative error number on failure.
5771 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5772 unsigned int new_extra_isize,
5773 struct ext4_iloc iloc,
5779 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5783 * In nojournal mode, we can immediately attempt to expand
5784 * the inode. When journaled, we first need to obtain extra
5785 * buffer credits since we may write into the EA block
5786 * with this same handle. If journal_extend fails, then it will
5787 * only result in a minor loss of functionality for that inode.
5788 * If this is felt to be critical, then e2fsck should be run to
5789 * force a large enough s_min_extra_isize.
5791 if (ext4_journal_extend(handle,
5792 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5795 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5798 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5799 handle, &no_expand);
5800 ext4_write_unlock_xattr(inode, &no_expand);
5805 int ext4_expand_extra_isize(struct inode *inode,
5806 unsigned int new_extra_isize,
5807 struct ext4_iloc *iloc)
5813 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5818 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5819 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5820 if (IS_ERR(handle)) {
5821 error = PTR_ERR(handle);
5826 ext4_write_lock_xattr(inode, &no_expand);
5828 BUFFER_TRACE(iloc->bh, "get_write_access");
5829 error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
5836 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5837 handle, &no_expand);
5839 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5844 ext4_write_unlock_xattr(inode, &no_expand);
5845 ext4_journal_stop(handle);
5850 * What we do here is to mark the in-core inode as clean with respect to inode
5851 * dirtiness (it may still be data-dirty).
5852 * This means that the in-core inode may be reaped by prune_icache
5853 * without having to perform any I/O. This is a very good thing,
5854 * because *any* task may call prune_icache - even ones which
5855 * have a transaction open against a different journal.
5857 * Is this cheating? Not really. Sure, we haven't written the
5858 * inode out, but prune_icache isn't a user-visible syncing function.
5859 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5860 * we start and wait on commits.
5862 int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5863 const char *func, unsigned int line)
5865 struct ext4_iloc iloc;
5866 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5870 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5871 err = ext4_reserve_inode_write(handle, inode, &iloc);
5875 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5876 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5879 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5882 ext4_error_inode_err(inode, func, line, 0, err,
5883 "mark_inode_dirty error");
5888 * ext4_dirty_inode() is called from __mark_inode_dirty()
5890 * We're really interested in the case where a file is being extended.
5891 * i_size has been changed by generic_commit_write() and we thus need
5892 * to include the updated inode in the current transaction.
5894 * Also, dquot_alloc_block() will always dirty the inode when blocks
5895 * are allocated to the file.
5897 * If the inode is marked synchronous, we don't honour that here - doing
5898 * so would cause a commit on atime updates, which we don't bother doing.
5899 * We handle synchronous inodes at the highest possible level.
5901 void ext4_dirty_inode(struct inode *inode, int flags)
5905 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5908 ext4_mark_inode_dirty(handle, inode);
5909 ext4_journal_stop(handle);
5912 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5917 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5920 * We have to be very careful here: changing a data block's
5921 * journaling status dynamically is dangerous. If we write a
5922 * data block to the journal, change the status and then delete
5923 * that block, we risk forgetting to revoke the old log record
5924 * from the journal and so a subsequent replay can corrupt data.
5925 * So, first we make sure that the journal is empty and that
5926 * nobody is changing anything.
5929 journal = EXT4_JOURNAL(inode);
5932 if (is_journal_aborted(journal))
5935 /* Wait for all existing dio workers */
5936 inode_dio_wait(inode);
5939 * Before flushing the journal and switching inode's aops, we have
5940 * to flush all dirty data the inode has. There can be outstanding
5941 * delayed allocations, there can be unwritten extents created by
5942 * fallocate or buffered writes in dioread_nolock mode covered by
5943 * dirty data which can be converted only after flushing the dirty
5944 * data (and journalled aops don't know how to handle these cases).
5947 filemap_invalidate_lock(inode->i_mapping);
5948 err = filemap_write_and_wait(inode->i_mapping);
5950 filemap_invalidate_unlock(inode->i_mapping);
5955 percpu_down_write(&sbi->s_writepages_rwsem);
5956 jbd2_journal_lock_updates(journal);
5959 * OK, there are no updates running now, and all cached data is
5960 * synced to disk. We are now in a completely consistent state
5961 * which doesn't have anything in the journal, and we know that
5962 * no filesystem updates are running, so it is safe to modify
5963 * the inode's in-core data-journaling state flag now.
