1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/inode.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
23 #include <linux/time.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/dax.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/bitops.h>
41 #include <linux/iomap.h>
42 #include <linux/iversion.h>
44 #include "ext4_jbd2.h"
49 #include <trace/events/ext4.h>
51 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
52 struct ext4_inode_info *ei)
54 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
57 int offset = offsetof(struct ext4_inode, i_checksum_lo);
58 unsigned int csum_size = sizeof(dummy_csum);
60 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
61 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
63 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
64 EXT4_GOOD_OLD_INODE_SIZE - offset);
66 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
67 offset = offsetof(struct ext4_inode, i_checksum_hi);
68 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
69 EXT4_GOOD_OLD_INODE_SIZE,
70 offset - EXT4_GOOD_OLD_INODE_SIZE);
71 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
72 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
76 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
77 EXT4_INODE_SIZE(inode->i_sb) - offset);
83 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
84 struct ext4_inode_info *ei)
86 __u32 provided, calculated;
88 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
89 cpu_to_le32(EXT4_OS_LINUX) ||
90 !ext4_has_metadata_csum(inode->i_sb))
93 provided = le16_to_cpu(raw->i_checksum_lo);
94 calculated = ext4_inode_csum(inode, raw, ei);
95 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
96 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
97 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
101 return provided == calculated;
104 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
105 struct ext4_inode_info *ei)
109 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
110 cpu_to_le32(EXT4_OS_LINUX) ||
111 !ext4_has_metadata_csum(inode->i_sb))
114 csum = ext4_inode_csum(inode, raw, ei);
115 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
116 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
117 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
118 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
121 static inline int ext4_begin_ordered_truncate(struct inode *inode,
124 trace_ext4_begin_ordered_truncate(inode, new_size);
126 * If jinode is zero, then we never opened the file for
127 * writing, so there's no need to call
128 * jbd2_journal_begin_ordered_truncate() since there's no
129 * outstanding writes we need to flush.
131 if (!EXT4_I(inode)->jinode)
133 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
134 EXT4_I(inode)->jinode,
138 static void ext4_invalidatepage(struct page *page, unsigned int offset,
139 unsigned int length);
140 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
141 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
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;
179 trace_ext4_evict_inode(inode);
181 if (inode->i_nlink) {
183 * When journalling data dirty buffers are tracked only in the
184 * journal. So although mm thinks everything is clean and
185 * ready for reaping the inode might still have some pages to
186 * write in the running transaction or waiting to be
187 * checkpointed. Thus calling jbd2_journal_invalidatepage()
188 * (via truncate_inode_pages()) to discard these buffers can
189 * cause data loss. Also even if we did not discard these
190 * buffers, we would have no way to find them after the inode
191 * is reaped and thus user could see stale data if he tries to
192 * read them before the transaction is checkpointed. So be
193 * careful and force everything to disk here... We use
194 * ei->i_datasync_tid to store the newest transaction
195 * containing inode's data.
197 * Note that directories do not have this problem because they
198 * don't use page cache.
200 if (inode->i_ino != EXT4_JOURNAL_INO &&
201 ext4_should_journal_data(inode) &&
202 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
203 inode->i_data.nrpages) {
204 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
205 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
207 jbd2_complete_transaction(journal, commit_tid);
208 filemap_write_and_wait(&inode->i_data);
210 truncate_inode_pages_final(&inode->i_data);
215 if (is_bad_inode(inode))
217 dquot_initialize(inode);
219 if (ext4_should_order_data(inode))
220 ext4_begin_ordered_truncate(inode, 0);
221 truncate_inode_pages_final(&inode->i_data);
224 * For inodes with journalled data, transaction commit could have
225 * dirtied the inode. Flush worker is ignoring it because of I_FREEING
226 * flag but we still need to remove the inode from the writeback lists.
228 if (!list_empty_careful(&inode->i_io_list)) {
229 WARN_ON_ONCE(!ext4_should_journal_data(inode));
230 inode_io_list_del(inode);
234 * Protect us against freezing - iput() caller didn't have to have any
235 * protection against it
237 sb_start_intwrite(inode->i_sb);
239 if (!IS_NOQUOTA(inode))
240 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
243 * Block bitmap, group descriptor, and inode are accounted in both
244 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
246 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
247 ext4_blocks_for_truncate(inode) + extra_credits - 3);
248 if (IS_ERR(handle)) {
249 ext4_std_error(inode->i_sb, PTR_ERR(handle));
251 * If we're going to skip the normal cleanup, we still need to
252 * make sure that the in-core orphan linked list is properly
255 ext4_orphan_del(NULL, inode);
256 sb_end_intwrite(inode->i_sb);
261 ext4_handle_sync(handle);
264 * Set inode->i_size to 0 before calling ext4_truncate(). We need
265 * special handling of symlinks here because i_size is used to
266 * determine whether ext4_inode_info->i_data contains symlink data or
267 * block mappings. Setting i_size to 0 will remove its fast symlink
268 * status. Erase i_data so that it becomes a valid empty block map.
270 if (ext4_inode_is_fast_symlink(inode))
271 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
273 err = ext4_mark_inode_dirty(handle, inode);
275 ext4_warning(inode->i_sb,
276 "couldn't mark inode dirty (err %d)", err);
279 if (inode->i_blocks) {
280 err = ext4_truncate(inode);
282 ext4_error_err(inode->i_sb, -err,
283 "couldn't truncate inode %lu (err %d)",
289 /* Remove xattr references. */
290 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
293 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
295 ext4_journal_stop(handle);
296 ext4_orphan_del(NULL, inode);
297 sb_end_intwrite(inode->i_sb);
298 ext4_xattr_inode_array_free(ea_inode_array);
303 * Kill off the orphan record which ext4_truncate created.
304 * AKPM: I think this can be inside the above `if'.
305 * Note that ext4_orphan_del() has to be able to cope with the
306 * deletion of a non-existent orphan - this is because we don't
307 * know if ext4_truncate() actually created an orphan record.
308 * (Well, we could do this if we need to, but heck - it works)
310 ext4_orphan_del(handle, inode);
311 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
314 * One subtle ordering requirement: if anything has gone wrong
315 * (transaction abort, IO errors, whatever), then we can still
316 * do these next steps (the fs will already have been marked as
317 * having errors), but we can't free the inode if the mark_dirty
320 if (ext4_mark_inode_dirty(handle, inode))
321 /* If that failed, just do the required in-core inode clear. */
322 ext4_clear_inode(inode);
324 ext4_free_inode(handle, inode);
325 ext4_journal_stop(handle);
326 sb_end_intwrite(inode->i_sb);
327 ext4_xattr_inode_array_free(ea_inode_array);
330 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
334 qsize_t *ext4_get_reserved_space(struct inode *inode)
336 return &EXT4_I(inode)->i_reserved_quota;
341 * Called with i_data_sem down, which is important since we can call
342 * ext4_discard_preallocations() from here.
344 void ext4_da_update_reserve_space(struct inode *inode,
345 int used, int quota_claim)
347 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
348 struct ext4_inode_info *ei = EXT4_I(inode);
350 spin_lock(&ei->i_block_reservation_lock);
351 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
352 if (unlikely(used > ei->i_reserved_data_blocks)) {
353 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
354 "with only %d reserved data blocks",
355 __func__, inode->i_ino, used,
356 ei->i_reserved_data_blocks);
358 used = ei->i_reserved_data_blocks;
361 /* Update per-inode reservations */
362 ei->i_reserved_data_blocks -= used;
363 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
365 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
367 /* Update quota subsystem for data blocks */
369 dquot_claim_block(inode, EXT4_C2B(sbi, used));
372 * We did fallocate with an offset that is already delayed
373 * allocated. So on delayed allocated writeback we should
374 * not re-claim the quota for fallocated blocks.
376 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
380 * If we have done all the pending block allocations and if
381 * there aren't any writers on the inode, we can discard the
382 * inode's preallocations.
384 if ((ei->i_reserved_data_blocks == 0) &&
385 !inode_is_open_for_write(inode))
386 ext4_discard_preallocations(inode, 0);
389 static int __check_block_validity(struct inode *inode, const char *func,
391 struct ext4_map_blocks *map)
393 if (ext4_has_feature_journal(inode->i_sb) &&
395 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
397 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
398 ext4_error_inode(inode, func, line, map->m_pblk,
399 "lblock %lu mapped to illegal pblock %llu "
400 "(length %d)", (unsigned long) map->m_lblk,
401 map->m_pblk, map->m_len);
402 return -EFSCORRUPTED;
407 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
412 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
413 return fscrypt_zeroout_range(inode, lblk, pblk, len);
415 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
422 #define check_block_validity(inode, map) \
423 __check_block_validity((inode), __func__, __LINE__, (map))
425 #ifdef ES_AGGRESSIVE_TEST
426 static void ext4_map_blocks_es_recheck(handle_t *handle,
428 struct ext4_map_blocks *es_map,
429 struct ext4_map_blocks *map,
436 * There is a race window that the result is not the same.
437 * e.g. xfstests #223 when dioread_nolock enables. The reason
438 * is that we lookup a block mapping in extent status tree with
439 * out taking i_data_sem. So at the time the unwritten extent
440 * could be converted.
442 down_read(&EXT4_I(inode)->i_data_sem);
443 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
444 retval = ext4_ext_map_blocks(handle, inode, map, 0);
446 retval = ext4_ind_map_blocks(handle, inode, map, 0);
448 up_read((&EXT4_I(inode)->i_data_sem));
451 * We don't check m_len because extent will be collpased in status
452 * tree. So the m_len might not equal.
454 if (es_map->m_lblk != map->m_lblk ||
455 es_map->m_flags != map->m_flags ||
456 es_map->m_pblk != map->m_pblk) {
457 printk("ES cache assertion failed for inode: %lu "
458 "es_cached ex [%d/%d/%llu/%x] != "
459 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
460 inode->i_ino, es_map->m_lblk, es_map->m_len,
461 es_map->m_pblk, es_map->m_flags, map->m_lblk,
462 map->m_len, map->m_pblk, map->m_flags,
466 #endif /* ES_AGGRESSIVE_TEST */
469 * The ext4_map_blocks() function tries to look up the requested blocks,
470 * and returns if the blocks are already mapped.
472 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
473 * and store the allocated blocks in the result buffer head and mark it
476 * If file type is extents based, it will call ext4_ext_map_blocks(),
477 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
480 * On success, it returns the number of blocks being mapped or allocated. if
481 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
482 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
484 * It returns 0 if plain look up failed (blocks have not been allocated), in
485 * that case, @map is returned as unmapped but we still do fill map->m_len to
486 * indicate the length of a hole starting at map->m_lblk.
488 * It returns the error in case of allocation failure.
490 int ext4_map_blocks(handle_t *handle, struct inode *inode,
491 struct ext4_map_blocks *map, int flags)
493 struct extent_status es;
496 #ifdef ES_AGGRESSIVE_TEST
497 struct ext4_map_blocks orig_map;
499 memcpy(&orig_map, map, sizeof(*map));
503 ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
504 flags, map->m_len, (unsigned long) map->m_lblk);
507 * ext4_map_blocks returns an int, and m_len is an unsigned int
509 if (unlikely(map->m_len > INT_MAX))
510 map->m_len = INT_MAX;
512 /* We can handle the block number less than EXT_MAX_BLOCKS */
513 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
514 return -EFSCORRUPTED;
516 /* Lookup extent status tree firstly */
517 if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
518 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
519 map->m_pblk = ext4_es_pblock(&es) +
520 map->m_lblk - es.es_lblk;
521 map->m_flags |= ext4_es_is_written(&es) ?
522 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
523 retval = es.es_len - (map->m_lblk - es.es_lblk);
524 if (retval > map->m_len)
527 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
529 retval = es.es_len - (map->m_lblk - es.es_lblk);
530 if (retval > map->m_len)
537 #ifdef ES_AGGRESSIVE_TEST
538 ext4_map_blocks_es_recheck(handle, inode, map,
545 * Try to see if we can get the block without requesting a new
548 down_read(&EXT4_I(inode)->i_data_sem);
549 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
550 retval = ext4_ext_map_blocks(handle, inode, map, 0);
552 retval = ext4_ind_map_blocks(handle, inode, map, 0);
557 if (unlikely(retval != map->m_len)) {
558 ext4_warning(inode->i_sb,
559 "ES len assertion failed for inode "
560 "%lu: retval %d != map->m_len %d",
561 inode->i_ino, retval, map->m_len);
565 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
566 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
567 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
568 !(status & EXTENT_STATUS_WRITTEN) &&
569 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
570 map->m_lblk + map->m_len - 1))
571 status |= EXTENT_STATUS_DELAYED;
572 ret = ext4_es_insert_extent(inode, map->m_lblk,
573 map->m_len, map->m_pblk, status);
577 up_read((&EXT4_I(inode)->i_data_sem));
580 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
581 ret = check_block_validity(inode, map);
586 /* If it is only a block(s) look up */
587 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
591 * Returns if the blocks have already allocated
593 * Note that if blocks have been preallocated
594 * ext4_ext_get_block() returns the create = 0
595 * with buffer head unmapped.
597 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
599 * If we need to convert extent to unwritten
600 * we continue and do the actual work in
601 * ext4_ext_map_blocks()
603 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
607 * Here we clear m_flags because after allocating an new extent,
608 * it will be set again.
610 map->m_flags &= ~EXT4_MAP_FLAGS;
613 * New blocks allocate and/or writing to unwritten extent
614 * will possibly result in updating i_data, so we take
615 * the write lock of i_data_sem, and call get_block()
616 * with create == 1 flag.
618 down_write(&EXT4_I(inode)->i_data_sem);
621 * We need to check for EXT4 here because migrate
622 * could have changed the inode type in between
624 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
625 retval = ext4_ext_map_blocks(handle, inode, map, flags);
627 retval = ext4_ind_map_blocks(handle, inode, map, flags);
629 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
631 * We allocated new blocks which will result in
632 * i_data's format changing. Force the migrate
633 * to fail by clearing migrate flags
635 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
639 * Update reserved blocks/metadata blocks after successful
640 * block allocation which had been deferred till now. We don't
641 * support fallocate for non extent files. So we can update
642 * reserve space here.
645 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
646 ext4_da_update_reserve_space(inode, retval, 1);
652 if (unlikely(retval != map->m_len)) {
653 ext4_warning(inode->i_sb,
654 "ES len assertion failed for inode "
655 "%lu: retval %d != map->m_len %d",
656 inode->i_ino, retval, map->m_len);
661 * We have to zeroout blocks before inserting them into extent
662 * status tree. Otherwise someone could look them up there and
663 * use them before they are really zeroed. We also have to
664 * unmap metadata before zeroing as otherwise writeback can
665 * overwrite zeros with stale data from block device.
667 if (flags & EXT4_GET_BLOCKS_ZERO &&
668 map->m_flags & EXT4_MAP_MAPPED &&
669 map->m_flags & EXT4_MAP_NEW) {
670 ret = ext4_issue_zeroout(inode, map->m_lblk,
671 map->m_pblk, map->m_len);
679 * If the extent has been zeroed out, we don't need to update
680 * extent status tree.
682 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
683 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
684 if (ext4_es_is_written(&es))
687 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
688 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
689 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
690 !(status & EXTENT_STATUS_WRITTEN) &&
691 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
692 map->m_lblk + map->m_len - 1))
693 status |= EXTENT_STATUS_DELAYED;
694 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
695 map->m_pblk, status);
703 up_write((&EXT4_I(inode)->i_data_sem));
704 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
705 ret = check_block_validity(inode, map);
710 * Inodes with freshly allocated blocks where contents will be
711 * visible after transaction commit must be on transaction's
714 if (map->m_flags & EXT4_MAP_NEW &&
715 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
716 !(flags & EXT4_GET_BLOCKS_ZERO) &&
717 !ext4_is_quota_file(inode) &&
718 ext4_should_order_data(inode)) {
720 (loff_t)map->m_lblk << inode->i_blkbits;
721 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
723 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
724 ret = ext4_jbd2_inode_add_wait(handle, inode,
727 ret = ext4_jbd2_inode_add_write(handle, inode,
732 ext4_fc_track_range(inode, map->m_lblk,
733 map->m_lblk + map->m_len - 1);
737 ext_debug(inode, "failed with err %d\n", retval);
742 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
743 * we have to be careful as someone else may be manipulating b_state as well.
745 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
747 unsigned long old_state;
748 unsigned long new_state;
750 flags &= EXT4_MAP_FLAGS;
752 /* Dummy buffer_head? Set non-atomically. */
754 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
758 * Someone else may be modifying b_state. Be careful! This is ugly but
759 * once we get rid of using bh as a container for mapping information
760 * to pass to / from get_block functions, this can go away.
763 old_state = READ_ONCE(bh->b_state);
764 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
766 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
769 static int _ext4_get_block(struct inode *inode, sector_t iblock,
770 struct buffer_head *bh, int flags)
772 struct ext4_map_blocks map;
775 if (ext4_has_inline_data(inode))
779 map.m_len = bh->b_size >> inode->i_blkbits;
781 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
784 map_bh(bh, inode->i_sb, map.m_pblk);
785 ext4_update_bh_state(bh, map.m_flags);
786 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
788 } else if (ret == 0) {
789 /* hole case, need to fill in bh->b_size */
790 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
795 int ext4_get_block(struct inode *inode, sector_t iblock,
796 struct buffer_head *bh, int create)
798 return _ext4_get_block(inode, iblock, bh,
799 create ? EXT4_GET_BLOCKS_CREATE : 0);
803 * Get block function used when preparing for buffered write if we require
804 * creating an unwritten extent if blocks haven't been allocated. The extent
805 * will be converted to written after the IO is complete.
