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 return S_ISLNK(inode->i_mode) && inode->i_size &&
154 (inode->i_size < EXT4_N_BLOCKS * 4);
158 * Restart the transaction associated with *handle. This does a commit,
159 * so before we call here everything must be consistently dirtied against
162 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
168 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
169 * moment, get_block can be called only for blocks inside i_size since
170 * page cache has been already dropped and writes are blocked by
171 * i_mutex. So we can safely drop the i_data_sem here.
173 BUG_ON(EXT4_JOURNAL(inode) == NULL);
174 jbd_debug(2, "restarting handle %p\n", handle);
175 up_write(&EXT4_I(inode)->i_data_sem);
176 ret = ext4_journal_restart(handle, nblocks);
177 down_write(&EXT4_I(inode)->i_data_sem);
178 ext4_discard_preallocations(inode);
184 * Called at the last iput() if i_nlink is zero.
186 void ext4_evict_inode(struct inode *inode)
190 int extra_credits = 3;
191 struct ext4_xattr_inode_array *ea_inode_array = NULL;
193 trace_ext4_evict_inode(inode);
195 if (inode->i_nlink) {
197 * When journalling data dirty buffers are tracked only in the
198 * journal. So although mm thinks everything is clean and
199 * ready for reaping the inode might still have some pages to
200 * write in the running transaction or waiting to be
201 * checkpointed. Thus calling jbd2_journal_invalidatepage()
202 * (via truncate_inode_pages()) to discard these buffers can
203 * cause data loss. Also even if we did not discard these
204 * buffers, we would have no way to find them after the inode
205 * is reaped and thus user could see stale data if he tries to
206 * read them before the transaction is checkpointed. So be
207 * careful and force everything to disk here... We use
208 * ei->i_datasync_tid to store the newest transaction
209 * containing inode's data.
211 * Note that directories do not have this problem because they
212 * don't use page cache.
214 if (inode->i_ino != EXT4_JOURNAL_INO &&
215 ext4_should_journal_data(inode) &&
216 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
217 inode->i_data.nrpages) {
218 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
219 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
221 jbd2_complete_transaction(journal, commit_tid);
222 filemap_write_and_wait(&inode->i_data);
224 truncate_inode_pages_final(&inode->i_data);
229 if (is_bad_inode(inode))
231 dquot_initialize(inode);
233 if (ext4_should_order_data(inode))
234 ext4_begin_ordered_truncate(inode, 0);
235 truncate_inode_pages_final(&inode->i_data);
238 * Protect us against freezing - iput() caller didn't have to have any
239 * protection against it
241 sb_start_intwrite(inode->i_sb);
243 if (!IS_NOQUOTA(inode))
244 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
246 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
247 ext4_blocks_for_truncate(inode)+extra_credits);
248 if (IS_ERR(handle)) {
249 ext4_std_error(inode->i_sb, PTR_ERR(handle));
251 * If we're going to skip the normal cleanup, we still need to
252 * make sure that the in-core orphan linked list is properly
255 ext4_orphan_del(NULL, inode);
256 sb_end_intwrite(inode->i_sb);
261 ext4_handle_sync(handle);
264 * Set inode->i_size to 0 before calling ext4_truncate(). We need
265 * special handling of symlinks here because i_size is used to
266 * determine whether ext4_inode_info->i_data contains symlink data or
267 * block mappings. Setting i_size to 0 will remove its fast symlink
268 * status. Erase i_data so that it becomes a valid empty block map.
270 if (ext4_inode_is_fast_symlink(inode))
271 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
273 err = ext4_mark_inode_dirty(handle, inode);
275 ext4_warning(inode->i_sb,
276 "couldn't mark inode dirty (err %d)", err);
279 if (inode->i_blocks) {
280 err = ext4_truncate(inode);
282 ext4_error(inode->i_sb,
283 "couldn't truncate inode %lu (err %d)",
289 /* Remove xattr references. */
290 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
293 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
295 ext4_journal_stop(handle);
296 ext4_orphan_del(NULL, inode);
297 sb_end_intwrite(inode->i_sb);
298 ext4_xattr_inode_array_free(ea_inode_array);
303 * Kill off the orphan record which ext4_truncate created.
304 * AKPM: I think this can be inside the above `if'.
305 * Note that ext4_orphan_del() has to be able to cope with the
306 * deletion of a non-existent orphan - this is because we don't
307 * know if ext4_truncate() actually created an orphan record.
308 * (Well, we could do this if we need to, but heck - it works)
310 ext4_orphan_del(handle, inode);
311 EXT4_I(inode)->i_dtime = get_seconds();
314 * One subtle ordering requirement: if anything has gone wrong
315 * (transaction abort, IO errors, whatever), then we can still
316 * do these next steps (the fs will already have been marked as
317 * having errors), but we can't free the inode if the mark_dirty
320 if (ext4_mark_inode_dirty(handle, inode))
321 /* If that failed, just do the required in-core inode clear. */
322 ext4_clear_inode(inode);
324 ext4_free_inode(handle, inode);
325 ext4_journal_stop(handle);
326 sb_end_intwrite(inode->i_sb);
327 ext4_xattr_inode_array_free(ea_inode_array);
330 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
334 qsize_t *ext4_get_reserved_space(struct inode *inode)
336 return &EXT4_I(inode)->i_reserved_quota;
341 * Called with i_data_sem down, which is important since we can call
342 * ext4_discard_preallocations() from here.
344 void ext4_da_update_reserve_space(struct inode *inode,
345 int used, int quota_claim)
347 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
348 struct ext4_inode_info *ei = EXT4_I(inode);
350 spin_lock(&ei->i_block_reservation_lock);
351 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
352 if (unlikely(used > ei->i_reserved_data_blocks)) {
353 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
354 "with only %d reserved data blocks",
355 __func__, inode->i_ino, used,
356 ei->i_reserved_data_blocks);
358 used = ei->i_reserved_data_blocks;
361 /* Update per-inode reservations */
362 ei->i_reserved_data_blocks -= used;
363 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
365 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
367 /* Update quota subsystem for data blocks */
369 dquot_claim_block(inode, EXT4_C2B(sbi, used));
372 * We did fallocate with an offset that is already delayed
373 * allocated. So on delayed allocated writeback we should
374 * not re-claim the quota for fallocated blocks.
376 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
380 * If we have done all the pending block allocations and if
381 * there aren't any writers on the inode, we can discard the
382 * inode's preallocations.
384 if ((ei->i_reserved_data_blocks == 0) &&
385 (atomic_read(&inode->i_writecount) == 0))
386 ext4_discard_preallocations(inode);
389 static int __check_block_validity(struct inode *inode, const char *func,
391 struct ext4_map_blocks *map)
393 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
395 ext4_error_inode(inode, func, line, map->m_pblk,
396 "lblock %lu mapped to illegal pblock "
397 "(length %d)", (unsigned long) map->m_lblk,
399 return -EFSCORRUPTED;
404 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
409 if (ext4_encrypted_inode(inode))
410 return fscrypt_zeroout_range(inode, lblk, pblk, len);
412 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
419 #define check_block_validity(inode, map) \
420 __check_block_validity((inode), __func__, __LINE__, (map))
422 #ifdef ES_AGGRESSIVE_TEST
423 static void ext4_map_blocks_es_recheck(handle_t *handle,
425 struct ext4_map_blocks *es_map,
426 struct ext4_map_blocks *map,
433 * There is a race window that the result is not the same.
434 * e.g. xfstests #223 when dioread_nolock enables. The reason
435 * is that we lookup a block mapping in extent status tree with
436 * out taking i_data_sem. So at the time the unwritten extent
437 * could be converted.
439 down_read(&EXT4_I(inode)->i_data_sem);
440 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
441 retval = ext4_ext_map_blocks(handle, inode, map, flags &
442 EXT4_GET_BLOCKS_KEEP_SIZE);
444 retval = ext4_ind_map_blocks(handle, inode, map, flags &
445 EXT4_GET_BLOCKS_KEEP_SIZE);
447 up_read((&EXT4_I(inode)->i_data_sem));
450 * We don't check m_len because extent will be collpased in status
451 * tree. So the m_len might not equal.
453 if (es_map->m_lblk != map->m_lblk ||
454 es_map->m_flags != map->m_flags ||
455 es_map->m_pblk != map->m_pblk) {
456 printk("ES cache assertion failed for inode: %lu "
457 "es_cached ex [%d/%d/%llu/%x] != "
458 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
459 inode->i_ino, es_map->m_lblk, es_map->m_len,
460 es_map->m_pblk, es_map->m_flags, map->m_lblk,
461 map->m_len, map->m_pblk, map->m_flags,
465 #endif /* ES_AGGRESSIVE_TEST */
468 * The ext4_map_blocks() function tries to look up the requested blocks,
469 * and returns if the blocks are already mapped.
471 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
472 * and store the allocated blocks in the result buffer head and mark it
475 * If file type is extents based, it will call ext4_ext_map_blocks(),
476 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
479 * On success, it returns the number of blocks being mapped or allocated. if
480 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
481 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
483 * It returns 0 if plain look up failed (blocks have not been allocated), in
484 * that case, @map is returned as unmapped but we still do fill map->m_len to
485 * indicate the length of a hole starting at map->m_lblk.
487 * It returns the error in case of allocation failure.
489 int ext4_map_blocks(handle_t *handle, struct inode *inode,
490 struct ext4_map_blocks *map, int flags)
492 struct extent_status es;
495 #ifdef ES_AGGRESSIVE_TEST
496 struct ext4_map_blocks orig_map;
498 memcpy(&orig_map, map, sizeof(*map));
502 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
503 "logical block %lu\n", inode->i_ino, flags, map->m_len,
504 (unsigned long) map->m_lblk);
507 * ext4_map_blocks returns an int, and m_len is an unsigned int
509 if (unlikely(map->m_len > INT_MAX))
510 map->m_len = INT_MAX;
512 /* We can handle the block number less than EXT_MAX_BLOCKS */
513 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
514 return -EFSCORRUPTED;
516 /* Lookup extent status tree firstly */
517 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
518 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
519 map->m_pblk = ext4_es_pblock(&es) +
520 map->m_lblk - es.es_lblk;
521 map->m_flags |= ext4_es_is_written(&es) ?
522 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
523 retval = es.es_len - (map->m_lblk - es.es_lblk);
524 if (retval > map->m_len)
527 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
529 retval = es.es_len - (map->m_lblk - es.es_lblk);
530 if (retval > map->m_len)
537 #ifdef ES_AGGRESSIVE_TEST
538 ext4_map_blocks_es_recheck(handle, inode, map,
545 * Try to see if we can get the block without requesting a new
548 down_read(&EXT4_I(inode)->i_data_sem);
549 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
550 retval = ext4_ext_map_blocks(handle, inode, map, flags &
551 EXT4_GET_BLOCKS_KEEP_SIZE);
553 retval = ext4_ind_map_blocks(handle, inode, map, flags &
554 EXT4_GET_BLOCKS_KEEP_SIZE);
559 if (unlikely(retval != map->m_len)) {
560 ext4_warning(inode->i_sb,
561 "ES len assertion failed for inode "
562 "%lu: retval %d != map->m_len %d",
563 inode->i_ino, retval, map->m_len);
567 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
568 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
569 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
570 !(status & EXTENT_STATUS_WRITTEN) &&
571 ext4_find_delalloc_range(inode, map->m_lblk,
572 map->m_lblk + map->m_len - 1))
573 status |= EXTENT_STATUS_DELAYED;
574 ret = ext4_es_insert_extent(inode, map->m_lblk,
575 map->m_len, map->m_pblk, status);
579 up_read((&EXT4_I(inode)->i_data_sem));
582 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
583 ret = check_block_validity(inode, map);
588 /* If it is only a block(s) look up */
589 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
593 * Returns if the blocks have already allocated
595 * Note that if blocks have been preallocated
596 * ext4_ext_get_block() returns the create = 0
597 * with buffer head unmapped.
599 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
601 * If we need to convert extent to unwritten
602 * we continue and do the actual work in
603 * ext4_ext_map_blocks()
605 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
609 * Here we clear m_flags because after allocating an new extent,
610 * it will be set again.
