1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/inode.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
23 #include <linux/time.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/dax.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/bitops.h>
41 #include <linux/iomap.h>
42 #include <linux/iversion.h>
44 #include "ext4_jbd2.h"
49 #include <trace/events/ext4.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
54 struct ext4_inode_info *ei)
56 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
59 int offset = offsetof(struct ext4_inode, i_checksum_lo);
60 unsigned int csum_size = sizeof(dummy_csum);
62 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
63 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
65 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
66 EXT4_GOOD_OLD_INODE_SIZE - offset);
68 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
69 offset = offsetof(struct ext4_inode, i_checksum_hi);
70 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
71 EXT4_GOOD_OLD_INODE_SIZE,
72 offset - EXT4_GOOD_OLD_INODE_SIZE);
73 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
74 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
78 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
79 EXT4_INODE_SIZE(inode->i_sb) - offset);
85 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
86 struct ext4_inode_info *ei)
88 __u32 provided, calculated;
90 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
91 cpu_to_le32(EXT4_OS_LINUX) ||
92 !ext4_has_metadata_csum(inode->i_sb))
95 provided = le16_to_cpu(raw->i_checksum_lo);
96 calculated = ext4_inode_csum(inode, raw, ei);
97 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
98 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
99 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
101 calculated &= 0xFFFF;
103 return provided == calculated;
106 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
107 struct ext4_inode_info *ei)
111 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
112 cpu_to_le32(EXT4_OS_LINUX) ||
113 !ext4_has_metadata_csum(inode->i_sb))
116 csum = ext4_inode_csum(inode, raw, ei);
117 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
118 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
119 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
120 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
123 static inline int ext4_begin_ordered_truncate(struct inode *inode,
126 trace_ext4_begin_ordered_truncate(inode, new_size);
128 * If jinode is zero, then we never opened the file for
129 * writing, so there's no need to call
130 * jbd2_journal_begin_ordered_truncate() since there's no
131 * outstanding writes we need to flush.
133 if (!EXT4_I(inode)->jinode)
135 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
136 EXT4_I(inode)->jinode,
140 static void ext4_invalidatepage(struct page *page, unsigned int offset,
141 unsigned int length);
142 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
143 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
144 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
148 * Test whether an inode is a fast symlink.
149 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
151 int ext4_inode_is_fast_symlink(struct inode *inode)
153 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
154 int ea_blocks = EXT4_I(inode)->i_file_acl ?
155 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
157 if (ext4_has_inline_data(inode))
160 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
162 return S_ISLNK(inode->i_mode) && inode->i_size &&
163 (inode->i_size < EXT4_N_BLOCKS * 4);
167 * Restart the transaction associated with *handle. This does a commit,
168 * so before we call here everything must be consistently dirtied against
171 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
177 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
178 * moment, get_block can be called only for blocks inside i_size since
179 * page cache has been already dropped and writes are blocked by
180 * i_mutex. So we can safely drop the i_data_sem here.
182 BUG_ON(EXT4_JOURNAL(inode) == NULL);
183 jbd_debug(2, "restarting handle %p\n", handle);
184 up_write(&EXT4_I(inode)->i_data_sem);
185 ret = ext4_journal_restart(handle, nblocks);
186 down_write(&EXT4_I(inode)->i_data_sem);
187 ext4_discard_preallocations(inode);
193 * Called at the last iput() if i_nlink is zero.
195 void ext4_evict_inode(struct inode *inode)
199 int extra_credits = 3;
200 struct ext4_xattr_inode_array *ea_inode_array = NULL;
202 trace_ext4_evict_inode(inode);
204 if (inode->i_nlink) {
206 * When journalling data dirty buffers are tracked only in the
207 * journal. So although mm thinks everything is clean and
208 * ready for reaping the inode might still have some pages to
209 * write in the running transaction or waiting to be
210 * checkpointed. Thus calling jbd2_journal_invalidatepage()
211 * (via truncate_inode_pages()) to discard these buffers can
212 * cause data loss. Also even if we did not discard these
213 * buffers, we would have no way to find them after the inode
214 * is reaped and thus user could see stale data if he tries to
215 * read them before the transaction is checkpointed. So be
216 * careful and force everything to disk here... We use
217 * ei->i_datasync_tid to store the newest transaction
218 * containing inode's data.
220 * Note that directories do not have this problem because they
221 * don't use page cache.
223 if (inode->i_ino != EXT4_JOURNAL_INO &&
224 ext4_should_journal_data(inode) &&
225 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
226 inode->i_data.nrpages) {
227 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
228 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
230 jbd2_complete_transaction(journal, commit_tid);
231 filemap_write_and_wait(&inode->i_data);
233 truncate_inode_pages_final(&inode->i_data);
238 if (is_bad_inode(inode))
240 dquot_initialize(inode);
242 if (ext4_should_order_data(inode))
243 ext4_begin_ordered_truncate(inode, 0);
244 truncate_inode_pages_final(&inode->i_data);
247 * Protect us against freezing - iput() caller didn't have to have any
248 * protection against it
250 sb_start_intwrite(inode->i_sb);
252 if (!IS_NOQUOTA(inode))
253 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
255 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
256 ext4_blocks_for_truncate(inode)+extra_credits);
257 if (IS_ERR(handle)) {
258 ext4_std_error(inode->i_sb, PTR_ERR(handle));
260 * If we're going to skip the normal cleanup, we still need to
261 * make sure that the in-core orphan linked list is properly
264 ext4_orphan_del(NULL, inode);
265 sb_end_intwrite(inode->i_sb);
270 ext4_handle_sync(handle);
273 * Set inode->i_size to 0 before calling ext4_truncate(). We need
274 * special handling of symlinks here because i_size is used to
275 * determine whether ext4_inode_info->i_data contains symlink data or
276 * block mappings. Setting i_size to 0 will remove its fast symlink
277 * status. Erase i_data so that it becomes a valid empty block map.
279 if (ext4_inode_is_fast_symlink(inode))
280 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
282 err = ext4_mark_inode_dirty(handle, inode);
284 ext4_warning(inode->i_sb,
285 "couldn't mark inode dirty (err %d)", err);
288 if (inode->i_blocks) {
289 err = ext4_truncate(inode);
291 ext4_error(inode->i_sb,
292 "couldn't truncate inode %lu (err %d)",
298 /* Remove xattr references. */
299 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
302 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
304 ext4_journal_stop(handle);
305 ext4_orphan_del(NULL, inode);
306 sb_end_intwrite(inode->i_sb);
307 ext4_xattr_inode_array_free(ea_inode_array);
312 * Kill off the orphan record which ext4_truncate created.
313 * AKPM: I think this can be inside the above `if'.
314 * Note that ext4_orphan_del() has to be able to cope with the
315 * deletion of a non-existent orphan - this is because we don't
316 * know if ext4_truncate() actually created an orphan record.
317 * (Well, we could do this if we need to, but heck - it works)
319 ext4_orphan_del(handle, inode);
320 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
323 * One subtle ordering requirement: if anything has gone wrong
324 * (transaction abort, IO errors, whatever), then we can still
325 * do these next steps (the fs will already have been marked as
326 * having errors), but we can't free the inode if the mark_dirty
329 if (ext4_mark_inode_dirty(handle, inode))
330 /* If that failed, just do the required in-core inode clear. */
331 ext4_clear_inode(inode);
333 ext4_free_inode(handle, inode);
334 ext4_journal_stop(handle);
335 sb_end_intwrite(inode->i_sb);
336 ext4_xattr_inode_array_free(ea_inode_array);
339 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
343 qsize_t *ext4_get_reserved_space(struct inode *inode)
345 return &EXT4_I(inode)->i_reserved_quota;
350 * Called with i_data_sem down, which is important since we can call
351 * ext4_discard_preallocations() from here.
353 void ext4_da_update_reserve_space(struct inode *inode,
354 int used, int quota_claim)
356 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
357 struct ext4_inode_info *ei = EXT4_I(inode);
359 spin_lock(&ei->i_block_reservation_lock);
360 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
361 if (unlikely(used > ei->i_reserved_data_blocks)) {
362 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
363 "with only %d reserved data blocks",
364 __func__, inode->i_ino, used,
365 ei->i_reserved_data_blocks);
367 used = ei->i_reserved_data_blocks;
370 /* Update per-inode reservations */
371 ei->i_reserved_data_blocks -= used;
372 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
374 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
376 /* Update quota subsystem for data blocks */
378 dquot_claim_block(inode, EXT4_C2B(sbi, used));
381 * We did fallocate with an offset that is already delayed
382 * allocated. So on delayed allocated writeback we should
383 * not re-claim the quota for fallocated blocks.
385 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
389 * If we have done all the pending block allocations and if
390 * there aren't any writers on the inode, we can discard the
391 * inode's preallocations.
393 if ((ei->i_reserved_data_blocks == 0) &&
394 !inode_is_open_for_write(inode))
395 ext4_discard_preallocations(inode);
398 static int __check_block_validity(struct inode *inode, const char *func,
400 struct ext4_map_blocks *map)
402 if (ext4_has_feature_journal(inode->i_sb) &&
404 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
406 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
408 ext4_error_inode(inode, func, line, map->m_pblk,
409 "lblock %lu mapped to illegal pblock %llu "
410 "(length %d)", (unsigned long) map->m_lblk,
411 map->m_pblk, map->m_len);
412 return -EFSCORRUPTED;
417 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
422 if (IS_ENCRYPTED(inode))
423 return fscrypt_zeroout_range(inode, lblk, pblk, len);
425 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
432 #define check_block_validity(inode, map) \
433 __check_block_validity((inode), __func__, __LINE__, (map))
435 #ifdef ES_AGGRESSIVE_TEST
436 static void ext4_map_blocks_es_recheck(handle_t *handle,
438 struct ext4_map_blocks *es_map,
439 struct ext4_map_blocks *map,
446 * There is a race window that the result is not the same.
447 * e.g. xfstests #223 when dioread_nolock enables. The reason
448 * is that we lookup a block mapping in extent status tree with
449 * out taking i_data_sem. So at the time the unwritten extent
450 * could be converted.
452 down_read(&EXT4_I(inode)->i_data_sem);
453 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
454 retval = ext4_ext_map_blocks(handle, inode, map, flags &
455 EXT4_GET_BLOCKS_KEEP_SIZE);
457 retval = ext4_ind_map_blocks(handle, inode, map, flags &
458 EXT4_GET_BLOCKS_KEEP_SIZE);
460 up_read((&EXT4_I(inode)->i_data_sem));
463 * We don't check m_len because extent will be collpased in status
464 * tree. So the m_len might not equal.
466 if (es_map->m_lblk != map->m_lblk ||
467 es_map->m_flags != map->m_flags ||
468 es_map->m_pblk != map->m_pblk) {
469 printk("ES cache assertion failed for inode: %lu "
470 "es_cached ex [%d/%d/%llu/%x] != "
471 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
472 inode->i_ino, es_map->m_lblk, es_map->m_len,
473 es_map->m_pblk, es_map->m_flags, map->m_lblk,
474 map->m_len, map->m_pblk, map->m_flags,
478 #endif /* ES_AGGRESSIVE_TEST */
481 * The ext4_map_blocks() function tries to look up the requested blocks,
482 * and returns if the blocks are already mapped.
484 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
485 * and store the allocated blocks in the result buffer head and mark it
488 * If file type is extents based, it will call ext4_ext_map_blocks(),
489 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
492 * On success, it returns the number of blocks being mapped or allocated. if
493 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
494 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
496 * It returns 0 if plain look up failed (blocks have not been allocated), in
497 * that case, @map is returned as unmapped but we still do fill map->m_len to
498 * indicate the length of a hole starting at map->m_lblk.
500 * It returns the error in case of allocation failure.
502 int ext4_map_blocks(handle_t *handle, struct inode *inode,
503 struct ext4_map_blocks *map, int flags)
505 struct extent_status es;
508 #ifdef ES_AGGRESSIVE_TEST
509 struct ext4_map_blocks orig_map;
511 memcpy(&orig_map, map, sizeof(*map));
515 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
516 "logical block %lu\n", inode->i_ino, flags, map->m_len,
517 (unsigned long) map->m_lblk);
520 * ext4_map_blocks returns an int, and m_len is an unsigned int
522 if (unlikely(map->m_len > INT_MAX))
523 map->m_len = INT_MAX;
525 /* We can handle the block number less than EXT_MAX_BLOCKS */
526 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
527 return -EFSCORRUPTED;
529 /* Lookup extent status tree firstly */
530 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
531 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
532 map->m_pblk = ext4_es_pblock(&es) +
533 map->m_lblk - es.es_lblk;
534 map->m_flags |= ext4_es_is_written(&es) ?
535 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
536 retval = es.es_len - (map->m_lblk - es.es_lblk);
537 if (retval > map->m_len)
540 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
542 retval = es.es_len - (map->m_lblk - es.es_lblk);
543 if (retval > map->m_len)
550 #ifdef ES_AGGRESSIVE_TEST
551 ext4_map_blocks_es_recheck(handle, inode, map,
558 * Try to see if we can get the block without requesting a new
561 down_read(&EXT4_I(inode)->i_data_sem);
562 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
563 retval = ext4_ext_map_blocks(handle, inode, map, flags &
564 EXT4_GET_BLOCKS_KEEP_SIZE);
566 retval = ext4_ind_map_blocks(handle, inode, map, flags &
567 EXT4_GET_BLOCKS_KEEP_SIZE);
572 if (unlikely(retval != map->m_len)) {
573 ext4_warning(inode->i_sb,
574 "ES len assertion failed for inode "
575 "%lu: retval %d != map->m_len %d",
576 inode->i_ino, retval, map->m_len);
580 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
581 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
582 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
583 !(status & EXTENT_STATUS_WRITTEN) &&
584 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
585 map->m_lblk + map->m_len - 1))
586 status |= EXTENT_STATUS_DELAYED;
587 ret = ext4_es_insert_extent(inode, map->m_lblk,
588 map->m_len, map->m_pblk, status);
592 up_read((&EXT4_I(inode)->i_data_sem));
595 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
596 ret = check_block_validity(inode, map);
601 /* If it is only a block(s) look up */
602 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
606 * Returns if the blocks have already allocated
608 * Note that if blocks have been preallocated
609 * ext4_ext_get_block() returns the create = 0
610 * with buffer head unmapped.
612 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
614 * If we need to convert extent to unwritten
615 * we continue and do the actual work in
616 * ext4_ext_map_blocks()
618 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
622 * Here we clear m_flags because after allocating an new extent,
623 * it will be set again.
625 map->m_flags &= ~EXT4_MAP_FLAGS;
628 * New blocks allocate and/or writing to unwritten extent
629 * will possibly result in updating i_data, so we take
630 * the write lock of i_data_sem, and call get_block()
631 * with create == 1 flag.
633 down_write(&EXT4_I(inode)->i_data_sem);
636 * We need to check for EXT4 here because migrate
637 * could have changed the inode type in between
639 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
640 retval = ext4_ext_map_blocks(handle, inode, map, flags);
642 retval = ext4_ind_map_blocks(handle, inode, map, flags);
644 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
646 * We allocated new blocks which will result in
647 * i_data's format changing. Force the migrate
648 * to fail by clearing migrate flags
650 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
654 * Update reserved blocks/metadata blocks after successful
655 * block allocation which had been deferred till now. We don't
656 * support fallocate for non extent files. So we can update
657 * reserve space here.
660 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
661 ext4_da_update_reserve_space(inode, retval, 1);
667 if (unlikely(retval != map->m_len)) {
668 ext4_warning(inode->i_sb,
669 "ES len assertion failed for inode "
670 "%lu: retval %d != map->m_len %d",
671 inode->i_ino, retval, map->m_len);
676 * We have to zeroout blocks before inserting them into extent
677 * status tree. Otherwise someone could look them up there and
678 * use them before they are really zeroed. We also have to
679 * unmap metadata before zeroing as otherwise writeback can
680 * overwrite zeros with stale data from block device.
682 if (flags & EXT4_GET_BLOCKS_ZERO &&
683 map->m_flags & EXT4_MAP_MAPPED &&
684 map->m_flags & EXT4_MAP_NEW) {
685 ret = ext4_issue_zeroout(inode, map->m_lblk,
686 map->m_pblk, map->m_len);
694 * If the extent has been zeroed out, we don't need to update
695 * extent status tree.
697 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
698 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
699 if (ext4_es_is_written(&es))
702 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
703 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
704 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
705 !(status & EXTENT_STATUS_WRITTEN) &&
706 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
707 map->m_lblk + map->m_len - 1))
708 status |= EXTENT_STATUS_DELAYED;
709 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
710 map->m_pblk, status);
718 up_write((&EXT4_I(inode)->i_data_sem));
719 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
720 ret = check_block_validity(inode, map);
725 * Inodes with freshly allocated blocks where contents will be
726 * visible after transaction commit must be on transaction's
729 if (map->m_flags & EXT4_MAP_NEW &&
730 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
731 !(flags & EXT4_GET_BLOCKS_ZERO) &&
732 !ext4_is_quota_file(inode) &&
733 ext4_should_order_data(inode)) {
735 (loff_t)map->m_lblk << inode->i_blkbits;
736 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
738 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
739 ret = ext4_jbd2_inode_add_wait(handle, inode,
742 ret = ext4_jbd2_inode_add_write(handle, inode,
752 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
753 * we have to be careful as someone else may be manipulating b_state as well.
755 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
757 unsigned long old_state;
758 unsigned long new_state;
760 flags &= EXT4_MAP_FLAGS;
762 /* Dummy buffer_head? Set non-atomically. */
764 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
768 * Someone else may be modifying b_state. Be careful! This is ugly but
769 * once we get rid of using bh as a container for mapping information
770 * to pass to / from get_block functions, this can go away.
773 old_state = READ_ONCE(bh->b_state);
774 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
776 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
779 static int _ext4_get_block(struct inode *inode, sector_t iblock,
780 struct buffer_head *bh, int flags)
782 struct ext4_map_blocks map;
785 if (ext4_has_inline_data(inode))
789 map.m_len = bh->b_size >> inode->i_blkbits;
791 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
794 map_bh(bh, inode->i_sb, map.m_pblk);
795 ext4_update_bh_state(bh, map.m_flags);
796 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
798 } else if (ret == 0) {
799 /* hole case, need to fill in bh->b_size */
800 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
805 int ext4_get_block(struct inode *inode, sector_t iblock,
806 struct buffer_head *bh, int create)
808 return _ext4_get_block(inode, iblock, bh,
809 create ? EXT4_GET_BLOCKS_CREATE : 0);
813 * Get block function used when preparing for buffered write if we require
814 * creating an unwritten extent if blocks haven't been allocated. The extent
815 * will be converted to written after the IO is complete.