807 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
808 struct buffer_head *bh_result, int create)
810 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
811 inode->i_ino, create);
812 return _ext4_get_block(inode, iblock, bh_result,
813 EXT4_GET_BLOCKS_IO_CREATE_EXT);
816 /* Maximum number of blocks we map for direct IO at once. */
817 #define DIO_MAX_BLOCKS 4096
820 * `handle' can be NULL if create is zero
822 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
823 ext4_lblk_t block, int map_flags)
825 struct ext4_map_blocks map;
826 struct buffer_head *bh;
827 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
830 J_ASSERT(handle != NULL || create == 0);
834 err = ext4_map_blocks(handle, inode, &map, map_flags);
837 return create ? ERR_PTR(-ENOSPC) : NULL;
841 bh = sb_getblk(inode->i_sb, map.m_pblk);
843 return ERR_PTR(-ENOMEM);
844 if (map.m_flags & EXT4_MAP_NEW) {
845 J_ASSERT(create != 0);
846 J_ASSERT(handle != NULL);
849 * Now that we do not always journal data, we should
850 * keep in mind whether this should always journal the
851 * new buffer as metadata. For now, regular file
852 * writes use ext4_get_block instead, so it's not a
856 BUFFER_TRACE(bh, "call get_create_access");
857 err = ext4_journal_get_create_access(handle, bh);
862 if (!buffer_uptodate(bh)) {
863 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
864 set_buffer_uptodate(bh);
867 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
868 err = ext4_handle_dirty_metadata(handle, inode, bh);
872 BUFFER_TRACE(bh, "not a new buffer");
879 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
880 ext4_lblk_t block, int map_flags)
882 struct buffer_head *bh;
885 bh = ext4_getblk(handle, inode, block, map_flags);
888 if (!bh || ext4_buffer_uptodate(bh))
891 ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
899 /* Read a contiguous batch of blocks. */
900 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
901 bool wait, struct buffer_head **bhs)
905 for (i = 0; i < bh_count; i++) {
906 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
907 if (IS_ERR(bhs[i])) {
908 err = PTR_ERR(bhs[i]);
914 for (i = 0; i < bh_count; i++)
915 /* Note that NULL bhs[i] is valid because of holes. */
916 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
917 ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
922 for (i = 0; i < bh_count; i++)
924 wait_on_buffer(bhs[i]);
926 for (i = 0; i < bh_count; i++) {
927 if (bhs[i] && !buffer_uptodate(bhs[i])) {
935 for (i = 0; i < bh_count; i++) {
942 int ext4_walk_page_buffers(handle_t *handle,
943 struct buffer_head *head,
947 int (*fn)(handle_t *handle,
948 struct buffer_head *bh))
950 struct buffer_head *bh;
951 unsigned block_start, block_end;
952 unsigned blocksize = head->b_size;
954 struct buffer_head *next;
956 for (bh = head, block_start = 0;
957 ret == 0 && (bh != head || !block_start);
958 block_start = block_end, bh = next) {
959 next = bh->b_this_page;
960 block_end = block_start + blocksize;
961 if (block_end <= from || block_start >= to) {
962 if (partial && !buffer_uptodate(bh))
966 err = (*fn)(handle, bh);
974 * To preserve ordering, it is essential that the hole instantiation and
975 * the data write be encapsulated in a single transaction. We cannot
976 * close off a transaction and start a new one between the ext4_get_block()
977 * and the commit_write(). So doing the jbd2_journal_start at the start of
978 * prepare_write() is the right place.
980 * Also, this function can nest inside ext4_writepage(). In that case, we
981 * *know* that ext4_writepage() has generated enough buffer credits to do the
982 * whole page. So we won't block on the journal in that case, which is good,
983 * because the caller may be PF_MEMALLOC.
985 * By accident, ext4 can be reentered when a transaction is open via
986 * quota file writes. If we were to commit the transaction while thus
987 * reentered, there can be a deadlock - we would be holding a quota
988 * lock, and the commit would never complete if another thread had a
989 * transaction open and was blocking on the quota lock - a ranking
992 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
993 * will _not_ run commit under these circumstances because handle->h_ref
994 * is elevated. We'll still have enough credits for the tiny quotafile
997 int do_journal_get_write_access(handle_t *handle,
998 struct buffer_head *bh)
1000 int dirty = buffer_dirty(bh);
1003 if (!buffer_mapped(bh) || buffer_freed(bh))
1006 * __block_write_begin() could have dirtied some buffers. Clean
1007 * the dirty bit as jbd2_journal_get_write_access() could complain
1008 * otherwise about fs integrity issues. Setting of the dirty bit
1009 * by __block_write_begin() isn't a real problem here as we clear
1010 * the bit before releasing a page lock and thus writeback cannot
1011 * ever write the buffer.
1014 clear_buffer_dirty(bh);
1015 BUFFER_TRACE(bh, "get write access");
1016 ret = ext4_journal_get_write_access(handle, bh);
1018 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1022 #ifdef CONFIG_FS_ENCRYPTION
1023 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1024 get_block_t *get_block)
1026 unsigned from = pos & (PAGE_SIZE - 1);
1027 unsigned to = from + len;
1028 struct inode *inode = page->mapping->host;
1029 unsigned block_start, block_end;
1032 unsigned blocksize = inode->i_sb->s_blocksize;
1034 struct buffer_head *bh, *head, *wait[2];
1038 BUG_ON(!PageLocked(page));
1039 BUG_ON(from > PAGE_SIZE);
1040 BUG_ON(to > PAGE_SIZE);
1043 if (!page_has_buffers(page))
1044 create_empty_buffers(page, blocksize, 0);
1045 head = page_buffers(page);
1046 bbits = ilog2(blocksize);
1047 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1049 for (bh = head, block_start = 0; bh != head || !block_start;
1050 block++, block_start = block_end, bh = bh->b_this_page) {
1051 block_end = block_start + blocksize;
1052 if (block_end <= from || block_start >= to) {
1053 if (PageUptodate(page)) {
1054 if (!buffer_uptodate(bh))
1055 set_buffer_uptodate(bh);
1060 clear_buffer_new(bh);
1061 if (!buffer_mapped(bh)) {
1062 WARN_ON(bh->b_size != blocksize);
1063 err = get_block(inode, block, bh, 1);
1066 if (buffer_new(bh)) {
1067 if (PageUptodate(page)) {
1068 clear_buffer_new(bh);
1069 set_buffer_uptodate(bh);
1070 mark_buffer_dirty(bh);
1073 if (block_end > to || block_start < from)
1074 zero_user_segments(page, to, block_end,
1079 if (PageUptodate(page)) {
1080 if (!buffer_uptodate(bh))
1081 set_buffer_uptodate(bh);
1084 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1085 !buffer_unwritten(bh) &&
1086 (block_start < from || block_end > to)) {
1087 ext4_read_bh_lock(bh, 0, false);
1088 wait[nr_wait++] = bh;
1092 * If we issued read requests, let them complete.
1094 for (i = 0; i < nr_wait; i++) {
1095 wait_on_buffer(wait[i]);
1096 if (!buffer_uptodate(wait[i]))
1099 if (unlikely(err)) {
1100 page_zero_new_buffers(page, from, to);
1101 } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1102 for (i = 0; i < nr_wait; i++) {
1105 err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1106 bh_offset(wait[i]));
1108 clear_buffer_uptodate(wait[i]);
1118 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1119 loff_t pos, unsigned len, unsigned flags,
1120 struct page **pagep, void **fsdata)
1122 struct inode *inode = mapping->host;
1123 int ret, needed_blocks;
1130 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1133 trace_ext4_write_begin(inode, pos, len, flags);
1135 * Reserve one block more for addition to orphan list in case
1136 * we allocate blocks but write fails for some reason
1138 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1139 index = pos >> PAGE_SHIFT;
1140 from = pos & (PAGE_SIZE - 1);
1143 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1144 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1153 * grab_cache_page_write_begin() can take a long time if the
1154 * system is thrashing due to memory pressure, or if the page
1155 * is being written back. So grab it first before we start
1156 * the transaction handle. This also allows us to allocate
1157 * the page (if needed) without using GFP_NOFS.
1160 page = grab_cache_page_write_begin(mapping, index, flags);
1166 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1167 if (IS_ERR(handle)) {
1169 return PTR_ERR(handle);
1173 if (page->mapping != mapping) {
1174 /* The page got truncated from under us */
1177 ext4_journal_stop(handle);
1180 /* In case writeback began while the page was unlocked */
1181 wait_for_stable_page(page);
1183 #ifdef CONFIG_FS_ENCRYPTION
1184 if (ext4_should_dioread_nolock(inode))
1185 ret = ext4_block_write_begin(page, pos, len,
1186 ext4_get_block_unwritten);
1188 ret = ext4_block_write_begin(page, pos, len,
1191 if (ext4_should_dioread_nolock(inode))
1192 ret = __block_write_begin(page, pos, len,
1193 ext4_get_block_unwritten);
1195 ret = __block_write_begin(page, pos, len, ext4_get_block);
1197 if (!ret && ext4_should_journal_data(inode)) {
1198 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1200 do_journal_get_write_access);
1204 bool extended = (pos + len > inode->i_size) &&
1205 !ext4_verity_in_progress(inode);
1209 * __block_write_begin may have instantiated a few blocks
1210 * outside i_size. Trim these off again. Don't need
1211 * i_size_read because we hold i_mutex.
1213 * Add inode to orphan list in case we crash before
1216 if (extended && ext4_can_truncate(inode))
1217 ext4_orphan_add(handle, inode);
1219 ext4_journal_stop(handle);
1221 ext4_truncate_failed_write(inode);
1223 * If truncate failed early the inode might
1224 * still be on the orphan list; we need to
1225 * make sure the inode is removed from the
1226 * orphan list in that case.
1229 ext4_orphan_del(NULL, inode);
1232 if (ret == -ENOSPC &&
1233 ext4_should_retry_alloc(inode->i_sb, &retries))
1242 /* For write_end() in data=journal mode */
1243 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1246 if (!buffer_mapped(bh) || buffer_freed(bh))
1248 set_buffer_uptodate(bh);
1249 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1250 clear_buffer_meta(bh);
1251 clear_buffer_prio(bh);
1256 * We need to pick up the new inode size which generic_commit_write gave us
1257 * `file' can be NULL - eg, when called from page_symlink().
1259 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1260 * buffers are managed internally.
1262 static int ext4_write_end(struct file *file,
1263 struct address_space *mapping,
1264 loff_t pos, unsigned len, unsigned copied,
1265 struct page *page, void *fsdata)
1267 handle_t *handle = ext4_journal_current_handle();
1268 struct inode *inode = mapping->host;
1269 loff_t old_size = inode->i_size;
1271 int i_size_changed = 0;
1272 int inline_data = ext4_has_inline_data(inode);
1273 bool verity = ext4_verity_in_progress(inode);
1275 trace_ext4_write_end(inode, pos, len, copied);
1277 ret = ext4_write_inline_data_end(inode, pos, len,
1286 copied = block_write_end(file, mapping, pos,
1287 len, copied, page, fsdata);
1289 * it's important to update i_size while still holding page lock:
1290 * page writeout could otherwise come in and zero beyond i_size.
1292 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1293 * blocks are being written past EOF, so skip the i_size update.
1296 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1300 if (old_size < pos && !verity)
1301 pagecache_isize_extended(inode, old_size, pos);
1303 * Don't mark the inode dirty under page lock. First, it unnecessarily
1304 * makes the holding time of page lock longer. Second, it forces lock
1305 * ordering of page lock and transaction start for journaling
1308 if (i_size_changed || inline_data)
1309 ret = ext4_mark_inode_dirty(handle, inode);
1311 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1312 /* if we have allocated more blocks and copied
1313 * less. We will have blocks allocated outside
1314 * inode->i_size. So truncate them
1316 ext4_orphan_add(handle, inode);
1318 ret2 = ext4_journal_stop(handle);
1322 if (pos + len > inode->i_size && !verity) {
1323 ext4_truncate_failed_write(inode);
1325 * If truncate failed early the inode might still be
1326 * on the orphan list; we need to make sure the inode
1327 * is removed from the orphan list in that case.
1330 ext4_orphan_del(NULL, inode);
1333 return ret ? ret : copied;
1337 * This is a private version of page_zero_new_buffers() which doesn't
1338 * set the buffer to be dirty, since in data=journalled mode we need
1339 * to call ext4_handle_dirty_metadata() instead.
1341 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1343 unsigned from, unsigned to)
1345 unsigned int block_start = 0, block_end;
1346 struct buffer_head *head, *bh;
1348 bh = head = page_buffers(page);
1350 block_end = block_start + bh->b_size;
1351 if (buffer_new(bh)) {
1352 if (block_end > from && block_start < to) {
1353 if (!PageUptodate(page)) {
1354 unsigned start, size;
1356 start = max(from, block_start);
1357 size = min(to, block_end) - start;
1359 zero_user(page, start, size);
1360 write_end_fn(handle, bh);
1362 clear_buffer_new(bh);
1365 block_start = block_end;
1366 bh = bh->b_this_page;
1367 } while (bh != head);
1370 static int ext4_journalled_write_end(struct file *file,
1371 struct address_space *mapping,
1372 loff_t pos, unsigned len, unsigned copied,
1373 struct page *page, void *fsdata)
1375 handle_t *handle = ext4_journal_current_handle();
1376 struct inode *inode = mapping->host;
1377 loff_t old_size = inode->i_size;
1381 int size_changed = 0;
1382 int inline_data = ext4_has_inline_data(inode);
1383 bool verity = ext4_verity_in_progress(inode);
1385 trace_ext4_journalled_write_end(inode, pos, len, copied);
1386 from = pos & (PAGE_SIZE - 1);
1389 BUG_ON(!ext4_handle_valid(handle));
1392 ret = ext4_write_inline_data_end(inode, pos, len,
1400 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1402 ext4_journalled_zero_new_buffers(handle, page, from, to);
1404 if (unlikely(copied < len))
1405 ext4_journalled_zero_new_buffers(handle, page,
1407 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1408 from + copied, &partial,
1411 SetPageUptodate(page);
1414 size_changed = ext4_update_inode_size(inode, pos + copied);
1415 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1416 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1420 if (old_size < pos && !verity)
1421 pagecache_isize_extended(inode, old_size, pos);
1423 if (size_changed || inline_data) {
1424 ret2 = ext4_mark_inode_dirty(handle, inode);
1429 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1430 /* if we have allocated more blocks and copied
1431 * less. We will have blocks allocated outside
1432 * inode->i_size. So truncate them
1434 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 buffer_head *bh)
1611 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1615 * ext4_insert_delayed_block - adds a delayed block to the extents status
1616 * tree, incrementing the reserved cluster/block
1617 * count or making a pending reservation
1620 * @inode - file containing the newly added block
1621 * @lblk - logical block to be added
1623 * Returns 0 on success, negative error code on failure.
1625 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1627 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1629 bool allocated = false;
1632 * If the cluster containing lblk is shared with a delayed,
1633 * written, or unwritten extent in a bigalloc file system, it's
1634 * already been accounted for and does not need to be reserved.
1635 * A pending reservation must be made for the cluster if it's
1636 * shared with a written or unwritten extent and doesn't already
1637 * have one. Written and unwritten extents can be purged from the
1638 * extents status tree if the system is under memory pressure, so
1639 * it's necessary to examine the extent tree if a search of the
1640 * extents status tree doesn't get a match.
1642 if (sbi->s_cluster_ratio == 1) {
1643 ret = ext4_da_reserve_space(inode);
1644 if (ret != 0) /* ENOSPC */
1646 } else { /* bigalloc */
1647 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1648 if (!ext4_es_scan_clu(inode,
1649 &ext4_es_is_mapped, lblk)) {
1650 ret = ext4_clu_mapped(inode,
1651 EXT4_B2C(sbi, lblk));
1655 ret = ext4_da_reserve_space(inode);
1656 if (ret != 0) /* ENOSPC */
1667 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1674 * This function is grabs code from the very beginning of
1675 * ext4_map_blocks, but assumes that the caller is from delayed write
1676 * time. This function looks up the requested blocks and sets the
1677 * buffer delay bit under the protection of i_data_sem.
1679 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1680 struct ext4_map_blocks *map,
1681 struct buffer_head *bh)
1683 struct extent_status es;
1685 sector_t invalid_block = ~((sector_t) 0xffff);
1686 #ifdef ES_AGGRESSIVE_TEST
1687 struct ext4_map_blocks orig_map;
1689 memcpy(&orig_map, map, sizeof(*map));
1692 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1696 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1697 (unsigned long) map->m_lblk);
1699 /* Lookup extent status tree firstly */
1700 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1701 if (ext4_es_is_hole(&es)) {
1703 down_read(&EXT4_I(inode)->i_data_sem);
1708 * Delayed extent could be allocated by fallocate.
1709 * So we need to check it.
1711 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1712 map_bh(bh, inode->i_sb, invalid_block);
1714 set_buffer_delay(bh);
1718 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1719 retval = es.es_len - (iblock - es.es_lblk);
1720 if (retval > map->m_len)
1721 retval = map->m_len;
1722 map->m_len = retval;
1723 if (ext4_es_is_written(&es))
1724 map->m_flags |= EXT4_MAP_MAPPED;
1725 else if (ext4_es_is_unwritten(&es))
1726 map->m_flags |= EXT4_MAP_UNWRITTEN;
1730 #ifdef ES_AGGRESSIVE_TEST
1731 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1737 * Try to see if we can get the block without requesting a new
1738 * file system block.
1740 down_read(&EXT4_I(inode)->i_data_sem);
1741 if (ext4_has_inline_data(inode))
1743 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1744 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1746 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1753 * XXX: __block_prepare_write() unmaps passed block,
1757 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1763 map_bh(bh, inode->i_sb, invalid_block);
1765 set_buffer_delay(bh);
1766 } else if (retval > 0) {
1768 unsigned int status;
1770 if (unlikely(retval != map->m_len)) {
1771 ext4_warning(inode->i_sb,
1772 "ES len assertion failed for inode "
1773 "%lu: retval %d != map->m_len %d",
1774 inode->i_ino, retval, map->m_len);
1778 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1779 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1780 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1781 map->m_pblk, status);
1787 up_read((&EXT4_I(inode)->i_data_sem));
1793 * This is a special get_block_t callback which is used by
1794 * ext4_da_write_begin(). It will either return mapped block or
1795 * reserve space for a single block.
1797 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1798 * We also have b_blocknr = -1 and b_bdev initialized properly
1800 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1801 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1802 * initialized properly.