612 map->m_flags &= ~EXT4_MAP_FLAGS;
615 * New blocks allocate and/or writing to unwritten extent
616 * will possibly result in updating i_data, so we take
617 * the write lock of i_data_sem, and call get_block()
618 * with create == 1 flag.
620 down_write(&EXT4_I(inode)->i_data_sem);
623 * We need to check for EXT4 here because migrate
624 * could have changed the inode type in between
626 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
627 retval = ext4_ext_map_blocks(handle, inode, map, flags);
629 retval = ext4_ind_map_blocks(handle, inode, map, flags);
631 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
633 * We allocated new blocks which will result in
634 * i_data's format changing. Force the migrate
635 * to fail by clearing migrate flags
637 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
641 * Update reserved blocks/metadata blocks after successful
642 * block allocation which had been deferred till now. We don't
643 * support fallocate for non extent files. So we can update
644 * reserve space here.
647 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
648 ext4_da_update_reserve_space(inode, retval, 1);
654 if (unlikely(retval != map->m_len)) {
655 ext4_warning(inode->i_sb,
656 "ES len assertion failed for inode "
657 "%lu: retval %d != map->m_len %d",
658 inode->i_ino, retval, map->m_len);
663 * We have to zeroout blocks before inserting them into extent
664 * status tree. Otherwise someone could look them up there and
665 * use them before they are really zeroed. We also have to
666 * unmap metadata before zeroing as otherwise writeback can
667 * overwrite zeros with stale data from block device.
669 if (flags & EXT4_GET_BLOCKS_ZERO &&
670 map->m_flags & EXT4_MAP_MAPPED &&
671 map->m_flags & EXT4_MAP_NEW) {
672 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
674 ret = ext4_issue_zeroout(inode, map->m_lblk,
675 map->m_pblk, map->m_len);
683 * If the extent has been zeroed out, we don't need to update
684 * extent status tree.
686 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
687 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
688 if (ext4_es_is_written(&es))
691 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
692 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
693 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
694 !(status & EXTENT_STATUS_WRITTEN) &&
695 ext4_find_delalloc_range(inode, map->m_lblk,
696 map->m_lblk + map->m_len - 1))
697 status |= EXTENT_STATUS_DELAYED;
698 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
699 map->m_pblk, status);
707 up_write((&EXT4_I(inode)->i_data_sem));
708 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
709 ret = check_block_validity(inode, map);
714 * Inodes with freshly allocated blocks where contents will be
715 * visible after transaction commit must be on transaction's
718 if (map->m_flags & EXT4_MAP_NEW &&
719 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
720 !(flags & EXT4_GET_BLOCKS_ZERO) &&
721 !ext4_is_quota_file(inode) &&
722 ext4_should_order_data(inode)) {
723 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
724 ret = ext4_jbd2_inode_add_wait(handle, inode);
726 ret = ext4_jbd2_inode_add_write(handle, inode);
735 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
736 * we have to be careful as someone else may be manipulating b_state as well.
738 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
740 unsigned long old_state;
741 unsigned long new_state;
743 flags &= EXT4_MAP_FLAGS;
745 /* Dummy buffer_head? Set non-atomically. */
747 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
751 * Someone else may be modifying b_state. Be careful! This is ugly but
752 * once we get rid of using bh as a container for mapping information
753 * to pass to / from get_block functions, this can go away.
756 old_state = READ_ONCE(bh->b_state);
757 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
759 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
762 static int _ext4_get_block(struct inode *inode, sector_t iblock,
763 struct buffer_head *bh, int flags)
765 struct ext4_map_blocks map;
768 if (ext4_has_inline_data(inode))
772 map.m_len = bh->b_size >> inode->i_blkbits;
774 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
777 map_bh(bh, inode->i_sb, map.m_pblk);
778 ext4_update_bh_state(bh, map.m_flags);
779 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
781 } else if (ret == 0) {
782 /* hole case, need to fill in bh->b_size */
783 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
788 int ext4_get_block(struct inode *inode, sector_t iblock,
789 struct buffer_head *bh, int create)
791 return _ext4_get_block(inode, iblock, bh,
792 create ? EXT4_GET_BLOCKS_CREATE : 0);
796 * Get block function used when preparing for buffered write if we require
797 * creating an unwritten extent if blocks haven't been allocated. The extent
798 * will be converted to written after the IO is complete.
800 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
801 struct buffer_head *bh_result, int create)
803 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
804 inode->i_ino, create);
805 return _ext4_get_block(inode, iblock, bh_result,
806 EXT4_GET_BLOCKS_IO_CREATE_EXT);
809 /* Maximum number of blocks we map for direct IO at once. */
810 #define DIO_MAX_BLOCKS 4096
813 * Get blocks function for the cases that need to start a transaction -
814 * generally difference cases of direct IO and DAX IO. It also handles retries
817 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
818 struct buffer_head *bh_result, int flags)
825 /* Trim mapping request to maximum we can map at once for DIO */
826 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
827 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
828 dio_credits = ext4_chunk_trans_blocks(inode,
829 bh_result->b_size >> inode->i_blkbits);
831 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
833 return PTR_ERR(handle);
835 ret = _ext4_get_block(inode, iblock, bh_result, flags);
836 ext4_journal_stop(handle);
838 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
843 /* Get block function for DIO reads and writes to inodes without extents */
844 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
845 struct buffer_head *bh, int create)
847 /* We don't expect handle for direct IO */
848 WARN_ON_ONCE(ext4_journal_current_handle());
851 return _ext4_get_block(inode, iblock, bh, 0);
852 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
856 * Get block function for AIO DIO writes when we create unwritten extent if
857 * blocks are not allocated yet. The extent will be converted to written
858 * after IO is complete.
860 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
861 sector_t iblock, struct buffer_head *bh_result, int create)
865 /* We don't expect handle for direct IO */
866 WARN_ON_ONCE(ext4_journal_current_handle());
868 ret = ext4_get_block_trans(inode, iblock, bh_result,
869 EXT4_GET_BLOCKS_IO_CREATE_EXT);
872 * When doing DIO using unwritten extents, we need io_end to convert
873 * unwritten extents to written on IO completion. We allocate io_end
874 * once we spot unwritten extent and store it in b_private. Generic
875 * DIO code keeps b_private set and furthermore passes the value to
876 * our completion callback in 'private' argument.
878 if (!ret && buffer_unwritten(bh_result)) {
879 if (!bh_result->b_private) {
880 ext4_io_end_t *io_end;
882 io_end = ext4_init_io_end(inode, GFP_KERNEL);
885 bh_result->b_private = io_end;
886 ext4_set_io_unwritten_flag(inode, io_end);
888 set_buffer_defer_completion(bh_result);
895 * Get block function for non-AIO DIO writes when we create unwritten extent if
896 * blocks are not allocated yet. The extent will be converted to written
897 * after IO is complete by ext4_direct_IO_write().
899 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
900 sector_t iblock, struct buffer_head *bh_result, int create)
904 /* We don't expect handle for direct IO */
905 WARN_ON_ONCE(ext4_journal_current_handle());
907 ret = ext4_get_block_trans(inode, iblock, bh_result,
908 EXT4_GET_BLOCKS_IO_CREATE_EXT);
911 * Mark inode as having pending DIO writes to unwritten extents.
912 * ext4_direct_IO_write() checks this flag and converts extents to
915 if (!ret && buffer_unwritten(bh_result))
916 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
921 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
922 struct buffer_head *bh_result, int create)
926 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
927 inode->i_ino, create);
928 /* We don't expect handle for direct IO */
929 WARN_ON_ONCE(ext4_journal_current_handle());
931 ret = _ext4_get_block(inode, iblock, bh_result, 0);
933 * Blocks should have been preallocated! ext4_file_write_iter() checks
936 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
943 * `handle' can be NULL if create is zero
945 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
946 ext4_lblk_t block, int map_flags)
948 struct ext4_map_blocks map;
949 struct buffer_head *bh;
950 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
953 J_ASSERT(handle != NULL || create == 0);
957 err = ext4_map_blocks(handle, inode, &map, map_flags);
960 return create ? ERR_PTR(-ENOSPC) : NULL;
964 bh = sb_getblk(inode->i_sb, map.m_pblk);
966 return ERR_PTR(-ENOMEM);
967 if (map.m_flags & EXT4_MAP_NEW) {
968 J_ASSERT(create != 0);
969 J_ASSERT(handle != NULL);
972 * Now that we do not always journal data, we should
973 * keep in mind whether this should always journal the
974 * new buffer as metadata. For now, regular file
975 * writes use ext4_get_block instead, so it's not a
979 BUFFER_TRACE(bh, "call get_create_access");
980 err = ext4_journal_get_create_access(handle, bh);
985 if (!buffer_uptodate(bh)) {
986 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
987 set_buffer_uptodate(bh);
990 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
991 err = ext4_handle_dirty_metadata(handle, inode, bh);
995 BUFFER_TRACE(bh, "not a new buffer");
1002 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1003 ext4_lblk_t block, int map_flags)
1005 struct buffer_head *bh;
1007 bh = ext4_getblk(handle, inode, block, map_flags);
1010 if (!bh || buffer_uptodate(bh))
1012 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1014 if (buffer_uptodate(bh))
1017 return ERR_PTR(-EIO);
1020 /* Read a contiguous batch of blocks. */
1021 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1022 bool wait, struct buffer_head **bhs)
1026 for (i = 0; i < bh_count; i++) {
1027 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1028 if (IS_ERR(bhs[i])) {
1029 err = PTR_ERR(bhs[i]);
1035 for (i = 0; i < bh_count; i++)
1036 /* Note that NULL bhs[i] is valid because of holes. */
1037 if (bhs[i] && !buffer_uptodate(bhs[i]))
1038 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1044 for (i = 0; i < bh_count; i++)
1046 wait_on_buffer(bhs[i]);
1048 for (i = 0; i < bh_count; i++) {
1049 if (bhs[i] && !buffer_uptodate(bhs[i])) {
1057 for (i = 0; i < bh_count; i++) {
1064 int ext4_walk_page_buffers(handle_t *handle,
1065 struct buffer_head *head,
1069 int (*fn)(handle_t *handle,
1070 struct buffer_head *bh))
1072 struct buffer_head *bh;
1073 unsigned block_start, block_end;
1074 unsigned blocksize = head->b_size;
1076 struct buffer_head *next;
1078 for (bh = head, block_start = 0;
1079 ret == 0 && (bh != head || !block_start);
1080 block_start = block_end, bh = next) {
1081 next = bh->b_this_page;
1082 block_end = block_start + blocksize;
1083 if (block_end <= from || block_start >= to) {
1084 if (partial && !buffer_uptodate(bh))
1088 err = (*fn)(handle, bh);
1096 * To preserve ordering, it is essential that the hole instantiation and
1097 * the data write be encapsulated in a single transaction. We cannot
1098 * close off a transaction and start a new one between the ext4_get_block()
1099 * and the commit_write(). So doing the jbd2_journal_start at the start of
1100 * prepare_write() is the right place.
1102 * Also, this function can nest inside ext4_writepage(). In that case, we
1103 * *know* that ext4_writepage() has generated enough buffer credits to do the
1104 * whole page. So we won't block on the journal in that case, which is good,
1105 * because the caller may be PF_MEMALLOC.
1107 * By accident, ext4 can be reentered when a transaction is open via
1108 * quota file writes. If we were to commit the transaction while thus
1109 * reentered, there can be a deadlock - we would be holding a quota
1110 * lock, and the commit would never complete if another thread had a
1111 * transaction open and was blocking on the quota lock - a ranking
1114 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1115 * will _not_ run commit under these circumstances because handle->h_ref
1116 * is elevated. We'll still have enough credits for the tiny quotafile
1119 int do_journal_get_write_access(handle_t *handle,
1120 struct buffer_head *bh)
1122 int dirty = buffer_dirty(bh);
1125 if (!buffer_mapped(bh) || buffer_freed(bh))
1128 * __block_write_begin() could have dirtied some buffers. Clean
1129 * the dirty bit as jbd2_journal_get_write_access() could complain
1130 * otherwise about fs integrity issues. Setting of the dirty bit
1131 * by __block_write_begin() isn't a real problem here as we clear
1132 * the bit before releasing a page lock and thus writeback cannot
1133 * ever write the buffer.
