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 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_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
518 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
519 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
520 map->m_pblk = ext4_es_pblock(&es) +
521 map->m_lblk - es.es_lblk;
522 map->m_flags |= ext4_es_is_written(&es) ?
523 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
524 retval = es.es_len - (map->m_lblk - es.es_lblk);
525 if (retval > map->m_len)
528 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
530 retval = es.es_len - (map->m_lblk - es.es_lblk);
531 if (retval > map->m_len)
538 #ifdef ES_AGGRESSIVE_TEST
539 ext4_map_blocks_es_recheck(handle, inode, map,
546 * Try to see if we can get the block without requesting a new
549 down_read(&EXT4_I(inode)->i_data_sem);
550 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
551 retval = ext4_ext_map_blocks(handle, inode, map, 0);
553 retval = ext4_ind_map_blocks(handle, inode, map, 0);
558 if (unlikely(retval != map->m_len)) {
559 ext4_warning(inode->i_sb,
560 "ES len assertion failed for inode "
561 "%lu: retval %d != map->m_len %d",
562 inode->i_ino, retval, map->m_len);
566 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
567 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
568 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
569 !(status & EXTENT_STATUS_WRITTEN) &&
570 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
571 map->m_lblk + map->m_len - 1))
572 status |= EXTENT_STATUS_DELAYED;
573 ret = ext4_es_insert_extent(inode, map->m_lblk,
574 map->m_len, map->m_pblk, status);
578 up_read((&EXT4_I(inode)->i_data_sem));
581 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
582 ret = check_block_validity(inode, map);
587 /* If it is only a block(s) look up */
588 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
592 * Returns if the blocks have already allocated
594 * Note that if blocks have been preallocated
595 * ext4_ext_get_block() returns the create = 0
596 * with buffer head unmapped.
598 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
600 * If we need to convert extent to unwritten
601 * we continue and do the actual work in
602 * ext4_ext_map_blocks()
604 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
608 * Here we clear m_flags because after allocating an new extent,
609 * it will be set again.
611 map->m_flags &= ~EXT4_MAP_FLAGS;
614 * New blocks allocate and/or writing to unwritten extent
615 * will possibly result in updating i_data, so we take
616 * the write lock of i_data_sem, and call get_block()
617 * with create == 1 flag.
619 down_write(&EXT4_I(inode)->i_data_sem);
622 * We need to check for EXT4 here because migrate
623 * could have changed the inode type in between
625 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
626 retval = ext4_ext_map_blocks(handle, inode, map, flags);
628 retval = ext4_ind_map_blocks(handle, inode, map, flags);
630 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
632 * We allocated new blocks which will result in
633 * i_data's format changing. Force the migrate
634 * to fail by clearing migrate flags
636 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
640 * Update reserved blocks/metadata blocks after successful
641 * block allocation which had been deferred till now. We don't
642 * support fallocate for non extent files. So we can update
643 * reserve space here.
646 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
647 ext4_da_update_reserve_space(inode, retval, 1);
653 if (unlikely(retval != map->m_len)) {
654 ext4_warning(inode->i_sb,
655 "ES len assertion failed for inode "
656 "%lu: retval %d != map->m_len %d",
657 inode->i_ino, retval, map->m_len);
662 * We have to zeroout blocks before inserting them into extent
663 * status tree. Otherwise someone could look them up there and
664 * use them before they are really zeroed. We also have to
665 * unmap metadata before zeroing as otherwise writeback can
666 * overwrite zeros with stale data from block device.
668 if (flags & EXT4_GET_BLOCKS_ZERO &&
669 map->m_flags & EXT4_MAP_MAPPED &&
670 map->m_flags & EXT4_MAP_NEW) {
671 ret = ext4_issue_zeroout(inode, map->m_lblk,
672 map->m_pblk, map->m_len);
680 * If the extent has been zeroed out, we don't need to update
681 * extent status tree.
683 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
684 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
685 if (ext4_es_is_written(&es))
688 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
689 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
690 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
691 !(status & EXTENT_STATUS_WRITTEN) &&
692 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
693 map->m_lblk + map->m_len - 1))
694 status |= EXTENT_STATUS_DELAYED;
695 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
696 map->m_pblk, status);
704 up_write((&EXT4_I(inode)->i_data_sem));
705 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
706 ret = check_block_validity(inode, map);
711 * Inodes with freshly allocated blocks where contents will be
712 * visible after transaction commit must be on transaction's
715 if (map->m_flags & EXT4_MAP_NEW &&
716 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
717 !(flags & EXT4_GET_BLOCKS_ZERO) &&
718 !ext4_is_quota_file(inode) &&
719 ext4_should_order_data(inode)) {
721 (loff_t)map->m_lblk << inode->i_blkbits;
722 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
724 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
725 ret = ext4_jbd2_inode_add_wait(handle, inode,
728 ret = ext4_jbd2_inode_add_write(handle, inode,
733 ext4_fc_track_range(inode, map->m_lblk,
734 map->m_lblk + map->m_len - 1);
738 ext_debug(inode, "failed with err %d\n", retval);
743 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
744 * we have to be careful as someone else may be manipulating b_state as well.
746 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
748 unsigned long old_state;
749 unsigned long new_state;
751 flags &= EXT4_MAP_FLAGS;
753 /* Dummy buffer_head? Set non-atomically. */
755 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
759 * Someone else may be modifying b_state. Be careful! This is ugly but
760 * once we get rid of using bh as a container for mapping information
761 * to pass to / from get_block functions, this can go away.
764 old_state = READ_ONCE(bh->b_state);
765 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
767 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
770 static int _ext4_get_block(struct inode *inode, sector_t iblock,
771 struct buffer_head *bh, int flags)
773 struct ext4_map_blocks map;
776 if (ext4_has_inline_data(inode))
780 map.m_len = bh->b_size >> inode->i_blkbits;
782 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
785 map_bh(bh, inode->i_sb, map.m_pblk);
786 ext4_update_bh_state(bh, map.m_flags);
787 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
789 } else if (ret == 0) {
790 /* hole case, need to fill in bh->b_size */
791 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
796 int ext4_get_block(struct inode *inode, sector_t iblock,
797 struct buffer_head *bh, int create)
799 return _ext4_get_block(inode, iblock, bh,
800 create ? EXT4_GET_BLOCKS_CREATE : 0);
804 * Get block function used when preparing for buffered write if we require
805 * creating an unwritten extent if blocks haven't been allocated. The extent
806 * will be converted to written after the IO is complete.
808 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
809 struct buffer_head *bh_result, int create)
811 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
812 inode->i_ino, create);
813 return _ext4_get_block(inode, iblock, bh_result,
814 EXT4_GET_BLOCKS_IO_CREATE_EXT);
817 /* Maximum number of blocks we map for direct IO at once. */
818 #define DIO_MAX_BLOCKS 4096
821 * `handle' can be NULL if create is zero
823 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
824 ext4_lblk_t block, int map_flags)
826 struct ext4_map_blocks map;
827 struct buffer_head *bh;
828 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
831 J_ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
832 || handle != NULL || create == 0);
836 err = ext4_map_blocks(handle, inode, &map, map_flags);
839 return create ? ERR_PTR(-ENOSPC) : NULL;
843 bh = sb_getblk(inode->i_sb, map.m_pblk);
845 return ERR_PTR(-ENOMEM);
846 if (map.m_flags & EXT4_MAP_NEW) {
847 J_ASSERT(create != 0);
848 J_ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
849 || (handle != NULL));
852 * Now that we do not always journal data, we should
853 * keep in mind whether this should always journal the
854 * new buffer as metadata. For now, regular file
855 * writes use ext4_get_block instead, so it's not a
859 BUFFER_TRACE(bh, "call get_create_access");
860 err = ext4_journal_get_create_access(handle, bh);
865 if (!buffer_uptodate(bh)) {
866 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
867 set_buffer_uptodate(bh);
870 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
871 err = ext4_handle_dirty_metadata(handle, inode, bh);
875 BUFFER_TRACE(bh, "not a new buffer");
882 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
883 ext4_lblk_t block, int map_flags)
885 struct buffer_head *bh;
888 bh = ext4_getblk(handle, inode, block, map_flags);
891 if (!bh || ext4_buffer_uptodate(bh))
894 ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
902 /* Read a contiguous batch of blocks. */
903 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
904 bool wait, struct buffer_head **bhs)
908 for (i = 0; i < bh_count; i++) {
909 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
910 if (IS_ERR(bhs[i])) {
911 err = PTR_ERR(bhs[i]);
917 for (i = 0; i < bh_count; i++)
918 /* Note that NULL bhs[i] is valid because of holes. */
919 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
920 ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
925 for (i = 0; i < bh_count; i++)
927 wait_on_buffer(bhs[i]);
929 for (i = 0; i < bh_count; i++) {
930 if (bhs[i] && !buffer_uptodate(bhs[i])) {
938 for (i = 0; i < bh_count; i++) {
945 int ext4_walk_page_buffers(handle_t *handle,
946 struct buffer_head *head,
950 int (*fn)(handle_t *handle,
951 struct buffer_head *bh))
953 struct buffer_head *bh;
954 unsigned block_start, block_end;
955 unsigned blocksize = head->b_size;
957 struct buffer_head *next;
959 for (bh = head, block_start = 0;
960 ret == 0 && (bh != head || !block_start);
961 block_start = block_end, bh = next) {
962 next = bh->b_this_page;
963 block_end = block_start + blocksize;
964 if (block_end <= from || block_start >= to) {
965 if (partial && !buffer_uptodate(bh))
969 err = (*fn)(handle, bh);
977 * To preserve ordering, it is essential that the hole instantiation and
978 * the data write be encapsulated in a single transaction. We cannot
979 * close off a transaction and start a new one between the ext4_get_block()
980 * and the commit_write(). So doing the jbd2_journal_start at the start of
981 * prepare_write() is the right place.
983 * Also, this function can nest inside ext4_writepage(). In that case, we
984 * *know* that ext4_writepage() has generated enough buffer credits to do the
985 * whole page. So we won't block on the journal in that case, which is good,
986 * because the caller may be PF_MEMALLOC.
988 * By accident, ext4 can be reentered when a transaction is open via
989 * quota file writes. If we were to commit the transaction while thus
990 * reentered, there can be a deadlock - we would be holding a quota
991 * lock, and the commit would never complete if another thread had a
992 * transaction open and was blocking on the quota lock - a ranking
995 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
996 * will _not_ run commit under these circumstances because handle->h_ref
997 * is elevated. We'll still have enough credits for the tiny quotafile
1000 int do_journal_get_write_access(handle_t *handle,
1001 struct buffer_head *bh)
1003 int dirty = buffer_dirty(bh);
1006 if (!buffer_mapped(bh) || buffer_freed(bh))
1009 * __block_write_begin() could have dirtied some buffers. Clean
1010 * the dirty bit as jbd2_journal_get_write_access() could complain
1011 * otherwise about fs integrity issues. Setting of the dirty bit
1012 * by __block_write_begin() isn't a real problem here as we clear
1013 * the bit before releasing a page lock and thus writeback cannot
1014 * ever write the buffer.
1017 clear_buffer_dirty(bh);
1018 BUFFER_TRACE(bh, "get write access");
1019 ret = ext4_journal_get_write_access(handle, bh);
1021 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1025 #ifdef CONFIG_FS_ENCRYPTION
1026 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1027 get_block_t *get_block)
1029 unsigned from = pos & (PAGE_SIZE - 1);
1030 unsigned to = from + len;
1031 struct inode *inode = page->mapping->host;
1032 unsigned block_start, block_end;
1035 unsigned blocksize = inode->i_sb->s_blocksize;
1037 struct buffer_head *bh, *head, *wait[2];
1041 BUG_ON(!PageLocked(page));
1042 BUG_ON(from > PAGE_SIZE);
1043 BUG_ON(to > PAGE_SIZE);
1046 if (!page_has_buffers(page))
1047 create_empty_buffers(page, blocksize, 0);
1048 head = page_buffers(page);
1049 bbits = ilog2(blocksize);
1050 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1052 for (bh = head, block_start = 0; bh != head || !block_start;
1053 block++, block_start = block_end, bh = bh->b_this_page) {
1054 block_end = block_start + blocksize;
1055 if (block_end <= from || block_start >= to) {
1056 if (PageUptodate(page)) {
1057 if (!buffer_uptodate(bh))
1058 set_buffer_uptodate(bh);
1063 clear_buffer_new(bh);
1064 if (!buffer_mapped(bh)) {
1065 WARN_ON(bh->b_size != blocksize);
1066 err = get_block(inode, block, bh, 1);
1069 if (buffer_new(bh)) {
1070 if (PageUptodate(page)) {
1071 clear_buffer_new(bh);
1072 set_buffer_uptodate(bh);
1073 mark_buffer_dirty(bh);
1076 if (block_end > to || block_start < from)
1077 zero_user_segments(page, to, block_end,
1082 if (PageUptodate(page)) {
1083 if (!buffer_uptodate(bh))
1084 set_buffer_uptodate(bh);
1087 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1088 !buffer_unwritten(bh) &&
1089 (block_start < from || block_end > to)) {
1090 ext4_read_bh_lock(bh, 0, false);
1091 wait[nr_wait++] = bh;
1095 * If we issued read requests, let them complete.
1097 for (i = 0; i < nr_wait; i++) {
1098 wait_on_buffer(wait[i]);
1099 if (!buffer_uptodate(wait[i]))
1102 if (unlikely(err)) {
1103 page_zero_new_buffers(page, from, to);
1104 } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1105 for (i = 0; i < nr_wait; i++) {
1108 err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1109 bh_offset(wait[i]));
1111 clear_buffer_uptodate(wait[i]);
1121 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1122 loff_t pos, unsigned len, unsigned flags,
1123 struct page **pagep, void **fsdata)
1125 struct inode *inode = mapping->host;
1126 int ret, needed_blocks;
1133 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1136 trace_ext4_write_begin(inode, pos, len, flags);
1138 * Reserve one block more for addition to orphan list in case
1139 * we allocate blocks but write fails for some reason
1141 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1142 index = pos >> PAGE_SHIFT;
1143 from = pos & (PAGE_SIZE - 1);
1146 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1147 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1156 * grab_cache_page_write_begin() can take a long time if the
1157 * system is thrashing due to memory pressure, or if the page
1158 * is being written back. So grab it first before we start
1159 * the transaction handle. This also allows us to allocate
1160 * the page (if needed) without using GFP_NOFS.
1163 page = grab_cache_page_write_begin(mapping, index, flags);
1169 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1170 if (IS_ERR(handle)) {
1172 return PTR_ERR(handle);
1176 if (page->mapping != mapping) {
1177 /* The page got truncated from under us */
1180 ext4_journal_stop(handle);
1183 /* In case writeback began while the page was unlocked */
1184 wait_for_stable_page(page);
1186 #ifdef CONFIG_FS_ENCRYPTION
1187 if (ext4_should_dioread_nolock(inode))
1188 ret = ext4_block_write_begin(page, pos, len,
1189 ext4_get_block_unwritten);
1191 ret = ext4_block_write_begin(page, pos, len,
1194 if (ext4_should_dioread_nolock(inode))
1195 ret = __block_write_begin(page, pos, len,
1196 ext4_get_block_unwritten);
1198 ret = __block_write_begin(page, pos, len, ext4_get_block);
1200 if (!ret && ext4_should_journal_data(inode)) {
1201 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1203 do_journal_get_write_access);
1207 bool extended = (pos + len > inode->i_size) &&
1208 !ext4_verity_in_progress(inode);
1212 * __block_write_begin may have instantiated a few blocks
1213 * outside i_size. Trim these off again. Don't need
1214 * i_size_read because we hold i_mutex.
1216 * Add inode to orphan list in case we crash before
1219 if (extended && ext4_can_truncate(inode))
1220 ext4_orphan_add(handle, inode);
1222 ext4_journal_stop(handle);
1224 ext4_truncate_failed_write(inode);
1226 * If truncate failed early the inode might
1227 * still be on the orphan list; we need to
1228 * make sure the inode is removed from the
1229 * orphan list in that case.
1232 ext4_orphan_del(NULL, inode);
1235 if (ret == -ENOSPC &&
1236 ext4_should_retry_alloc(inode->i_sb, &retries))
1245 /* For write_end() in data=journal mode */
1246 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1249 if (!buffer_mapped(bh) || buffer_freed(bh))
1251 set_buffer_uptodate(bh);
1252 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1253 clear_buffer_meta(bh);
1254 clear_buffer_prio(bh);
1259 * We need to pick up the new inode size which generic_commit_write gave us
1260 * `file' can be NULL - eg, when called from page_symlink().
1262 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1263 * buffers are managed internally.
1265 static int ext4_write_end(struct file *file,
1266 struct address_space *mapping,
1267 loff_t pos, unsigned len, unsigned copied,
1268 struct page *page, void *fsdata)
1270 handle_t *handle = ext4_journal_current_handle();
1271 struct inode *inode = mapping->host;
1272 loff_t old_size = inode->i_size;
1274 int i_size_changed = 0;
1275 int inline_data = ext4_has_inline_data(inode);
1276 bool verity = ext4_verity_in_progress(inode);
1278 trace_ext4_write_end(inode, pos, len, copied);
1280 ret = ext4_write_inline_data_end(inode, pos, len,
1289 copied = block_write_end(file, mapping, pos,
1290 len, copied, page, fsdata);
1292 * it's important to update i_size while still holding page lock:
1293 * page writeout could otherwise come in and zero beyond i_size.