817 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
818 struct buffer_head *bh_result, int create)
820 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
821 inode->i_ino, create);
822 return _ext4_get_block(inode, iblock, bh_result,
823 EXT4_GET_BLOCKS_IO_CREATE_EXT);
826 /* Maximum number of blocks we map for direct IO at once. */
827 #define DIO_MAX_BLOCKS 4096
830 * Get blocks function for the cases that need to start a transaction -
831 * generally difference cases of direct IO and DAX IO. It also handles retries
834 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
835 struct buffer_head *bh_result, int flags)
842 /* Trim mapping request to maximum we can map at once for DIO */
843 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
844 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
845 dio_credits = ext4_chunk_trans_blocks(inode,
846 bh_result->b_size >> inode->i_blkbits);
848 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
850 return PTR_ERR(handle);
852 ret = _ext4_get_block(inode, iblock, bh_result, flags);
853 ext4_journal_stop(handle);
855 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
860 /* Get block function for DIO reads and writes to inodes without extents */
861 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
862 struct buffer_head *bh, int create)
864 /* We don't expect handle for direct IO */
865 WARN_ON_ONCE(ext4_journal_current_handle());
868 return _ext4_get_block(inode, iblock, bh, 0);
869 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
873 * Get block function for AIO DIO writes when we create unwritten extent if
874 * blocks are not allocated yet. The extent will be converted to written
875 * after IO is complete.
877 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
878 sector_t iblock, struct buffer_head *bh_result, int create)
882 /* We don't expect handle for direct IO */
883 WARN_ON_ONCE(ext4_journal_current_handle());
885 ret = ext4_get_block_trans(inode, iblock, bh_result,
886 EXT4_GET_BLOCKS_IO_CREATE_EXT);
889 * When doing DIO using unwritten extents, we need io_end to convert
890 * unwritten extents to written on IO completion. We allocate io_end
891 * once we spot unwritten extent and store it in b_private. Generic
892 * DIO code keeps b_private set and furthermore passes the value to
893 * our completion callback in 'private' argument.
895 if (!ret && buffer_unwritten(bh_result)) {
896 if (!bh_result->b_private) {
897 ext4_io_end_t *io_end;
899 io_end = ext4_init_io_end(inode, GFP_KERNEL);
902 bh_result->b_private = io_end;
903 ext4_set_io_unwritten_flag(inode, io_end);
905 set_buffer_defer_completion(bh_result);
912 * Get block function for non-AIO DIO writes when we create unwritten extent if
913 * blocks are not allocated yet. The extent will be converted to written
914 * after IO is complete by ext4_direct_IO_write().
916 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
917 sector_t iblock, struct buffer_head *bh_result, int create)
921 /* We don't expect handle for direct IO */
922 WARN_ON_ONCE(ext4_journal_current_handle());
924 ret = ext4_get_block_trans(inode, iblock, bh_result,
925 EXT4_GET_BLOCKS_IO_CREATE_EXT);
928 * Mark inode as having pending DIO writes to unwritten extents.
929 * ext4_direct_IO_write() checks this flag and converts extents to
932 if (!ret && buffer_unwritten(bh_result))
933 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
938 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
939 struct buffer_head *bh_result, int create)
943 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
944 inode->i_ino, create);
945 /* We don't expect handle for direct IO */
946 WARN_ON_ONCE(ext4_journal_current_handle());
948 ret = _ext4_get_block(inode, iblock, bh_result, 0);
950 * Blocks should have been preallocated! ext4_file_write_iter() checks
953 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
960 * `handle' can be NULL if create is zero
962 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
963 ext4_lblk_t block, int map_flags)
965 struct ext4_map_blocks map;
966 struct buffer_head *bh;
967 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
970 J_ASSERT(handle != NULL || create == 0);
974 err = ext4_map_blocks(handle, inode, &map, map_flags);
977 return create ? ERR_PTR(-ENOSPC) : NULL;
981 bh = sb_getblk(inode->i_sb, map.m_pblk);
983 return ERR_PTR(-ENOMEM);
984 if (map.m_flags & EXT4_MAP_NEW) {
985 J_ASSERT(create != 0);
986 J_ASSERT(handle != NULL);
989 * Now that we do not always journal data, we should
990 * keep in mind whether this should always journal the
991 * new buffer as metadata. For now, regular file
992 * writes use ext4_get_block instead, so it's not a
996 BUFFER_TRACE(bh, "call get_create_access");
997 err = ext4_journal_get_create_access(handle, bh);
1002 if (!buffer_uptodate(bh)) {
1003 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1004 set_buffer_uptodate(bh);
1007 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1008 err = ext4_handle_dirty_metadata(handle, inode, bh);
1012 BUFFER_TRACE(bh, "not a new buffer");
1016 return ERR_PTR(err);
1019 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1020 ext4_lblk_t block, int map_flags)
1022 struct buffer_head *bh;
1024 bh = ext4_getblk(handle, inode, block, map_flags);
1027 if (!bh || buffer_uptodate(bh))
1029 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1031 if (buffer_uptodate(bh))
1034 return ERR_PTR(-EIO);
1037 /* Read a contiguous batch of blocks. */
1038 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1039 bool wait, struct buffer_head **bhs)
1043 for (i = 0; i < bh_count; i++) {
1044 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1045 if (IS_ERR(bhs[i])) {
1046 err = PTR_ERR(bhs[i]);
1052 for (i = 0; i < bh_count; i++)
1053 /* Note that NULL bhs[i] is valid because of holes. */
1054 if (bhs[i] && !buffer_uptodate(bhs[i]))
1055 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1061 for (i = 0; i < bh_count; i++)
1063 wait_on_buffer(bhs[i]);
1065 for (i = 0; i < bh_count; i++) {
1066 if (bhs[i] && !buffer_uptodate(bhs[i])) {
1074 for (i = 0; i < bh_count; i++) {
1081 int ext4_walk_page_buffers(handle_t *handle,
1082 struct buffer_head *head,
1086 int (*fn)(handle_t *handle,
1087 struct buffer_head *bh))
1089 struct buffer_head *bh;
1090 unsigned block_start, block_end;
1091 unsigned blocksize = head->b_size;
1093 struct buffer_head *next;
1095 for (bh = head, block_start = 0;
1096 ret == 0 && (bh != head || !block_start);
1097 block_start = block_end, bh = next) {
1098 next = bh->b_this_page;
1099 block_end = block_start + blocksize;
1100 if (block_end <= from || block_start >= to) {
1101 if (partial && !buffer_uptodate(bh))
1105 err = (*fn)(handle, bh);
1113 * To preserve ordering, it is essential that the hole instantiation and
1114 * the data write be encapsulated in a single transaction. We cannot
1115 * close off a transaction and start a new one between the ext4_get_block()
1116 * and the commit_write(). So doing the jbd2_journal_start at the start of
1117 * prepare_write() is the right place.
1119 * Also, this function can nest inside ext4_writepage(). In that case, we
1120 * *know* that ext4_writepage() has generated enough buffer credits to do the
1121 * whole page. So we won't block on the journal in that case, which is good,
1122 * because the caller may be PF_MEMALLOC.
1124 * By accident, ext4 can be reentered when a transaction is open via
1125 * quota file writes. If we were to commit the transaction while thus
1126 * reentered, there can be a deadlock - we would be holding a quota
1127 * lock, and the commit would never complete if another thread had a
1128 * transaction open and was blocking on the quota lock - a ranking
1131 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1132 * will _not_ run commit under these circumstances because handle->h_ref
1133 * is elevated. We'll still have enough credits for the tiny quotafile
1136 int do_journal_get_write_access(handle_t *handle,
1137 struct buffer_head *bh)
1139 int dirty = buffer_dirty(bh);
1142 if (!buffer_mapped(bh) || buffer_freed(bh))
1145 * __block_write_begin() could have dirtied some buffers. Clean
1146 * the dirty bit as jbd2_journal_get_write_access() could complain
1147 * otherwise about fs integrity issues. Setting of the dirty bit
1148 * by __block_write_begin() isn't a real problem here as we clear
1149 * the bit before releasing a page lock and thus writeback cannot
1150 * ever write the buffer.
1153 clear_buffer_dirty(bh);
1154 BUFFER_TRACE(bh, "get write access");
1155 ret = ext4_journal_get_write_access(handle, bh);
1157 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1161 #ifdef CONFIG_FS_ENCRYPTION
1162 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1163 get_block_t *get_block)
1165 unsigned from = pos & (PAGE_SIZE - 1);
1166 unsigned to = from + len;
1167 struct inode *inode = page->mapping->host;
1168 unsigned block_start, block_end;
1171 unsigned blocksize = inode->i_sb->s_blocksize;
1173 struct buffer_head *bh, *head, *wait[2];
1177 BUG_ON(!PageLocked(page));
1178 BUG_ON(from > PAGE_SIZE);
1179 BUG_ON(to > PAGE_SIZE);
1182 if (!page_has_buffers(page))
1183 create_empty_buffers(page, blocksize, 0);
1184 head = page_buffers(page);
1185 bbits = ilog2(blocksize);
1186 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1188 for (bh = head, block_start = 0; bh != head || !block_start;
1189 block++, block_start = block_end, bh = bh->b_this_page) {
1190 block_end = block_start + blocksize;
1191 if (block_end <= from || block_start >= to) {
1192 if (PageUptodate(page)) {
1193 if (!buffer_uptodate(bh))
1194 set_buffer_uptodate(bh);
1199 clear_buffer_new(bh);
1200 if (!buffer_mapped(bh)) {
1201 WARN_ON(bh->b_size != blocksize);
1202 err = get_block(inode, block, bh, 1);
1205 if (buffer_new(bh)) {
1206 if (PageUptodate(page)) {
1207 clear_buffer_new(bh);
1208 set_buffer_uptodate(bh);
1209 mark_buffer_dirty(bh);
1212 if (block_end > to || block_start < from)
1213 zero_user_segments(page, to, block_end,
1218 if (PageUptodate(page)) {
1219 if (!buffer_uptodate(bh))
1220 set_buffer_uptodate(bh);
1223 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1224 !buffer_unwritten(bh) &&
1225 (block_start < from || block_end > to)) {
1226 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1227 wait[nr_wait++] = bh;
1231 * If we issued read requests, let them complete.
1233 for (i = 0; i < nr_wait; i++) {
1234 wait_on_buffer(wait[i]);
1235 if (!buffer_uptodate(wait[i]))
1238 if (unlikely(err)) {
1239 page_zero_new_buffers(page, from, to);
1240 } else if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
1241 for (i = 0; i < nr_wait; i++) {
1244 err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1245 bh_offset(wait[i]));
1247 clear_buffer_uptodate(wait[i]);
1257 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1258 loff_t pos, unsigned len, unsigned flags,
1259 struct page **pagep, void **fsdata)
1261 struct inode *inode = mapping->host;
1262 int ret, needed_blocks;
1269 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1272 trace_ext4_write_begin(inode, pos, len, flags);
1274 * Reserve one block more for addition to orphan list in case
1275 * we allocate blocks but write fails for some reason
1277 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1278 index = pos >> PAGE_SHIFT;
1279 from = pos & (PAGE_SIZE - 1);
1282 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1283 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1292 * grab_cache_page_write_begin() can take a long time if the
1293 * system is thrashing due to memory pressure, or if the page
1294 * is being written back. So grab it first before we start
1295 * the transaction handle. This also allows us to allocate
1296 * the page (if needed) without using GFP_NOFS.
1299 page = grab_cache_page_write_begin(mapping, index, flags);
1305 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1306 if (IS_ERR(handle)) {
1308 return PTR_ERR(handle);
1312 if (page->mapping != mapping) {
1313 /* The page got truncated from under us */
1316 ext4_journal_stop(handle);
1319 /* In case writeback began while the page was unlocked */
1320 wait_for_stable_page(page);
1322 #ifdef CONFIG_FS_ENCRYPTION
1323 if (ext4_should_dioread_nolock(inode))
1324 ret = ext4_block_write_begin(page, pos, len,
1325 ext4_get_block_unwritten);
1327 ret = ext4_block_write_begin(page, pos, len,
1330 if (ext4_should_dioread_nolock(inode))
1331 ret = __block_write_begin(page, pos, len,
1332 ext4_get_block_unwritten);
1334 ret = __block_write_begin(page, pos, len, ext4_get_block);
1336 if (!ret && ext4_should_journal_data(inode)) {
1337 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1339 do_journal_get_write_access);
1345 * __block_write_begin may have instantiated a few blocks
1346 * outside i_size. Trim these off again. Don't need
1347 * i_size_read because we hold i_mutex.
1349 * Add inode to orphan list in case we crash before
1352 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1353 ext4_orphan_add(handle, inode);
1355 ext4_journal_stop(handle);
1356 if (pos + len > inode->i_size) {
1357 ext4_truncate_failed_write(inode);
1359 * If truncate failed early the inode might
1360 * still be on the orphan list; we need to
1361 * make sure the inode is removed from the
1362 * orphan list in that case.
1365 ext4_orphan_del(NULL, inode);
1368 if (ret == -ENOSPC &&
1369 ext4_should_retry_alloc(inode->i_sb, &retries))
1378 /* For write_end() in data=journal mode */
1379 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1382 if (!buffer_mapped(bh) || buffer_freed(bh))
1384 set_buffer_uptodate(bh);
1385 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1386 clear_buffer_meta(bh);
1387 clear_buffer_prio(bh);
1392 * We need to pick up the new inode size which generic_commit_write gave us
1393 * `file' can be NULL - eg, when called from page_symlink().
1395 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1396 * buffers are managed internally.
1398 static int ext4_write_end(struct file *file,
1399 struct address_space *mapping,
1400 loff_t pos, unsigned len, unsigned copied,
1401 struct page *page, void *fsdata)
1403 handle_t *handle = ext4_journal_current_handle();
1404 struct inode *inode = mapping->host;
1405 loff_t old_size = inode->i_size;
1407 int i_size_changed = 0;
1408 int inline_data = ext4_has_inline_data(inode);
1410 trace_ext4_write_end(inode, pos, len, copied);
1412 ret = ext4_write_inline_data_end(inode, pos, len,
1421 copied = block_write_end(file, mapping, pos,
1422 len, copied, page, fsdata);
1424 * it's important to update i_size while still holding page lock:
1425 * page writeout could otherwise come in and zero beyond i_size.
1427 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1432 pagecache_isize_extended(inode, old_size, pos);
1434 * Don't mark the inode dirty under page lock. First, it unnecessarily
1435 * makes the holding time of page lock longer. Second, it forces lock
1436 * ordering of page lock and transaction start for journaling
1439 if (i_size_changed || inline_data)
1440 ext4_mark_inode_dirty(handle, inode);
1442 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1443 /* if we have allocated more blocks and copied
1444 * less. We will have blocks allocated outside
1445 * inode->i_size. So truncate them
1447 ext4_orphan_add(handle, inode);
1449 ret2 = ext4_journal_stop(handle);
1453 if (pos + len > inode->i_size) {
1454 ext4_truncate_failed_write(inode);
1456 * If truncate failed early the inode might still be
1457 * on the orphan list; we need to make sure the inode
1458 * is removed from the orphan list in that case.
1461 ext4_orphan_del(NULL, inode);
1464 return ret ? ret : copied;
1468 * This is a private version of page_zero_new_buffers() which doesn't
1469 * set the buffer to be dirty, since in data=journalled mode we need
1470 * to call ext4_handle_dirty_metadata() instead.
1472 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1474 unsigned from, unsigned to)
1476 unsigned int block_start = 0, block_end;
1477 struct buffer_head *head, *bh;
1479 bh = head = page_buffers(page);
1481 block_end = block_start + bh->b_size;
1482 if (buffer_new(bh)) {
1483 if (block_end > from && block_start < to) {
1484 if (!PageUptodate(page)) {
1485 unsigned start, size;
1487 start = max(from, block_start);
1488 size = min(to, block_end) - start;
1490 zero_user(page, start, size);
1491 write_end_fn(handle, bh);
1493 clear_buffer_new(bh);
1496 block_start = block_end;
1497 bh = bh->b_this_page;
1498 } while (bh != head);
1501 static int ext4_journalled_write_end(struct file *file,
1502 struct address_space *mapping,
1503 loff_t pos, unsigned len, unsigned copied,
1504 struct page *page, void *fsdata)
1506 handle_t *handle = ext4_journal_current_handle();
1507 struct inode *inode = mapping->host;
1508 loff_t old_size = inode->i_size;
1512 int size_changed = 0;
1513 int inline_data = ext4_has_inline_data(inode);
1515 trace_ext4_journalled_write_end(inode, pos, len, copied);
1516 from = pos & (PAGE_SIZE - 1);
1519 BUG_ON(!ext4_handle_valid(handle));
1522 ret = ext4_write_inline_data_end(inode, pos, len,
1530 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1532 ext4_journalled_zero_new_buffers(handle, page, from, to);
1534 if (unlikely(copied < len))
1535 ext4_journalled_zero_new_buffers(handle, page,
1537 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1538 from + copied, &partial,
1541 SetPageUptodate(page);
1543 size_changed = ext4_update_inode_size(inode, pos + copied);
1544 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1545 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1550 pagecache_isize_extended(inode, old_size, pos);
1552 if (size_changed || inline_data) {
1553 ret2 = ext4_mark_inode_dirty(handle, inode);
1558 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1559 /* if we have allocated more blocks and copied
1560 * less. We will have blocks allocated outside
1561 * inode->i_size. So truncate them
1563 ext4_orphan_add(handle, inode);
1566 ret2 = ext4_journal_stop(handle);
1569 if (pos + len > inode->i_size) {
1570 ext4_truncate_failed_write(inode);
1572 * If truncate failed early the inode might still be
1573 * on the orphan list; we need to make sure the inode
1574 * is removed from the orphan list in that case.
1577 ext4_orphan_del(NULL, inode);
1580 return ret ? ret : copied;
1584 * Reserve space for a single cluster
1586 static int ext4_da_reserve_space(struct inode *inode)
1588 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1589 struct ext4_inode_info *ei = EXT4_I(inode);
1593 * We will charge metadata quota at writeout time; this saves
1594 * us from metadata over-estimation, though we may go over by
1595 * a small amount in the end. Here we just reserve for data.
1597 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1601 spin_lock(&ei->i_block_reservation_lock);
1602 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1603 spin_unlock(&ei->i_block_reservation_lock);
1604 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1607 ei->i_reserved_data_blocks++;
1608 trace_ext4_da_reserve_space(inode);
1609 spin_unlock(&ei->i_block_reservation_lock);
1611 return 0; /* success */
1614 void ext4_da_release_space(struct inode *inode, int to_free)
1616 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1617 struct ext4_inode_info *ei = EXT4_I(inode);
1620 return; /* Nothing to release, exit */
1622 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1624 trace_ext4_da_release_space(inode, to_free);
1625 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1627 * if there aren't enough reserved blocks, then the
1628 * counter is messed up somewhere. Since this
1629 * function is called from invalidate page, it's
1630 * harmless to return without any action.
