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_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
404 ext4_error_inode(inode, func, line, map->m_pblk,
405 "lblock %lu mapped to illegal pblock %llu "
406 "(length %d)", (unsigned long) map->m_lblk,
407 map->m_pblk, map->m_len);
408 return -EFSCORRUPTED;
413 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
418 if (ext4_encrypted_inode(inode))
419 return fscrypt_zeroout_range(inode, lblk, pblk, len);
421 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
428 #define check_block_validity(inode, map) \
429 __check_block_validity((inode), __func__, __LINE__, (map))
431 #ifdef ES_AGGRESSIVE_TEST
432 static void ext4_map_blocks_es_recheck(handle_t *handle,
434 struct ext4_map_blocks *es_map,
435 struct ext4_map_blocks *map,
442 * There is a race window that the result is not the same.
443 * e.g. xfstests #223 when dioread_nolock enables. The reason
444 * is that we lookup a block mapping in extent status tree with
445 * out taking i_data_sem. So at the time the unwritten extent
446 * could be converted.
448 down_read(&EXT4_I(inode)->i_data_sem);
449 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
450 retval = ext4_ext_map_blocks(handle, inode, map, flags &
451 EXT4_GET_BLOCKS_KEEP_SIZE);
453 retval = ext4_ind_map_blocks(handle, inode, map, flags &
454 EXT4_GET_BLOCKS_KEEP_SIZE);
456 up_read((&EXT4_I(inode)->i_data_sem));
459 * We don't check m_len because extent will be collpased in status
460 * tree. So the m_len might not equal.
462 if (es_map->m_lblk != map->m_lblk ||
463 es_map->m_flags != map->m_flags ||
464 es_map->m_pblk != map->m_pblk) {
465 printk("ES cache assertion failed for inode: %lu "
466 "es_cached ex [%d/%d/%llu/%x] != "
467 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
468 inode->i_ino, es_map->m_lblk, es_map->m_len,
469 es_map->m_pblk, es_map->m_flags, map->m_lblk,
470 map->m_len, map->m_pblk, map->m_flags,
474 #endif /* ES_AGGRESSIVE_TEST */
477 * The ext4_map_blocks() function tries to look up the requested blocks,
478 * and returns if the blocks are already mapped.
480 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
481 * and store the allocated blocks in the result buffer head and mark it
484 * If file type is extents based, it will call ext4_ext_map_blocks(),
485 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
488 * On success, it returns the number of blocks being mapped or allocated. if
489 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
490 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
492 * It returns 0 if plain look up failed (blocks have not been allocated), in
493 * that case, @map is returned as unmapped but we still do fill map->m_len to
494 * indicate the length of a hole starting at map->m_lblk.
496 * It returns the error in case of allocation failure.
498 int ext4_map_blocks(handle_t *handle, struct inode *inode,
499 struct ext4_map_blocks *map, int flags)
501 struct extent_status es;
504 #ifdef ES_AGGRESSIVE_TEST
505 struct ext4_map_blocks orig_map;
507 memcpy(&orig_map, map, sizeof(*map));
511 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
512 "logical block %lu\n", inode->i_ino, flags, map->m_len,
513 (unsigned long) map->m_lblk);
516 * ext4_map_blocks returns an int, and m_len is an unsigned int
518 if (unlikely(map->m_len > INT_MAX))
519 map->m_len = INT_MAX;
521 /* We can handle the block number less than EXT_MAX_BLOCKS */
522 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
523 return -EFSCORRUPTED;
525 /* Lookup extent status tree firstly */
526 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
527 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
528 map->m_pblk = ext4_es_pblock(&es) +
529 map->m_lblk - es.es_lblk;
530 map->m_flags |= ext4_es_is_written(&es) ?
531 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
532 retval = es.es_len - (map->m_lblk - es.es_lblk);
533 if (retval > map->m_len)
536 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
538 retval = es.es_len - (map->m_lblk - es.es_lblk);
539 if (retval > map->m_len)
546 #ifdef ES_AGGRESSIVE_TEST
547 ext4_map_blocks_es_recheck(handle, inode, map,
554 * Try to see if we can get the block without requesting a new
557 down_read(&EXT4_I(inode)->i_data_sem);
558 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
559 retval = ext4_ext_map_blocks(handle, inode, map, flags &
560 EXT4_GET_BLOCKS_KEEP_SIZE);
562 retval = ext4_ind_map_blocks(handle, inode, map, flags &
563 EXT4_GET_BLOCKS_KEEP_SIZE);
568 if (unlikely(retval != map->m_len)) {
569 ext4_warning(inode->i_sb,
570 "ES len assertion failed for inode "
571 "%lu: retval %d != map->m_len %d",
572 inode->i_ino, retval, map->m_len);
576 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
577 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
578 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
579 !(status & EXTENT_STATUS_WRITTEN) &&
580 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
581 map->m_lblk + map->m_len - 1))
582 status |= EXTENT_STATUS_DELAYED;
583 ret = ext4_es_insert_extent(inode, map->m_lblk,
584 map->m_len, map->m_pblk, status);
588 up_read((&EXT4_I(inode)->i_data_sem));
591 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
592 ret = check_block_validity(inode, map);
597 /* If it is only a block(s) look up */
598 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
602 * Returns if the blocks have already allocated
604 * Note that if blocks have been preallocated
605 * ext4_ext_get_block() returns the create = 0
606 * with buffer head unmapped.
608 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
610 * If we need to convert extent to unwritten
611 * we continue and do the actual work in
612 * ext4_ext_map_blocks()
614 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
618 * Here we clear m_flags because after allocating an new extent,
619 * it will be set again.
621 map->m_flags &= ~EXT4_MAP_FLAGS;
624 * New blocks allocate and/or writing to unwritten extent
625 * will possibly result in updating i_data, so we take
626 * the write lock of i_data_sem, and call get_block()
627 * with create == 1 flag.
629 down_write(&EXT4_I(inode)->i_data_sem);
632 * We need to check for EXT4 here because migrate
633 * could have changed the inode type in between
635 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
636 retval = ext4_ext_map_blocks(handle, inode, map, flags);
638 retval = ext4_ind_map_blocks(handle, inode, map, flags);
640 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
642 * We allocated new blocks which will result in
643 * i_data's format changing. Force the migrate
644 * to fail by clearing migrate flags
646 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
650 * Update reserved blocks/metadata blocks after successful
651 * block allocation which had been deferred till now. We don't
652 * support fallocate for non extent files. So we can update
653 * reserve space here.
656 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
657 ext4_da_update_reserve_space(inode, retval, 1);
663 if (unlikely(retval != map->m_len)) {
664 ext4_warning(inode->i_sb,
665 "ES len assertion failed for inode "
666 "%lu: retval %d != map->m_len %d",
667 inode->i_ino, retval, map->m_len);
672 * We have to zeroout blocks before inserting them into extent
673 * status tree. Otherwise someone could look them up there and
674 * use them before they are really zeroed. We also have to
675 * unmap metadata before zeroing as otherwise writeback can
676 * overwrite zeros with stale data from block device.
678 if (flags & EXT4_GET_BLOCKS_ZERO &&
679 map->m_flags & EXT4_MAP_MAPPED &&
680 map->m_flags & EXT4_MAP_NEW) {
681 ret = ext4_issue_zeroout(inode, map->m_lblk,
682 map->m_pblk, map->m_len);
690 * If the extent has been zeroed out, we don't need to update
691 * extent status tree.
693 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
694 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
695 if (ext4_es_is_written(&es))
698 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
699 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
700 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
701 !(status & EXTENT_STATUS_WRITTEN) &&
702 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
703 map->m_lblk + map->m_len - 1))
704 status |= EXTENT_STATUS_DELAYED;
705 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
706 map->m_pblk, status);
714 up_write((&EXT4_I(inode)->i_data_sem));
715 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
716 ret = check_block_validity(inode, map);
721 * Inodes with freshly allocated blocks where contents will be
722 * visible after transaction commit must be on transaction's
725 if (map->m_flags & EXT4_MAP_NEW &&
726 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
727 !(flags & EXT4_GET_BLOCKS_ZERO) &&
728 !ext4_is_quota_file(inode) &&
729 ext4_should_order_data(inode)) {
730 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
731 ret = ext4_jbd2_inode_add_wait(handle, inode);
733 ret = ext4_jbd2_inode_add_write(handle, inode);
742 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
743 * we have to be careful as someone else may be manipulating b_state as well.
745 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
747 unsigned long old_state;
748 unsigned long new_state;
750 flags &= EXT4_MAP_FLAGS;
752 /* Dummy buffer_head? Set non-atomically. */
754 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
758 * Someone else may be modifying b_state. Be careful! This is ugly but
759 * once we get rid of using bh as a container for mapping information
760 * to pass to / from get_block functions, this can go away.
763 old_state = READ_ONCE(bh->b_state);
764 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
766 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
769 static int _ext4_get_block(struct inode *inode, sector_t iblock,
770 struct buffer_head *bh, int flags)
772 struct ext4_map_blocks map;
775 if (ext4_has_inline_data(inode))
779 map.m_len = bh->b_size >> inode->i_blkbits;
781 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
784 map_bh(bh, inode->i_sb, map.m_pblk);
785 ext4_update_bh_state(bh, map.m_flags);
786 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
788 } else if (ret == 0) {
789 /* hole case, need to fill in bh->b_size */
790 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
795 int ext4_get_block(struct inode *inode, sector_t iblock,
796 struct buffer_head *bh, int create)
798 return _ext4_get_block(inode, iblock, bh,
799 create ? EXT4_GET_BLOCKS_CREATE : 0);
803 * Get block function used when preparing for buffered write if we require
804 * creating an unwritten extent if blocks haven't been allocated. The extent
805 * will be converted to written after the IO is complete.
807 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
808 struct buffer_head *bh_result, int create)
810 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
811 inode->i_ino, create);
812 return _ext4_get_block(inode, iblock, bh_result,
813 EXT4_GET_BLOCKS_IO_CREATE_EXT);
816 /* Maximum number of blocks we map for direct IO at once. */
817 #define DIO_MAX_BLOCKS 4096
820 * Get blocks function for the cases that need to start a transaction -
821 * generally difference cases of direct IO and DAX IO. It also handles retries
824 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
825 struct buffer_head *bh_result, int flags)
832 /* Trim mapping request to maximum we can map at once for DIO */
833 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
834 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
835 dio_credits = ext4_chunk_trans_blocks(inode,
836 bh_result->b_size >> inode->i_blkbits);
838 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
840 return PTR_ERR(handle);
842 ret = _ext4_get_block(inode, iblock, bh_result, flags);
843 ext4_journal_stop(handle);
845 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
850 /* Get block function for DIO reads and writes to inodes without extents */
851 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
852 struct buffer_head *bh, int create)
854 /* We don't expect handle for direct IO */
855 WARN_ON_ONCE(ext4_journal_current_handle());
858 return _ext4_get_block(inode, iblock, bh, 0);
859 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
863 * Get block function for AIO DIO writes when we create unwritten extent if
864 * blocks are not allocated yet. The extent will be converted to written
865 * after IO is complete.
867 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
868 sector_t iblock, struct buffer_head *bh_result, int create)
872 /* We don't expect handle for direct IO */
873 WARN_ON_ONCE(ext4_journal_current_handle());
875 ret = ext4_get_block_trans(inode, iblock, bh_result,
876 EXT4_GET_BLOCKS_IO_CREATE_EXT);
879 * When doing DIO using unwritten extents, we need io_end to convert
880 * unwritten extents to written on IO completion. We allocate io_end
881 * once we spot unwritten extent and store it in b_private. Generic
882 * DIO code keeps b_private set and furthermore passes the value to
883 * our completion callback in 'private' argument.
885 if (!ret && buffer_unwritten(bh_result)) {
886 if (!bh_result->b_private) {
887 ext4_io_end_t *io_end;
889 io_end = ext4_init_io_end(inode, GFP_KERNEL);
892 bh_result->b_private = io_end;
893 ext4_set_io_unwritten_flag(inode, io_end);
895 set_buffer_defer_completion(bh_result);
902 * Get block function for non-AIO DIO writes when we create unwritten extent if
903 * blocks are not allocated yet. The extent will be converted to written
904 * after IO is complete by ext4_direct_IO_write().
906 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
907 sector_t iblock, struct buffer_head *bh_result, int create)
911 /* We don't expect handle for direct IO */
912 WARN_ON_ONCE(ext4_journal_current_handle());
914 ret = ext4_get_block_trans(inode, iblock, bh_result,
915 EXT4_GET_BLOCKS_IO_CREATE_EXT);
918 * Mark inode as having pending DIO writes to unwritten extents.
919 * ext4_direct_IO_write() checks this flag and converts extents to
922 if (!ret && buffer_unwritten(bh_result))
923 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
928 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
929 struct buffer_head *bh_result, int create)
933 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
934 inode->i_ino, create);
935 /* We don't expect handle for direct IO */
936 WARN_ON_ONCE(ext4_journal_current_handle());
938 ret = _ext4_get_block(inode, iblock, bh_result, 0);
940 * Blocks should have been preallocated! ext4_file_write_iter() checks
943 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
950 * `handle' can be NULL if create is zero
952 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
953 ext4_lblk_t block, int map_flags)
955 struct ext4_map_blocks map;
956 struct buffer_head *bh;
957 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
960 J_ASSERT(handle != NULL || create == 0);
964 err = ext4_map_blocks(handle, inode, &map, map_flags);
967 return create ? ERR_PTR(-ENOSPC) : NULL;
971 bh = sb_getblk(inode->i_sb, map.m_pblk);
973 return ERR_PTR(-ENOMEM);
974 if (map.m_flags & EXT4_MAP_NEW) {
975 J_ASSERT(create != 0);
976 J_ASSERT(handle != NULL);
979 * Now that we do not always journal data, we should
980 * keep in mind whether this should always journal the
981 * new buffer as metadata. For now, regular file
982 * writes use ext4_get_block instead, so it's not a
986 BUFFER_TRACE(bh, "call get_create_access");
987 err = ext4_journal_get_create_access(handle, bh);
992 if (!buffer_uptodate(bh)) {
993 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
994 set_buffer_uptodate(bh);
997 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
998 err = ext4_handle_dirty_metadata(handle, inode, bh);
1002 BUFFER_TRACE(bh, "not a new buffer");
1006 return ERR_PTR(err);
1009 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1010 ext4_lblk_t block, int map_flags)
1012 struct buffer_head *bh;
1014 bh = ext4_getblk(handle, inode, block, map_flags);
1017 if (!bh || buffer_uptodate(bh))
1019 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1021 if (buffer_uptodate(bh))
1024 return ERR_PTR(-EIO);
1027 /* Read a contiguous batch of blocks. */
1028 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1029 bool wait, struct buffer_head **bhs)
1033 for (i = 0; i < bh_count; i++) {
1034 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1035 if (IS_ERR(bhs[i])) {
1036 err = PTR_ERR(bhs[i]);
1042 for (i = 0; i < bh_count; i++)
1043 /* Note that NULL bhs[i] is valid because of holes. */
1044 if (bhs[i] && !buffer_uptodate(bhs[i]))
1045 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1051 for (i = 0; i < bh_count; i++)
1053 wait_on_buffer(bhs[i]);
1055 for (i = 0; i < bh_count; i++) {
1056 if (bhs[i] && !buffer_uptodate(bhs[i])) {
1064 for (i = 0; i < bh_count; i++) {
1071 int ext4_walk_page_buffers(handle_t *handle,
1072 struct buffer_head *head,
1076 int (*fn)(handle_t *handle,
1077 struct buffer_head *bh))
1079 struct buffer_head *bh;
1080 unsigned block_start, block_end;
1081 unsigned blocksize = head->b_size;
1083 struct buffer_head *next;
1085 for (bh = head, block_start = 0;
1086 ret == 0 && (bh != head || !block_start);
1087 block_start = block_end, bh = next) {
1088 next = bh->b_this_page;
1089 block_end = block_start + blocksize;
1090 if (block_end <= from || block_start >= to) {
1091 if (partial && !buffer_uptodate(bh))
1095 err = (*fn)(handle, bh);
1103 * To preserve ordering, it is essential that the hole instantiation and
1104 * the data write be encapsulated in a single transaction. We cannot
1105 * close off a transaction and start a new one between the ext4_get_block()
1106 * and the commit_write(). So doing the jbd2_journal_start at the start of
1107 * prepare_write() is the right place.
1109 * Also, this function can nest inside ext4_writepage(). In that case, we
1110 * *know* that ext4_writepage() has generated enough buffer credits to do the
1111 * whole page. So we won't block on the journal in that case, which is good,
1112 * because the caller may be PF_MEMALLOC.
1114 * By accident, ext4 can be reentered when a transaction is open via
1115 * quota file writes. If we were to commit the transaction while thus
1116 * reentered, there can be a deadlock - we would be holding a quota
1117 * lock, and the commit would never complete if another thread had a
1118 * transaction open and was blocking on the quota lock - a ranking
1121 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1122 * will _not_ run commit under these circumstances because handle->h_ref
1123 * is elevated. We'll still have enough credits for the tiny quotafile
1126 int do_journal_get_write_access(handle_t *handle,
1127 struct buffer_head *bh)
1129 int dirty = buffer_dirty(bh);
1132 if (!buffer_mapped(bh) || buffer_freed(bh))
1135 * __block_write_begin() could have dirtied some buffers. Clean
1136 * the dirty bit as jbd2_journal_get_write_access() could complain
1137 * otherwise about fs integrity issues. Setting of the dirty bit
1138 * by __block_write_begin() isn't a real problem here as we clear
1139 * the bit before releasing a page lock and thus writeback cannot
1140 * ever write the buffer.