1295 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1296 * blocks are being written past EOF, so skip the i_size update.
1299 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1303 if (old_size < pos && !verity)
1304 pagecache_isize_extended(inode, old_size, pos);
1306 * Don't mark the inode dirty under page lock. First, it unnecessarily
1307 * makes the holding time of page lock longer. Second, it forces lock
1308 * ordering of page lock and transaction start for journaling
1311 if (i_size_changed || inline_data)
1312 ret = ext4_mark_inode_dirty(handle, inode);
1314 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1315 /* if we have allocated more blocks and copied
1316 * less. We will have blocks allocated outside
1317 * inode->i_size. So truncate them
1319 ext4_orphan_add(handle, inode);
1321 ret2 = ext4_journal_stop(handle);
1325 if (pos + len > inode->i_size && !verity) {
1326 ext4_truncate_failed_write(inode);
1328 * If truncate failed early the inode might still be
1329 * on the orphan list; we need to make sure the inode
1330 * is removed from the orphan list in that case.
1333 ext4_orphan_del(NULL, inode);
1336 return ret ? ret : copied;
1340 * This is a private version of page_zero_new_buffers() which doesn't
1341 * set the buffer to be dirty, since in data=journalled mode we need
1342 * to call ext4_handle_dirty_metadata() instead.
1344 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1346 unsigned from, unsigned to)
1348 unsigned int block_start = 0, block_end;
1349 struct buffer_head *head, *bh;
1351 bh = head = page_buffers(page);
1353 block_end = block_start + bh->b_size;
1354 if (buffer_new(bh)) {
1355 if (block_end > from && block_start < to) {
1356 if (!PageUptodate(page)) {
1357 unsigned start, size;
1359 start = max(from, block_start);
1360 size = min(to, block_end) - start;
1362 zero_user(page, start, size);
1363 write_end_fn(handle, bh);
1365 clear_buffer_new(bh);
1368 block_start = block_end;
1369 bh = bh->b_this_page;
1370 } while (bh != head);
1373 static int ext4_journalled_write_end(struct file *file,
1374 struct address_space *mapping,
1375 loff_t pos, unsigned len, unsigned copied,
1376 struct page *page, void *fsdata)
1378 handle_t *handle = ext4_journal_current_handle();
1379 struct inode *inode = mapping->host;
1380 loff_t old_size = inode->i_size;
1384 int size_changed = 0;
1385 int inline_data = ext4_has_inline_data(inode);
1386 bool verity = ext4_verity_in_progress(inode);
1388 trace_ext4_journalled_write_end(inode, pos, len, copied);
1389 from = pos & (PAGE_SIZE - 1);
1392 BUG_ON(!ext4_handle_valid(handle));
1395 ret = ext4_write_inline_data_end(inode, pos, len,
1403 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1405 ext4_journalled_zero_new_buffers(handle, page, from, to);
1407 if (unlikely(copied < len))
1408 ext4_journalled_zero_new_buffers(handle, page,
1410 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1411 from + copied, &partial,
1414 SetPageUptodate(page);
1417 size_changed = ext4_update_inode_size(inode, pos + copied);
1418 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1419 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1423 if (old_size < pos && !verity)
1424 pagecache_isize_extended(inode, old_size, pos);
1426 if (size_changed || inline_data) {
1427 ret2 = ext4_mark_inode_dirty(handle, inode);
1432 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1433 /* if we have allocated more blocks and copied
1434 * less. We will have blocks allocated outside
1435 * inode->i_size. So truncate them
1437 ext4_orphan_add(handle, inode);
1440 ret2 = ext4_journal_stop(handle);
1443 if (pos + len > inode->i_size && !verity) {
1444 ext4_truncate_failed_write(inode);
1446 * If truncate failed early the inode might still be
1447 * on the orphan list; we need to make sure the inode
1448 * is removed from the orphan list in that case.
1451 ext4_orphan_del(NULL, inode);
1454 return ret ? ret : copied;
1458 * Reserve space for a single cluster
1460 static int ext4_da_reserve_space(struct inode *inode)
1462 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1463 struct ext4_inode_info *ei = EXT4_I(inode);
1467 * We will charge metadata quota at writeout time; this saves
1468 * us from metadata over-estimation, though we may go over by
1469 * a small amount in the end. Here we just reserve for data.
1471 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1475 spin_lock(&ei->i_block_reservation_lock);
1476 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1477 spin_unlock(&ei->i_block_reservation_lock);
1478 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1481 ei->i_reserved_data_blocks++;
1482 trace_ext4_da_reserve_space(inode);
1483 spin_unlock(&ei->i_block_reservation_lock);
1485 return 0; /* success */
1488 void ext4_da_release_space(struct inode *inode, int to_free)
1490 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1491 struct ext4_inode_info *ei = EXT4_I(inode);
1494 return; /* Nothing to release, exit */
1496 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1498 trace_ext4_da_release_space(inode, to_free);
1499 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1501 * if there aren't enough reserved blocks, then the
1502 * counter is messed up somewhere. Since this
1503 * function is called from invalidate page, it's
1504 * harmless to return without any action.
1506 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1507 "ino %lu, to_free %d with only %d reserved "
1508 "data blocks", inode->i_ino, to_free,
1509 ei->i_reserved_data_blocks);
1511 to_free = ei->i_reserved_data_blocks;
1513 ei->i_reserved_data_blocks -= to_free;
1515 /* update fs dirty data blocks counter */
1516 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1518 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1520 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1524 * Delayed allocation stuff
1527 struct mpage_da_data {
1528 struct inode *inode;
1529 struct writeback_control *wbc;
1531 pgoff_t first_page; /* The first page to write */
1532 pgoff_t next_page; /* Current page to examine */
1533 pgoff_t last_page; /* Last page to examine */
1535 * Extent to map - this can be after first_page because that can be
1536 * fully mapped. We somewhat abuse m_flags to store whether the extent
1537 * is delalloc or unwritten.
1539 struct ext4_map_blocks map;
1540 struct ext4_io_submit io_submit; /* IO submission data */
1541 unsigned int do_map:1;
1542 unsigned int scanned_until_end:1;
1545 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1550 struct pagevec pvec;
1551 struct inode *inode = mpd->inode;
1552 struct address_space *mapping = inode->i_mapping;
1554 /* This is necessary when next_page == 0. */
1555 if (mpd->first_page >= mpd->next_page)
1558 mpd->scanned_until_end = 0;
1559 index = mpd->first_page;
1560 end = mpd->next_page - 1;
1562 ext4_lblk_t start, last;
1563 start = index << (PAGE_SHIFT - inode->i_blkbits);
1564 last = end << (PAGE_SHIFT - inode->i_blkbits);
1565 ext4_es_remove_extent(inode, start, last - start + 1);
1568 pagevec_init(&pvec);
1569 while (index <= end) {
1570 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1573 for (i = 0; i < nr_pages; i++) {
1574 struct page *page = pvec.pages[i];
1576 BUG_ON(!PageLocked(page));
1577 BUG_ON(PageWriteback(page));
1579 if (page_mapped(page))
1580 clear_page_dirty_for_io(page);
1581 block_invalidatepage(page, 0, PAGE_SIZE);
1582 ClearPageUptodate(page);
1586 pagevec_release(&pvec);
1590 static void ext4_print_free_blocks(struct inode *inode)
1592 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1593 struct super_block *sb = inode->i_sb;
1594 struct ext4_inode_info *ei = EXT4_I(inode);
1596 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1597 EXT4_C2B(EXT4_SB(inode->i_sb),
1598 ext4_count_free_clusters(sb)));
1599 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1600 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1601 (long long) EXT4_C2B(EXT4_SB(sb),
1602 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1603 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1604 (long long) EXT4_C2B(EXT4_SB(sb),
1605 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1606 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1607 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1608 ei->i_reserved_data_blocks);
1612 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1614 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1618 * ext4_insert_delayed_block - adds a delayed block to the extents status
1619 * tree, incrementing the reserved cluster/block
1620 * count or making a pending reservation
1623 * @inode - file containing the newly added block
1624 * @lblk - logical block to be added
1626 * Returns 0 on success, negative error code on failure.
1628 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1630 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1632 bool allocated = false;
1635 * If the cluster containing lblk is shared with a delayed,
1636 * written, or unwritten extent in a bigalloc file system, it's
1637 * already been accounted for and does not need to be reserved.
1638 * A pending reservation must be made for the cluster if it's
1639 * shared with a written or unwritten extent and doesn't already
1640 * have one. Written and unwritten extents can be purged from the
1641 * extents status tree if the system is under memory pressure, so
1642 * it's necessary to examine the extent tree if a search of the
1643 * extents status tree doesn't get a match.
1645 if (sbi->s_cluster_ratio == 1) {
1646 ret = ext4_da_reserve_space(inode);
1647 if (ret != 0) /* ENOSPC */
1649 } else { /* bigalloc */
1650 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1651 if (!ext4_es_scan_clu(inode,
1652 &ext4_es_is_mapped, lblk)) {
1653 ret = ext4_clu_mapped(inode,
1654 EXT4_B2C(sbi, lblk));
1658 ret = ext4_da_reserve_space(inode);
1659 if (ret != 0) /* ENOSPC */
1670 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1677 * This function is grabs code from the very beginning of
1678 * ext4_map_blocks, but assumes that the caller is from delayed write
1679 * time. This function looks up the requested blocks and sets the
1680 * buffer delay bit under the protection of i_data_sem.
1682 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1683 struct ext4_map_blocks *map,
1684 struct buffer_head *bh)
1686 struct extent_status es;
1688 sector_t invalid_block = ~((sector_t) 0xffff);
1689 #ifdef ES_AGGRESSIVE_TEST
1690 struct ext4_map_blocks orig_map;
1692 memcpy(&orig_map, map, sizeof(*map));
1695 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1699 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1700 (unsigned long) map->m_lblk);
1702 /* Lookup extent status tree firstly */
1703 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1704 if (ext4_es_is_hole(&es)) {
1706 down_read(&EXT4_I(inode)->i_data_sem);
1711 * Delayed extent could be allocated by fallocate.
1712 * So we need to check it.
1714 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1715 map_bh(bh, inode->i_sb, invalid_block);
1717 set_buffer_delay(bh);
1721 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1722 retval = es.es_len - (iblock - es.es_lblk);
1723 if (retval > map->m_len)
1724 retval = map->m_len;
1725 map->m_len = retval;
1726 if (ext4_es_is_written(&es))
1727 map->m_flags |= EXT4_MAP_MAPPED;
1728 else if (ext4_es_is_unwritten(&es))
1729 map->m_flags |= EXT4_MAP_UNWRITTEN;
1733 #ifdef ES_AGGRESSIVE_TEST
1734 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1740 * Try to see if we can get the block without requesting a new
1741 * file system block.
1743 down_read(&EXT4_I(inode)->i_data_sem);
1744 if (ext4_has_inline_data(inode))
1746 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1747 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1749 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1756 * XXX: __block_prepare_write() unmaps passed block,
1760 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1766 map_bh(bh, inode->i_sb, invalid_block);
1768 set_buffer_delay(bh);
1769 } else if (retval > 0) {
1771 unsigned int status;
1773 if (unlikely(retval != map->m_len)) {
1774 ext4_warning(inode->i_sb,
1775 "ES len assertion failed for inode "
1776 "%lu: retval %d != map->m_len %d",
1777 inode->i_ino, retval, map->m_len);
1781 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1782 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1783 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1784 map->m_pblk, status);
1790 up_read((&EXT4_I(inode)->i_data_sem));
1796 * This is a special get_block_t callback which is used by
1797 * ext4_da_write_begin(). It will either return mapped block or
1798 * reserve space for a single block.
1800 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1801 * We also have b_blocknr = -1 and b_bdev initialized properly
1803 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1804 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1805 * initialized properly.
1807 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1808 struct buffer_head *bh, int create)
1810 struct ext4_map_blocks map;
1813 BUG_ON(create == 0);
1814 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1816 map.m_lblk = iblock;
1820 * first, we need to know whether the block is allocated already
1821 * preallocated blocks are unmapped but should treated
1822 * the same as allocated blocks.
1824 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1828 map_bh(bh, inode->i_sb, map.m_pblk);
1829 ext4_update_bh_state(bh, map.m_flags);
1831 if (buffer_unwritten(bh)) {
1832 /* A delayed write to unwritten bh should be marked
1833 * new and mapped. Mapped ensures that we don't do
1834 * get_block multiple times when we write to the same
1835 * offset and new ensures that we do proper zero out
1836 * for partial write.
1839 set_buffer_mapped(bh);
1844 static int bget_one(handle_t *handle, struct buffer_head *bh)
1850 static int bput_one(handle_t *handle, struct buffer_head *bh)
1856 static int __ext4_journalled_writepage(struct page *page,
1859 struct address_space *mapping = page->mapping;
1860 struct inode *inode = mapping->host;
1861 struct buffer_head *page_bufs = NULL;
1862 handle_t *handle = NULL;
1863 int ret = 0, err = 0;
1864 int inline_data = ext4_has_inline_data(inode);
1865 struct buffer_head *inode_bh = NULL;
1867 ClearPageChecked(page);
1870 BUG_ON(page->index != 0);
1871 BUG_ON(len > ext4_get_max_inline_size(inode));
1872 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1873 if (inode_bh == NULL)
1876 page_bufs = page_buffers(page);
1881 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1885 * We need to release the page lock before we start the
1886 * journal, so grab a reference so the page won't disappear
1887 * out from under us.
1892 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1893 ext4_writepage_trans_blocks(inode));
1894 if (IS_ERR(handle)) {
1895 ret = PTR_ERR(handle);
1897 goto out_no_pagelock;
1899 BUG_ON(!ext4_handle_valid(handle));
1903 if (page->mapping != mapping) {
1904 /* The page got truncated from under us */
1905 ext4_journal_stop(handle);
1911 ret = ext4_mark_inode_dirty(handle, inode);
1913 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1914 do_journal_get_write_access);
1916 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1921 err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len);
1924 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1925 err = ext4_journal_stop(handle);
1929 if (!ext4_has_inline_data(inode))
1930 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1932 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1941 * Note that we don't need to start a transaction unless we're journaling data
1942 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1943 * need to file the inode to the transaction's list in ordered mode because if
1944 * we are writing back data added by write(), the inode is already there and if
1945 * we are writing back data modified via mmap(), no one guarantees in which
1946 * transaction the data will hit the disk. In case we are journaling data, we
1947 * cannot start transaction directly because transaction start ranks above page
1948 * lock so we have to do some magic.
1950 * This function can get called via...
1951 * - ext4_writepages after taking page lock (have journal handle)
1952 * - journal_submit_inode_data_buffers (no journal handle)
1953 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1954 * - grab_page_cache when doing write_begin (have journal handle)
1956 * We don't do any block allocation in this function. If we have page with
1957 * multiple blocks we need to write those buffer_heads that are mapped. This
1958 * is important for mmaped based write. So if we do with blocksize 1K
1959 * truncate(f, 1024);
1960 * a = mmap(f, 0, 4096);
1962 * truncate(f, 4096);
1963 * we have in the page first buffer_head mapped via page_mkwrite call back
1964 * but other buffer_heads would be unmapped but dirty (dirty done via the
1965 * do_wp_page). So writepage should write the first block. If we modify
1966 * the mmap area beyond 1024 we will again get a page_fault and the
1967 * page_mkwrite callback will do the block allocation and mark the
1968 * buffer_heads mapped.
1970 * We redirty the page if we have any buffer_heads that is either delay or
1971 * unwritten in the page.
1973 * We can get recursively called as show below.
1975 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1978 * But since we don't do any block allocation we should not deadlock.
1979 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1981 static int ext4_writepage(struct page *page,
1982 struct writeback_control *wbc)
1987 struct buffer_head *page_bufs = NULL;
1988 struct inode *inode = page->mapping->host;
1989 struct ext4_io_submit io_submit;
1990 bool keep_towrite = false;
1992 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
1993 inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
1998 trace_ext4_writepage(page);
1999 size = i_size_read(inode);
2000 if (page->index == size >> PAGE_SHIFT &&
2001 !ext4_verity_in_progress(inode))
2002 len = size & ~PAGE_MASK;
2006 page_bufs = page_buffers(page);
2008 * We cannot do block allocation or other extent handling in this
2009 * function. If there are buffers needing that, we have to redirty
2010 * the page. But we may reach here when we do a journal commit via
2011 * journal_submit_inode_data_buffers() and in that case we must write
2012 * allocated buffers to achieve data=ordered mode guarantees.
2014 * Also, if there is only one buffer per page (the fs block
2015 * size == the page size), if one buffer needs block
2016 * allocation or needs to modify the extent tree to clear the
2017 * unwritten flag, we know that the page can't be written at
2018 * all, so we might as well refuse the write immediately.
2019 * Unfortunately if the block size != page size, we can't as
2020 * easily detect this case using ext4_walk_page_buffers(), but
2021 * for the extremely common case, this is an optimization that
2022 * skips a useless round trip through ext4_bio_write_page().
2024 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2025 ext4_bh_delay_or_unwritten)) {
2026 redirty_page_for_writepage(wbc, page);
2027 if ((current->flags & PF_MEMALLOC) ||
2028 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2030 * For memory cleaning there's no point in writing only
2031 * some buffers. So just bail out. Warn if we came here
2032 * from direct reclaim.