1136 clear_buffer_dirty(bh);
1137 BUFFER_TRACE(bh, "get write access");
1138 ret = ext4_journal_get_write_access(handle, bh);
1140 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1144 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1145 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1146 get_block_t *get_block)
1148 unsigned from = pos & (PAGE_SIZE - 1);
1149 unsigned to = from + len;
1150 struct inode *inode = page->mapping->host;
1151 unsigned block_start, block_end;
1154 unsigned blocksize = inode->i_sb->s_blocksize;
1156 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1157 bool decrypt = false;
1159 BUG_ON(!PageLocked(page));
1160 BUG_ON(from > PAGE_SIZE);
1161 BUG_ON(to > PAGE_SIZE);
1164 if (!page_has_buffers(page))
1165 create_empty_buffers(page, blocksize, 0);
1166 head = page_buffers(page);
1167 bbits = ilog2(blocksize);
1168 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1170 for (bh = head, block_start = 0; bh != head || !block_start;
1171 block++, block_start = block_end, bh = bh->b_this_page) {
1172 block_end = block_start + blocksize;
1173 if (block_end <= from || block_start >= to) {
1174 if (PageUptodate(page)) {
1175 if (!buffer_uptodate(bh))
1176 set_buffer_uptodate(bh);
1181 clear_buffer_new(bh);
1182 if (!buffer_mapped(bh)) {
1183 WARN_ON(bh->b_size != blocksize);
1184 err = get_block(inode, block, bh, 1);
1187 if (buffer_new(bh)) {
1188 clean_bdev_bh_alias(bh);
1189 if (PageUptodate(page)) {
1190 clear_buffer_new(bh);
1191 set_buffer_uptodate(bh);
1192 mark_buffer_dirty(bh);
1195 if (block_end > to || block_start < from)
1196 zero_user_segments(page, to, block_end,
1201 if (PageUptodate(page)) {
1202 if (!buffer_uptodate(bh))
1203 set_buffer_uptodate(bh);
1206 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1207 !buffer_unwritten(bh) &&
1208 (block_start < from || block_end > to)) {
1209 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1211 decrypt = ext4_encrypted_inode(inode) &&
1212 S_ISREG(inode->i_mode);
1216 * If we issued read requests, let them complete.
1218 while (wait_bh > wait) {
1219 wait_on_buffer(*--wait_bh);
1220 if (!buffer_uptodate(*wait_bh))
1224 page_zero_new_buffers(page, from, to);
1226 err = fscrypt_decrypt_page(page->mapping->host, page,
1227 PAGE_SIZE, 0, page->index);
1232 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1233 loff_t pos, unsigned len, unsigned flags,
1234 struct page **pagep, void **fsdata)
1236 struct inode *inode = mapping->host;
1237 int ret, needed_blocks;
1244 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1247 trace_ext4_write_begin(inode, pos, len, flags);
1249 * Reserve one block more for addition to orphan list in case
1250 * we allocate blocks but write fails for some reason
1252 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1253 index = pos >> PAGE_SHIFT;
1254 from = pos & (PAGE_SIZE - 1);
1257 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1258 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1267 * grab_cache_page_write_begin() can take a long time if the
1268 * system is thrashing due to memory pressure, or if the page
1269 * is being written back. So grab it first before we start
1270 * the transaction handle. This also allows us to allocate
1271 * the page (if needed) without using GFP_NOFS.
1274 page = grab_cache_page_write_begin(mapping, index, flags);
1280 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1281 if (IS_ERR(handle)) {
1283 return PTR_ERR(handle);
1287 if (page->mapping != mapping) {
1288 /* The page got truncated from under us */
1291 ext4_journal_stop(handle);
1294 /* In case writeback began while the page was unlocked */
1295 wait_for_stable_page(page);
1297 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1298 if (ext4_should_dioread_nolock(inode))
1299 ret = ext4_block_write_begin(page, pos, len,
1300 ext4_get_block_unwritten);
1302 ret = ext4_block_write_begin(page, pos, len,
1305 if (ext4_should_dioread_nolock(inode))
1306 ret = __block_write_begin(page, pos, len,
1307 ext4_get_block_unwritten);
1309 ret = __block_write_begin(page, pos, len, ext4_get_block);
1311 if (!ret && ext4_should_journal_data(inode)) {
1312 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1314 do_journal_get_write_access);
1320 * __block_write_begin may have instantiated a few blocks
1321 * outside i_size. Trim these off again. Don't need
1322 * i_size_read because we hold i_mutex.
1324 * Add inode to orphan list in case we crash before
1327 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1328 ext4_orphan_add(handle, inode);
1330 ext4_journal_stop(handle);
1331 if (pos + len > inode->i_size) {
1332 ext4_truncate_failed_write(inode);
1334 * If truncate failed early the inode might
1335 * still be on the orphan list; we need to
1336 * make sure the inode is removed from the
1337 * orphan list in that case.
1340 ext4_orphan_del(NULL, inode);
1343 if (ret == -ENOSPC &&
1344 ext4_should_retry_alloc(inode->i_sb, &retries))
1353 /* For write_end() in data=journal mode */
1354 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1357 if (!buffer_mapped(bh) || buffer_freed(bh))
1359 set_buffer_uptodate(bh);
1360 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1361 clear_buffer_meta(bh);
1362 clear_buffer_prio(bh);
1367 * We need to pick up the new inode size which generic_commit_write gave us
1368 * `file' can be NULL - eg, when called from page_symlink().
1370 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1371 * buffers are managed internally.
1373 static int ext4_write_end(struct file *file,
1374 struct address_space *mapping,
1375 loff_t pos, unsigned len, unsigned copied,
1376 struct page *page, void *fsdata)
1378 handle_t *handle = ext4_journal_current_handle();
1379 struct inode *inode = mapping->host;
1380 loff_t old_size = inode->i_size;
1382 int i_size_changed = 0;
1384 trace_ext4_write_end(inode, pos, len, copied);
1385 if (ext4_has_inline_data(inode)) {
1386 ret = ext4_write_inline_data_end(inode, pos, len,
1395 copied = block_write_end(file, mapping, pos,
1396 len, copied, page, fsdata);
1398 * it's important to update i_size while still holding page lock:
1399 * page writeout could otherwise come in and zero beyond i_size.
1401 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1406 pagecache_isize_extended(inode, old_size, pos);
1408 * Don't mark the inode dirty under page lock. First, it unnecessarily
1409 * makes the holding time of page lock longer. Second, it forces lock
1410 * ordering of page lock and transaction start for journaling
1414 ext4_mark_inode_dirty(handle, inode);
1416 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1417 /* if we have allocated more blocks and copied
1418 * less. We will have blocks allocated outside
1419 * inode->i_size. So truncate them
1421 ext4_orphan_add(handle, inode);
1423 ret2 = ext4_journal_stop(handle);
1427 if (pos + len > inode->i_size) {
1428 ext4_truncate_failed_write(inode);
1430 * If truncate failed early the inode might still be
1431 * on the orphan list; we need to make sure the inode
1432 * is removed from the orphan list in that case.
1435 ext4_orphan_del(NULL, inode);
1438 return ret ? ret : copied;
1442 * This is a private version of page_zero_new_buffers() which doesn't
1443 * set the buffer to be dirty, since in data=journalled mode we need
1444 * to call ext4_handle_dirty_metadata() instead.
1446 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1448 unsigned from, unsigned to)
1450 unsigned int block_start = 0, block_end;
1451 struct buffer_head *head, *bh;
1453 bh = head = page_buffers(page);
1455 block_end = block_start + bh->b_size;
1456 if (buffer_new(bh)) {
1457 if (block_end > from && block_start < to) {
1458 if (!PageUptodate(page)) {
1459 unsigned start, size;
1461 start = max(from, block_start);
1462 size = min(to, block_end) - start;
1464 zero_user(page, start, size);
1465 write_end_fn(handle, bh);
1467 clear_buffer_new(bh);
1470 block_start = block_end;
1471 bh = bh->b_this_page;
1472 } while (bh != head);
1475 static int ext4_journalled_write_end(struct file *file,
1476 struct address_space *mapping,
1477 loff_t pos, unsigned len, unsigned copied,
1478 struct page *page, void *fsdata)
1480 handle_t *handle = ext4_journal_current_handle();
1481 struct inode *inode = mapping->host;
1482 loff_t old_size = inode->i_size;
1486 int size_changed = 0;
1488 trace_ext4_journalled_write_end(inode, pos, len, copied);
1489 from = pos & (PAGE_SIZE - 1);
1492 BUG_ON(!ext4_handle_valid(handle));
1494 if (ext4_has_inline_data(inode)) {
1495 ret = ext4_write_inline_data_end(inode, pos, len,
1503 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1505 ext4_journalled_zero_new_buffers(handle, page, from, to);
1507 if (unlikely(copied < len))
1508 ext4_journalled_zero_new_buffers(handle, page,
1510 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1511 from + copied, &partial,
1514 SetPageUptodate(page);
1516 size_changed = ext4_update_inode_size(inode, pos + copied);
1517 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1518 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1523 pagecache_isize_extended(inode, old_size, pos);
1526 ret2 = ext4_mark_inode_dirty(handle, inode);
1531 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1532 /* if we have allocated more blocks and copied
1533 * less. We will have blocks allocated outside
1534 * inode->i_size. So truncate them
1536 ext4_orphan_add(handle, inode);
1539 ret2 = ext4_journal_stop(handle);
1542 if (pos + len > inode->i_size) {
1543 ext4_truncate_failed_write(inode);
1545 * If truncate failed early the inode might still be
1546 * on the orphan list; we need to make sure the inode
1547 * is removed from the orphan list in that case.
1550 ext4_orphan_del(NULL, inode);
1553 return ret ? ret : copied;
1557 * Reserve space for a single cluster
1559 static int ext4_da_reserve_space(struct inode *inode)
1561 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1562 struct ext4_inode_info *ei = EXT4_I(inode);
1566 * We will charge metadata quota at writeout time; this saves
1567 * us from metadata over-estimation, though we may go over by
1568 * a small amount in the end. Here we just reserve for data.
1570 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1574 spin_lock(&ei->i_block_reservation_lock);
1575 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1576 spin_unlock(&ei->i_block_reservation_lock);
1577 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1580 ei->i_reserved_data_blocks++;
1581 trace_ext4_da_reserve_space(inode);
1582 spin_unlock(&ei->i_block_reservation_lock);
1584 return 0; /* success */
1587 static void ext4_da_release_space(struct inode *inode, int to_free)
1589 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1590 struct ext4_inode_info *ei = EXT4_I(inode);
1593 return; /* Nothing to release, exit */
1595 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1597 trace_ext4_da_release_space(inode, to_free);
1598 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1600 * if there aren't enough reserved blocks, then the
1601 * counter is messed up somewhere. Since this
1602 * function is called from invalidate page, it's
1603 * harmless to return without any action.
1605 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1606 "ino %lu, to_free %d with only %d reserved "
1607 "data blocks", inode->i_ino, to_free,
1608 ei->i_reserved_data_blocks);
1610 to_free = ei->i_reserved_data_blocks;
1612 ei->i_reserved_data_blocks -= to_free;
1614 /* update fs dirty data blocks counter */
1615 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1617 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1619 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1622 static void ext4_da_page_release_reservation(struct page *page,
1623 unsigned int offset,
1624 unsigned int length)
1626 int to_release = 0, contiguous_blks = 0;
1627 struct buffer_head *head, *bh;
1628 unsigned int curr_off = 0;
1629 struct inode *inode = page->mapping->host;
1630 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1631 unsigned int stop = offset + length;
1635 BUG_ON(stop > PAGE_SIZE || stop < length);
1637 head = page_buffers(page);
1640 unsigned int next_off = curr_off + bh->b_size;
1642 if (next_off > stop)
1645 if ((offset <= curr_off) && (buffer_delay(bh))) {
1648 clear_buffer_delay(bh);
1649 } else if (contiguous_blks) {
1650 lblk = page->index <<
1651 (PAGE_SHIFT - inode->i_blkbits);
1652 lblk += (curr_off >> inode->i_blkbits) -
1654 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1655 contiguous_blks = 0;
1657 curr_off = next_off;
1658 } while ((bh = bh->b_this_page) != head);
1660 if (contiguous_blks) {
1661 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1662 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1663 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1666 /* If we have released all the blocks belonging to a cluster, then we
1667 * need to release the reserved space for that cluster. */
1668 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1669 while (num_clusters > 0) {
1670 lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1671 ((num_clusters - 1) << sbi->s_cluster_bits);
1672 if (sbi->s_cluster_ratio == 1 ||
1673 !ext4_find_delalloc_cluster(inode, lblk))
1674 ext4_da_release_space(inode, 1);
1681 * Delayed allocation stuff
1684 struct mpage_da_data {
1685 struct inode *inode;
1686 struct writeback_control *wbc;
1688 pgoff_t first_page; /* The first page to write */
1689 pgoff_t next_page; /* Current page to examine */
1690 pgoff_t last_page; /* Last page to examine */
1692 * Extent to map - this can be after first_page because that can be
1693 * fully mapped. We somewhat abuse m_flags to store whether the extent
1694 * is delalloc or unwritten.