1143 clear_buffer_dirty(bh);
1144 BUFFER_TRACE(bh, "get write access");
1145 ret = ext4_journal_get_write_access(handle, bh);
1147 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1151 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1152 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1153 get_block_t *get_block)
1155 unsigned from = pos & (PAGE_SIZE - 1);
1156 unsigned to = from + len;
1157 struct inode *inode = page->mapping->host;
1158 unsigned block_start, block_end;
1161 unsigned blocksize = inode->i_sb->s_blocksize;
1163 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1164 bool decrypt = false;
1166 BUG_ON(!PageLocked(page));
1167 BUG_ON(from > PAGE_SIZE);
1168 BUG_ON(to > PAGE_SIZE);
1171 if (!page_has_buffers(page))
1172 create_empty_buffers(page, blocksize, 0);
1173 head = page_buffers(page);
1174 bbits = ilog2(blocksize);
1175 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1177 for (bh = head, block_start = 0; bh != head || !block_start;
1178 block++, block_start = block_end, bh = bh->b_this_page) {
1179 block_end = block_start + blocksize;
1180 if (block_end <= from || block_start >= to) {
1181 if (PageUptodate(page)) {
1182 if (!buffer_uptodate(bh))
1183 set_buffer_uptodate(bh);
1188 clear_buffer_new(bh);
1189 if (!buffer_mapped(bh)) {
1190 WARN_ON(bh->b_size != blocksize);
1191 err = get_block(inode, block, bh, 1);
1194 if (buffer_new(bh)) {
1195 if (PageUptodate(page)) {
1196 clear_buffer_new(bh);
1197 set_buffer_uptodate(bh);
1198 mark_buffer_dirty(bh);
1201 if (block_end > to || block_start < from)
1202 zero_user_segments(page, to, block_end,
1207 if (PageUptodate(page)) {
1208 if (!buffer_uptodate(bh))
1209 set_buffer_uptodate(bh);
1212 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1213 !buffer_unwritten(bh) &&
1214 (block_start < from || block_end > to)) {
1215 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1217 decrypt = ext4_encrypted_inode(inode) &&
1218 S_ISREG(inode->i_mode);
1222 * If we issued read requests, let them complete.
1224 while (wait_bh > wait) {
1225 wait_on_buffer(*--wait_bh);
1226 if (!buffer_uptodate(*wait_bh))
1230 page_zero_new_buffers(page, from, to);
1232 err = fscrypt_decrypt_page(page->mapping->host, page,
1233 PAGE_SIZE, 0, page->index);
1238 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1239 loff_t pos, unsigned len, unsigned flags,
1240 struct page **pagep, void **fsdata)
1242 struct inode *inode = mapping->host;
1243 int ret, needed_blocks;
1250 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1253 trace_ext4_write_begin(inode, pos, len, flags);
1255 * Reserve one block more for addition to orphan list in case
1256 * we allocate blocks but write fails for some reason
1258 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1259 index = pos >> PAGE_SHIFT;
1260 from = pos & (PAGE_SIZE - 1);
1263 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1264 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1273 * grab_cache_page_write_begin() can take a long time if the
1274 * system is thrashing due to memory pressure, or if the page
1275 * is being written back. So grab it first before we start
1276 * the transaction handle. This also allows us to allocate
1277 * the page (if needed) without using GFP_NOFS.
1280 page = grab_cache_page_write_begin(mapping, index, flags);
1286 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1287 if (IS_ERR(handle)) {
1289 return PTR_ERR(handle);
1293 if (page->mapping != mapping) {
1294 /* The page got truncated from under us */
1297 ext4_journal_stop(handle);
1300 /* In case writeback began while the page was unlocked */
1301 wait_for_stable_page(page);
1303 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1304 if (ext4_should_dioread_nolock(inode))
1305 ret = ext4_block_write_begin(page, pos, len,
1306 ext4_get_block_unwritten);
1308 ret = ext4_block_write_begin(page, pos, len,
1311 if (ext4_should_dioread_nolock(inode))
1312 ret = __block_write_begin(page, pos, len,
1313 ext4_get_block_unwritten);
1315 ret = __block_write_begin(page, pos, len, ext4_get_block);
1317 if (!ret && ext4_should_journal_data(inode)) {
1318 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1320 do_journal_get_write_access);
1326 * __block_write_begin may have instantiated a few blocks
1327 * outside i_size. Trim these off again. Don't need
1328 * i_size_read because we hold i_mutex.
1330 * Add inode to orphan list in case we crash before
1333 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1334 ext4_orphan_add(handle, inode);
1336 ext4_journal_stop(handle);
1337 if (pos + len > inode->i_size) {
1338 ext4_truncate_failed_write(inode);
1340 * If truncate failed early the inode might
1341 * still be on the orphan list; we need to
1342 * make sure the inode is removed from the
1343 * orphan list in that case.
1346 ext4_orphan_del(NULL, inode);
1349 if (ret == -ENOSPC &&
1350 ext4_should_retry_alloc(inode->i_sb, &retries))
1359 /* For write_end() in data=journal mode */
1360 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1363 if (!buffer_mapped(bh) || buffer_freed(bh))
1365 set_buffer_uptodate(bh);
1366 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1367 clear_buffer_meta(bh);
1368 clear_buffer_prio(bh);
1373 * We need to pick up the new inode size which generic_commit_write gave us
1374 * `file' can be NULL - eg, when called from page_symlink().
1376 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1377 * buffers are managed internally.
1379 static int ext4_write_end(struct file *file,
1380 struct address_space *mapping,
1381 loff_t pos, unsigned len, unsigned copied,
1382 struct page *page, void *fsdata)
1384 handle_t *handle = ext4_journal_current_handle();
1385 struct inode *inode = mapping->host;
1386 loff_t old_size = inode->i_size;
1388 int i_size_changed = 0;
1389 int inline_data = ext4_has_inline_data(inode);
1391 trace_ext4_write_end(inode, pos, len, copied);
1393 ret = ext4_write_inline_data_end(inode, pos, len,
1402 copied = block_write_end(file, mapping, pos,
1403 len, copied, page, fsdata);
1405 * it's important to update i_size while still holding page lock:
1406 * page writeout could otherwise come in and zero beyond i_size.
1408 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1413 pagecache_isize_extended(inode, old_size, pos);
1415 * Don't mark the inode dirty under page lock. First, it unnecessarily
1416 * makes the holding time of page lock longer. Second, it forces lock
1417 * ordering of page lock and transaction start for journaling
1420 if (i_size_changed || inline_data)
1421 ext4_mark_inode_dirty(handle, inode);
1423 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1424 /* if we have allocated more blocks and copied
1425 * less. We will have blocks allocated outside
1426 * inode->i_size. So truncate them
1428 ext4_orphan_add(handle, inode);
1430 ret2 = ext4_journal_stop(handle);
1434 if (pos + len > inode->i_size) {
1435 ext4_truncate_failed_write(inode);
1437 * If truncate failed early the inode might still be
1438 * on the orphan list; we need to make sure the inode
1439 * is removed from the orphan list in that case.
1442 ext4_orphan_del(NULL, inode);
1445 return ret ? ret : copied;
1449 * This is a private version of page_zero_new_buffers() which doesn't
1450 * set the buffer to be dirty, since in data=journalled mode we need
1451 * to call ext4_handle_dirty_metadata() instead.
1453 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1455 unsigned from, unsigned to)
1457 unsigned int block_start = 0, block_end;
1458 struct buffer_head *head, *bh;
1460 bh = head = page_buffers(page);
1462 block_end = block_start + bh->b_size;
1463 if (buffer_new(bh)) {
1464 if (block_end > from && block_start < to) {
1465 if (!PageUptodate(page)) {
1466 unsigned start, size;
1468 start = max(from, block_start);
1469 size = min(to, block_end) - start;
1471 zero_user(page, start, size);
1472 write_end_fn(handle, bh);
1474 clear_buffer_new(bh);
1477 block_start = block_end;
1478 bh = bh->b_this_page;
1479 } while (bh != head);
1482 static int ext4_journalled_write_end(struct file *file,
1483 struct address_space *mapping,
1484 loff_t pos, unsigned len, unsigned copied,
1485 struct page *page, void *fsdata)
1487 handle_t *handle = ext4_journal_current_handle();
1488 struct inode *inode = mapping->host;
1489 loff_t old_size = inode->i_size;
1493 int size_changed = 0;
1494 int inline_data = ext4_has_inline_data(inode);
1496 trace_ext4_journalled_write_end(inode, pos, len, copied);
1497 from = pos & (PAGE_SIZE - 1);
1500 BUG_ON(!ext4_handle_valid(handle));
1503 ret = ext4_write_inline_data_end(inode, pos, len,
1511 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1513 ext4_journalled_zero_new_buffers(handle, page, from, to);
1515 if (unlikely(copied < len))
1516 ext4_journalled_zero_new_buffers(handle, page,
1518 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1519 from + copied, &partial,
1522 SetPageUptodate(page);
1524 size_changed = ext4_update_inode_size(inode, pos + copied);
1525 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1526 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1531 pagecache_isize_extended(inode, old_size, pos);
1533 if (size_changed || inline_data) {
1534 ret2 = ext4_mark_inode_dirty(handle, inode);
1539 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1540 /* if we have allocated more blocks and copied
1541 * less. We will have blocks allocated outside
1542 * inode->i_size. So truncate them
1544 ext4_orphan_add(handle, inode);
1547 ret2 = ext4_journal_stop(handle);
1550 if (pos + len > inode->i_size) {
1551 ext4_truncate_failed_write(inode);
1553 * If truncate failed early the inode might still be
1554 * on the orphan list; we need to make sure the inode
1555 * is removed from the orphan list in that case.
1558 ext4_orphan_del(NULL, inode);
1561 return ret ? ret : copied;
1565 * Reserve space for a single cluster
1567 static int ext4_da_reserve_space(struct inode *inode)
1569 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1570 struct ext4_inode_info *ei = EXT4_I(inode);
1574 * We will charge metadata quota at writeout time; this saves
1575 * us from metadata over-estimation, though we may go over by
1576 * a small amount in the end. Here we just reserve for data.
1578 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1582 spin_lock(&ei->i_block_reservation_lock);
1583 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1584 spin_unlock(&ei->i_block_reservation_lock);
1585 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1588 ei->i_reserved_data_blocks++;
1589 trace_ext4_da_reserve_space(inode);
1590 spin_unlock(&ei->i_block_reservation_lock);
1592 return 0; /* success */
1595 void ext4_da_release_space(struct inode *inode, int to_free)
1597 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1598 struct ext4_inode_info *ei = EXT4_I(inode);
1601 return; /* Nothing to release, exit */
1603 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1605 trace_ext4_da_release_space(inode, to_free);
1606 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1608 * if there aren't enough reserved blocks, then the
1609 * counter is messed up somewhere. Since this
1610 * function is called from invalidate page, it's
1611 * harmless to return without any action.
1613 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1614 "ino %lu, to_free %d with only %d reserved "
1615 "data blocks", inode->i_ino, to_free,
1616 ei->i_reserved_data_blocks);
1618 to_free = ei->i_reserved_data_blocks;
1620 ei->i_reserved_data_blocks -= to_free;
1622 /* update fs dirty data blocks counter */
1623 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1625 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1627 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1630 static void ext4_da_page_release_reservation(struct page *page,
1631 unsigned int offset,
1632 unsigned int length)
1634 int contiguous_blks = 0;
1635 struct buffer_head *head, *bh;
1636 unsigned int curr_off = 0;
1637 struct inode *inode = page->mapping->host;
1638 unsigned int stop = offset + length;
1641 BUG_ON(stop > PAGE_SIZE || stop < length);
1643 head = page_buffers(page);
1646 unsigned int next_off = curr_off + bh->b_size;
1648 if (next_off > stop)
1651 if ((offset <= curr_off) && (buffer_delay(bh))) {
1653 clear_buffer_delay(bh);
1654 } else if (contiguous_blks) {
1655 lblk = page->index <<
1656 (PAGE_SHIFT - inode->i_blkbits);
1657 lblk += (curr_off >> inode->i_blkbits) -
1659 ext4_es_remove_blks(inode, lblk, contiguous_blks);
1660 contiguous_blks = 0;
1662 curr_off = next_off;
1663 } while ((bh = bh->b_this_page) != head);
1665 if (contiguous_blks) {
1666 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1667 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1668 ext4_es_remove_blks(inode, lblk, contiguous_blks);
1674 * Delayed allocation stuff
1677 struct mpage_da_data {
1678 struct inode *inode;
1679 struct writeback_control *wbc;
1681 pgoff_t first_page; /* The first page to write */
1682 pgoff_t next_page; /* Current page to examine */
1683 pgoff_t last_page; /* Last page to examine */
1685 * Extent to map - this can be after first_page because that can be
1686 * fully mapped. We somewhat abuse m_flags to store whether the extent
1687 * is delalloc or unwritten.
1689 struct ext4_map_blocks map;
1690 struct ext4_io_submit io_submit; /* IO submission data */
1691 unsigned int do_map:1;
1694 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1699 struct pagevec pvec;
1700 struct inode *inode = mpd->inode;
1701 struct address_space *mapping = inode->i_mapping;
1703 /* This is necessary when next_page == 0. */
1704 if (mpd->first_page >= mpd->next_page)
1707 index = mpd->first_page;
1708 end = mpd->next_page - 1;
1710 ext4_lblk_t start, last;
1711 start = index << (PAGE_SHIFT - inode->i_blkbits);
1712 last = end << (PAGE_SHIFT - inode->i_blkbits);
1713 ext4_es_remove_extent(inode, start, last - start + 1);
1716 pagevec_init(&pvec);
1717 while (index <= end) {
1718 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1721 for (i = 0; i < nr_pages; i++) {
1722 struct page *page = pvec.pages[i];
1724 BUG_ON(!PageLocked(page));
1725 BUG_ON(PageWriteback(page));
1727 if (page_mapped(page))
1728 clear_page_dirty_for_io(page);
1729 block_invalidatepage(page, 0, PAGE_SIZE);
1730 ClearPageUptodate(page);
1734 pagevec_release(&pvec);
1738 static void ext4_print_free_blocks(struct inode *inode)
1740 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1741 struct super_block *sb = inode->i_sb;
1742 struct ext4_inode_info *ei = EXT4_I(inode);
1744 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1745 EXT4_C2B(EXT4_SB(inode->i_sb),
1746 ext4_count_free_clusters(sb)));
1747 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1748 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1749 (long long) EXT4_C2B(EXT4_SB(sb),
1750 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1751 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1752 (long long) EXT4_C2B(EXT4_SB(sb),
1753 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1754 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1755 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1756 ei->i_reserved_data_blocks);
1760 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1762 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1766 * ext4_insert_delayed_block - adds a delayed block to the extents status
1767 * tree, incrementing the reserved cluster/block
1768 * count or making a pending reservation
1771 * @inode - file containing the newly added block
1772 * @lblk - logical block to be added
1774 * Returns 0 on success, negative error code on failure.
1776 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1778 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1780 bool allocated = false;
1783 * If the cluster containing lblk is shared with a delayed,
1784 * written, or unwritten extent in a bigalloc file system, it's
1785 * already been accounted for and does not need to be reserved.
1786 * A pending reservation must be made for the cluster if it's
1787 * shared with a written or unwritten extent and doesn't already
1788 * have one. Written and unwritten extents can be purged from the
1789 * extents status tree if the system is under memory pressure, so
1790 * it's necessary to examine the extent tree if a search of the
1791 * extents status tree doesn't get a match.
1793 if (sbi->s_cluster_ratio == 1) {
1794 ret = ext4_da_reserve_space(inode);
1795 if (ret != 0) /* ENOSPC */
1797 } else { /* bigalloc */
1798 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1799 if (!ext4_es_scan_clu(inode,
1800 &ext4_es_is_mapped, lblk)) {
1801 ret = ext4_clu_mapped(inode,
1802 EXT4_B2C(sbi, lblk));
1806 ret = ext4_da_reserve_space(inode);
1807 if (ret != 0) /* ENOSPC */
1818 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1825 * This function is grabs code from the very beginning of
1826 * ext4_map_blocks, but assumes that the caller is from delayed write
1827 * time. This function looks up the requested blocks and sets the
1828 * buffer delay bit under the protection of i_data_sem.
1830 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1831 struct ext4_map_blocks *map,
1832 struct buffer_head *bh)
1834 struct extent_status es;
1836 sector_t invalid_block = ~((sector_t) 0xffff);
1837 #ifdef ES_AGGRESSIVE_TEST
1838 struct ext4_map_blocks orig_map;
1840 memcpy(&orig_map, map, sizeof(*map));
1843 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1847 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1848 "logical block %lu\n", inode->i_ino, map->m_len,
1849 (unsigned long) map->m_lblk);
1851 /* Lookup extent status tree firstly */
1852 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1853 if (ext4_es_is_hole(&es)) {
1855 down_read(&EXT4_I(inode)->i_data_sem);
1860 * Delayed extent could be allocated by fallocate.
1861 * So we need to check it.
1863 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1864 map_bh(bh, inode->i_sb, invalid_block);
1866 set_buffer_delay(bh);
1870 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1871 retval = es.es_len - (iblock - es.es_lblk);
1872 if (retval > map->m_len)
1873 retval = map->m_len;
1874 map->m_len = retval;
1875 if (ext4_es_is_written(&es))
1876 map->m_flags |= EXT4_MAP_MAPPED;
1877 else if (ext4_es_is_unwritten(&es))
1878 map->m_flags |= EXT4_MAP_UNWRITTEN;
1882 #ifdef ES_AGGRESSIVE_TEST
1883 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1889 * Try to see if we can get the block without requesting a new
1890 * file system block.
1892 down_read(&EXT4_I(inode)->i_data_sem);
1893 if (ext4_has_inline_data(inode))
1895 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1896 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1898 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1905 * XXX: __block_prepare_write() unmaps passed block,
1909 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1915 map_bh(bh, inode->i_sb, invalid_block);
1917 set_buffer_delay(bh);
1918 } else if (retval > 0) {
1920 unsigned int status;
1922 if (unlikely(retval != map->m_len)) {
1923 ext4_warning(inode->i_sb,
1924 "ES len assertion failed for inode "
1925 "%lu: retval %d != map->m_len %d",
1926 inode->i_ino, retval, map->m_len);
1930 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1931 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1932 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1933 map->m_pblk, status);
1939 up_read((&EXT4_I(inode)->i_data_sem));
1945 * This is a special get_block_t callback which is used by
1946 * ext4_da_write_begin(). It will either return mapped block or
1947 * reserve space for a single block.
1949 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1950 * We also have b_blocknr = -1 and b_bdev initialized properly
1952 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1953 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1954 * initialized properly.