1632 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1633 "ino %lu, to_free %d with only %d reserved "
1634 "data blocks", inode->i_ino, to_free,
1635 ei->i_reserved_data_blocks);
1637 to_free = ei->i_reserved_data_blocks;
1639 ei->i_reserved_data_blocks -= to_free;
1641 /* update fs dirty data blocks counter */
1642 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1644 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1646 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1649 static void ext4_da_page_release_reservation(struct page *page,
1650 unsigned int offset,
1651 unsigned int length)
1653 int contiguous_blks = 0;
1654 struct buffer_head *head, *bh;
1655 unsigned int curr_off = 0;
1656 struct inode *inode = page->mapping->host;
1657 unsigned int stop = offset + length;
1660 BUG_ON(stop > PAGE_SIZE || stop < length);
1662 head = page_buffers(page);
1665 unsigned int next_off = curr_off + bh->b_size;
1667 if (next_off > stop)
1670 if ((offset <= curr_off) && (buffer_delay(bh))) {
1672 clear_buffer_delay(bh);
1673 } else if (contiguous_blks) {
1674 lblk = page->index <<
1675 (PAGE_SHIFT - inode->i_blkbits);
1676 lblk += (curr_off >> inode->i_blkbits) -
1678 ext4_es_remove_blks(inode, lblk, contiguous_blks);
1679 contiguous_blks = 0;
1681 curr_off = next_off;
1682 } while ((bh = bh->b_this_page) != head);
1684 if (contiguous_blks) {
1685 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1686 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1687 ext4_es_remove_blks(inode, lblk, contiguous_blks);
1693 * Delayed allocation stuff
1696 struct mpage_da_data {
1697 struct inode *inode;
1698 struct writeback_control *wbc;
1700 pgoff_t first_page; /* The first page to write */
1701 pgoff_t next_page; /* Current page to examine */
1702 pgoff_t last_page; /* Last page to examine */
1704 * Extent to map - this can be after first_page because that can be
1705 * fully mapped. We somewhat abuse m_flags to store whether the extent
1706 * is delalloc or unwritten.
1708 struct ext4_map_blocks map;
1709 struct ext4_io_submit io_submit; /* IO submission data */
1710 unsigned int do_map:1;
1713 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1718 struct pagevec pvec;
1719 struct inode *inode = mpd->inode;
1720 struct address_space *mapping = inode->i_mapping;
1722 /* This is necessary when next_page == 0. */
1723 if (mpd->first_page >= mpd->next_page)
1726 index = mpd->first_page;
1727 end = mpd->next_page - 1;
1729 ext4_lblk_t start, last;
1730 start = index << (PAGE_SHIFT - inode->i_blkbits);
1731 last = end << (PAGE_SHIFT - inode->i_blkbits);
1732 ext4_es_remove_extent(inode, start, last - start + 1);
1735 pagevec_init(&pvec);
1736 while (index <= end) {
1737 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1740 for (i = 0; i < nr_pages; i++) {
1741 struct page *page = pvec.pages[i];
1743 BUG_ON(!PageLocked(page));
1744 BUG_ON(PageWriteback(page));
1746 if (page_mapped(page))
1747 clear_page_dirty_for_io(page);
1748 block_invalidatepage(page, 0, PAGE_SIZE);
1749 ClearPageUptodate(page);
1753 pagevec_release(&pvec);
1757 static void ext4_print_free_blocks(struct inode *inode)
1759 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1760 struct super_block *sb = inode->i_sb;
1761 struct ext4_inode_info *ei = EXT4_I(inode);
1763 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1764 EXT4_C2B(EXT4_SB(inode->i_sb),
1765 ext4_count_free_clusters(sb)));
1766 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1767 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1768 (long long) EXT4_C2B(EXT4_SB(sb),
1769 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1770 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1771 (long long) EXT4_C2B(EXT4_SB(sb),
1772 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1773 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1774 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1775 ei->i_reserved_data_blocks);
1779 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1781 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1785 * ext4_insert_delayed_block - adds a delayed block to the extents status
1786 * tree, incrementing the reserved cluster/block
1787 * count or making a pending reservation
1790 * @inode - file containing the newly added block
1791 * @lblk - logical block to be added
1793 * Returns 0 on success, negative error code on failure.
1795 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1797 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1799 bool allocated = false;
1802 * If the cluster containing lblk is shared with a delayed,
1803 * written, or unwritten extent in a bigalloc file system, it's
1804 * already been accounted for and does not need to be reserved.
1805 * A pending reservation must be made for the cluster if it's
1806 * shared with a written or unwritten extent and doesn't already
1807 * have one. Written and unwritten extents can be purged from the
1808 * extents status tree if the system is under memory pressure, so
1809 * it's necessary to examine the extent tree if a search of the
1810 * extents status tree doesn't get a match.
1812 if (sbi->s_cluster_ratio == 1) {
1813 ret = ext4_da_reserve_space(inode);
1814 if (ret != 0) /* ENOSPC */
1816 } else { /* bigalloc */
1817 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1818 if (!ext4_es_scan_clu(inode,
1819 &ext4_es_is_mapped, lblk)) {
1820 ret = ext4_clu_mapped(inode,
1821 EXT4_B2C(sbi, lblk));
1825 ret = ext4_da_reserve_space(inode);
1826 if (ret != 0) /* ENOSPC */
1837 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1844 * This function is grabs code from the very beginning of
1845 * ext4_map_blocks, but assumes that the caller is from delayed write
1846 * time. This function looks up the requested blocks and sets the
1847 * buffer delay bit under the protection of i_data_sem.
1849 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1850 struct ext4_map_blocks *map,
1851 struct buffer_head *bh)
1853 struct extent_status es;
1855 sector_t invalid_block = ~((sector_t) 0xffff);
1856 #ifdef ES_AGGRESSIVE_TEST
1857 struct ext4_map_blocks orig_map;
1859 memcpy(&orig_map, map, sizeof(*map));
1862 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1866 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1867 "logical block %lu\n", inode->i_ino, map->m_len,
1868 (unsigned long) map->m_lblk);
1870 /* Lookup extent status tree firstly */
1871 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1872 if (ext4_es_is_hole(&es)) {
1874 down_read(&EXT4_I(inode)->i_data_sem);
1879 * Delayed extent could be allocated by fallocate.
1880 * So we need to check it.
1882 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1883 map_bh(bh, inode->i_sb, invalid_block);
1885 set_buffer_delay(bh);
1889 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1890 retval = es.es_len - (iblock - es.es_lblk);
1891 if (retval > map->m_len)
1892 retval = map->m_len;
1893 map->m_len = retval;
1894 if (ext4_es_is_written(&es))
1895 map->m_flags |= EXT4_MAP_MAPPED;
1896 else if (ext4_es_is_unwritten(&es))
1897 map->m_flags |= EXT4_MAP_UNWRITTEN;
1901 #ifdef ES_AGGRESSIVE_TEST
1902 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1908 * Try to see if we can get the block without requesting a new
1909 * file system block.
1911 down_read(&EXT4_I(inode)->i_data_sem);
1912 if (ext4_has_inline_data(inode))
1914 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1915 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1917 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1924 * XXX: __block_prepare_write() unmaps passed block,
1928 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1934 map_bh(bh, inode->i_sb, invalid_block);
1936 set_buffer_delay(bh);
1937 } else if (retval > 0) {
1939 unsigned int status;
1941 if (unlikely(retval != map->m_len)) {
1942 ext4_warning(inode->i_sb,
1943 "ES len assertion failed for inode "
1944 "%lu: retval %d != map->m_len %d",
1945 inode->i_ino, retval, map->m_len);
1949 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1950 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1951 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1952 map->m_pblk, status);
1958 up_read((&EXT4_I(inode)->i_data_sem));
1964 * This is a special get_block_t callback which is used by
1965 * ext4_da_write_begin(). It will either return mapped block or
1966 * reserve space for a single block.
1968 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1969 * We also have b_blocknr = -1 and b_bdev initialized properly
1971 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1972 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1973 * initialized properly.
1975 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1976 struct buffer_head *bh, int create)
1978 struct ext4_map_blocks map;
1981 BUG_ON(create == 0);
1982 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1984 map.m_lblk = iblock;
1988 * first, we need to know whether the block is allocated already
1989 * preallocated blocks are unmapped but should treated
1990 * the same as allocated blocks.
1992 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1996 map_bh(bh, inode->i_sb, map.m_pblk);
1997 ext4_update_bh_state(bh, map.m_flags);
1999 if (buffer_unwritten(bh)) {
2000 /* A delayed write to unwritten bh should be marked
2001 * new and mapped. Mapped ensures that we don't do
2002 * get_block multiple times when we write to the same
2003 * offset and new ensures that we do proper zero out
2004 * for partial write.
2007 set_buffer_mapped(bh);
2012 static int bget_one(handle_t *handle, struct buffer_head *bh)
2018 static int bput_one(handle_t *handle, struct buffer_head *bh)
2024 static int __ext4_journalled_writepage(struct page *page,
2027 struct address_space *mapping = page->mapping;
2028 struct inode *inode = mapping->host;
2029 struct buffer_head *page_bufs = NULL;
2030 handle_t *handle = NULL;
2031 int ret = 0, err = 0;
2032 int inline_data = ext4_has_inline_data(inode);
2033 struct buffer_head *inode_bh = NULL;
2035 ClearPageChecked(page);
2038 BUG_ON(page->index != 0);
2039 BUG_ON(len > ext4_get_max_inline_size(inode));
2040 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2041 if (inode_bh == NULL)
2044 page_bufs = page_buffers(page);
2049 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2053 * We need to release the page lock before we start the
2054 * journal, so grab a reference so the page won't disappear
2055 * out from under us.
2060 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2061 ext4_writepage_trans_blocks(inode));
2062 if (IS_ERR(handle)) {
2063 ret = PTR_ERR(handle);
2065 goto out_no_pagelock;
2067 BUG_ON(!ext4_handle_valid(handle));
2071 if (page->mapping != mapping) {
2072 /* The page got truncated from under us */
2073 ext4_journal_stop(handle);
2079 ret = ext4_mark_inode_dirty(handle, inode);
2081 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2082 do_journal_get_write_access);
2084 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2089 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2090 err = ext4_journal_stop(handle);
2094 if (!ext4_has_inline_data(inode))
2095 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2097 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2106 * Note that we don't need to start a transaction unless we're journaling data
2107 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2108 * need to file the inode to the transaction's list in ordered mode because if
2109 * we are writing back data added by write(), the inode is already there and if
2110 * we are writing back data modified via mmap(), no one guarantees in which
2111 * transaction the data will hit the disk. In case we are journaling data, we
2112 * cannot start transaction directly because transaction start ranks above page
2113 * lock so we have to do some magic.
2115 * This function can get called via...
2116 * - ext4_writepages after taking page lock (have journal handle)
2117 * - journal_submit_inode_data_buffers (no journal handle)
2118 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2119 * - grab_page_cache when doing write_begin (have journal handle)
2121 * We don't do any block allocation in this function. If we have page with
2122 * multiple blocks we need to write those buffer_heads that are mapped. This
2123 * is important for mmaped based write. So if we do with blocksize 1K
2124 * truncate(f, 1024);
2125 * a = mmap(f, 0, 4096);
2127 * truncate(f, 4096);
2128 * we have in the page first buffer_head mapped via page_mkwrite call back
2129 * but other buffer_heads would be unmapped but dirty (dirty done via the
2130 * do_wp_page). So writepage should write the first block. If we modify
2131 * the mmap area beyond 1024 we will again get a page_fault and the
2132 * page_mkwrite callback will do the block allocation and mark the
2133 * buffer_heads mapped.
2135 * We redirty the page if we have any buffer_heads that is either delay or
2136 * unwritten in the page.
2138 * We can get recursively called as show below.
2140 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2143 * But since we don't do any block allocation we should not deadlock.
2144 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2146 static int ext4_writepage(struct page *page,
2147 struct writeback_control *wbc)
2152 struct buffer_head *page_bufs = NULL;
2153 struct inode *inode = page->mapping->host;
2154 struct ext4_io_submit io_submit;
2155 bool keep_towrite = false;
2157 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2158 ext4_invalidatepage(page, 0, PAGE_SIZE);
2163 trace_ext4_writepage(page);
2164 size = i_size_read(inode);
2165 if (page->index == size >> PAGE_SHIFT)
2166 len = size & ~PAGE_MASK;
2170 page_bufs = page_buffers(page);
2172 * We cannot do block allocation or other extent handling in this
2173 * function. If there are buffers needing that, we have to redirty
2174 * the page. But we may reach here when we do a journal commit via
2175 * journal_submit_inode_data_buffers() and in that case we must write
2176 * allocated buffers to achieve data=ordered mode guarantees.
2178 * Also, if there is only one buffer per page (the fs block
2179 * size == the page size), if one buffer needs block
2180 * allocation or needs to modify the extent tree to clear the
2181 * unwritten flag, we know that the page can't be written at
2182 * all, so we might as well refuse the write immediately.
2183 * Unfortunately if the block size != page size, we can't as
2184 * easily detect this case using ext4_walk_page_buffers(), but
2185 * for the extremely common case, this is an optimization that
2186 * skips a useless round trip through ext4_bio_write_page().
2188 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2189 ext4_bh_delay_or_unwritten)) {
2190 redirty_page_for_writepage(wbc, page);
2191 if ((current->flags & PF_MEMALLOC) ||
2192 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2194 * For memory cleaning there's no point in writing only
2195 * some buffers. So just bail out. Warn if we came here
2196 * from direct reclaim.
2198 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2203 keep_towrite = true;
2206 if (PageChecked(page) && ext4_should_journal_data(inode))
2208 * It's mmapped pagecache. Add buffers and journal it. There
2209 * doesn't seem much point in redirtying the page here.
2211 return __ext4_journalled_writepage(page, len);
2213 ext4_io_submit_init(&io_submit, wbc);
2214 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2215 if (!io_submit.io_end) {
2216 redirty_page_for_writepage(wbc, page);
2220 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2221 ext4_io_submit(&io_submit);
2222 /* Drop io_end reference we got from init */
2223 ext4_put_io_end_defer(io_submit.io_end);
2227 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2233 BUG_ON(page->index != mpd->first_page);
2234 clear_page_dirty_for_io(page);
2236 * We have to be very careful here! Nothing protects writeback path
2237 * against i_size changes and the page can be writeably mapped into
2238 * page tables. So an application can be growing i_size and writing
2239 * data through mmap while writeback runs. clear_page_dirty_for_io()
2240 * write-protects our page in page tables and the page cannot get
2241 * written to again until we release page lock. So only after
2242 * clear_page_dirty_for_io() we are safe to sample i_size for
2243 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2244 * on the barrier provided by TestClearPageDirty in
2245 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2246 * after page tables are updated.
2248 size = i_size_read(mpd->inode);
2249 if (page->index == size >> PAGE_SHIFT)
2250 len = size & ~PAGE_MASK;
2253 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2255 mpd->wbc->nr_to_write--;
2261 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2264 * mballoc gives us at most this number of blocks...
2265 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2266 * The rest of mballoc seems to handle chunks up to full group size.
2268 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2271 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2273 * @mpd - extent of blocks
2274 * @lblk - logical number of the block in the file
2275 * @bh - buffer head we want to add to the extent
2277 * The function is used to collect contig. blocks in the same state. If the
2278 * buffer doesn't require mapping for writeback and we haven't started the
2279 * extent of buffers to map yet, the function returns 'true' immediately - the
2280 * caller can write the buffer right away. Otherwise the function returns true
2281 * if the block has been added to the extent, false if the block couldn't be
2284 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2285 struct buffer_head *bh)
2287 struct ext4_map_blocks *map = &mpd->map;
2289 /* Buffer that doesn't need mapping for writeback? */
2290 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2291 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2292 /* So far no extent to map => we write the buffer right away */
2293 if (map->m_len == 0)
2298 /* First block in the extent? */
2299 if (map->m_len == 0) {
2300 /* We cannot map unless handle is started... */
2305 map->m_flags = bh->b_state & BH_FLAGS;
2309 /* Don't go larger than mballoc is willing to allocate */
2310 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2313 /* Can we merge the block to our big extent? */
2314 if (lblk == map->m_lblk + map->m_len &&
2315 (bh->b_state & BH_FLAGS) == map->m_flags) {
2323 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2325 * @mpd - extent of blocks for mapping
2326 * @head - the first buffer in the page
2327 * @bh - buffer we should start processing from
2328 * @lblk - logical number of the block in the file corresponding to @bh
2330 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2331 * the page for IO if all buffers in this page were mapped and there's no
2332 * accumulated extent of buffers to map or add buffers in the page to the
2333 * extent of buffers to map. The function returns 1 if the caller can continue
2334 * by processing the next page, 0 if it should stop adding buffers to the
2335 * extent to map because we cannot extend it anymore. It can also return value
2336 * < 0 in case of error during IO submission.
2338 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2339 struct buffer_head *head,
2340 struct buffer_head *bh,
2343 struct inode *inode = mpd->inode;
2345 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2346 >> inode->i_blkbits;
2349 BUG_ON(buffer_locked(bh));
2351 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2352 /* Found extent to map? */
2355 /* Buffer needs mapping and handle is not started? */
2358 /* Everything mapped so far and we hit EOF */
2361 } while (lblk++, (bh = bh->b_this_page) != head);
2362 /* So far everything mapped? Submit the page for IO. */
2363 if (mpd->map.m_len == 0) {
2364 err = mpage_submit_page(mpd, head->b_page);
2368 return lblk < blocks;
2372 * mpage_map_buffers - update buffers corresponding to changed extent and
2373 * submit fully mapped pages for IO
2375 * @mpd - description of extent to map, on return next extent to map
2377 * Scan buffers corresponding to changed extent (we expect corresponding pages
2378 * to be already locked) and update buffer state according to new extent state.
2379 * We map delalloc buffers to their physical location, clear unwritten bits,
2380 * and mark buffers as uninit when we perform writes to unwritten extents
2381 * and do extent conversion after IO is finished. If the last page is not fully
2382 * mapped, we update @map to the next extent in the last page that needs
2383 * mapping. Otherwise we submit the page for IO.
2385 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2387 struct pagevec pvec;
2389 struct inode *inode = mpd->inode;
2390 struct buffer_head *head, *bh;
2391 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2397 start = mpd->map.m_lblk >> bpp_bits;
2398 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2399 lblk = start << bpp_bits;
2400 pblock = mpd->map.m_pblk;
2402 pagevec_init(&pvec);
2403 while (start <= end) {
2404 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2408 for (i = 0; i < nr_pages; i++) {
2409 struct page *page = pvec.pages[i];
2411 bh = head = page_buffers(page);
2413 if (lblk < mpd->map.m_lblk)
2415 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2417 * Buffer after end of mapped extent.
2418 * Find next buffer in the page to map.
2421 mpd->map.m_flags = 0;
2423 * FIXME: If dioread_nolock supports
2424 * blocksize < pagesize, we need to make
2425 * sure we add size mapped so far to
2426 * io_end->size as the following call
2427 * can submit the page for IO.
2429 err = mpage_process_page_bufs(mpd, head,
2431 pagevec_release(&pvec);
2436 if (buffer_delay(bh)) {
2437 clear_buffer_delay(bh);
2438 bh->b_blocknr = pblock++;
2440 clear_buffer_unwritten(bh);
2441 } while (lblk++, (bh = bh->b_this_page) != head);
2444 * FIXME: This is going to break if dioread_nolock
2445 * supports blocksize < pagesize as we will try to
2446 * convert potentially unmapped parts of inode.
2448 mpd->io_submit.io_end->size += PAGE_SIZE;
2449 /* Page fully mapped - let IO run! */
2450 err = mpage_submit_page(mpd, page);
2452 pagevec_release(&pvec);
2456 pagevec_release(&pvec);
2458 /* Extent fully mapped and matches with page boundary. We are done. */
2460 mpd->map.m_flags = 0;
2464 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2466 struct inode *inode = mpd->inode;
2467 struct ext4_map_blocks *map = &mpd->map;
2468 int get_blocks_flags;
2469 int err, dioread_nolock;
2471 trace_ext4_da_write_pages_extent(inode, map);
2473 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2474 * to convert an unwritten extent to be initialized (in the case
2475 * where we have written into one or more preallocated blocks). It is
2476 * possible that we're going to need more metadata blocks than
2477 * previously reserved. However we must not fail because we're in
2478 * writeback and there is nothing we can do about it so it might result
2479 * in data loss. So use reserved blocks to allocate metadata if
2482 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2483 * the blocks in question are delalloc blocks. This indicates
2484 * that the blocks and quotas has already been checked when
2485 * the data was copied into the page cache.