1804 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1805 struct buffer_head *bh, int create)
1807 struct ext4_map_blocks map;
1810 BUG_ON(create == 0);
1811 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1813 map.m_lblk = iblock;
1817 * first, we need to know whether the block is allocated already
1818 * preallocated blocks are unmapped but should treated
1819 * the same as allocated blocks.
1821 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1825 map_bh(bh, inode->i_sb, map.m_pblk);
1826 ext4_update_bh_state(bh, map.m_flags);
1828 if (buffer_unwritten(bh)) {
1829 /* A delayed write to unwritten bh should be marked
1830 * new and mapped. Mapped ensures that we don't do
1831 * get_block multiple times when we write to the same
1832 * offset and new ensures that we do proper zero out
1833 * for partial write.
1836 set_buffer_mapped(bh);
1841 static int bget_one(handle_t *handle, struct buffer_head *bh)
1847 static int bput_one(handle_t *handle, struct buffer_head *bh)
1853 static int __ext4_journalled_writepage(struct page *page,
1856 struct address_space *mapping = page->mapping;
1857 struct inode *inode = mapping->host;
1858 struct buffer_head *page_bufs = NULL;
1859 handle_t *handle = NULL;
1860 int ret = 0, err = 0;
1861 int inline_data = ext4_has_inline_data(inode);
1862 struct buffer_head *inode_bh = NULL;
1864 ClearPageChecked(page);
1867 BUG_ON(page->index != 0);
1868 BUG_ON(len > ext4_get_max_inline_size(inode));
1869 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1870 if (inode_bh == NULL)
1873 page_bufs = page_buffers(page);
1878 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1882 * We need to release the page lock before we start the
1883 * journal, so grab a reference so the page won't disappear
1884 * out from under us.
1889 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1890 ext4_writepage_trans_blocks(inode));
1891 if (IS_ERR(handle)) {
1892 ret = PTR_ERR(handle);
1894 goto out_no_pagelock;
1896 BUG_ON(!ext4_handle_valid(handle));
1900 if (page->mapping != mapping) {
1901 /* The page got truncated from under us */
1902 ext4_journal_stop(handle);
1908 ret = ext4_mark_inode_dirty(handle, inode);
1910 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1911 do_journal_get_write_access);
1913 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1918 err = ext4_jbd2_inode_add_write(handle, inode, 0, len);
1921 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1922 err = ext4_journal_stop(handle);
1926 if (!ext4_has_inline_data(inode))
1927 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1929 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1938 * Note that we don't need to start a transaction unless we're journaling data
1939 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1940 * need to file the inode to the transaction's list in ordered mode because if
1941 * we are writing back data added by write(), the inode is already there and if
1942 * we are writing back data modified via mmap(), no one guarantees in which
1943 * transaction the data will hit the disk. In case we are journaling data, we
1944 * cannot start transaction directly because transaction start ranks above page
1945 * lock so we have to do some magic.
1947 * This function can get called via...
1948 * - ext4_writepages after taking page lock (have journal handle)
1949 * - journal_submit_inode_data_buffers (no journal handle)
1950 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1951 * - grab_page_cache when doing write_begin (have journal handle)
1953 * We don't do any block allocation in this function. If we have page with
1954 * multiple blocks we need to write those buffer_heads that are mapped. This
1955 * is important for mmaped based write. So if we do with blocksize 1K
1956 * truncate(f, 1024);
1957 * a = mmap(f, 0, 4096);
1959 * truncate(f, 4096);
1960 * we have in the page first buffer_head mapped via page_mkwrite call back
1961 * but other buffer_heads would be unmapped but dirty (dirty done via the
1962 * do_wp_page). So writepage should write the first block. If we modify
1963 * the mmap area beyond 1024 we will again get a page_fault and the
1964 * page_mkwrite callback will do the block allocation and mark the
1965 * buffer_heads mapped.
1967 * We redirty the page if we have any buffer_heads that is either delay or
1968 * unwritten in the page.
1970 * We can get recursively called as show below.
1972 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1975 * But since we don't do any block allocation we should not deadlock.
1976 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1978 static int ext4_writepage(struct page *page,
1979 struct writeback_control *wbc)
1984 struct buffer_head *page_bufs = NULL;
1985 struct inode *inode = page->mapping->host;
1986 struct ext4_io_submit io_submit;
1987 bool keep_towrite = false;
1989 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
1990 inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
1995 trace_ext4_writepage(page);
1996 size = i_size_read(inode);
1997 if (page->index == size >> PAGE_SHIFT &&
1998 !ext4_verity_in_progress(inode))
1999 len = size & ~PAGE_MASK;
2003 page_bufs = page_buffers(page);
2005 * We cannot do block allocation or other extent handling in this
2006 * function. If there are buffers needing that, we have to redirty
2007 * the page. But we may reach here when we do a journal commit via
2008 * journal_submit_inode_data_buffers() and in that case we must write
2009 * allocated buffers to achieve data=ordered mode guarantees.
2011 * Also, if there is only one buffer per page (the fs block
2012 * size == the page size), if one buffer needs block
2013 * allocation or needs to modify the extent tree to clear the
2014 * unwritten flag, we know that the page can't be written at
2015 * all, so we might as well refuse the write immediately.
2016 * Unfortunately if the block size != page size, we can't as
2017 * easily detect this case using ext4_walk_page_buffers(), but
2018 * for the extremely common case, this is an optimization that
2019 * skips a useless round trip through ext4_bio_write_page().
2021 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2022 ext4_bh_delay_or_unwritten)) {
2023 redirty_page_for_writepage(wbc, page);
2024 if ((current->flags & PF_MEMALLOC) ||
2025 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2027 * For memory cleaning there's no point in writing only
2028 * some buffers. So just bail out. Warn if we came here
2029 * from direct reclaim.
2031 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2036 keep_towrite = true;
2039 if (PageChecked(page) && ext4_should_journal_data(inode))
2041 * It's mmapped pagecache. Add buffers and journal it. There
2042 * doesn't seem much point in redirtying the page here.
2044 return __ext4_journalled_writepage(page, len);
2046 ext4_io_submit_init(&io_submit, wbc);
2047 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2048 if (!io_submit.io_end) {
2049 redirty_page_for_writepage(wbc, page);
2053 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2054 ext4_io_submit(&io_submit);
2055 /* Drop io_end reference we got from init */
2056 ext4_put_io_end_defer(io_submit.io_end);
2060 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2066 BUG_ON(page->index != mpd->first_page);
2067 clear_page_dirty_for_io(page);
2069 * We have to be very careful here! Nothing protects writeback path
2070 * against i_size changes and the page can be writeably mapped into
2071 * page tables. So an application can be growing i_size and writing
2072 * data through mmap while writeback runs. clear_page_dirty_for_io()
2073 * write-protects our page in page tables and the page cannot get
2074 * written to again until we release page lock. So only after
2075 * clear_page_dirty_for_io() we are safe to sample i_size for
2076 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2077 * on the barrier provided by TestClearPageDirty in
2078 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2079 * after page tables are updated.
2081 size = i_size_read(mpd->inode);
2082 if (page->index == size >> PAGE_SHIFT &&
2083 !ext4_verity_in_progress(mpd->inode))
2084 len = size & ~PAGE_MASK;
2087 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2089 mpd->wbc->nr_to_write--;
2095 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2098 * mballoc gives us at most this number of blocks...
2099 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2100 * The rest of mballoc seems to handle chunks up to full group size.
2102 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2105 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2107 * @mpd - extent of blocks
2108 * @lblk - logical number of the block in the file
2109 * @bh - buffer head we want to add to the extent
2111 * The function is used to collect contig. blocks in the same state. If the
2112 * buffer doesn't require mapping for writeback and we haven't started the
2113 * extent of buffers to map yet, the function returns 'true' immediately - the
2114 * caller can write the buffer right away. Otherwise the function returns true
2115 * if the block has been added to the extent, false if the block couldn't be
2118 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2119 struct buffer_head *bh)
2121 struct ext4_map_blocks *map = &mpd->map;
2123 /* Buffer that doesn't need mapping for writeback? */
2124 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2125 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2126 /* So far no extent to map => we write the buffer right away */
2127 if (map->m_len == 0)
2132 /* First block in the extent? */
2133 if (map->m_len == 0) {
2134 /* We cannot map unless handle is started... */
2139 map->m_flags = bh->b_state & BH_FLAGS;
2143 /* Don't go larger than mballoc is willing to allocate */
2144 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2147 /* Can we merge the block to our big extent? */
2148 if (lblk == map->m_lblk + map->m_len &&
2149 (bh->b_state & BH_FLAGS) == map->m_flags) {
2157 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2159 * @mpd - extent of blocks for mapping
2160 * @head - the first buffer in the page
2161 * @bh - buffer we should start processing from
2162 * @lblk - logical number of the block in the file corresponding to @bh
2164 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2165 * the page for IO if all buffers in this page were mapped and there's no
2166 * accumulated extent of buffers to map or add buffers in the page to the
2167 * extent of buffers to map. The function returns 1 if the caller can continue
2168 * by processing the next page, 0 if it should stop adding buffers to the
2169 * extent to map because we cannot extend it anymore. It can also return value
2170 * < 0 in case of error during IO submission.
2172 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2173 struct buffer_head *head,
2174 struct buffer_head *bh,
2177 struct inode *inode = mpd->inode;
2179 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2180 >> inode->i_blkbits;
2182 if (ext4_verity_in_progress(inode))
2183 blocks = EXT_MAX_BLOCKS;
2186 BUG_ON(buffer_locked(bh));
2188 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2189 /* Found extent to map? */
2192 /* Buffer needs mapping and handle is not started? */
2195 /* Everything mapped so far and we hit EOF */
2198 } while (lblk++, (bh = bh->b_this_page) != head);
2199 /* So far everything mapped? Submit the page for IO. */
2200 if (mpd->map.m_len == 0) {
2201 err = mpage_submit_page(mpd, head->b_page);
2205 if (lblk >= blocks) {
2206 mpd->scanned_until_end = 1;
2213 * mpage_process_page - update page buffers corresponding to changed extent and
2214 * may submit fully mapped page for IO
2216 * @mpd - description of extent to map, on return next extent to map
2217 * @m_lblk - logical block mapping.
2218 * @m_pblk - corresponding physical mapping.
2219 * @map_bh - determines on return whether this page requires any further
2221 * Scan given page buffers corresponding to changed extent and update buffer
2222 * state according to new extent state.
2223 * We map delalloc buffers to their physical location, clear unwritten bits.
2224 * If the given page is not fully mapped, we update @map to the next extent in
2225 * the given page that needs mapping & return @map_bh as true.
2227 static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2228 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2231 struct buffer_head *head, *bh;
2232 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2233 ext4_lblk_t lblk = *m_lblk;
2234 ext4_fsblk_t pblock = *m_pblk;
2236 int blkbits = mpd->inode->i_blkbits;
2237 ssize_t io_end_size = 0;
2238 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2240 bh = head = page_buffers(page);
2242 if (lblk < mpd->map.m_lblk)
2244 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2246 * Buffer after end of mapped extent.
2247 * Find next buffer in the page to map.
2250 mpd->map.m_flags = 0;
2251 io_end_vec->size += io_end_size;
2254 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2257 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2258 io_end_vec = ext4_alloc_io_end_vec(io_end);
2259 if (IS_ERR(io_end_vec)) {
2260 err = PTR_ERR(io_end_vec);
2263 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2268 if (buffer_delay(bh)) {
2269 clear_buffer_delay(bh);
2270 bh->b_blocknr = pblock++;
2272 clear_buffer_unwritten(bh);
2273 io_end_size += (1 << blkbits);
2274 } while (lblk++, (bh = bh->b_this_page) != head);
2276 io_end_vec->size += io_end_size;
2286 * mpage_map_buffers - update buffers corresponding to changed extent and
2287 * submit fully mapped pages for IO
2289 * @mpd - description of extent to map, on return next extent to map
2291 * Scan buffers corresponding to changed extent (we expect corresponding pages
2292 * to be already locked) and update buffer state according to new extent state.
2293 * We map delalloc buffers to their physical location, clear unwritten bits,
2294 * and mark buffers as uninit when we perform writes to unwritten extents
2295 * and do extent conversion after IO is finished. If the last page is not fully
2296 * mapped, we update @map to the next extent in the last page that needs
2297 * mapping. Otherwise we submit the page for IO.
2299 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2301 struct pagevec pvec;
2303 struct inode *inode = mpd->inode;
2304 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2307 ext4_fsblk_t pblock;
2309 bool map_bh = false;
2311 start = mpd->map.m_lblk >> bpp_bits;
2312 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2313 lblk = start << bpp_bits;
2314 pblock = mpd->map.m_pblk;
2316 pagevec_init(&pvec);
2317 while (start <= end) {
2318 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2322 for (i = 0; i < nr_pages; i++) {
2323 struct page *page = pvec.pages[i];
2325 err = mpage_process_page(mpd, page, &lblk, &pblock,
2328 * If map_bh is true, means page may require further bh
2329 * mapping, or maybe the page was submitted for IO.
2330 * So we return to call further extent mapping.
2332 if (err < 0 || map_bh)
2334 /* Page fully mapped - let IO run! */
2335 err = mpage_submit_page(mpd, page);
2339 pagevec_release(&pvec);
2341 /* Extent fully mapped and matches with page boundary. We are done. */
2343 mpd->map.m_flags = 0;
2346 pagevec_release(&pvec);
2350 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2352 struct inode *inode = mpd->inode;
2353 struct ext4_map_blocks *map = &mpd->map;
2354 int get_blocks_flags;
2355 int err, dioread_nolock;
2357 trace_ext4_da_write_pages_extent(inode, map);
2359 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2360 * to convert an unwritten extent to be initialized (in the case
2361 * where we have written into one or more preallocated blocks). It is
2362 * possible that we're going to need more metadata blocks than
2363 * previously reserved. However we must not fail because we're in
2364 * writeback and there is nothing we can do about it so it might result
2365 * in data loss. So use reserved blocks to allocate metadata if
2368 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2369 * the blocks in question are delalloc blocks. This indicates
2370 * that the blocks and quotas has already been checked when
2371 * the data was copied into the page cache.
2373 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2374 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2375 EXT4_GET_BLOCKS_IO_SUBMIT;
2376 dioread_nolock = ext4_should_dioread_nolock(inode);
2378 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2379 if (map->m_flags & BIT(BH_Delay))
2380 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2382 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2385 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2386 if (!mpd->io_submit.io_end->handle &&
2387 ext4_handle_valid(handle)) {
2388 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2389 handle->h_rsv_handle = NULL;
2391 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2394 BUG_ON(map->m_len == 0);
2399 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2400 * mpd->len and submit pages underlying it for IO
2402 * @handle - handle for journal operations
2403 * @mpd - extent to map
2404 * @give_up_on_write - we set this to true iff there is a fatal error and there
2405 * is no hope of writing the data. The caller should discard
2406 * dirty pages to avoid infinite loops.
2408 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2409 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2410 * them to initialized or split the described range from larger unwritten
2411 * extent. Note that we need not map all the described range since allocation
2412 * can return less blocks or the range is covered by more unwritten extents. We
2413 * cannot map more because we are limited by reserved transaction credits. On
2414 * the other hand we always make sure that the last touched page is fully
2415 * mapped so that it can be written out (and thus forward progress is
2416 * guaranteed). After mapping we submit all mapped pages for IO.
2418 static int mpage_map_and_submit_extent(handle_t *handle,
2419 struct mpage_da_data *mpd,
2420 bool *give_up_on_write)
2422 struct inode *inode = mpd->inode;
2423 struct ext4_map_blocks *map = &mpd->map;
2427 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2428 struct ext4_io_end_vec *io_end_vec;
2430 io_end_vec = ext4_alloc_io_end_vec(io_end);
2431 if (IS_ERR(io_end_vec))
2432 return PTR_ERR(io_end_vec);
2433 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2435 err = mpage_map_one_extent(handle, mpd);
2437 struct super_block *sb = inode->i_sb;
2439 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2440 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2441 goto invalidate_dirty_pages;
2443 * Let the uper layers retry transient errors.
2444 * In the case of ENOSPC, if ext4_count_free_blocks()
2445 * is non-zero, a commit should free up blocks.
2447 if ((err == -ENOMEM) ||
2448 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2450 goto update_disksize;
2453 ext4_msg(sb, KERN_CRIT,
2454 "Delayed block allocation failed for "
2455 "inode %lu at logical offset %llu with"
2456 " max blocks %u with error %d",
2458 (unsigned long long)map->m_lblk,
2459 (unsigned)map->m_len, -err);
2460 ext4_msg(sb, KERN_CRIT,
2461 "This should not happen!! Data will "
2464 ext4_print_free_blocks(inode);
2465 invalidate_dirty_pages:
2466 *give_up_on_write = true;
2471 * Update buffer state, submit mapped pages, and get us new
2474 err = mpage_map_and_submit_buffers(mpd);
2476 goto update_disksize;
2477 } while (map->m_len);
2481 * Update on-disk size after IO is submitted. Races with
2482 * truncate are avoided by checking i_size under i_data_sem.
2484 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2485 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2489 down_write(&EXT4_I(inode)->i_data_sem);
2490 i_size = i_size_read(inode);
2491 if (disksize > i_size)
2493 if (disksize > EXT4_I(inode)->i_disksize)
2494 EXT4_I(inode)->i_disksize = disksize;
2495 up_write(&EXT4_I(inode)->i_data_sem);
2496 err2 = ext4_mark_inode_dirty(handle, inode);
2498 ext4_error_err(inode->i_sb, -err2,
2499 "Failed to mark inode %lu dirty",
2509 * Calculate the total number of credits to reserve for one writepages
2510 * iteration. This is called from ext4_writepages(). We map an extent of
2511 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2512 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2513 * bpp - 1 blocks in bpp different extents.
2515 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2517 int bpp = ext4_journal_blocks_per_page(inode);
2519 return ext4_meta_trans_blocks(inode,
2520 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2524 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2525 * and underlying extent to map
2527 * @mpd - where to look for pages
2529 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2530 * IO immediately. When we find a page which isn't mapped we start accumulating
2531 * extent of buffers underlying these pages that needs mapping (formed by
2532 * either delayed or unwritten buffers). We also lock the pages containing
2533 * these buffers. The extent found is returned in @mpd structure (starting at
2534 * mpd->lblk with length mpd->len blocks).