1956 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1957 struct buffer_head *bh, int create)
1959 struct ext4_map_blocks map;
1962 BUG_ON(create == 0);
1963 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1965 map.m_lblk = iblock;
1969 * first, we need to know whether the block is allocated already
1970 * preallocated blocks are unmapped but should treated
1971 * the same as allocated blocks.
1973 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1977 map_bh(bh, inode->i_sb, map.m_pblk);
1978 ext4_update_bh_state(bh, map.m_flags);
1980 if (buffer_unwritten(bh)) {
1981 /* A delayed write to unwritten bh should be marked
1982 * new and mapped. Mapped ensures that we don't do
1983 * get_block multiple times when we write to the same
1984 * offset and new ensures that we do proper zero out
1985 * for partial write.
1988 set_buffer_mapped(bh);
1993 static int bget_one(handle_t *handle, struct buffer_head *bh)
1999 static int bput_one(handle_t *handle, struct buffer_head *bh)
2005 static int __ext4_journalled_writepage(struct page *page,
2008 struct address_space *mapping = page->mapping;
2009 struct inode *inode = mapping->host;
2010 struct buffer_head *page_bufs = NULL;
2011 handle_t *handle = NULL;
2012 int ret = 0, err = 0;
2013 int inline_data = ext4_has_inline_data(inode);
2014 struct buffer_head *inode_bh = NULL;
2016 ClearPageChecked(page);
2019 BUG_ON(page->index != 0);
2020 BUG_ON(len > ext4_get_max_inline_size(inode));
2021 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2022 if (inode_bh == NULL)
2025 page_bufs = page_buffers(page);
2030 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2034 * We need to release the page lock before we start the
2035 * journal, so grab a reference so the page won't disappear
2036 * out from under us.
2041 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2042 ext4_writepage_trans_blocks(inode));
2043 if (IS_ERR(handle)) {
2044 ret = PTR_ERR(handle);
2046 goto out_no_pagelock;
2048 BUG_ON(!ext4_handle_valid(handle));
2052 if (page->mapping != mapping) {
2053 /* The page got truncated from under us */
2054 ext4_journal_stop(handle);
2060 ret = ext4_mark_inode_dirty(handle, inode);
2062 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2063 do_journal_get_write_access);
2065 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2070 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2071 err = ext4_journal_stop(handle);
2075 if (!ext4_has_inline_data(inode))
2076 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2078 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2087 * Note that we don't need to start a transaction unless we're journaling data
2088 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2089 * need to file the inode to the transaction's list in ordered mode because if
2090 * we are writing back data added by write(), the inode is already there and if
2091 * we are writing back data modified via mmap(), no one guarantees in which
2092 * transaction the data will hit the disk. In case we are journaling data, we
2093 * cannot start transaction directly because transaction start ranks above page
2094 * lock so we have to do some magic.
2096 * This function can get called via...
2097 * - ext4_writepages after taking page lock (have journal handle)
2098 * - journal_submit_inode_data_buffers (no journal handle)
2099 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2100 * - grab_page_cache when doing write_begin (have journal handle)
2102 * We don't do any block allocation in this function. If we have page with
2103 * multiple blocks we need to write those buffer_heads that are mapped. This
2104 * is important for mmaped based write. So if we do with blocksize 1K
2105 * truncate(f, 1024);
2106 * a = mmap(f, 0, 4096);
2108 * truncate(f, 4096);
2109 * we have in the page first buffer_head mapped via page_mkwrite call back
2110 * but other buffer_heads would be unmapped but dirty (dirty done via the
2111 * do_wp_page). So writepage should write the first block. If we modify
2112 * the mmap area beyond 1024 we will again get a page_fault and the
2113 * page_mkwrite callback will do the block allocation and mark the
2114 * buffer_heads mapped.
2116 * We redirty the page if we have any buffer_heads that is either delay or
2117 * unwritten in the page.
2119 * We can get recursively called as show below.
2121 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2124 * But since we don't do any block allocation we should not deadlock.
2125 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2127 static int ext4_writepage(struct page *page,
2128 struct writeback_control *wbc)
2133 struct buffer_head *page_bufs = NULL;
2134 struct inode *inode = page->mapping->host;
2135 struct ext4_io_submit io_submit;
2136 bool keep_towrite = false;
2138 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2139 ext4_invalidatepage(page, 0, PAGE_SIZE);
2144 trace_ext4_writepage(page);
2145 size = i_size_read(inode);
2146 if (page->index == size >> PAGE_SHIFT)
2147 len = size & ~PAGE_MASK;
2151 page_bufs = page_buffers(page);
2153 * We cannot do block allocation or other extent handling in this
2154 * function. If there are buffers needing that, we have to redirty
2155 * the page. But we may reach here when we do a journal commit via
2156 * journal_submit_inode_data_buffers() and in that case we must write
2157 * allocated buffers to achieve data=ordered mode guarantees.
2159 * Also, if there is only one buffer per page (the fs block
2160 * size == the page size), if one buffer needs block
2161 * allocation or needs to modify the extent tree to clear the
2162 * unwritten flag, we know that the page can't be written at
2163 * all, so we might as well refuse the write immediately.
2164 * Unfortunately if the block size != page size, we can't as
2165 * easily detect this case using ext4_walk_page_buffers(), but
2166 * for the extremely common case, this is an optimization that
2167 * skips a useless round trip through ext4_bio_write_page().
2169 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2170 ext4_bh_delay_or_unwritten)) {
2171 redirty_page_for_writepage(wbc, page);
2172 if ((current->flags & PF_MEMALLOC) ||
2173 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2175 * For memory cleaning there's no point in writing only
2176 * some buffers. So just bail out. Warn if we came here
2177 * from direct reclaim.
2179 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2184 keep_towrite = true;
2187 if (PageChecked(page) && ext4_should_journal_data(inode))
2189 * It's mmapped pagecache. Add buffers and journal it. There
2190 * doesn't seem much point in redirtying the page here.
2192 return __ext4_journalled_writepage(page, len);
2194 ext4_io_submit_init(&io_submit, wbc);
2195 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2196 if (!io_submit.io_end) {
2197 redirty_page_for_writepage(wbc, page);
2201 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2202 ext4_io_submit(&io_submit);
2203 /* Drop io_end reference we got from init */
2204 ext4_put_io_end_defer(io_submit.io_end);
2208 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2214 BUG_ON(page->index != mpd->first_page);
2215 clear_page_dirty_for_io(page);
2217 * We have to be very careful here! Nothing protects writeback path
2218 * against i_size changes and the page can be writeably mapped into
2219 * page tables. So an application can be growing i_size and writing
2220 * data through mmap while writeback runs. clear_page_dirty_for_io()
2221 * write-protects our page in page tables and the page cannot get
2222 * written to again until we release page lock. So only after
2223 * clear_page_dirty_for_io() we are safe to sample i_size for
2224 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2225 * on the barrier provided by TestClearPageDirty in
2226 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2227 * after page tables are updated.
2229 size = i_size_read(mpd->inode);
2230 if (page->index == size >> PAGE_SHIFT)
2231 len = size & ~PAGE_MASK;
2234 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2236 mpd->wbc->nr_to_write--;
2242 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2245 * mballoc gives us at most this number of blocks...
2246 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2247 * The rest of mballoc seems to handle chunks up to full group size.
2249 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2252 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2254 * @mpd - extent of blocks
2255 * @lblk - logical number of the block in the file
2256 * @bh - buffer head we want to add to the extent
2258 * The function is used to collect contig. blocks in the same state. If the
2259 * buffer doesn't require mapping for writeback and we haven't started the
2260 * extent of buffers to map yet, the function returns 'true' immediately - the
2261 * caller can write the buffer right away. Otherwise the function returns true
2262 * if the block has been added to the extent, false if the block couldn't be
2265 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2266 struct buffer_head *bh)
2268 struct ext4_map_blocks *map = &mpd->map;
2270 /* Buffer that doesn't need mapping for writeback? */
2271 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2272 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2273 /* So far no extent to map => we write the buffer right away */
2274 if (map->m_len == 0)
2279 /* First block in the extent? */
2280 if (map->m_len == 0) {
2281 /* We cannot map unless handle is started... */
2286 map->m_flags = bh->b_state & BH_FLAGS;
2290 /* Don't go larger than mballoc is willing to allocate */
2291 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2294 /* Can we merge the block to our big extent? */
2295 if (lblk == map->m_lblk + map->m_len &&
2296 (bh->b_state & BH_FLAGS) == map->m_flags) {
2304 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2306 * @mpd - extent of blocks for mapping
2307 * @head - the first buffer in the page
2308 * @bh - buffer we should start processing from
2309 * @lblk - logical number of the block in the file corresponding to @bh
2311 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2312 * the page for IO if all buffers in this page were mapped and there's no
2313 * accumulated extent of buffers to map or add buffers in the page to the
2314 * extent of buffers to map. The function returns 1 if the caller can continue
2315 * by processing the next page, 0 if it should stop adding buffers to the
2316 * extent to map because we cannot extend it anymore. It can also return value
2317 * < 0 in case of error during IO submission.
2319 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2320 struct buffer_head *head,
2321 struct buffer_head *bh,
2324 struct inode *inode = mpd->inode;
2326 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2327 >> inode->i_blkbits;
2330 BUG_ON(buffer_locked(bh));
2332 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2333 /* Found extent to map? */
2336 /* Buffer needs mapping and handle is not started? */
2339 /* Everything mapped so far and we hit EOF */
2342 } while (lblk++, (bh = bh->b_this_page) != head);
2343 /* So far everything mapped? Submit the page for IO. */
2344 if (mpd->map.m_len == 0) {
2345 err = mpage_submit_page(mpd, head->b_page);
2349 return lblk < blocks;
2353 * mpage_map_buffers - update buffers corresponding to changed extent and
2354 * submit fully mapped pages for IO
2356 * @mpd - description of extent to map, on return next extent to map
2358 * Scan buffers corresponding to changed extent (we expect corresponding pages
2359 * to be already locked) and update buffer state according to new extent state.
2360 * We map delalloc buffers to their physical location, clear unwritten bits,
2361 * and mark buffers as uninit when we perform writes to unwritten extents
2362 * and do extent conversion after IO is finished. If the last page is not fully
2363 * mapped, we update @map to the next extent in the last page that needs
2364 * mapping. Otherwise we submit the page for IO.
2366 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2368 struct pagevec pvec;
2370 struct inode *inode = mpd->inode;
2371 struct buffer_head *head, *bh;
2372 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2378 start = mpd->map.m_lblk >> bpp_bits;
2379 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2380 lblk = start << bpp_bits;
2381 pblock = mpd->map.m_pblk;
2383 pagevec_init(&pvec);
2384 while (start <= end) {
2385 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2389 for (i = 0; i < nr_pages; i++) {
2390 struct page *page = pvec.pages[i];
2392 bh = head = page_buffers(page);
2394 if (lblk < mpd->map.m_lblk)
2396 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2398 * Buffer after end of mapped extent.
2399 * Find next buffer in the page to map.
2402 mpd->map.m_flags = 0;
2404 * FIXME: If dioread_nolock supports
2405 * blocksize < pagesize, we need to make
2406 * sure we add size mapped so far to
2407 * io_end->size as the following call
2408 * can submit the page for IO.
2410 err = mpage_process_page_bufs(mpd, head,
2412 pagevec_release(&pvec);
2417 if (buffer_delay(bh)) {
2418 clear_buffer_delay(bh);
2419 bh->b_blocknr = pblock++;
2421 clear_buffer_unwritten(bh);
2422 } while (lblk++, (bh = bh->b_this_page) != head);
2425 * FIXME: This is going to break if dioread_nolock
2426 * supports blocksize < pagesize as we will try to
2427 * convert potentially unmapped parts of inode.
2429 mpd->io_submit.io_end->size += PAGE_SIZE;
2430 /* Page fully mapped - let IO run! */
2431 err = mpage_submit_page(mpd, page);
2433 pagevec_release(&pvec);
2437 pagevec_release(&pvec);
2439 /* Extent fully mapped and matches with page boundary. We are done. */
2441 mpd->map.m_flags = 0;
2445 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2447 struct inode *inode = mpd->inode;
2448 struct ext4_map_blocks *map = &mpd->map;
2449 int get_blocks_flags;
2450 int err, dioread_nolock;
2452 trace_ext4_da_write_pages_extent(inode, map);
2454 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2455 * to convert an unwritten extent to be initialized (in the case
2456 * where we have written into one or more preallocated blocks). It is
2457 * possible that we're going to need more metadata blocks than
2458 * previously reserved. However we must not fail because we're in
2459 * writeback and there is nothing we can do about it so it might result
2460 * in data loss. So use reserved blocks to allocate metadata if
2463 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2464 * the blocks in question are delalloc blocks. This indicates
2465 * that the blocks and quotas has already been checked when
2466 * the data was copied into the page cache.
2468 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2469 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2470 EXT4_GET_BLOCKS_IO_SUBMIT;
2471 dioread_nolock = ext4_should_dioread_nolock(inode);
2473 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2474 if (map->m_flags & (1 << BH_Delay))
2475 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2477 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2480 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2481 if (!mpd->io_submit.io_end->handle &&
2482 ext4_handle_valid(handle)) {
2483 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2484 handle->h_rsv_handle = NULL;
2486 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2489 BUG_ON(map->m_len == 0);
2494 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2495 * mpd->len and submit pages underlying it for IO
2497 * @handle - handle for journal operations
2498 * @mpd - extent to map
2499 * @give_up_on_write - we set this to true iff there is a fatal error and there
2500 * is no hope of writing the data. The caller should discard
2501 * dirty pages to avoid infinite loops.
2503 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2504 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2505 * them to initialized or split the described range from larger unwritten
2506 * extent. Note that we need not map all the described range since allocation
2507 * can return less blocks or the range is covered by more unwritten extents. We
2508 * cannot map more because we are limited by reserved transaction credits. On
2509 * the other hand we always make sure that the last touched page is fully
2510 * mapped so that it can be written out (and thus forward progress is
2511 * guaranteed). After mapping we submit all mapped pages for IO.
2513 static int mpage_map_and_submit_extent(handle_t *handle,
2514 struct mpage_da_data *mpd,
2515 bool *give_up_on_write)
2517 struct inode *inode = mpd->inode;
2518 struct ext4_map_blocks *map = &mpd->map;
2523 mpd->io_submit.io_end->offset =
2524 ((loff_t)map->m_lblk) << inode->i_blkbits;
2526 err = mpage_map_one_extent(handle, mpd);
2528 struct super_block *sb = inode->i_sb;
2530 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2531 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2532 goto invalidate_dirty_pages;
2534 * Let the uper layers retry transient errors.
2535 * In the case of ENOSPC, if ext4_count_free_blocks()
2536 * is non-zero, a commit should free up blocks.
2538 if ((err == -ENOMEM) ||
2539 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2541 goto update_disksize;
2544 ext4_msg(sb, KERN_CRIT,
2545 "Delayed block allocation failed for "
2546 "inode %lu at logical offset %llu with"
2547 " max blocks %u with error %d",
2549 (unsigned long long)map->m_lblk,
2550 (unsigned)map->m_len, -err);
2551 ext4_msg(sb, KERN_CRIT,
2552 "This should not happen!! Data will "
2555 ext4_print_free_blocks(inode);
2556 invalidate_dirty_pages:
2557 *give_up_on_write = true;
2562 * Update buffer state, submit mapped pages, and get us new
2565 err = mpage_map_and_submit_buffers(mpd);
2567 goto update_disksize;
2568 } while (map->m_len);
2572 * Update on-disk size after IO is submitted. Races with
2573 * truncate are avoided by checking i_size under i_data_sem.
2575 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2576 if (disksize > EXT4_I(inode)->i_disksize) {
2580 down_write(&EXT4_I(inode)->i_data_sem);
2581 i_size = i_size_read(inode);
2582 if (disksize > i_size)
2584 if (disksize > EXT4_I(inode)->i_disksize)
2585 EXT4_I(inode)->i_disksize = disksize;
2586 up_write(&EXT4_I(inode)->i_data_sem);
2587 err2 = ext4_mark_inode_dirty(handle, inode);
2589 ext4_error(inode->i_sb,
2590 "Failed to mark inode %lu dirty",
2599 * Calculate the total number of credits to reserve for one writepages
2600 * iteration. This is called from ext4_writepages(). We map an extent of
2601 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2602 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2603 * bpp - 1 blocks in bpp different extents.
2605 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2607 int bpp = ext4_journal_blocks_per_page(inode);
2609 return ext4_meta_trans_blocks(inode,
2610 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2614 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2615 * and underlying extent to map
2617 * @mpd - where to look for pages
2619 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2620 * IO immediately. When we find a page which isn't mapped we start accumulating
2621 * extent of buffers underlying these pages that needs mapping (formed by
2622 * either delayed or unwritten buffers). We also lock the pages containing
2623 * these buffers. The extent found is returned in @mpd structure (starting at
2624 * mpd->lblk with length mpd->len blocks).
2626 * Note that this function can attach bios to one io_end structure which are
2627 * neither logically nor physically contiguous. Although it may seem as an
2628 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2629 * case as we need to track IO to all buffers underlying a page in one io_end.
2631 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2633 struct address_space *mapping = mpd->inode->i_mapping;
2634 struct pagevec pvec;
2635 unsigned int nr_pages;
2636 long left = mpd->wbc->nr_to_write;
2637 pgoff_t index = mpd->first_page;
2638 pgoff_t end = mpd->last_page;
2641 int blkbits = mpd->inode->i_blkbits;
2643 struct buffer_head *head;
2645 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2646 tag = PAGECACHE_TAG_TOWRITE;
2648 tag = PAGECACHE_TAG_DIRTY;
2650 pagevec_init(&pvec);
2652 mpd->next_page = index;
2653 while (index <= end) {
2654 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2659 for (i = 0; i < nr_pages; i++) {
2660 struct page *page = pvec.pages[i];
2663 * Accumulated enough dirty pages? This doesn't apply
2664 * to WB_SYNC_ALL mode. For integrity sync we have to
2665 * keep going because someone may be concurrently
2666 * dirtying pages, and we might have synced a lot of
2667 * newly appeared dirty pages, but have not synced all
2668 * of the old dirty pages.