2034 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2039 keep_towrite = true;
2042 if (PageChecked(page) && ext4_should_journal_data(inode))
2044 * It's mmapped pagecache. Add buffers and journal it. There
2045 * doesn't seem much point in redirtying the page here.
2047 return __ext4_journalled_writepage(page, len);
2049 ext4_io_submit_init(&io_submit, wbc);
2050 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2051 if (!io_submit.io_end) {
2052 redirty_page_for_writepage(wbc, page);
2056 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2057 ext4_io_submit(&io_submit);
2058 /* Drop io_end reference we got from init */
2059 ext4_put_io_end_defer(io_submit.io_end);
2063 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2069 BUG_ON(page->index != mpd->first_page);
2070 clear_page_dirty_for_io(page);
2072 * We have to be very careful here! Nothing protects writeback path
2073 * against i_size changes and the page can be writeably mapped into
2074 * page tables. So an application can be growing i_size and writing
2075 * data through mmap while writeback runs. clear_page_dirty_for_io()
2076 * write-protects our page in page tables and the page cannot get
2077 * written to again until we release page lock. So only after
2078 * clear_page_dirty_for_io() we are safe to sample i_size for
2079 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2080 * on the barrier provided by TestClearPageDirty in
2081 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2082 * after page tables are updated.
2084 size = i_size_read(mpd->inode);
2085 if (page->index == size >> PAGE_SHIFT &&
2086 !ext4_verity_in_progress(mpd->inode))
2087 len = size & ~PAGE_MASK;
2090 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2092 mpd->wbc->nr_to_write--;
2098 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2101 * mballoc gives us at most this number of blocks...
2102 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2103 * The rest of mballoc seems to handle chunks up to full group size.
2105 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2108 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2110 * @mpd - extent of blocks
2111 * @lblk - logical number of the block in the file
2112 * @bh - buffer head we want to add to the extent
2114 * The function is used to collect contig. blocks in the same state. If the
2115 * buffer doesn't require mapping for writeback and we haven't started the
2116 * extent of buffers to map yet, the function returns 'true' immediately - the
2117 * caller can write the buffer right away. Otherwise the function returns true
2118 * if the block has been added to the extent, false if the block couldn't be
2121 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2122 struct buffer_head *bh)
2124 struct ext4_map_blocks *map = &mpd->map;
2126 /* Buffer that doesn't need mapping for writeback? */
2127 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2128 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2129 /* So far no extent to map => we write the buffer right away */
2130 if (map->m_len == 0)
2135 /* First block in the extent? */
2136 if (map->m_len == 0) {
2137 /* We cannot map unless handle is started... */
2142 map->m_flags = bh->b_state & BH_FLAGS;
2146 /* Don't go larger than mballoc is willing to allocate */
2147 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2150 /* Can we merge the block to our big extent? */
2151 if (lblk == map->m_lblk + map->m_len &&
2152 (bh->b_state & BH_FLAGS) == map->m_flags) {
2160 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2162 * @mpd - extent of blocks for mapping
2163 * @head - the first buffer in the page
2164 * @bh - buffer we should start processing from
2165 * @lblk - logical number of the block in the file corresponding to @bh
2167 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2168 * the page for IO if all buffers in this page were mapped and there's no
2169 * accumulated extent of buffers to map or add buffers in the page to the
2170 * extent of buffers to map. The function returns 1 if the caller can continue
2171 * by processing the next page, 0 if it should stop adding buffers to the
2172 * extent to map because we cannot extend it anymore. It can also return value
2173 * < 0 in case of error during IO submission.
2175 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2176 struct buffer_head *head,
2177 struct buffer_head *bh,
2180 struct inode *inode = mpd->inode;
2182 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2183 >> inode->i_blkbits;
2185 if (ext4_verity_in_progress(inode))
2186 blocks = EXT_MAX_BLOCKS;
2189 BUG_ON(buffer_locked(bh));
2191 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2192 /* Found extent to map? */
2195 /* Buffer needs mapping and handle is not started? */
2198 /* Everything mapped so far and we hit EOF */
2201 } while (lblk++, (bh = bh->b_this_page) != head);
2202 /* So far everything mapped? Submit the page for IO. */
2203 if (mpd->map.m_len == 0) {
2204 err = mpage_submit_page(mpd, head->b_page);
2208 if (lblk >= blocks) {
2209 mpd->scanned_until_end = 1;
2216 * mpage_process_page - update page buffers corresponding to changed extent and
2217 * may submit fully mapped page for IO
2219 * @mpd - description of extent to map, on return next extent to map
2220 * @m_lblk - logical block mapping.
2221 * @m_pblk - corresponding physical mapping.
2222 * @map_bh - determines on return whether this page requires any further
2224 * Scan given page buffers corresponding to changed extent and update buffer
2225 * state according to new extent state.
2226 * We map delalloc buffers to their physical location, clear unwritten bits.
2227 * If the given page is not fully mapped, we update @map to the next extent in
2228 * the given page that needs mapping & return @map_bh as true.
2230 static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2231 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2234 struct buffer_head *head, *bh;
2235 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2236 ext4_lblk_t lblk = *m_lblk;
2237 ext4_fsblk_t pblock = *m_pblk;
2239 int blkbits = mpd->inode->i_blkbits;
2240 ssize_t io_end_size = 0;
2241 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2243 bh = head = page_buffers(page);
2245 if (lblk < mpd->map.m_lblk)
2247 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2249 * Buffer after end of mapped extent.
2250 * Find next buffer in the page to map.
2253 mpd->map.m_flags = 0;
2254 io_end_vec->size += io_end_size;
2257 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2260 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2261 io_end_vec = ext4_alloc_io_end_vec(io_end);
2262 if (IS_ERR(io_end_vec)) {
2263 err = PTR_ERR(io_end_vec);
2266 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2271 if (buffer_delay(bh)) {
2272 clear_buffer_delay(bh);
2273 bh->b_blocknr = pblock++;
2275 clear_buffer_unwritten(bh);
2276 io_end_size += (1 << blkbits);
2277 } while (lblk++, (bh = bh->b_this_page) != head);
2279 io_end_vec->size += io_end_size;
2289 * mpage_map_buffers - update buffers corresponding to changed extent and
2290 * submit fully mapped pages for IO
2292 * @mpd - description of extent to map, on return next extent to map
2294 * Scan buffers corresponding to changed extent (we expect corresponding pages
2295 * to be already locked) and update buffer state according to new extent state.
2296 * We map delalloc buffers to their physical location, clear unwritten bits,
2297 * and mark buffers as uninit when we perform writes to unwritten extents
2298 * and do extent conversion after IO is finished. If the last page is not fully
2299 * mapped, we update @map to the next extent in the last page that needs
2300 * mapping. Otherwise we submit the page for IO.
2302 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2304 struct pagevec pvec;
2306 struct inode *inode = mpd->inode;
2307 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2310 ext4_fsblk_t pblock;
2312 bool map_bh = false;
2314 start = mpd->map.m_lblk >> bpp_bits;
2315 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2316 lblk = start << bpp_bits;
2317 pblock = mpd->map.m_pblk;
2319 pagevec_init(&pvec);
2320 while (start <= end) {
2321 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2325 for (i = 0; i < nr_pages; i++) {
2326 struct page *page = pvec.pages[i];
2328 err = mpage_process_page(mpd, page, &lblk, &pblock,
2331 * If map_bh is true, means page may require further bh
2332 * mapping, or maybe the page was submitted for IO.
2333 * So we return to call further extent mapping.
2335 if (err < 0 || map_bh)
2337 /* Page fully mapped - let IO run! */
2338 err = mpage_submit_page(mpd, page);
2342 pagevec_release(&pvec);
2344 /* Extent fully mapped and matches with page boundary. We are done. */
2346 mpd->map.m_flags = 0;
2349 pagevec_release(&pvec);
2353 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2355 struct inode *inode = mpd->inode;
2356 struct ext4_map_blocks *map = &mpd->map;
2357 int get_blocks_flags;
2358 int err, dioread_nolock;
2360 trace_ext4_da_write_pages_extent(inode, map);
2362 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2363 * to convert an unwritten extent to be initialized (in the case
2364 * where we have written into one or more preallocated blocks). It is
2365 * possible that we're going to need more metadata blocks than
2366 * previously reserved. However we must not fail because we're in
2367 * writeback and there is nothing we can do about it so it might result
2368 * in data loss. So use reserved blocks to allocate metadata if
2371 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2372 * the blocks in question are delalloc blocks. This indicates
2373 * that the blocks and quotas has already been checked when
2374 * the data was copied into the page cache.
2376 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2377 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2378 EXT4_GET_BLOCKS_IO_SUBMIT;
2379 dioread_nolock = ext4_should_dioread_nolock(inode);
2381 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2382 if (map->m_flags & BIT(BH_Delay))
2383 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2385 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2388 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2389 if (!mpd->io_submit.io_end->handle &&
2390 ext4_handle_valid(handle)) {
2391 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2392 handle->h_rsv_handle = NULL;
2394 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2397 BUG_ON(map->m_len == 0);
2402 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2403 * mpd->len and submit pages underlying it for IO
2405 * @handle - handle for journal operations
2406 * @mpd - extent to map
2407 * @give_up_on_write - we set this to true iff there is a fatal error and there
2408 * is no hope of writing the data. The caller should discard
2409 * dirty pages to avoid infinite loops.
2411 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2412 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2413 * them to initialized or split the described range from larger unwritten
2414 * extent. Note that we need not map all the described range since allocation
2415 * can return less blocks or the range is covered by more unwritten extents. We
2416 * cannot map more because we are limited by reserved transaction credits. On
2417 * the other hand we always make sure that the last touched page is fully
2418 * mapped so that it can be written out (and thus forward progress is
2419 * guaranteed). After mapping we submit all mapped pages for IO.
2421 static int mpage_map_and_submit_extent(handle_t *handle,
2422 struct mpage_da_data *mpd,
2423 bool *give_up_on_write)
2425 struct inode *inode = mpd->inode;
2426 struct ext4_map_blocks *map = &mpd->map;
2430 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2431 struct ext4_io_end_vec *io_end_vec;
2433 io_end_vec = ext4_alloc_io_end_vec(io_end);
2434 if (IS_ERR(io_end_vec))
2435 return PTR_ERR(io_end_vec);
2436 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2438 err = mpage_map_one_extent(handle, mpd);
2440 struct super_block *sb = inode->i_sb;
2442 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2443 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2444 goto invalidate_dirty_pages;
2446 * Let the uper layers retry transient errors.
2447 * In the case of ENOSPC, if ext4_count_free_blocks()
2448 * is non-zero, a commit should free up blocks.
2450 if ((err == -ENOMEM) ||
2451 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2453 goto update_disksize;
2456 ext4_msg(sb, KERN_CRIT,
2457 "Delayed block allocation failed for "
2458 "inode %lu at logical offset %llu with"
2459 " max blocks %u with error %d",
2461 (unsigned long long)map->m_lblk,
2462 (unsigned)map->m_len, -err);
2463 ext4_msg(sb, KERN_CRIT,
2464 "This should not happen!! Data will "
2467 ext4_print_free_blocks(inode);
2468 invalidate_dirty_pages:
2469 *give_up_on_write = true;
2474 * Update buffer state, submit mapped pages, and get us new
2477 err = mpage_map_and_submit_buffers(mpd);
2479 goto update_disksize;
2480 } while (map->m_len);
2484 * Update on-disk size after IO is submitted. Races with
2485 * truncate are avoided by checking i_size under i_data_sem.
2487 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2488 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2492 down_write(&EXT4_I(inode)->i_data_sem);
2493 i_size = i_size_read(inode);
2494 if (disksize > i_size)
2496 if (disksize > EXT4_I(inode)->i_disksize)
2497 EXT4_I(inode)->i_disksize = disksize;
2498 up_write(&EXT4_I(inode)->i_data_sem);
2499 err2 = ext4_mark_inode_dirty(handle, inode);
2501 ext4_error_err(inode->i_sb, -err2,
2502 "Failed to mark inode %lu dirty",
2512 * Calculate the total number of credits to reserve for one writepages
2513 * iteration. This is called from ext4_writepages(). We map an extent of
2514 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2515 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2516 * bpp - 1 blocks in bpp different extents.
2518 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2520 int bpp = ext4_journal_blocks_per_page(inode);
2522 return ext4_meta_trans_blocks(inode,
2523 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2527 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2528 * and underlying extent to map
2530 * @mpd - where to look for pages
2532 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2533 * IO immediately. When we find a page which isn't mapped we start accumulating
2534 * extent of buffers underlying these pages that needs mapping (formed by
2535 * either delayed or unwritten buffers). We also lock the pages containing
2536 * these buffers. The extent found is returned in @mpd structure (starting at
2537 * mpd->lblk with length mpd->len blocks).
2539 * Note that this function can attach bios to one io_end structure which are
2540 * neither logically nor physically contiguous. Although it may seem as an
2541 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2542 * case as we need to track IO to all buffers underlying a page in one io_end.
2544 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2546 struct address_space *mapping = mpd->inode->i_mapping;
2547 struct pagevec pvec;
2548 unsigned int nr_pages;
2549 long left = mpd->wbc->nr_to_write;
2550 pgoff_t index = mpd->first_page;
2551 pgoff_t end = mpd->last_page;
2554 int blkbits = mpd->inode->i_blkbits;
2556 struct buffer_head *head;
2558 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2559 tag = PAGECACHE_TAG_TOWRITE;
2561 tag = PAGECACHE_TAG_DIRTY;
2563 pagevec_init(&pvec);
2565 mpd->next_page = index;
2566 while (index <= end) {
2567 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2572 for (i = 0; i < nr_pages; i++) {
2573 struct page *page = pvec.pages[i];
2576 * Accumulated enough dirty pages? This doesn't apply
2577 * to WB_SYNC_ALL mode. For integrity sync we have to
2578 * keep going because someone may be concurrently
2579 * dirtying pages, and we might have synced a lot of
2580 * newly appeared dirty pages, but have not synced all
2581 * of the old dirty pages.
2583 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2586 /* If we can't merge this page, we are done. */
2587 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2592 * If the page is no longer dirty, or its mapping no
2593 * longer corresponds to inode we are writing (which
2594 * means it has been truncated or invalidated), or the
2595 * page is already under writeback and we are not doing
2596 * a data integrity writeback, skip the page
2598 if (!PageDirty(page) ||
2599 (PageWriteback(page) &&
2600 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2601 unlikely(page->mapping != mapping)) {
2606 wait_on_page_writeback(page);
2607 BUG_ON(PageWriteback(page));
2609 if (mpd->map.m_len == 0)
2610 mpd->first_page = page->index;
2611 mpd->next_page = page->index + 1;
2612 /* Add all dirty buffers to mpd */
2613 lblk = ((ext4_lblk_t)page->index) <<
2614 (PAGE_SHIFT - blkbits);
2615 head = page_buffers(page);
2616 err = mpage_process_page_bufs(mpd, head, head, lblk);
2622 pagevec_release(&pvec);
2625 mpd->scanned_until_end = 1;
2628 pagevec_release(&pvec);
2632 static int ext4_writepages(struct address_space *mapping,
2633 struct writeback_control *wbc)
2635 pgoff_t writeback_index = 0;
2636 long nr_to_write = wbc->nr_to_write;
2637 int range_whole = 0;
2639 handle_t *handle = NULL;
2640 struct mpage_da_data mpd;
2641 struct inode *inode = mapping->host;
2642 int needed_blocks, rsv_blocks = 0, ret = 0;
2643 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2644 struct blk_plug plug;
2645 bool give_up_on_write = false;
2647 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2650 percpu_down_read(&sbi->s_writepages_rwsem);
2651 trace_ext4_writepages(inode, wbc);
2654 * No pages to write? This is mainly a kludge to avoid starting
2655 * a transaction for special inodes like journal inode on last iput()
2656 * because that could violate lock ordering on umount
2658 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2659 goto out_writepages;
2661 if (ext4_should_journal_data(inode)) {
2662 ret = generic_writepages(mapping, wbc);
2663 goto out_writepages;
2667 * If the filesystem has aborted, it is read-only, so return
2668 * right away instead of dumping stack traces later on that
2669 * will obscure the real source of the problem. We test
2670 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2671 * the latter could be true if the filesystem is mounted
2672 * read-only, and in that case, ext4_writepages should
2673 * *never* be called, so if that ever happens, we would want
2676 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2677 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2679 goto out_writepages;
2683 * If we have inline data and arrive here, it means that
2684 * we will soon create the block for the 1st page, so
2685 * we'd better clear the inline data here.
2687 if (ext4_has_inline_data(inode)) {
2688 /* Just inode will be modified... */
2689 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2690 if (IS_ERR(handle)) {
2691 ret = PTR_ERR(handle);
2692 goto out_writepages;
2694 BUG_ON(ext4_test_inode_state(inode,
2695 EXT4_STATE_MAY_INLINE_DATA));
2696 ext4_destroy_inline_data(handle, inode);
2697 ext4_journal_stop(handle);
2700 if (ext4_should_dioread_nolock(inode)) {
2702 * We may need to convert up to one extent per block in
2703 * the page and we may dirty the inode.