2536 * Note that this function can attach bios to one io_end structure which are
2537 * neither logically nor physically contiguous. Although it may seem as an
2538 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2539 * case as we need to track IO to all buffers underlying a page in one io_end.
2541 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2543 struct address_space *mapping = mpd->inode->i_mapping;
2544 struct pagevec pvec;
2545 unsigned int nr_pages;
2546 long left = mpd->wbc->nr_to_write;
2547 pgoff_t index = mpd->first_page;
2548 pgoff_t end = mpd->last_page;
2551 int blkbits = mpd->inode->i_blkbits;
2553 struct buffer_head *head;
2555 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2556 tag = PAGECACHE_TAG_TOWRITE;
2558 tag = PAGECACHE_TAG_DIRTY;
2560 pagevec_init(&pvec);
2562 mpd->next_page = index;
2563 while (index <= end) {
2564 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2569 for (i = 0; i < nr_pages; i++) {
2570 struct page *page = pvec.pages[i];
2573 * Accumulated enough dirty pages? This doesn't apply
2574 * to WB_SYNC_ALL mode. For integrity sync we have to
2575 * keep going because someone may be concurrently
2576 * dirtying pages, and we might have synced a lot of
2577 * newly appeared dirty pages, but have not synced all
2578 * of the old dirty pages.
2580 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2583 /* If we can't merge this page, we are done. */
2584 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2589 * If the page is no longer dirty, or its mapping no
2590 * longer corresponds to inode we are writing (which
2591 * means it has been truncated or invalidated), or the
2592 * page is already under writeback and we are not doing
2593 * a data integrity writeback, skip the page
2595 if (!PageDirty(page) ||
2596 (PageWriteback(page) &&
2597 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2598 unlikely(page->mapping != mapping)) {
2603 wait_on_page_writeback(page);
2604 BUG_ON(PageWriteback(page));
2606 if (mpd->map.m_len == 0)
2607 mpd->first_page = page->index;
2608 mpd->next_page = page->index + 1;
2609 /* Add all dirty buffers to mpd */
2610 lblk = ((ext4_lblk_t)page->index) <<
2611 (PAGE_SHIFT - blkbits);
2612 head = page_buffers(page);
2613 err = mpage_process_page_bufs(mpd, head, head, lblk);
2619 pagevec_release(&pvec);
2622 mpd->scanned_until_end = 1;
2625 pagevec_release(&pvec);
2629 static int ext4_writepages(struct address_space *mapping,
2630 struct writeback_control *wbc)
2632 pgoff_t writeback_index = 0;
2633 long nr_to_write = wbc->nr_to_write;
2634 int range_whole = 0;
2636 handle_t *handle = NULL;
2637 struct mpage_da_data mpd;
2638 struct inode *inode = mapping->host;
2639 int needed_blocks, rsv_blocks = 0, ret = 0;
2640 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2641 struct blk_plug plug;
2642 bool give_up_on_write = false;
2644 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2647 percpu_down_read(&sbi->s_writepages_rwsem);
2648 trace_ext4_writepages(inode, wbc);
2651 * No pages to write? This is mainly a kludge to avoid starting
2652 * a transaction for special inodes like journal inode on last iput()
2653 * because that could violate lock ordering on umount
2655 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2656 goto out_writepages;
2658 if (ext4_should_journal_data(inode)) {
2659 ret = generic_writepages(mapping, wbc);
2660 goto out_writepages;
2664 * If the filesystem has aborted, it is read-only, so return
2665 * right away instead of dumping stack traces later on that
2666 * will obscure the real source of the problem. We test
2667 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2668 * the latter could be true if the filesystem is mounted
2669 * read-only, and in that case, ext4_writepages should
2670 * *never* be called, so if that ever happens, we would want
2673 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2674 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2676 goto out_writepages;
2680 * If we have inline data and arrive here, it means that
2681 * we will soon create the block for the 1st page, so
2682 * we'd better clear the inline data here.
2684 if (ext4_has_inline_data(inode)) {
2685 /* Just inode will be modified... */
2686 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2687 if (IS_ERR(handle)) {
2688 ret = PTR_ERR(handle);
2689 goto out_writepages;
2691 BUG_ON(ext4_test_inode_state(inode,
2692 EXT4_STATE_MAY_INLINE_DATA));
2693 ext4_destroy_inline_data(handle, inode);
2694 ext4_journal_stop(handle);
2697 if (ext4_should_dioread_nolock(inode)) {
2699 * We may need to convert up to one extent per block in
2700 * the page and we may dirty the inode.
2702 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2703 PAGE_SIZE >> inode->i_blkbits);
2706 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2709 if (wbc->range_cyclic) {
2710 writeback_index = mapping->writeback_index;
2711 if (writeback_index)
2713 mpd.first_page = writeback_index;
2716 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2717 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2722 ext4_io_submit_init(&mpd.io_submit, wbc);
2724 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2725 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2726 blk_start_plug(&plug);
2729 * First writeback pages that don't need mapping - we can avoid
2730 * starting a transaction unnecessarily and also avoid being blocked
2731 * in the block layer on device congestion while having transaction
2735 mpd.scanned_until_end = 0;
2736 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2737 if (!mpd.io_submit.io_end) {
2741 ret = mpage_prepare_extent_to_map(&mpd);
2742 /* Unlock pages we didn't use */
2743 mpage_release_unused_pages(&mpd, false);
2744 /* Submit prepared bio */
2745 ext4_io_submit(&mpd.io_submit);
2746 ext4_put_io_end_defer(mpd.io_submit.io_end);
2747 mpd.io_submit.io_end = NULL;
2751 while (!mpd.scanned_until_end && wbc->nr_to_write > 0) {
2752 /* For each extent of pages we use new io_end */
2753 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2754 if (!mpd.io_submit.io_end) {
2760 * We have two constraints: We find one extent to map and we
2761 * must always write out whole page (makes a difference when
2762 * blocksize < pagesize) so that we don't block on IO when we
2763 * try to write out the rest of the page. Journalled mode is
2764 * not supported by delalloc.
2766 BUG_ON(ext4_should_journal_data(inode));
2767 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2769 /* start a new transaction */
2770 handle = ext4_journal_start_with_reserve(inode,
2771 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2772 if (IS_ERR(handle)) {
2773 ret = PTR_ERR(handle);
2774 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2775 "%ld pages, ino %lu; err %d", __func__,
2776 wbc->nr_to_write, inode->i_ino, ret);
2777 /* Release allocated io_end */
2778 ext4_put_io_end(mpd.io_submit.io_end);
2779 mpd.io_submit.io_end = NULL;
2784 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2785 ret = mpage_prepare_extent_to_map(&mpd);
2786 if (!ret && mpd.map.m_len)
2787 ret = mpage_map_and_submit_extent(handle, &mpd,
2790 * Caution: If the handle is synchronous,
2791 * ext4_journal_stop() can wait for transaction commit
2792 * to finish which may depend on writeback of pages to
2793 * complete or on page lock to be released. In that
2794 * case, we have to wait until after we have
2795 * submitted all the IO, released page locks we hold,
2796 * and dropped io_end reference (for extent conversion
2797 * to be able to complete) before stopping the handle.
2799 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2800 ext4_journal_stop(handle);
2804 /* Unlock pages we didn't use */
2805 mpage_release_unused_pages(&mpd, give_up_on_write);
2806 /* Submit prepared bio */
2807 ext4_io_submit(&mpd.io_submit);
2810 * Drop our io_end reference we got from init. We have
2811 * to be careful and use deferred io_end finishing if
2812 * we are still holding the transaction as we can
2813 * release the last reference to io_end which may end
2814 * up doing unwritten extent conversion.
2817 ext4_put_io_end_defer(mpd.io_submit.io_end);
2818 ext4_journal_stop(handle);
2820 ext4_put_io_end(mpd.io_submit.io_end);
2821 mpd.io_submit.io_end = NULL;
2823 if (ret == -ENOSPC && sbi->s_journal) {
2825 * Commit the transaction which would
2826 * free blocks released in the transaction
2829 jbd2_journal_force_commit_nested(sbi->s_journal);
2833 /* Fatal error - ENOMEM, EIO... */
2838 blk_finish_plug(&plug);
2839 if (!ret && !cycled && wbc->nr_to_write > 0) {
2841 mpd.last_page = writeback_index - 1;
2847 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2849 * Set the writeback_index so that range_cyclic
2850 * mode will write it back later
2852 mapping->writeback_index = mpd.first_page;
2855 trace_ext4_writepages_result(inode, wbc, ret,
2856 nr_to_write - wbc->nr_to_write);
2857 percpu_up_read(&sbi->s_writepages_rwsem);
2861 static int ext4_dax_writepages(struct address_space *mapping,
2862 struct writeback_control *wbc)
2865 long nr_to_write = wbc->nr_to_write;
2866 struct inode *inode = mapping->host;
2867 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2869 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2872 percpu_down_read(&sbi->s_writepages_rwsem);
2873 trace_ext4_writepages(inode, wbc);
2875 ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
2876 trace_ext4_writepages_result(inode, wbc, ret,
2877 nr_to_write - wbc->nr_to_write);
2878 percpu_up_read(&sbi->s_writepages_rwsem);
2882 static int ext4_nonda_switch(struct super_block *sb)
2884 s64 free_clusters, dirty_clusters;
2885 struct ext4_sb_info *sbi = EXT4_SB(sb);
2888 * switch to non delalloc mode if we are running low
2889 * on free block. The free block accounting via percpu
2890 * counters can get slightly wrong with percpu_counter_batch getting
2891 * accumulated on each CPU without updating global counters
2892 * Delalloc need an accurate free block accounting. So switch
2893 * to non delalloc when we are near to error range.
2896 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2898 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2900 * Start pushing delalloc when 1/2 of free blocks are dirty.
2902 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2903 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2905 if (2 * free_clusters < 3 * dirty_clusters ||
2906 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2908 * free block count is less than 150% of dirty blocks
2909 * or free blocks is less than watermark
2916 /* We always reserve for an inode update; the superblock could be there too */
2917 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2919 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2922 if (pos + len <= 0x7fffffffULL)
2925 /* We might need to update the superblock to set LARGE_FILE */
2929 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2930 loff_t pos, unsigned len, unsigned flags,
2931 struct page **pagep, void **fsdata)
2933 int ret, retries = 0;
2936 struct inode *inode = mapping->host;
2939 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2942 index = pos >> PAGE_SHIFT;
2944 if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) ||
2945 ext4_verity_in_progress(inode)) {
2946 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2947 return ext4_write_begin(file, mapping, pos,
2948 len, flags, pagep, fsdata);
2950 *fsdata = (void *)0;
2951 trace_ext4_da_write_begin(inode, pos, len, flags);
2953 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2954 ret = ext4_da_write_inline_data_begin(mapping, inode,
2964 * grab_cache_page_write_begin() can take a long time if the
2965 * system is thrashing due to memory pressure, or if the page
2966 * is being written back. So grab it first before we start
2967 * the transaction handle. This also allows us to allocate
2968 * the page (if needed) without using GFP_NOFS.
2971 page = grab_cache_page_write_begin(mapping, index, flags);
2977 * With delayed allocation, we don't log the i_disksize update
2978 * if there is delayed block allocation. But we still need
2979 * to journalling the i_disksize update if writes to the end
2980 * of file which has an already mapped buffer.
2983 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2984 ext4_da_write_credits(inode, pos, len));
2985 if (IS_ERR(handle)) {
2987 return PTR_ERR(handle);
2991 if (page->mapping != mapping) {
2992 /* The page got truncated from under us */
2995 ext4_journal_stop(handle);
2998 /* In case writeback began while the page was unlocked */
2999 wait_for_stable_page(page);
3001 #ifdef CONFIG_FS_ENCRYPTION
3002 ret = ext4_block_write_begin(page, pos, len,
3003 ext4_da_get_block_prep);
3005 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3009 ext4_journal_stop(handle);
3011 * block_write_begin may have instantiated a few blocks
3012 * outside i_size. Trim these off again. Don't need
3013 * i_size_read because we hold i_mutex.
3015 if (pos + len > inode->i_size)
3016 ext4_truncate_failed_write(inode);
3018 if (ret == -ENOSPC &&
3019 ext4_should_retry_alloc(inode->i_sb, &retries))
3031 * Check if we should update i_disksize
3032 * when write to the end of file but not require block allocation
3034 static int ext4_da_should_update_i_disksize(struct page *page,
3035 unsigned long offset)
3037 struct buffer_head *bh;
3038 struct inode *inode = page->mapping->host;
3042 bh = page_buffers(page);
3043 idx = offset >> inode->i_blkbits;
3045 for (i = 0; i < idx; i++)
3046 bh = bh->b_this_page;
3048 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3053 static int ext4_da_write_end(struct file *file,
3054 struct address_space *mapping,
3055 loff_t pos, unsigned len, unsigned copied,
3056 struct page *page, void *fsdata)
3058 struct inode *inode = mapping->host;
3060 handle_t *handle = ext4_journal_current_handle();
3062 unsigned long start, end;
3063 int write_mode = (int)(unsigned long)fsdata;
3065 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3066 return ext4_write_end(file, mapping, pos,
3067 len, copied, page, fsdata);
3069 trace_ext4_da_write_end(inode, pos, len, copied);
3070 start = pos & (PAGE_SIZE - 1);
3071 end = start + copied - 1;
3074 * generic_write_end() will run mark_inode_dirty() if i_size
3075 * changes. So let's piggyback the i_disksize mark_inode_dirty
3078 new_i_size = pos + copied;
3079 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3080 if (ext4_has_inline_data(inode) ||
3081 ext4_da_should_update_i_disksize(page, end)) {
3082 ext4_update_i_disksize(inode, new_i_size);
3083 /* We need to mark inode dirty even if
3084 * new_i_size is less that inode->i_size
3085 * bu greater than i_disksize.(hint delalloc)
3087 ret = ext4_mark_inode_dirty(handle, inode);
3091 if (write_mode != CONVERT_INLINE_DATA &&
3092 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3093 ext4_has_inline_data(inode))
3094 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3097 ret2 = generic_write_end(file, mapping, pos, len, copied,
3103 ret2 = ext4_journal_stop(handle);
3104 if (unlikely(ret2 && !ret))
3107 return ret ? ret : copied;
3111 * Force all delayed allocation blocks to be allocated for a given inode.
3113 int ext4_alloc_da_blocks(struct inode *inode)
3115 trace_ext4_alloc_da_blocks(inode);
3117 if (!EXT4_I(inode)->i_reserved_data_blocks)
3121 * We do something simple for now. The filemap_flush() will
3122 * also start triggering a write of the data blocks, which is
3123 * not strictly speaking necessary (and for users of
3124 * laptop_mode, not even desirable). However, to do otherwise
3125 * would require replicating code paths in:
3127 * ext4_writepages() ->
3128 * write_cache_pages() ---> (via passed in callback function)
3129 * __mpage_da_writepage() -->
3130 * mpage_add_bh_to_extent()
3131 * mpage_da_map_blocks()
3133 * The problem is that write_cache_pages(), located in
3134 * mm/page-writeback.c, marks pages clean in preparation for
3135 * doing I/O, which is not desirable if we're not planning on
3138 * We could call write_cache_pages(), and then redirty all of
3139 * the pages by calling redirty_page_for_writepage() but that
3140 * would be ugly in the extreme. So instead we would need to
3141 * replicate parts of the code in the above functions,
3142 * simplifying them because we wouldn't actually intend to
3143 * write out the pages, but rather only collect contiguous
3144 * logical block extents, call the multi-block allocator, and
3145 * then update the buffer heads with the block allocations.
3147 * For now, though, we'll cheat by calling filemap_flush(),
3148 * which will map the blocks, and start the I/O, but not
3149 * actually wait for the I/O to complete.
3151 return filemap_flush(inode->i_mapping);
3155 * bmap() is special. It gets used by applications such as lilo and by
3156 * the swapper to find the on-disk block of a specific piece of data.
3158 * Naturally, this is dangerous if the block concerned is still in the
3159 * journal. If somebody makes a swapfile on an ext4 data-journaling
3160 * filesystem and enables swap, then they may get a nasty shock when the
3161 * data getting swapped to that swapfile suddenly gets overwritten by
3162 * the original zero's written out previously to the journal and
3163 * awaiting writeback in the kernel's buffer cache.
3165 * So, if we see any bmap calls here on a modified, data-journaled file,
3166 * take extra steps to flush any blocks which might be in the cache.
3168 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3170 struct inode *inode = mapping->host;
3175 * We can get here for an inline file via the FIBMAP ioctl
3177 if (ext4_has_inline_data(inode))
3180 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3181 test_opt(inode->i_sb, DELALLOC)) {
3183 * With delalloc we want to sync the file
3184 * so that we can make sure we allocate
3187 filemap_write_and_wait(mapping);
3190 if (EXT4_JOURNAL(inode) &&
3191 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3193 * This is a REALLY heavyweight approach, but the use of
3194 * bmap on dirty files is expected to be extremely rare:
3195 * only if we run lilo or swapon on a freshly made file
3196 * do we expect this to happen.
3198 * (bmap requires CAP_SYS_RAWIO so this does not
3199 * represent an unprivileged user DOS attack --- we'd be
3200 * in trouble if mortal users could trigger this path at
3203 * NB. EXT4_STATE_JDATA is not set on files other than
3204 * regular files. If somebody wants to bmap a directory
3205 * or symlink and gets confused because the buffer
3206 * hasn't yet been flushed to disk, they deserve
3207 * everything they get.