1696 struct ext4_map_blocks map;
1697 struct ext4_io_submit io_submit; /* IO submission data */
1698 unsigned int do_map:1;
1701 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1706 struct pagevec pvec;
1707 struct inode *inode = mpd->inode;
1708 struct address_space *mapping = inode->i_mapping;
1710 /* This is necessary when next_page == 0. */
1711 if (mpd->first_page >= mpd->next_page)
1714 index = mpd->first_page;
1715 end = mpd->next_page - 1;
1717 ext4_lblk_t start, last;
1718 start = index << (PAGE_SHIFT - inode->i_blkbits);
1719 last = end << (PAGE_SHIFT - inode->i_blkbits);
1720 ext4_es_remove_extent(inode, start, last - start + 1);
1723 pagevec_init(&pvec);
1724 while (index <= end) {
1725 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1728 for (i = 0; i < nr_pages; i++) {
1729 struct page *page = pvec.pages[i];
1731 BUG_ON(!PageLocked(page));
1732 BUG_ON(PageWriteback(page));
1734 if (page_mapped(page))
1735 clear_page_dirty_for_io(page);
1736 block_invalidatepage(page, 0, PAGE_SIZE);
1737 ClearPageUptodate(page);
1741 pagevec_release(&pvec);
1745 static void ext4_print_free_blocks(struct inode *inode)
1747 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1748 struct super_block *sb = inode->i_sb;
1749 struct ext4_inode_info *ei = EXT4_I(inode);
1751 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1752 EXT4_C2B(EXT4_SB(inode->i_sb),
1753 ext4_count_free_clusters(sb)));
1754 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1755 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1756 (long long) EXT4_C2B(EXT4_SB(sb),
1757 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1758 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1759 (long long) EXT4_C2B(EXT4_SB(sb),
1760 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1761 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1762 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1763 ei->i_reserved_data_blocks);
1767 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1769 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1773 * This function is grabs code from the very beginning of
1774 * ext4_map_blocks, but assumes that the caller is from delayed write
1775 * time. This function looks up the requested blocks and sets the
1776 * buffer delay bit under the protection of i_data_sem.
1778 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1779 struct ext4_map_blocks *map,
1780 struct buffer_head *bh)
1782 struct extent_status es;
1784 sector_t invalid_block = ~((sector_t) 0xffff);
1785 #ifdef ES_AGGRESSIVE_TEST
1786 struct ext4_map_blocks orig_map;
1788 memcpy(&orig_map, map, sizeof(*map));
1791 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1795 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1796 "logical block %lu\n", inode->i_ino, map->m_len,
1797 (unsigned long) map->m_lblk);
1799 /* Lookup extent status tree firstly */
1800 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1801 if (ext4_es_is_hole(&es)) {
1803 down_read(&EXT4_I(inode)->i_data_sem);
1808 * Delayed extent could be allocated by fallocate.
1809 * So we need to check it.
1811 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1812 map_bh(bh, inode->i_sb, invalid_block);
1814 set_buffer_delay(bh);
1818 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1819 retval = es.es_len - (iblock - es.es_lblk);
1820 if (retval > map->m_len)
1821 retval = map->m_len;
1822 map->m_len = retval;
1823 if (ext4_es_is_written(&es))
1824 map->m_flags |= EXT4_MAP_MAPPED;
1825 else if (ext4_es_is_unwritten(&es))
1826 map->m_flags |= EXT4_MAP_UNWRITTEN;
1830 #ifdef ES_AGGRESSIVE_TEST
1831 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1837 * Try to see if we can get the block without requesting a new
1838 * file system block.
1840 down_read(&EXT4_I(inode)->i_data_sem);
1841 if (ext4_has_inline_data(inode))
1843 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1844 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1846 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1852 * XXX: __block_prepare_write() unmaps passed block,
1856 * If the block was allocated from previously allocated cluster,
1857 * then we don't need to reserve it again. However we still need
1858 * to reserve metadata for every block we're going to write.
1860 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1861 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1862 ret = ext4_da_reserve_space(inode);
1864 /* not enough space to reserve */
1870 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1871 ~0, EXTENT_STATUS_DELAYED);
1877 map_bh(bh, inode->i_sb, invalid_block);
1879 set_buffer_delay(bh);
1880 } else if (retval > 0) {
1882 unsigned int status;
1884 if (unlikely(retval != map->m_len)) {
1885 ext4_warning(inode->i_sb,
1886 "ES len assertion failed for inode "
1887 "%lu: retval %d != map->m_len %d",
1888 inode->i_ino, retval, map->m_len);
1892 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1893 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1894 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1895 map->m_pblk, status);
1901 up_read((&EXT4_I(inode)->i_data_sem));
1907 * This is a special get_block_t callback which is used by
1908 * ext4_da_write_begin(). It will either return mapped block or
1909 * reserve space for a single block.
1911 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1912 * We also have b_blocknr = -1 and b_bdev initialized properly
1914 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1915 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1916 * initialized properly.
1918 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1919 struct buffer_head *bh, int create)
1921 struct ext4_map_blocks map;
1924 BUG_ON(create == 0);
1925 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1927 map.m_lblk = iblock;
1931 * first, we need to know whether the block is allocated already
1932 * preallocated blocks are unmapped but should treated
1933 * the same as allocated blocks.
1935 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1939 map_bh(bh, inode->i_sb, map.m_pblk);
1940 ext4_update_bh_state(bh, map.m_flags);
1942 if (buffer_unwritten(bh)) {
1943 /* A delayed write to unwritten bh should be marked
1944 * new and mapped. Mapped ensures that we don't do
1945 * get_block multiple times when we write to the same
1946 * offset and new ensures that we do proper zero out
1947 * for partial write.
1950 set_buffer_mapped(bh);
1955 static int bget_one(handle_t *handle, struct buffer_head *bh)
1961 static int bput_one(handle_t *handle, struct buffer_head *bh)
1967 static int __ext4_journalled_writepage(struct page *page,
1970 struct address_space *mapping = page->mapping;
1971 struct inode *inode = mapping->host;
1972 struct buffer_head *page_bufs = NULL;
1973 handle_t *handle = NULL;
1974 int ret = 0, err = 0;
1975 int inline_data = ext4_has_inline_data(inode);
1976 struct buffer_head *inode_bh = NULL;
1978 ClearPageChecked(page);
1981 BUG_ON(page->index != 0);
1982 BUG_ON(len > ext4_get_max_inline_size(inode));
1983 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1984 if (inode_bh == NULL)
1987 page_bufs = page_buffers(page);
1992 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1996 * We need to release the page lock before we start the
1997 * journal, so grab a reference so the page won't disappear
1998 * out from under us.
2003 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2004 ext4_writepage_trans_blocks(inode));
2005 if (IS_ERR(handle)) {
2006 ret = PTR_ERR(handle);
2008 goto out_no_pagelock;
2010 BUG_ON(!ext4_handle_valid(handle));
2014 if (page->mapping != mapping) {
2015 /* The page got truncated from under us */
2016 ext4_journal_stop(handle);
2022 BUFFER_TRACE(inode_bh, "get write access");
2023 ret = ext4_journal_get_write_access(handle, inode_bh);
2025 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2028 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2029 do_journal_get_write_access);
2031 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2036 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2037 err = ext4_journal_stop(handle);
2041 if (!ext4_has_inline_data(inode))
2042 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2044 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2053 * Note that we don't need to start a transaction unless we're journaling data
2054 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2055 * need to file the inode to the transaction's list in ordered mode because if
2056 * we are writing back data added by write(), the inode is already there and if
2057 * we are writing back data modified via mmap(), no one guarantees in which
2058 * transaction the data will hit the disk. In case we are journaling data, we
2059 * cannot start transaction directly because transaction start ranks above page
2060 * lock so we have to do some magic.
2062 * This function can get called via...
2063 * - ext4_writepages after taking page lock (have journal handle)
2064 * - journal_submit_inode_data_buffers (no journal handle)
2065 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2066 * - grab_page_cache when doing write_begin (have journal handle)
2068 * We don't do any block allocation in this function. If we have page with
2069 * multiple blocks we need to write those buffer_heads that are mapped. This
2070 * is important for mmaped based write. So if we do with blocksize 1K
2071 * truncate(f, 1024);
2072 * a = mmap(f, 0, 4096);
2074 * truncate(f, 4096);
2075 * we have in the page first buffer_head mapped via page_mkwrite call back
2076 * but other buffer_heads would be unmapped but dirty (dirty done via the
2077 * do_wp_page). So writepage should write the first block. If we modify
2078 * the mmap area beyond 1024 we will again get a page_fault and the
2079 * page_mkwrite callback will do the block allocation and mark the
2080 * buffer_heads mapped.
2082 * We redirty the page if we have any buffer_heads that is either delay or
2083 * unwritten in the page.
2085 * We can get recursively called as show below.
2087 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2090 * But since we don't do any block allocation we should not deadlock.
2091 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2093 static int ext4_writepage(struct page *page,
2094 struct writeback_control *wbc)
2099 struct buffer_head *page_bufs = NULL;
2100 struct inode *inode = page->mapping->host;
2101 struct ext4_io_submit io_submit;
2102 bool keep_towrite = false;
2104 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2105 ext4_invalidatepage(page, 0, PAGE_SIZE);
2110 trace_ext4_writepage(page);
2111 size = i_size_read(inode);
2112 if (page->index == size >> PAGE_SHIFT)
2113 len = size & ~PAGE_MASK;
2117 page_bufs = page_buffers(page);
2119 * We cannot do block allocation or other extent handling in this
2120 * function. If there are buffers needing that, we have to redirty
2121 * the page. But we may reach here when we do a journal commit via
2122 * journal_submit_inode_data_buffers() and in that case we must write
2123 * allocated buffers to achieve data=ordered mode guarantees.
2125 * Also, if there is only one buffer per page (the fs block
2126 * size == the page size), if one buffer needs block
2127 * allocation or needs to modify the extent tree to clear the
2128 * unwritten flag, we know that the page can't be written at
2129 * all, so we might as well refuse the write immediately.
2130 * Unfortunately if the block size != page size, we can't as
2131 * easily detect this case using ext4_walk_page_buffers(), but
2132 * for the extremely common case, this is an optimization that
2133 * skips a useless round trip through ext4_bio_write_page().
2135 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2136 ext4_bh_delay_or_unwritten)) {
2137 redirty_page_for_writepage(wbc, page);
2138 if ((current->flags & PF_MEMALLOC) ||
2139 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2141 * For memory cleaning there's no point in writing only
2142 * some buffers. So just bail out. Warn if we came here
2143 * from direct reclaim.
2145 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2150 keep_towrite = true;
2153 if (PageChecked(page) && ext4_should_journal_data(inode))
2155 * It's mmapped pagecache. Add buffers and journal it. There
2156 * doesn't seem much point in redirtying the page here.