2705 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2706 PAGE_SIZE >> inode->i_blkbits);
2709 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2712 if (wbc->range_cyclic) {
2713 writeback_index = mapping->writeback_index;
2714 if (writeback_index)
2716 mpd.first_page = writeback_index;
2719 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2720 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2725 ext4_io_submit_init(&mpd.io_submit, wbc);
2727 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2728 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2729 blk_start_plug(&plug);
2732 * First writeback pages that don't need mapping - we can avoid
2733 * starting a transaction unnecessarily and also avoid being blocked
2734 * in the block layer on device congestion while having transaction
2738 mpd.scanned_until_end = 0;
2739 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2740 if (!mpd.io_submit.io_end) {
2744 ret = mpage_prepare_extent_to_map(&mpd);
2745 /* Unlock pages we didn't use */
2746 mpage_release_unused_pages(&mpd, false);
2747 /* Submit prepared bio */
2748 ext4_io_submit(&mpd.io_submit);
2749 ext4_put_io_end_defer(mpd.io_submit.io_end);
2750 mpd.io_submit.io_end = NULL;
2754 while (!mpd.scanned_until_end && wbc->nr_to_write > 0) {
2755 /* For each extent of pages we use new io_end */
2756 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2757 if (!mpd.io_submit.io_end) {
2763 * We have two constraints: We find one extent to map and we
2764 * must always write out whole page (makes a difference when
2765 * blocksize < pagesize) so that we don't block on IO when we
2766 * try to write out the rest of the page. Journalled mode is
2767 * not supported by delalloc.
2769 BUG_ON(ext4_should_journal_data(inode));
2770 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2772 /* start a new transaction */
2773 handle = ext4_journal_start_with_reserve(inode,
2774 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2775 if (IS_ERR(handle)) {
2776 ret = PTR_ERR(handle);
2777 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2778 "%ld pages, ino %lu; err %d", __func__,
2779 wbc->nr_to_write, inode->i_ino, ret);
2780 /* Release allocated io_end */
2781 ext4_put_io_end(mpd.io_submit.io_end);
2782 mpd.io_submit.io_end = NULL;
2787 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2788 ret = mpage_prepare_extent_to_map(&mpd);
2789 if (!ret && mpd.map.m_len)
2790 ret = mpage_map_and_submit_extent(handle, &mpd,
2793 * Caution: If the handle is synchronous,
2794 * ext4_journal_stop() can wait for transaction commit
2795 * to finish which may depend on writeback of pages to
2796 * complete or on page lock to be released. In that
2797 * case, we have to wait until after we have
2798 * submitted all the IO, released page locks we hold,
2799 * and dropped io_end reference (for extent conversion
2800 * to be able to complete) before stopping the handle.
2802 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2803 ext4_journal_stop(handle);
2807 /* Unlock pages we didn't use */
2808 mpage_release_unused_pages(&mpd, give_up_on_write);
2809 /* Submit prepared bio */
2810 ext4_io_submit(&mpd.io_submit);
2813 * Drop our io_end reference we got from init. We have
2814 * to be careful and use deferred io_end finishing if
2815 * we are still holding the transaction as we can
2816 * release the last reference to io_end which may end
2817 * up doing unwritten extent conversion.
2820 ext4_put_io_end_defer(mpd.io_submit.io_end);
2821 ext4_journal_stop(handle);
2823 ext4_put_io_end(mpd.io_submit.io_end);
2824 mpd.io_submit.io_end = NULL;
2826 if (ret == -ENOSPC && sbi->s_journal) {
2828 * Commit the transaction which would
2829 * free blocks released in the transaction
2832 jbd2_journal_force_commit_nested(sbi->s_journal);
2836 /* Fatal error - ENOMEM, EIO... */
2841 blk_finish_plug(&plug);
2842 if (!ret && !cycled && wbc->nr_to_write > 0) {
2844 mpd.last_page = writeback_index - 1;
2850 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2852 * Set the writeback_index so that range_cyclic
2853 * mode will write it back later
2855 mapping->writeback_index = mpd.first_page;
2858 trace_ext4_writepages_result(inode, wbc, ret,
2859 nr_to_write - wbc->nr_to_write);
2860 percpu_up_read(&sbi->s_writepages_rwsem);
2864 static int ext4_dax_writepages(struct address_space *mapping,
2865 struct writeback_control *wbc)
2868 long nr_to_write = wbc->nr_to_write;
2869 struct inode *inode = mapping->host;
2870 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2872 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2875 percpu_down_read(&sbi->s_writepages_rwsem);
2876 trace_ext4_writepages(inode, wbc);
2878 ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
2879 trace_ext4_writepages_result(inode, wbc, ret,
2880 nr_to_write - wbc->nr_to_write);
2881 percpu_up_read(&sbi->s_writepages_rwsem);
2885 static int ext4_nonda_switch(struct super_block *sb)
2887 s64 free_clusters, dirty_clusters;
2888 struct ext4_sb_info *sbi = EXT4_SB(sb);
2891 * switch to non delalloc mode if we are running low
2892 * on free block. The free block accounting via percpu
2893 * counters can get slightly wrong with percpu_counter_batch getting
2894 * accumulated on each CPU without updating global counters
2895 * Delalloc need an accurate free block accounting. So switch
2896 * to non delalloc when we are near to error range.
2899 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2901 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2903 * Start pushing delalloc when 1/2 of free blocks are dirty.
2905 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2906 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2908 if (2 * free_clusters < 3 * dirty_clusters ||
2909 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2911 * free block count is less than 150% of dirty blocks
2912 * or free blocks is less than watermark
2919 /* We always reserve for an inode update; the superblock could be there too */
2920 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2922 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2925 if (pos + len <= 0x7fffffffULL)
2928 /* We might need to update the superblock to set LARGE_FILE */
2932 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2933 loff_t pos, unsigned len, unsigned flags,
2934 struct page **pagep, void **fsdata)
2936 int ret, retries = 0;
2939 struct inode *inode = mapping->host;
2942 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2945 index = pos >> PAGE_SHIFT;
2947 if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) ||
2948 ext4_verity_in_progress(inode)) {
2949 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2950 return ext4_write_begin(file, mapping, pos,
2951 len, flags, pagep, fsdata);
2953 *fsdata = (void *)0;
2954 trace_ext4_da_write_begin(inode, pos, len, flags);
2956 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2957 ret = ext4_da_write_inline_data_begin(mapping, inode,
2967 * grab_cache_page_write_begin() can take a long time if the
2968 * system is thrashing due to memory pressure, or if the page
2969 * is being written back. So grab it first before we start
2970 * the transaction handle. This also allows us to allocate
2971 * the page (if needed) without using GFP_NOFS.
2974 page = grab_cache_page_write_begin(mapping, index, flags);
2980 * With delayed allocation, we don't log the i_disksize update
2981 * if there is delayed block allocation. But we still need
2982 * to journalling the i_disksize update if writes to the end
2983 * of file which has an already mapped buffer.
2986 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2987 ext4_da_write_credits(inode, pos, len));
2988 if (IS_ERR(handle)) {
2990 return PTR_ERR(handle);
2994 if (page->mapping != mapping) {
2995 /* The page got truncated from under us */
2998 ext4_journal_stop(handle);
3001 /* In case writeback began while the page was unlocked */
3002 wait_for_stable_page(page);
3004 #ifdef CONFIG_FS_ENCRYPTION
3005 ret = ext4_block_write_begin(page, pos, len,
3006 ext4_da_get_block_prep);
3008 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3012 ext4_journal_stop(handle);
3014 * block_write_begin may have instantiated a few blocks
3015 * outside i_size. Trim these off again. Don't need
3016 * i_size_read because we hold i_mutex.
3018 if (pos + len > inode->i_size)
3019 ext4_truncate_failed_write(inode);
3021 if (ret == -ENOSPC &&
3022 ext4_should_retry_alloc(inode->i_sb, &retries))
3034 * Check if we should update i_disksize
3035 * when write to the end of file but not require block allocation
3037 static int ext4_da_should_update_i_disksize(struct page *page,
3038 unsigned long offset)
3040 struct buffer_head *bh;
3041 struct inode *inode = page->mapping->host;
3045 bh = page_buffers(page);
3046 idx = offset >> inode->i_blkbits;
3048 for (i = 0; i < idx; i++)
3049 bh = bh->b_this_page;
3051 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3056 static int ext4_da_write_end(struct file *file,
3057 struct address_space *mapping,
3058 loff_t pos, unsigned len, unsigned copied,
3059 struct page *page, void *fsdata)
3061 struct inode *inode = mapping->host;
3063 handle_t *handle = ext4_journal_current_handle();
3065 unsigned long start, end;
3066 int write_mode = (int)(unsigned long)fsdata;
3068 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3069 return ext4_write_end(file, mapping, pos,
3070 len, copied, page, fsdata);
3072 trace_ext4_da_write_end(inode, pos, len, copied);
3073 start = pos & (PAGE_SIZE - 1);
3074 end = start + copied - 1;
3077 * generic_write_end() will run mark_inode_dirty() if i_size
3078 * changes. So let's piggyback the i_disksize mark_inode_dirty
3081 new_i_size = pos + copied;
3082 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3083 if (ext4_has_inline_data(inode) ||
3084 ext4_da_should_update_i_disksize(page, end)) {
3085 ext4_update_i_disksize(inode, new_i_size);
3086 /* We need to mark inode dirty even if
3087 * new_i_size is less that inode->i_size
3088 * bu greater than i_disksize.(hint delalloc)
3090 ret = ext4_mark_inode_dirty(handle, inode);
3094 if (write_mode != CONVERT_INLINE_DATA &&
3095 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3096 ext4_has_inline_data(inode))
3097 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3100 ret2 = generic_write_end(file, mapping, pos, len, copied,
3106 ret2 = ext4_journal_stop(handle);
3107 if (unlikely(ret2 && !ret))
3110 return ret ? ret : copied;
3114 * Force all delayed allocation blocks to be allocated for a given inode.
3116 int ext4_alloc_da_blocks(struct inode *inode)
3118 trace_ext4_alloc_da_blocks(inode);
3120 if (!EXT4_I(inode)->i_reserved_data_blocks)
3124 * We do something simple for now. The filemap_flush() will
3125 * also start triggering a write of the data blocks, which is
3126 * not strictly speaking necessary (and for users of
3127 * laptop_mode, not even desirable). However, to do otherwise
3128 * would require replicating code paths in:
3130 * ext4_writepages() ->
3131 * write_cache_pages() ---> (via passed in callback function)
3132 * __mpage_da_writepage() -->
3133 * mpage_add_bh_to_extent()
3134 * mpage_da_map_blocks()
3136 * The problem is that write_cache_pages(), located in
3137 * mm/page-writeback.c, marks pages clean in preparation for
3138 * doing I/O, which is not desirable if we're not planning on
3141 * We could call write_cache_pages(), and then redirty all of
3142 * the pages by calling redirty_page_for_writepage() but that
3143 * would be ugly in the extreme. So instead we would need to
3144 * replicate parts of the code in the above functions,
3145 * simplifying them because we wouldn't actually intend to
3146 * write out the pages, but rather only collect contiguous
3147 * logical block extents, call the multi-block allocator, and
3148 * then update the buffer heads with the block allocations.
3150 * For now, though, we'll cheat by calling filemap_flush(),
3151 * which will map the blocks, and start the I/O, but not
3152 * actually wait for the I/O to complete.
3154 return filemap_flush(inode->i_mapping);
3158 * bmap() is special. It gets used by applications such as lilo and by
3159 * the swapper to find the on-disk block of a specific piece of data.
3161 * Naturally, this is dangerous if the block concerned is still in the
3162 * journal. If somebody makes a swapfile on an ext4 data-journaling
3163 * filesystem and enables swap, then they may get a nasty shock when the
3164 * data getting swapped to that swapfile suddenly gets overwritten by
3165 * the original zero's written out previously to the journal and
3166 * awaiting writeback in the kernel's buffer cache.
3168 * So, if we see any bmap calls here on a modified, data-journaled file,
3169 * take extra steps to flush any blocks which might be in the cache.
3171 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3173 struct inode *inode = mapping->host;
3178 * We can get here for an inline file via the FIBMAP ioctl
3180 if (ext4_has_inline_data(inode))
3183 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3184 test_opt(inode->i_sb, DELALLOC)) {
3186 * With delalloc we want to sync the file
3187 * so that we can make sure we allocate
3190 filemap_write_and_wait(mapping);
3193 if (EXT4_JOURNAL(inode) &&
3194 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3196 * This is a REALLY heavyweight approach, but the use of
3197 * bmap on dirty files is expected to be extremely rare:
3198 * only if we run lilo or swapon on a freshly made file
3199 * do we expect this to happen.
3201 * (bmap requires CAP_SYS_RAWIO so this does not
3202 * represent an unprivileged user DOS attack --- we'd be
3203 * in trouble if mortal users could trigger this path at
3206 * NB. EXT4_STATE_JDATA is not set on files other than
3207 * regular files. If somebody wants to bmap a directory
3208 * or symlink and gets confused because the buffer
3209 * hasn't yet been flushed to disk, they deserve
3210 * everything they get.
3213 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3214 journal = EXT4_JOURNAL(inode);
3215 jbd2_journal_lock_updates(journal);
3216 err = jbd2_journal_flush(journal);
3217 jbd2_journal_unlock_updates(journal);
3223 return iomap_bmap(mapping, block, &ext4_iomap_ops);
3226 static int ext4_readpage(struct file *file, struct page *page)
3229 struct inode *inode = page->mapping->host;
3231 trace_ext4_readpage(page);
3233 if (ext4_has_inline_data(inode))
3234 ret = ext4_readpage_inline(inode, page);
3237 return ext4_mpage_readpages(inode, NULL, page);
3242 static void ext4_readahead(struct readahead_control *rac)
3244 struct inode *inode = rac->mapping->host;
3246 /* If the file has inline data, no need to do readahead. */
3247 if (ext4_has_inline_data(inode))
3250 ext4_mpage_readpages(inode, rac, NULL);
3253 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3254 unsigned int length)
3256 trace_ext4_invalidatepage(page, offset, length);
3258 /* No journalling happens on data buffers when this function is used */
3259 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3261 block_invalidatepage(page, offset, length);
3264 static int __ext4_journalled_invalidatepage(struct page *page,
3265 unsigned int offset,
3266 unsigned int length)
3268 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3270 trace_ext4_journalled_invalidatepage(page, offset, length);
3273 * If it's a full truncate we just forget about the pending dirtying
3275 if (offset == 0 && length == PAGE_SIZE)
3276 ClearPageChecked(page);
3278 return jbd2_journal_invalidatepage(journal, page, offset, length);
3281 /* Wrapper for aops... */
3282 static void ext4_journalled_invalidatepage(struct page *page,
3283 unsigned int offset,
3284 unsigned int length)
3286 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3289 static int ext4_releasepage(struct page *page, gfp_t wait)
3291 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3293 trace_ext4_releasepage(page);
3295 /* Page has dirty journalled data -> cannot release */
3296 if (PageChecked(page))
3299 return jbd2_journal_try_to_free_buffers(journal, page);
3301 return try_to_free_buffers(page);
3304 static bool ext4_inode_datasync_dirty(struct inode *inode)
3306 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3309 if (jbd2_transaction_committed(journal,
3310 EXT4_I(inode)->i_datasync_tid))
3312 if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3313 return atomic_read(&EXT4_SB(inode->i_sb)->s_fc_subtid) <
3314 EXT4_I(inode)->i_fc_committed_subtid;
3318 /* Any metadata buffers to write? */
3319 if (!list_empty(&inode->i_mapping->private_list))
3321 return inode->i_state & I_DIRTY_DATASYNC;
3324 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3325 struct ext4_map_blocks *map, loff_t offset,
3328 u8 blkbits = inode->i_blkbits;
3331 * Writes that span EOF might trigger an I/O size update on completion,
3332 * so consider them to be dirty for the purpose of O_DSYNC, even if
3333 * there is no other metadata changes being made or are pending.
3336 if (ext4_inode_datasync_dirty(inode) ||
3337 offset + length > i_size_read(inode))
3338 iomap->flags |= IOMAP_F_DIRTY;
3340 if (map->m_flags & EXT4_MAP_NEW)
3341 iomap->flags |= IOMAP_F_NEW;
3343 iomap->bdev = inode->i_sb->s_bdev;
3344 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3345 iomap->offset = (u64) map->m_lblk << blkbits;
3346 iomap->length = (u64) map->m_len << blkbits;
3348 if ((map->m_flags & EXT4_MAP_MAPPED) &&
3349 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3350 iomap->flags |= IOMAP_F_MERGED;
3353 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3354 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3355 * set. In order for any allocated unwritten extents to be converted
3356 * into written extents correctly within the ->end_io() handler, we
3357 * need to ensure that the iomap->type is set appropriately. Hence, the
3358 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3361 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3362 iomap->type = IOMAP_UNWRITTEN;
3363 iomap->addr = (u64) map->m_pblk << blkbits;
3364 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3365 iomap->type = IOMAP_MAPPED;
3366 iomap->addr = (u64) map->m_pblk << blkbits;
3368 iomap->type = IOMAP_HOLE;
3369 iomap->addr = IOMAP_NULL_ADDR;
3373 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3377 u8 blkbits = inode->i_blkbits;
3378 int ret, dio_credits, m_flags = 0, retries = 0;
3381 * Trim the mapping request to the maximum value that we can map at
3382 * once for direct I/O.
3384 if (map->m_len > DIO_MAX_BLOCKS)
3385 map->m_len = DIO_MAX_BLOCKS;
3386 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3390 * Either we allocate blocks and then don't get an unwritten extent, so
3391 * in that case we have reserved enough credits. Or, the blocks are
3392 * already allocated and unwritten. In that case, the extent conversion
3393 * fits into the credits as well.