3210 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3211 journal = EXT4_JOURNAL(inode);
3212 jbd2_journal_lock_updates(journal);
3213 err = jbd2_journal_flush(journal);
3214 jbd2_journal_unlock_updates(journal);
3220 return iomap_bmap(mapping, block, &ext4_iomap_ops);
3223 static int ext4_readpage(struct file *file, struct page *page)
3226 struct inode *inode = page->mapping->host;
3228 trace_ext4_readpage(page);
3230 if (ext4_has_inline_data(inode))
3231 ret = ext4_readpage_inline(inode, page);
3234 return ext4_mpage_readpages(inode, NULL, page);
3239 static void ext4_readahead(struct readahead_control *rac)
3241 struct inode *inode = rac->mapping->host;
3243 /* If the file has inline data, no need to do readahead. */
3244 if (ext4_has_inline_data(inode))
3247 ext4_mpage_readpages(inode, rac, NULL);
3250 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3251 unsigned int length)
3253 trace_ext4_invalidatepage(page, offset, length);
3255 /* No journalling happens on data buffers when this function is used */
3256 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3258 block_invalidatepage(page, offset, length);
3261 static int __ext4_journalled_invalidatepage(struct page *page,
3262 unsigned int offset,
3263 unsigned int length)
3265 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3267 trace_ext4_journalled_invalidatepage(page, offset, length);
3270 * If it's a full truncate we just forget about the pending dirtying
3272 if (offset == 0 && length == PAGE_SIZE)
3273 ClearPageChecked(page);
3275 return jbd2_journal_invalidatepage(journal, page, offset, length);
3278 /* Wrapper for aops... */
3279 static void ext4_journalled_invalidatepage(struct page *page,
3280 unsigned int offset,
3281 unsigned int length)
3283 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3286 static int ext4_releasepage(struct page *page, gfp_t wait)
3288 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3290 trace_ext4_releasepage(page);
3292 /* Page has dirty journalled data -> cannot release */
3293 if (PageChecked(page))
3296 return jbd2_journal_try_to_free_buffers(journal, page);
3298 return try_to_free_buffers(page);
3301 static bool ext4_inode_datasync_dirty(struct inode *inode)
3303 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3306 if (jbd2_transaction_committed(journal,
3307 EXT4_I(inode)->i_datasync_tid))
3309 return atomic_read(&EXT4_SB(inode->i_sb)->s_fc_subtid) >=
3310 EXT4_I(inode)->i_fc_committed_subtid;
3313 /* Any metadata buffers to write? */
3314 if (!list_empty(&inode->i_mapping->private_list))
3316 return inode->i_state & I_DIRTY_DATASYNC;
3319 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3320 struct ext4_map_blocks *map, loff_t offset,
3323 u8 blkbits = inode->i_blkbits;
3326 * Writes that span EOF might trigger an I/O size update on completion,
3327 * so consider them to be dirty for the purpose of O_DSYNC, even if
3328 * there is no other metadata changes being made or are pending.
3331 if (ext4_inode_datasync_dirty(inode) ||
3332 offset + length > i_size_read(inode))
3333 iomap->flags |= IOMAP_F_DIRTY;
3335 if (map->m_flags & EXT4_MAP_NEW)
3336 iomap->flags |= IOMAP_F_NEW;
3338 iomap->bdev = inode->i_sb->s_bdev;
3339 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3340 iomap->offset = (u64) map->m_lblk << blkbits;
3341 iomap->length = (u64) map->m_len << blkbits;
3343 if ((map->m_flags & EXT4_MAP_MAPPED) &&
3344 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3345 iomap->flags |= IOMAP_F_MERGED;
3348 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3349 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3350 * set. In order for any allocated unwritten extents to be converted
3351 * into written extents correctly within the ->end_io() handler, we
3352 * need to ensure that the iomap->type is set appropriately. Hence, the
3353 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3356 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3357 iomap->type = IOMAP_UNWRITTEN;
3358 iomap->addr = (u64) map->m_pblk << blkbits;
3359 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3360 iomap->type = IOMAP_MAPPED;
3361 iomap->addr = (u64) map->m_pblk << blkbits;
3363 iomap->type = IOMAP_HOLE;
3364 iomap->addr = IOMAP_NULL_ADDR;
3368 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3372 u8 blkbits = inode->i_blkbits;
3373 int ret, dio_credits, m_flags = 0, retries = 0;
3376 * Trim the mapping request to the maximum value that we can map at
3377 * once for direct I/O.
3379 if (map->m_len > DIO_MAX_BLOCKS)
3380 map->m_len = DIO_MAX_BLOCKS;
3381 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3385 * Either we allocate blocks and then don't get an unwritten extent, so
3386 * in that case we have reserved enough credits. Or, the blocks are
3387 * already allocated and unwritten. In that case, the extent conversion
3388 * fits into the credits as well.
3390 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3392 return PTR_ERR(handle);
3395 * DAX and direct I/O are the only two operations that are currently
3396 * supported with IOMAP_WRITE.
3398 WARN_ON(!IS_DAX(inode) && !(flags & IOMAP_DIRECT));
3400 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3402 * We use i_size instead of i_disksize here because delalloc writeback
3403 * can complete at any point during the I/O and subsequently push the
3404 * i_disksize out to i_size. This could be beyond where direct I/O is
3405 * happening and thus expose allocated blocks to direct I/O reads.
3407 else if ((map->m_lblk * (1 << blkbits)) >= i_size_read(inode))
3408 m_flags = EXT4_GET_BLOCKS_CREATE;
3409 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3410 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3412 ret = ext4_map_blocks(handle, inode, map, m_flags);
3415 * We cannot fill holes in indirect tree based inodes as that could
3416 * expose stale data in the case of a crash. Use the magic error code
3417 * to fallback to buffered I/O.
3419 if (!m_flags && !ret)
3422 ext4_journal_stop(handle);
3423 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3430 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3431 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3434 struct ext4_map_blocks map;
3435 u8 blkbits = inode->i_blkbits;
3437 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3440 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3444 * Calculate the first and last logical blocks respectively.
3446 map.m_lblk = offset >> blkbits;
3447 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3448 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3450 if (flags & IOMAP_WRITE) {
3452 * We check here if the blocks are already allocated, then we
3453 * don't need to start a journal txn and we can directly return
3454 * the mapping information. This could boost performance
3455 * especially in multi-threaded overwrite requests.
3457 if (offset + length <= i_size_read(inode)) {
3458 ret = ext4_map_blocks(NULL, inode, &map, 0);
3459 if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3462 ret = ext4_iomap_alloc(inode, &map, flags);
3464 ret = ext4_map_blocks(NULL, inode, &map, 0);
3470 ext4_set_iomap(inode, iomap, &map, offset, length);
3475 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3476 loff_t length, unsigned flags, struct iomap *iomap,
3477 struct iomap *srcmap)
3482 * Even for writes we don't need to allocate blocks, so just pretend
3483 * we are reading to save overhead of starting a transaction.
3485 flags &= ~IOMAP_WRITE;
3486 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3487 WARN_ON_ONCE(iomap->type != IOMAP_MAPPED);
3491 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3492 ssize_t written, unsigned flags, struct iomap *iomap)
3495 * Check to see whether an error occurred while writing out the data to
3496 * the allocated blocks. If so, return the magic error code so that we
3497 * fallback to buffered I/O and attempt to complete the remainder of
3498 * the I/O. Any blocks that may have been allocated in preparation for
3499 * the direct I/O will be reused during buffered I/O.
3501 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3507 const struct iomap_ops ext4_iomap_ops = {
3508 .iomap_begin = ext4_iomap_begin,
3509 .iomap_end = ext4_iomap_end,
3512 const struct iomap_ops ext4_iomap_overwrite_ops = {
3513 .iomap_begin = ext4_iomap_overwrite_begin,
3514 .iomap_end = ext4_iomap_end,
3517 static bool ext4_iomap_is_delalloc(struct inode *inode,
3518 struct ext4_map_blocks *map)
3520 struct extent_status es;
3521 ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3523 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3524 map->m_lblk, end, &es);
3526 if (!es.es_len || es.es_lblk > end)
3529 if (es.es_lblk > map->m_lblk) {
3530 map->m_len = es.es_lblk - map->m_lblk;
3534 offset = map->m_lblk - es.es_lblk;
3535 map->m_len = es.es_len - offset;
3540 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3541 loff_t length, unsigned int flags,
3542 struct iomap *iomap, struct iomap *srcmap)
3545 bool delalloc = false;
3546 struct ext4_map_blocks map;
3547 u8 blkbits = inode->i_blkbits;
3549 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3552 if (ext4_has_inline_data(inode)) {
3553 ret = ext4_inline_data_iomap(inode, iomap);
3554 if (ret != -EAGAIN) {
3555 if (ret == 0 && offset >= iomap->length)
3562 * Calculate the first and last logical block respectively.
3564 map.m_lblk = offset >> blkbits;
3565 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3566 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3569 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3570 * So handle it here itself instead of querying ext4_map_blocks().
3571 * Since ext4_map_blocks() will warn about it and will return
3574 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3575 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3577 if (offset >= sbi->s_bitmap_maxbytes) {
3583 ret = ext4_map_blocks(NULL, inode, &map, 0);
3587 delalloc = ext4_iomap_is_delalloc(inode, &map);
3590 ext4_set_iomap(inode, iomap, &map, offset, length);
3591 if (delalloc && iomap->type == IOMAP_HOLE)
3592 iomap->type = IOMAP_DELALLOC;
3597 const struct iomap_ops ext4_iomap_report_ops = {
3598 .iomap_begin = ext4_iomap_begin_report,
3602 * Pages can be marked dirty completely asynchronously from ext4's journalling
3603 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3604 * much here because ->set_page_dirty is called under VFS locks. The page is
3605 * not necessarily locked.
3607 * We cannot just dirty the page and leave attached buffers clean, because the
3608 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3609 * or jbddirty because all the journalling code will explode.
3611 * So what we do is to mark the page "pending dirty" and next time writepage
3612 * is called, propagate that into the buffers appropriately.
3614 static int ext4_journalled_set_page_dirty(struct page *page)
3616 SetPageChecked(page);
3617 return __set_page_dirty_nobuffers(page);
3620 static int ext4_set_page_dirty(struct page *page)
3622 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3623 WARN_ON_ONCE(!page_has_buffers(page));
3624 return __set_page_dirty_buffers(page);
3627 static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3628 struct file *file, sector_t *span)
3630 return iomap_swapfile_activate(sis, file, span,
3631 &ext4_iomap_report_ops);
3634 static const struct address_space_operations ext4_aops = {
3635 .readpage = ext4_readpage,
3636 .readahead = ext4_readahead,
3637 .writepage = ext4_writepage,
3638 .writepages = ext4_writepages,
3639 .write_begin = ext4_write_begin,
3640 .write_end = ext4_write_end,
3641 .set_page_dirty = ext4_set_page_dirty,
3643 .invalidatepage = ext4_invalidatepage,
3644 .releasepage = ext4_releasepage,
3645 .direct_IO = noop_direct_IO,
3646 .migratepage = buffer_migrate_page,
3647 .is_partially_uptodate = block_is_partially_uptodate,
3648 .error_remove_page = generic_error_remove_page,
3649 .swap_activate = ext4_iomap_swap_activate,
3652 static const struct address_space_operations ext4_journalled_aops = {
3653 .readpage = ext4_readpage,
3654 .readahead = ext4_readahead,
3655 .writepage = ext4_writepage,
3656 .writepages = ext4_writepages,
3657 .write_begin = ext4_write_begin,
3658 .write_end = ext4_journalled_write_end,
3659 .set_page_dirty = ext4_journalled_set_page_dirty,
3661 .invalidatepage = ext4_journalled_invalidatepage,
3662 .releasepage = ext4_releasepage,
3663 .direct_IO = noop_direct_IO,
3664 .is_partially_uptodate = block_is_partially_uptodate,
3665 .error_remove_page = generic_error_remove_page,
3666 .swap_activate = ext4_iomap_swap_activate,
3669 static const struct address_space_operations ext4_da_aops = {
3670 .readpage = ext4_readpage,
3671 .readahead = ext4_readahead,
3672 .writepage = ext4_writepage,
3673 .writepages = ext4_writepages,
3674 .write_begin = ext4_da_write_begin,
3675 .write_end = ext4_da_write_end,
3676 .set_page_dirty = ext4_set_page_dirty,
3678 .invalidatepage = ext4_invalidatepage,
3679 .releasepage = ext4_releasepage,
3680 .direct_IO = noop_direct_IO,
3681 .migratepage = buffer_migrate_page,
3682 .is_partially_uptodate = block_is_partially_uptodate,
3683 .error_remove_page = generic_error_remove_page,
3684 .swap_activate = ext4_iomap_swap_activate,
3687 static const struct address_space_operations ext4_dax_aops = {
3688 .writepages = ext4_dax_writepages,
3689 .direct_IO = noop_direct_IO,
3690 .set_page_dirty = noop_set_page_dirty,
3692 .invalidatepage = noop_invalidatepage,
3693 .swap_activate = ext4_iomap_swap_activate,
3696 void ext4_set_aops(struct inode *inode)
3698 switch (ext4_inode_journal_mode(inode)) {
3699 case EXT4_INODE_ORDERED_DATA_MODE:
3700 case EXT4_INODE_WRITEBACK_DATA_MODE:
3702 case EXT4_INODE_JOURNAL_DATA_MODE:
3703 inode->i_mapping->a_ops = &ext4_journalled_aops;
3709 inode->i_mapping->a_ops = &ext4_dax_aops;
3710 else if (test_opt(inode->i_sb, DELALLOC))
3711 inode->i_mapping->a_ops = &ext4_da_aops;
3713 inode->i_mapping->a_ops = &ext4_aops;
3716 static int __ext4_block_zero_page_range(handle_t *handle,
3717 struct address_space *mapping, loff_t from, loff_t length)
3719 ext4_fsblk_t index = from >> PAGE_SHIFT;
3720 unsigned offset = from & (PAGE_SIZE-1);
3721 unsigned blocksize, pos;
3723 struct inode *inode = mapping->host;
3724 struct buffer_head *bh;
3728 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3729 mapping_gfp_constraint(mapping, ~__GFP_FS));
3733 blocksize = inode->i_sb->s_blocksize;
3735 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3737 if (!page_has_buffers(page))
3738 create_empty_buffers(page, blocksize, 0);
3740 /* Find the buffer that contains "offset" */
3741 bh = page_buffers(page);
3743 while (offset >= pos) {
3744 bh = bh->b_this_page;
3748 if (buffer_freed(bh)) {
3749 BUFFER_TRACE(bh, "freed: skip");
3752 if (!buffer_mapped(bh)) {
3753 BUFFER_TRACE(bh, "unmapped");
3754 ext4_get_block(inode, iblock, bh, 0);
3755 /* unmapped? It's a hole - nothing to do */
3756 if (!buffer_mapped(bh)) {
3757 BUFFER_TRACE(bh, "still unmapped");
3762 /* Ok, it's mapped. Make sure it's up-to-date */
3763 if (PageUptodate(page))
3764 set_buffer_uptodate(bh);
3766 if (!buffer_uptodate(bh)) {
3767 err = ext4_read_bh_lock(bh, 0, true);
3770 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3771 /* We expect the key to be set. */
3772 BUG_ON(!fscrypt_has_encryption_key(inode));
3773 err = fscrypt_decrypt_pagecache_blocks(page, blocksize,
3776 clear_buffer_uptodate(bh);
3781 if (ext4_should_journal_data(inode)) {
3782 BUFFER_TRACE(bh, "get write access");
3783 err = ext4_journal_get_write_access(handle, bh);
3787 zero_user(page, offset, length);
3788 BUFFER_TRACE(bh, "zeroed end of block");
3790 if (ext4_should_journal_data(inode)) {
3791 err = ext4_handle_dirty_metadata(handle, inode, bh);
3794 mark_buffer_dirty(bh);
3795 if (ext4_should_order_data(inode))
3796 err = ext4_jbd2_inode_add_write(handle, inode, from,
3807 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3808 * starting from file offset 'from'. The range to be zero'd must
3809 * be contained with in one block. If the specified range exceeds
3810 * the end of the block it will be shortened to end of the block
3811 * that cooresponds to 'from'
3813 static int ext4_block_zero_page_range(handle_t *handle,
3814 struct address_space *mapping, loff_t from, loff_t length)
3816 struct inode *inode = mapping->host;
3817 unsigned offset = from & (PAGE_SIZE-1);
3818 unsigned blocksize = inode->i_sb->s_blocksize;
3819 unsigned max = blocksize - (offset & (blocksize - 1));
3822 * correct length if it does not fall between
3823 * 'from' and the end of the block
3825 if (length > max || length < 0)
3828 if (IS_DAX(inode)) {
3829 return iomap_zero_range(inode, from, length, NULL,
3832 return __ext4_block_zero_page_range(handle, mapping, from, length);
3836 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3837 * up to the end of the block which corresponds to `from'.
3838 * This required during truncate. We need to physically zero the tail end
3839 * of that block so it doesn't yield old data if the file is later grown.
3841 static int ext4_block_truncate_page(handle_t *handle,
3842 struct address_space *mapping, loff_t from)
3844 unsigned offset = from & (PAGE_SIZE-1);
3847 struct inode *inode = mapping->host;
3849 /* If we are processing an encrypted inode during orphan list handling */
3850 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3853 blocksize = inode->i_sb->s_blocksize;
3854 length = blocksize - (offset & (blocksize - 1));
3856 return ext4_block_zero_page_range(handle, mapping, from, length);
3859 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3860 loff_t lstart, loff_t length)
3862 struct super_block *sb = inode->i_sb;
3863 struct address_space *mapping = inode->i_mapping;
3864 unsigned partial_start, partial_end;
3865 ext4_fsblk_t start, end;
3866 loff_t byte_end = (lstart + length - 1);
3869 partial_start = lstart & (sb->s_blocksize - 1);
3870 partial_end = byte_end & (sb->s_blocksize - 1);
3872 start = lstart >> sb->s_blocksize_bits;
3873 end = byte_end >> sb->s_blocksize_bits;
3875 /* Handle partial zero within the single block */
3877 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3878 err = ext4_block_zero_page_range(handle, mapping,
3882 /* Handle partial zero out on the start of the range */
3883 if (partial_start) {
3884 err = ext4_block_zero_page_range(handle, mapping,
3885 lstart, sb->s_blocksize);
3889 /* Handle partial zero out on the end of the range */
3890 if (partial_end != sb->s_blocksize - 1)
3891 err = ext4_block_zero_page_range(handle, mapping,
3892 byte_end - partial_end,
3897 int ext4_can_truncate(struct inode *inode)
3899 if (S_ISREG(inode->i_mode))
3901 if (S_ISDIR(inode->i_mode))
3903 if (S_ISLNK(inode->i_mode))
3904 return !ext4_inode_is_fast_symlink(inode);
3909 * We have to make sure i_disksize gets properly updated before we truncate
3910 * page cache due to hole punching or zero range. Otherwise i_disksize update
3911 * can get lost as it may have been postponed to submission of writeback but
3912 * that will never happen after we truncate page cache.