2158 return __ext4_journalled_writepage(page, len);
2160 ext4_io_submit_init(&io_submit, wbc);
2161 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2162 if (!io_submit.io_end) {
2163 redirty_page_for_writepage(wbc, page);
2167 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2168 ext4_io_submit(&io_submit);
2169 /* Drop io_end reference we got from init */
2170 ext4_put_io_end_defer(io_submit.io_end);
2174 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2180 BUG_ON(page->index != mpd->first_page);
2181 clear_page_dirty_for_io(page);
2183 * We have to be very careful here! Nothing protects writeback path
2184 * against i_size changes and the page can be writeably mapped into
2185 * page tables. So an application can be growing i_size and writing
2186 * data through mmap while writeback runs. clear_page_dirty_for_io()
2187 * write-protects our page in page tables and the page cannot get
2188 * written to again until we release page lock. So only after
2189 * clear_page_dirty_for_io() we are safe to sample i_size for
2190 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2191 * on the barrier provided by TestClearPageDirty in
2192 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2193 * after page tables are updated.
2195 size = i_size_read(mpd->inode);
2196 if (page->index == size >> PAGE_SHIFT)
2197 len = size & ~PAGE_MASK;
2200 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2202 mpd->wbc->nr_to_write--;
2208 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2211 * mballoc gives us at most this number of blocks...
2212 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2213 * The rest of mballoc seems to handle chunks up to full group size.
2215 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2218 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2220 * @mpd - extent of blocks
2221 * @lblk - logical number of the block in the file
2222 * @bh - buffer head we want to add to the extent
2224 * The function is used to collect contig. blocks in the same state. If the
2225 * buffer doesn't require mapping for writeback and we haven't started the
2226 * extent of buffers to map yet, the function returns 'true' immediately - the
2227 * caller can write the buffer right away. Otherwise the function returns true
2228 * if the block has been added to the extent, false if the block couldn't be
2231 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2232 struct buffer_head *bh)
2234 struct ext4_map_blocks *map = &mpd->map;
2236 /* Buffer that doesn't need mapping for writeback? */
2237 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2238 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2239 /* So far no extent to map => we write the buffer right away */
2240 if (map->m_len == 0)
2245 /* First block in the extent? */
2246 if (map->m_len == 0) {
2247 /* We cannot map unless handle is started... */
2252 map->m_flags = bh->b_state & BH_FLAGS;
2256 /* Don't go larger than mballoc is willing to allocate */
2257 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2260 /* Can we merge the block to our big extent? */
2261 if (lblk == map->m_lblk + map->m_len &&
2262 (bh->b_state & BH_FLAGS) == map->m_flags) {
2270 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2272 * @mpd - extent of blocks for mapping
2273 * @head - the first buffer in the page
2274 * @bh - buffer we should start processing from
2275 * @lblk - logical number of the block in the file corresponding to @bh
2277 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2278 * the page for IO if all buffers in this page were mapped and there's no
2279 * accumulated extent of buffers to map or add buffers in the page to the
2280 * extent of buffers to map. The function returns 1 if the caller can continue
2281 * by processing the next page, 0 if it should stop adding buffers to the
2282 * extent to map because we cannot extend it anymore. It can also return value
2283 * < 0 in case of error during IO submission.
2285 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2286 struct buffer_head *head,
2287 struct buffer_head *bh,
2290 struct inode *inode = mpd->inode;
2292 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2293 >> inode->i_blkbits;
2296 BUG_ON(buffer_locked(bh));
2298 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2299 /* Found extent to map? */
2302 /* Buffer needs mapping and handle is not started? */
2305 /* Everything mapped so far and we hit EOF */
2308 } while (lblk++, (bh = bh->b_this_page) != head);
2309 /* So far everything mapped? Submit the page for IO. */
2310 if (mpd->map.m_len == 0) {
2311 err = mpage_submit_page(mpd, head->b_page);
2315 return lblk < blocks;
2319 * mpage_map_buffers - update buffers corresponding to changed extent and
2320 * submit fully mapped pages for IO
2322 * @mpd - description of extent to map, on return next extent to map
2324 * Scan buffers corresponding to changed extent (we expect corresponding pages
2325 * to be already locked) and update buffer state according to new extent state.
2326 * We map delalloc buffers to their physical location, clear unwritten bits,
2327 * and mark buffers as uninit when we perform writes to unwritten extents
2328 * and do extent conversion after IO is finished. If the last page is not fully
2329 * mapped, we update @map to the next extent in the last page that needs
2330 * mapping. Otherwise we submit the page for IO.
2332 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2334 struct pagevec pvec;
2336 struct inode *inode = mpd->inode;
2337 struct buffer_head *head, *bh;
2338 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2344 start = mpd->map.m_lblk >> bpp_bits;
2345 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2346 lblk = start << bpp_bits;
2347 pblock = mpd->map.m_pblk;
2349 pagevec_init(&pvec);
2350 while (start <= end) {
2351 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2355 for (i = 0; i < nr_pages; i++) {
2356 struct page *page = pvec.pages[i];
2358 bh = head = page_buffers(page);
2360 if (lblk < mpd->map.m_lblk)
2362 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2364 * Buffer after end of mapped extent.
2365 * Find next buffer in the page to map.
2368 mpd->map.m_flags = 0;
2370 * FIXME: If dioread_nolock supports
2371 * blocksize < pagesize, we need to make
2372 * sure we add size mapped so far to
2373 * io_end->size as the following call
2374 * can submit the page for IO.
2376 err = mpage_process_page_bufs(mpd, head,
2378 pagevec_release(&pvec);
2383 if (buffer_delay(bh)) {
2384 clear_buffer_delay(bh);
2385 bh->b_blocknr = pblock++;
2387 clear_buffer_unwritten(bh);
2388 } while (lblk++, (bh = bh->b_this_page) != head);
2391 * FIXME: This is going to break if dioread_nolock
2392 * supports blocksize < pagesize as we will try to
2393 * convert potentially unmapped parts of inode.
2395 mpd->io_submit.io_end->size += PAGE_SIZE;
2396 /* Page fully mapped - let IO run! */
2397 err = mpage_submit_page(mpd, page);
2399 pagevec_release(&pvec);
2403 pagevec_release(&pvec);
2405 /* Extent fully mapped and matches with page boundary. We are done. */
2407 mpd->map.m_flags = 0;
2411 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2413 struct inode *inode = mpd->inode;
2414 struct ext4_map_blocks *map = &mpd->map;
2415 int get_blocks_flags;
2416 int err, dioread_nolock;
2418 trace_ext4_da_write_pages_extent(inode, map);
2420 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2421 * to convert an unwritten extent to be initialized (in the case
2422 * where we have written into one or more preallocated blocks). It is
2423 * possible that we're going to need more metadata blocks than
2424 * previously reserved. However we must not fail because we're in
2425 * writeback and there is nothing we can do about it so it might result
2426 * in data loss. So use reserved blocks to allocate metadata if
2429 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2430 * the blocks in question are delalloc blocks. This indicates
2431 * that the blocks and quotas has already been checked when
2432 * the data was copied into the page cache.
2434 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2435 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2436 EXT4_GET_BLOCKS_IO_SUBMIT;
2437 dioread_nolock = ext4_should_dioread_nolock(inode);
2439 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2440 if (map->m_flags & (1 << BH_Delay))
2441 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2443 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2446 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2447 if (!mpd->io_submit.io_end->handle &&
2448 ext4_handle_valid(handle)) {
2449 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2450 handle->h_rsv_handle = NULL;
2452 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2455 BUG_ON(map->m_len == 0);
2456 if (map->m_flags & EXT4_MAP_NEW) {
2457 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
2464 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2465 * mpd->len and submit pages underlying it for IO
2467 * @handle - handle for journal operations
2468 * @mpd - extent to map
2469 * @give_up_on_write - we set this to true iff there is a fatal error and there
2470 * is no hope of writing the data. The caller should discard
2471 * dirty pages to avoid infinite loops.
2473 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2474 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2475 * them to initialized or split the described range from larger unwritten
2476 * extent. Note that we need not map all the described range since allocation
2477 * can return less blocks or the range is covered by more unwritten extents. We
2478 * cannot map more because we are limited by reserved transaction credits. On
2479 * the other hand we always make sure that the last touched page is fully
2480 * mapped so that it can be written out (and thus forward progress is
2481 * guaranteed). After mapping we submit all mapped pages for IO.
2483 static int mpage_map_and_submit_extent(handle_t *handle,
2484 struct mpage_da_data *mpd,
2485 bool *give_up_on_write)
2487 struct inode *inode = mpd->inode;
2488 struct ext4_map_blocks *map = &mpd->map;
2493 mpd->io_submit.io_end->offset =
2494 ((loff_t)map->m_lblk) << inode->i_blkbits;
2496 err = mpage_map_one_extent(handle, mpd);
2498 struct super_block *sb = inode->i_sb;
2500 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2501 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2502 goto invalidate_dirty_pages;
2504 * Let the uper layers retry transient errors.
2505 * In the case of ENOSPC, if ext4_count_free_blocks()
2506 * is non-zero, a commit should free up blocks.
2508 if ((err == -ENOMEM) ||
2509 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2511 goto update_disksize;
2514 ext4_msg(sb, KERN_CRIT,
2515 "Delayed block allocation failed for "
2516 "inode %lu at logical offset %llu with"
2517 " max blocks %u with error %d",
2519 (unsigned long long)map->m_lblk,
2520 (unsigned)map->m_len, -err);
2521 ext4_msg(sb, KERN_CRIT,
2522 "This should not happen!! Data will "
2525 ext4_print_free_blocks(inode);
2526 invalidate_dirty_pages:
2527 *give_up_on_write = true;
2532 * Update buffer state, submit mapped pages, and get us new
2535 err = mpage_map_and_submit_buffers(mpd);
2537 goto update_disksize;
2538 } while (map->m_len);
2542 * Update on-disk size after IO is submitted. Races with
2543 * truncate are avoided by checking i_size under i_data_sem.
2545 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2546 if (disksize > EXT4_I(inode)->i_disksize) {
2550 down_write(&EXT4_I(inode)->i_data_sem);
2551 i_size = i_size_read(inode);
2552 if (disksize > i_size)
2554 if (disksize > EXT4_I(inode)->i_disksize)
2555 EXT4_I(inode)->i_disksize = disksize;
2556 up_write(&EXT4_I(inode)->i_data_sem);
2557 err2 = ext4_mark_inode_dirty(handle, inode);
2559 ext4_error(inode->i_sb,
2560 "Failed to mark inode %lu dirty",
2569 * Calculate the total number of credits to reserve for one writepages
2570 * iteration. This is called from ext4_writepages(). We map an extent of
2571 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2572 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2573 * bpp - 1 blocks in bpp different extents.
2575 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2577 int bpp = ext4_journal_blocks_per_page(inode);
2579 return ext4_meta_trans_blocks(inode,
2580 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2584 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2585 * and underlying extent to map
2587 * @mpd - where to look for pages
2589 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2590 * IO immediately. When we find a page which isn't mapped we start accumulating
2591 * extent of buffers underlying these pages that needs mapping (formed by
2592 * either delayed or unwritten buffers). We also lock the pages containing
2593 * these buffers. The extent found is returned in @mpd structure (starting at
2594 * mpd->lblk with length mpd->len blocks).
2596 * Note that this function can attach bios to one io_end structure which are
2597 * neither logically nor physically contiguous. Although it may seem as an
2598 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2599 * case as we need to track IO to all buffers underlying a page in one io_end.
2601 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2603 struct address_space *mapping = mpd->inode->i_mapping;
2604 struct pagevec pvec;
2605 unsigned int nr_pages;
2606 long left = mpd->wbc->nr_to_write;
2607 pgoff_t index = mpd->first_page;
2608 pgoff_t end = mpd->last_page;
2611 int blkbits = mpd->inode->i_blkbits;
2613 struct buffer_head *head;
2615 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2616 tag = PAGECACHE_TAG_TOWRITE;
2618 tag = PAGECACHE_TAG_DIRTY;
2620 pagevec_init(&pvec);
2622 mpd->next_page = index;
2623 while (index <= end) {
2624 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2629 for (i = 0; i < nr_pages; i++) {
2630 struct page *page = pvec.pages[i];
2633 * Accumulated enough dirty pages? This doesn't apply
2634 * to WB_SYNC_ALL mode. For integrity sync we have to
2635 * keep going because someone may be concurrently
2636 * dirtying pages, and we might have synced a lot of
2637 * newly appeared dirty pages, but have not synced all
2638 * of the old dirty pages.