3395 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3397 return PTR_ERR(handle);
3400 * DAX and direct I/O are the only two operations that are currently
3401 * supported with IOMAP_WRITE.
3403 WARN_ON(!IS_DAX(inode) && !(flags & IOMAP_DIRECT));
3405 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3407 * We use i_size instead of i_disksize here because delalloc writeback
3408 * can complete at any point during the I/O and subsequently push the
3409 * i_disksize out to i_size. This could be beyond where direct I/O is
3410 * happening and thus expose allocated blocks to direct I/O reads.
3412 else if ((map->m_lblk * (1 << blkbits)) >= i_size_read(inode))
3413 m_flags = EXT4_GET_BLOCKS_CREATE;
3414 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3415 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3417 ret = ext4_map_blocks(handle, inode, map, m_flags);
3420 * We cannot fill holes in indirect tree based inodes as that could
3421 * expose stale data in the case of a crash. Use the magic error code
3422 * to fallback to buffered I/O.
3424 if (!m_flags && !ret)
3427 ext4_journal_stop(handle);
3428 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3435 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3436 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3439 struct ext4_map_blocks map;
3440 u8 blkbits = inode->i_blkbits;
3442 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3445 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3449 * Calculate the first and last logical blocks respectively.
3451 map.m_lblk = offset >> blkbits;
3452 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3453 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3455 if (flags & IOMAP_WRITE) {
3457 * We check here if the blocks are already allocated, then we
3458 * don't need to start a journal txn and we can directly return
3459 * the mapping information. This could boost performance
3460 * especially in multi-threaded overwrite requests.
3462 if (offset + length <= i_size_read(inode)) {
3463 ret = ext4_map_blocks(NULL, inode, &map, 0);
3464 if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3467 ret = ext4_iomap_alloc(inode, &map, flags);
3469 ret = ext4_map_blocks(NULL, inode, &map, 0);
3475 ext4_set_iomap(inode, iomap, &map, offset, length);
3480 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3481 loff_t length, unsigned flags, struct iomap *iomap,
3482 struct iomap *srcmap)
3487 * Even for writes we don't need to allocate blocks, so just pretend
3488 * we are reading to save overhead of starting a transaction.
3490 flags &= ~IOMAP_WRITE;
3491 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3492 WARN_ON_ONCE(iomap->type != IOMAP_MAPPED);
3496 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3497 ssize_t written, unsigned flags, struct iomap *iomap)
3500 * Check to see whether an error occurred while writing out the data to
3501 * the allocated blocks. If so, return the magic error code so that we
3502 * fallback to buffered I/O and attempt to complete the remainder of
3503 * the I/O. Any blocks that may have been allocated in preparation for
3504 * the direct I/O will be reused during buffered I/O.
3506 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3512 const struct iomap_ops ext4_iomap_ops = {
3513 .iomap_begin = ext4_iomap_begin,
3514 .iomap_end = ext4_iomap_end,
3517 const struct iomap_ops ext4_iomap_overwrite_ops = {
3518 .iomap_begin = ext4_iomap_overwrite_begin,
3519 .iomap_end = ext4_iomap_end,
3522 static bool ext4_iomap_is_delalloc(struct inode *inode,
3523 struct ext4_map_blocks *map)
3525 struct extent_status es;
3526 ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3528 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3529 map->m_lblk, end, &es);
3531 if (!es.es_len || es.es_lblk > end)
3534 if (es.es_lblk > map->m_lblk) {
3535 map->m_len = es.es_lblk - map->m_lblk;
3539 offset = map->m_lblk - es.es_lblk;
3540 map->m_len = es.es_len - offset;
3545 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3546 loff_t length, unsigned int flags,
3547 struct iomap *iomap, struct iomap *srcmap)
3550 bool delalloc = false;
3551 struct ext4_map_blocks map;
3552 u8 blkbits = inode->i_blkbits;
3554 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3557 if (ext4_has_inline_data(inode)) {
3558 ret = ext4_inline_data_iomap(inode, iomap);
3559 if (ret != -EAGAIN) {
3560 if (ret == 0 && offset >= iomap->length)
3567 * Calculate the first and last logical block respectively.
3569 map.m_lblk = offset >> blkbits;
3570 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3571 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3574 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3575 * So handle it here itself instead of querying ext4_map_blocks().
3576 * Since ext4_map_blocks() will warn about it and will return
3579 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3580 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3582 if (offset >= sbi->s_bitmap_maxbytes) {
3588 ret = ext4_map_blocks(NULL, inode, &map, 0);
3592 delalloc = ext4_iomap_is_delalloc(inode, &map);
3595 ext4_set_iomap(inode, iomap, &map, offset, length);
3596 if (delalloc && iomap->type == IOMAP_HOLE)
3597 iomap->type = IOMAP_DELALLOC;
3602 const struct iomap_ops ext4_iomap_report_ops = {
3603 .iomap_begin = ext4_iomap_begin_report,
3607 * Pages can be marked dirty completely asynchronously from ext4's journalling
3608 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3609 * much here because ->set_page_dirty is called under VFS locks. The page is
3610 * not necessarily locked.
3612 * We cannot just dirty the page and leave attached buffers clean, because the
3613 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3614 * or jbddirty because all the journalling code will explode.
3616 * So what we do is to mark the page "pending dirty" and next time writepage
3617 * is called, propagate that into the buffers appropriately.
3619 static int ext4_journalled_set_page_dirty(struct page *page)
3621 SetPageChecked(page);
3622 return __set_page_dirty_nobuffers(page);
3625 static int ext4_set_page_dirty(struct page *page)
3627 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3628 WARN_ON_ONCE(!page_has_buffers(page));
3629 return __set_page_dirty_buffers(page);
3632 static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3633 struct file *file, sector_t *span)
3635 return iomap_swapfile_activate(sis, file, span,
3636 &ext4_iomap_report_ops);
3639 static const struct address_space_operations ext4_aops = {
3640 .readpage = ext4_readpage,
3641 .readahead = ext4_readahead,
3642 .writepage = ext4_writepage,
3643 .writepages = ext4_writepages,
3644 .write_begin = ext4_write_begin,
3645 .write_end = ext4_write_end,
3646 .set_page_dirty = ext4_set_page_dirty,
3648 .invalidatepage = ext4_invalidatepage,
3649 .releasepage = ext4_releasepage,
3650 .direct_IO = noop_direct_IO,
3651 .migratepage = buffer_migrate_page,
3652 .is_partially_uptodate = block_is_partially_uptodate,
3653 .error_remove_page = generic_error_remove_page,
3654 .swap_activate = ext4_iomap_swap_activate,
3657 static const struct address_space_operations ext4_journalled_aops = {
3658 .readpage = ext4_readpage,
3659 .readahead = ext4_readahead,
3660 .writepage = ext4_writepage,
3661 .writepages = ext4_writepages,
3662 .write_begin = ext4_write_begin,
3663 .write_end = ext4_journalled_write_end,
3664 .set_page_dirty = ext4_journalled_set_page_dirty,
3666 .invalidatepage = ext4_journalled_invalidatepage,
3667 .releasepage = ext4_releasepage,
3668 .direct_IO = noop_direct_IO,
3669 .is_partially_uptodate = block_is_partially_uptodate,
3670 .error_remove_page = generic_error_remove_page,
3671 .swap_activate = ext4_iomap_swap_activate,
3674 static const struct address_space_operations ext4_da_aops = {
3675 .readpage = ext4_readpage,
3676 .readahead = ext4_readahead,
3677 .writepage = ext4_writepage,
3678 .writepages = ext4_writepages,
3679 .write_begin = ext4_da_write_begin,
3680 .write_end = ext4_da_write_end,
3681 .set_page_dirty = ext4_set_page_dirty,
3683 .invalidatepage = ext4_invalidatepage,
3684 .releasepage = ext4_releasepage,
3685 .direct_IO = noop_direct_IO,
3686 .migratepage = buffer_migrate_page,
3687 .is_partially_uptodate = block_is_partially_uptodate,
3688 .error_remove_page = generic_error_remove_page,
3689 .swap_activate = ext4_iomap_swap_activate,
3692 static const struct address_space_operations ext4_dax_aops = {
3693 .writepages = ext4_dax_writepages,
3694 .direct_IO = noop_direct_IO,
3695 .set_page_dirty = noop_set_page_dirty,
3697 .invalidatepage = noop_invalidatepage,
3698 .swap_activate = ext4_iomap_swap_activate,
3701 void ext4_set_aops(struct inode *inode)
3703 switch (ext4_inode_journal_mode(inode)) {
3704 case EXT4_INODE_ORDERED_DATA_MODE:
3705 case EXT4_INODE_WRITEBACK_DATA_MODE:
3707 case EXT4_INODE_JOURNAL_DATA_MODE:
3708 inode->i_mapping->a_ops = &ext4_journalled_aops;
3714 inode->i_mapping->a_ops = &ext4_dax_aops;
3715 else if (test_opt(inode->i_sb, DELALLOC))
3716 inode->i_mapping->a_ops = &ext4_da_aops;
3718 inode->i_mapping->a_ops = &ext4_aops;
3721 static int __ext4_block_zero_page_range(handle_t *handle,
3722 struct address_space *mapping, loff_t from, loff_t length)
3724 ext4_fsblk_t index = from >> PAGE_SHIFT;
3725 unsigned offset = from & (PAGE_SIZE-1);
3726 unsigned blocksize, pos;
3728 struct inode *inode = mapping->host;
3729 struct buffer_head *bh;
3733 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3734 mapping_gfp_constraint(mapping, ~__GFP_FS));
3738 blocksize = inode->i_sb->s_blocksize;
3740 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3742 if (!page_has_buffers(page))
3743 create_empty_buffers(page, blocksize, 0);
3745 /* Find the buffer that contains "offset" */
3746 bh = page_buffers(page);
3748 while (offset >= pos) {
3749 bh = bh->b_this_page;
3753 if (buffer_freed(bh)) {
3754 BUFFER_TRACE(bh, "freed: skip");
3757 if (!buffer_mapped(bh)) {
3758 BUFFER_TRACE(bh, "unmapped");
3759 ext4_get_block(inode, iblock, bh, 0);
3760 /* unmapped? It's a hole - nothing to do */
3761 if (!buffer_mapped(bh)) {
3762 BUFFER_TRACE(bh, "still unmapped");
3767 /* Ok, it's mapped. Make sure it's up-to-date */
3768 if (PageUptodate(page))
3769 set_buffer_uptodate(bh);
3771 if (!buffer_uptodate(bh)) {
3772 err = ext4_read_bh_lock(bh, 0, true);
3775 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3776 /* We expect the key to be set. */
3777 BUG_ON(!fscrypt_has_encryption_key(inode));
3778 err = fscrypt_decrypt_pagecache_blocks(page, blocksize,
3781 clear_buffer_uptodate(bh);
3786 if (ext4_should_journal_data(inode)) {
3787 BUFFER_TRACE(bh, "get write access");
3788 err = ext4_journal_get_write_access(handle, bh);
3792 zero_user(page, offset, length);
3793 BUFFER_TRACE(bh, "zeroed end of block");
3795 if (ext4_should_journal_data(inode)) {
3796 err = ext4_handle_dirty_metadata(handle, inode, bh);
3799 mark_buffer_dirty(bh);
3800 if (ext4_should_order_data(inode))
3801 err = ext4_jbd2_inode_add_write(handle, inode, from,
3812 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3813 * starting from file offset 'from'. The range to be zero'd must
3814 * be contained with in one block. If the specified range exceeds
3815 * the end of the block it will be shortened to end of the block
3816 * that cooresponds to 'from'
3818 static int ext4_block_zero_page_range(handle_t *handle,
3819 struct address_space *mapping, loff_t from, loff_t length)
3821 struct inode *inode = mapping->host;
3822 unsigned offset = from & (PAGE_SIZE-1);
3823 unsigned blocksize = inode->i_sb->s_blocksize;
3824 unsigned max = blocksize - (offset & (blocksize - 1));
3827 * correct length if it does not fall between
3828 * 'from' and the end of the block
3830 if (length > max || length < 0)
3833 if (IS_DAX(inode)) {
3834 return iomap_zero_range(inode, from, length, NULL,
3837 return __ext4_block_zero_page_range(handle, mapping, from, length);
3841 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3842 * up to the end of the block which corresponds to `from'.
3843 * This required during truncate. We need to physically zero the tail end
3844 * of that block so it doesn't yield old data if the file is later grown.
3846 static int ext4_block_truncate_page(handle_t *handle,
3847 struct address_space *mapping, loff_t from)
3849 unsigned offset = from & (PAGE_SIZE-1);
3852 struct inode *inode = mapping->host;
3854 /* If we are processing an encrypted inode during orphan list handling */
3855 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3858 blocksize = inode->i_sb->s_blocksize;
3859 length = blocksize - (offset & (blocksize - 1));
3861 return ext4_block_zero_page_range(handle, mapping, from, length);
3864 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3865 loff_t lstart, loff_t length)
3867 struct super_block *sb = inode->i_sb;
3868 struct address_space *mapping = inode->i_mapping;
3869 unsigned partial_start, partial_end;
3870 ext4_fsblk_t start, end;
3871 loff_t byte_end = (lstart + length - 1);
3874 partial_start = lstart & (sb->s_blocksize - 1);
3875 partial_end = byte_end & (sb->s_blocksize - 1);
3877 start = lstart >> sb->s_blocksize_bits;
3878 end = byte_end >> sb->s_blocksize_bits;
3880 /* Handle partial zero within the single block */
3882 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3883 err = ext4_block_zero_page_range(handle, mapping,
3887 /* Handle partial zero out on the start of the range */
3888 if (partial_start) {
3889 err = ext4_block_zero_page_range(handle, mapping,
3890 lstart, sb->s_blocksize);
3894 /* Handle partial zero out on the end of the range */
3895 if (partial_end != sb->s_blocksize - 1)
3896 err = ext4_block_zero_page_range(handle, mapping,
3897 byte_end - partial_end,
3902 int ext4_can_truncate(struct inode *inode)
3904 if (S_ISREG(inode->i_mode))
3906 if (S_ISDIR(inode->i_mode))
3908 if (S_ISLNK(inode->i_mode))
3909 return !ext4_inode_is_fast_symlink(inode);
3914 * We have to make sure i_disksize gets properly updated before we truncate
3915 * page cache due to hole punching or zero range. Otherwise i_disksize update
3916 * can get lost as it may have been postponed to submission of writeback but
3917 * that will never happen after we truncate page cache.
3919 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3925 loff_t size = i_size_read(inode);
3927 WARN_ON(!inode_is_locked(inode));
3928 if (offset > size || offset + len < size)
3931 if (EXT4_I(inode)->i_disksize >= size)
3934 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3936 return PTR_ERR(handle);
3937 ext4_update_i_disksize(inode, size);
3938 ret = ext4_mark_inode_dirty(handle, inode);
3939 ext4_journal_stop(handle);
3944 static void ext4_wait_dax_page(struct ext4_inode_info *ei)
3946 up_write(&ei->i_mmap_sem);
3948 down_write(&ei->i_mmap_sem);
3951 int ext4_break_layouts(struct inode *inode)
3953 struct ext4_inode_info *ei = EXT4_I(inode);
3957 if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
3961 page = dax_layout_busy_page(inode->i_mapping);
3965 error = ___wait_var_event(&page->_refcount,
3966 atomic_read(&page->_refcount) == 1,
3967 TASK_INTERRUPTIBLE, 0, 0,
3968 ext4_wait_dax_page(ei));
3969 } while (error == 0);
3975 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3976 * associated with the given offset and length
3978 * @inode: File inode
3979 * @offset: The offset where the hole will begin
3980 * @len: The length of the hole
3982 * Returns: 0 on success or negative on failure
3985 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3987 struct super_block *sb = inode->i_sb;
3988 ext4_lblk_t first_block, stop_block;
3989 struct address_space *mapping = inode->i_mapping;
3990 loff_t first_block_offset, last_block_offset;
3992 unsigned int credits;
3993 int ret = 0, ret2 = 0;
3995 trace_ext4_punch_hole(inode, offset, length, 0);
3997 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
3998 if (ext4_has_inline_data(inode)) {
3999 down_write(&EXT4_I(inode)->i_mmap_sem);
4000 ret = ext4_convert_inline_data(inode);
4001 up_write(&EXT4_I(inode)->i_mmap_sem);
4007 * Write out all dirty pages to avoid race conditions
4008 * Then release them.