2487 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2488 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2489 EXT4_GET_BLOCKS_IO_SUBMIT;
2490 dioread_nolock = ext4_should_dioread_nolock(inode);
2492 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2493 if (map->m_flags & (1 << BH_Delay))
2494 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2496 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2499 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2500 if (!mpd->io_submit.io_end->handle &&
2501 ext4_handle_valid(handle)) {
2502 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2503 handle->h_rsv_handle = NULL;
2505 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2508 BUG_ON(map->m_len == 0);
2513 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2514 * mpd->len and submit pages underlying it for IO
2516 * @handle - handle for journal operations
2517 * @mpd - extent to map
2518 * @give_up_on_write - we set this to true iff there is a fatal error and there
2519 * is no hope of writing the data. The caller should discard
2520 * dirty pages to avoid infinite loops.
2522 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2523 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2524 * them to initialized or split the described range from larger unwritten
2525 * extent. Note that we need not map all the described range since allocation
2526 * can return less blocks or the range is covered by more unwritten extents. We
2527 * cannot map more because we are limited by reserved transaction credits. On
2528 * the other hand we always make sure that the last touched page is fully
2529 * mapped so that it can be written out (and thus forward progress is
2530 * guaranteed). After mapping we submit all mapped pages for IO.
2532 static int mpage_map_and_submit_extent(handle_t *handle,
2533 struct mpage_da_data *mpd,
2534 bool *give_up_on_write)
2536 struct inode *inode = mpd->inode;
2537 struct ext4_map_blocks *map = &mpd->map;
2542 mpd->io_submit.io_end->offset =
2543 ((loff_t)map->m_lblk) << inode->i_blkbits;
2545 err = mpage_map_one_extent(handle, mpd);
2547 struct super_block *sb = inode->i_sb;
2549 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2550 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2551 goto invalidate_dirty_pages;
2553 * Let the uper layers retry transient errors.
2554 * In the case of ENOSPC, if ext4_count_free_blocks()
2555 * is non-zero, a commit should free up blocks.
2557 if ((err == -ENOMEM) ||
2558 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2560 goto update_disksize;
2563 ext4_msg(sb, KERN_CRIT,
2564 "Delayed block allocation failed for "
2565 "inode %lu at logical offset %llu with"
2566 " max blocks %u with error %d",
2568 (unsigned long long)map->m_lblk,
2569 (unsigned)map->m_len, -err);
2570 ext4_msg(sb, KERN_CRIT,
2571 "This should not happen!! Data will "
2574 ext4_print_free_blocks(inode);
2575 invalidate_dirty_pages:
2576 *give_up_on_write = true;
2581 * Update buffer state, submit mapped pages, and get us new
2584 err = mpage_map_and_submit_buffers(mpd);
2586 goto update_disksize;
2587 } while (map->m_len);
2591 * Update on-disk size after IO is submitted. Races with
2592 * truncate are avoided by checking i_size under i_data_sem.
2594 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2595 if (disksize > EXT4_I(inode)->i_disksize) {
2599 down_write(&EXT4_I(inode)->i_data_sem);
2600 i_size = i_size_read(inode);
2601 if (disksize > i_size)
2603 if (disksize > EXT4_I(inode)->i_disksize)
2604 EXT4_I(inode)->i_disksize = disksize;
2605 up_write(&EXT4_I(inode)->i_data_sem);
2606 err2 = ext4_mark_inode_dirty(handle, inode);
2608 ext4_error(inode->i_sb,
2609 "Failed to mark inode %lu dirty",
2618 * Calculate the total number of credits to reserve for one writepages
2619 * iteration. This is called from ext4_writepages(). We map an extent of
2620 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2621 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2622 * bpp - 1 blocks in bpp different extents.
2624 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2626 int bpp = ext4_journal_blocks_per_page(inode);
2628 return ext4_meta_trans_blocks(inode,
2629 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2633 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2634 * and underlying extent to map
2636 * @mpd - where to look for pages
2638 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2639 * IO immediately. When we find a page which isn't mapped we start accumulating
2640 * extent of buffers underlying these pages that needs mapping (formed by
2641 * either delayed or unwritten buffers). We also lock the pages containing
2642 * these buffers. The extent found is returned in @mpd structure (starting at
2643 * mpd->lblk with length mpd->len blocks).
2645 * Note that this function can attach bios to one io_end structure which are
2646 * neither logically nor physically contiguous. Although it may seem as an
2647 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2648 * case as we need to track IO to all buffers underlying a page in one io_end.
2650 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2652 struct address_space *mapping = mpd->inode->i_mapping;
2653 struct pagevec pvec;
2654 unsigned int nr_pages;
2655 long left = mpd->wbc->nr_to_write;
2656 pgoff_t index = mpd->first_page;
2657 pgoff_t end = mpd->last_page;
2660 int blkbits = mpd->inode->i_blkbits;
2662 struct buffer_head *head;
2664 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2665 tag = PAGECACHE_TAG_TOWRITE;
2667 tag = PAGECACHE_TAG_DIRTY;
2669 pagevec_init(&pvec);
2671 mpd->next_page = index;
2672 while (index <= end) {
2673 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2678 for (i = 0; i < nr_pages; i++) {
2679 struct page *page = pvec.pages[i];
2682 * Accumulated enough dirty pages? This doesn't apply
2683 * to WB_SYNC_ALL mode. For integrity sync we have to
2684 * keep going because someone may be concurrently
2685 * dirtying pages, and we might have synced a lot of
2686 * newly appeared dirty pages, but have not synced all
2687 * of the old dirty pages.
2689 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2692 /* If we can't merge this page, we are done. */
2693 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2698 * If the page is no longer dirty, or its mapping no
2699 * longer corresponds to inode we are writing (which
2700 * means it has been truncated or invalidated), or the
2701 * page is already under writeback and we are not doing
2702 * a data integrity writeback, skip the page
2704 if (!PageDirty(page) ||
2705 (PageWriteback(page) &&
2706 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2707 unlikely(page->mapping != mapping)) {
2712 wait_on_page_writeback(page);
2713 BUG_ON(PageWriteback(page));
2715 if (mpd->map.m_len == 0)
2716 mpd->first_page = page->index;
2717 mpd->next_page = page->index + 1;
2718 /* Add all dirty buffers to mpd */
2719 lblk = ((ext4_lblk_t)page->index) <<
2720 (PAGE_SHIFT - blkbits);
2721 head = page_buffers(page);
2722 err = mpage_process_page_bufs(mpd, head, head, lblk);
2728 pagevec_release(&pvec);
2733 pagevec_release(&pvec);
2737 static int ext4_writepages(struct address_space *mapping,
2738 struct writeback_control *wbc)
2740 pgoff_t writeback_index = 0;
2741 long nr_to_write = wbc->nr_to_write;
2742 int range_whole = 0;
2744 handle_t *handle = NULL;
2745 struct mpage_da_data mpd;
2746 struct inode *inode = mapping->host;
2747 int needed_blocks, rsv_blocks = 0, ret = 0;
2748 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2750 struct blk_plug plug;
2751 bool give_up_on_write = false;
2753 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2756 percpu_down_read(&sbi->s_journal_flag_rwsem);
2757 trace_ext4_writepages(inode, wbc);
2760 * No pages to write? This is mainly a kludge to avoid starting
2761 * a transaction for special inodes like journal inode on last iput()
2762 * because that could violate lock ordering on umount
2764 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2765 goto out_writepages;
2767 if (ext4_should_journal_data(inode)) {
2768 ret = generic_writepages(mapping, wbc);
2769 goto out_writepages;
2773 * If the filesystem has aborted, it is read-only, so return
2774 * right away instead of dumping stack traces later on that
2775 * will obscure the real source of the problem. We test
2776 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2777 * the latter could be true if the filesystem is mounted
2778 * read-only, and in that case, ext4_writepages should
2779 * *never* be called, so if that ever happens, we would want
2782 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2783 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2785 goto out_writepages;
2788 if (ext4_should_dioread_nolock(inode)) {
2790 * We may need to convert up to one extent per block in
2791 * the page and we may dirty the inode.
2793 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2794 PAGE_SIZE >> inode->i_blkbits);
2798 * If we have inline data and arrive here, it means that
2799 * we will soon create the block for the 1st page, so
2800 * we'd better clear the inline data here.
2802 if (ext4_has_inline_data(inode)) {
2803 /* Just inode will be modified... */
2804 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2805 if (IS_ERR(handle)) {
2806 ret = PTR_ERR(handle);
2807 goto out_writepages;
2809 BUG_ON(ext4_test_inode_state(inode,
2810 EXT4_STATE_MAY_INLINE_DATA));
2811 ext4_destroy_inline_data(handle, inode);
2812 ext4_journal_stop(handle);
2815 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2818 if (wbc->range_cyclic) {
2819 writeback_index = mapping->writeback_index;
2820 if (writeback_index)
2822 mpd.first_page = writeback_index;
2825 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2826 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2831 ext4_io_submit_init(&mpd.io_submit, wbc);
2833 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2834 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2836 blk_start_plug(&plug);
2839 * First writeback pages that don't need mapping - we can avoid
2840 * starting a transaction unnecessarily and also avoid being blocked
2841 * in the block layer on device congestion while having transaction
2845 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2846 if (!mpd.io_submit.io_end) {
2850 ret = mpage_prepare_extent_to_map(&mpd);
2851 /* Unlock pages we didn't use */
2852 mpage_release_unused_pages(&mpd, false);
2853 /* Submit prepared bio */
2854 ext4_io_submit(&mpd.io_submit);
2855 ext4_put_io_end_defer(mpd.io_submit.io_end);
2856 mpd.io_submit.io_end = NULL;
2860 while (!done && mpd.first_page <= mpd.last_page) {
2861 /* For each extent of pages we use new io_end */
2862 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2863 if (!mpd.io_submit.io_end) {
2869 * We have two constraints: We find one extent to map and we
2870 * must always write out whole page (makes a difference when
2871 * blocksize < pagesize) so that we don't block on IO when we
2872 * try to write out the rest of the page. Journalled mode is
2873 * not supported by delalloc.
2875 BUG_ON(ext4_should_journal_data(inode));
2876 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2878 /* start a new transaction */
2879 handle = ext4_journal_start_with_reserve(inode,
2880 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2881 if (IS_ERR(handle)) {
2882 ret = PTR_ERR(handle);
2883 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2884 "%ld pages, ino %lu; err %d", __func__,
2885 wbc->nr_to_write, inode->i_ino, ret);
2886 /* Release allocated io_end */
2887 ext4_put_io_end(mpd.io_submit.io_end);
2888 mpd.io_submit.io_end = NULL;
2893 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2894 ret = mpage_prepare_extent_to_map(&mpd);
2897 ret = mpage_map_and_submit_extent(handle, &mpd,
2901 * We scanned the whole range (or exhausted
2902 * nr_to_write), submitted what was mapped and
2903 * didn't find anything needing mapping. We are
2910 * Caution: If the handle is synchronous,
2911 * ext4_journal_stop() can wait for transaction commit
2912 * to finish which may depend on writeback of pages to
2913 * complete or on page lock to be released. In that
2914 * case, we have to wait until after after we have
2915 * submitted all the IO, released page locks we hold,
2916 * and dropped io_end reference (for extent conversion
2917 * to be able to complete) before stopping the handle.
2919 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2920 ext4_journal_stop(handle);
2924 /* Unlock pages we didn't use */
2925 mpage_release_unused_pages(&mpd, give_up_on_write);
2926 /* Submit prepared bio */
2927 ext4_io_submit(&mpd.io_submit);
2930 * Drop our io_end reference we got from init. We have
2931 * to be careful and use deferred io_end finishing if
2932 * we are still holding the transaction as we can
2933 * release the last reference to io_end which may end
2934 * up doing unwritten extent conversion.
2937 ext4_put_io_end_defer(mpd.io_submit.io_end);
2938 ext4_journal_stop(handle);
2940 ext4_put_io_end(mpd.io_submit.io_end);
2941 mpd.io_submit.io_end = NULL;
2943 if (ret == -ENOSPC && sbi->s_journal) {
2945 * Commit the transaction which would
2946 * free blocks released in the transaction
2949 jbd2_journal_force_commit_nested(sbi->s_journal);
2953 /* Fatal error - ENOMEM, EIO... */
2958 blk_finish_plug(&plug);
2959 if (!ret && !cycled && wbc->nr_to_write > 0) {
2961 mpd.last_page = writeback_index - 1;
2967 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2969 * Set the writeback_index so that range_cyclic
2970 * mode will write it back later
2972 mapping->writeback_index = mpd.first_page;
2975 trace_ext4_writepages_result(inode, wbc, ret,
2976 nr_to_write - wbc->nr_to_write);
2977 percpu_up_read(&sbi->s_journal_flag_rwsem);
2981 static int ext4_dax_writepages(struct address_space *mapping,
2982 struct writeback_control *wbc)
2985 long nr_to_write = wbc->nr_to_write;
2986 struct inode *inode = mapping->host;
2987 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2989 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2992 percpu_down_read(&sbi->s_journal_flag_rwsem);
2993 trace_ext4_writepages(inode, wbc);
2995 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, wbc);
2996 trace_ext4_writepages_result(inode, wbc, ret,
2997 nr_to_write - wbc->nr_to_write);
2998 percpu_up_read(&sbi->s_journal_flag_rwsem);
3002 static int ext4_nonda_switch(struct super_block *sb)
3004 s64 free_clusters, dirty_clusters;
3005 struct ext4_sb_info *sbi = EXT4_SB(sb);
3008 * switch to non delalloc mode if we are running low
3009 * on free block. The free block accounting via percpu
3010 * counters can get slightly wrong with percpu_counter_batch getting
3011 * accumulated on each CPU without updating global counters
3012 * Delalloc need an accurate free block accounting. So switch
3013 * to non delalloc when we are near to error range.
3016 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
3018 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
3020 * Start pushing delalloc when 1/2 of free blocks are dirty.
3022 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
3023 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
3025 if (2 * free_clusters < 3 * dirty_clusters ||
3026 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
3028 * free block count is less than 150% of dirty blocks
3029 * or free blocks is less than watermark
3036 /* We always reserve for an inode update; the superblock could be there too */
3037 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
3039 if (likely(ext4_has_feature_large_file(inode->i_sb)))
3042 if (pos + len <= 0x7fffffffULL)
3045 /* We might need to update the superblock to set LARGE_FILE */
3049 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3050 loff_t pos, unsigned len, unsigned flags,
3051 struct page **pagep, void **fsdata)
3053 int ret, retries = 0;
3056 struct inode *inode = mapping->host;
3059 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3062 index = pos >> PAGE_SHIFT;
3064 if (ext4_nonda_switch(inode->i_sb) ||
3065 S_ISLNK(inode->i_mode)) {
3066 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3067 return ext4_write_begin(file, mapping, pos,
3068 len, flags, pagep, fsdata);
3070 *fsdata = (void *)0;
3071 trace_ext4_da_write_begin(inode, pos, len, flags);
3073 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3074 ret = ext4_da_write_inline_data_begin(mapping, inode,
3084 * grab_cache_page_write_begin() can take a long time if the
3085 * system is thrashing due to memory pressure, or if the page
3086 * is being written back. So grab it first before we start
3087 * the transaction handle. This also allows us to allocate
3088 * the page (if needed) without using GFP_NOFS.
3091 page = grab_cache_page_write_begin(mapping, index, flags);
3097 * With delayed allocation, we don't log the i_disksize update
3098 * if there is delayed block allocation. But we still need
3099 * to journalling the i_disksize update if writes to the end
3100 * of file which has an already mapped buffer.
3103 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3104 ext4_da_write_credits(inode, pos, len));
3105 if (IS_ERR(handle)) {
3107 return PTR_ERR(handle);
3111 if (page->mapping != mapping) {
3112 /* The page got truncated from under us */
3115 ext4_journal_stop(handle);
3118 /* In case writeback began while the page was unlocked */
3119 wait_for_stable_page(page);
3121 #ifdef CONFIG_FS_ENCRYPTION
3122 ret = ext4_block_write_begin(page, pos, len,
3123 ext4_da_get_block_prep);
3125 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3129 ext4_journal_stop(handle);
3131 * block_write_begin may have instantiated a few blocks
3132 * outside i_size. Trim these off again. Don't need
3133 * i_size_read because we hold i_mutex.
3135 if (pos + len > inode->i_size)
3136 ext4_truncate_failed_write(inode);
3138 if (ret == -ENOSPC &&
3139 ext4_should_retry_alloc(inode->i_sb, &retries))
3151 * Check if we should update i_disksize
3152 * when write to the end of file but not require block allocation
3154 static int ext4_da_should_update_i_disksize(struct page *page,
3155 unsigned long offset)
3157 struct buffer_head *bh;
3158 struct inode *inode = page->mapping->host;
3162 bh = page_buffers(page);
3163 idx = offset >> inode->i_blkbits;
3165 for (i = 0; i < idx; i++)
3166 bh = bh->b_this_page;
3168 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3173 static int ext4_da_write_end(struct file *file,
3174 struct address_space *mapping,
3175 loff_t pos, unsigned len, unsigned copied,
3176 struct page *page, void *fsdata)
3178 struct inode *inode = mapping->host;
3180 handle_t *handle = ext4_journal_current_handle();
3182 unsigned long start, end;
3183 int write_mode = (int)(unsigned long)fsdata;
3185 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3186 return ext4_write_end(file, mapping, pos,
3187 len, copied, page, fsdata);
3189 trace_ext4_da_write_end(inode, pos, len, copied);
3190 start = pos & (PAGE_SIZE - 1);
3191 end = start + copied - 1;
3194 * generic_write_end() will run mark_inode_dirty() if i_size
3195 * changes. So let's piggyback the i_disksize mark_inode_dirty
3198 new_i_size = pos + copied;
3199 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3200 if (ext4_has_inline_data(inode) ||
3201 ext4_da_should_update_i_disksize(page, end)) {
3202 ext4_update_i_disksize(inode, new_i_size);
3203 /* We need to mark inode dirty even if
3204 * new_i_size is less that inode->i_size
3205 * bu greater than i_disksize.(hint delalloc)
3207 ext4_mark_inode_dirty(handle, inode);
3211 if (write_mode != CONVERT_INLINE_DATA &&
3212 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3213 ext4_has_inline_data(inode))
3214 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3217 ret2 = generic_write_end(file, mapping, pos, len, copied,
3223 ret2 = ext4_journal_stop(handle);
3227 return ret ? ret : copied;
3230 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3231 unsigned int length)
3234 * Drop reserved blocks
3236 BUG_ON(!PageLocked(page));
3237 if (!page_has_buffers(page))
3240 ext4_da_page_release_reservation(page, offset, length);
3243 ext4_invalidatepage(page, offset, length);
3249 * Force all delayed allocation blocks to be allocated for a given inode.