2640 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2643 /* If we can't merge this page, we are done. */
2644 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2649 * If the page is no longer dirty, or its mapping no
2650 * longer corresponds to inode we are writing (which
2651 * means it has been truncated or invalidated), or the
2652 * page is already under writeback and we are not doing
2653 * a data integrity writeback, skip the page
2655 if (!PageDirty(page) ||
2656 (PageWriteback(page) &&
2657 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2658 unlikely(page->mapping != mapping)) {
2663 wait_on_page_writeback(page);
2664 BUG_ON(PageWriteback(page));
2666 if (mpd->map.m_len == 0)
2667 mpd->first_page = page->index;
2668 mpd->next_page = page->index + 1;
2669 /* Add all dirty buffers to mpd */
2670 lblk = ((ext4_lblk_t)page->index) <<
2671 (PAGE_SHIFT - blkbits);
2672 head = page_buffers(page);
2673 err = mpage_process_page_bufs(mpd, head, head, lblk);
2679 pagevec_release(&pvec);
2684 pagevec_release(&pvec);
2688 static int __writepage(struct page *page, struct writeback_control *wbc,
2691 struct address_space *mapping = data;
2692 int ret = ext4_writepage(page, wbc);
2693 mapping_set_error(mapping, ret);
2697 static int ext4_writepages(struct address_space *mapping,
2698 struct writeback_control *wbc)
2700 pgoff_t writeback_index = 0;
2701 long nr_to_write = wbc->nr_to_write;
2702 int range_whole = 0;
2704 handle_t *handle = NULL;
2705 struct mpage_da_data mpd;
2706 struct inode *inode = mapping->host;
2707 int needed_blocks, rsv_blocks = 0, ret = 0;
2708 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2710 struct blk_plug plug;
2711 bool give_up_on_write = false;
2713 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2716 percpu_down_read(&sbi->s_journal_flag_rwsem);
2717 trace_ext4_writepages(inode, wbc);
2719 if (dax_mapping(mapping)) {
2720 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev,
2722 goto out_writepages;
2726 * No pages to write? This is mainly a kludge to avoid starting
2727 * a transaction for special inodes like journal inode on last iput()
2728 * because that could violate lock ordering on umount
2730 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2731 goto out_writepages;
2733 if (ext4_should_journal_data(inode)) {
2734 struct blk_plug plug;
2736 blk_start_plug(&plug);
2737 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2738 blk_finish_plug(&plug);
2739 goto out_writepages;
2743 * If the filesystem has aborted, it is read-only, so return
2744 * right away instead of dumping stack traces later on that
2745 * will obscure the real source of the problem. We test
2746 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2747 * the latter could be true if the filesystem is mounted
2748 * read-only, and in that case, ext4_writepages should
2749 * *never* be called, so if that ever happens, we would want
2752 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2753 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2755 goto out_writepages;
2758 if (ext4_should_dioread_nolock(inode)) {
2760 * We may need to convert up to one extent per block in
2761 * the page and we may dirty the inode.
2763 rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits);
2767 * If we have inline data and arrive here, it means that
2768 * we will soon create the block for the 1st page, so
2769 * we'd better clear the inline data here.
2771 if (ext4_has_inline_data(inode)) {
2772 /* Just inode will be modified... */
2773 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2774 if (IS_ERR(handle)) {
2775 ret = PTR_ERR(handle);
2776 goto out_writepages;
2778 BUG_ON(ext4_test_inode_state(inode,
2779 EXT4_STATE_MAY_INLINE_DATA));
2780 ext4_destroy_inline_data(handle, inode);
2781 ext4_journal_stop(handle);
2784 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2787 if (wbc->range_cyclic) {
2788 writeback_index = mapping->writeback_index;
2789 if (writeback_index)
2791 mpd.first_page = writeback_index;
2794 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2795 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2800 ext4_io_submit_init(&mpd.io_submit, wbc);
2802 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2803 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2805 blk_start_plug(&plug);
2808 * First writeback pages that don't need mapping - we can avoid
2809 * starting a transaction unnecessarily and also avoid being blocked
2810 * in the block layer on device congestion while having transaction
2814 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2815 if (!mpd.io_submit.io_end) {
2819 ret = mpage_prepare_extent_to_map(&mpd);
2820 /* Submit prepared bio */
2821 ext4_io_submit(&mpd.io_submit);
2822 ext4_put_io_end_defer(mpd.io_submit.io_end);
2823 mpd.io_submit.io_end = NULL;
2824 /* Unlock pages we didn't use */
2825 mpage_release_unused_pages(&mpd, false);
2829 while (!done && mpd.first_page <= mpd.last_page) {
2830 /* For each extent of pages we use new io_end */
2831 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2832 if (!mpd.io_submit.io_end) {
2838 * We have two constraints: We find one extent to map and we
2839 * must always write out whole page (makes a difference when
2840 * blocksize < pagesize) so that we don't block on IO when we
2841 * try to write out the rest of the page. Journalled mode is
2842 * not supported by delalloc.
2844 BUG_ON(ext4_should_journal_data(inode));
2845 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2847 /* start a new transaction */
2848 handle = ext4_journal_start_with_reserve(inode,
2849 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2850 if (IS_ERR(handle)) {
2851 ret = PTR_ERR(handle);
2852 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2853 "%ld pages, ino %lu; err %d", __func__,
2854 wbc->nr_to_write, inode->i_ino, ret);
2855 /* Release allocated io_end */
2856 ext4_put_io_end(mpd.io_submit.io_end);
2857 mpd.io_submit.io_end = NULL;
2862 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2863 ret = mpage_prepare_extent_to_map(&mpd);
2866 ret = mpage_map_and_submit_extent(handle, &mpd,
2870 * We scanned the whole range (or exhausted
2871 * nr_to_write), submitted what was mapped and
2872 * didn't find anything needing mapping. We are
2879 * Caution: If the handle is synchronous,
2880 * ext4_journal_stop() can wait for transaction commit
2881 * to finish which may depend on writeback of pages to
2882 * complete or on page lock to be released. In that
2883 * case, we have to wait until after after we have
2884 * submitted all the IO, released page locks we hold,
2885 * and dropped io_end reference (for extent conversion
2886 * to be able to complete) before stopping the handle.
2888 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2889 ext4_journal_stop(handle);
2893 /* Submit prepared bio */
2894 ext4_io_submit(&mpd.io_submit);
2895 /* Unlock pages we didn't use */
2896 mpage_release_unused_pages(&mpd, give_up_on_write);
2898 * Drop our io_end reference we got from init. We have
2899 * to be careful and use deferred io_end finishing if
2900 * we are still holding the transaction as we can
2901 * release the last reference to io_end which may end
2902 * up doing unwritten extent conversion.
2905 ext4_put_io_end_defer(mpd.io_submit.io_end);
2906 ext4_journal_stop(handle);
2908 ext4_put_io_end(mpd.io_submit.io_end);
2909 mpd.io_submit.io_end = NULL;
2911 if (ret == -ENOSPC && sbi->s_journal) {
2913 * Commit the transaction which would
2914 * free blocks released in the transaction
2917 jbd2_journal_force_commit_nested(sbi->s_journal);
2921 /* Fatal error - ENOMEM, EIO... */
2926 blk_finish_plug(&plug);
2927 if (!ret && !cycled && wbc->nr_to_write > 0) {
2929 mpd.last_page = writeback_index - 1;
2935 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2937 * Set the writeback_index so that range_cyclic
2938 * mode will write it back later
2940 mapping->writeback_index = mpd.first_page;
2943 trace_ext4_writepages_result(inode, wbc, ret,
2944 nr_to_write - wbc->nr_to_write);
2945 percpu_up_read(&sbi->s_journal_flag_rwsem);
2949 static int ext4_nonda_switch(struct super_block *sb)
2951 s64 free_clusters, dirty_clusters;
2952 struct ext4_sb_info *sbi = EXT4_SB(sb);
2955 * switch to non delalloc mode if we are running low
2956 * on free block. The free block accounting via percpu
2957 * counters can get slightly wrong with percpu_counter_batch getting
2958 * accumulated on each CPU without updating global counters
2959 * Delalloc need an accurate free block accounting. So switch
2960 * to non delalloc when we are near to error range.
2963 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2965 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2967 * Start pushing delalloc when 1/2 of free blocks are dirty.
2969 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2970 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2972 if (2 * free_clusters < 3 * dirty_clusters ||
2973 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2975 * free block count is less than 150% of dirty blocks
2976 * or free blocks is less than watermark
2983 /* We always reserve for an inode update; the superblock could be there too */
2984 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2986 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2989 if (pos + len <= 0x7fffffffULL)
2992 /* We might need to update the superblock to set LARGE_FILE */
2996 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2997 loff_t pos, unsigned len, unsigned flags,
2998 struct page **pagep, void **fsdata)
3000 int ret, retries = 0;
3003 struct inode *inode = mapping->host;
3006 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3009 index = pos >> PAGE_SHIFT;
3011 if (ext4_nonda_switch(inode->i_sb) ||
3012 S_ISLNK(inode->i_mode)) {
3013 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3014 return ext4_write_begin(file, mapping, pos,
3015 len, flags, pagep, fsdata);
3017 *fsdata = (void *)0;
3018 trace_ext4_da_write_begin(inode, pos, len, flags);
3020 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3021 ret = ext4_da_write_inline_data_begin(mapping, inode,
3031 * grab_cache_page_write_begin() can take a long time if the
3032 * system is thrashing due to memory pressure, or if the page
3033 * is being written back. So grab it first before we start
3034 * the transaction handle. This also allows us to allocate
3035 * the page (if needed) without using GFP_NOFS.
3038 page = grab_cache_page_write_begin(mapping, index, flags);
3044 * With delayed allocation, we don't log the i_disksize update
3045 * if there is delayed block allocation. But we still need
3046 * to journalling the i_disksize update if writes to the end
3047 * of file which has an already mapped buffer.
3050 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3051 ext4_da_write_credits(inode, pos, len));
3052 if (IS_ERR(handle)) {
3054 return PTR_ERR(handle);
3058 if (page->mapping != mapping) {
3059 /* The page got truncated from under us */
3062 ext4_journal_stop(handle);
3065 /* In case writeback began while the page was unlocked */
3066 wait_for_stable_page(page);
3068 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3069 ret = ext4_block_write_begin(page, pos, len,
3070 ext4_da_get_block_prep);
3072 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3076 ext4_journal_stop(handle);
3078 * block_write_begin may have instantiated a few blocks
3079 * outside i_size. Trim these off again. Don't need
3080 * i_size_read because we hold i_mutex.
3082 if (pos + len > inode->i_size)
3083 ext4_truncate_failed_write(inode);
3085 if (ret == -ENOSPC &&
3086 ext4_should_retry_alloc(inode->i_sb, &retries))
3098 * Check if we should update i_disksize
3099 * when write to the end of file but not require block allocation
3101 static int ext4_da_should_update_i_disksize(struct page *page,
3102 unsigned long offset)
3104 struct buffer_head *bh;
3105 struct inode *inode = page->mapping->host;
3109 bh = page_buffers(page);
3110 idx = offset >> inode->i_blkbits;
3112 for (i = 0; i < idx; i++)
3113 bh = bh->b_this_page;
3115 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3120 static int ext4_da_write_end(struct file *file,
3121 struct address_space *mapping,
3122 loff_t pos, unsigned len, unsigned copied,
3123 struct page *page, void *fsdata)
3125 struct inode *inode = mapping->host;
3127 handle_t *handle = ext4_journal_current_handle();
3129 unsigned long start, end;
3130 int write_mode = (int)(unsigned long)fsdata;
3132 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3133 return ext4_write_end(file, mapping, pos,
3134 len, copied, page, fsdata);
3136 trace_ext4_da_write_end(inode, pos, len, copied);
3137 start = pos & (PAGE_SIZE - 1);
3138 end = start + copied - 1;
3141 * generic_write_end() will run mark_inode_dirty() if i_size
3142 * changes. So let's piggyback the i_disksize mark_inode_dirty
3145 new_i_size = pos + copied;
3146 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3147 if (ext4_has_inline_data(inode) ||
3148 ext4_da_should_update_i_disksize(page, end)) {
3149 ext4_update_i_disksize(inode, new_i_size);
3150 /* We need to mark inode dirty even if
3151 * new_i_size is less that inode->i_size
3152 * bu greater than i_disksize.(hint delalloc)
3154 ext4_mark_inode_dirty(handle, inode);
3158 if (write_mode != CONVERT_INLINE_DATA &&
3159 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3160 ext4_has_inline_data(inode))
3161 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3164 ret2 = generic_write_end(file, mapping, pos, len, copied,
3170 ret2 = ext4_journal_stop(handle);
3174 return ret ? ret : copied;
3177 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3178 unsigned int length)
3181 * Drop reserved blocks
3183 BUG_ON(!PageLocked(page));
3184 if (!page_has_buffers(page))
3187 ext4_da_page_release_reservation(page, offset, length);
3190 ext4_invalidatepage(page, offset, length);
3196 * Force all delayed allocation blocks to be allocated for a given inode.