2670 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2673 /* If we can't merge this page, we are done. */
2674 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2679 * If the page is no longer dirty, or its mapping no
2680 * longer corresponds to inode we are writing (which
2681 * means it has been truncated or invalidated), or the
2682 * page is already under writeback and we are not doing
2683 * a data integrity writeback, skip the page
2685 if (!PageDirty(page) ||
2686 (PageWriteback(page) &&
2687 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2688 unlikely(page->mapping != mapping)) {
2693 wait_on_page_writeback(page);
2694 BUG_ON(PageWriteback(page));
2696 if (mpd->map.m_len == 0)
2697 mpd->first_page = page->index;
2698 mpd->next_page = page->index + 1;
2699 /* Add all dirty buffers to mpd */
2700 lblk = ((ext4_lblk_t)page->index) <<
2701 (PAGE_SHIFT - blkbits);
2702 head = page_buffers(page);
2703 err = mpage_process_page_bufs(mpd, head, head, lblk);
2709 pagevec_release(&pvec);
2714 pagevec_release(&pvec);
2718 static int ext4_writepages(struct address_space *mapping,
2719 struct writeback_control *wbc)
2721 pgoff_t writeback_index = 0;
2722 long nr_to_write = wbc->nr_to_write;
2723 int range_whole = 0;
2725 handle_t *handle = NULL;
2726 struct mpage_da_data mpd;
2727 struct inode *inode = mapping->host;
2728 int needed_blocks, rsv_blocks = 0, ret = 0;
2729 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2731 struct blk_plug plug;
2732 bool give_up_on_write = false;
2734 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2737 percpu_down_read(&sbi->s_journal_flag_rwsem);
2738 trace_ext4_writepages(inode, wbc);
2741 * No pages to write? This is mainly a kludge to avoid starting
2742 * a transaction for special inodes like journal inode on last iput()
2743 * because that could violate lock ordering on umount
2745 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2746 goto out_writepages;
2748 if (ext4_should_journal_data(inode)) {
2749 ret = generic_writepages(mapping, wbc);
2750 goto out_writepages;
2754 * If the filesystem has aborted, it is read-only, so return
2755 * right away instead of dumping stack traces later on that
2756 * will obscure the real source of the problem. We test
2757 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2758 * the latter could be true if the filesystem is mounted
2759 * read-only, and in that case, ext4_writepages should
2760 * *never* be called, so if that ever happens, we would want
2763 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2764 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2766 goto out_writepages;
2769 if (ext4_should_dioread_nolock(inode)) {
2771 * We may need to convert up to one extent per block in
2772 * the page and we may dirty the inode.
2774 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2775 PAGE_SIZE >> inode->i_blkbits);
2779 * If we have inline data and arrive here, it means that
2780 * we will soon create the block for the 1st page, so
2781 * we'd better clear the inline data here.
2783 if (ext4_has_inline_data(inode)) {
2784 /* Just inode will be modified... */
2785 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2786 if (IS_ERR(handle)) {
2787 ret = PTR_ERR(handle);
2788 goto out_writepages;
2790 BUG_ON(ext4_test_inode_state(inode,
2791 EXT4_STATE_MAY_INLINE_DATA));
2792 ext4_destroy_inline_data(handle, inode);
2793 ext4_journal_stop(handle);
2796 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2799 if (wbc->range_cyclic) {
2800 writeback_index = mapping->writeback_index;
2801 if (writeback_index)
2803 mpd.first_page = writeback_index;
2806 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2807 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2812 ext4_io_submit_init(&mpd.io_submit, wbc);
2814 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2815 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2817 blk_start_plug(&plug);
2820 * First writeback pages that don't need mapping - we can avoid
2821 * starting a transaction unnecessarily and also avoid being blocked
2822 * in the block layer on device congestion while having transaction
2826 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2827 if (!mpd.io_submit.io_end) {
2831 ret = mpage_prepare_extent_to_map(&mpd);
2832 /* Unlock pages we didn't use */
2833 mpage_release_unused_pages(&mpd, false);
2834 /* Submit prepared bio */
2835 ext4_io_submit(&mpd.io_submit);
2836 ext4_put_io_end_defer(mpd.io_submit.io_end);
2837 mpd.io_submit.io_end = NULL;
2841 while (!done && mpd.first_page <= mpd.last_page) {
2842 /* For each extent of pages we use new io_end */
2843 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2844 if (!mpd.io_submit.io_end) {
2850 * We have two constraints: We find one extent to map and we
2851 * must always write out whole page (makes a difference when
2852 * blocksize < pagesize) so that we don't block on IO when we
2853 * try to write out the rest of the page. Journalled mode is
2854 * not supported by delalloc.
2856 BUG_ON(ext4_should_journal_data(inode));
2857 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2859 /* start a new transaction */
2860 handle = ext4_journal_start_with_reserve(inode,
2861 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2862 if (IS_ERR(handle)) {
2863 ret = PTR_ERR(handle);
2864 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2865 "%ld pages, ino %lu; err %d", __func__,
2866 wbc->nr_to_write, inode->i_ino, ret);
2867 /* Release allocated io_end */
2868 ext4_put_io_end(mpd.io_submit.io_end);
2869 mpd.io_submit.io_end = NULL;
2874 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2875 ret = mpage_prepare_extent_to_map(&mpd);
2878 ret = mpage_map_and_submit_extent(handle, &mpd,
2882 * We scanned the whole range (or exhausted
2883 * nr_to_write), submitted what was mapped and
2884 * didn't find anything needing mapping. We are
2891 * Caution: If the handle is synchronous,
2892 * ext4_journal_stop() can wait for transaction commit
2893 * to finish which may depend on writeback of pages to
2894 * complete or on page lock to be released. In that
2895 * case, we have to wait until after after we have
2896 * submitted all the IO, released page locks we hold,
2897 * and dropped io_end reference (for extent conversion
2898 * to be able to complete) before stopping the handle.
2900 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2901 ext4_journal_stop(handle);
2905 /* Unlock pages we didn't use */
2906 mpage_release_unused_pages(&mpd, give_up_on_write);
2907 /* Submit prepared bio */
2908 ext4_io_submit(&mpd.io_submit);
2911 * Drop our io_end reference we got from init. We have
2912 * to be careful and use deferred io_end finishing if
2913 * we are still holding the transaction as we can
2914 * release the last reference to io_end which may end
2915 * up doing unwritten extent conversion.
2918 ext4_put_io_end_defer(mpd.io_submit.io_end);
2919 ext4_journal_stop(handle);
2921 ext4_put_io_end(mpd.io_submit.io_end);
2922 mpd.io_submit.io_end = NULL;
2924 if (ret == -ENOSPC && sbi->s_journal) {
2926 * Commit the transaction which would
2927 * free blocks released in the transaction
2930 jbd2_journal_force_commit_nested(sbi->s_journal);
2934 /* Fatal error - ENOMEM, EIO... */
2939 blk_finish_plug(&plug);
2940 if (!ret && !cycled && wbc->nr_to_write > 0) {
2942 mpd.last_page = writeback_index - 1;
2948 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2950 * Set the writeback_index so that range_cyclic
2951 * mode will write it back later
2953 mapping->writeback_index = mpd.first_page;
2956 trace_ext4_writepages_result(inode, wbc, ret,
2957 nr_to_write - wbc->nr_to_write);
2958 percpu_up_read(&sbi->s_journal_flag_rwsem);
2962 static int ext4_dax_writepages(struct address_space *mapping,
2963 struct writeback_control *wbc)
2966 long nr_to_write = wbc->nr_to_write;
2967 struct inode *inode = mapping->host;
2968 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2970 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2973 percpu_down_read(&sbi->s_journal_flag_rwsem);
2974 trace_ext4_writepages(inode, wbc);
2976 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, wbc);
2977 trace_ext4_writepages_result(inode, wbc, ret,
2978 nr_to_write - wbc->nr_to_write);
2979 percpu_up_read(&sbi->s_journal_flag_rwsem);
2983 static int ext4_nonda_switch(struct super_block *sb)
2985 s64 free_clusters, dirty_clusters;
2986 struct ext4_sb_info *sbi = EXT4_SB(sb);
2989 * switch to non delalloc mode if we are running low
2990 * on free block. The free block accounting via percpu
2991 * counters can get slightly wrong with percpu_counter_batch getting
2992 * accumulated on each CPU without updating global counters
2993 * Delalloc need an accurate free block accounting. So switch
2994 * to non delalloc when we are near to error range.
2997 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2999 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
3001 * Start pushing delalloc when 1/2 of free blocks are dirty.
3003 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
3004 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
3006 if (2 * free_clusters < 3 * dirty_clusters ||
3007 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
3009 * free block count is less than 150% of dirty blocks
3010 * or free blocks is less than watermark
3017 /* We always reserve for an inode update; the superblock could be there too */
3018 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
3020 if (likely(ext4_has_feature_large_file(inode->i_sb)))
3023 if (pos + len <= 0x7fffffffULL)
3026 /* We might need to update the superblock to set LARGE_FILE */
3030 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3031 loff_t pos, unsigned len, unsigned flags,
3032 struct page **pagep, void **fsdata)
3034 int ret, retries = 0;
3037 struct inode *inode = mapping->host;
3040 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3043 index = pos >> PAGE_SHIFT;
3045 if (ext4_nonda_switch(inode->i_sb) ||
3046 S_ISLNK(inode->i_mode)) {
3047 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3048 return ext4_write_begin(file, mapping, pos,
3049 len, flags, pagep, fsdata);
3051 *fsdata = (void *)0;
3052 trace_ext4_da_write_begin(inode, pos, len, flags);
3054 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3055 ret = ext4_da_write_inline_data_begin(mapping, inode,
3065 * grab_cache_page_write_begin() can take a long time if the
3066 * system is thrashing due to memory pressure, or if the page
3067 * is being written back. So grab it first before we start
3068 * the transaction handle. This also allows us to allocate
3069 * the page (if needed) without using GFP_NOFS.
3072 page = grab_cache_page_write_begin(mapping, index, flags);
3078 * With delayed allocation, we don't log the i_disksize update
3079 * if there is delayed block allocation. But we still need
3080 * to journalling the i_disksize update if writes to the end
3081 * of file which has an already mapped buffer.
3084 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3085 ext4_da_write_credits(inode, pos, len));
3086 if (IS_ERR(handle)) {
3088 return PTR_ERR(handle);
3092 if (page->mapping != mapping) {
3093 /* The page got truncated from under us */
3096 ext4_journal_stop(handle);
3099 /* In case writeback began while the page was unlocked */
3100 wait_for_stable_page(page);
3102 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3103 ret = ext4_block_write_begin(page, pos, len,
3104 ext4_da_get_block_prep);
3106 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3110 ext4_journal_stop(handle);
3112 * block_write_begin may have instantiated a few blocks
3113 * outside i_size. Trim these off again. Don't need
3114 * i_size_read because we hold i_mutex.
3116 if (pos + len > inode->i_size)
3117 ext4_truncate_failed_write(inode);
3119 if (ret == -ENOSPC &&
3120 ext4_should_retry_alloc(inode->i_sb, &retries))
3132 * Check if we should update i_disksize
3133 * when write to the end of file but not require block allocation
3135 static int ext4_da_should_update_i_disksize(struct page *page,
3136 unsigned long offset)
3138 struct buffer_head *bh;
3139 struct inode *inode = page->mapping->host;
3143 bh = page_buffers(page);
3144 idx = offset >> inode->i_blkbits;
3146 for (i = 0; i < idx; i++)
3147 bh = bh->b_this_page;
3149 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3154 static int ext4_da_write_end(struct file *file,
3155 struct address_space *mapping,
3156 loff_t pos, unsigned len, unsigned copied,
3157 struct page *page, void *fsdata)
3159 struct inode *inode = mapping->host;
3161 handle_t *handle = ext4_journal_current_handle();
3163 unsigned long start, end;
3164 int write_mode = (int)(unsigned long)fsdata;
3166 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3167 return ext4_write_end(file, mapping, pos,
3168 len, copied, page, fsdata);
3170 trace_ext4_da_write_end(inode, pos, len, copied);
3171 start = pos & (PAGE_SIZE - 1);
3172 end = start + copied - 1;
3175 * generic_write_end() will run mark_inode_dirty() if i_size
3176 * changes. So let's piggyback the i_disksize mark_inode_dirty
3179 new_i_size = pos + copied;
3180 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3181 if (ext4_has_inline_data(inode) ||
3182 ext4_da_should_update_i_disksize(page, end)) {
3183 ext4_update_i_disksize(inode, new_i_size);
3184 /* We need to mark inode dirty even if
3185 * new_i_size is less that inode->i_size
3186 * bu greater than i_disksize.(hint delalloc)
3188 ext4_mark_inode_dirty(handle, inode);
3192 if (write_mode != CONVERT_INLINE_DATA &&
3193 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3194 ext4_has_inline_data(inode))
3195 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3198 ret2 = generic_write_end(file, mapping, pos, len, copied,
3204 ret2 = ext4_journal_stop(handle);
3208 return ret ? ret : copied;
3211 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3212 unsigned int length)
3215 * Drop reserved blocks
3217 BUG_ON(!PageLocked(page));
3218 if (!page_has_buffers(page))
3221 ext4_da_page_release_reservation(page, offset, length);
3224 ext4_invalidatepage(page, offset, length);
3230 * Force all delayed allocation blocks to be allocated for a given inode.
3232 int ext4_alloc_da_blocks(struct inode *inode)
3234 trace_ext4_alloc_da_blocks(inode);
3236 if (!EXT4_I(inode)->i_reserved_data_blocks)
3240 * We do something simple for now. The filemap_flush() will
3241 * also start triggering a write of the data blocks, which is
3242 * not strictly speaking necessary (and for users of
3243 * laptop_mode, not even desirable). However, to do otherwise
3244 * would require replicating code paths in:
3246 * ext4_writepages() ->
3247 * write_cache_pages() ---> (via passed in callback function)
3248 * __mpage_da_writepage() -->
3249 * mpage_add_bh_to_extent()
3250 * mpage_da_map_blocks()
3252 * The problem is that write_cache_pages(), located in
3253 * mm/page-writeback.c, marks pages clean in preparation for
3254 * doing I/O, which is not desirable if we're not planning on
3257 * We could call write_cache_pages(), and then redirty all of
3258 * the pages by calling redirty_page_for_writepage() but that
3259 * would be ugly in the extreme. So instead we would need to
3260 * replicate parts of the code in the above functions,
3261 * simplifying them because we wouldn't actually intend to
3262 * write out the pages, but rather only collect contiguous
3263 * logical block extents, call the multi-block allocator, and
3264 * then update the buffer heads with the block allocations.
3266 * For now, though, we'll cheat by calling filemap_flush(),
3267 * which will map the blocks, and start the I/O, but not
3268 * actually wait for the I/O to complete.
3270 return filemap_flush(inode->i_mapping);
3274 * bmap() is special. It gets used by applications such as lilo and by
3275 * the swapper to find the on-disk block of a specific piece of data.
3277 * Naturally, this is dangerous if the block concerned is still in the
3278 * journal. If somebody makes a swapfile on an ext4 data-journaling
3279 * filesystem and enables swap, then they may get a nasty shock when the
3280 * data getting swapped to that swapfile suddenly gets overwritten by
3281 * the original zero's written out previously to the journal and
3282 * awaiting writeback in the kernel's buffer cache.
3284 * So, if we see any bmap calls here on a modified, data-journaled file,
3285 * take extra steps to flush any blocks which might be in the cache.
3287 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3289 struct inode *inode = mapping->host;
3294 * We can get here for an inline file via the FIBMAP ioctl
3296 if (ext4_has_inline_data(inode))
3299 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3300 test_opt(inode->i_sb, DELALLOC)) {
3302 * With delalloc we want to sync the file
3303 * so that we can make sure we allocate
3306 filemap_write_and_wait(mapping);
3309 if (EXT4_JOURNAL(inode) &&
3310 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3312 * This is a REALLY heavyweight approach, but the use of
3313 * bmap on dirty files is expected to be extremely rare:
3314 * only if we run lilo or swapon on a freshly made file
3315 * do we expect this to happen.
3317 * (bmap requires CAP_SYS_RAWIO so this does not
3318 * represent an unprivileged user DOS attack --- we'd be
3319 * in trouble if mortal users could trigger this path at
3322 * NB. EXT4_STATE_JDATA is not set on files other than
3323 * regular files. If somebody wants to bmap a directory
3324 * or symlink and gets confused because the buffer
3325 * hasn't yet been flushed to disk, they deserve
3326 * everything they get.