4010 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4011 ret = filemap_write_and_wait_range(mapping, offset,
4012 offset + length - 1);
4019 /* No need to punch hole beyond i_size */
4020 if (offset >= inode->i_size)
4024 * If the hole extends beyond i_size, set the hole
4025 * to end after the page that contains i_size
4027 if (offset + length > inode->i_size) {
4028 length = inode->i_size +
4029 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4033 if (offset & (sb->s_blocksize - 1) ||
4034 (offset + length) & (sb->s_blocksize - 1)) {
4036 * Attach jinode to inode for jbd2 if we do any zeroing of
4039 ret = ext4_inode_attach_jinode(inode);
4045 /* Wait all existing dio workers, newcomers will block on i_mutex */
4046 inode_dio_wait(inode);
4049 * Prevent page faults from reinstantiating pages we have released from
4052 down_write(&EXT4_I(inode)->i_mmap_sem);
4054 ret = ext4_break_layouts(inode);
4058 first_block_offset = round_up(offset, sb->s_blocksize);
4059 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4061 /* Now release the pages and zero block aligned part of pages*/
4062 if (last_block_offset > first_block_offset) {
4063 ret = ext4_update_disksize_before_punch(inode, offset, length);
4066 truncate_pagecache_range(inode, first_block_offset,
4070 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4071 credits = ext4_writepage_trans_blocks(inode);
4073 credits = ext4_blocks_for_truncate(inode);
4074 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4075 if (IS_ERR(handle)) {
4076 ret = PTR_ERR(handle);
4077 ext4_std_error(sb, ret);
4081 ret = ext4_zero_partial_blocks(handle, inode, offset,
4086 first_block = (offset + sb->s_blocksize - 1) >>
4087 EXT4_BLOCK_SIZE_BITS(sb);
4088 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4090 /* If there are blocks to remove, do it */
4091 if (stop_block > first_block) {
4093 down_write(&EXT4_I(inode)->i_data_sem);
4094 ext4_discard_preallocations(inode, 0);
4096 ret = ext4_es_remove_extent(inode, first_block,
4097 stop_block - first_block);
4099 up_write(&EXT4_I(inode)->i_data_sem);
4103 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4104 ret = ext4_ext_remove_space(inode, first_block,
4107 ret = ext4_ind_remove_space(handle, inode, first_block,
4110 up_write(&EXT4_I(inode)->i_data_sem);
4112 ext4_fc_track_range(inode, first_block, stop_block);
4114 ext4_handle_sync(handle);
4116 inode->i_mtime = inode->i_ctime = current_time(inode);
4117 ret2 = ext4_mark_inode_dirty(handle, inode);
4121 ext4_update_inode_fsync_trans(handle, inode, 1);
4123 ext4_journal_stop(handle);
4125 up_write(&EXT4_I(inode)->i_mmap_sem);
4127 inode_unlock(inode);
4131 int ext4_inode_attach_jinode(struct inode *inode)
4133 struct ext4_inode_info *ei = EXT4_I(inode);
4134 struct jbd2_inode *jinode;
4136 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4139 jinode = jbd2_alloc_inode(GFP_KERNEL);
4140 spin_lock(&inode->i_lock);
4143 spin_unlock(&inode->i_lock);
4146 ei->jinode = jinode;
4147 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4150 spin_unlock(&inode->i_lock);
4151 if (unlikely(jinode != NULL))
4152 jbd2_free_inode(jinode);
4159 * We block out ext4_get_block() block instantiations across the entire
4160 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4161 * simultaneously on behalf of the same inode.
4163 * As we work through the truncate and commit bits of it to the journal there
4164 * is one core, guiding principle: the file's tree must always be consistent on
4165 * disk. We must be able to restart the truncate after a crash.
4167 * The file's tree may be transiently inconsistent in memory (although it
4168 * probably isn't), but whenever we close off and commit a journal transaction,
4169 * the contents of (the filesystem + the journal) must be consistent and
4170 * restartable. It's pretty simple, really: bottom up, right to left (although
4171 * left-to-right works OK too).
4173 * Note that at recovery time, journal replay occurs *before* the restart of
4174 * truncate against the orphan inode list.
4176 * The committed inode has the new, desired i_size (which is the same as
4177 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4178 * that this inode's truncate did not complete and it will again call
4179 * ext4_truncate() to have another go. So there will be instantiated blocks
4180 * to the right of the truncation point in a crashed ext4 filesystem. But
4181 * that's fine - as long as they are linked from the inode, the post-crash
4182 * ext4_truncate() run will find them and release them.
4184 int ext4_truncate(struct inode *inode)
4186 struct ext4_inode_info *ei = EXT4_I(inode);
4187 unsigned int credits;
4190 struct address_space *mapping = inode->i_mapping;
4193 * There is a possibility that we're either freeing the inode
4194 * or it's a completely new inode. In those cases we might not
4195 * have i_mutex locked because it's not necessary.
4197 if (!(inode->i_state & (I_NEW|I_FREEING)))
4198 WARN_ON(!inode_is_locked(inode));
4199 trace_ext4_truncate_enter(inode);
4201 if (!ext4_can_truncate(inode))
4204 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4205 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4207 if (ext4_has_inline_data(inode)) {
4210 err = ext4_inline_data_truncate(inode, &has_inline);
4211 if (err || has_inline)
4215 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4216 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4217 if (ext4_inode_attach_jinode(inode) < 0)
4221 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4222 credits = ext4_writepage_trans_blocks(inode);
4224 credits = ext4_blocks_for_truncate(inode);
4226 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4227 if (IS_ERR(handle)) {
4228 err = PTR_ERR(handle);
4232 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4233 ext4_block_truncate_page(handle, mapping, inode->i_size);
4236 * We add the inode to the orphan list, so that if this
4237 * truncate spans multiple transactions, and we crash, we will
4238 * resume the truncate when the filesystem recovers. It also
4239 * marks the inode dirty, to catch the new size.
4241 * Implication: the file must always be in a sane, consistent
4242 * truncatable state while each transaction commits.
4244 err = ext4_orphan_add(handle, inode);
4248 down_write(&EXT4_I(inode)->i_data_sem);
4250 ext4_discard_preallocations(inode, 0);
4252 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4253 err = ext4_ext_truncate(handle, inode);
4255 ext4_ind_truncate(handle, inode);
4257 up_write(&ei->i_data_sem);
4262 ext4_handle_sync(handle);
4266 * If this was a simple ftruncate() and the file will remain alive,
4267 * then we need to clear up the orphan record which we created above.
4268 * However, if this was a real unlink then we were called by
4269 * ext4_evict_inode(), and we allow that function to clean up the
4270 * orphan info for us.
4273 ext4_orphan_del(handle, inode);
4275 inode->i_mtime = inode->i_ctime = current_time(inode);
4276 err2 = ext4_mark_inode_dirty(handle, inode);
4277 if (unlikely(err2 && !err))
4279 ext4_journal_stop(handle);
4282 trace_ext4_truncate_exit(inode);
4287 * ext4_get_inode_loc returns with an extra refcount against the inode's
4288 * underlying buffer_head on success. If 'in_mem' is true, we have all
4289 * data in memory that is needed to recreate the on-disk version of this
4292 static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4293 struct ext4_iloc *iloc, int in_mem,
4294 ext4_fsblk_t *ret_block)
4296 struct ext4_group_desc *gdp;
4297 struct buffer_head *bh;
4299 struct blk_plug plug;
4300 int inodes_per_block, inode_offset;
4303 if (ino < EXT4_ROOT_INO ||
4304 ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4305 return -EFSCORRUPTED;
4307 iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4308 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4313 * Figure out the offset within the block group inode table
4315 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4316 inode_offset = ((ino - 1) %
4317 EXT4_INODES_PER_GROUP(sb));
4318 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4319 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4321 bh = sb_getblk(sb, block);
4324 if (ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO))
4326 if (!buffer_uptodate(bh)) {
4329 if (ext4_buffer_uptodate(bh)) {
4330 /* someone brought it uptodate while we waited */
4336 * If we have all information of the inode in memory and this
4337 * is the only valid inode in the block, we need not read the
4341 struct buffer_head *bitmap_bh;
4344 start = inode_offset & ~(inodes_per_block - 1);
4346 /* Is the inode bitmap in cache? */
4347 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4348 if (unlikely(!bitmap_bh))
4352 * If the inode bitmap isn't in cache then the
4353 * optimisation may end up performing two reads instead
4354 * of one, so skip it.
4356 if (!buffer_uptodate(bitmap_bh)) {
4360 for (i = start; i < start + inodes_per_block; i++) {
4361 if (i == inode_offset)
4363 if (ext4_test_bit(i, bitmap_bh->b_data))
4367 if (i == start + inodes_per_block) {
4368 /* all other inodes are free, so skip I/O */
4369 memset(bh->b_data, 0, bh->b_size);
4370 set_buffer_uptodate(bh);
4378 * If we need to do any I/O, try to pre-readahead extra
4379 * blocks from the inode table.
4381 blk_start_plug(&plug);
4382 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4383 ext4_fsblk_t b, end, table;
4385 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4387 table = ext4_inode_table(sb, gdp);
4388 /* s_inode_readahead_blks is always a power of 2 */
4389 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4393 num = EXT4_INODES_PER_GROUP(sb);
4394 if (ext4_has_group_desc_csum(sb))
4395 num -= ext4_itable_unused_count(sb, gdp);
4396 table += num / inodes_per_block;
4400 ext4_sb_breadahead_unmovable(sb, b++);
4404 * There are other valid inodes in the buffer, this inode
4405 * has in-inode xattrs, or we don't have this inode in memory.
4406 * Read the block from disk.
4408 trace_ext4_load_inode(sb, ino);
4409 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4410 blk_finish_plug(&plug);
4412 if (!buffer_uptodate(bh)) {
4425 static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4426 struct ext4_iloc *iloc)
4428 ext4_fsblk_t err_blk;
4431 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc, 0,
4435 ext4_error_inode_block(inode, err_blk, EIO,
4436 "unable to read itable block");
4441 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4443 ext4_fsblk_t err_blk;
4446 /* We have all inode data except xattrs in memory here. */
4447 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc,
4448 !ext4_test_inode_state(inode, EXT4_STATE_XATTR), &err_blk);
4451 ext4_error_inode_block(inode, err_blk, EIO,
4452 "unable to read itable block");
4458 int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4459 struct ext4_iloc *iloc)
4461 return __ext4_get_inode_loc(sb, ino, iloc, 0, NULL);
4464 static bool ext4_should_enable_dax(struct inode *inode)
4466 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4468 if (test_opt2(inode->i_sb, DAX_NEVER))
4470 if (!S_ISREG(inode->i_mode))
4472 if (ext4_should_journal_data(inode))
4474 if (ext4_has_inline_data(inode))
4476 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4478 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4480 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4482 if (test_opt(inode->i_sb, DAX_ALWAYS))
4485 return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4488 void ext4_set_inode_flags(struct inode *inode, bool init)
4490 unsigned int flags = EXT4_I(inode)->i_flags;
4491 unsigned int new_fl = 0;
4493 WARN_ON_ONCE(IS_DAX(inode) && init);
4495 if (flags & EXT4_SYNC_FL)
4497 if (flags & EXT4_APPEND_FL)
4499 if (flags & EXT4_IMMUTABLE_FL)
4500 new_fl |= S_IMMUTABLE;
4501 if (flags & EXT4_NOATIME_FL)
4502 new_fl |= S_NOATIME;
4503 if (flags & EXT4_DIRSYNC_FL)
4504 new_fl |= S_DIRSYNC;
4506 /* Because of the way inode_set_flags() works we must preserve S_DAX
4507 * here if already set. */
4508 new_fl |= (inode->i_flags & S_DAX);
4509 if (init && ext4_should_enable_dax(inode))
4512 if (flags & EXT4_ENCRYPT_FL)
4513 new_fl |= S_ENCRYPTED;
4514 if (flags & EXT4_CASEFOLD_FL)
4515 new_fl |= S_CASEFOLD;
4516 if (flags & EXT4_VERITY_FL)
4518 inode_set_flags(inode, new_fl,
4519 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4520 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4523 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4524 struct ext4_inode_info *ei)
4527 struct inode *inode = &(ei->vfs_inode);
4528 struct super_block *sb = inode->i_sb;
4530 if (ext4_has_feature_huge_file(sb)) {
4531 /* we are using combined 48 bit field */
4532 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4533 le32_to_cpu(raw_inode->i_blocks_lo);
4534 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4535 /* i_blocks represent file system block size */
4536 return i_blocks << (inode->i_blkbits - 9);
4541 return le32_to_cpu(raw_inode->i_blocks_lo);
4545 static inline int ext4_iget_extra_inode(struct inode *inode,
4546 struct ext4_inode *raw_inode,
4547 struct ext4_inode_info *ei)
4549 __le32 *magic = (void *)raw_inode +
4550 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4552 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4553 EXT4_INODE_SIZE(inode->i_sb) &&
4554 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4555 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4556 return ext4_find_inline_data_nolock(inode);
4558 EXT4_I(inode)->i_inline_off = 0;
4562 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4564 if (!ext4_has_feature_project(inode->i_sb))
4566 *projid = EXT4_I(inode)->i_projid;
4571 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4572 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4575 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4577 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4578 inode_set_iversion_raw(inode, val);
4580 inode_set_iversion_queried(inode, val);
4582 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4584 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4585 return inode_peek_iversion_raw(inode);
4587 return inode_peek_iversion(inode);
4590 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4591 ext4_iget_flags flags, const char *function,
4594 struct ext4_iloc iloc;
4595 struct ext4_inode *raw_inode;
4596 struct ext4_inode_info *ei;
4597 struct inode *inode;
4598 journal_t *journal = EXT4_SB(sb)->s_journal;
4606 if ((!(flags & EXT4_IGET_SPECIAL) &&
4607 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4608 (ino < EXT4_ROOT_INO) ||
4609 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4610 if (flags & EXT4_IGET_HANDLE)
4611 return ERR_PTR(-ESTALE);
4612 __ext4_error(sb, function, line, EFSCORRUPTED, 0,
4613 "inode #%lu: comm %s: iget: illegal inode #",
4614 ino, current->comm);
4615 return ERR_PTR(-EFSCORRUPTED);
4618 inode = iget_locked(sb, ino);
4620 return ERR_PTR(-ENOMEM);
4621 if (!(inode->i_state & I_NEW))
4627 ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4630 raw_inode = ext4_raw_inode(&iloc);
4632 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4633 ext4_error_inode(inode, function, line, 0,
4634 "iget: root inode unallocated");
4635 ret = -EFSCORRUPTED;
4639 if ((flags & EXT4_IGET_HANDLE) &&
4640 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4645 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4646 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4647 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4648 EXT4_INODE_SIZE(inode->i_sb) ||
4649 (ei->i_extra_isize & 3)) {
4650 ext4_error_inode(inode, function, line, 0,
4651 "iget: bad extra_isize %u "
4654 EXT4_INODE_SIZE(inode->i_sb));
4655 ret = -EFSCORRUPTED;
4659 ei->i_extra_isize = 0;
4661 /* Precompute checksum seed for inode metadata */
4662 if (ext4_has_metadata_csum(sb)) {
4663 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4665 __le32 inum = cpu_to_le32(inode->i_ino);
4666 __le32 gen = raw_inode->i_generation;
4667 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4669 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4673 if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4674 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4675 (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4676 ext4_error_inode_err(inode, function, line, 0,
4677 EFSBADCRC, "iget: checksum invalid");
4682 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4683 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4684 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4685 if (ext4_has_feature_project(sb) &&
4686 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4687 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4688 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4690 i_projid = EXT4_DEF_PROJID;
4692 if (!(test_opt(inode->i_sb, NO_UID32))) {
4693 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4694 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4696 i_uid_write(inode, i_uid);
4697 i_gid_write(inode, i_gid);
4698 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4699 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4701 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4702 ei->i_inline_off = 0;
4703 ei->i_dir_start_lookup = 0;
4704 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4705 /* We now have enough fields to check if the inode was active or not.
4706 * This is needed because nfsd might try to access dead inodes
4707 * the test is that same one that e2fsck uses
4708 * NeilBrown 1999oct15
4710 if (inode->i_nlink == 0) {
4711 if ((inode->i_mode == 0 ||
4712 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4713 ino != EXT4_BOOT_LOADER_INO) {
4714 /* this inode is deleted */
4718 /* The only unlinked inodes we let through here have
4719 * valid i_mode and are being read by the orphan
4720 * recovery code: that's fine, we're about to complete
4721 * the process of deleting those.
4722 * OR it is the EXT4_BOOT_LOADER_INO which is
4723 * not initialized on a new filesystem. */
4725 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4726 ext4_set_inode_flags(inode, true);
4727 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4728 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4729 if (ext4_has_feature_64bit(sb))
4731 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4732 inode->i_size = ext4_isize(sb, raw_inode);
4733 if ((size = i_size_read(inode)) < 0) {
4734 ext4_error_inode(inode, function, line, 0,
4735 "iget: bad i_size value: %lld", size);
4736 ret = -EFSCORRUPTED;
4740 * If dir_index is not enabled but there's dir with INDEX flag set,
4741 * we'd normally treat htree data as empty space. But with metadata
4742 * checksumming that corrupts checksums so forbid that.
4744 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4745 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4746 ext4_error_inode(inode, function, line, 0,
4747 "iget: Dir with htree data on filesystem without dir_index feature.");
4748 ret = -EFSCORRUPTED;
4751 ei->i_disksize = inode->i_size;
4753 ei->i_reserved_quota = 0;
4755 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4756 ei->i_block_group = iloc.block_group;
4757 ei->i_last_alloc_group = ~0;
4759 * NOTE! The in-memory inode i_data array is in little-endian order
4760 * even on big-endian machines: we do NOT byteswap the block numbers!
4762 for (block = 0; block < EXT4_N_BLOCKS; block++)
4763 ei->i_data[block] = raw_inode->i_block[block];
4764 INIT_LIST_HEAD(&ei->i_orphan);
4765 ext4_fc_init_inode(&ei->vfs_inode);
4768 * Set transaction id's of transactions that have to be committed
4769 * to finish f[data]sync. We set them to currently running transaction
4770 * as we cannot be sure that the inode or some of its metadata isn't
4771 * part of the transaction - the inode could have been reclaimed and
4772 * now it is reread from disk.