3198 int ext4_alloc_da_blocks(struct inode *inode)
3200 trace_ext4_alloc_da_blocks(inode);
3202 if (!EXT4_I(inode)->i_reserved_data_blocks)
3206 * We do something simple for now. The filemap_flush() will
3207 * also start triggering a write of the data blocks, which is
3208 * not strictly speaking necessary (and for users of
3209 * laptop_mode, not even desirable). However, to do otherwise
3210 * would require replicating code paths in:
3212 * ext4_writepages() ->
3213 * write_cache_pages() ---> (via passed in callback function)
3214 * __mpage_da_writepage() -->
3215 * mpage_add_bh_to_extent()
3216 * mpage_da_map_blocks()
3218 * The problem is that write_cache_pages(), located in
3219 * mm/page-writeback.c, marks pages clean in preparation for
3220 * doing I/O, which is not desirable if we're not planning on
3223 * We could call write_cache_pages(), and then redirty all of
3224 * the pages by calling redirty_page_for_writepage() but that
3225 * would be ugly in the extreme. So instead we would need to
3226 * replicate parts of the code in the above functions,
3227 * simplifying them because we wouldn't actually intend to
3228 * write out the pages, but rather only collect contiguous
3229 * logical block extents, call the multi-block allocator, and
3230 * then update the buffer heads with the block allocations.
3232 * For now, though, we'll cheat by calling filemap_flush(),
3233 * which will map the blocks, and start the I/O, but not
3234 * actually wait for the I/O to complete.
3236 return filemap_flush(inode->i_mapping);
3240 * bmap() is special. It gets used by applications such as lilo and by
3241 * the swapper to find the on-disk block of a specific piece of data.
3243 * Naturally, this is dangerous if the block concerned is still in the
3244 * journal. If somebody makes a swapfile on an ext4 data-journaling
3245 * filesystem and enables swap, then they may get a nasty shock when the
3246 * data getting swapped to that swapfile suddenly gets overwritten by
3247 * the original zero's written out previously to the journal and
3248 * awaiting writeback in the kernel's buffer cache.
3250 * So, if we see any bmap calls here on a modified, data-journaled file,
3251 * take extra steps to flush any blocks which might be in the cache.
3253 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3255 struct inode *inode = mapping->host;
3260 * We can get here for an inline file via the FIBMAP ioctl
3262 if (ext4_has_inline_data(inode))
3265 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3266 test_opt(inode->i_sb, DELALLOC)) {
3268 * With delalloc we want to sync the file
3269 * so that we can make sure we allocate
3272 filemap_write_and_wait(mapping);
3275 if (EXT4_JOURNAL(inode) &&
3276 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3278 * This is a REALLY heavyweight approach, but the use of
3279 * bmap on dirty files is expected to be extremely rare:
3280 * only if we run lilo or swapon on a freshly made file
3281 * do we expect this to happen.
3283 * (bmap requires CAP_SYS_RAWIO so this does not
3284 * represent an unprivileged user DOS attack --- we'd be
3285 * in trouble if mortal users could trigger this path at
3288 * NB. EXT4_STATE_JDATA is not set on files other than
3289 * regular files. If somebody wants to bmap a directory
3290 * or symlink and gets confused because the buffer
3291 * hasn't yet been flushed to disk, they deserve
3292 * everything they get.
3295 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3296 journal = EXT4_JOURNAL(inode);
3297 jbd2_journal_lock_updates(journal);
3298 err = jbd2_journal_flush(journal);
3299 jbd2_journal_unlock_updates(journal);
3305 return generic_block_bmap(mapping, block, ext4_get_block);
3308 static int ext4_readpage(struct file *file, struct page *page)
3311 struct inode *inode = page->mapping->host;
3313 trace_ext4_readpage(page);
3315 if (ext4_has_inline_data(inode))
3316 ret = ext4_readpage_inline(inode, page);
3319 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3325 ext4_readpages(struct file *file, struct address_space *mapping,
3326 struct list_head *pages, unsigned nr_pages)
3328 struct inode *inode = mapping->host;
3330 /* If the file has inline data, no need to do readpages. */
3331 if (ext4_has_inline_data(inode))
3334 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3337 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3338 unsigned int length)
3340 trace_ext4_invalidatepage(page, offset, length);
3342 /* No journalling happens on data buffers when this function is used */
3343 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3345 block_invalidatepage(page, offset, length);
3348 static int __ext4_journalled_invalidatepage(struct page *page,
3349 unsigned int offset,
3350 unsigned int length)
3352 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3354 trace_ext4_journalled_invalidatepage(page, offset, length);
3357 * If it's a full truncate we just forget about the pending dirtying
3359 if (offset == 0 && length == PAGE_SIZE)
3360 ClearPageChecked(page);
3362 return jbd2_journal_invalidatepage(journal, page, offset, length);
3365 /* Wrapper for aops... */
3366 static void ext4_journalled_invalidatepage(struct page *page,
3367 unsigned int offset,
3368 unsigned int length)
3370 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3373 static int ext4_releasepage(struct page *page, gfp_t wait)
3375 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3377 trace_ext4_releasepage(page);
3379 /* Page has dirty journalled data -> cannot release */
3380 if (PageChecked(page))
3383 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3385 return try_to_free_buffers(page);
3388 static bool ext4_inode_datasync_dirty(struct inode *inode)
3390 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3393 return !jbd2_transaction_committed(journal,
3394 EXT4_I(inode)->i_datasync_tid);
3395 /* Any metadata buffers to write? */
3396 if (!list_empty(&inode->i_mapping->private_list))
3398 return inode->i_state & I_DIRTY_DATASYNC;
3401 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3402 unsigned flags, struct iomap *iomap)
3404 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3405 unsigned int blkbits = inode->i_blkbits;
3406 unsigned long first_block = offset >> blkbits;
3407 unsigned long last_block = (offset + length - 1) >> blkbits;
3408 struct ext4_map_blocks map;
3409 bool delalloc = false;
3413 if (flags & IOMAP_REPORT) {
3414 if (ext4_has_inline_data(inode)) {
3415 ret = ext4_inline_data_iomap(inode, iomap);
3416 if (ret != -EAGAIN) {
3417 if (ret == 0 && offset >= iomap->length)
3423 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3427 map.m_lblk = first_block;
3428 map.m_len = last_block - first_block + 1;
3430 if (flags & IOMAP_REPORT) {
3431 ret = ext4_map_blocks(NULL, inode, &map, 0);
3436 ext4_lblk_t end = map.m_lblk + map.m_len - 1;
3437 struct extent_status es;
3439 ext4_es_find_delayed_extent_range(inode, map.m_lblk, end, &es);
3441 if (!es.es_len || es.es_lblk > end) {
3442 /* entire range is a hole */
3443 } else if (es.es_lblk > map.m_lblk) {
3444 /* range starts with a hole */
3445 map.m_len = es.es_lblk - map.m_lblk;
3447 ext4_lblk_t offs = 0;
3449 if (es.es_lblk < map.m_lblk)
3450 offs = map.m_lblk - es.es_lblk;
3451 map.m_lblk = es.es_lblk + offs;
3452 map.m_len = es.es_len - offs;
3456 } else if (flags & IOMAP_WRITE) {
3461 /* Trim mapping request to maximum we can map at once for DIO */
3462 if (map.m_len > DIO_MAX_BLOCKS)
3463 map.m_len = DIO_MAX_BLOCKS;
3464 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3467 * Either we allocate blocks and then we don't get unwritten
3468 * extent so we have reserved enough credits, or the blocks
3469 * are already allocated and unwritten and in that case
3470 * extent conversion fits in the credits as well.
3472 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3475 return PTR_ERR(handle);
3477 ret = ext4_map_blocks(handle, inode, &map,
3478 EXT4_GET_BLOCKS_CREATE_ZERO);
3480 ext4_journal_stop(handle);
3481 if (ret == -ENOSPC &&
3482 ext4_should_retry_alloc(inode->i_sb, &retries))
3488 * If we added blocks beyond i_size, we need to make sure they
3489 * will get truncated if we crash before updating i_size in
3490 * ext4_iomap_end(). For faults we don't need to do that (and
3491 * even cannot because for orphan list operations inode_lock is
3492 * required) - if we happen to instantiate block beyond i_size,
3493 * it is because we race with truncate which has already added
3494 * the inode to the orphan list.
3496 if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3497 (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3500 err = ext4_orphan_add(handle, inode);
3502 ext4_journal_stop(handle);
3506 ext4_journal_stop(handle);
3508 ret = ext4_map_blocks(NULL, inode, &map, 0);
3514 if (ext4_inode_datasync_dirty(inode))
3515 iomap->flags |= IOMAP_F_DIRTY;
3516 iomap->bdev = inode->i_sb->s_bdev;
3517 iomap->dax_dev = sbi->s_daxdev;
3518 iomap->offset = first_block << blkbits;
3519 iomap->length = (u64)map.m_len << blkbits;
3522 iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE;
3523 iomap->addr = IOMAP_NULL_ADDR;
3525 if (map.m_flags & EXT4_MAP_MAPPED) {
3526 iomap->type = IOMAP_MAPPED;
3527 } else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3528 iomap->type = IOMAP_UNWRITTEN;
3533 iomap->addr = (u64)map.m_pblk << blkbits;
3536 if (map.m_flags & EXT4_MAP_NEW)
3537 iomap->flags |= IOMAP_F_NEW;
3542 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3543 ssize_t written, unsigned flags, struct iomap *iomap)
3547 int blkbits = inode->i_blkbits;
3548 bool truncate = false;
3550 if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3553 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3554 if (IS_ERR(handle)) {
3555 ret = PTR_ERR(handle);
3558 if (ext4_update_inode_size(inode, offset + written))
3559 ext4_mark_inode_dirty(handle, inode);
3561 * We may need to truncate allocated but not written blocks beyond EOF.
3563 if (iomap->offset + iomap->length >
3564 ALIGN(inode->i_size, 1 << blkbits)) {
3565 ext4_lblk_t written_blk, end_blk;
3567 written_blk = (offset + written) >> blkbits;
3568 end_blk = (offset + length) >> blkbits;
3569 if (written_blk < end_blk && ext4_can_truncate(inode))
3573 * Remove inode from orphan list if we were extending a inode and
3574 * everything went fine.
3576 if (!truncate && inode->i_nlink &&
3577 !list_empty(&EXT4_I(inode)->i_orphan))
3578 ext4_orphan_del(handle, inode);
3579 ext4_journal_stop(handle);
3581 ext4_truncate_failed_write(inode);
3584 * If truncate failed early the inode might still be on the
3585 * orphan list; we need to make sure the inode is removed from
3586 * the orphan list in that case.
3589 ext4_orphan_del(NULL, inode);
3594 const struct iomap_ops ext4_iomap_ops = {
3595 .iomap_begin = ext4_iomap_begin,
3596 .iomap_end = ext4_iomap_end,
3599 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3600 ssize_t size, void *private)
3602 ext4_io_end_t *io_end = private;
3604 /* if not async direct IO just return */
3608 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3609 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3610 io_end, io_end->inode->i_ino, iocb, offset, size);
3613 * Error during AIO DIO. We cannot convert unwritten extents as the
3614 * data was not written. Just clear the unwritten flag and drop io_end.