3251 int ext4_alloc_da_blocks(struct inode *inode)
3253 trace_ext4_alloc_da_blocks(inode);
3255 if (!EXT4_I(inode)->i_reserved_data_blocks)
3259 * We do something simple for now. The filemap_flush() will
3260 * also start triggering a write of the data blocks, which is
3261 * not strictly speaking necessary (and for users of
3262 * laptop_mode, not even desirable). However, to do otherwise
3263 * would require replicating code paths in:
3265 * ext4_writepages() ->
3266 * write_cache_pages() ---> (via passed in callback function)
3267 * __mpage_da_writepage() -->
3268 * mpage_add_bh_to_extent()
3269 * mpage_da_map_blocks()
3271 * The problem is that write_cache_pages(), located in
3272 * mm/page-writeback.c, marks pages clean in preparation for
3273 * doing I/O, which is not desirable if we're not planning on
3276 * We could call write_cache_pages(), and then redirty all of
3277 * the pages by calling redirty_page_for_writepage() but that
3278 * would be ugly in the extreme. So instead we would need to
3279 * replicate parts of the code in the above functions,
3280 * simplifying them because we wouldn't actually intend to
3281 * write out the pages, but rather only collect contiguous
3282 * logical block extents, call the multi-block allocator, and
3283 * then update the buffer heads with the block allocations.
3285 * For now, though, we'll cheat by calling filemap_flush(),
3286 * which will map the blocks, and start the I/O, but not
3287 * actually wait for the I/O to complete.
3289 return filemap_flush(inode->i_mapping);
3293 * bmap() is special. It gets used by applications such as lilo and by
3294 * the swapper to find the on-disk block of a specific piece of data.
3296 * Naturally, this is dangerous if the block concerned is still in the
3297 * journal. If somebody makes a swapfile on an ext4 data-journaling
3298 * filesystem and enables swap, then they may get a nasty shock when the
3299 * data getting swapped to that swapfile suddenly gets overwritten by
3300 * the original zero's written out previously to the journal and
3301 * awaiting writeback in the kernel's buffer cache.
3303 * So, if we see any bmap calls here on a modified, data-journaled file,
3304 * take extra steps to flush any blocks which might be in the cache.
3306 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3308 struct inode *inode = mapping->host;
3313 * We can get here for an inline file via the FIBMAP ioctl
3315 if (ext4_has_inline_data(inode))
3318 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3319 test_opt(inode->i_sb, DELALLOC)) {
3321 * With delalloc we want to sync the file
3322 * so that we can make sure we allocate
3325 filemap_write_and_wait(mapping);
3328 if (EXT4_JOURNAL(inode) &&
3329 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3331 * This is a REALLY heavyweight approach, but the use of
3332 * bmap on dirty files is expected to be extremely rare:
3333 * only if we run lilo or swapon on a freshly made file
3334 * do we expect this to happen.
3336 * (bmap requires CAP_SYS_RAWIO so this does not
3337 * represent an unprivileged user DOS attack --- we'd be
3338 * in trouble if mortal users could trigger this path at
3341 * NB. EXT4_STATE_JDATA is not set on files other than
3342 * regular files. If somebody wants to bmap a directory
3343 * or symlink and gets confused because the buffer
3344 * hasn't yet been flushed to disk, they deserve
3345 * everything they get.
3348 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3349 journal = EXT4_JOURNAL(inode);
3350 jbd2_journal_lock_updates(journal);
3351 err = jbd2_journal_flush(journal);
3352 jbd2_journal_unlock_updates(journal);
3358 return generic_block_bmap(mapping, block, ext4_get_block);
3361 static int ext4_readpage(struct file *file, struct page *page)
3364 struct inode *inode = page->mapping->host;
3366 trace_ext4_readpage(page);
3368 if (ext4_has_inline_data(inode))
3369 ret = ext4_readpage_inline(inode, page);
3372 return ext4_mpage_readpages(page->mapping, NULL, page, 1,
3379 ext4_readpages(struct file *file, struct address_space *mapping,
3380 struct list_head *pages, unsigned nr_pages)
3382 struct inode *inode = mapping->host;
3384 /* If the file has inline data, no need to do readpages. */
3385 if (ext4_has_inline_data(inode))
3388 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages, true);
3391 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3392 unsigned int length)
3394 trace_ext4_invalidatepage(page, offset, length);
3396 /* No journalling happens on data buffers when this function is used */
3397 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3399 block_invalidatepage(page, offset, length);
3402 static int __ext4_journalled_invalidatepage(struct page *page,
3403 unsigned int offset,
3404 unsigned int length)
3406 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3408 trace_ext4_journalled_invalidatepage(page, offset, length);
3411 * If it's a full truncate we just forget about the pending dirtying
3413 if (offset == 0 && length == PAGE_SIZE)
3414 ClearPageChecked(page);
3416 return jbd2_journal_invalidatepage(journal, page, offset, length);
3419 /* Wrapper for aops... */
3420 static void ext4_journalled_invalidatepage(struct page *page,
3421 unsigned int offset,
3422 unsigned int length)
3424 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3427 static int ext4_releasepage(struct page *page, gfp_t wait)
3429 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3431 trace_ext4_releasepage(page);
3433 /* Page has dirty journalled data -> cannot release */
3434 if (PageChecked(page))
3437 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3439 return try_to_free_buffers(page);
3442 static bool ext4_inode_datasync_dirty(struct inode *inode)
3444 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3447 return !jbd2_transaction_committed(journal,
3448 EXT4_I(inode)->i_datasync_tid);
3449 /* Any metadata buffers to write? */
3450 if (!list_empty(&inode->i_mapping->private_list))
3452 return inode->i_state & I_DIRTY_DATASYNC;
3455 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3456 unsigned flags, struct iomap *iomap)
3458 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3459 unsigned int blkbits = inode->i_blkbits;
3460 unsigned long first_block, last_block;
3461 struct ext4_map_blocks map;
3462 bool delalloc = false;
3465 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3467 first_block = offset >> blkbits;
3468 last_block = min_t(loff_t, (offset + length - 1) >> blkbits,
3469 EXT4_MAX_LOGICAL_BLOCK);
3471 if (flags & IOMAP_REPORT) {
3472 if (ext4_has_inline_data(inode)) {
3473 ret = ext4_inline_data_iomap(inode, iomap);
3474 if (ret != -EAGAIN) {
3475 if (ret == 0 && offset >= iomap->length)
3481 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3485 map.m_lblk = first_block;
3486 map.m_len = last_block - first_block + 1;
3488 if (flags & IOMAP_REPORT) {
3489 ret = ext4_map_blocks(NULL, inode, &map, 0);
3494 ext4_lblk_t end = map.m_lblk + map.m_len - 1;
3495 struct extent_status es;
3497 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3498 map.m_lblk, end, &es);
3500 if (!es.es_len || es.es_lblk > end) {
3501 /* entire range is a hole */
3502 } else if (es.es_lblk > map.m_lblk) {
3503 /* range starts with a hole */
3504 map.m_len = es.es_lblk - map.m_lblk;
3506 ext4_lblk_t offs = 0;
3508 if (es.es_lblk < map.m_lblk)
3509 offs = map.m_lblk - es.es_lblk;
3510 map.m_lblk = es.es_lblk + offs;
3511 map.m_len = es.es_len - offs;
3515 } else if (flags & IOMAP_WRITE) {
3520 /* Trim mapping request to maximum we can map at once for DIO */
3521 if (map.m_len > DIO_MAX_BLOCKS)
3522 map.m_len = DIO_MAX_BLOCKS;
3523 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3526 * Either we allocate blocks and then we don't get unwritten
3527 * extent so we have reserved enough credits, or the blocks
3528 * are already allocated and unwritten and in that case
3529 * extent conversion fits in the credits as well.
3531 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3534 return PTR_ERR(handle);
3536 ret = ext4_map_blocks(handle, inode, &map,
3537 EXT4_GET_BLOCKS_CREATE_ZERO);
3539 ext4_journal_stop(handle);
3540 if (ret == -ENOSPC &&
3541 ext4_should_retry_alloc(inode->i_sb, &retries))
3547 * If we added blocks beyond i_size, we need to make sure they
3548 * will get truncated if we crash before updating i_size in
3549 * ext4_iomap_end(). For faults we don't need to do that (and
3550 * even cannot because for orphan list operations inode_lock is
3551 * required) - if we happen to instantiate block beyond i_size,
3552 * it is because we race with truncate which has already added
3553 * the inode to the orphan list.
3555 if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3556 (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3559 err = ext4_orphan_add(handle, inode);
3561 ext4_journal_stop(handle);
3565 ext4_journal_stop(handle);
3567 ret = ext4_map_blocks(NULL, inode, &map, 0);
3573 if (ext4_inode_datasync_dirty(inode))
3574 iomap->flags |= IOMAP_F_DIRTY;
3575 iomap->bdev = inode->i_sb->s_bdev;
3576 iomap->dax_dev = sbi->s_daxdev;
3577 iomap->offset = (u64)first_block << blkbits;
3578 iomap->length = (u64)map.m_len << blkbits;
3581 iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE;
3582 iomap->addr = IOMAP_NULL_ADDR;
3584 if (map.m_flags & EXT4_MAP_MAPPED) {
3585 iomap->type = IOMAP_MAPPED;
3586 } else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3587 iomap->type = IOMAP_UNWRITTEN;
3592 iomap->addr = (u64)map.m_pblk << blkbits;
3595 if (map.m_flags & EXT4_MAP_NEW)
3596 iomap->flags |= IOMAP_F_NEW;
3601 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3602 ssize_t written, unsigned flags, struct iomap *iomap)
3606 int blkbits = inode->i_blkbits;
3607 bool truncate = false;
3609 if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3612 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3613 if (IS_ERR(handle)) {
3614 ret = PTR_ERR(handle);
3617 if (ext4_update_inode_size(inode, offset + written))
3618 ext4_mark_inode_dirty(handle, inode);
3620 * We may need to truncate allocated but not written blocks beyond EOF.
3622 if (iomap->offset + iomap->length >
3623 ALIGN(inode->i_size, 1 << blkbits)) {
3624 ext4_lblk_t written_blk, end_blk;
3626 written_blk = (offset + written) >> blkbits;
3627 end_blk = (offset + length) >> blkbits;
3628 if (written_blk < end_blk && ext4_can_truncate(inode))
3632 * Remove inode from orphan list if we were extending a inode and
3633 * everything went fine.
3635 if (!truncate && inode->i_nlink &&
3636 !list_empty(&EXT4_I(inode)->i_orphan))
3637 ext4_orphan_del(handle, inode);
3638 ext4_journal_stop(handle);
3640 ext4_truncate_failed_write(inode);
3643 * If truncate failed early the inode might still be on the
3644 * orphan list; we need to make sure the inode is removed from
3645 * the orphan list in that case.
3648 ext4_orphan_del(NULL, inode);
3653 const struct iomap_ops ext4_iomap_ops = {
3654 .iomap_begin = ext4_iomap_begin,
3655 .iomap_end = ext4_iomap_end,
3658 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3659 ssize_t size, void *private)
3661 ext4_io_end_t *io_end = private;
3663 /* if not async direct IO just return */
3667 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3668 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3669 io_end, io_end->inode->i_ino, iocb, offset, size);
3672 * Error during AIO DIO. We cannot convert unwritten extents as the
3673 * data was not written. Just clear the unwritten flag and drop io_end.
3676 ext4_clear_io_unwritten_flag(io_end);
3679 io_end->offset = offset;
3680 io_end->size = size;
3681 ext4_put_io_end(io_end);
3687 * Handling of direct IO writes.
3689 * For ext4 extent files, ext4 will do direct-io write even to holes,
3690 * preallocated extents, and those write extend the file, no need to
3691 * fall back to buffered IO.
3693 * For holes, we fallocate those blocks, mark them as unwritten
3694 * If those blocks were preallocated, we mark sure they are split, but
3695 * still keep the range to write as unwritten.
3697 * The unwritten extents will be converted to written when DIO is completed.
3698 * For async direct IO, since the IO may still pending when return, we
3699 * set up an end_io call back function, which will do the conversion
3700 * when async direct IO completed.
3702 * If the O_DIRECT write will extend the file then add this inode to the
3703 * orphan list. So recovery will truncate it back to the original size
3704 * if the machine crashes during the write.
3707 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3709 struct file *file = iocb->ki_filp;
3710 struct inode *inode = file->f_mapping->host;
3711 struct ext4_inode_info *ei = EXT4_I(inode);
3713 loff_t offset = iocb->ki_pos;
3714 size_t count = iov_iter_count(iter);
3716 get_block_t *get_block_func = NULL;
3718 loff_t final_size = offset + count;
3722 if (final_size > inode->i_size || final_size > ei->i_disksize) {
3723 /* Credits for sb + inode write */
3724 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3725 if (IS_ERR(handle)) {
3726 ret = PTR_ERR(handle);
3729 ret = ext4_orphan_add(handle, inode);
3731 ext4_journal_stop(handle);
3735 ext4_update_i_disksize(inode, inode->i_size);
3736 ext4_journal_stop(handle);
3739 BUG_ON(iocb->private == NULL);
3742 * Make all waiters for direct IO properly wait also for extent
3743 * conversion. This also disallows race between truncate() and
3744 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3746 inode_dio_begin(inode);
3748 /* If we do a overwrite dio, i_mutex locking can be released */
3749 overwrite = *((int *)iocb->private);
3752 inode_unlock(inode);
3755 * For extent mapped files we could direct write to holes and fallocate.
3757 * Allocated blocks to fill the hole are marked as unwritten to prevent
3758 * parallel buffered read to expose the stale data before DIO complete
3761 * As to previously fallocated extents, ext4 get_block will just simply
3762 * mark the buffer mapped but still keep the extents unwritten.
3764 * For non AIO case, we will convert those unwritten extents to written
3765 * after return back from blockdev_direct_IO. That way we save us from
3766 * allocating io_end structure and also the overhead of offloading
3767 * the extent convertion to a workqueue.
3769 * For async DIO, the conversion needs to be deferred when the
3770 * IO is completed. The ext4 end_io callback function will be
3771 * called to take care of the conversion work. Here for async
3772 * case, we allocate an io_end structure to hook to the iocb.
3774 iocb->private = NULL;
3776 get_block_func = ext4_dio_get_block_overwrite;
3777 else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3778 round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3779 get_block_func = ext4_dio_get_block;
3780 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3781 } else if (is_sync_kiocb(iocb)) {
3782 get_block_func = ext4_dio_get_block_unwritten_sync;
3783 dio_flags = DIO_LOCKING;
3785 get_block_func = ext4_dio_get_block_unwritten_async;
3786 dio_flags = DIO_LOCKING;
3788 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3789 get_block_func, ext4_end_io_dio, NULL,
3792 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3793 EXT4_STATE_DIO_UNWRITTEN)) {
3796 * for non AIO case, since the IO is already
3797 * completed, we could do the conversion right here
3799 err = ext4_convert_unwritten_extents(NULL, inode,
3803 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3806 inode_dio_end(inode);
3807 /* take i_mutex locking again if we do a ovewrite dio */
3811 if (ret < 0 && final_size > inode->i_size)
3812 ext4_truncate_failed_write(inode);
3814 /* Handle extending of i_size after direct IO write */
3818 /* Credits for sb + inode write */
3819 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3820 if (IS_ERR(handle)) {
3822 * We wrote the data but cannot extend
3823 * i_size. Bail out. In async io case, we do
3824 * not return error here because we have
3825 * already submmitted the corresponding
3826 * bio. Returning error here makes the caller
3827 * think that this IO is done and failed
3828 * resulting in race with bio's completion
3832 ret = PTR_ERR(handle);
3834 ext4_orphan_del(NULL, inode);
3839 ext4_orphan_del(handle, inode);
3841 loff_t end = offset + ret;
3842 if (end > inode->i_size || end > ei->i_disksize) {
3843 ext4_update_i_disksize(inode, end);
3844 if (end > inode->i_size)
3845 i_size_write(inode, end);
3847 * We're going to return a positive `ret'
3848 * here due to non-zero-length I/O, so there's
3849 * no way of reporting error returns from
3850 * ext4_mark_inode_dirty() to userspace. So
3853 ext4_mark_inode_dirty(handle, inode);
3856 err = ext4_journal_stop(handle);
3864 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3866 struct address_space *mapping = iocb->ki_filp->f_mapping;
3867 struct inode *inode = mapping->host;
3868 size_t count = iov_iter_count(iter);
3872 * Shared inode_lock is enough for us - it protects against concurrent
3873 * writes & truncates and since we take care of writing back page cache,
3874 * we are protected against page writeback as well.
3876 inode_lock_shared(inode);
3877 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3878 iocb->ki_pos + count - 1);
3881 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3882 iter, ext4_dio_get_block, NULL, NULL, 0);
3884 inode_unlock_shared(inode);
3888 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3890 struct file *file = iocb->ki_filp;
3891 struct inode *inode = file->f_mapping->host;
3892 size_t count = iov_iter_count(iter);
3893 loff_t offset = iocb->ki_pos;
3896 #ifdef CONFIG_FS_ENCRYPTION
3897 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
3902 * If we are doing data journalling we don't support O_DIRECT
3904 if (ext4_should_journal_data(inode))
3907 /* Let buffer I/O handle the inline data case. */
3908 if (ext4_has_inline_data(inode))
3911 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3912 if (iov_iter_rw(iter) == READ)
3913 ret = ext4_direct_IO_read(iocb, iter);
3915 ret = ext4_direct_IO_write(iocb, iter);
3916 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3921 * Pages can be marked dirty completely asynchronously from ext4's journalling
3922 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3923 * much here because ->set_page_dirty is called under VFS locks. The page is
3924 * not necessarily locked.
3926 * We cannot just dirty the page and leave attached buffers clean, because the
3927 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3928 * or jbddirty because all the journalling code will explode.
3930 * So what we do is to mark the page "pending dirty" and next time writepage
3931 * is called, propagate that into the buffers appropriately.