4775 transaction_t *transaction;
4778 read_lock(&journal->j_state_lock);
4779 if (journal->j_running_transaction)
4780 transaction = journal->j_running_transaction;
4782 transaction = journal->j_committing_transaction;
4784 tid = transaction->t_tid;
4786 tid = journal->j_commit_sequence;
4787 read_unlock(&journal->j_state_lock);
4788 ei->i_sync_tid = tid;
4789 ei->i_datasync_tid = tid;
4792 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4793 if (ei->i_extra_isize == 0) {
4794 /* The extra space is currently unused. Use it. */
4795 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4796 ei->i_extra_isize = sizeof(struct ext4_inode) -
4797 EXT4_GOOD_OLD_INODE_SIZE;
4799 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4805 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4806 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4807 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4808 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4810 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4811 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4813 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4814 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4816 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4818 ext4_inode_set_iversion_queried(inode, ivers);
4822 if (ei->i_file_acl &&
4823 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4824 ext4_error_inode(inode, function, line, 0,
4825 "iget: bad extended attribute block %llu",
4827 ret = -EFSCORRUPTED;
4829 } else if (!ext4_has_inline_data(inode)) {
4830 /* validate the block references in the inode */
4831 if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4832 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4833 (S_ISLNK(inode->i_mode) &&
4834 !ext4_inode_is_fast_symlink(inode)))) {
4835 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4836 ret = ext4_ext_check_inode(inode);
4838 ret = ext4_ind_check_inode(inode);
4844 if (S_ISREG(inode->i_mode)) {
4845 inode->i_op = &ext4_file_inode_operations;
4846 inode->i_fop = &ext4_file_operations;
4847 ext4_set_aops(inode);
4848 } else if (S_ISDIR(inode->i_mode)) {
4849 inode->i_op = &ext4_dir_inode_operations;
4850 inode->i_fop = &ext4_dir_operations;
4851 } else if (S_ISLNK(inode->i_mode)) {
4852 /* VFS does not allow setting these so must be corruption */
4853 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4854 ext4_error_inode(inode, function, line, 0,
4855 "iget: immutable or append flags "
4856 "not allowed on symlinks");
4857 ret = -EFSCORRUPTED;
4860 if (IS_ENCRYPTED(inode)) {
4861 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4862 ext4_set_aops(inode);
4863 } else if (ext4_inode_is_fast_symlink(inode)) {
4864 inode->i_link = (char *)ei->i_data;
4865 inode->i_op = &ext4_fast_symlink_inode_operations;
4866 nd_terminate_link(ei->i_data, inode->i_size,
4867 sizeof(ei->i_data) - 1);
4869 inode->i_op = &ext4_symlink_inode_operations;
4870 ext4_set_aops(inode);
4872 inode_nohighmem(inode);
4873 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4874 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4875 inode->i_op = &ext4_special_inode_operations;
4876 if (raw_inode->i_block[0])
4877 init_special_inode(inode, inode->i_mode,
4878 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4880 init_special_inode(inode, inode->i_mode,
4881 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4882 } else if (ino == EXT4_BOOT_LOADER_INO) {
4883 make_bad_inode(inode);
4885 ret = -EFSCORRUPTED;
4886 ext4_error_inode(inode, function, line, 0,
4887 "iget: bogus i_mode (%o)", inode->i_mode);
4890 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
4891 ext4_error_inode(inode, function, line, 0,
4892 "casefold flag without casefold feature");
4895 unlock_new_inode(inode);
4901 return ERR_PTR(ret);
4904 static int ext4_inode_blocks_set(handle_t *handle,
4905 struct ext4_inode *raw_inode,
4906 struct ext4_inode_info *ei)
4908 struct inode *inode = &(ei->vfs_inode);
4909 u64 i_blocks = READ_ONCE(inode->i_blocks);
4910 struct super_block *sb = inode->i_sb;
4912 if (i_blocks <= ~0U) {
4914 * i_blocks can be represented in a 32 bit variable
4915 * as multiple of 512 bytes
4917 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4918 raw_inode->i_blocks_high = 0;
4919 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4922 if (!ext4_has_feature_huge_file(sb))
4925 if (i_blocks <= 0xffffffffffffULL) {
4927 * i_blocks can be represented in a 48 bit variable
4928 * as multiple of 512 bytes
4930 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4931 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4932 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4934 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4935 /* i_block is stored in file system block size */
4936 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4937 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4938 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4943 static void __ext4_update_other_inode_time(struct super_block *sb,
4944 unsigned long orig_ino,
4946 struct ext4_inode *raw_inode)
4948 struct inode *inode;
4950 inode = find_inode_by_ino_rcu(sb, ino);
4954 if ((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4956 ((inode->i_state & I_DIRTY_TIME) == 0))
4959 spin_lock(&inode->i_lock);
4960 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4961 I_DIRTY_INODE)) == 0) &&
4962 (inode->i_state & I_DIRTY_TIME)) {
4963 struct ext4_inode_info *ei = EXT4_I(inode);
4965 inode->i_state &= ~I_DIRTY_TIME;
4966 spin_unlock(&inode->i_lock);
4968 spin_lock(&ei->i_raw_lock);
4969 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4970 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4971 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4972 ext4_inode_csum_set(inode, raw_inode, ei);
4973 spin_unlock(&ei->i_raw_lock);
4974 trace_ext4_other_inode_update_time(inode, orig_ino);
4977 spin_unlock(&inode->i_lock);
4981 * Opportunistically update the other time fields for other inodes in
4982 * the same inode table block.
4984 static void ext4_update_other_inodes_time(struct super_block *sb,
4985 unsigned long orig_ino, char *buf)
4988 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4989 int inode_size = EXT4_INODE_SIZE(sb);
4992 * Calculate the first inode in the inode table block. Inode
4993 * numbers are one-based. That is, the first inode in a block
4994 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4996 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4998 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4999 if (ino == orig_ino)
5001 __ext4_update_other_inode_time(sb, orig_ino, ino,
5002 (struct ext4_inode *)buf);
5008 * Post the struct inode info into an on-disk inode location in the
5009 * buffer-cache. This gobbles the caller's reference to the
5010 * buffer_head in the inode location struct.
5012 * The caller must have write access to iloc->bh.
5014 static int ext4_do_update_inode(handle_t *handle,
5015 struct inode *inode,
5016 struct ext4_iloc *iloc)
5018 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5019 struct ext4_inode_info *ei = EXT4_I(inode);
5020 struct buffer_head *bh = iloc->bh;
5021 struct super_block *sb = inode->i_sb;
5022 int err = 0, rc, block;
5023 int need_datasync = 0, set_large_file = 0;
5028 spin_lock(&ei->i_raw_lock);
5030 /* For fields not tracked in the in-memory inode,
5031 * initialise them to zero for new inodes. */
5032 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5033 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5035 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5037 spin_unlock(&ei->i_raw_lock);
5041 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5042 i_uid = i_uid_read(inode);
5043 i_gid = i_gid_read(inode);
5044 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5045 if (!(test_opt(inode->i_sb, NO_UID32))) {
5046 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5047 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5049 * Fix up interoperability with old kernels. Otherwise, old inodes get
5050 * re-used with the upper 16 bits of the uid/gid intact
5052 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5053 raw_inode->i_uid_high = 0;
5054 raw_inode->i_gid_high = 0;
5056 raw_inode->i_uid_high =
5057 cpu_to_le16(high_16_bits(i_uid));
5058 raw_inode->i_gid_high =
5059 cpu_to_le16(high_16_bits(i_gid));
5062 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5063 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5064 raw_inode->i_uid_high = 0;
5065 raw_inode->i_gid_high = 0;
5067 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5069 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5070 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5071 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5072 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5074 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5075 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5076 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5077 raw_inode->i_file_acl_high =
5078 cpu_to_le16(ei->i_file_acl >> 32);
5079 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5080 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) {
5081 ext4_isize_set(raw_inode, ei->i_disksize);
5084 if (ei->i_disksize > 0x7fffffffULL) {
5085 if (!ext4_has_feature_large_file(sb) ||
5086 EXT4_SB(sb)->s_es->s_rev_level ==
5087 cpu_to_le32(EXT4_GOOD_OLD_REV))
5090 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5091 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5092 if (old_valid_dev(inode->i_rdev)) {
5093 raw_inode->i_block[0] =
5094 cpu_to_le32(old_encode_dev(inode->i_rdev));
5095 raw_inode->i_block[1] = 0;
5097 raw_inode->i_block[0] = 0;
5098 raw_inode->i_block[1] =
5099 cpu_to_le32(new_encode_dev(inode->i_rdev));
5100 raw_inode->i_block[2] = 0;
5102 } else if (!ext4_has_inline_data(inode)) {
5103 for (block = 0; block < EXT4_N_BLOCKS; block++)
5104 raw_inode->i_block[block] = ei->i_data[block];
5107 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5108 u64 ivers = ext4_inode_peek_iversion(inode);
5110 raw_inode->i_disk_version = cpu_to_le32(ivers);
5111 if (ei->i_extra_isize) {
5112 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5113 raw_inode->i_version_hi =
5114 cpu_to_le32(ivers >> 32);
5115 raw_inode->i_extra_isize =
5116 cpu_to_le16(ei->i_extra_isize);
5120 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5121 i_projid != EXT4_DEF_PROJID);
5123 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5124 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5125 raw_inode->i_projid = cpu_to_le32(i_projid);
5127 ext4_inode_csum_set(inode, raw_inode, ei);
5128 spin_unlock(&ei->i_raw_lock);
5129 if (inode->i_sb->s_flags & SB_LAZYTIME)
5130 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5133 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5134 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5137 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5138 if (set_large_file) {
5139 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5140 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5143 ext4_set_feature_large_file(sb);
5144 ext4_handle_sync(handle);
5145 err = ext4_handle_dirty_super(handle, sb);
5147 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5150 ext4_std_error(inode->i_sb, err);
5155 * ext4_write_inode()
5157 * We are called from a few places:
5159 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5160 * Here, there will be no transaction running. We wait for any running
5161 * transaction to commit.
5163 * - Within flush work (sys_sync(), kupdate and such).
5164 * We wait on commit, if told to.
5166 * - Within iput_final() -> write_inode_now()
5167 * We wait on commit, if told to.
5169 * In all cases it is actually safe for us to return without doing anything,
5170 * because the inode has been copied into a raw inode buffer in
5171 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5174 * Note that we are absolutely dependent upon all inode dirtiers doing the
5175 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5176 * which we are interested.
5178 * It would be a bug for them to not do this. The code:
5180 * mark_inode_dirty(inode)
5182 * inode->i_size = expr;
5184 * is in error because write_inode() could occur while `stuff()' is running,
5185 * and the new i_size will be lost. Plus the inode will no longer be on the
5186 * superblock's dirty inode list.
5188 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5192 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5193 sb_rdonly(inode->i_sb))
5196 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5199 if (EXT4_SB(inode->i_sb)->s_journal) {
5200 if (ext4_journal_current_handle()) {
5201 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5207 * No need to force transaction in WB_SYNC_NONE mode. Also
5208 * ext4_sync_fs() will force the commit after everything is
5211 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5214 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5215 EXT4_I(inode)->i_sync_tid);
5217 struct ext4_iloc iloc;
5219 err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5223 * sync(2) will flush the whole buffer cache. No need to do
5224 * it here separately for each inode.
5226 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5227 sync_dirty_buffer(iloc.bh);
5228 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5229 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5230 "IO error syncing inode");
5239 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5240 * buffers that are attached to a page stradding i_size and are undergoing
5241 * commit. In that case we have to wait for commit to finish and try again.
5243 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5247 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5248 tid_t commit_tid = 0;
5251 offset = inode->i_size & (PAGE_SIZE - 1);
5253 * If the page is fully truncated, we don't need to wait for any commit
5254 * (and we even should not as __ext4_journalled_invalidatepage() may
5255 * strip all buffers from the page but keep the page dirty which can then
5256 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5257 * buffers). Also we don't need to wait for any commit if all buffers in
5258 * the page remain valid. This is most beneficial for the common case of
5259 * blocksize == PAGESIZE.
5261 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5264 page = find_lock_page(inode->i_mapping,
5265 inode->i_size >> PAGE_SHIFT);
5268 ret = __ext4_journalled_invalidatepage(page, offset,
5269 PAGE_SIZE - offset);
5275 read_lock(&journal->j_state_lock);
5276 if (journal->j_committing_transaction)
5277 commit_tid = journal->j_committing_transaction->t_tid;
5278 read_unlock(&journal->j_state_lock);
5280 jbd2_log_wait_commit(journal, commit_tid);
5287 * Called from notify_change.
5289 * We want to trap VFS attempts to truncate the file as soon as
5290 * possible. In particular, we want to make sure that when the VFS
5291 * shrinks i_size, we put the inode on the orphan list and modify
5292 * i_disksize immediately, so that during the subsequent flushing of
5293 * dirty pages and freeing of disk blocks, we can guarantee that any
5294 * commit will leave the blocks being flushed in an unused state on
5295 * disk. (On recovery, the inode will get truncated and the blocks will
5296 * be freed, so we have a strong guarantee that no future commit will
5297 * leave these blocks visible to the user.)
5299 * Another thing we have to assure is that if we are in ordered mode
5300 * and inode is still attached to the committing transaction, we must
5301 * we start writeout of all the dirty pages which are being truncated.
5302 * This way we are sure that all the data written in the previous
5303 * transaction are already on disk (truncate waits for pages under
5306 * Called with inode->i_mutex down.
5308 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5310 struct inode *inode = d_inode(dentry);
5313 const unsigned int ia_valid = attr->ia_valid;
5315 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5318 if (unlikely(IS_IMMUTABLE(inode)))
5321 if (unlikely(IS_APPEND(inode) &&
5322 (ia_valid & (ATTR_MODE | ATTR_UID |
5323 ATTR_GID | ATTR_TIMES_SET))))
5326 error = setattr_prepare(dentry, attr);
5330 error = fscrypt_prepare_setattr(dentry, attr);
5334 error = fsverity_prepare_setattr(dentry, attr);
5338 if (is_quota_modification(inode, attr)) {
5339 error = dquot_initialize(inode);
5343 ext4_fc_start_update(inode);
5344 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5345 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5348 /* (user+group)*(old+new) structure, inode write (sb,
5349 * inode block, ? - but truncate inode update has it) */
5350 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5351 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5352 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5353 if (IS_ERR(handle)) {
5354 error = PTR_ERR(handle);
5358 /* dquot_transfer() calls back ext4_get_inode_usage() which
5359 * counts xattr inode references.
5361 down_read(&EXT4_I(inode)->xattr_sem);
5362 error = dquot_transfer(inode, attr);
5363 up_read(&EXT4_I(inode)->xattr_sem);
5366 ext4_journal_stop(handle);
5367 ext4_fc_stop_update(inode);
5370 /* Update corresponding info in inode so that everything is in
5371 * one transaction */
5372 if (attr->ia_valid & ATTR_UID)
5373 inode->i_uid = attr->ia_uid;
5374 if (attr->ia_valid & ATTR_GID)
5375 inode->i_gid = attr->ia_gid;
5376 error = ext4_mark_inode_dirty(handle, inode);
5377 ext4_journal_stop(handle);
5378 if (unlikely(error))
5382 if (attr->ia_valid & ATTR_SIZE) {
5384 loff_t oldsize = inode->i_size;
5385 int shrink = (attr->ia_size < inode->i_size);
5387 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5388 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5390 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5391 ext4_fc_stop_update(inode);
5395 if (!S_ISREG(inode->i_mode)) {
5396 ext4_fc_stop_update(inode);
5400 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5401 inode_inc_iversion(inode);
5404 if (ext4_should_order_data(inode)) {
5405 error = ext4_begin_ordered_truncate(inode,
5411 * Blocks are going to be removed from the inode. Wait
5412 * for dio in flight.
5414 inode_dio_wait(inode);
5417 down_write(&EXT4_I(inode)->i_mmap_sem);
5419 rc = ext4_break_layouts(inode);
5421 up_write(&EXT4_I(inode)->i_mmap_sem);
5425 if (attr->ia_size != inode->i_size) {
5426 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5427 if (IS_ERR(handle)) {
5428 error = PTR_ERR(handle);
5431 if (ext4_handle_valid(handle) && shrink) {
5432 error = ext4_orphan_add(handle, inode);
5436 * Update c/mtime on truncate up, ext4_truncate() will
5437 * update c/mtime in shrink case below
5440 inode->i_mtime = current_time(inode);
5441 inode->i_ctime = inode->i_mtime;
5445 ext4_fc_track_range(inode,
5446 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5447 inode->i_sb->s_blocksize_bits,
5448 (oldsize > 0 ? oldsize - 1 : 0) >>
5449 inode->i_sb->s_blocksize_bits);
5451 ext4_fc_track_range(
5453 (oldsize > 0 ? oldsize - 1 : oldsize) >>
5454 inode->i_sb->s_blocksize_bits,
5455 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5456 inode->i_sb->s_blocksize_bits);
5458 down_write(&EXT4_I(inode)->i_data_sem);
5459 EXT4_I(inode)->i_disksize = attr->ia_size;
5460 rc = ext4_mark_inode_dirty(handle, inode);
5464 * We have to update i_size under i_data_sem together
5465 * with i_disksize to avoid races with writeback code
5466 * running ext4_wb_update_i_disksize().