3914 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3920 loff_t size = i_size_read(inode);
3922 WARN_ON(!inode_is_locked(inode));
3923 if (offset > size || offset + len < size)
3926 if (EXT4_I(inode)->i_disksize >= size)
3929 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3931 return PTR_ERR(handle);
3932 ext4_update_i_disksize(inode, size);
3933 ret = ext4_mark_inode_dirty(handle, inode);
3934 ext4_journal_stop(handle);
3939 static void ext4_wait_dax_page(struct ext4_inode_info *ei)
3941 up_write(&ei->i_mmap_sem);
3943 down_write(&ei->i_mmap_sem);
3946 int ext4_break_layouts(struct inode *inode)
3948 struct ext4_inode_info *ei = EXT4_I(inode);
3952 if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
3956 page = dax_layout_busy_page(inode->i_mapping);
3960 error = ___wait_var_event(&page->_refcount,
3961 atomic_read(&page->_refcount) == 1,
3962 TASK_INTERRUPTIBLE, 0, 0,
3963 ext4_wait_dax_page(ei));
3964 } while (error == 0);
3970 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3971 * associated with the given offset and length
3973 * @inode: File inode
3974 * @offset: The offset where the hole will begin
3975 * @len: The length of the hole
3977 * Returns: 0 on success or negative on failure
3980 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3982 struct super_block *sb = inode->i_sb;
3983 ext4_lblk_t first_block, stop_block;
3984 struct address_space *mapping = inode->i_mapping;
3985 loff_t first_block_offset, last_block_offset;
3987 unsigned int credits;
3988 int ret = 0, ret2 = 0;
3990 trace_ext4_punch_hole(inode, offset, length, 0);
3992 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
3993 if (ext4_has_inline_data(inode)) {
3994 down_write(&EXT4_I(inode)->i_mmap_sem);
3995 ret = ext4_convert_inline_data(inode);
3996 up_write(&EXT4_I(inode)->i_mmap_sem);
4002 * Write out all dirty pages to avoid race conditions
4003 * Then release them.
4005 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4006 ret = filemap_write_and_wait_range(mapping, offset,
4007 offset + length - 1);
4014 /* No need to punch hole beyond i_size */
4015 if (offset >= inode->i_size)
4019 * If the hole extends beyond i_size, set the hole
4020 * to end after the page that contains i_size
4022 if (offset + length > inode->i_size) {
4023 length = inode->i_size +
4024 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4028 if (offset & (sb->s_blocksize - 1) ||
4029 (offset + length) & (sb->s_blocksize - 1)) {
4031 * Attach jinode to inode for jbd2 if we do any zeroing of
4034 ret = ext4_inode_attach_jinode(inode);
4040 /* Wait all existing dio workers, newcomers will block on i_mutex */
4041 inode_dio_wait(inode);
4044 * Prevent page faults from reinstantiating pages we have released from
4047 down_write(&EXT4_I(inode)->i_mmap_sem);
4049 ret = ext4_break_layouts(inode);
4053 first_block_offset = round_up(offset, sb->s_blocksize);
4054 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4056 /* Now release the pages and zero block aligned part of pages*/
4057 if (last_block_offset > first_block_offset) {
4058 ret = ext4_update_disksize_before_punch(inode, offset, length);
4061 truncate_pagecache_range(inode, first_block_offset,
4065 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4066 credits = ext4_writepage_trans_blocks(inode);
4068 credits = ext4_blocks_for_truncate(inode);
4069 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4070 if (IS_ERR(handle)) {
4071 ret = PTR_ERR(handle);
4072 ext4_std_error(sb, ret);
4076 ret = ext4_zero_partial_blocks(handle, inode, offset,
4081 first_block = (offset + sb->s_blocksize - 1) >>
4082 EXT4_BLOCK_SIZE_BITS(sb);
4083 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4085 /* If there are blocks to remove, do it */
4086 if (stop_block > first_block) {
4088 down_write(&EXT4_I(inode)->i_data_sem);
4089 ext4_discard_preallocations(inode, 0);
4091 ret = ext4_es_remove_extent(inode, first_block,
4092 stop_block - first_block);
4094 up_write(&EXT4_I(inode)->i_data_sem);
4098 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4099 ret = ext4_ext_remove_space(inode, first_block,
4102 ret = ext4_ind_remove_space(handle, inode, first_block,
4105 up_write(&EXT4_I(inode)->i_data_sem);
4107 ext4_fc_track_range(inode, first_block, stop_block);
4109 ext4_handle_sync(handle);
4111 inode->i_mtime = inode->i_ctime = current_time(inode);
4112 ret2 = ext4_mark_inode_dirty(handle, inode);
4116 ext4_update_inode_fsync_trans(handle, inode, 1);
4118 ext4_journal_stop(handle);
4120 up_write(&EXT4_I(inode)->i_mmap_sem);
4122 inode_unlock(inode);
4126 int ext4_inode_attach_jinode(struct inode *inode)
4128 struct ext4_inode_info *ei = EXT4_I(inode);
4129 struct jbd2_inode *jinode;
4131 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4134 jinode = jbd2_alloc_inode(GFP_KERNEL);
4135 spin_lock(&inode->i_lock);
4138 spin_unlock(&inode->i_lock);
4141 ei->jinode = jinode;
4142 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4145 spin_unlock(&inode->i_lock);
4146 if (unlikely(jinode != NULL))
4147 jbd2_free_inode(jinode);
4154 * We block out ext4_get_block() block instantiations across the entire
4155 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4156 * simultaneously on behalf of the same inode.
4158 * As we work through the truncate and commit bits of it to the journal there
4159 * is one core, guiding principle: the file's tree must always be consistent on
4160 * disk. We must be able to restart the truncate after a crash.
4162 * The file's tree may be transiently inconsistent in memory (although it
4163 * probably isn't), but whenever we close off and commit a journal transaction,
4164 * the contents of (the filesystem + the journal) must be consistent and
4165 * restartable. It's pretty simple, really: bottom up, right to left (although
4166 * left-to-right works OK too).
4168 * Note that at recovery time, journal replay occurs *before* the restart of
4169 * truncate against the orphan inode list.
4171 * The committed inode has the new, desired i_size (which is the same as
4172 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4173 * that this inode's truncate did not complete and it will again call
4174 * ext4_truncate() to have another go. So there will be instantiated blocks
4175 * to the right of the truncation point in a crashed ext4 filesystem. But
4176 * that's fine - as long as they are linked from the inode, the post-crash
4177 * ext4_truncate() run will find them and release them.
4179 int ext4_truncate(struct inode *inode)
4181 struct ext4_inode_info *ei = EXT4_I(inode);
4182 unsigned int credits;
4185 struct address_space *mapping = inode->i_mapping;
4188 * There is a possibility that we're either freeing the inode
4189 * or it's a completely new inode. In those cases we might not
4190 * have i_mutex locked because it's not necessary.
4192 if (!(inode->i_state & (I_NEW|I_FREEING)))
4193 WARN_ON(!inode_is_locked(inode));
4194 trace_ext4_truncate_enter(inode);
4196 if (!ext4_can_truncate(inode))
4199 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4200 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4202 if (ext4_has_inline_data(inode)) {
4205 err = ext4_inline_data_truncate(inode, &has_inline);
4206 if (err || has_inline)
4210 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4211 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4212 if (ext4_inode_attach_jinode(inode) < 0)
4216 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4217 credits = ext4_writepage_trans_blocks(inode);
4219 credits = ext4_blocks_for_truncate(inode);
4221 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4222 if (IS_ERR(handle)) {
4223 err = PTR_ERR(handle);
4227 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4228 ext4_block_truncate_page(handle, mapping, inode->i_size);
4231 * We add the inode to the orphan list, so that if this
4232 * truncate spans multiple transactions, and we crash, we will
4233 * resume the truncate when the filesystem recovers. It also
4234 * marks the inode dirty, to catch the new size.
4236 * Implication: the file must always be in a sane, consistent
4237 * truncatable state while each transaction commits.
4239 err = ext4_orphan_add(handle, inode);
4243 down_write(&EXT4_I(inode)->i_data_sem);
4245 ext4_discard_preallocations(inode, 0);
4247 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4248 err = ext4_ext_truncate(handle, inode);
4250 ext4_ind_truncate(handle, inode);
4252 up_write(&ei->i_data_sem);
4257 ext4_handle_sync(handle);
4261 * If this was a simple ftruncate() and the file will remain alive,
4262 * then we need to clear up the orphan record which we created above.
4263 * However, if this was a real unlink then we were called by
4264 * ext4_evict_inode(), and we allow that function to clean up the
4265 * orphan info for us.
4268 ext4_orphan_del(handle, inode);
4270 inode->i_mtime = inode->i_ctime = current_time(inode);
4271 err2 = ext4_mark_inode_dirty(handle, inode);
4272 if (unlikely(err2 && !err))
4274 ext4_journal_stop(handle);
4277 trace_ext4_truncate_exit(inode);
4282 * ext4_get_inode_loc returns with an extra refcount against the inode's
4283 * underlying buffer_head on success. If 'in_mem' is true, we have all
4284 * data in memory that is needed to recreate the on-disk version of this
4287 static int __ext4_get_inode_loc(struct inode *inode,
4288 struct ext4_iloc *iloc, int in_mem)
4290 struct ext4_group_desc *gdp;
4291 struct buffer_head *bh;
4292 struct super_block *sb = inode->i_sb;
4294 struct blk_plug plug;
4295 int inodes_per_block, inode_offset;
4298 if (inode->i_ino < EXT4_ROOT_INO ||
4299 inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4300 return -EFSCORRUPTED;
4302 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4303 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4308 * Figure out the offset within the block group inode table
4310 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4311 inode_offset = ((inode->i_ino - 1) %
4312 EXT4_INODES_PER_GROUP(sb));
4313 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4314 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4316 bh = sb_getblk(sb, block);
4319 if (ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO))
4321 if (!buffer_uptodate(bh)) {
4324 if (ext4_buffer_uptodate(bh)) {
4325 /* someone brought it uptodate while we waited */
4331 * If we have all information of the inode in memory and this
4332 * is the only valid inode in the block, we need not read the
4336 struct buffer_head *bitmap_bh;
4339 start = inode_offset & ~(inodes_per_block - 1);
4341 /* Is the inode bitmap in cache? */
4342 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4343 if (unlikely(!bitmap_bh))
4347 * If the inode bitmap isn't in cache then the
4348 * optimisation may end up performing two reads instead
4349 * of one, so skip it.
4351 if (!buffer_uptodate(bitmap_bh)) {
4355 for (i = start; i < start + inodes_per_block; i++) {
4356 if (i == inode_offset)
4358 if (ext4_test_bit(i, bitmap_bh->b_data))
4362 if (i == start + inodes_per_block) {
4363 /* all other inodes are free, so skip I/O */
4364 memset(bh->b_data, 0, bh->b_size);
4365 set_buffer_uptodate(bh);
4373 * If we need to do any I/O, try to pre-readahead extra
4374 * blocks from the inode table.
4376 blk_start_plug(&plug);
4377 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4378 ext4_fsblk_t b, end, table;
4380 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4382 table = ext4_inode_table(sb, gdp);
4383 /* s_inode_readahead_blks is always a power of 2 */
4384 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4388 num = EXT4_INODES_PER_GROUP(sb);
4389 if (ext4_has_group_desc_csum(sb))
4390 num -= ext4_itable_unused_count(sb, gdp);
4391 table += num / inodes_per_block;
4395 ext4_sb_breadahead_unmovable(sb, b++);
4399 * There are other valid inodes in the buffer, this inode
4400 * has in-inode xattrs, or we don't have this inode in memory.
4401 * Read the block from disk.
4403 trace_ext4_load_inode(inode);
4404 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4405 blk_finish_plug(&plug);
4407 if (!buffer_uptodate(bh)) {
4409 ext4_error_inode_block(inode, block, EIO,
4410 "unable to read itable block");
4420 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4422 /* We have all inode data except xattrs in memory here. */
4423 return __ext4_get_inode_loc(inode, iloc,
4424 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4427 static bool ext4_should_enable_dax(struct inode *inode)
4429 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4431 if (test_opt2(inode->i_sb, DAX_NEVER))
4433 if (!S_ISREG(inode->i_mode))
4435 if (ext4_should_journal_data(inode))
4437 if (ext4_has_inline_data(inode))
4439 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4441 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4443 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4445 if (test_opt(inode->i_sb, DAX_ALWAYS))
4448 return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4451 void ext4_set_inode_flags(struct inode *inode, bool init)
4453 unsigned int flags = EXT4_I(inode)->i_flags;
4454 unsigned int new_fl = 0;
4456 WARN_ON_ONCE(IS_DAX(inode) && init);
4458 if (flags & EXT4_SYNC_FL)
4460 if (flags & EXT4_APPEND_FL)
4462 if (flags & EXT4_IMMUTABLE_FL)
4463 new_fl |= S_IMMUTABLE;
4464 if (flags & EXT4_NOATIME_FL)
4465 new_fl |= S_NOATIME;
4466 if (flags & EXT4_DIRSYNC_FL)
4467 new_fl |= S_DIRSYNC;
4469 /* Because of the way inode_set_flags() works we must preserve S_DAX
4470 * here if already set. */
4471 new_fl |= (inode->i_flags & S_DAX);
4472 if (init && ext4_should_enable_dax(inode))
4475 if (flags & EXT4_ENCRYPT_FL)
4476 new_fl |= S_ENCRYPTED;
4477 if (flags & EXT4_CASEFOLD_FL)
4478 new_fl |= S_CASEFOLD;
4479 if (flags & EXT4_VERITY_FL)
4481 inode_set_flags(inode, new_fl,
4482 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4483 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4486 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4487 struct ext4_inode_info *ei)
4490 struct inode *inode = &(ei->vfs_inode);
4491 struct super_block *sb = inode->i_sb;
4493 if (ext4_has_feature_huge_file(sb)) {
4494 /* we are using combined 48 bit field */
4495 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4496 le32_to_cpu(raw_inode->i_blocks_lo);
4497 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4498 /* i_blocks represent file system block size */
4499 return i_blocks << (inode->i_blkbits - 9);
4504 return le32_to_cpu(raw_inode->i_blocks_lo);
4508 static inline int ext4_iget_extra_inode(struct inode *inode,
4509 struct ext4_inode *raw_inode,
4510 struct ext4_inode_info *ei)
4512 __le32 *magic = (void *)raw_inode +
4513 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4515 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4516 EXT4_INODE_SIZE(inode->i_sb) &&
4517 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4518 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4519 return ext4_find_inline_data_nolock(inode);
4521 EXT4_I(inode)->i_inline_off = 0;
4525 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4527 if (!ext4_has_feature_project(inode->i_sb))
4529 *projid = EXT4_I(inode)->i_projid;
4534 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4535 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4538 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4540 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4541 inode_set_iversion_raw(inode, val);
4543 inode_set_iversion_queried(inode, val);
4545 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4547 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4548 return inode_peek_iversion_raw(inode);
4550 return inode_peek_iversion(inode);
4553 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4554 ext4_iget_flags flags, const char *function,
4557 struct ext4_iloc iloc;
4558 struct ext4_inode *raw_inode;
4559 struct ext4_inode_info *ei;
4560 struct inode *inode;
4561 journal_t *journal = EXT4_SB(sb)->s_journal;
4569 if ((!(flags & EXT4_IGET_SPECIAL) &&
4570 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4571 (ino < EXT4_ROOT_INO) ||
4572 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4573 if (flags & EXT4_IGET_HANDLE)
4574 return ERR_PTR(-ESTALE);
4575 __ext4_error(sb, function, line, EFSCORRUPTED, 0,
4576 "inode #%lu: comm %s: iget: illegal inode #",
4577 ino, current->comm);
4578 return ERR_PTR(-EFSCORRUPTED);
4581 inode = iget_locked(sb, ino);
4583 return ERR_PTR(-ENOMEM);
4584 if (!(inode->i_state & I_NEW))
4590 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4593 raw_inode = ext4_raw_inode(&iloc);
4595 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4596 ext4_error_inode(inode, function, line, 0,
4597 "iget: root inode unallocated");
4598 ret = -EFSCORRUPTED;
4602 if ((flags & EXT4_IGET_HANDLE) &&
4603 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4608 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4609 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4610 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4611 EXT4_INODE_SIZE(inode->i_sb) ||
4612 (ei->i_extra_isize & 3)) {
4613 ext4_error_inode(inode, function, line, 0,
4614 "iget: bad extra_isize %u "
4617 EXT4_INODE_SIZE(inode->i_sb));
4618 ret = -EFSCORRUPTED;
4622 ei->i_extra_isize = 0;
4624 /* Precompute checksum seed for inode metadata */
4625 if (ext4_has_metadata_csum(sb)) {
4626 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4628 __le32 inum = cpu_to_le32(inode->i_ino);
4629 __le32 gen = raw_inode->i_generation;
4630 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4632 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4636 if (!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4637 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) {
4638 ext4_error_inode_err(inode, function, line, 0, EFSBADCRC,
4639 "iget: checksum invalid");
4644 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4645 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4646 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4647 if (ext4_has_feature_project(sb) &&
4648 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4649 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4650 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4652 i_projid = EXT4_DEF_PROJID;
4654 if (!(test_opt(inode->i_sb, NO_UID32))) {
4655 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4656 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4658 i_uid_write(inode, i_uid);
4659 i_gid_write(inode, i_gid);
4660 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4661 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4663 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4664 ei->i_inline_off = 0;
4665 ei->i_dir_start_lookup = 0;
4666 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4667 /* We now have enough fields to check if the inode was active or not.