3933 static int ext4_journalled_set_page_dirty(struct page *page)
3935 SetPageChecked(page);
3936 return __set_page_dirty_nobuffers(page);
3939 static int ext4_set_page_dirty(struct page *page)
3941 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3942 WARN_ON_ONCE(!page_has_buffers(page));
3943 return __set_page_dirty_buffers(page);
3946 static const struct address_space_operations ext4_aops = {
3947 .readpage = ext4_readpage,
3948 .readpages = ext4_readpages,
3949 .writepage = ext4_writepage,
3950 .writepages = ext4_writepages,
3951 .write_begin = ext4_write_begin,
3952 .write_end = ext4_write_end,
3953 .set_page_dirty = ext4_set_page_dirty,
3955 .invalidatepage = ext4_invalidatepage,
3956 .releasepage = ext4_releasepage,
3957 .direct_IO = ext4_direct_IO,
3958 .migratepage = buffer_migrate_page,
3959 .is_partially_uptodate = block_is_partially_uptodate,
3960 .error_remove_page = generic_error_remove_page,
3963 static const struct address_space_operations ext4_journalled_aops = {
3964 .readpage = ext4_readpage,
3965 .readpages = ext4_readpages,
3966 .writepage = ext4_writepage,
3967 .writepages = ext4_writepages,
3968 .write_begin = ext4_write_begin,
3969 .write_end = ext4_journalled_write_end,
3970 .set_page_dirty = ext4_journalled_set_page_dirty,
3972 .invalidatepage = ext4_journalled_invalidatepage,
3973 .releasepage = ext4_releasepage,
3974 .direct_IO = ext4_direct_IO,
3975 .is_partially_uptodate = block_is_partially_uptodate,
3976 .error_remove_page = generic_error_remove_page,
3979 static const struct address_space_operations ext4_da_aops = {
3980 .readpage = ext4_readpage,
3981 .readpages = ext4_readpages,
3982 .writepage = ext4_writepage,
3983 .writepages = ext4_writepages,
3984 .write_begin = ext4_da_write_begin,
3985 .write_end = ext4_da_write_end,
3986 .set_page_dirty = ext4_set_page_dirty,
3988 .invalidatepage = ext4_da_invalidatepage,
3989 .releasepage = ext4_releasepage,
3990 .direct_IO = ext4_direct_IO,
3991 .migratepage = buffer_migrate_page,
3992 .is_partially_uptodate = block_is_partially_uptodate,
3993 .error_remove_page = generic_error_remove_page,
3996 static const struct address_space_operations ext4_dax_aops = {
3997 .writepages = ext4_dax_writepages,
3998 .direct_IO = noop_direct_IO,
3999 .set_page_dirty = noop_set_page_dirty,
4001 .invalidatepage = noop_invalidatepage,
4004 void ext4_set_aops(struct inode *inode)
4006 switch (ext4_inode_journal_mode(inode)) {
4007 case EXT4_INODE_ORDERED_DATA_MODE:
4008 case EXT4_INODE_WRITEBACK_DATA_MODE:
4010 case EXT4_INODE_JOURNAL_DATA_MODE:
4011 inode->i_mapping->a_ops = &ext4_journalled_aops;
4017 inode->i_mapping->a_ops = &ext4_dax_aops;
4018 else if (test_opt(inode->i_sb, DELALLOC))
4019 inode->i_mapping->a_ops = &ext4_da_aops;
4021 inode->i_mapping->a_ops = &ext4_aops;
4024 static int __ext4_block_zero_page_range(handle_t *handle,
4025 struct address_space *mapping, loff_t from, loff_t length)
4027 ext4_fsblk_t index = from >> PAGE_SHIFT;
4028 unsigned offset = from & (PAGE_SIZE-1);
4029 unsigned blocksize, pos;
4031 struct inode *inode = mapping->host;
4032 struct buffer_head *bh;
4036 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
4037 mapping_gfp_constraint(mapping, ~__GFP_FS));
4041 blocksize = inode->i_sb->s_blocksize;
4043 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
4045 if (!page_has_buffers(page))
4046 create_empty_buffers(page, blocksize, 0);
4048 /* Find the buffer that contains "offset" */
4049 bh = page_buffers(page);
4051 while (offset >= pos) {
4052 bh = bh->b_this_page;
4056 if (buffer_freed(bh)) {
4057 BUFFER_TRACE(bh, "freed: skip");
4060 if (!buffer_mapped(bh)) {
4061 BUFFER_TRACE(bh, "unmapped");
4062 ext4_get_block(inode, iblock, bh, 0);
4063 /* unmapped? It's a hole - nothing to do */
4064 if (!buffer_mapped(bh)) {
4065 BUFFER_TRACE(bh, "still unmapped");
4070 /* Ok, it's mapped. Make sure it's up-to-date */
4071 if (PageUptodate(page))
4072 set_buffer_uptodate(bh);
4074 if (!buffer_uptodate(bh)) {
4076 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
4078 /* Uhhuh. Read error. Complain and punt. */
4079 if (!buffer_uptodate(bh))
4081 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
4082 /* We expect the key to be set. */
4083 BUG_ON(!fscrypt_has_encryption_key(inode));
4084 WARN_ON_ONCE(fscrypt_decrypt_pagecache_blocks(
4085 page, blocksize, bh_offset(bh)));
4088 if (ext4_should_journal_data(inode)) {
4089 BUFFER_TRACE(bh, "get write access");
4090 err = ext4_journal_get_write_access(handle, bh);
4094 zero_user(page, offset, length);
4095 BUFFER_TRACE(bh, "zeroed end of block");
4097 if (ext4_should_journal_data(inode)) {
4098 err = ext4_handle_dirty_metadata(handle, inode, bh);
4101 mark_buffer_dirty(bh);
4102 if (ext4_should_order_data(inode))
4103 err = ext4_jbd2_inode_add_write(handle, inode, from,
4114 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4115 * starting from file offset 'from'. The range to be zero'd must
4116 * be contained with in one block. If the specified range exceeds
4117 * the end of the block it will be shortened to end of the block
4118 * that cooresponds to 'from'
4120 static int ext4_block_zero_page_range(handle_t *handle,
4121 struct address_space *mapping, loff_t from, loff_t length)
4123 struct inode *inode = mapping->host;
4124 unsigned offset = from & (PAGE_SIZE-1);
4125 unsigned blocksize = inode->i_sb->s_blocksize;
4126 unsigned max = blocksize - (offset & (blocksize - 1));
4129 * correct length if it does not fall between
4130 * 'from' and the end of the block
4132 if (length > max || length < 0)
4135 if (IS_DAX(inode)) {
4136 return iomap_zero_range(inode, from, length, NULL,
4139 return __ext4_block_zero_page_range(handle, mapping, from, length);
4143 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4144 * up to the end of the block which corresponds to `from'.
4145 * This required during truncate. We need to physically zero the tail end
4146 * of that block so it doesn't yield old data if the file is later grown.
4148 static int ext4_block_truncate_page(handle_t *handle,
4149 struct address_space *mapping, loff_t from)
4151 unsigned offset = from & (PAGE_SIZE-1);
4154 struct inode *inode = mapping->host;
4156 /* If we are processing an encrypted inode during orphan list handling */
4157 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
4160 blocksize = inode->i_sb->s_blocksize;
4161 length = blocksize - (offset & (blocksize - 1));
4163 return ext4_block_zero_page_range(handle, mapping, from, length);
4166 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4167 loff_t lstart, loff_t length)
4169 struct super_block *sb = inode->i_sb;
4170 struct address_space *mapping = inode->i_mapping;
4171 unsigned partial_start, partial_end;
4172 ext4_fsblk_t start, end;
4173 loff_t byte_end = (lstart + length - 1);
4176 partial_start = lstart & (sb->s_blocksize - 1);
4177 partial_end = byte_end & (sb->s_blocksize - 1);
4179 start = lstart >> sb->s_blocksize_bits;
4180 end = byte_end >> sb->s_blocksize_bits;
4182 /* Handle partial zero within the single block */
4184 (partial_start || (partial_end != sb->s_blocksize - 1))) {
4185 err = ext4_block_zero_page_range(handle, mapping,
4189 /* Handle partial zero out on the start of the range */
4190 if (partial_start) {
4191 err = ext4_block_zero_page_range(handle, mapping,
4192 lstart, sb->s_blocksize);
4196 /* Handle partial zero out on the end of the range */
4197 if (partial_end != sb->s_blocksize - 1)
4198 err = ext4_block_zero_page_range(handle, mapping,
4199 byte_end - partial_end,
4204 int ext4_can_truncate(struct inode *inode)
4206 if (S_ISREG(inode->i_mode))
4208 if (S_ISDIR(inode->i_mode))
4210 if (S_ISLNK(inode->i_mode))
4211 return !ext4_inode_is_fast_symlink(inode);
4216 * We have to make sure i_disksize gets properly updated before we truncate
4217 * page cache due to hole punching or zero range. Otherwise i_disksize update
4218 * can get lost as it may have been postponed to submission of writeback but
4219 * that will never happen after we truncate page cache.
4221 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4225 loff_t size = i_size_read(inode);
4227 WARN_ON(!inode_is_locked(inode));
4228 if (offset > size || offset + len < size)
4231 if (EXT4_I(inode)->i_disksize >= size)
4234 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4236 return PTR_ERR(handle);
4237 ext4_update_i_disksize(inode, size);
4238 ext4_mark_inode_dirty(handle, inode);
4239 ext4_journal_stop(handle);
4244 static void ext4_wait_dax_page(struct ext4_inode_info *ei)
4246 up_write(&ei->i_mmap_sem);
4248 down_write(&ei->i_mmap_sem);
4251 int ext4_break_layouts(struct inode *inode)
4253 struct ext4_inode_info *ei = EXT4_I(inode);
4257 if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
4261 page = dax_layout_busy_page(inode->i_mapping);
4265 error = ___wait_var_event(&page->_refcount,
4266 atomic_read(&page->_refcount) == 1,
4267 TASK_INTERRUPTIBLE, 0, 0,
4268 ext4_wait_dax_page(ei));
4269 } while (error == 0);
4275 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4276 * associated with the given offset and length
4278 * @inode: File inode
4279 * @offset: The offset where the hole will begin
4280 * @len: The length of the hole
4282 * Returns: 0 on success or negative on failure
4285 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4287 struct super_block *sb = inode->i_sb;
4288 ext4_lblk_t first_block, stop_block;
4289 struct address_space *mapping = inode->i_mapping;
4290 loff_t first_block_offset, last_block_offset;
4292 unsigned int credits;
4295 if (!S_ISREG(inode->i_mode))
4298 trace_ext4_punch_hole(inode, offset, length, 0);
4301 * Write out all dirty pages to avoid race conditions
4302 * Then release them.
4304 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4305 ret = filemap_write_and_wait_range(mapping, offset,
4306 offset + length - 1);
4313 /* No need to punch hole beyond i_size */
4314 if (offset >= inode->i_size)
4318 * If the hole extends beyond i_size, set the hole
4319 * to end after the page that contains i_size
4321 if (offset + length > inode->i_size) {
4322 length = inode->i_size +
4323 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4327 if (offset & (sb->s_blocksize - 1) ||
4328 (offset + length) & (sb->s_blocksize - 1)) {
4330 * Attach jinode to inode for jbd2 if we do any zeroing of
4333 ret = ext4_inode_attach_jinode(inode);
4339 /* Wait all existing dio workers, newcomers will block on i_mutex */
4340 inode_dio_wait(inode);
4343 * Prevent page faults from reinstantiating pages we have released from
4346 down_write(&EXT4_I(inode)->i_mmap_sem);
4348 ret = ext4_break_layouts(inode);
4352 first_block_offset = round_up(offset, sb->s_blocksize);
4353 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4355 /* Now release the pages and zero block aligned part of pages*/
4356 if (last_block_offset > first_block_offset) {
4357 ret = ext4_update_disksize_before_punch(inode, offset, length);
4360 truncate_pagecache_range(inode, first_block_offset,
4364 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4365 credits = ext4_writepage_trans_blocks(inode);
4367 credits = ext4_blocks_for_truncate(inode);
4368 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4369 if (IS_ERR(handle)) {
4370 ret = PTR_ERR(handle);
4371 ext4_std_error(sb, ret);
4375 ret = ext4_zero_partial_blocks(handle, inode, offset,
4380 first_block = (offset + sb->s_blocksize - 1) >>
4381 EXT4_BLOCK_SIZE_BITS(sb);
4382 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4384 /* If there are blocks to remove, do it */
4385 if (stop_block > first_block) {
4387 down_write(&EXT4_I(inode)->i_data_sem);
4388 ext4_discard_preallocations(inode);
4390 ret = ext4_es_remove_extent(inode, first_block,
4391 stop_block - first_block);
4393 up_write(&EXT4_I(inode)->i_data_sem);
4397 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4398 ret = ext4_ext_remove_space(inode, first_block,
4401 ret = ext4_ind_remove_space(handle, inode, first_block,
4404 up_write(&EXT4_I(inode)->i_data_sem);
4407 ext4_handle_sync(handle);
4409 inode->i_mtime = inode->i_ctime = current_time(inode);
4410 ext4_mark_inode_dirty(handle, inode);
4412 ext4_update_inode_fsync_trans(handle, inode, 1);
4414 ext4_journal_stop(handle);
4416 up_write(&EXT4_I(inode)->i_mmap_sem);
4418 inode_unlock(inode);
4422 int ext4_inode_attach_jinode(struct inode *inode)
4424 struct ext4_inode_info *ei = EXT4_I(inode);
4425 struct jbd2_inode *jinode;
4427 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4430 jinode = jbd2_alloc_inode(GFP_KERNEL);
4431 spin_lock(&inode->i_lock);
4434 spin_unlock(&inode->i_lock);
4437 ei->jinode = jinode;
4438 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4441 spin_unlock(&inode->i_lock);
4442 if (unlikely(jinode != NULL))
4443 jbd2_free_inode(jinode);
4450 * We block out ext4_get_block() block instantiations across the entire
4451 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4452 * simultaneously on behalf of the same inode.
4454 * As we work through the truncate and commit bits of it to the journal there
4455 * is one core, guiding principle: the file's tree must always be consistent on
4456 * disk. We must be able to restart the truncate after a crash.
4458 * The file's tree may be transiently inconsistent in memory (although it
4459 * probably isn't), but whenever we close off and commit a journal transaction,
4460 * the contents of (the filesystem + the journal) must be consistent and
4461 * restartable. It's pretty simple, really: bottom up, right to left (although
4462 * left-to-right works OK too).
4464 * Note that at recovery time, journal replay occurs *before* the restart of
4465 * truncate against the orphan inode list.
4467 * The committed inode has the new, desired i_size (which is the same as
4468 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4469 * that this inode's truncate did not complete and it will again call
4470 * ext4_truncate() to have another go. So there will be instantiated blocks
4471 * to the right of the truncation point in a crashed ext4 filesystem. But
4472 * that's fine - as long as they are linked from the inode, the post-crash
4473 * ext4_truncate() run will find them and release them.
4475 int ext4_truncate(struct inode *inode)
4477 struct ext4_inode_info *ei = EXT4_I(inode);
4478 unsigned int credits;
4481 struct address_space *mapping = inode->i_mapping;
4484 * There is a possibility that we're either freeing the inode
4485 * or it's a completely new inode. In those cases we might not
4486 * have i_mutex locked because it's not necessary.
4488 if (!(inode->i_state & (I_NEW|I_FREEING)))
4489 WARN_ON(!inode_is_locked(inode));
4490 trace_ext4_truncate_enter(inode);
4492 if (!ext4_can_truncate(inode))
4495 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4497 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4498 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4500 if (ext4_has_inline_data(inode)) {
4503 err = ext4_inline_data_truncate(inode, &has_inline);
4510 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4511 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4512 if (ext4_inode_attach_jinode(inode) < 0)
4516 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4517 credits = ext4_writepage_trans_blocks(inode);
4519 credits = ext4_blocks_for_truncate(inode);
4521 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4523 return PTR_ERR(handle);
4525 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4526 ext4_block_truncate_page(handle, mapping, inode->i_size);
4529 * We add the inode to the orphan list, so that if this
4530 * truncate spans multiple transactions, and we crash, we will
4531 * resume the truncate when the filesystem recovers. It also
4532 * marks the inode dirty, to catch the new size.
4534 * Implication: the file must always be in a sane, consistent
4535 * truncatable state while each transaction commits.
4537 err = ext4_orphan_add(handle, inode);
4541 down_write(&EXT4_I(inode)->i_data_sem);
4543 ext4_discard_preallocations(inode);
4545 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4546 err = ext4_ext_truncate(handle, inode);
4548 ext4_ind_truncate(handle, inode);
4550 up_write(&ei->i_data_sem);
4555 ext4_handle_sync(handle);
4559 * If this was a simple ftruncate() and the file will remain alive,
4560 * then we need to clear up the orphan record which we created above.
4561 * However, if this was a real unlink then we were called by
4562 * ext4_evict_inode(), and we allow that function to clean up the
4563 * orphan info for us.
4566 ext4_orphan_del(handle, inode);
4568 inode->i_mtime = inode->i_ctime = current_time(inode);
4569 ext4_mark_inode_dirty(handle, inode);
4570 ext4_journal_stop(handle);
4572 trace_ext4_truncate_exit(inode);
4577 * ext4_get_inode_loc returns with an extra refcount against the inode's
4578 * underlying buffer_head on success. If 'in_mem' is true, we have all
4579 * data in memory that is needed to recreate the on-disk version of this
4582 static int __ext4_get_inode_loc(struct inode *inode,
4583 struct ext4_iloc *iloc, int in_mem)
4585 struct ext4_group_desc *gdp;
4586 struct buffer_head *bh;
4587 struct super_block *sb = inode->i_sb;
4589 int inodes_per_block, inode_offset;
4592 if (inode->i_ino < EXT4_ROOT_INO ||
4593 inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4594 return -EFSCORRUPTED;
4596 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4597 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4602 * Figure out the offset within the block group inode table
4604 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4605 inode_offset = ((inode->i_ino - 1) %
4606 EXT4_INODES_PER_GROUP(sb));
4607 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4608 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4610 bh = sb_getblk(sb, block);
4613 if (!buffer_uptodate(bh)) {
4617 * If the buffer has the write error flag, we have failed
4618 * to write out another inode in the same block. In this
4619 * case, we don't have to read the block because we may
4620 * read the old inode data successfully.
4622 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4623 set_buffer_uptodate(bh);
4625 if (buffer_uptodate(bh)) {
4626 /* someone brought it uptodate while we waited */
4632 * If we have all information of the inode in memory and this
4633 * is the only valid inode in the block, we need not read the
4637 struct buffer_head *bitmap_bh;
4640 start = inode_offset & ~(inodes_per_block - 1);
4642 /* Is the inode bitmap in cache? */
4643 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4644 if (unlikely(!bitmap_bh))
4648 * If the inode bitmap isn't in cache then the
4649 * optimisation may end up performing two reads instead
4650 * of one, so skip it.
4652 if (!buffer_uptodate(bitmap_bh)) {
4656 for (i = start; i < start + inodes_per_block; i++) {
4657 if (i == inode_offset)
4659 if (ext4_test_bit(i, bitmap_bh->b_data))
4663 if (i == start + inodes_per_block) {
4664 /* all other inodes are free, so skip I/O */
4665 memset(bh->b_data, 0, bh->b_size);
4666 set_buffer_uptodate(bh);
4674 * If we need to do any I/O, try to pre-readahead extra
4675 * blocks from the inode table.
4677 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4678 ext4_fsblk_t b, end, table;
4680 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4682 table = ext4_inode_table(sb, gdp);
4683 /* s_inode_readahead_blks is always a power of 2 */
4684 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4688 num = EXT4_INODES_PER_GROUP(sb);
4689 if (ext4_has_group_desc_csum(sb))
4690 num -= ext4_itable_unused_count(sb, gdp);
4691 table += num / inodes_per_block;
4695 sb_breadahead(sb, b++);
4699 * There are other valid inodes in the buffer, this inode
4700 * has in-inode xattrs, or we don't have this inode in memory.
4701 * Read the block from disk.