3329 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3330 journal = EXT4_JOURNAL(inode);
3331 jbd2_journal_lock_updates(journal);
3332 err = jbd2_journal_flush(journal);
3333 jbd2_journal_unlock_updates(journal);
3339 return generic_block_bmap(mapping, block, ext4_get_block);
3342 static int ext4_readpage(struct file *file, struct page *page)
3345 struct inode *inode = page->mapping->host;
3347 trace_ext4_readpage(page);
3349 if (ext4_has_inline_data(inode))
3350 ret = ext4_readpage_inline(inode, page);
3353 return ext4_mpage_readpages(page->mapping, NULL, page, 1,
3360 ext4_readpages(struct file *file, struct address_space *mapping,
3361 struct list_head *pages, unsigned nr_pages)
3363 struct inode *inode = mapping->host;
3365 /* If the file has inline data, no need to do readpages. */
3366 if (ext4_has_inline_data(inode))
3369 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages, true);
3372 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3373 unsigned int length)
3375 trace_ext4_invalidatepage(page, offset, length);
3377 /* No journalling happens on data buffers when this function is used */
3378 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3380 block_invalidatepage(page, offset, length);
3383 static int __ext4_journalled_invalidatepage(struct page *page,
3384 unsigned int offset,
3385 unsigned int length)
3387 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3389 trace_ext4_journalled_invalidatepage(page, offset, length);
3392 * If it's a full truncate we just forget about the pending dirtying
3394 if (offset == 0 && length == PAGE_SIZE)
3395 ClearPageChecked(page);
3397 return jbd2_journal_invalidatepage(journal, page, offset, length);
3400 /* Wrapper for aops... */
3401 static void ext4_journalled_invalidatepage(struct page *page,
3402 unsigned int offset,
3403 unsigned int length)
3405 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3408 static int ext4_releasepage(struct page *page, gfp_t wait)
3410 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3412 trace_ext4_releasepage(page);
3414 /* Page has dirty journalled data -> cannot release */
3415 if (PageChecked(page))
3418 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3420 return try_to_free_buffers(page);
3423 static bool ext4_inode_datasync_dirty(struct inode *inode)
3425 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3428 return !jbd2_transaction_committed(journal,
3429 EXT4_I(inode)->i_datasync_tid);
3430 /* Any metadata buffers to write? */
3431 if (!list_empty(&inode->i_mapping->private_list))
3433 return inode->i_state & I_DIRTY_DATASYNC;
3436 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3437 unsigned flags, struct iomap *iomap)
3439 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3440 unsigned int blkbits = inode->i_blkbits;
3441 unsigned long first_block, last_block;
3442 struct ext4_map_blocks map;
3443 bool delalloc = false;
3446 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3448 first_block = offset >> blkbits;
3449 last_block = min_t(loff_t, (offset + length - 1) >> blkbits,
3450 EXT4_MAX_LOGICAL_BLOCK);
3452 if (flags & IOMAP_REPORT) {
3453 if (ext4_has_inline_data(inode)) {
3454 ret = ext4_inline_data_iomap(inode, iomap);
3455 if (ret != -EAGAIN) {
3456 if (ret == 0 && offset >= iomap->length)
3462 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3466 map.m_lblk = first_block;
3467 map.m_len = last_block - first_block + 1;
3469 if (flags & IOMAP_REPORT) {
3470 ret = ext4_map_blocks(NULL, inode, &map, 0);
3475 ext4_lblk_t end = map.m_lblk + map.m_len - 1;
3476 struct extent_status es;
3478 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3479 map.m_lblk, end, &es);
3481 if (!es.es_len || es.es_lblk > end) {
3482 /* entire range is a hole */
3483 } else if (es.es_lblk > map.m_lblk) {
3484 /* range starts with a hole */
3485 map.m_len = es.es_lblk - map.m_lblk;
3487 ext4_lblk_t offs = 0;
3489 if (es.es_lblk < map.m_lblk)
3490 offs = map.m_lblk - es.es_lblk;
3491 map.m_lblk = es.es_lblk + offs;
3492 map.m_len = es.es_len - offs;
3496 } else if (flags & IOMAP_WRITE) {
3501 /* Trim mapping request to maximum we can map at once for DIO */
3502 if (map.m_len > DIO_MAX_BLOCKS)
3503 map.m_len = DIO_MAX_BLOCKS;
3504 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3507 * Either we allocate blocks and then we don't get unwritten
3508 * extent so we have reserved enough credits, or the blocks
3509 * are already allocated and unwritten and in that case
3510 * extent conversion fits in the credits as well.
3512 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3515 return PTR_ERR(handle);
3517 ret = ext4_map_blocks(handle, inode, &map,
3518 EXT4_GET_BLOCKS_CREATE_ZERO);
3520 ext4_journal_stop(handle);
3521 if (ret == -ENOSPC &&
3522 ext4_should_retry_alloc(inode->i_sb, &retries))
3528 * If we added blocks beyond i_size, we need to make sure they
3529 * will get truncated if we crash before updating i_size in
3530 * ext4_iomap_end(). For faults we don't need to do that (and
3531 * even cannot because for orphan list operations inode_lock is
3532 * required) - if we happen to instantiate block beyond i_size,
3533 * it is because we race with truncate which has already added
3534 * the inode to the orphan list.
3536 if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3537 (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3540 err = ext4_orphan_add(handle, inode);
3542 ext4_journal_stop(handle);
3546 ext4_journal_stop(handle);
3548 ret = ext4_map_blocks(NULL, inode, &map, 0);
3554 if (ext4_inode_datasync_dirty(inode))
3555 iomap->flags |= IOMAP_F_DIRTY;
3556 iomap->bdev = inode->i_sb->s_bdev;
3557 iomap->dax_dev = sbi->s_daxdev;
3558 iomap->offset = (u64)first_block << blkbits;
3559 iomap->length = (u64)map.m_len << blkbits;
3562 iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE;
3563 iomap->addr = IOMAP_NULL_ADDR;
3565 if (map.m_flags & EXT4_MAP_MAPPED) {
3566 iomap->type = IOMAP_MAPPED;
3567 } else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3568 iomap->type = IOMAP_UNWRITTEN;
3573 iomap->addr = (u64)map.m_pblk << blkbits;
3576 if (map.m_flags & EXT4_MAP_NEW)
3577 iomap->flags |= IOMAP_F_NEW;
3582 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3583 ssize_t written, unsigned flags, struct iomap *iomap)
3587 int blkbits = inode->i_blkbits;
3588 bool truncate = false;
3590 if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3593 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3594 if (IS_ERR(handle)) {
3595 ret = PTR_ERR(handle);
3598 if (ext4_update_inode_size(inode, offset + written))
3599 ext4_mark_inode_dirty(handle, inode);
3601 * We may need to truncate allocated but not written blocks beyond EOF.
3603 if (iomap->offset + iomap->length >
3604 ALIGN(inode->i_size, 1 << blkbits)) {
3605 ext4_lblk_t written_blk, end_blk;
3607 written_blk = (offset + written) >> blkbits;
3608 end_blk = (offset + length) >> blkbits;
3609 if (written_blk < end_blk && ext4_can_truncate(inode))
3613 * Remove inode from orphan list if we were extending a inode and
3614 * everything went fine.
3616 if (!truncate && inode->i_nlink &&
3617 !list_empty(&EXT4_I(inode)->i_orphan))
3618 ext4_orphan_del(handle, inode);
3619 ext4_journal_stop(handle);
3621 ext4_truncate_failed_write(inode);
3624 * If truncate failed early the inode might still be on the
3625 * orphan list; we need to make sure the inode is removed from
3626 * the orphan list in that case.
3629 ext4_orphan_del(NULL, inode);
3634 const struct iomap_ops ext4_iomap_ops = {
3635 .iomap_begin = ext4_iomap_begin,
3636 .iomap_end = ext4_iomap_end,
3639 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3640 ssize_t size, void *private)
3642 ext4_io_end_t *io_end = private;
3644 /* if not async direct IO just return */
3648 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3649 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3650 io_end, io_end->inode->i_ino, iocb, offset, size);
3653 * Error during AIO DIO. We cannot convert unwritten extents as the
3654 * data was not written. Just clear the unwritten flag and drop io_end.
3657 ext4_clear_io_unwritten_flag(io_end);
3660 io_end->offset = offset;
3661 io_end->size = size;
3662 ext4_put_io_end(io_end);
3668 * Handling of direct IO writes.
3670 * For ext4 extent files, ext4 will do direct-io write even to holes,
3671 * preallocated extents, and those write extend the file, no need to
3672 * fall back to buffered IO.
3674 * For holes, we fallocate those blocks, mark them as unwritten
3675 * If those blocks were preallocated, we mark sure they are split, but
3676 * still keep the range to write as unwritten.
3678 * The unwritten extents will be converted to written when DIO is completed.
3679 * For async direct IO, since the IO may still pending when return, we
3680 * set up an end_io call back function, which will do the conversion
3681 * when async direct IO completed.
3683 * If the O_DIRECT write will extend the file then add this inode to the
3684 * orphan list. So recovery will truncate it back to the original size
3685 * if the machine crashes during the write.
3688 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3690 struct file *file = iocb->ki_filp;
3691 struct inode *inode = file->f_mapping->host;
3692 struct ext4_inode_info *ei = EXT4_I(inode);
3694 loff_t offset = iocb->ki_pos;
3695 size_t count = iov_iter_count(iter);
3697 get_block_t *get_block_func = NULL;
3699 loff_t final_size = offset + count;
3703 if (final_size > inode->i_size || final_size > ei->i_disksize) {
3704 /* Credits for sb + inode write */
3705 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3706 if (IS_ERR(handle)) {
3707 ret = PTR_ERR(handle);
3710 ret = ext4_orphan_add(handle, inode);
3712 ext4_journal_stop(handle);
3716 ext4_update_i_disksize(inode, inode->i_size);
3717 ext4_journal_stop(handle);
3720 BUG_ON(iocb->private == NULL);
3723 * Make all waiters for direct IO properly wait also for extent
3724 * conversion. This also disallows race between truncate() and
3725 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3727 inode_dio_begin(inode);
3729 /* If we do a overwrite dio, i_mutex locking can be released */
3730 overwrite = *((int *)iocb->private);
3733 inode_unlock(inode);
3736 * For extent mapped files we could direct write to holes and fallocate.
3738 * Allocated blocks to fill the hole are marked as unwritten to prevent
3739 * parallel buffered read to expose the stale data before DIO complete
3742 * As to previously fallocated extents, ext4 get_block will just simply
3743 * mark the buffer mapped but still keep the extents unwritten.
3745 * For non AIO case, we will convert those unwritten extents to written
3746 * after return back from blockdev_direct_IO. That way we save us from
3747 * allocating io_end structure and also the overhead of offloading
3748 * the extent convertion to a workqueue.
3750 * For async DIO, the conversion needs to be deferred when the
3751 * IO is completed. The ext4 end_io callback function will be
3752 * called to take care of the conversion work. Here for async
3753 * case, we allocate an io_end structure to hook to the iocb.
3755 iocb->private = NULL;
3757 get_block_func = ext4_dio_get_block_overwrite;
3758 else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3759 round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3760 get_block_func = ext4_dio_get_block;
3761 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3762 } else if (is_sync_kiocb(iocb)) {
3763 get_block_func = ext4_dio_get_block_unwritten_sync;
3764 dio_flags = DIO_LOCKING;
3766 get_block_func = ext4_dio_get_block_unwritten_async;
3767 dio_flags = DIO_LOCKING;
3769 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3770 get_block_func, ext4_end_io_dio, NULL,
3773 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3774 EXT4_STATE_DIO_UNWRITTEN)) {
3777 * for non AIO case, since the IO is already
3778 * completed, we could do the conversion right here
3780 err = ext4_convert_unwritten_extents(NULL, inode,
3784 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3787 inode_dio_end(inode);
3788 /* take i_mutex locking again if we do a ovewrite dio */
3792 if (ret < 0 && final_size > inode->i_size)
3793 ext4_truncate_failed_write(inode);
3795 /* Handle extending of i_size after direct IO write */
3799 /* Credits for sb + inode write */
3800 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3801 if (IS_ERR(handle)) {
3803 * We wrote the data but cannot extend
3804 * i_size. Bail out. In async io case, we do
3805 * not return error here because we have
3806 * already submmitted the corresponding
3807 * bio. Returning error here makes the caller
3808 * think that this IO is done and failed
3809 * resulting in race with bio's completion
3813 ret = PTR_ERR(handle);
3815 ext4_orphan_del(NULL, inode);
3820 ext4_orphan_del(handle, inode);
3822 loff_t end = offset + ret;
3823 if (end > inode->i_size || end > ei->i_disksize) {
3824 ext4_update_i_disksize(inode, end);
3825 if (end > inode->i_size)
3826 i_size_write(inode, end);
3828 * We're going to return a positive `ret'
3829 * here due to non-zero-length I/O, so there's
3830 * no way of reporting error returns from
3831 * ext4_mark_inode_dirty() to userspace. So
3834 ext4_mark_inode_dirty(handle, inode);
3837 err = ext4_journal_stop(handle);
3845 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3847 struct address_space *mapping = iocb->ki_filp->f_mapping;
3848 struct inode *inode = mapping->host;
3849 size_t count = iov_iter_count(iter);
3853 * Shared inode_lock is enough for us - it protects against concurrent
3854 * writes & truncates and since we take care of writing back page cache,
3855 * we are protected against page writeback as well.
3857 inode_lock_shared(inode);
3858 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3859 iocb->ki_pos + count - 1);
3862 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3863 iter, ext4_dio_get_block, NULL, NULL, 0);
3865 inode_unlock_shared(inode);
3869 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3871 struct file *file = iocb->ki_filp;
3872 struct inode *inode = file->f_mapping->host;
3873 size_t count = iov_iter_count(iter);
3874 loff_t offset = iocb->ki_pos;
3877 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3878 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3883 * If we are doing data journalling we don't support O_DIRECT
3885 if (ext4_should_journal_data(inode))
3888 /* Let buffer I/O handle the inline data case. */
3889 if (ext4_has_inline_data(inode))
3892 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3893 if (iov_iter_rw(iter) == READ)
3894 ret = ext4_direct_IO_read(iocb, iter);
3896 ret = ext4_direct_IO_write(iocb, iter);
3897 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3902 * Pages can be marked dirty completely asynchronously from ext4's journalling
3903 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3904 * much here because ->set_page_dirty is called under VFS locks. The page is
3905 * not necessarily locked.
3907 * We cannot just dirty the page and leave attached buffers clean, because the
3908 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3909 * or jbddirty because all the journalling code will explode.
3911 * So what we do is to mark the page "pending dirty" and next time writepage
3912 * is called, propagate that into the buffers appropriately.
3914 static int ext4_journalled_set_page_dirty(struct page *page)
3916 SetPageChecked(page);
3917 return __set_page_dirty_nobuffers(page);
3920 static int ext4_set_page_dirty(struct page *page)
3922 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3923 WARN_ON_ONCE(!page_has_buffers(page));
3924 return __set_page_dirty_buffers(page);
3927 static const struct address_space_operations ext4_aops = {
3928 .readpage = ext4_readpage,
3929 .readpages = ext4_readpages,
3930 .writepage = ext4_writepage,
3931 .writepages = ext4_writepages,
3932 .write_begin = ext4_write_begin,
3933 .write_end = ext4_write_end,
3934 .set_page_dirty = ext4_set_page_dirty,
3936 .invalidatepage = ext4_invalidatepage,
3937 .releasepage = ext4_releasepage,
3938 .direct_IO = ext4_direct_IO,
3939 .migratepage = buffer_migrate_page,
3940 .is_partially_uptodate = block_is_partially_uptodate,
3941 .error_remove_page = generic_error_remove_page,
3944 static const struct address_space_operations ext4_journalled_aops = {
3945 .readpage = ext4_readpage,
3946 .readpages = ext4_readpages,
3947 .writepage = ext4_writepage,
3948 .writepages = ext4_writepages,
3949 .write_begin = ext4_write_begin,
3950 .write_end = ext4_journalled_write_end,
3951 .set_page_dirty = ext4_journalled_set_page_dirty,
3953 .invalidatepage = ext4_journalled_invalidatepage,
3954 .releasepage = ext4_releasepage,
3955 .direct_IO = ext4_direct_IO,
3956 .is_partially_uptodate = block_is_partially_uptodate,
3957 .error_remove_page = generic_error_remove_page,
3960 static const struct address_space_operations ext4_da_aops = {
3961 .readpage = ext4_readpage,
3962 .readpages = ext4_readpages,
3963 .writepage = ext4_writepage,
3964 .writepages = ext4_writepages,
3965 .write_begin = ext4_da_write_begin,
3966 .write_end = ext4_da_write_end,
3967 .set_page_dirty = ext4_set_page_dirty,
3969 .invalidatepage = ext4_da_invalidatepage,
3970 .releasepage = ext4_releasepage,
3971 .direct_IO = ext4_direct_IO,
3972 .migratepage = buffer_migrate_page,
3973 .is_partially_uptodate = block_is_partially_uptodate,
3974 .error_remove_page = generic_error_remove_page,
3977 static const struct address_space_operations ext4_dax_aops = {
3978 .writepages = ext4_dax_writepages,
3979 .direct_IO = noop_direct_IO,
3980 .set_page_dirty = noop_set_page_dirty,
3982 .invalidatepage = noop_invalidatepage,
3985 void ext4_set_aops(struct inode *inode)
3987 switch (ext4_inode_journal_mode(inode)) {
3988 case EXT4_INODE_ORDERED_DATA_MODE:
3989 case EXT4_INODE_WRITEBACK_DATA_MODE:
3991 case EXT4_INODE_JOURNAL_DATA_MODE:
3992 inode->i_mapping->a_ops = &ext4_journalled_aops;
3998 inode->i_mapping->a_ops = &ext4_dax_aops;
3999 else if (test_opt(inode->i_sb, DELALLOC))
4000 inode->i_mapping->a_ops = &ext4_da_aops;
4002 inode->i_mapping->a_ops = &ext4_aops;
4005 static int __ext4_block_zero_page_range(handle_t *handle,
4006 struct address_space *mapping, loff_t from, loff_t length)
4008 ext4_fsblk_t index = from >> PAGE_SHIFT;
4009 unsigned offset = from & (PAGE_SIZE-1);
4010 unsigned blocksize, pos;
4012 struct inode *inode = mapping->host;
4013 struct buffer_head *bh;
4017 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
4018 mapping_gfp_constraint(mapping, ~__GFP_FS));
4022 blocksize = inode->i_sb->s_blocksize;
4024 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
4026 if (!page_has_buffers(page))
4027 create_empty_buffers(page, blocksize, 0);
4029 /* Find the buffer that contains "offset" */
4030 bh = page_buffers(page);
4032 while (offset >= pos) {
4033 bh = bh->b_this_page;
4037 if (buffer_freed(bh)) {
4038 BUFFER_TRACE(bh, "freed: skip");
4041 if (!buffer_mapped(bh)) {
4042 BUFFER_TRACE(bh, "unmapped");
4043 ext4_get_block(inode, iblock, bh, 0);
4044 /* unmapped? It's a hole - nothing to do */
4045 if (!buffer_mapped(bh)) {
4046 BUFFER_TRACE(bh, "still unmapped");
4051 /* Ok, it's mapped. Make sure it's up-to-date */
4052 if (PageUptodate(page))
4053 set_buffer_uptodate(bh);
4055 if (!buffer_uptodate(bh)) {
4057 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
4059 /* Uhhuh. Read error. Complain and punt. */
4060 if (!buffer_uptodate(bh))
4062 if (S_ISREG(inode->i_mode) &&
4063 ext4_encrypted_inode(inode)) {
4064 /* We expect the key to be set. */
4065 BUG_ON(!fscrypt_has_encryption_key(inode));
4066 BUG_ON(blocksize != PAGE_SIZE);
4067 WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host,
4068 page, PAGE_SIZE, 0, page->index));
4071 if (ext4_should_journal_data(inode)) {
4072 BUFFER_TRACE(bh, "get write access");
4073 err = ext4_journal_get_write_access(handle, bh);
4077 zero_user(page, offset, length);
4078 BUFFER_TRACE(bh, "zeroed end of block");
4080 if (ext4_should_journal_data(inode)) {
4081 err = ext4_handle_dirty_metadata(handle, inode, bh);
4084 mark_buffer_dirty(bh);
4085 if (ext4_should_order_data(inode))
4086 err = ext4_jbd2_inode_add_write(handle, inode);
4096 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4097 * starting from file offset 'from'. The range to be zero'd must
4098 * be contained with in one block. If the specified range exceeds
4099 * the end of the block it will be shortened to end of the block
4100 * that cooresponds to 'from'
4102 static int ext4_block_zero_page_range(handle_t *handle,
4103 struct address_space *mapping, loff_t from, loff_t length)
4105 struct inode *inode = mapping->host;
4106 unsigned offset = from & (PAGE_SIZE-1);
4107 unsigned blocksize = inode->i_sb->s_blocksize;
4108 unsigned max = blocksize - (offset & (blocksize - 1));
4111 * correct length if it does not fall between
4112 * 'from' and the end of the block
4114 if (length > max || length < 0)
4117 if (IS_DAX(inode)) {
4118 return iomap_zero_range(inode, from, length, NULL,
4121 return __ext4_block_zero_page_range(handle, mapping, from, length);
4125 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4126 * up to the end of the block which corresponds to `from'.