4668 * This is needed because nfsd might try to access dead inodes
4669 * the test is that same one that e2fsck uses
4670 * NeilBrown 1999oct15
4672 if (inode->i_nlink == 0) {
4673 if ((inode->i_mode == 0 ||
4674 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4675 ino != EXT4_BOOT_LOADER_INO) {
4676 /* this inode is deleted */
4680 /* The only unlinked inodes we let through here have
4681 * valid i_mode and are being read by the orphan
4682 * recovery code: that's fine, we're about to complete
4683 * the process of deleting those.
4684 * OR it is the EXT4_BOOT_LOADER_INO which is
4685 * not initialized on a new filesystem. */
4687 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4688 ext4_set_inode_flags(inode, true);
4689 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4690 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4691 if (ext4_has_feature_64bit(sb))
4693 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4694 inode->i_size = ext4_isize(sb, raw_inode);
4695 if ((size = i_size_read(inode)) < 0) {
4696 ext4_error_inode(inode, function, line, 0,
4697 "iget: bad i_size value: %lld", size);
4698 ret = -EFSCORRUPTED;
4702 * If dir_index is not enabled but there's dir with INDEX flag set,
4703 * we'd normally treat htree data as empty space. But with metadata
4704 * checksumming that corrupts checksums so forbid that.
4706 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4707 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4708 ext4_error_inode(inode, function, line, 0,
4709 "iget: Dir with htree data on filesystem without dir_index feature.");
4710 ret = -EFSCORRUPTED;
4713 ei->i_disksize = inode->i_size;
4715 ei->i_reserved_quota = 0;
4717 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4718 ei->i_block_group = iloc.block_group;
4719 ei->i_last_alloc_group = ~0;
4721 * NOTE! The in-memory inode i_data array is in little-endian order
4722 * even on big-endian machines: we do NOT byteswap the block numbers!
4724 for (block = 0; block < EXT4_N_BLOCKS; block++)
4725 ei->i_data[block] = raw_inode->i_block[block];
4726 INIT_LIST_HEAD(&ei->i_orphan);
4727 ext4_fc_init_inode(&ei->vfs_inode);
4730 * Set transaction id's of transactions that have to be committed
4731 * to finish f[data]sync. We set them to currently running transaction
4732 * as we cannot be sure that the inode or some of its metadata isn't
4733 * part of the transaction - the inode could have been reclaimed and
4734 * now it is reread from disk.
4737 transaction_t *transaction;
4740 read_lock(&journal->j_state_lock);
4741 if (journal->j_running_transaction)
4742 transaction = journal->j_running_transaction;
4744 transaction = journal->j_committing_transaction;
4746 tid = transaction->t_tid;
4748 tid = journal->j_commit_sequence;
4749 read_unlock(&journal->j_state_lock);
4750 ei->i_sync_tid = tid;
4751 ei->i_datasync_tid = tid;
4754 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4755 if (ei->i_extra_isize == 0) {
4756 /* The extra space is currently unused. Use it. */
4757 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4758 ei->i_extra_isize = sizeof(struct ext4_inode) -
4759 EXT4_GOOD_OLD_INODE_SIZE;
4761 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4767 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4768 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4769 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4770 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4772 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4773 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4775 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4776 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4778 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4780 ext4_inode_set_iversion_queried(inode, ivers);
4784 if (ei->i_file_acl &&
4785 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4786 ext4_error_inode(inode, function, line, 0,
4787 "iget: bad extended attribute block %llu",
4789 ret = -EFSCORRUPTED;
4791 } else if (!ext4_has_inline_data(inode)) {
4792 /* validate the block references in the inode */
4793 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4794 (S_ISLNK(inode->i_mode) &&
4795 !ext4_inode_is_fast_symlink(inode))) {
4796 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4797 ret = ext4_ext_check_inode(inode);
4799 ret = ext4_ind_check_inode(inode);
4805 if (S_ISREG(inode->i_mode)) {
4806 inode->i_op = &ext4_file_inode_operations;
4807 inode->i_fop = &ext4_file_operations;
4808 ext4_set_aops(inode);
4809 } else if (S_ISDIR(inode->i_mode)) {
4810 inode->i_op = &ext4_dir_inode_operations;
4811 inode->i_fop = &ext4_dir_operations;
4812 } else if (S_ISLNK(inode->i_mode)) {
4813 /* VFS does not allow setting these so must be corruption */
4814 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4815 ext4_error_inode(inode, function, line, 0,
4816 "iget: immutable or append flags "
4817 "not allowed on symlinks");
4818 ret = -EFSCORRUPTED;
4821 if (IS_ENCRYPTED(inode)) {
4822 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4823 ext4_set_aops(inode);
4824 } else if (ext4_inode_is_fast_symlink(inode)) {
4825 inode->i_link = (char *)ei->i_data;
4826 inode->i_op = &ext4_fast_symlink_inode_operations;
4827 nd_terminate_link(ei->i_data, inode->i_size,
4828 sizeof(ei->i_data) - 1);
4830 inode->i_op = &ext4_symlink_inode_operations;
4831 ext4_set_aops(inode);
4833 inode_nohighmem(inode);
4834 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4835 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4836 inode->i_op = &ext4_special_inode_operations;
4837 if (raw_inode->i_block[0])
4838 init_special_inode(inode, inode->i_mode,
4839 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4841 init_special_inode(inode, inode->i_mode,
4842 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4843 } else if (ino == EXT4_BOOT_LOADER_INO) {
4844 make_bad_inode(inode);
4846 ret = -EFSCORRUPTED;
4847 ext4_error_inode(inode, function, line, 0,
4848 "iget: bogus i_mode (%o)", inode->i_mode);
4851 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
4852 ext4_error_inode(inode, function, line, 0,
4853 "casefold flag without casefold feature");
4856 unlock_new_inode(inode);
4862 return ERR_PTR(ret);
4865 static int ext4_inode_blocks_set(handle_t *handle,
4866 struct ext4_inode *raw_inode,
4867 struct ext4_inode_info *ei)
4869 struct inode *inode = &(ei->vfs_inode);
4870 u64 i_blocks = READ_ONCE(inode->i_blocks);
4871 struct super_block *sb = inode->i_sb;
4873 if (i_blocks <= ~0U) {
4875 * i_blocks can be represented in a 32 bit variable
4876 * as multiple of 512 bytes
4878 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4879 raw_inode->i_blocks_high = 0;
4880 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4883 if (!ext4_has_feature_huge_file(sb))
4886 if (i_blocks <= 0xffffffffffffULL) {
4888 * i_blocks can be represented in a 48 bit variable
4889 * as multiple of 512 bytes
4891 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4892 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4893 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4895 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4896 /* i_block is stored in file system block size */
4897 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4898 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4899 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4904 static void __ext4_update_other_inode_time(struct super_block *sb,
4905 unsigned long orig_ino,
4907 struct ext4_inode *raw_inode)
4909 struct inode *inode;
4911 inode = find_inode_by_ino_rcu(sb, ino);
4915 if ((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4917 ((inode->i_state & I_DIRTY_TIME) == 0))
4920 spin_lock(&inode->i_lock);
4921 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4922 I_DIRTY_INODE)) == 0) &&
4923 (inode->i_state & I_DIRTY_TIME)) {
4924 struct ext4_inode_info *ei = EXT4_I(inode);
4926 inode->i_state &= ~I_DIRTY_TIME;
4927 spin_unlock(&inode->i_lock);
4929 spin_lock(&ei->i_raw_lock);
4930 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4931 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4932 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4933 ext4_inode_csum_set(inode, raw_inode, ei);
4934 spin_unlock(&ei->i_raw_lock);
4935 trace_ext4_other_inode_update_time(inode, orig_ino);
4938 spin_unlock(&inode->i_lock);
4942 * Opportunistically update the other time fields for other inodes in
4943 * the same inode table block.
4945 static void ext4_update_other_inodes_time(struct super_block *sb,
4946 unsigned long orig_ino, char *buf)
4949 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4950 int inode_size = EXT4_INODE_SIZE(sb);
4953 * Calculate the first inode in the inode table block. Inode
4954 * numbers are one-based. That is, the first inode in a block
4955 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4957 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4959 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4960 if (ino == orig_ino)
4962 __ext4_update_other_inode_time(sb, orig_ino, ino,
4963 (struct ext4_inode *)buf);
4969 * Post the struct inode info into an on-disk inode location in the
4970 * buffer-cache. This gobbles the caller's reference to the
4971 * buffer_head in the inode location struct.
4973 * The caller must have write access to iloc->bh.
4975 static int ext4_do_update_inode(handle_t *handle,
4976 struct inode *inode,
4977 struct ext4_iloc *iloc)
4979 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4980 struct ext4_inode_info *ei = EXT4_I(inode);
4981 struct buffer_head *bh = iloc->bh;
4982 struct super_block *sb = inode->i_sb;
4983 int err = 0, rc, block;
4984 int need_datasync = 0, set_large_file = 0;
4989 spin_lock(&ei->i_raw_lock);
4991 /* For fields not tracked in the in-memory inode,
4992 * initialise them to zero for new inodes. */
4993 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4994 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4996 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4997 i_uid = i_uid_read(inode);
4998 i_gid = i_gid_read(inode);
4999 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5000 if (!(test_opt(inode->i_sb, NO_UID32))) {
5001 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5002 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5004 * Fix up interoperability with old kernels. Otherwise, old inodes get
5005 * re-used with the upper 16 bits of the uid/gid intact
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 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5031 spin_unlock(&ei->i_raw_lock);
5034 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5035 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5036 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5037 raw_inode->i_file_acl_high =
5038 cpu_to_le16(ei->i_file_acl >> 32);
5039 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5040 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) {
5041 ext4_isize_set(raw_inode, ei->i_disksize);
5044 if (ei->i_disksize > 0x7fffffffULL) {
5045 if (!ext4_has_feature_large_file(sb) ||
5046 EXT4_SB(sb)->s_es->s_rev_level ==
5047 cpu_to_le32(EXT4_GOOD_OLD_REV))
5050 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5051 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5052 if (old_valid_dev(inode->i_rdev)) {
5053 raw_inode->i_block[0] =
5054 cpu_to_le32(old_encode_dev(inode->i_rdev));
5055 raw_inode->i_block[1] = 0;
5057 raw_inode->i_block[0] = 0;
5058 raw_inode->i_block[1] =
5059 cpu_to_le32(new_encode_dev(inode->i_rdev));
5060 raw_inode->i_block[2] = 0;
5062 } else if (!ext4_has_inline_data(inode)) {
5063 for (block = 0; block < EXT4_N_BLOCKS; block++)
5064 raw_inode->i_block[block] = ei->i_data[block];
5067 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5068 u64 ivers = ext4_inode_peek_iversion(inode);
5070 raw_inode->i_disk_version = cpu_to_le32(ivers);
5071 if (ei->i_extra_isize) {
5072 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5073 raw_inode->i_version_hi =
5074 cpu_to_le32(ivers >> 32);
5075 raw_inode->i_extra_isize =
5076 cpu_to_le16(ei->i_extra_isize);
5080 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5081 i_projid != EXT4_DEF_PROJID);
5083 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5084 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5085 raw_inode->i_projid = cpu_to_le32(i_projid);
5087 ext4_inode_csum_set(inode, raw_inode, ei);
5088 spin_unlock(&ei->i_raw_lock);
5089 if (inode->i_sb->s_flags & SB_LAZYTIME)
5090 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5093 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5094 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5097 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5098 if (set_large_file) {
5099 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5100 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5103 ext4_set_feature_large_file(sb);
5104 ext4_handle_sync(handle);
5105 err = ext4_handle_dirty_super(handle, sb);
5107 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5110 ext4_std_error(inode->i_sb, err);
5115 * ext4_write_inode()
5117 * We are called from a few places:
5119 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5120 * Here, there will be no transaction running. We wait for any running
5121 * transaction to commit.
5123 * - Within flush work (sys_sync(), kupdate and such).
5124 * We wait on commit, if told to.
5126 * - Within iput_final() -> write_inode_now()
5127 * We wait on commit, if told to.
5129 * In all cases it is actually safe for us to return without doing anything,
5130 * because the inode has been copied into a raw inode buffer in
5131 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5134 * Note that we are absolutely dependent upon all inode dirtiers doing the
5135 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5136 * which we are interested.
5138 * It would be a bug for them to not do this. The code:
5140 * mark_inode_dirty(inode)
5142 * inode->i_size = expr;
5144 * is in error because write_inode() could occur while `stuff()' is running,
5145 * and the new i_size will be lost. Plus the inode will no longer be on the
5146 * superblock's dirty inode list.
5148 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5152 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5153 sb_rdonly(inode->i_sb))
5156 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5159 if (EXT4_SB(inode->i_sb)->s_journal) {
5160 if (ext4_journal_current_handle()) {
5161 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5167 * No need to force transaction in WB_SYNC_NONE mode. Also
5168 * ext4_sync_fs() will force the commit after everything is
5171 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5174 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5175 EXT4_I(inode)->i_sync_tid);
5177 struct ext4_iloc iloc;
5179 err = __ext4_get_inode_loc(inode, &iloc, 0);
5183 * sync(2) will flush the whole buffer cache. No need to do
5184 * it here separately for each inode.
5186 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5187 sync_dirty_buffer(iloc.bh);
5188 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5189 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5190 "IO error syncing inode");
5199 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5200 * buffers that are attached to a page stradding i_size and are undergoing
5201 * commit. In that case we have to wait for commit to finish and try again.
5203 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5207 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5208 tid_t commit_tid = 0;
5211 offset = inode->i_size & (PAGE_SIZE - 1);
5213 * If the page is fully truncated, we don't need to wait for any commit
5214 * (and we even should not as __ext4_journalled_invalidatepage() may
5215 * strip all buffers from the page but keep the page dirty which can then
5216 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5217 * buffers). Also we don't need to wait for any commit if all buffers in
5218 * the page remain valid. This is most beneficial for the common case of
5219 * blocksize == PAGESIZE.
5221 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5224 page = find_lock_page(inode->i_mapping,
5225 inode->i_size >> PAGE_SHIFT);
5228 ret = __ext4_journalled_invalidatepage(page, offset,
5229 PAGE_SIZE - offset);
5235 read_lock(&journal->j_state_lock);
5236 if (journal->j_committing_transaction)
5237 commit_tid = journal->j_committing_transaction->t_tid;
5238 read_unlock(&journal->j_state_lock);
5240 jbd2_log_wait_commit(journal, commit_tid);
5247 * Called from notify_change.
5249 * We want to trap VFS attempts to truncate the file as soon as
5250 * possible. In particular, we want to make sure that when the VFS
5251 * shrinks i_size, we put the inode on the orphan list and modify
5252 * i_disksize immediately, so that during the subsequent flushing of
5253 * dirty pages and freeing of disk blocks, we can guarantee that any
5254 * commit will leave the blocks being flushed in an unused state on
5255 * disk. (On recovery, the inode will get truncated and the blocks will
5256 * be freed, so we have a strong guarantee that no future commit will
5257 * leave these blocks visible to the user.)
5259 * Another thing we have to assure is that if we are in ordered mode
5260 * and inode is still attached to the committing transaction, we must
5261 * we start writeout of all the dirty pages which are being truncated.
5262 * This way we are sure that all the data written in the previous
5263 * transaction are already on disk (truncate waits for pages under
5266 * Called with inode->i_mutex down.
5268 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5270 struct inode *inode = d_inode(dentry);
5273 const unsigned int ia_valid = attr->ia_valid;
5275 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5278 if (unlikely(IS_IMMUTABLE(inode)))
5281 if (unlikely(IS_APPEND(inode) &&
5282 (ia_valid & (ATTR_MODE | ATTR_UID |
5283 ATTR_GID | ATTR_TIMES_SET))))
5286 error = setattr_prepare(dentry, attr);
5290 error = fscrypt_prepare_setattr(dentry, attr);
5294 error = fsverity_prepare_setattr(dentry, attr);
5298 if (is_quota_modification(inode, attr)) {
5299 error = dquot_initialize(inode);
5303 ext4_fc_start_update(inode);
5304 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5305 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5308 /* (user+group)*(old+new) structure, inode write (sb,
5309 * inode block, ? - but truncate inode update has it) */
5310 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5311 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5312 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5313 if (IS_ERR(handle)) {
5314 error = PTR_ERR(handle);
5318 /* dquot_transfer() calls back ext4_get_inode_usage() which
5319 * counts xattr inode references.
5321 down_read(&EXT4_I(inode)->xattr_sem);
5322 error = dquot_transfer(inode, attr);
5323 up_read(&EXT4_I(inode)->xattr_sem);
5326 ext4_journal_stop(handle);
5327 ext4_fc_stop_update(inode);
5330 /* Update corresponding info in inode so that everything is in
5331 * one transaction */
5332 if (attr->ia_valid & ATTR_UID)
5333 inode->i_uid = attr->ia_uid;
5334 if (attr->ia_valid & ATTR_GID)
5335 inode->i_gid = attr->ia_gid;
5336 error = ext4_mark_inode_dirty(handle, inode);
5337 ext4_journal_stop(handle);
5338 if (unlikely(error))
5342 if (attr->ia_valid & ATTR_SIZE) {
5344 loff_t oldsize = inode->i_size;
5345 int shrink = (attr->ia_size < inode->i_size);
5347 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5348 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5350 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5351 ext4_fc_stop_update(inode);
5355 if (!S_ISREG(inode->i_mode)) {
5356 ext4_fc_stop_update(inode);
5360 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5361 inode_inc_iversion(inode);
5364 if (ext4_should_order_data(inode)) {
5365 error = ext4_begin_ordered_truncate(inode,
5371 * Blocks are going to be removed from the inode. Wait
5372 * for dio in flight.