4703 trace_ext4_load_inode(inode);
4705 bh->b_end_io = end_buffer_read_sync;
4706 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4708 if (!buffer_uptodate(bh)) {
4709 EXT4_ERROR_INODE_BLOCK(inode, block,
4710 "unable to read itable block");
4720 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4722 /* We have all inode data except xattrs in memory here. */
4723 return __ext4_get_inode_loc(inode, iloc,
4724 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4727 static bool ext4_should_use_dax(struct inode *inode)
4729 if (!test_opt(inode->i_sb, DAX))
4731 if (!S_ISREG(inode->i_mode))
4733 if (ext4_should_journal_data(inode))
4735 if (ext4_has_inline_data(inode))
4737 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4742 void ext4_set_inode_flags(struct inode *inode)
4744 unsigned int flags = EXT4_I(inode)->i_flags;
4745 unsigned int new_fl = 0;
4747 if (flags & EXT4_SYNC_FL)
4749 if (flags & EXT4_APPEND_FL)
4751 if (flags & EXT4_IMMUTABLE_FL)
4752 new_fl |= S_IMMUTABLE;
4753 if (flags & EXT4_NOATIME_FL)
4754 new_fl |= S_NOATIME;
4755 if (flags & EXT4_DIRSYNC_FL)
4756 new_fl |= S_DIRSYNC;
4757 if (ext4_should_use_dax(inode))
4759 if (flags & EXT4_ENCRYPT_FL)
4760 new_fl |= S_ENCRYPTED;
4761 if (flags & EXT4_CASEFOLD_FL)
4762 new_fl |= S_CASEFOLD;
4763 inode_set_flags(inode, new_fl,
4764 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4765 S_ENCRYPTED|S_CASEFOLD);
4768 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4769 struct ext4_inode_info *ei)
4772 struct inode *inode = &(ei->vfs_inode);
4773 struct super_block *sb = inode->i_sb;
4775 if (ext4_has_feature_huge_file(sb)) {
4776 /* we are using combined 48 bit field */
4777 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4778 le32_to_cpu(raw_inode->i_blocks_lo);
4779 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4780 /* i_blocks represent file system block size */
4781 return i_blocks << (inode->i_blkbits - 9);
4786 return le32_to_cpu(raw_inode->i_blocks_lo);
4790 static inline int ext4_iget_extra_inode(struct inode *inode,
4791 struct ext4_inode *raw_inode,
4792 struct ext4_inode_info *ei)
4794 __le32 *magic = (void *)raw_inode +
4795 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4797 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4798 EXT4_INODE_SIZE(inode->i_sb) &&
4799 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4800 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4801 return ext4_find_inline_data_nolock(inode);
4803 EXT4_I(inode)->i_inline_off = 0;
4807 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4809 if (!ext4_has_feature_project(inode->i_sb))
4811 *projid = EXT4_I(inode)->i_projid;
4816 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4817 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4820 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4822 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4823 inode_set_iversion_raw(inode, val);
4825 inode_set_iversion_queried(inode, val);
4827 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4829 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4830 return inode_peek_iversion_raw(inode);
4832 return inode_peek_iversion(inode);
4835 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4836 ext4_iget_flags flags, const char *function,
4839 struct ext4_iloc iloc;
4840 struct ext4_inode *raw_inode;
4841 struct ext4_inode_info *ei;
4842 struct inode *inode;
4843 journal_t *journal = EXT4_SB(sb)->s_journal;
4851 if ((!(flags & EXT4_IGET_SPECIAL) &&
4852 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4853 (ino < EXT4_ROOT_INO) ||
4854 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4855 if (flags & EXT4_IGET_HANDLE)
4856 return ERR_PTR(-ESTALE);
4857 __ext4_error(sb, function, line,
4858 "inode #%lu: comm %s: iget: illegal inode #",
4859 ino, current->comm);
4860 return ERR_PTR(-EFSCORRUPTED);
4863 inode = iget_locked(sb, ino);
4865 return ERR_PTR(-ENOMEM);
4866 if (!(inode->i_state & I_NEW))
4872 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4875 raw_inode = ext4_raw_inode(&iloc);
4877 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4878 ext4_error_inode(inode, function, line, 0,
4879 "iget: root inode unallocated");
4880 ret = -EFSCORRUPTED;
4884 if ((flags & EXT4_IGET_HANDLE) &&
4885 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4890 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4891 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4892 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4893 EXT4_INODE_SIZE(inode->i_sb) ||
4894 (ei->i_extra_isize & 3)) {
4895 ext4_error_inode(inode, function, line, 0,
4896 "iget: bad extra_isize %u "
4899 EXT4_INODE_SIZE(inode->i_sb));
4900 ret = -EFSCORRUPTED;
4904 ei->i_extra_isize = 0;
4906 /* Precompute checksum seed for inode metadata */
4907 if (ext4_has_metadata_csum(sb)) {
4908 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4910 __le32 inum = cpu_to_le32(inode->i_ino);
4911 __le32 gen = raw_inode->i_generation;
4912 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4914 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4918 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4919 ext4_error_inode(inode, function, line, 0,
4920 "iget: checksum invalid");
4925 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4926 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4927 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4928 if (ext4_has_feature_project(sb) &&
4929 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4930 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4931 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4933 i_projid = EXT4_DEF_PROJID;
4935 if (!(test_opt(inode->i_sb, NO_UID32))) {
4936 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4937 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4939 i_uid_write(inode, i_uid);
4940 i_gid_write(inode, i_gid);
4941 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4942 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4944 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4945 ei->i_inline_off = 0;
4946 ei->i_dir_start_lookup = 0;
4947 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4948 /* We now have enough fields to check if the inode was active or not.
4949 * This is needed because nfsd might try to access dead inodes
4950 * the test is that same one that e2fsck uses
4951 * NeilBrown 1999oct15
4953 if (inode->i_nlink == 0) {
4954 if ((inode->i_mode == 0 ||
4955 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4956 ino != EXT4_BOOT_LOADER_INO) {
4957 /* this inode is deleted */
4961 /* The only unlinked inodes we let through here have
4962 * valid i_mode and are being read by the orphan
4963 * recovery code: that's fine, we're about to complete
4964 * the process of deleting those.
4965 * OR it is the EXT4_BOOT_LOADER_INO which is
4966 * not initialized on a new filesystem. */
4968 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4969 ext4_set_inode_flags(inode);
4970 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4971 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4972 if (ext4_has_feature_64bit(sb))
4974 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4975 inode->i_size = ext4_isize(sb, raw_inode);
4976 if ((size = i_size_read(inode)) < 0) {
4977 ext4_error_inode(inode, function, line, 0,
4978 "iget: bad i_size value: %lld", size);
4979 ret = -EFSCORRUPTED;
4982 ei->i_disksize = inode->i_size;
4984 ei->i_reserved_quota = 0;
4986 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4987 ei->i_block_group = iloc.block_group;
4988 ei->i_last_alloc_group = ~0;
4990 * NOTE! The in-memory inode i_data array is in little-endian order
4991 * even on big-endian machines: we do NOT byteswap the block numbers!
4993 for (block = 0; block < EXT4_N_BLOCKS; block++)
4994 ei->i_data[block] = raw_inode->i_block[block];
4995 INIT_LIST_HEAD(&ei->i_orphan);
4998 * Set transaction id's of transactions that have to be committed
4999 * to finish f[data]sync. We set them to currently running transaction
5000 * as we cannot be sure that the inode or some of its metadata isn't
5001 * part of the transaction - the inode could have been reclaimed and
5002 * now it is reread from disk.
5005 transaction_t *transaction;
5008 read_lock(&journal->j_state_lock);
5009 if (journal->j_running_transaction)
5010 transaction = journal->j_running_transaction;
5012 transaction = journal->j_committing_transaction;
5014 tid = transaction->t_tid;
5016 tid = journal->j_commit_sequence;
5017 read_unlock(&journal->j_state_lock);
5018 ei->i_sync_tid = tid;
5019 ei->i_datasync_tid = tid;
5022 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5023 if (ei->i_extra_isize == 0) {
5024 /* The extra space is currently unused. Use it. */
5025 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
5026 ei->i_extra_isize = sizeof(struct ext4_inode) -
5027 EXT4_GOOD_OLD_INODE_SIZE;
5029 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
5035 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5036 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5037 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5038 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5040 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5041 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
5043 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5044 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5046 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5048 ext4_inode_set_iversion_queried(inode, ivers);
5052 if (ei->i_file_acl &&
5053 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5054 ext4_error_inode(inode, function, line, 0,
5055 "iget: bad extended attribute block %llu",
5057 ret = -EFSCORRUPTED;
5059 } else if (!ext4_has_inline_data(inode)) {
5060 /* validate the block references in the inode */
5061 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5062 (S_ISLNK(inode->i_mode) &&
5063 !ext4_inode_is_fast_symlink(inode))) {
5064 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5065 ret = ext4_ext_check_inode(inode);
5067 ret = ext4_ind_check_inode(inode);
5073 if (S_ISREG(inode->i_mode)) {
5074 inode->i_op = &ext4_file_inode_operations;
5075 inode->i_fop = &ext4_file_operations;
5076 ext4_set_aops(inode);
5077 } else if (S_ISDIR(inode->i_mode)) {
5078 inode->i_op = &ext4_dir_inode_operations;
5079 inode->i_fop = &ext4_dir_operations;
5080 } else if (S_ISLNK(inode->i_mode)) {
5081 /* VFS does not allow setting these so must be corruption */
5082 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5083 ext4_error_inode(inode, function, line, 0,
5084 "iget: immutable or append flags "
5085 "not allowed on symlinks");
5086 ret = -EFSCORRUPTED;
5089 if (IS_ENCRYPTED(inode)) {
5090 inode->i_op = &ext4_encrypted_symlink_inode_operations;
5091 ext4_set_aops(inode);
5092 } else if (ext4_inode_is_fast_symlink(inode)) {
5093 inode->i_link = (char *)ei->i_data;
5094 inode->i_op = &ext4_fast_symlink_inode_operations;
5095 nd_terminate_link(ei->i_data, inode->i_size,
5096 sizeof(ei->i_data) - 1);
5098 inode->i_op = &ext4_symlink_inode_operations;
5099 ext4_set_aops(inode);
5101 inode_nohighmem(inode);
5102 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5103 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5104 inode->i_op = &ext4_special_inode_operations;
5105 if (raw_inode->i_block[0])
5106 init_special_inode(inode, inode->i_mode,
5107 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5109 init_special_inode(inode, inode->i_mode,
5110 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5111 } else if (ino == EXT4_BOOT_LOADER_INO) {
5112 make_bad_inode(inode);
5114 ret = -EFSCORRUPTED;
5115 ext4_error_inode(inode, function, line, 0,
5116 "iget: bogus i_mode (%o)", inode->i_mode);
5121 unlock_new_inode(inode);
5127 return ERR_PTR(ret);
5130 static int ext4_inode_blocks_set(handle_t *handle,
5131 struct ext4_inode *raw_inode,
5132 struct ext4_inode_info *ei)
5134 struct inode *inode = &(ei->vfs_inode);
5135 u64 i_blocks = inode->i_blocks;
5136 struct super_block *sb = inode->i_sb;
5138 if (i_blocks <= ~0U) {
5140 * i_blocks can be represented in a 32 bit variable
5141 * as multiple of 512 bytes
5143 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5144 raw_inode->i_blocks_high = 0;
5145 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5148 if (!ext4_has_feature_huge_file(sb))
5151 if (i_blocks <= 0xffffffffffffULL) {
5153 * i_blocks can be represented in a 48 bit variable
5154 * as multiple of 512 bytes
5156 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5157 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5158 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5160 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5161 /* i_block is stored in file system block size */
5162 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5163 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5164 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5169 struct other_inode {
5170 unsigned long orig_ino;
5171 struct ext4_inode *raw_inode;
5174 static int other_inode_match(struct inode * inode, unsigned long ino,
5177 struct other_inode *oi = (struct other_inode *) data;
5179 if ((inode->i_ino != ino) ||
5180 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5182 ((inode->i_state & I_DIRTY_TIME) == 0))
5184 spin_lock(&inode->i_lock);
5185 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5186 I_DIRTY_INODE)) == 0) &&
5187 (inode->i_state & I_DIRTY_TIME)) {
5188 struct ext4_inode_info *ei = EXT4_I(inode);
5190 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
5191 spin_unlock(&inode->i_lock);
5193 spin_lock(&ei->i_raw_lock);
5194 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
5195 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
5196 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
5197 ext4_inode_csum_set(inode, oi->raw_inode, ei);
5198 spin_unlock(&ei->i_raw_lock);
5199 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
5202 spin_unlock(&inode->i_lock);
5207 * Opportunistically update the other time fields for other inodes in
5208 * the same inode table block.
5210 static void ext4_update_other_inodes_time(struct super_block *sb,
5211 unsigned long orig_ino, char *buf)
5213 struct other_inode oi;
5215 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5216 int inode_size = EXT4_INODE_SIZE(sb);
5218 oi.orig_ino = orig_ino;
5220 * Calculate the first inode in the inode table block. Inode
5221 * numbers are one-based. That is, the first inode in a block
5222 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5224 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5225 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5226 if (ino == orig_ino)
5228 oi.raw_inode = (struct ext4_inode *) buf;
5229 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
5234 * Post the struct inode info into an on-disk inode location in the
5235 * buffer-cache. This gobbles the caller's reference to the
5236 * buffer_head in the inode location struct.
5238 * The caller must have write access to iloc->bh.
5240 static int ext4_do_update_inode(handle_t *handle,
5241 struct inode *inode,
5242 struct ext4_iloc *iloc)
5244 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5245 struct ext4_inode_info *ei = EXT4_I(inode);
5246 struct buffer_head *bh = iloc->bh;
5247 struct super_block *sb = inode->i_sb;
5248 int err = 0, rc, block;
5249 int need_datasync = 0, set_large_file = 0;
5254 spin_lock(&ei->i_raw_lock);
5256 /* For fields not tracked in the in-memory inode,
5257 * initialise them to zero for new inodes. */
5258 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5259 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5261 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5262 i_uid = i_uid_read(inode);
5263 i_gid = i_gid_read(inode);
5264 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5265 if (!(test_opt(inode->i_sb, NO_UID32))) {
5266 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5267 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5269 * Fix up interoperability with old kernels. Otherwise, old inodes get
5270 * re-used with the upper 16 bits of the uid/gid intact
5272 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5273 raw_inode->i_uid_high = 0;
5274 raw_inode->i_gid_high = 0;
5276 raw_inode->i_uid_high =
5277 cpu_to_le16(high_16_bits(i_uid));
5278 raw_inode->i_gid_high =
5279 cpu_to_le16(high_16_bits(i_gid));
5282 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5283 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5284 raw_inode->i_uid_high = 0;
5285 raw_inode->i_gid_high = 0;
5287 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5289 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5290 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5291 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5292 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5294 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5296 spin_unlock(&ei->i_raw_lock);
5299 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5300 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5301 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5302 raw_inode->i_file_acl_high =
5303 cpu_to_le16(ei->i_file_acl >> 32);
5304 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5305 if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
5306 ext4_isize_set(raw_inode, ei->i_disksize);
5309 if (ei->i_disksize > 0x7fffffffULL) {
5310 if (!ext4_has_feature_large_file(sb) ||
5311 EXT4_SB(sb)->s_es->s_rev_level ==
5312 cpu_to_le32(EXT4_GOOD_OLD_REV))
5315 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5316 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5317 if (old_valid_dev(inode->i_rdev)) {
5318 raw_inode->i_block[0] =
5319 cpu_to_le32(old_encode_dev(inode->i_rdev));
5320 raw_inode->i_block[1] = 0;
5322 raw_inode->i_block[0] = 0;
5323 raw_inode->i_block[1] =
5324 cpu_to_le32(new_encode_dev(inode->i_rdev));
5325 raw_inode->i_block[2] = 0;
5327 } else if (!ext4_has_inline_data(inode)) {
5328 for (block = 0; block < EXT4_N_BLOCKS; block++)
5329 raw_inode->i_block[block] = ei->i_data[block];
5332 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5333 u64 ivers = ext4_inode_peek_iversion(inode);
5335 raw_inode->i_disk_version = cpu_to_le32(ivers);
5336 if (ei->i_extra_isize) {
5337 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5338 raw_inode->i_version_hi =
5339 cpu_to_le32(ivers >> 32);
5340 raw_inode->i_extra_isize =
5341 cpu_to_le16(ei->i_extra_isize);
5345 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5346 i_projid != EXT4_DEF_PROJID);
5348 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5349 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5350 raw_inode->i_projid = cpu_to_le32(i_projid);
5352 ext4_inode_csum_set(inode, raw_inode, ei);
5353 spin_unlock(&ei->i_raw_lock);
5354 if (inode->i_sb->s_flags & SB_LAZYTIME)
5355 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5358 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5359 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5362 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5363 if (set_large_file) {
5364 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5365 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5368 ext4_set_feature_large_file(sb);
5369 ext4_handle_sync(handle);
5370 err = ext4_handle_dirty_super(handle, sb);
5372 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5375 ext4_std_error(inode->i_sb, err);
5380 * ext4_write_inode()
5382 * We are called from a few places:
5384 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5385 * Here, there will be no transaction running. We wait for any running
5386 * transaction to commit.
5388 * - Within flush work (sys_sync(), kupdate and such).
5389 * We wait on commit, if told to.
5391 * - Within iput_final() -> write_inode_now()
5392 * We wait on commit, if told to.
5394 * In all cases it is actually safe for us to return without doing anything,
5395 * because the inode has been copied into a raw inode buffer in
5396 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5399 * Note that we are absolutely dependent upon all inode dirtiers doing the
5400 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5401 * which we are interested.
5403 * It would be a bug for them to not do this. The code:
5405 * mark_inode_dirty(inode)
5407 * inode->i_size = expr;
5409 * is in error because write_inode() could occur while `stuff()' is running,
5410 * and the new i_size will be lost. Plus the inode will no longer be on the
5411 * superblock's dirty inode list.
5413 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5417 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5418 sb_rdonly(inode->i_sb))
5421 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5424 if (EXT4_SB(inode->i_sb)->s_journal) {
5425 if (ext4_journal_current_handle()) {
5426 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5432 * No need to force transaction in WB_SYNC_NONE mode. Also
5433 * ext4_sync_fs() will force the commit after everything is
5436 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5439 err = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5440 EXT4_I(inode)->i_sync_tid);
5442 struct ext4_iloc iloc;
5444 err = __ext4_get_inode_loc(inode, &iloc, 0);
5448 * sync(2) will flush the whole buffer cache. No need to do
5449 * it here separately for each inode.
5451 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5452 sync_dirty_buffer(iloc.bh);
5453 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5454 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5455 "IO error syncing inode");
5464 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5465 * buffers that are attached to a page stradding i_size and are undergoing
5466 * commit. In that case we have to wait for commit to finish and try again.
5468 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5472 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5473 tid_t commit_tid = 0;
5476 offset = inode->i_size & (PAGE_SIZE - 1);
5478 * All buffers in the last page remain valid? Then there's nothing to
5479 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5482 if (offset > PAGE_SIZE - i_blocksize(inode))
5485 page = find_lock_page(inode->i_mapping,
5486 inode->i_size >> PAGE_SHIFT);
5489 ret = __ext4_journalled_invalidatepage(page, offset,
5490 PAGE_SIZE - offset);
5496 read_lock(&journal->j_state_lock);
5497 if (journal->j_committing_transaction)
5498 commit_tid = journal->j_committing_transaction->t_tid;
5499 read_unlock(&journal->j_state_lock);
5501 jbd2_log_wait_commit(journal, commit_tid);
5508 * Called from notify_change.