4127 * This required during truncate. We need to physically zero the tail end
4128 * of that block so it doesn't yield old data if the file is later grown.
4130 static int ext4_block_truncate_page(handle_t *handle,
4131 struct address_space *mapping, loff_t from)
4133 unsigned offset = from & (PAGE_SIZE-1);
4136 struct inode *inode = mapping->host;
4138 /* If we are processing an encrypted inode during orphan list handling */
4139 if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode))
4142 blocksize = inode->i_sb->s_blocksize;
4143 length = blocksize - (offset & (blocksize - 1));
4145 return ext4_block_zero_page_range(handle, mapping, from, length);
4148 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4149 loff_t lstart, loff_t length)
4151 struct super_block *sb = inode->i_sb;
4152 struct address_space *mapping = inode->i_mapping;
4153 unsigned partial_start, partial_end;
4154 ext4_fsblk_t start, end;
4155 loff_t byte_end = (lstart + length - 1);
4158 partial_start = lstart & (sb->s_blocksize - 1);
4159 partial_end = byte_end & (sb->s_blocksize - 1);
4161 start = lstart >> sb->s_blocksize_bits;
4162 end = byte_end >> sb->s_blocksize_bits;
4164 /* Handle partial zero within the single block */
4166 (partial_start || (partial_end != sb->s_blocksize - 1))) {
4167 err = ext4_block_zero_page_range(handle, mapping,
4171 /* Handle partial zero out on the start of the range */
4172 if (partial_start) {
4173 err = ext4_block_zero_page_range(handle, mapping,
4174 lstart, sb->s_blocksize);
4178 /* Handle partial zero out on the end of the range */
4179 if (partial_end != sb->s_blocksize - 1)
4180 err = ext4_block_zero_page_range(handle, mapping,
4181 byte_end - partial_end,
4186 int ext4_can_truncate(struct inode *inode)
4188 if (S_ISREG(inode->i_mode))
4190 if (S_ISDIR(inode->i_mode))
4192 if (S_ISLNK(inode->i_mode))
4193 return !ext4_inode_is_fast_symlink(inode);
4198 * We have to make sure i_disksize gets properly updated before we truncate
4199 * page cache due to hole punching or zero range. Otherwise i_disksize update
4200 * can get lost as it may have been postponed to submission of writeback but
4201 * that will never happen after we truncate page cache.
4203 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4207 loff_t size = i_size_read(inode);
4209 WARN_ON(!inode_is_locked(inode));
4210 if (offset > size || offset + len < size)
4213 if (EXT4_I(inode)->i_disksize >= size)
4216 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4218 return PTR_ERR(handle);
4219 ext4_update_i_disksize(inode, size);
4220 ext4_mark_inode_dirty(handle, inode);
4221 ext4_journal_stop(handle);
4226 static void ext4_wait_dax_page(struct ext4_inode_info *ei)
4228 up_write(&ei->i_mmap_sem);
4230 down_write(&ei->i_mmap_sem);
4233 int ext4_break_layouts(struct inode *inode)
4235 struct ext4_inode_info *ei = EXT4_I(inode);
4239 if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
4243 page = dax_layout_busy_page(inode->i_mapping);
4247 error = ___wait_var_event(&page->_refcount,
4248 atomic_read(&page->_refcount) == 1,
4249 TASK_INTERRUPTIBLE, 0, 0,
4250 ext4_wait_dax_page(ei));
4251 } while (error == 0);
4257 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4258 * associated with the given offset and length
4260 * @inode: File inode
4261 * @offset: The offset where the hole will begin
4262 * @len: The length of the hole
4264 * Returns: 0 on success or negative on failure
4267 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4269 struct super_block *sb = inode->i_sb;
4270 ext4_lblk_t first_block, stop_block;
4271 struct address_space *mapping = inode->i_mapping;
4272 loff_t first_block_offset, last_block_offset;
4274 unsigned int credits;
4277 if (!S_ISREG(inode->i_mode))
4280 trace_ext4_punch_hole(inode, offset, length, 0);
4283 * Write out all dirty pages to avoid race conditions
4284 * Then release them.
4286 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4287 ret = filemap_write_and_wait_range(mapping, offset,
4288 offset + length - 1);
4295 /* No need to punch hole beyond i_size */
4296 if (offset >= inode->i_size)
4300 * If the hole extends beyond i_size, set the hole
4301 * to end after the page that contains i_size
4303 if (offset + length > inode->i_size) {
4304 length = inode->i_size +
4305 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4309 if (offset & (sb->s_blocksize - 1) ||
4310 (offset + length) & (sb->s_blocksize - 1)) {
4312 * Attach jinode to inode for jbd2 if we do any zeroing of
4315 ret = ext4_inode_attach_jinode(inode);
4321 /* Wait all existing dio workers, newcomers will block on i_mutex */
4322 inode_dio_wait(inode);
4325 * Prevent page faults from reinstantiating pages we have released from
4328 down_write(&EXT4_I(inode)->i_mmap_sem);
4330 ret = ext4_break_layouts(inode);
4334 first_block_offset = round_up(offset, sb->s_blocksize);
4335 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4337 /* Now release the pages and zero block aligned part of pages*/
4338 if (last_block_offset > first_block_offset) {
4339 ret = ext4_update_disksize_before_punch(inode, offset, length);
4342 truncate_pagecache_range(inode, first_block_offset,
4346 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4347 credits = ext4_writepage_trans_blocks(inode);
4349 credits = ext4_blocks_for_truncate(inode);
4350 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4351 if (IS_ERR(handle)) {
4352 ret = PTR_ERR(handle);
4353 ext4_std_error(sb, ret);
4357 ret = ext4_zero_partial_blocks(handle, inode, offset,
4362 first_block = (offset + sb->s_blocksize - 1) >>
4363 EXT4_BLOCK_SIZE_BITS(sb);
4364 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4366 /* If there are blocks to remove, do it */
4367 if (stop_block > first_block) {
4369 down_write(&EXT4_I(inode)->i_data_sem);
4370 ext4_discard_preallocations(inode);
4372 ret = ext4_es_remove_extent(inode, first_block,
4373 stop_block - first_block);
4375 up_write(&EXT4_I(inode)->i_data_sem);
4379 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4380 ret = ext4_ext_remove_space(inode, first_block,
4383 ret = ext4_ind_remove_space(handle, inode, first_block,
4386 up_write(&EXT4_I(inode)->i_data_sem);
4389 ext4_handle_sync(handle);
4391 inode->i_mtime = inode->i_ctime = current_time(inode);
4392 ext4_mark_inode_dirty(handle, inode);
4394 ext4_update_inode_fsync_trans(handle, inode, 1);
4396 ext4_journal_stop(handle);
4398 up_write(&EXT4_I(inode)->i_mmap_sem);
4400 inode_unlock(inode);
4404 int ext4_inode_attach_jinode(struct inode *inode)
4406 struct ext4_inode_info *ei = EXT4_I(inode);
4407 struct jbd2_inode *jinode;
4409 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4412 jinode = jbd2_alloc_inode(GFP_KERNEL);
4413 spin_lock(&inode->i_lock);
4416 spin_unlock(&inode->i_lock);
4419 ei->jinode = jinode;
4420 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4423 spin_unlock(&inode->i_lock);
4424 if (unlikely(jinode != NULL))
4425 jbd2_free_inode(jinode);
4432 * We block out ext4_get_block() block instantiations across the entire
4433 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4434 * simultaneously on behalf of the same inode.
4436 * As we work through the truncate and commit bits of it to the journal there
4437 * is one core, guiding principle: the file's tree must always be consistent on
4438 * disk. We must be able to restart the truncate after a crash.
4440 * The file's tree may be transiently inconsistent in memory (although it
4441 * probably isn't), but whenever we close off and commit a journal transaction,
4442 * the contents of (the filesystem + the journal) must be consistent and
4443 * restartable. It's pretty simple, really: bottom up, right to left (although
4444 * left-to-right works OK too).
4446 * Note that at recovery time, journal replay occurs *before* the restart of
4447 * truncate against the orphan inode list.
4449 * The committed inode has the new, desired i_size (which is the same as
4450 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4451 * that this inode's truncate did not complete and it will again call
4452 * ext4_truncate() to have another go. So there will be instantiated blocks
4453 * to the right of the truncation point in a crashed ext4 filesystem. But
4454 * that's fine - as long as they are linked from the inode, the post-crash
4455 * ext4_truncate() run will find them and release them.
4457 int ext4_truncate(struct inode *inode)
4459 struct ext4_inode_info *ei = EXT4_I(inode);
4460 unsigned int credits;
4463 struct address_space *mapping = inode->i_mapping;
4466 * There is a possibility that we're either freeing the inode
4467 * or it's a completely new inode. In those cases we might not
4468 * have i_mutex locked because it's not necessary.
4470 if (!(inode->i_state & (I_NEW|I_FREEING)))
4471 WARN_ON(!inode_is_locked(inode));
4472 trace_ext4_truncate_enter(inode);
4474 if (!ext4_can_truncate(inode))
4477 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4479 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4480 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4482 if (ext4_has_inline_data(inode)) {
4485 err = ext4_inline_data_truncate(inode, &has_inline);
4492 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4493 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4494 if (ext4_inode_attach_jinode(inode) < 0)
4498 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4499 credits = ext4_writepage_trans_blocks(inode);
4501 credits = ext4_blocks_for_truncate(inode);
4503 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4505 return PTR_ERR(handle);
4507 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4508 ext4_block_truncate_page(handle, mapping, inode->i_size);
4511 * We add the inode to the orphan list, so that if this
4512 * truncate spans multiple transactions, and we crash, we will
4513 * resume the truncate when the filesystem recovers. It also
4514 * marks the inode dirty, to catch the new size.
4516 * Implication: the file must always be in a sane, consistent
4517 * truncatable state while each transaction commits.
4519 err = ext4_orphan_add(handle, inode);
4523 down_write(&EXT4_I(inode)->i_data_sem);
4525 ext4_discard_preallocations(inode);
4527 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4528 err = ext4_ext_truncate(handle, inode);
4530 ext4_ind_truncate(handle, inode);
4532 up_write(&ei->i_data_sem);
4537 ext4_handle_sync(handle);
4541 * If this was a simple ftruncate() and the file will remain alive,
4542 * then we need to clear up the orphan record which we created above.
4543 * However, if this was a real unlink then we were called by
4544 * ext4_evict_inode(), and we allow that function to clean up the
4545 * orphan info for us.
4548 ext4_orphan_del(handle, inode);
4550 inode->i_mtime = inode->i_ctime = current_time(inode);
4551 ext4_mark_inode_dirty(handle, inode);
4552 ext4_journal_stop(handle);
4554 trace_ext4_truncate_exit(inode);
4559 * ext4_get_inode_loc returns with an extra refcount against the inode's
4560 * underlying buffer_head on success. If 'in_mem' is true, we have all
4561 * data in memory that is needed to recreate the on-disk version of this
4564 static int __ext4_get_inode_loc(struct inode *inode,
4565 struct ext4_iloc *iloc, int in_mem)
4567 struct ext4_group_desc *gdp;
4568 struct buffer_head *bh;
4569 struct super_block *sb = inode->i_sb;
4571 int inodes_per_block, inode_offset;
4574 if (inode->i_ino < EXT4_ROOT_INO ||
4575 inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4576 return -EFSCORRUPTED;
4578 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4579 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4584 * Figure out the offset within the block group inode table
4586 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4587 inode_offset = ((inode->i_ino - 1) %
4588 EXT4_INODES_PER_GROUP(sb));
4589 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4590 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4592 bh = sb_getblk(sb, block);
4595 if (!buffer_uptodate(bh)) {
4599 * If the buffer has the write error flag, we have failed
4600 * to write out another inode in the same block. In this
4601 * case, we don't have to read the block because we may
4602 * read the old inode data successfully.
4604 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4605 set_buffer_uptodate(bh);
4607 if (buffer_uptodate(bh)) {
4608 /* someone brought it uptodate while we waited */
4614 * If we have all information of the inode in memory and this
4615 * is the only valid inode in the block, we need not read the
4619 struct buffer_head *bitmap_bh;
4622 start = inode_offset & ~(inodes_per_block - 1);
4624 /* Is the inode bitmap in cache? */
4625 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4626 if (unlikely(!bitmap_bh))
4630 * If the inode bitmap isn't in cache then the
4631 * optimisation may end up performing two reads instead
4632 * of one, so skip it.
4634 if (!buffer_uptodate(bitmap_bh)) {
4638 for (i = start; i < start + inodes_per_block; i++) {
4639 if (i == inode_offset)
4641 if (ext4_test_bit(i, bitmap_bh->b_data))
4645 if (i == start + inodes_per_block) {
4646 /* all other inodes are free, so skip I/O */
4647 memset(bh->b_data, 0, bh->b_size);
4648 set_buffer_uptodate(bh);
4656 * If we need to do any I/O, try to pre-readahead extra
4657 * blocks from the inode table.
4659 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4660 ext4_fsblk_t b, end, table;
4662 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4664 table = ext4_inode_table(sb, gdp);
4665 /* s_inode_readahead_blks is always a power of 2 */
4666 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4670 num = EXT4_INODES_PER_GROUP(sb);
4671 if (ext4_has_group_desc_csum(sb))
4672 num -= ext4_itable_unused_count(sb, gdp);
4673 table += num / inodes_per_block;
4677 sb_breadahead(sb, b++);
4681 * There are other valid inodes in the buffer, this inode
4682 * has in-inode xattrs, or we don't have this inode in memory.
4683 * Read the block from disk.
4685 trace_ext4_load_inode(inode);
4687 bh->b_end_io = end_buffer_read_sync;
4688 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4690 if (!buffer_uptodate(bh)) {
4691 EXT4_ERROR_INODE_BLOCK(inode, block,
4692 "unable to read itable block");
4702 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4704 /* We have all inode data except xattrs in memory here. */
4705 return __ext4_get_inode_loc(inode, iloc,
4706 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4709 static bool ext4_should_use_dax(struct inode *inode)
4711 if (!test_opt(inode->i_sb, DAX))
4713 if (!S_ISREG(inode->i_mode))
4715 if (ext4_should_journal_data(inode))
4717 if (ext4_has_inline_data(inode))
4719 if (ext4_encrypted_inode(inode))
4724 void ext4_set_inode_flags(struct inode *inode)
4726 unsigned int flags = EXT4_I(inode)->i_flags;
4727 unsigned int new_fl = 0;
4729 if (flags & EXT4_SYNC_FL)
4731 if (flags & EXT4_APPEND_FL)
4733 if (flags & EXT4_IMMUTABLE_FL)
4734 new_fl |= S_IMMUTABLE;
4735 if (flags & EXT4_NOATIME_FL)
4736 new_fl |= S_NOATIME;
4737 if (flags & EXT4_DIRSYNC_FL)
4738 new_fl |= S_DIRSYNC;
4739 if (ext4_should_use_dax(inode))
4741 if (flags & EXT4_ENCRYPT_FL)
4742 new_fl |= S_ENCRYPTED;
4743 inode_set_flags(inode, new_fl,
4744 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4748 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4749 struct ext4_inode_info *ei)
4752 struct inode *inode = &(ei->vfs_inode);
4753 struct super_block *sb = inode->i_sb;
4755 if (ext4_has_feature_huge_file(sb)) {
4756 /* we are using combined 48 bit field */
4757 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4758 le32_to_cpu(raw_inode->i_blocks_lo);
4759 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4760 /* i_blocks represent file system block size */
4761 return i_blocks << (inode->i_blkbits - 9);
4766 return le32_to_cpu(raw_inode->i_blocks_lo);
4770 static inline int ext4_iget_extra_inode(struct inode *inode,
4771 struct ext4_inode *raw_inode,
4772 struct ext4_inode_info *ei)
4774 __le32 *magic = (void *)raw_inode +
4775 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4777 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4778 EXT4_INODE_SIZE(inode->i_sb) &&
4779 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4780 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4781 return ext4_find_inline_data_nolock(inode);
4783 EXT4_I(inode)->i_inline_off = 0;
4787 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4789 if (!ext4_has_feature_project(inode->i_sb))
4791 *projid = EXT4_I(inode)->i_projid;
4796 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4797 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4800 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4802 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4803 inode_set_iversion_raw(inode, val);
4805 inode_set_iversion_queried(inode, val);
4807 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4809 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4810 return inode_peek_iversion_raw(inode);
4812 return inode_peek_iversion(inode);
4815 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4816 ext4_iget_flags flags, const char *function,
4819 struct ext4_iloc iloc;
4820 struct ext4_inode *raw_inode;
4821 struct ext4_inode_info *ei;
4822 struct inode *inode;
4823 journal_t *journal = EXT4_SB(sb)->s_journal;
4831 if ((!(flags & EXT4_IGET_SPECIAL) &&
4832 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4833 (ino < EXT4_ROOT_INO) ||
4834 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4835 if (flags & EXT4_IGET_HANDLE)
4836 return ERR_PTR(-ESTALE);
4837 __ext4_error(sb, function, line,
4838 "inode #%lu: comm %s: iget: illegal inode #",
4839 ino, current->comm);
4840 return ERR_PTR(-EFSCORRUPTED);
4843 inode = iget_locked(sb, ino);
4845 return ERR_PTR(-ENOMEM);
4846 if (!(inode->i_state & I_NEW))
4852 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4855 raw_inode = ext4_raw_inode(&iloc);
4857 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4858 ext4_error_inode(inode, function, line, 0,
4859 "iget: root inode unallocated");
4860 ret = -EFSCORRUPTED;
4864 if ((flags & EXT4_IGET_HANDLE) &&
4865 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4870 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4871 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4872 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4873 EXT4_INODE_SIZE(inode->i_sb) ||
4874 (ei->i_extra_isize & 3)) {
4875 ext4_error_inode(inode, function, line, 0,
4876 "iget: bad extra_isize %u "
4879 EXT4_INODE_SIZE(inode->i_sb));
4880 ret = -EFSCORRUPTED;
4884 ei->i_extra_isize = 0;
4886 /* Precompute checksum seed for inode metadata */
4887 if (ext4_has_metadata_csum(sb)) {
4888 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4890 __le32 inum = cpu_to_le32(inode->i_ino);
4891 __le32 gen = raw_inode->i_generation;
4892 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4894 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4898 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4899 ext4_error_inode(inode, function, line, 0,
4900 "iget: checksum invalid");
4905 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4906 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4907 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4908 if (ext4_has_feature_project(sb) &&
4909 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4910 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4911 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4913 i_projid = EXT4_DEF_PROJID;
4915 if (!(test_opt(inode->i_sb, NO_UID32))) {
4916 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4917 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4919 i_uid_write(inode, i_uid);
4920 i_gid_write(inode, i_gid);
4921 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4922 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4924 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4925 ei->i_inline_off = 0;
4926 ei->i_dir_start_lookup = 0;
4927 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4928 /* We now have enough fields to check if the inode was active or not.