5967 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5969 err = jbd2_journal_flush(journal, 0);
5971 jbd2_journal_unlock_updates(journal);
5972 percpu_up_write(&sbi->s_writepages_rwsem);
5975 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5977 ext4_set_aops(inode);
5979 jbd2_journal_unlock_updates(journal);
5980 percpu_up_write(&sbi->s_writepages_rwsem);
5983 filemap_invalidate_unlock(inode->i_mapping);
5985 /* Finally we can mark the inode as dirty. */
5987 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5989 return PTR_ERR(handle);
5991 ext4_fc_mark_ineligible(inode->i_sb,
5992 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE);
5993 err = ext4_mark_inode_dirty(handle, inode);
5994 ext4_handle_sync(handle);
5995 ext4_journal_stop(handle);
5996 ext4_std_error(inode->i_sb, err);
6001 static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6002 struct buffer_head *bh)
6004 return !buffer_mapped(bh);
6007 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6009 struct vm_area_struct *vma = vmf->vma;
6010 struct page *page = vmf->page;
6015 struct file *file = vma->vm_file;
6016 struct inode *inode = file_inode(file);
6017 struct address_space *mapping = inode->i_mapping;
6019 get_block_t *get_block;
6022 if (unlikely(IS_IMMUTABLE(inode)))
6023 return VM_FAULT_SIGBUS;
6025 sb_start_pagefault(inode->i_sb);
6026 file_update_time(vma->vm_file);
6028 filemap_invalidate_lock_shared(mapping);
6030 err = ext4_convert_inline_data(inode);
6035 * On data journalling we skip straight to the transaction handle:
6036 * there's no delalloc; page truncated will be checked later; the
6037 * early return w/ all buffers mapped (calculates size/len) can't
6038 * be used; and there's no dioread_nolock, so only ext4_get_block.
6040 if (ext4_should_journal_data(inode))
6043 /* Delalloc case is easy... */
6044 if (test_opt(inode->i_sb, DELALLOC) &&
6045 !ext4_nonda_switch(inode->i_sb)) {
6047 err = block_page_mkwrite(vma, vmf,
6048 ext4_da_get_block_prep);
6049 } while (err == -ENOSPC &&
6050 ext4_should_retry_alloc(inode->i_sb, &retries));
6055 size = i_size_read(inode);
6056 /* Page got truncated from under us? */
6057 if (page->mapping != mapping || page_offset(page) > size) {
6059 ret = VM_FAULT_NOPAGE;
6063 if (page->index == size >> PAGE_SHIFT)
6064 len = size & ~PAGE_MASK;
6068 * Return if we have all the buffers mapped. This avoids the need to do
6069 * journal_start/journal_stop which can block and take a long time
6071 * This cannot be done for data journalling, as we have to add the
6072 * inode to the transaction's list to writeprotect pages on commit.
6074 if (page_has_buffers(page)) {
6075 if (!ext4_walk_page_buffers(NULL, inode, page_buffers(page),
6077 ext4_bh_unmapped)) {
6078 /* Wait so that we don't change page under IO */
6079 wait_for_stable_page(page);
6080 ret = VM_FAULT_LOCKED;
6085 /* OK, we need to fill the hole... */
6086 if (ext4_should_dioread_nolock(inode))
6087 get_block = ext4_get_block_unwritten;
6089 get_block = ext4_get_block;
6091 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6092 ext4_writepage_trans_blocks(inode));
6093 if (IS_ERR(handle)) {
6094 ret = VM_FAULT_SIGBUS;
6098 * Data journalling can't use block_page_mkwrite() because it
6099 * will set_buffer_dirty() before do_journal_get_write_access()
6100 * thus might hit warning messages for dirty metadata buffers.
6102 if (!ext4_should_journal_data(inode)) {
6103 err = block_page_mkwrite(vma, vmf, get_block);
6106 size = i_size_read(inode);
6107 /* Page got truncated from under us? */
6108 if (page->mapping != mapping || page_offset(page) > size) {
6109 ret = VM_FAULT_NOPAGE;
6113 if (page->index == size >> PAGE_SHIFT)
6114 len = size & ~PAGE_MASK;
6118 err = __block_write_begin(page, 0, len, ext4_get_block);
6120 ret = VM_FAULT_SIGBUS;
6121 if (ext4_walk_page_buffers(handle, inode,
6122 page_buffers(page), 0, len, NULL,
6123 do_journal_get_write_access))
6125 if (ext4_walk_page_buffers(handle, inode,
6126 page_buffers(page), 0, len, NULL,
6129 if (ext4_jbd2_inode_add_write(handle, inode,
6130 page_offset(page), len))
6132 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6137 ext4_journal_stop(handle);
6138 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6141 ret = block_page_mkwrite_return(err);
6143 filemap_invalidate_unlock_shared(mapping);
6144 sb_end_pagefault(inode->i_sb);
6148 ext4_journal_stop(handle);