5510 * We want to trap VFS attempts to truncate the file as soon as
5511 * possible. In particular, we want to make sure that when the VFS
5512 * shrinks i_size, we put the inode on the orphan list and modify
5513 * i_disksize immediately, so that during the subsequent flushing of
5514 * dirty pages and freeing of disk blocks, we can guarantee that any
5515 * commit will leave the blocks being flushed in an unused state on
5516 * disk. (On recovery, the inode will get truncated and the blocks will
5517 * be freed, so we have a strong guarantee that no future commit will
5518 * leave these blocks visible to the user.)
5520 * Another thing we have to assure is that if we are in ordered mode
5521 * and inode is still attached to the committing transaction, we must
5522 * we start writeout of all the dirty pages which are being truncated.
5523 * This way we are sure that all the data written in the previous
5524 * transaction are already on disk (truncate waits for pages under
5527 * Called with inode->i_mutex down.
5529 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5531 struct inode *inode = d_inode(dentry);
5534 const unsigned int ia_valid = attr->ia_valid;
5536 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5539 if (unlikely(IS_IMMUTABLE(inode)))
5542 if (unlikely(IS_APPEND(inode) &&
5543 (ia_valid & (ATTR_MODE | ATTR_UID |
5544 ATTR_GID | ATTR_TIMES_SET))))
5547 error = setattr_prepare(dentry, attr);
5551 error = fscrypt_prepare_setattr(dentry, attr);
5555 if (is_quota_modification(inode, attr)) {
5556 error = dquot_initialize(inode);
5560 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5561 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5564 /* (user+group)*(old+new) structure, inode write (sb,
5565 * inode block, ? - but truncate inode update has it) */
5566 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5567 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5568 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5569 if (IS_ERR(handle)) {
5570 error = PTR_ERR(handle);
5574 /* dquot_transfer() calls back ext4_get_inode_usage() which
5575 * counts xattr inode references.
5577 down_read(&EXT4_I(inode)->xattr_sem);
5578 error = dquot_transfer(inode, attr);
5579 up_read(&EXT4_I(inode)->xattr_sem);
5582 ext4_journal_stop(handle);
5585 /* Update corresponding info in inode so that everything is in
5586 * one transaction */
5587 if (attr->ia_valid & ATTR_UID)
5588 inode->i_uid = attr->ia_uid;
5589 if (attr->ia_valid & ATTR_GID)
5590 inode->i_gid = attr->ia_gid;
5591 error = ext4_mark_inode_dirty(handle, inode);
5592 ext4_journal_stop(handle);
5595 if (attr->ia_valid & ATTR_SIZE) {
5597 loff_t oldsize = inode->i_size;
5598 int shrink = (attr->ia_size < inode->i_size);
5600 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5601 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5603 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5606 if (!S_ISREG(inode->i_mode))
5609 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5610 inode_inc_iversion(inode);
5613 if (ext4_should_order_data(inode)) {
5614 error = ext4_begin_ordered_truncate(inode,
5620 * Blocks are going to be removed from the inode. Wait
5621 * for dio in flight.
5623 inode_dio_wait(inode);
5626 down_write(&EXT4_I(inode)->i_mmap_sem);
5628 rc = ext4_break_layouts(inode);
5630 up_write(&EXT4_I(inode)->i_mmap_sem);
5634 if (attr->ia_size != inode->i_size) {
5635 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5636 if (IS_ERR(handle)) {
5637 error = PTR_ERR(handle);
5640 if (ext4_handle_valid(handle) && shrink) {
5641 error = ext4_orphan_add(handle, inode);
5645 * Update c/mtime on truncate up, ext4_truncate() will
5646 * update c/mtime in shrink case below
5649 inode->i_mtime = current_time(inode);
5650 inode->i_ctime = inode->i_mtime;
5652 down_write(&EXT4_I(inode)->i_data_sem);
5653 EXT4_I(inode)->i_disksize = attr->ia_size;
5654 rc = ext4_mark_inode_dirty(handle, inode);
5658 * We have to update i_size under i_data_sem together
5659 * with i_disksize to avoid races with writeback code
5660 * running ext4_wb_update_i_disksize().
5663 i_size_write(inode, attr->ia_size);
5664 up_write(&EXT4_I(inode)->i_data_sem);
5665 ext4_journal_stop(handle);
5669 pagecache_isize_extended(inode, oldsize,
5671 } else if (ext4_should_journal_data(inode)) {
5672 ext4_wait_for_tail_page_commit(inode);
5677 * Truncate pagecache after we've waited for commit
5678 * in data=journal mode to make pages freeable.
5680 truncate_pagecache(inode, inode->i_size);
5682 * Call ext4_truncate() even if i_size didn't change to
5683 * truncate possible preallocated blocks.
5685 if (attr->ia_size <= oldsize) {
5686 rc = ext4_truncate(inode);
5691 up_write(&EXT4_I(inode)->i_mmap_sem);
5695 setattr_copy(inode, attr);
5696 mark_inode_dirty(inode);
5700 * If the call to ext4_truncate failed to get a transaction handle at
5701 * all, we need to clean up the in-core orphan list manually.
5703 if (orphan && inode->i_nlink)
5704 ext4_orphan_del(NULL, inode);
5706 if (!error && (ia_valid & ATTR_MODE))
5707 rc = posix_acl_chmod(inode, inode->i_mode);
5710 ext4_std_error(inode->i_sb, error);
5716 int ext4_getattr(const struct path *path, struct kstat *stat,
5717 u32 request_mask, unsigned int query_flags)
5719 struct inode *inode = d_inode(path->dentry);
5720 struct ext4_inode *raw_inode;
5721 struct ext4_inode_info *ei = EXT4_I(inode);
5724 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5725 stat->result_mask |= STATX_BTIME;
5726 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5727 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5730 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5731 if (flags & EXT4_APPEND_FL)
5732 stat->attributes |= STATX_ATTR_APPEND;
5733 if (flags & EXT4_COMPR_FL)
5734 stat->attributes |= STATX_ATTR_COMPRESSED;
5735 if (flags & EXT4_ENCRYPT_FL)
5736 stat->attributes |= STATX_ATTR_ENCRYPTED;
5737 if (flags & EXT4_IMMUTABLE_FL)
5738 stat->attributes |= STATX_ATTR_IMMUTABLE;
5739 if (flags & EXT4_NODUMP_FL)
5740 stat->attributes |= STATX_ATTR_NODUMP;
5742 stat->attributes_mask |= (STATX_ATTR_APPEND |
5743 STATX_ATTR_COMPRESSED |
5744 STATX_ATTR_ENCRYPTED |
5745 STATX_ATTR_IMMUTABLE |
5748 generic_fillattr(inode, stat);
5752 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5753 u32 request_mask, unsigned int query_flags)
5755 struct inode *inode = d_inode(path->dentry);
5756 u64 delalloc_blocks;
5758 ext4_getattr(path, stat, request_mask, query_flags);
5761 * If there is inline data in the inode, the inode will normally not
5762 * have data blocks allocated (it may have an external xattr block).
5763 * Report at least one sector for such files, so tools like tar, rsync,
5764 * others don't incorrectly think the file is completely sparse.
5766 if (unlikely(ext4_has_inline_data(inode)))
5767 stat->blocks += (stat->size + 511) >> 9;
5770 * We can't update i_blocks if the block allocation is delayed
5771 * otherwise in the case of system crash before the real block
5772 * allocation is done, we will have i_blocks inconsistent with
5773 * on-disk file blocks.
5774 * We always keep i_blocks updated together with real
5775 * allocation. But to not confuse with user, stat
5776 * will return the blocks that include the delayed allocation
5777 * blocks for this file.
5779 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5780 EXT4_I(inode)->i_reserved_data_blocks);
5781 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5785 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5788 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5789 return ext4_ind_trans_blocks(inode, lblocks);
5790 return ext4_ext_index_trans_blocks(inode, pextents);
5794 * Account for index blocks, block groups bitmaps and block group
5795 * descriptor blocks if modify datablocks and index blocks
5796 * worse case, the indexs blocks spread over different block groups
5798 * If datablocks are discontiguous, they are possible to spread over
5799 * different block groups too. If they are contiguous, with flexbg,
5800 * they could still across block group boundary.
5802 * Also account for superblock, inode, quota and xattr blocks
5804 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5807 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5813 * How many index blocks need to touch to map @lblocks logical blocks
5814 * to @pextents physical extents?
5816 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5821 * Now let's see how many group bitmaps and group descriptors need
5824 groups = idxblocks + pextents;
5826 if (groups > ngroups)
5828 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5829 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5831 /* bitmaps and block group descriptor blocks */
5832 ret += groups + gdpblocks;
5834 /* Blocks for super block, inode, quota and xattr blocks */
5835 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5841 * Calculate the total number of credits to reserve to fit
5842 * the modification of a single pages into a single transaction,
5843 * which may include multiple chunks of block allocations.
5845 * This could be called via ext4_write_begin()
5847 * We need to consider the worse case, when
5848 * one new block per extent.
5850 int ext4_writepage_trans_blocks(struct inode *inode)
5852 int bpp = ext4_journal_blocks_per_page(inode);
5855 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5857 /* Account for data blocks for journalled mode */
5858 if (ext4_should_journal_data(inode))
5864 * Calculate the journal credits for a chunk of data modification.
5866 * This is called from DIO, fallocate or whoever calling
5867 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5869 * journal buffers for data blocks are not included here, as DIO
5870 * and fallocate do no need to journal data buffers.
5872 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5874 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5878 * The caller must have previously called ext4_reserve_inode_write().
5879 * Give this, we know that the caller already has write access to iloc->bh.
5881 int ext4_mark_iloc_dirty(handle_t *handle,
5882 struct inode *inode, struct ext4_iloc *iloc)
5886 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5890 if (IS_I_VERSION(inode))
5891 inode_inc_iversion(inode);
5893 /* the do_update_inode consumes one bh->b_count */
5896 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5897 err = ext4_do_update_inode(handle, inode, iloc);
5903 * On success, We end up with an outstanding reference count against
5904 * iloc->bh. This _must_ be cleaned up later.
5908 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5909 struct ext4_iloc *iloc)
5913 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5916 err = ext4_get_inode_loc(inode, iloc);
5918 BUFFER_TRACE(iloc->bh, "get_write_access");
5919 err = ext4_journal_get_write_access(handle, iloc->bh);
5925 ext4_std_error(inode->i_sb, err);
5929 static int __ext4_expand_extra_isize(struct inode *inode,
5930 unsigned int new_extra_isize,
5931 struct ext4_iloc *iloc,
5932 handle_t *handle, int *no_expand)
5934 struct ext4_inode *raw_inode;
5935 struct ext4_xattr_ibody_header *header;
5938 raw_inode = ext4_raw_inode(iloc);
5940 header = IHDR(inode, raw_inode);
5942 /* No extended attributes present */
5943 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5944 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5945 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5946 EXT4_I(inode)->i_extra_isize, 0,
5947 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5948 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5952 /* try to expand with EAs present */
5953 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5957 * Inode size expansion failed; don't try again
5966 * Expand an inode by new_extra_isize bytes.
5967 * Returns 0 on success or negative error number on failure.
5969 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5970 unsigned int new_extra_isize,
5971 struct ext4_iloc iloc,
5977 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5981 * In nojournal mode, we can immediately attempt to expand
5982 * the inode. When journaled, we first need to obtain extra
5983 * buffer credits since we may write into the EA block
5984 * with this same handle. If journal_extend fails, then it will
5985 * only result in a minor loss of functionality for that inode.
5986 * If this is felt to be critical, then e2fsck should be run to
5987 * force a large enough s_min_extra_isize.
5989 if (ext4_handle_valid(handle) &&
5990 jbd2_journal_extend(handle,
5991 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0)
5994 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5997 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5998 handle, &no_expand);
5999 ext4_write_unlock_xattr(inode, &no_expand);
6004 int ext4_expand_extra_isize(struct inode *inode,
6005 unsigned int new_extra_isize,
6006 struct ext4_iloc *iloc)
6012 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
6017 handle = ext4_journal_start(inode, EXT4_HT_INODE,
6018 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
6019 if (IS_ERR(handle)) {
6020 error = PTR_ERR(handle);
6025 ext4_write_lock_xattr(inode, &no_expand);
6027 BUFFER_TRACE(iloc->bh, "get_write_access");
6028 error = ext4_journal_get_write_access(handle, iloc->bh);
6034 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
6035 handle, &no_expand);
6037 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
6041 ext4_write_unlock_xattr(inode, &no_expand);
6043 ext4_journal_stop(handle);
6048 * What we do here is to mark the in-core inode as clean with respect to inode
6049 * dirtiness (it may still be data-dirty).
6050 * This means that the in-core inode may be reaped by prune_icache
6051 * without having to perform any I/O. This is a very good thing,
6052 * because *any* task may call prune_icache - even ones which
6053 * have a transaction open against a different journal.
6055 * Is this cheating? Not really. Sure, we haven't written the
6056 * inode out, but prune_icache isn't a user-visible syncing function.
6057 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6058 * we start and wait on commits.
6060 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
6062 struct ext4_iloc iloc;
6063 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6067 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6068 err = ext4_reserve_inode_write(handle, inode, &iloc);
6072 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6073 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6076 return ext4_mark_iloc_dirty(handle, inode, &iloc);
6080 * ext4_dirty_inode() is called from __mark_inode_dirty()
6082 * We're really interested in the case where a file is being extended.
6083 * i_size has been changed by generic_commit_write() and we thus need
6084 * to include the updated inode in the current transaction.
6086 * Also, dquot_alloc_block() will always dirty the inode when blocks
6087 * are allocated to the file.
6089 * If the inode is marked synchronous, we don't honour that here - doing
6090 * so would cause a commit on atime updates, which we don't bother doing.
6091 * We handle synchronous inodes at the highest possible level.
6093 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
6094 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
6095 * to copy into the on-disk inode structure are the timestamp files.
6097 void ext4_dirty_inode(struct inode *inode, int flags)
6101 if (flags == I_DIRTY_TIME)
6103 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6107 ext4_mark_inode_dirty(handle, inode);
6109 ext4_journal_stop(handle);
6114 int ext4_change_inode_journal_flag(struct inode *inode, int val)
6119 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6122 * We have to be very careful here: changing a data block's
6123 * journaling status dynamically is dangerous. If we write a
6124 * data block to the journal, change the status and then delete
6125 * that block, we risk forgetting to revoke the old log record
6126 * from the journal and so a subsequent replay can corrupt data.
6127 * So, first we make sure that the journal is empty and that
6128 * nobody is changing anything.
6131 journal = EXT4_JOURNAL(inode);
6134 if (is_journal_aborted(journal))
6137 /* Wait for all existing dio workers */
6138 inode_dio_wait(inode);
6141 * Before flushing the journal and switching inode's aops, we have
6142 * to flush all dirty data the inode has. There can be outstanding
6143 * delayed allocations, there can be unwritten extents created by
6144 * fallocate or buffered writes in dioread_nolock mode covered by
6145 * dirty data which can be converted only after flushing the dirty
6146 * data (and journalled aops don't know how to handle these cases).
6149 down_write(&EXT4_I(inode)->i_mmap_sem);
6150 err = filemap_write_and_wait(inode->i_mapping);
6152 up_write(&EXT4_I(inode)->i_mmap_sem);
6157 percpu_down_write(&sbi->s_journal_flag_rwsem);
6158 jbd2_journal_lock_updates(journal);
6161 * OK, there are no updates running now, and all cached data is
6162 * synced to disk. We are now in a completely consistent state
6163 * which doesn't have anything in the journal, and we know that
6164 * no filesystem updates are running, so it is safe to modify
6165 * the inode's in-core data-journaling state flag now.
6169 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6171 err = jbd2_journal_flush(journal);
6173 jbd2_journal_unlock_updates(journal);
6174 percpu_up_write(&sbi->s_journal_flag_rwsem);
6177 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6179 ext4_set_aops(inode);
6181 jbd2_journal_unlock_updates(journal);
6182 percpu_up_write(&sbi->s_journal_flag_rwsem);
6185 up_write(&EXT4_I(inode)->i_mmap_sem);
6187 /* Finally we can mark the inode as dirty. */
6189 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6191 return PTR_ERR(handle);
6193 err = ext4_mark_inode_dirty(handle, inode);
6194 ext4_handle_sync(handle);
6195 ext4_journal_stop(handle);
6196 ext4_std_error(inode->i_sb, err);
6201 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6203 return !buffer_mapped(bh);
6206 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6208 struct vm_area_struct *vma = vmf->vma;
6209 struct page *page = vmf->page;
6214 struct file *file = vma->vm_file;
6215 struct inode *inode = file_inode(file);
6216 struct address_space *mapping = inode->i_mapping;
6218 get_block_t *get_block;
6221 if (unlikely(IS_IMMUTABLE(inode)))
6222 return VM_FAULT_SIGBUS;
6224 sb_start_pagefault(inode->i_sb);
6225 file_update_time(vma->vm_file);
6227 down_read(&EXT4_I(inode)->i_mmap_sem);
6229 err = ext4_convert_inline_data(inode);
6233 /* Delalloc case is easy... */
6234 if (test_opt(inode->i_sb, DELALLOC) &&
6235 !ext4_should_journal_data(inode) &&
6236 !ext4_nonda_switch(inode->i_sb)) {
6238 err = block_page_mkwrite(vma, vmf,
6239 ext4_da_get_block_prep);
6240 } while (err == -ENOSPC &&
6241 ext4_should_retry_alloc(inode->i_sb, &retries));
6246 size = i_size_read(inode);
6247 /* Page got truncated from under us? */
6248 if (page->mapping != mapping || page_offset(page) > size) {
6250 ret = VM_FAULT_NOPAGE;
6254 if (page->index == size >> PAGE_SHIFT)
6255 len = size & ~PAGE_MASK;
6259 * Return if we have all the buffers mapped. This avoids the need to do
6260 * journal_start/journal_stop which can block and take a long time
6262 if (page_has_buffers(page)) {
6263 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6265 ext4_bh_unmapped)) {
6266 /* Wait so that we don't change page under IO */
6267 wait_for_stable_page(page);
6268 ret = VM_FAULT_LOCKED;
6273 /* OK, we need to fill the hole... */
6274 if (ext4_should_dioread_nolock(inode))
6275 get_block = ext4_get_block_unwritten;
6277 get_block = ext4_get_block;
6279 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6280 ext4_writepage_trans_blocks(inode));
6281 if (IS_ERR(handle)) {
6282 ret = VM_FAULT_SIGBUS;
6285 err = block_page_mkwrite(vma, vmf, get_block);
6286 if (!err && ext4_should_journal_data(inode)) {
6287 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
6288 PAGE_SIZE, NULL, do_journal_get_write_access)) {
6290 ret = VM_FAULT_SIGBUS;
6291 ext4_journal_stop(handle);
6294 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6296 ext4_journal_stop(handle);
6297 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6300 ret = block_page_mkwrite_return(err);
6302 up_read(&EXT4_I(inode)->i_mmap_sem);
6303 sb_end_pagefault(inode->i_sb);
6307 vm_fault_t ext4_filemap_fault(struct vm_fault *vmf)
6309 struct inode *inode = file_inode(vmf->vma->vm_file);
6312 down_read(&EXT4_I(inode)->i_mmap_sem);
6313 ret = filemap_fault(vmf);
6314 up_read(&EXT4_I(inode)->i_mmap_sem);