4929 * This is needed because nfsd might try to access dead inodes
4930 * the test is that same one that e2fsck uses
4931 * NeilBrown 1999oct15
4933 if (inode->i_nlink == 0) {
4934 if ((inode->i_mode == 0 ||
4935 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4936 ino != EXT4_BOOT_LOADER_INO) {
4937 /* this inode is deleted */
4941 /* The only unlinked inodes we let through here have
4942 * valid i_mode and are being read by the orphan
4943 * recovery code: that's fine, we're about to complete
4944 * the process of deleting those.
4945 * OR it is the EXT4_BOOT_LOADER_INO which is
4946 * not initialized on a new filesystem. */
4948 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4949 ext4_set_inode_flags(inode);
4950 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4951 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4952 if (ext4_has_feature_64bit(sb))
4954 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4955 inode->i_size = ext4_isize(sb, raw_inode);
4956 if ((size = i_size_read(inode)) < 0) {
4957 ext4_error_inode(inode, function, line, 0,
4958 "iget: bad i_size value: %lld", size);
4959 ret = -EFSCORRUPTED;
4962 ei->i_disksize = inode->i_size;
4964 ei->i_reserved_quota = 0;
4966 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4967 ei->i_block_group = iloc.block_group;
4968 ei->i_last_alloc_group = ~0;
4970 * NOTE! The in-memory inode i_data array is in little-endian order
4971 * even on big-endian machines: we do NOT byteswap the block numbers!
4973 for (block = 0; block < EXT4_N_BLOCKS; block++)
4974 ei->i_data[block] = raw_inode->i_block[block];
4975 INIT_LIST_HEAD(&ei->i_orphan);
4978 * Set transaction id's of transactions that have to be committed
4979 * to finish f[data]sync. We set them to currently running transaction
4980 * as we cannot be sure that the inode or some of its metadata isn't
4981 * part of the transaction - the inode could have been reclaimed and
4982 * now it is reread from disk.
4985 transaction_t *transaction;
4988 read_lock(&journal->j_state_lock);
4989 if (journal->j_running_transaction)
4990 transaction = journal->j_running_transaction;
4992 transaction = journal->j_committing_transaction;
4994 tid = transaction->t_tid;
4996 tid = journal->j_commit_sequence;
4997 read_unlock(&journal->j_state_lock);
4998 ei->i_sync_tid = tid;
4999 ei->i_datasync_tid = tid;
5002 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5003 if (ei->i_extra_isize == 0) {
5004 /* The extra space is currently unused. Use it. */
5005 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
5006 ei->i_extra_isize = sizeof(struct ext4_inode) -
5007 EXT4_GOOD_OLD_INODE_SIZE;
5009 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
5015 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5016 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5017 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5018 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5020 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5021 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
5023 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5024 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5026 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5028 ext4_inode_set_iversion_queried(inode, ivers);
5032 if (ei->i_file_acl &&
5033 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5034 ext4_error_inode(inode, function, line, 0,
5035 "iget: bad extended attribute block %llu",
5037 ret = -EFSCORRUPTED;
5039 } else if (!ext4_has_inline_data(inode)) {
5040 /* validate the block references in the inode */
5041 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5042 (S_ISLNK(inode->i_mode) &&
5043 !ext4_inode_is_fast_symlink(inode))) {
5044 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5045 ret = ext4_ext_check_inode(inode);
5047 ret = ext4_ind_check_inode(inode);
5053 if (S_ISREG(inode->i_mode)) {
5054 inode->i_op = &ext4_file_inode_operations;
5055 inode->i_fop = &ext4_file_operations;
5056 ext4_set_aops(inode);
5057 } else if (S_ISDIR(inode->i_mode)) {
5058 inode->i_op = &ext4_dir_inode_operations;
5059 inode->i_fop = &ext4_dir_operations;
5060 } else if (S_ISLNK(inode->i_mode)) {
5061 /* VFS does not allow setting these so must be corruption */
5062 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5063 ext4_error_inode(inode, function, line, 0,
5064 "iget: immutable or append flags "
5065 "not allowed on symlinks");
5066 ret = -EFSCORRUPTED;
5069 if (ext4_encrypted_inode(inode)) {
5070 inode->i_op = &ext4_encrypted_symlink_inode_operations;
5071 ext4_set_aops(inode);
5072 } else if (ext4_inode_is_fast_symlink(inode)) {
5073 inode->i_link = (char *)ei->i_data;
5074 inode->i_op = &ext4_fast_symlink_inode_operations;
5075 nd_terminate_link(ei->i_data, inode->i_size,
5076 sizeof(ei->i_data) - 1);
5078 inode->i_op = &ext4_symlink_inode_operations;
5079 ext4_set_aops(inode);
5081 inode_nohighmem(inode);
5082 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5083 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5084 inode->i_op = &ext4_special_inode_operations;
5085 if (raw_inode->i_block[0])
5086 init_special_inode(inode, inode->i_mode,
5087 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5089 init_special_inode(inode, inode->i_mode,
5090 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5091 } else if (ino == EXT4_BOOT_LOADER_INO) {
5092 make_bad_inode(inode);
5094 ret = -EFSCORRUPTED;
5095 ext4_error_inode(inode, function, line, 0,
5096 "iget: bogus i_mode (%o)", inode->i_mode);
5101 unlock_new_inode(inode);
5107 return ERR_PTR(ret);
5110 static int ext4_inode_blocks_set(handle_t *handle,
5111 struct ext4_inode *raw_inode,
5112 struct ext4_inode_info *ei)
5114 struct inode *inode = &(ei->vfs_inode);
5115 u64 i_blocks = inode->i_blocks;
5116 struct super_block *sb = inode->i_sb;
5118 if (i_blocks <= ~0U) {
5120 * i_blocks can be represented in a 32 bit variable
5121 * as multiple of 512 bytes
5123 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5124 raw_inode->i_blocks_high = 0;
5125 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5128 if (!ext4_has_feature_huge_file(sb))
5131 if (i_blocks <= 0xffffffffffffULL) {
5133 * i_blocks can be represented in a 48 bit variable
5134 * as multiple of 512 bytes
5136 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5137 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5138 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5140 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5141 /* i_block is stored in file system block size */
5142 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5143 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5144 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5149 struct other_inode {
5150 unsigned long orig_ino;
5151 struct ext4_inode *raw_inode;
5154 static int other_inode_match(struct inode * inode, unsigned long ino,
5157 struct other_inode *oi = (struct other_inode *) data;
5159 if ((inode->i_ino != ino) ||
5160 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5162 ((inode->i_state & I_DIRTY_TIME) == 0))
5164 spin_lock(&inode->i_lock);
5165 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5166 I_DIRTY_INODE)) == 0) &&
5167 (inode->i_state & I_DIRTY_TIME)) {
5168 struct ext4_inode_info *ei = EXT4_I(inode);
5170 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
5171 spin_unlock(&inode->i_lock);
5173 spin_lock(&ei->i_raw_lock);
5174 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
5175 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
5176 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
5177 ext4_inode_csum_set(inode, oi->raw_inode, ei);
5178 spin_unlock(&ei->i_raw_lock);
5179 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
5182 spin_unlock(&inode->i_lock);
5187 * Opportunistically update the other time fields for other inodes in
5188 * the same inode table block.
5190 static void ext4_update_other_inodes_time(struct super_block *sb,
5191 unsigned long orig_ino, char *buf)
5193 struct other_inode oi;
5195 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5196 int inode_size = EXT4_INODE_SIZE(sb);
5198 oi.orig_ino = orig_ino;
5200 * Calculate the first inode in the inode table block. Inode
5201 * numbers are one-based. That is, the first inode in a block
5202 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5204 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5205 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5206 if (ino == orig_ino)
5208 oi.raw_inode = (struct ext4_inode *) buf;
5209 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
5214 * Post the struct inode info into an on-disk inode location in the
5215 * buffer-cache. This gobbles the caller's reference to the
5216 * buffer_head in the inode location struct.
5218 * The caller must have write access to iloc->bh.
5220 static int ext4_do_update_inode(handle_t *handle,
5221 struct inode *inode,
5222 struct ext4_iloc *iloc)
5224 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5225 struct ext4_inode_info *ei = EXT4_I(inode);
5226 struct buffer_head *bh = iloc->bh;
5227 struct super_block *sb = inode->i_sb;
5228 int err = 0, rc, block;
5229 int need_datasync = 0, set_large_file = 0;
5234 spin_lock(&ei->i_raw_lock);
5236 /* For fields not tracked in the in-memory inode,
5237 * initialise them to zero for new inodes. */
5238 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5239 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5241 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5242 i_uid = i_uid_read(inode);
5243 i_gid = i_gid_read(inode);
5244 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5245 if (!(test_opt(inode->i_sb, NO_UID32))) {
5246 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5247 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5249 * Fix up interoperability with old kernels. Otherwise, old inodes get
5250 * re-used with the upper 16 bits of the uid/gid intact
5252 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5253 raw_inode->i_uid_high = 0;
5254 raw_inode->i_gid_high = 0;
5256 raw_inode->i_uid_high =
5257 cpu_to_le16(high_16_bits(i_uid));
5258 raw_inode->i_gid_high =
5259 cpu_to_le16(high_16_bits(i_gid));
5262 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5263 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5264 raw_inode->i_uid_high = 0;
5265 raw_inode->i_gid_high = 0;
5267 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5269 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5270 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5271 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5272 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5274 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5276 spin_unlock(&ei->i_raw_lock);
5279 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5280 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5281 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5282 raw_inode->i_file_acl_high =
5283 cpu_to_le16(ei->i_file_acl >> 32);
5284 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5285 if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
5286 ext4_isize_set(raw_inode, ei->i_disksize);
5289 if (ei->i_disksize > 0x7fffffffULL) {
5290 if (!ext4_has_feature_large_file(sb) ||
5291 EXT4_SB(sb)->s_es->s_rev_level ==
5292 cpu_to_le32(EXT4_GOOD_OLD_REV))
5295 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5296 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5297 if (old_valid_dev(inode->i_rdev)) {
5298 raw_inode->i_block[0] =
5299 cpu_to_le32(old_encode_dev(inode->i_rdev));
5300 raw_inode->i_block[1] = 0;
5302 raw_inode->i_block[0] = 0;
5303 raw_inode->i_block[1] =
5304 cpu_to_le32(new_encode_dev(inode->i_rdev));
5305 raw_inode->i_block[2] = 0;
5307 } else if (!ext4_has_inline_data(inode)) {
5308 for (block = 0; block < EXT4_N_BLOCKS; block++)
5309 raw_inode->i_block[block] = ei->i_data[block];
5312 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5313 u64 ivers = ext4_inode_peek_iversion(inode);
5315 raw_inode->i_disk_version = cpu_to_le32(ivers);
5316 if (ei->i_extra_isize) {
5317 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5318 raw_inode->i_version_hi =
5319 cpu_to_le32(ivers >> 32);
5320 raw_inode->i_extra_isize =
5321 cpu_to_le16(ei->i_extra_isize);
5325 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5326 i_projid != EXT4_DEF_PROJID);
5328 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5329 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5330 raw_inode->i_projid = cpu_to_le32(i_projid);
5332 ext4_inode_csum_set(inode, raw_inode, ei);
5333 spin_unlock(&ei->i_raw_lock);
5334 if (inode->i_sb->s_flags & SB_LAZYTIME)
5335 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5338 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5339 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5342 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5343 if (set_large_file) {
5344 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5345 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5348 ext4_set_feature_large_file(sb);
5349 ext4_handle_sync(handle);
5350 err = ext4_handle_dirty_super(handle, sb);
5352 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5355 ext4_std_error(inode->i_sb, err);
5360 * ext4_write_inode()
5362 * We are called from a few places:
5364 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5365 * Here, there will be no transaction running. We wait for any running
5366 * transaction to commit.
5368 * - Within flush work (sys_sync(), kupdate and such).
5369 * We wait on commit, if told to.
5371 * - Within iput_final() -> write_inode_now()
5372 * We wait on commit, if told to.
5374 * In all cases it is actually safe for us to return without doing anything,
5375 * because the inode has been copied into a raw inode buffer in
5376 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5379 * Note that we are absolutely dependent upon all inode dirtiers doing the
5380 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5381 * which we are interested.
5383 * It would be a bug for them to not do this. The code:
5385 * mark_inode_dirty(inode)
5387 * inode->i_size = expr;
5389 * is in error because write_inode() could occur while `stuff()' is running,
5390 * and the new i_size will be lost. Plus the inode will no longer be on the
5391 * superblock's dirty inode list.
5393 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5397 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5398 sb_rdonly(inode->i_sb))
5401 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5404 if (EXT4_SB(inode->i_sb)->s_journal) {
5405 if (ext4_journal_current_handle()) {
5406 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5412 * No need to force transaction in WB_SYNC_NONE mode. Also
5413 * ext4_sync_fs() will force the commit after everything is
5416 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5419 err = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5420 EXT4_I(inode)->i_sync_tid);
5422 struct ext4_iloc iloc;
5424 err = __ext4_get_inode_loc(inode, &iloc, 0);
5428 * sync(2) will flush the whole buffer cache. No need to do
5429 * it here separately for each inode.
5431 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5432 sync_dirty_buffer(iloc.bh);
5433 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5434 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5435 "IO error syncing inode");
5444 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5445 * buffers that are attached to a page stradding i_size and are undergoing
5446 * commit. In that case we have to wait for commit to finish and try again.
5448 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5452 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5453 tid_t commit_tid = 0;
5456 offset = inode->i_size & (PAGE_SIZE - 1);
5458 * All buffers in the last page remain valid? Then there's nothing to
5459 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5462 if (offset > PAGE_SIZE - i_blocksize(inode))
5465 page = find_lock_page(inode->i_mapping,
5466 inode->i_size >> PAGE_SHIFT);
5469 ret = __ext4_journalled_invalidatepage(page, offset,
5470 PAGE_SIZE - offset);
5476 read_lock(&journal->j_state_lock);
5477 if (journal->j_committing_transaction)
5478 commit_tid = journal->j_committing_transaction->t_tid;
5479 read_unlock(&journal->j_state_lock);
5481 jbd2_log_wait_commit(journal, commit_tid);
5488 * Called from notify_change.
5490 * We want to trap VFS attempts to truncate the file as soon as
5491 * possible. In particular, we want to make sure that when the VFS
5492 * shrinks i_size, we put the inode on the orphan list and modify
5493 * i_disksize immediately, so that during the subsequent flushing of
5494 * dirty pages and freeing of disk blocks, we can guarantee that any
5495 * commit will leave the blocks being flushed in an unused state on
5496 * disk. (On recovery, the inode will get truncated and the blocks will
5497 * be freed, so we have a strong guarantee that no future commit will
5498 * leave these blocks visible to the user.)
5500 * Another thing we have to assure is that if we are in ordered mode
5501 * and inode is still attached to the committing transaction, we must
5502 * we start writeout of all the dirty pages which are being truncated.
5503 * This way we are sure that all the data written in the previous
5504 * transaction are already on disk (truncate waits for pages under
5507 * Called with inode->i_mutex down.
5509 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5511 struct inode *inode = d_inode(dentry);
5514 const unsigned int ia_valid = attr->ia_valid;
5516 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5519 error = setattr_prepare(dentry, attr);
5523 error = fscrypt_prepare_setattr(dentry, attr);
5527 if (is_quota_modification(inode, attr)) {
5528 error = dquot_initialize(inode);
5532 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5533 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5536 /* (user+group)*(old+new) structure, inode write (sb,
5537 * inode block, ? - but truncate inode update has it) */
5538 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5539 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5540 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5541 if (IS_ERR(handle)) {
5542 error = PTR_ERR(handle);
5546 /* dquot_transfer() calls back ext4_get_inode_usage() which
5547 * counts xattr inode references.