5469 i_size_write(inode, attr->ia_size);
5470 up_write(&EXT4_I(inode)->i_data_sem);
5471 ext4_journal_stop(handle);
5475 pagecache_isize_extended(inode, oldsize,
5477 } else if (ext4_should_journal_data(inode)) {
5478 ext4_wait_for_tail_page_commit(inode);
5483 * Truncate pagecache after we've waited for commit
5484 * in data=journal mode to make pages freeable.
5486 truncate_pagecache(inode, inode->i_size);
5488 * Call ext4_truncate() even if i_size didn't change to
5489 * truncate possible preallocated blocks.
5491 if (attr->ia_size <= oldsize) {
5492 rc = ext4_truncate(inode);
5497 up_write(&EXT4_I(inode)->i_mmap_sem);
5501 setattr_copy(inode, attr);
5502 mark_inode_dirty(inode);
5506 * If the call to ext4_truncate failed to get a transaction handle at
5507 * all, we need to clean up the in-core orphan list manually.
5509 if (orphan && inode->i_nlink)
5510 ext4_orphan_del(NULL, inode);
5512 if (!error && (ia_valid & ATTR_MODE))
5513 rc = posix_acl_chmod(inode, inode->i_mode);
5517 ext4_std_error(inode->i_sb, error);
5520 ext4_fc_stop_update(inode);
5524 int ext4_getattr(const struct path *path, struct kstat *stat,
5525 u32 request_mask, unsigned int query_flags)
5527 struct inode *inode = d_inode(path->dentry);
5528 struct ext4_inode *raw_inode;
5529 struct ext4_inode_info *ei = EXT4_I(inode);
5532 if ((request_mask & STATX_BTIME) &&
5533 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5534 stat->result_mask |= STATX_BTIME;
5535 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5536 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5539 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5540 if (flags & EXT4_APPEND_FL)
5541 stat->attributes |= STATX_ATTR_APPEND;
5542 if (flags & EXT4_COMPR_FL)
5543 stat->attributes |= STATX_ATTR_COMPRESSED;
5544 if (flags & EXT4_ENCRYPT_FL)
5545 stat->attributes |= STATX_ATTR_ENCRYPTED;
5546 if (flags & EXT4_IMMUTABLE_FL)
5547 stat->attributes |= STATX_ATTR_IMMUTABLE;
5548 if (flags & EXT4_NODUMP_FL)
5549 stat->attributes |= STATX_ATTR_NODUMP;
5550 if (flags & EXT4_VERITY_FL)
5551 stat->attributes |= STATX_ATTR_VERITY;
5553 stat->attributes_mask |= (STATX_ATTR_APPEND |
5554 STATX_ATTR_COMPRESSED |
5555 STATX_ATTR_ENCRYPTED |
5556 STATX_ATTR_IMMUTABLE |
5560 generic_fillattr(inode, stat);
5564 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5565 u32 request_mask, unsigned int query_flags)
5567 struct inode *inode = d_inode(path->dentry);
5568 u64 delalloc_blocks;
5570 ext4_getattr(path, stat, request_mask, query_flags);
5573 * If there is inline data in the inode, the inode will normally not
5574 * have data blocks allocated (it may have an external xattr block).
5575 * Report at least one sector for such files, so tools like tar, rsync,
5576 * others don't incorrectly think the file is completely sparse.
5578 if (unlikely(ext4_has_inline_data(inode)))
5579 stat->blocks += (stat->size + 511) >> 9;
5582 * We can't update i_blocks if the block allocation is delayed
5583 * otherwise in the case of system crash before the real block
5584 * allocation is done, we will have i_blocks inconsistent with
5585 * on-disk file blocks.
5586 * We always keep i_blocks updated together with real
5587 * allocation. But to not confuse with user, stat
5588 * will return the blocks that include the delayed allocation
5589 * blocks for this file.
5591 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5592 EXT4_I(inode)->i_reserved_data_blocks);
5593 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5597 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5600 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5601 return ext4_ind_trans_blocks(inode, lblocks);
5602 return ext4_ext_index_trans_blocks(inode, pextents);
5606 * Account for index blocks, block groups bitmaps and block group
5607 * descriptor blocks if modify datablocks and index blocks
5608 * worse case, the indexs blocks spread over different block groups
5610 * If datablocks are discontiguous, they are possible to spread over
5611 * different block groups too. If they are contiguous, with flexbg,
5612 * they could still across block group boundary.
5614 * Also account for superblock, inode, quota and xattr blocks
5616 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5619 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5625 * How many index blocks need to touch to map @lblocks logical blocks
5626 * to @pextents physical extents?
5628 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5633 * Now let's see how many group bitmaps and group descriptors need
5636 groups = idxblocks + pextents;
5638 if (groups > ngroups)
5640 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5641 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5643 /* bitmaps and block group descriptor blocks */
5644 ret += groups + gdpblocks;
5646 /* Blocks for super block, inode, quota and xattr blocks */
5647 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5653 * Calculate the total number of credits to reserve to fit
5654 * the modification of a single pages into a single transaction,
5655 * which may include multiple chunks of block allocations.
5657 * This could be called via ext4_write_begin()
5659 * We need to consider the worse case, when
5660 * one new block per extent.
5662 int ext4_writepage_trans_blocks(struct inode *inode)
5664 int bpp = ext4_journal_blocks_per_page(inode);
5667 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5669 /* Account for data blocks for journalled mode */
5670 if (ext4_should_journal_data(inode))
5676 * Calculate the journal credits for a chunk of data modification.
5678 * This is called from DIO, fallocate or whoever calling
5679 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5681 * journal buffers for data blocks are not included here, as DIO
5682 * and fallocate do no need to journal data buffers.
5684 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5686 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5690 * The caller must have previously called ext4_reserve_inode_write().
5691 * Give this, we know that the caller already has write access to iloc->bh.
5693 int ext4_mark_iloc_dirty(handle_t *handle,
5694 struct inode *inode, struct ext4_iloc *iloc)
5698 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5702 ext4_fc_track_inode(inode);
5704 if (IS_I_VERSION(inode))
5705 inode_inc_iversion(inode);
5707 /* the do_update_inode consumes one bh->b_count */
5710 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5711 err = ext4_do_update_inode(handle, inode, iloc);
5717 * On success, We end up with an outstanding reference count against
5718 * iloc->bh. This _must_ be cleaned up later.
5722 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5723 struct ext4_iloc *iloc)
5727 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5730 err = ext4_get_inode_loc(inode, iloc);
5732 BUFFER_TRACE(iloc->bh, "get_write_access");
5733 err = ext4_journal_get_write_access(handle, iloc->bh);
5739 ext4_std_error(inode->i_sb, err);
5743 static int __ext4_expand_extra_isize(struct inode *inode,
5744 unsigned int new_extra_isize,
5745 struct ext4_iloc *iloc,
5746 handle_t *handle, int *no_expand)
5748 struct ext4_inode *raw_inode;
5749 struct ext4_xattr_ibody_header *header;
5750 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5751 struct ext4_inode_info *ei = EXT4_I(inode);
5754 /* this was checked at iget time, but double check for good measure */
5755 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5756 (ei->i_extra_isize & 3)) {
5757 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5759 EXT4_INODE_SIZE(inode->i_sb));
5760 return -EFSCORRUPTED;
5762 if ((new_extra_isize < ei->i_extra_isize) ||
5763 (new_extra_isize < 4) ||
5764 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5765 return -EINVAL; /* Should never happen */
5767 raw_inode = ext4_raw_inode(iloc);
5769 header = IHDR(inode, raw_inode);
5771 /* No extended attributes present */
5772 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5773 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5774 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5775 EXT4_I(inode)->i_extra_isize, 0,
5776 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5777 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5781 /* try to expand with EAs present */
5782 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5786 * Inode size expansion failed; don't try again
5795 * Expand an inode by new_extra_isize bytes.
5796 * Returns 0 on success or negative error number on failure.
5798 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5799 unsigned int new_extra_isize,
5800 struct ext4_iloc iloc,
5806 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5810 * In nojournal mode, we can immediately attempt to expand
5811 * the inode. When journaled, we first need to obtain extra
5812 * buffer credits since we may write into the EA block
5813 * with this same handle. If journal_extend fails, then it will
5814 * only result in a minor loss of functionality for that inode.
5815 * If this is felt to be critical, then e2fsck should be run to
5816 * force a large enough s_min_extra_isize.
5818 if (ext4_journal_extend(handle,
5819 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5822 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5825 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5826 handle, &no_expand);
5827 ext4_write_unlock_xattr(inode, &no_expand);
5832 int ext4_expand_extra_isize(struct inode *inode,
5833 unsigned int new_extra_isize,
5834 struct ext4_iloc *iloc)
5840 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5845 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5846 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5847 if (IS_ERR(handle)) {
5848 error = PTR_ERR(handle);
5853 ext4_write_lock_xattr(inode, &no_expand);
5855 BUFFER_TRACE(iloc->bh, "get_write_access");
5856 error = ext4_journal_get_write_access(handle, iloc->bh);
5862 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5863 handle, &no_expand);
5865 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5870 ext4_write_unlock_xattr(inode, &no_expand);
5871 ext4_journal_stop(handle);
5876 * What we do here is to mark the in-core inode as clean with respect to inode
5877 * dirtiness (it may still be data-dirty).
5878 * This means that the in-core inode may be reaped by prune_icache
5879 * without having to perform any I/O. This is a very good thing,
5880 * because *any* task may call prune_icache - even ones which
5881 * have a transaction open against a different journal.
5883 * Is this cheating? Not really. Sure, we haven't written the
5884 * inode out, but prune_icache isn't a user-visible syncing function.
5885 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5886 * we start and wait on commits.
5888 int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5889 const char *func, unsigned int line)
5891 struct ext4_iloc iloc;
5892 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5896 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5897 err = ext4_reserve_inode_write(handle, inode, &iloc);
5901 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5902 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5905 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5908 ext4_error_inode_err(inode, func, line, 0, err,
5909 "mark_inode_dirty error");
5914 * ext4_dirty_inode() is called from __mark_inode_dirty()
5916 * We're really interested in the case where a file is being extended.
5917 * i_size has been changed by generic_commit_write() and we thus need
5918 * to include the updated inode in the current transaction.
5920 * Also, dquot_alloc_block() will always dirty the inode when blocks
5921 * are allocated to the file.
5923 * If the inode is marked synchronous, we don't honour that here - doing
5924 * so would cause a commit on atime updates, which we don't bother doing.
5925 * We handle synchronous inodes at the highest possible level.
5927 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5928 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5929 * to copy into the on-disk inode structure are the timestamp files.
5931 void ext4_dirty_inode(struct inode *inode, int flags)
5935 if (flags == I_DIRTY_TIME)
5937 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5941 ext4_mark_inode_dirty(handle, inode);
5943 ext4_journal_stop(handle);
5948 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5953 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5956 * We have to be very careful here: changing a data block's
5957 * journaling status dynamically is dangerous. If we write a
5958 * data block to the journal, change the status and then delete
5959 * that block, we risk forgetting to revoke the old log record
5960 * from the journal and so a subsequent replay can corrupt data.
5961 * So, first we make sure that the journal is empty and that
5962 * nobody is changing anything.
5965 journal = EXT4_JOURNAL(inode);
5968 if (is_journal_aborted(journal))
5971 /* Wait for all existing dio workers */
5972 inode_dio_wait(inode);
5975 * Before flushing the journal and switching inode's aops, we have
5976 * to flush all dirty data the inode has. There can be outstanding
5977 * delayed allocations, there can be unwritten extents created by
5978 * fallocate or buffered writes in dioread_nolock mode covered by
5979 * dirty data which can be converted only after flushing the dirty
5980 * data (and journalled aops don't know how to handle these cases).
5983 down_write(&EXT4_I(inode)->i_mmap_sem);
5984 err = filemap_write_and_wait(inode->i_mapping);
5986 up_write(&EXT4_I(inode)->i_mmap_sem);
5991 percpu_down_write(&sbi->s_writepages_rwsem);
5992 jbd2_journal_lock_updates(journal);
5995 * OK, there are no updates running now, and all cached data is
5996 * synced to disk. We are now in a completely consistent state
5997 * which doesn't have anything in the journal, and we know that
5998 * no filesystem updates are running, so it is safe to modify
5999 * the inode's in-core data-journaling state flag now.
6003 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6005 err = jbd2_journal_flush(journal);
6007 jbd2_journal_unlock_updates(journal);
6008 percpu_up_write(&sbi->s_writepages_rwsem);
6011 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6013 ext4_set_aops(inode);
6015 jbd2_journal_unlock_updates(journal);
6016 percpu_up_write(&sbi->s_writepages_rwsem);
6019 up_write(&EXT4_I(inode)->i_mmap_sem);
6021 /* Finally we can mark the inode as dirty. */
6023 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6025 return PTR_ERR(handle);
6027 ext4_fc_mark_ineligible(inode->i_sb,
6028 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE);
6029 err = ext4_mark_inode_dirty(handle, inode);
6030 ext4_handle_sync(handle);
6031 ext4_journal_stop(handle);
6032 ext4_std_error(inode->i_sb, err);
6037 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6039 return !buffer_mapped(bh);
6042 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6044 struct vm_area_struct *vma = vmf->vma;
6045 struct page *page = vmf->page;
6050 struct file *file = vma->vm_file;
6051 struct inode *inode = file_inode(file);
6052 struct address_space *mapping = inode->i_mapping;
6054 get_block_t *get_block;
6057 if (unlikely(IS_IMMUTABLE(inode)))
6058 return VM_FAULT_SIGBUS;
6060 sb_start_pagefault(inode->i_sb);
6061 file_update_time(vma->vm_file);
6063 down_read(&EXT4_I(inode)->i_mmap_sem);
6065 err = ext4_convert_inline_data(inode);
6070 * On data journalling we skip straight to the transaction handle:
6071 * there's no delalloc; page truncated will be checked later; the
6072 * early return w/ all buffers mapped (calculates size/len) can't
6073 * be used; and there's no dioread_nolock, so only ext4_get_block.
6075 if (ext4_should_journal_data(inode))
6078 /* Delalloc case is easy... */
6079 if (test_opt(inode->i_sb, DELALLOC) &&
6080 !ext4_nonda_switch(inode->i_sb)) {
6082 err = block_page_mkwrite(vma, vmf,
6083 ext4_da_get_block_prep);
6084 } while (err == -ENOSPC &&
6085 ext4_should_retry_alloc(inode->i_sb, &retries));
6090 size = i_size_read(inode);
6091 /* Page got truncated from under us? */
6092 if (page->mapping != mapping || page_offset(page) > size) {
6094 ret = VM_FAULT_NOPAGE;
6098 if (page->index == size >> PAGE_SHIFT)
6099 len = size & ~PAGE_MASK;
6103 * Return if we have all the buffers mapped. This avoids the need to do
6104 * journal_start/journal_stop which can block and take a long time
6106 * This cannot be done for data journalling, as we have to add the
6107 * inode to the transaction's list to writeprotect pages on commit.
6109 if (page_has_buffers(page)) {
6110 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6112 ext4_bh_unmapped)) {
6113 /* Wait so that we don't change page under IO */
6114 wait_for_stable_page(page);
6115 ret = VM_FAULT_LOCKED;
6120 /* OK, we need to fill the hole... */
6121 if (ext4_should_dioread_nolock(inode))
6122 get_block = ext4_get_block_unwritten;
6124 get_block = ext4_get_block;
6126 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6127 ext4_writepage_trans_blocks(inode));
6128 if (IS_ERR(handle)) {
6129 ret = VM_FAULT_SIGBUS;
6133 * Data journalling can't use block_page_mkwrite() because it
6134 * will set_buffer_dirty() before do_journal_get_write_access()
6135 * thus might hit warning messages for dirty metadata buffers.
6137 if (!ext4_should_journal_data(inode)) {
6138 err = block_page_mkwrite(vma, vmf, get_block);
6141 size = i_size_read(inode);
6142 /* Page got truncated from under us? */
6143 if (page->mapping != mapping || page_offset(page) > size) {
6144 ret = VM_FAULT_NOPAGE;
6148 if (page->index == size >> PAGE_SHIFT)
6149 len = size & ~PAGE_MASK;
6153 err = __block_write_begin(page, 0, len, ext4_get_block);
6155 ret = VM_FAULT_SIGBUS;
6156 if (ext4_walk_page_buffers(handle, page_buffers(page),
6157 0, len, NULL, do_journal_get_write_access))
6159 if (ext4_walk_page_buffers(handle, page_buffers(page),
6160 0, len, NULL, write_end_fn))
6162 if (ext4_jbd2_inode_add_write(handle, inode,
6163 page_offset(page), len))
6165 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6170 ext4_journal_stop(handle);
6171 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6174 ret = block_page_mkwrite_return(err);
6176 up_read(&EXT4_I(inode)->i_mmap_sem);
6177 sb_end_pagefault(inode->i_sb);
6181 ext4_journal_stop(handle);
6185 vm_fault_t ext4_filemap_fault(struct vm_fault *vmf)
6187 struct inode *inode = file_inode(vmf->vma->vm_file);
6190 down_read(&EXT4_I(inode)->i_mmap_sem);
6191 ret = filemap_fault(vmf);
6192 up_read(&EXT4_I(inode)->i_mmap_sem);