5374 inode_dio_wait(inode);
5377 down_write(&EXT4_I(inode)->i_mmap_sem);
5379 rc = ext4_break_layouts(inode);
5381 up_write(&EXT4_I(inode)->i_mmap_sem);
5385 if (attr->ia_size != inode->i_size) {
5386 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5387 if (IS_ERR(handle)) {
5388 error = PTR_ERR(handle);
5391 if (ext4_handle_valid(handle) && shrink) {
5392 error = ext4_orphan_add(handle, inode);
5396 * Update c/mtime on truncate up, ext4_truncate() will
5397 * update c/mtime in shrink case below
5400 inode->i_mtime = current_time(inode);
5401 inode->i_ctime = inode->i_mtime;
5405 ext4_fc_track_range(inode,
5406 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5407 inode->i_sb->s_blocksize_bits,
5408 (oldsize > 0 ? oldsize - 1 : 0) >>
5409 inode->i_sb->s_blocksize_bits);
5411 ext4_fc_track_range(
5413 (oldsize > 0 ? oldsize - 1 : oldsize) >>
5414 inode->i_sb->s_blocksize_bits,
5415 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5416 inode->i_sb->s_blocksize_bits);
5418 down_write(&EXT4_I(inode)->i_data_sem);
5419 EXT4_I(inode)->i_disksize = attr->ia_size;
5420 rc = ext4_mark_inode_dirty(handle, inode);
5424 * We have to update i_size under i_data_sem together
5425 * with i_disksize to avoid races with writeback code
5426 * running ext4_wb_update_i_disksize().
5429 i_size_write(inode, attr->ia_size);
5430 up_write(&EXT4_I(inode)->i_data_sem);
5431 ext4_journal_stop(handle);
5435 pagecache_isize_extended(inode, oldsize,
5437 } else if (ext4_should_journal_data(inode)) {
5438 ext4_wait_for_tail_page_commit(inode);
5443 * Truncate pagecache after we've waited for commit
5444 * in data=journal mode to make pages freeable.
5446 truncate_pagecache(inode, inode->i_size);
5448 * Call ext4_truncate() even if i_size didn't change to
5449 * truncate possible preallocated blocks.
5451 if (attr->ia_size <= oldsize) {
5452 rc = ext4_truncate(inode);
5457 up_write(&EXT4_I(inode)->i_mmap_sem);
5461 setattr_copy(inode, attr);
5462 mark_inode_dirty(inode);
5466 * If the call to ext4_truncate failed to get a transaction handle at
5467 * all, we need to clean up the in-core orphan list manually.
5469 if (orphan && inode->i_nlink)
5470 ext4_orphan_del(NULL, inode);
5472 if (!error && (ia_valid & ATTR_MODE))
5473 rc = posix_acl_chmod(inode, inode->i_mode);
5477 ext4_std_error(inode->i_sb, error);
5480 ext4_fc_stop_update(inode);
5484 int ext4_getattr(const struct path *path, struct kstat *stat,
5485 u32 request_mask, unsigned int query_flags)
5487 struct inode *inode = d_inode(path->dentry);
5488 struct ext4_inode *raw_inode;
5489 struct ext4_inode_info *ei = EXT4_I(inode);
5492 if ((request_mask & STATX_BTIME) &&
5493 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5494 stat->result_mask |= STATX_BTIME;
5495 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5496 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5499 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5500 if (flags & EXT4_APPEND_FL)
5501 stat->attributes |= STATX_ATTR_APPEND;
5502 if (flags & EXT4_COMPR_FL)
5503 stat->attributes |= STATX_ATTR_COMPRESSED;
5504 if (flags & EXT4_ENCRYPT_FL)
5505 stat->attributes |= STATX_ATTR_ENCRYPTED;
5506 if (flags & EXT4_IMMUTABLE_FL)
5507 stat->attributes |= STATX_ATTR_IMMUTABLE;
5508 if (flags & EXT4_NODUMP_FL)
5509 stat->attributes |= STATX_ATTR_NODUMP;
5510 if (flags & EXT4_VERITY_FL)
5511 stat->attributes |= STATX_ATTR_VERITY;
5513 stat->attributes_mask |= (STATX_ATTR_APPEND |
5514 STATX_ATTR_COMPRESSED |
5515 STATX_ATTR_ENCRYPTED |
5516 STATX_ATTR_IMMUTABLE |
5520 generic_fillattr(inode, stat);
5524 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5525 u32 request_mask, unsigned int query_flags)
5527 struct inode *inode = d_inode(path->dentry);
5528 u64 delalloc_blocks;
5530 ext4_getattr(path, stat, request_mask, query_flags);
5533 * If there is inline data in the inode, the inode will normally not
5534 * have data blocks allocated (it may have an external xattr block).
5535 * Report at least one sector for such files, so tools like tar, rsync,
5536 * others don't incorrectly think the file is completely sparse.
5538 if (unlikely(ext4_has_inline_data(inode)))
5539 stat->blocks += (stat->size + 511) >> 9;
5542 * We can't update i_blocks if the block allocation is delayed
5543 * otherwise in the case of system crash before the real block
5544 * allocation is done, we will have i_blocks inconsistent with
5545 * on-disk file blocks.
5546 * We always keep i_blocks updated together with real
5547 * allocation. But to not confuse with user, stat
5548 * will return the blocks that include the delayed allocation
5549 * blocks for this file.
5551 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5552 EXT4_I(inode)->i_reserved_data_blocks);
5553 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5557 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5560 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5561 return ext4_ind_trans_blocks(inode, lblocks);
5562 return ext4_ext_index_trans_blocks(inode, pextents);
5566 * Account for index blocks, block groups bitmaps and block group
5567 * descriptor blocks if modify datablocks and index blocks
5568 * worse case, the indexs blocks spread over different block groups
5570 * If datablocks are discontiguous, they are possible to spread over
5571 * different block groups too. If they are contiguous, with flexbg,
5572 * they could still across block group boundary.
5574 * Also account for superblock, inode, quota and xattr blocks
5576 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5579 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5585 * How many index blocks need to touch to map @lblocks logical blocks
5586 * to @pextents physical extents?
5588 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5593 * Now let's see how many group bitmaps and group descriptors need
5596 groups = idxblocks + pextents;
5598 if (groups > ngroups)
5600 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5601 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5603 /* bitmaps and block group descriptor blocks */
5604 ret += groups + gdpblocks;
5606 /* Blocks for super block, inode, quota and xattr blocks */
5607 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5613 * Calculate the total number of credits to reserve to fit
5614 * the modification of a single pages into a single transaction,
5615 * which may include multiple chunks of block allocations.
5617 * This could be called via ext4_write_begin()
5619 * We need to consider the worse case, when
5620 * one new block per extent.
5622 int ext4_writepage_trans_blocks(struct inode *inode)
5624 int bpp = ext4_journal_blocks_per_page(inode);
5627 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5629 /* Account for data blocks for journalled mode */
5630 if (ext4_should_journal_data(inode))
5636 * Calculate the journal credits for a chunk of data modification.
5638 * This is called from DIO, fallocate or whoever calling
5639 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5641 * journal buffers for data blocks are not included here, as DIO
5642 * and fallocate do no need to journal data buffers.
5644 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5646 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5650 * The caller must have previously called ext4_reserve_inode_write().
5651 * Give this, we know that the caller already has write access to iloc->bh.
5653 int ext4_mark_iloc_dirty(handle_t *handle,
5654 struct inode *inode, struct ext4_iloc *iloc)
5658 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5662 ext4_fc_track_inode(inode);
5664 if (IS_I_VERSION(inode))
5665 inode_inc_iversion(inode);
5667 /* the do_update_inode consumes one bh->b_count */
5670 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5671 err = ext4_do_update_inode(handle, inode, iloc);
5677 * On success, We end up with an outstanding reference count against
5678 * iloc->bh. This _must_ be cleaned up later.
5682 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5683 struct ext4_iloc *iloc)
5687 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5690 err = ext4_get_inode_loc(inode, iloc);
5692 BUFFER_TRACE(iloc->bh, "get_write_access");
5693 err = ext4_journal_get_write_access(handle, iloc->bh);
5699 ext4_std_error(inode->i_sb, err);
5703 static int __ext4_expand_extra_isize(struct inode *inode,
5704 unsigned int new_extra_isize,
5705 struct ext4_iloc *iloc,
5706 handle_t *handle, int *no_expand)
5708 struct ext4_inode *raw_inode;
5709 struct ext4_xattr_ibody_header *header;
5710 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5711 struct ext4_inode_info *ei = EXT4_I(inode);
5714 /* this was checked at iget time, but double check for good measure */
5715 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5716 (ei->i_extra_isize & 3)) {
5717 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5719 EXT4_INODE_SIZE(inode->i_sb));
5720 return -EFSCORRUPTED;
5722 if ((new_extra_isize < ei->i_extra_isize) ||
5723 (new_extra_isize < 4) ||
5724 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5725 return -EINVAL; /* Should never happen */
5727 raw_inode = ext4_raw_inode(iloc);
5729 header = IHDR(inode, raw_inode);
5731 /* No extended attributes present */
5732 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5733 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5734 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5735 EXT4_I(inode)->i_extra_isize, 0,
5736 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5737 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5741 /* try to expand with EAs present */
5742 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5746 * Inode size expansion failed; don't try again
5755 * Expand an inode by new_extra_isize bytes.
5756 * Returns 0 on success or negative error number on failure.
5758 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5759 unsigned int new_extra_isize,
5760 struct ext4_iloc iloc,
5766 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5770 * In nojournal mode, we can immediately attempt to expand
5771 * the inode. When journaled, we first need to obtain extra
5772 * buffer credits since we may write into the EA block
5773 * with this same handle. If journal_extend fails, then it will
5774 * only result in a minor loss of functionality for that inode.
5775 * If this is felt to be critical, then e2fsck should be run to
5776 * force a large enough s_min_extra_isize.
5778 if (ext4_journal_extend(handle,
5779 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5782 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5785 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5786 handle, &no_expand);
5787 ext4_write_unlock_xattr(inode, &no_expand);
5792 int ext4_expand_extra_isize(struct inode *inode,
5793 unsigned int new_extra_isize,
5794 struct ext4_iloc *iloc)
5800 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5805 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5806 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5807 if (IS_ERR(handle)) {
5808 error = PTR_ERR(handle);
5813 ext4_write_lock_xattr(inode, &no_expand);
5815 BUFFER_TRACE(iloc->bh, "get_write_access");
5816 error = ext4_journal_get_write_access(handle, iloc->bh);
5822 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5823 handle, &no_expand);
5825 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5830 ext4_write_unlock_xattr(inode, &no_expand);
5831 ext4_journal_stop(handle);
5836 * What we do here is to mark the in-core inode as clean with respect to inode
5837 * dirtiness (it may still be data-dirty).
5838 * This means that the in-core inode may be reaped by prune_icache
5839 * without having to perform any I/O. This is a very good thing,
5840 * because *any* task may call prune_icache - even ones which
5841 * have a transaction open against a different journal.
5843 * Is this cheating? Not really. Sure, we haven't written the
5844 * inode out, but prune_icache isn't a user-visible syncing function.
5845 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5846 * we start and wait on commits.
5848 int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5849 const char *func, unsigned int line)
5851 struct ext4_iloc iloc;
5852 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5856 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5857 err = ext4_reserve_inode_write(handle, inode, &iloc);
5861 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5862 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5865 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5868 ext4_error_inode_err(inode, func, line, 0, err,
5869 "mark_inode_dirty error");
5874 * ext4_dirty_inode() is called from __mark_inode_dirty()
5876 * We're really interested in the case where a file is being extended.
5877 * i_size has been changed by generic_commit_write() and we thus need
5878 * to include the updated inode in the current transaction.
5880 * Also, dquot_alloc_block() will always dirty the inode when blocks
5881 * are allocated to the file.
5883 * If the inode is marked synchronous, we don't honour that here - doing
5884 * so would cause a commit on atime updates, which we don't bother doing.
5885 * We handle synchronous inodes at the highest possible level.
5887 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5888 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5889 * to copy into the on-disk inode structure are the timestamp files.
5891 void ext4_dirty_inode(struct inode *inode, int flags)
5895 if (flags == I_DIRTY_TIME)
5897 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5901 ext4_mark_inode_dirty(handle, inode);
5903 ext4_journal_stop(handle);
5908 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5913 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5916 * We have to be very careful here: changing a data block's
5917 * journaling status dynamically is dangerous. If we write a
5918 * data block to the journal, change the status and then delete
5919 * that block, we risk forgetting to revoke the old log record
5920 * from the journal and so a subsequent replay can corrupt data.
5921 * So, first we make sure that the journal is empty and that
5922 * nobody is changing anything.
5925 journal = EXT4_JOURNAL(inode);
5928 if (is_journal_aborted(journal))
5931 /* Wait for all existing dio workers */
5932 inode_dio_wait(inode);
5935 * Before flushing the journal and switching inode's aops, we have
5936 * to flush all dirty data the inode has. There can be outstanding
5937 * delayed allocations, there can be unwritten extents created by
5938 * fallocate or buffered writes in dioread_nolock mode covered by
5939 * dirty data which can be converted only after flushing the dirty
5940 * data (and journalled aops don't know how to handle these cases).
5943 down_write(&EXT4_I(inode)->i_mmap_sem);
5944 err = filemap_write_and_wait(inode->i_mapping);
5946 up_write(&EXT4_I(inode)->i_mmap_sem);
5951 percpu_down_write(&sbi->s_writepages_rwsem);
5952 jbd2_journal_lock_updates(journal);
5955 * OK, there are no updates running now, and all cached data is
5956 * synced to disk. We are now in a completely consistent state
5957 * which doesn't have anything in the journal, and we know that
5958 * no filesystem updates are running, so it is safe to modify
5959 * the inode's in-core data-journaling state flag now.
5963 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5965 err = jbd2_journal_flush(journal);
5967 jbd2_journal_unlock_updates(journal);
5968 percpu_up_write(&sbi->s_writepages_rwsem);
5971 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5973 ext4_set_aops(inode);
5975 jbd2_journal_unlock_updates(journal);
5976 percpu_up_write(&sbi->s_writepages_rwsem);
5979 up_write(&EXT4_I(inode)->i_mmap_sem);
5981 /* Finally we can mark the inode as dirty. */
5983 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5985 return PTR_ERR(handle);
5987 ext4_fc_mark_ineligible(inode->i_sb,
5988 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE);
5989 err = ext4_mark_inode_dirty(handle, inode);
5990 ext4_handle_sync(handle);
5991 ext4_journal_stop(handle);
5992 ext4_std_error(inode->i_sb, err);
5997 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5999 return !buffer_mapped(bh);
6002 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6004 struct vm_area_struct *vma = vmf->vma;
6005 struct page *page = vmf->page;
6010 struct file *file = vma->vm_file;
6011 struct inode *inode = file_inode(file);
6012 struct address_space *mapping = inode->i_mapping;
6014 get_block_t *get_block;
6017 if (unlikely(IS_IMMUTABLE(inode)))
6018 return VM_FAULT_SIGBUS;
6020 sb_start_pagefault(inode->i_sb);
6021 file_update_time(vma->vm_file);
6023 down_read(&EXT4_I(inode)->i_mmap_sem);
6025 err = ext4_convert_inline_data(inode);
6030 * On data journalling we skip straight to the transaction handle:
6031 * there's no delalloc; page truncated will be checked later; the
6032 * early return w/ all buffers mapped (calculates size/len) can't
6033 * be used; and there's no dioread_nolock, so only ext4_get_block.
6035 if (ext4_should_journal_data(inode))
6038 /* Delalloc case is easy... */
6039 if (test_opt(inode->i_sb, DELALLOC) &&
6040 !ext4_nonda_switch(inode->i_sb)) {
6042 err = block_page_mkwrite(vma, vmf,
6043 ext4_da_get_block_prep);
6044 } while (err == -ENOSPC &&
6045 ext4_should_retry_alloc(inode->i_sb, &retries));
6050 size = i_size_read(inode);
6051 /* Page got truncated from under us? */
6052 if (page->mapping != mapping || page_offset(page) > size) {
6054 ret = VM_FAULT_NOPAGE;
6058 if (page->index == size >> PAGE_SHIFT)
6059 len = size & ~PAGE_MASK;
6063 * Return if we have all the buffers mapped. This avoids the need to do
6064 * journal_start/journal_stop which can block and take a long time
6066 * This cannot be done for data journalling, as we have to add the
6067 * inode to the transaction's list to writeprotect pages on commit.
6069 if (page_has_buffers(page)) {
6070 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6072 ext4_bh_unmapped)) {
6073 /* Wait so that we don't change page under IO */
6074 wait_for_stable_page(page);
6075 ret = VM_FAULT_LOCKED;
6080 /* OK, we need to fill the hole... */
6081 if (ext4_should_dioread_nolock(inode))
6082 get_block = ext4_get_block_unwritten;
6084 get_block = ext4_get_block;
6086 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6087 ext4_writepage_trans_blocks(inode));
6088 if (IS_ERR(handle)) {
6089 ret = VM_FAULT_SIGBUS;
6093 * Data journalling can't use block_page_mkwrite() because it
6094 * will set_buffer_dirty() before do_journal_get_write_access()
6095 * thus might hit warning messages for dirty metadata buffers.
6097 if (!ext4_should_journal_data(inode)) {
6098 err = block_page_mkwrite(vma, vmf, get_block);
6101 size = i_size_read(inode);
6102 /* Page got truncated from under us? */
6103 if (page->mapping != mapping || page_offset(page) > size) {
6104 ret = VM_FAULT_NOPAGE;
6108 if (page->index == size >> PAGE_SHIFT)
6109 len = size & ~PAGE_MASK;
6113 err = __block_write_begin(page, 0, len, ext4_get_block);
6115 ret = VM_FAULT_SIGBUS;
6116 if (ext4_walk_page_buffers(handle, page_buffers(page),
6117 0, len, NULL, do_journal_get_write_access))
6119 if (ext4_walk_page_buffers(handle, page_buffers(page),
6120 0, len, NULL, write_end_fn))
6122 if (ext4_jbd2_inode_add_write(handle, inode, 0, len))
6124 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6129 ext4_journal_stop(handle);
6130 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6133 ret = block_page_mkwrite_return(err);
6135 up_read(&EXT4_I(inode)->i_mmap_sem);
6136 sb_end_pagefault(inode->i_sb);
6140 ext4_journal_stop(handle);
6144 vm_fault_t ext4_filemap_fault(struct vm_fault *vmf)
6146 struct inode *inode = file_inode(vmf->vma->vm_file);
6149 down_read(&EXT4_I(inode)->i_mmap_sem);
6150 ret = filemap_fault(vmf);
6151 up_read(&EXT4_I(inode)->i_mmap_sem);