5549 down_read(&EXT4_I(inode)->xattr_sem);
5550 error = dquot_transfer(inode, attr);
5551 up_read(&EXT4_I(inode)->xattr_sem);
5554 ext4_journal_stop(handle);
5557 /* Update corresponding info in inode so that everything is in
5558 * one transaction */
5559 if (attr->ia_valid & ATTR_UID)
5560 inode->i_uid = attr->ia_uid;
5561 if (attr->ia_valid & ATTR_GID)
5562 inode->i_gid = attr->ia_gid;
5563 error = ext4_mark_inode_dirty(handle, inode);
5564 ext4_journal_stop(handle);
5567 if (attr->ia_valid & ATTR_SIZE) {
5569 loff_t oldsize = inode->i_size;
5570 int shrink = (attr->ia_size <= inode->i_size);
5572 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5573 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5575 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5578 if (!S_ISREG(inode->i_mode))
5581 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5582 inode_inc_iversion(inode);
5584 if (ext4_should_order_data(inode) &&
5585 (attr->ia_size < inode->i_size)) {
5586 error = ext4_begin_ordered_truncate(inode,
5591 if (attr->ia_size != inode->i_size) {
5592 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5593 if (IS_ERR(handle)) {
5594 error = PTR_ERR(handle);
5597 if (ext4_handle_valid(handle) && shrink) {
5598 error = ext4_orphan_add(handle, inode);
5602 * Update c/mtime on truncate up, ext4_truncate() will
5603 * update c/mtime in shrink case below
5606 inode->i_mtime = current_time(inode);
5607 inode->i_ctime = inode->i_mtime;
5609 down_write(&EXT4_I(inode)->i_data_sem);
5610 EXT4_I(inode)->i_disksize = attr->ia_size;
5611 rc = ext4_mark_inode_dirty(handle, inode);
5615 * We have to update i_size under i_data_sem together
5616 * with i_disksize to avoid races with writeback code
5617 * running ext4_wb_update_i_disksize().
5620 i_size_write(inode, attr->ia_size);
5621 up_write(&EXT4_I(inode)->i_data_sem);
5622 ext4_journal_stop(handle);
5625 ext4_orphan_del(NULL, inode);
5630 pagecache_isize_extended(inode, oldsize, inode->i_size);
5633 * Blocks are going to be removed from the inode. Wait
5634 * for dio in flight. Temporarily disable
5635 * dioread_nolock to prevent livelock.
5638 if (!ext4_should_journal_data(inode)) {
5639 inode_dio_wait(inode);
5641 ext4_wait_for_tail_page_commit(inode);
5643 down_write(&EXT4_I(inode)->i_mmap_sem);
5645 rc = ext4_break_layouts(inode);
5647 up_write(&EXT4_I(inode)->i_mmap_sem);
5653 * Truncate pagecache after we've waited for commit
5654 * in data=journal mode to make pages freeable.
5656 truncate_pagecache(inode, inode->i_size);
5658 rc = ext4_truncate(inode);
5662 up_write(&EXT4_I(inode)->i_mmap_sem);
5666 setattr_copy(inode, attr);
5667 mark_inode_dirty(inode);
5671 * If the call to ext4_truncate failed to get a transaction handle at
5672 * all, we need to clean up the in-core orphan list manually.
5674 if (orphan && inode->i_nlink)
5675 ext4_orphan_del(NULL, inode);
5677 if (!error && (ia_valid & ATTR_MODE))
5678 rc = posix_acl_chmod(inode, inode->i_mode);
5681 ext4_std_error(inode->i_sb, error);
5687 int ext4_getattr(const struct path *path, struct kstat *stat,
5688 u32 request_mask, unsigned int query_flags)
5690 struct inode *inode = d_inode(path->dentry);
5691 struct ext4_inode *raw_inode;
5692 struct ext4_inode_info *ei = EXT4_I(inode);
5695 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5696 stat->result_mask |= STATX_BTIME;
5697 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5698 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5701 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5702 if (flags & EXT4_APPEND_FL)
5703 stat->attributes |= STATX_ATTR_APPEND;
5704 if (flags & EXT4_COMPR_FL)
5705 stat->attributes |= STATX_ATTR_COMPRESSED;
5706 if (flags & EXT4_ENCRYPT_FL)
5707 stat->attributes |= STATX_ATTR_ENCRYPTED;
5708 if (flags & EXT4_IMMUTABLE_FL)
5709 stat->attributes |= STATX_ATTR_IMMUTABLE;
5710 if (flags & EXT4_NODUMP_FL)
5711 stat->attributes |= STATX_ATTR_NODUMP;
5713 stat->attributes_mask |= (STATX_ATTR_APPEND |
5714 STATX_ATTR_COMPRESSED |
5715 STATX_ATTR_ENCRYPTED |
5716 STATX_ATTR_IMMUTABLE |
5719 generic_fillattr(inode, stat);
5723 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5724 u32 request_mask, unsigned int query_flags)
5726 struct inode *inode = d_inode(path->dentry);
5727 u64 delalloc_blocks;
5729 ext4_getattr(path, stat, request_mask, query_flags);
5732 * If there is inline data in the inode, the inode will normally not
5733 * have data blocks allocated (it may have an external xattr block).
5734 * Report at least one sector for such files, so tools like tar, rsync,
5735 * others don't incorrectly think the file is completely sparse.
5737 if (unlikely(ext4_has_inline_data(inode)))
5738 stat->blocks += (stat->size + 511) >> 9;
5741 * We can't update i_blocks if the block allocation is delayed
5742 * otherwise in the case of system crash before the real block
5743 * allocation is done, we will have i_blocks inconsistent with
5744 * on-disk file blocks.
5745 * We always keep i_blocks updated together with real
5746 * allocation. But to not confuse with user, stat
5747 * will return the blocks that include the delayed allocation
5748 * blocks for this file.
5750 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5751 EXT4_I(inode)->i_reserved_data_blocks);
5752 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5756 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5759 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5760 return ext4_ind_trans_blocks(inode, lblocks);
5761 return ext4_ext_index_trans_blocks(inode, pextents);
5765 * Account for index blocks, block groups bitmaps and block group
5766 * descriptor blocks if modify datablocks and index blocks
5767 * worse case, the indexs blocks spread over different block groups
5769 * If datablocks are discontiguous, they are possible to spread over
5770 * different block groups too. If they are contiguous, with flexbg,
5771 * they could still across block group boundary.
5773 * Also account for superblock, inode, quota and xattr blocks
5775 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5778 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5784 * How many index blocks need to touch to map @lblocks logical blocks
5785 * to @pextents physical extents?
5787 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5792 * Now let's see how many group bitmaps and group descriptors need
5795 groups = idxblocks + pextents;
5797 if (groups > ngroups)
5799 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5800 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5802 /* bitmaps and block group descriptor blocks */
5803 ret += groups + gdpblocks;
5805 /* Blocks for super block, inode, quota and xattr blocks */
5806 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5812 * Calculate the total number of credits to reserve to fit
5813 * the modification of a single pages into a single transaction,
5814 * which may include multiple chunks of block allocations.
5816 * This could be called via ext4_write_begin()
5818 * We need to consider the worse case, when
5819 * one new block per extent.
5821 int ext4_writepage_trans_blocks(struct inode *inode)
5823 int bpp = ext4_journal_blocks_per_page(inode);
5826 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5828 /* Account for data blocks for journalled mode */
5829 if (ext4_should_journal_data(inode))
5835 * Calculate the journal credits for a chunk of data modification.
5837 * This is called from DIO, fallocate or whoever calling
5838 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5840 * journal buffers for data blocks are not included here, as DIO
5841 * and fallocate do no need to journal data buffers.
5843 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5845 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5849 * The caller must have previously called ext4_reserve_inode_write().
5850 * Give this, we know that the caller already has write access to iloc->bh.
5852 int ext4_mark_iloc_dirty(handle_t *handle,
5853 struct inode *inode, struct ext4_iloc *iloc)
5857 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5861 if (IS_I_VERSION(inode))
5862 inode_inc_iversion(inode);
5864 /* the do_update_inode consumes one bh->b_count */
5867 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5868 err = ext4_do_update_inode(handle, inode, iloc);
5874 * On success, We end up with an outstanding reference count against
5875 * iloc->bh. This _must_ be cleaned up later.
5879 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5880 struct ext4_iloc *iloc)
5884 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5887 err = ext4_get_inode_loc(inode, iloc);
5889 BUFFER_TRACE(iloc->bh, "get_write_access");
5890 err = ext4_journal_get_write_access(handle, iloc->bh);
5896 ext4_std_error(inode->i_sb, err);
5900 static int __ext4_expand_extra_isize(struct inode *inode,
5901 unsigned int new_extra_isize,
5902 struct ext4_iloc *iloc,
5903 handle_t *handle, int *no_expand)
5905 struct ext4_inode *raw_inode;
5906 struct ext4_xattr_ibody_header *header;
5909 raw_inode = ext4_raw_inode(iloc);
5911 header = IHDR(inode, raw_inode);
5913 /* No extended attributes present */
5914 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5915 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5916 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5917 EXT4_I(inode)->i_extra_isize, 0,
5918 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5919 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5923 /* try to expand with EAs present */
5924 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5928 * Inode size expansion failed; don't try again
5937 * Expand an inode by new_extra_isize bytes.
5938 * Returns 0 on success or negative error number on failure.
5940 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5941 unsigned int new_extra_isize,
5942 struct ext4_iloc iloc,
5948 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5952 * In nojournal mode, we can immediately attempt to expand
5953 * the inode. When journaled, we first need to obtain extra
5954 * buffer credits since we may write into the EA block
5955 * with this same handle. If journal_extend fails, then it will
5956 * only result in a minor loss of functionality for that inode.
5957 * If this is felt to be critical, then e2fsck should be run to
5958 * force a large enough s_min_extra_isize.
5960 if (ext4_handle_valid(handle) &&
5961 jbd2_journal_extend(handle,
5962 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0)
5965 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5968 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5969 handle, &no_expand);
5970 ext4_write_unlock_xattr(inode, &no_expand);
5975 int ext4_expand_extra_isize(struct inode *inode,
5976 unsigned int new_extra_isize,
5977 struct ext4_iloc *iloc)
5983 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5988 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5989 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5990 if (IS_ERR(handle)) {
5991 error = PTR_ERR(handle);
5996 ext4_write_lock_xattr(inode, &no_expand);
5998 BUFFER_TRACE(iloc.bh, "get_write_access");
5999 error = ext4_journal_get_write_access(handle, iloc->bh);
6005 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
6006 handle, &no_expand);
6008 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
6012 ext4_write_unlock_xattr(inode, &no_expand);
6014 ext4_journal_stop(handle);
6019 * What we do here is to mark the in-core inode as clean with respect to inode
6020 * dirtiness (it may still be data-dirty).
6021 * This means that the in-core inode may be reaped by prune_icache
6022 * without having to perform any I/O. This is a very good thing,
6023 * because *any* task may call prune_icache - even ones which
6024 * have a transaction open against a different journal.
6026 * Is this cheating? Not really. Sure, we haven't written the
6027 * inode out, but prune_icache isn't a user-visible syncing function.
6028 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6029 * we start and wait on commits.
6031 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
6033 struct ext4_iloc iloc;
6034 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6038 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6039 err = ext4_reserve_inode_write(handle, inode, &iloc);
6043 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6044 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6047 return ext4_mark_iloc_dirty(handle, inode, &iloc);
6051 * ext4_dirty_inode() is called from __mark_inode_dirty()
6053 * We're really interested in the case where a file is being extended.
6054 * i_size has been changed by generic_commit_write() and we thus need
6055 * to include the updated inode in the current transaction.
6057 * Also, dquot_alloc_block() will always dirty the inode when blocks
6058 * are allocated to the file.
6060 * If the inode is marked synchronous, we don't honour that here - doing
6061 * so would cause a commit on atime updates, which we don't bother doing.
6062 * We handle synchronous inodes at the highest possible level.
6064 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
6065 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
6066 * to copy into the on-disk inode structure are the timestamp files.
6068 void ext4_dirty_inode(struct inode *inode, int flags)
6072 if (flags == I_DIRTY_TIME)
6074 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6078 ext4_mark_inode_dirty(handle, inode);
6080 ext4_journal_stop(handle);
6087 * Bind an inode's backing buffer_head into this transaction, to prevent
6088 * it from being flushed to disk early. Unlike
6089 * ext4_reserve_inode_write, this leaves behind no bh reference and
6090 * returns no iloc structure, so the caller needs to repeat the iloc
6091 * lookup to mark the inode dirty later.
6093 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
6095 struct ext4_iloc iloc;
6099 err = ext4_get_inode_loc(inode, &iloc);
6101 BUFFER_TRACE(iloc.bh, "get_write_access");
6102 err = jbd2_journal_get_write_access(handle, iloc.bh);
6104 err = ext4_handle_dirty_metadata(handle,
6110 ext4_std_error(inode->i_sb, err);
6115 int ext4_change_inode_journal_flag(struct inode *inode, int val)
6120 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6123 * We have to be very careful here: changing a data block's
6124 * journaling status dynamically is dangerous. If we write a
6125 * data block to the journal, change the status and then delete
6126 * that block, we risk forgetting to revoke the old log record
6127 * from the journal and so a subsequent replay can corrupt data.
6128 * So, first we make sure that the journal is empty and that
6129 * nobody is changing anything.
6132 journal = EXT4_JOURNAL(inode);
6135 if (is_journal_aborted(journal))
6138 /* Wait for all existing dio workers */
6139 inode_dio_wait(inode);
6142 * Before flushing the journal and switching inode's aops, we have
6143 * to flush all dirty data the inode has. There can be outstanding
6144 * delayed allocations, there can be unwritten extents created by
6145 * fallocate or buffered writes in dioread_nolock mode covered by
6146 * dirty data which can be converted only after flushing the dirty
6147 * data (and journalled aops don't know how to handle these cases).
6150 down_write(&EXT4_I(inode)->i_mmap_sem);
6151 err = filemap_write_and_wait(inode->i_mapping);
6153 up_write(&EXT4_I(inode)->i_mmap_sem);
6158 percpu_down_write(&sbi->s_journal_flag_rwsem);
6159 jbd2_journal_lock_updates(journal);
6162 * OK, there are no updates running now, and all cached data is
6163 * synced to disk. We are now in a completely consistent state
6164 * which doesn't have anything in the journal, and we know that
6165 * no filesystem updates are running, so it is safe to modify
6166 * the inode's in-core data-journaling state flag now.
6170 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6172 err = jbd2_journal_flush(journal);
6174 jbd2_journal_unlock_updates(journal);
6175 percpu_up_write(&sbi->s_journal_flag_rwsem);
6178 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6180 ext4_set_aops(inode);
6182 jbd2_journal_unlock_updates(journal);
6183 percpu_up_write(&sbi->s_journal_flag_rwsem);
6186 up_write(&EXT4_I(inode)->i_mmap_sem);
6188 /* Finally we can mark the inode as dirty. */
6190 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6192 return PTR_ERR(handle);
6194 err = ext4_mark_inode_dirty(handle, inode);
6195 ext4_handle_sync(handle);
6196 ext4_journal_stop(handle);
6197 ext4_std_error(inode->i_sb, err);
6202 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6204 return !buffer_mapped(bh);
6207 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6209 struct vm_area_struct *vma = vmf->vma;
6210 struct page *page = vmf->page;
6215 struct file *file = vma->vm_file;
6216 struct inode *inode = file_inode(file);
6217 struct address_space *mapping = inode->i_mapping;
6219 get_block_t *get_block;
6222 sb_start_pagefault(inode->i_sb);
6223 file_update_time(vma->vm_file);
6225 down_read(&EXT4_I(inode)->i_mmap_sem);
6227 err = ext4_convert_inline_data(inode);
6231 /* Delalloc case is easy... */
6232 if (test_opt(inode->i_sb, DELALLOC) &&
6233 !ext4_should_journal_data(inode) &&
6234 !ext4_nonda_switch(inode->i_sb)) {
6236 err = block_page_mkwrite(vma, vmf,
6237 ext4_da_get_block_prep);
6238 } while (err == -ENOSPC &&
6239 ext4_should_retry_alloc(inode->i_sb, &retries));
6244 size = i_size_read(inode);
6245 /* Page got truncated from under us? */
6246 if (page->mapping != mapping || page_offset(page) > size) {
6248 ret = VM_FAULT_NOPAGE;
6252 if (page->index == size >> PAGE_SHIFT)
6253 len = size & ~PAGE_MASK;
6257 * Return if we have all the buffers mapped. This avoids the need to do
6258 * journal_start/journal_stop which can block and take a long time
6260 if (page_has_buffers(page)) {
6261 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6263 ext4_bh_unmapped)) {
6264 /* Wait so that we don't change page under IO */
6265 wait_for_stable_page(page);
6266 ret = VM_FAULT_LOCKED;
6271 /* OK, we need to fill the hole... */
6272 if (ext4_should_dioread_nolock(inode))
6273 get_block = ext4_get_block_unwritten;
6275 get_block = ext4_get_block;
6277 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6278 ext4_writepage_trans_blocks(inode));
6279 if (IS_ERR(handle)) {
6280 ret = VM_FAULT_SIGBUS;
6283 err = block_page_mkwrite(vma, vmf, get_block);
6284 if (!err && ext4_should_journal_data(inode)) {
6285 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
6286 PAGE_SIZE, NULL, do_journal_get_write_access)) {
6288 ret = VM_FAULT_SIGBUS;
6289 ext4_journal_stop(handle);
6292 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6294 ext4_journal_stop(handle);
6295 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6298 ret = block_page_mkwrite_return(err);
6300 up_read(&EXT4_I(inode)->i_mmap_sem);
6301 sb_end_pagefault(inode->i_sb);
6305 vm_fault_t ext4_filemap_fault(struct vm_fault *vmf)
6307 struct inode *inode = file_inode(vmf->vma->vm_file);
6310 down_read(&EXT4_I(inode)->i_mmap_sem);
6311 ret = filemap_fault(vmf);
6312 up_read(&EXT4_I(inode)->i_mmap_sem);