3617 ext4_clear_io_unwritten_flag(io_end);
3620 io_end->offset = offset;
3621 io_end->size = size;
3622 ext4_put_io_end(io_end);
3628 * Handling of direct IO writes.
3630 * For ext4 extent files, ext4 will do direct-io write even to holes,
3631 * preallocated extents, and those write extend the file, no need to
3632 * fall back to buffered IO.
3634 * For holes, we fallocate those blocks, mark them as unwritten
3635 * If those blocks were preallocated, we mark sure they are split, but
3636 * still keep the range to write as unwritten.
3638 * The unwritten extents will be converted to written when DIO is completed.
3639 * For async direct IO, since the IO may still pending when return, we
3640 * set up an end_io call back function, which will do the conversion
3641 * when async direct IO completed.
3643 * If the O_DIRECT write will extend the file then add this inode to the
3644 * orphan list. So recovery will truncate it back to the original size
3645 * if the machine crashes during the write.
3648 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3650 struct file *file = iocb->ki_filp;
3651 struct inode *inode = file->f_mapping->host;
3652 struct ext4_inode_info *ei = EXT4_I(inode);
3654 loff_t offset = iocb->ki_pos;
3655 size_t count = iov_iter_count(iter);
3657 get_block_t *get_block_func = NULL;
3659 loff_t final_size = offset + count;
3663 if (final_size > inode->i_size) {
3664 /* Credits for sb + inode write */
3665 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3666 if (IS_ERR(handle)) {
3667 ret = PTR_ERR(handle);
3670 ret = ext4_orphan_add(handle, inode);
3672 ext4_journal_stop(handle);
3676 ei->i_disksize = inode->i_size;
3677 ext4_journal_stop(handle);
3680 BUG_ON(iocb->private == NULL);
3683 * Make all waiters for direct IO properly wait also for extent
3684 * conversion. This also disallows race between truncate() and
3685 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3687 inode_dio_begin(inode);
3689 /* If we do a overwrite dio, i_mutex locking can be released */
3690 overwrite = *((int *)iocb->private);
3693 inode_unlock(inode);
3696 * For extent mapped files we could direct write to holes and fallocate.
3698 * Allocated blocks to fill the hole are marked as unwritten to prevent
3699 * parallel buffered read to expose the stale data before DIO complete
3702 * As to previously fallocated extents, ext4 get_block will just simply
3703 * mark the buffer mapped but still keep the extents unwritten.
3705 * For non AIO case, we will convert those unwritten extents to written
3706 * after return back from blockdev_direct_IO. That way we save us from
3707 * allocating io_end structure and also the overhead of offloading
3708 * the extent convertion to a workqueue.
3710 * For async DIO, the conversion needs to be deferred when the
3711 * IO is completed. The ext4 end_io callback function will be
3712 * called to take care of the conversion work. Here for async
3713 * case, we allocate an io_end structure to hook to the iocb.
3715 iocb->private = NULL;
3717 get_block_func = ext4_dio_get_block_overwrite;
3718 else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3719 round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3720 get_block_func = ext4_dio_get_block;
3721 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3722 } else if (is_sync_kiocb(iocb)) {
3723 get_block_func = ext4_dio_get_block_unwritten_sync;
3724 dio_flags = DIO_LOCKING;
3726 get_block_func = ext4_dio_get_block_unwritten_async;
3727 dio_flags = DIO_LOCKING;
3729 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3730 get_block_func, ext4_end_io_dio, NULL,
3733 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3734 EXT4_STATE_DIO_UNWRITTEN)) {
3737 * for non AIO case, since the IO is already
3738 * completed, we could do the conversion right here
3740 err = ext4_convert_unwritten_extents(NULL, inode,
3744 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3747 inode_dio_end(inode);
3748 /* take i_mutex locking again if we do a ovewrite dio */
3752 if (ret < 0 && final_size > inode->i_size)
3753 ext4_truncate_failed_write(inode);
3755 /* Handle extending of i_size after direct IO write */
3759 /* Credits for sb + inode write */
3760 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3761 if (IS_ERR(handle)) {
3762 /* This is really bad luck. We've written the data
3763 * but cannot extend i_size. Bail out and pretend
3764 * the write failed... */
3765 ret = PTR_ERR(handle);
3767 ext4_orphan_del(NULL, inode);
3772 ext4_orphan_del(handle, inode);
3774 loff_t end = offset + ret;
3775 if (end > inode->i_size) {
3776 ei->i_disksize = end;
3777 i_size_write(inode, end);
3779 * We're going to return a positive `ret'
3780 * here due to non-zero-length I/O, so there's
3781 * no way of reporting error returns from
3782 * ext4_mark_inode_dirty() to userspace. So
3785 ext4_mark_inode_dirty(handle, inode);
3788 err = ext4_journal_stop(handle);
3796 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3798 struct address_space *mapping = iocb->ki_filp->f_mapping;
3799 struct inode *inode = mapping->host;
3800 size_t count = iov_iter_count(iter);
3804 * Shared inode_lock is enough for us - it protects against concurrent
3805 * writes & truncates and since we take care of writing back page cache,
3806 * we are protected against page writeback as well.
3808 inode_lock_shared(inode);
3809 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3810 iocb->ki_pos + count - 1);
3813 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3814 iter, ext4_dio_get_block, NULL, NULL, 0);
3816 inode_unlock_shared(inode);
3820 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3822 struct file *file = iocb->ki_filp;
3823 struct inode *inode = file->f_mapping->host;
3824 size_t count = iov_iter_count(iter);
3825 loff_t offset = iocb->ki_pos;
3828 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3829 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3834 * If we are doing data journalling we don't support O_DIRECT
3836 if (ext4_should_journal_data(inode))
3839 /* Let buffer I/O handle the inline data case. */
3840 if (ext4_has_inline_data(inode))
3843 /* DAX uses iomap path now */
3844 if (WARN_ON_ONCE(IS_DAX(inode)))
3847 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3848 if (iov_iter_rw(iter) == READ)
3849 ret = ext4_direct_IO_read(iocb, iter);
3851 ret = ext4_direct_IO_write(iocb, iter);
3852 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3857 * Pages can be marked dirty completely asynchronously from ext4's journalling
3858 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3859 * much here because ->set_page_dirty is called under VFS locks. The page is
3860 * not necessarily locked.
3862 * We cannot just dirty the page and leave attached buffers clean, because the
3863 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3864 * or jbddirty because all the journalling code will explode.
3866 * So what we do is to mark the page "pending dirty" and next time writepage
3867 * is called, propagate that into the buffers appropriately.
3869 static int ext4_journalled_set_page_dirty(struct page *page)
3871 SetPageChecked(page);
3872 return __set_page_dirty_nobuffers(page);
3875 static int ext4_set_page_dirty(struct page *page)
3877 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3878 WARN_ON_ONCE(!page_has_buffers(page));
3879 return __set_page_dirty_buffers(page);
3882 static const struct address_space_operations ext4_aops = {
3883 .readpage = ext4_readpage,
3884 .readpages = ext4_readpages,
3885 .writepage = ext4_writepage,
3886 .writepages = ext4_writepages,
3887 .write_begin = ext4_write_begin,
3888 .write_end = ext4_write_end,
3889 .set_page_dirty = ext4_set_page_dirty,
3891 .invalidatepage = ext4_invalidatepage,
3892 .releasepage = ext4_releasepage,
3893 .direct_IO = ext4_direct_IO,
3894 .migratepage = buffer_migrate_page,
3895 .is_partially_uptodate = block_is_partially_uptodate,
3896 .error_remove_page = generic_error_remove_page,
3899 static const struct address_space_operations ext4_journalled_aops = {
3900 .readpage = ext4_readpage,
3901 .readpages = ext4_readpages,
3902 .writepage = ext4_writepage,
3903 .writepages = ext4_writepages,
3904 .write_begin = ext4_write_begin,
3905 .write_end = ext4_journalled_write_end,
3906 .set_page_dirty = ext4_journalled_set_page_dirty,
3908 .invalidatepage = ext4_journalled_invalidatepage,
3909 .releasepage = ext4_releasepage,
3910 .direct_IO = ext4_direct_IO,
3911 .is_partially_uptodate = block_is_partially_uptodate,
3912 .error_remove_page = generic_error_remove_page,
3915 static const struct address_space_operations ext4_da_aops = {
3916 .readpage = ext4_readpage,
3917 .readpages = ext4_readpages,
3918 .writepage = ext4_writepage,
3919 .writepages = ext4_writepages,
3920 .write_begin = ext4_da_write_begin,
3921 .write_end = ext4_da_write_end,
3922 .set_page_dirty = ext4_set_page_dirty,
3924 .invalidatepage = ext4_da_invalidatepage,
3925 .releasepage = ext4_releasepage,
3926 .direct_IO = ext4_direct_IO,
3927 .migratepage = buffer_migrate_page,
3928 .is_partially_uptodate = block_is_partially_uptodate,
3929 .error_remove_page = generic_error_remove_page,
3932 void ext4_set_aops(struct inode *inode)
3934 switch (ext4_inode_journal_mode(inode)) {
3935 case EXT4_INODE_ORDERED_DATA_MODE:
3936 case EXT4_INODE_WRITEBACK_DATA_MODE:
3938 case EXT4_INODE_JOURNAL_DATA_MODE:
3939 inode->i_mapping->a_ops = &ext4_journalled_aops;
3944 if (test_opt(inode->i_sb, DELALLOC))
3945 inode->i_mapping->a_ops = &ext4_da_aops;
3947 inode->i_mapping->a_ops = &ext4_aops;
3950 static int __ext4_block_zero_page_range(handle_t *handle,
3951 struct address_space *mapping, loff_t from, loff_t length)
3953 ext4_fsblk_t index = from >> PAGE_SHIFT;
3954 unsigned offset = from & (PAGE_SIZE-1);
3955 unsigned blocksize, pos;
3957 struct inode *inode = mapping->host;
3958 struct buffer_head *bh;
3962 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3963 mapping_gfp_constraint(mapping, ~__GFP_FS));
3967 blocksize = inode->i_sb->s_blocksize;
3969 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3971 if (!page_has_buffers(page))
3972 create_empty_buffers(page, blocksize, 0);
3974 /* Find the buffer that contains "offset" */
3975 bh = page_buffers(page);
3977 while (offset >= pos) {
3978 bh = bh->b_this_page;
3982 if (buffer_freed(bh)) {
3983 BUFFER_TRACE(bh, "freed: skip");
3986 if (!buffer_mapped(bh)) {
3987 BUFFER_TRACE(bh, "unmapped");
3988 ext4_get_block(inode, iblock, bh, 0);
3989 /* unmapped? It's a hole - nothing to do */
3990 if (!buffer_mapped(bh)) {
3991 BUFFER_TRACE(bh, "still unmapped");
3996 /* Ok, it's mapped. Make sure it's up-to-date */
3997 if (PageUptodate(page))
3998 set_buffer_uptodate(bh);
4000 if (!buffer_uptodate(bh)) {
4002 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
4004 /* Uhhuh. Read error. Complain and punt. */
4005 if (!buffer_uptodate(bh))
4007 if (S_ISREG(inode->i_mode) &&
4008 ext4_encrypted_inode(inode)) {
4009 /* We expect the key to be set. */
4010 BUG_ON(!fscrypt_has_encryption_key(inode));
4011 BUG_ON(blocksize != PAGE_SIZE);
4012 WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host,
4013 page, PAGE_SIZE, 0, page->index));
4016 if (ext4_should_journal_data(inode)) {
4017 BUFFER_TRACE(bh, "get write access");
4018 err = ext4_journal_get_write_access(handle, bh);
4022 zero_user(page, offset, length);
4023 BUFFER_TRACE(bh, "zeroed end of block");
4025 if (ext4_should_journal_data(inode)) {
4026 err = ext4_handle_dirty_metadata(handle, inode, bh);
4029 mark_buffer_dirty(bh);
4030 if (ext4_should_order_data(inode))
4031 err = ext4_jbd2_inode_add_write(handle, inode);
4041 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4042 * starting from file offset 'from'. The range to be zero'd must
4043 * be contained with in one block. If the specified range exceeds
4044 * the end of the block it will be shortened to end of the block
4045 * that cooresponds to 'from'
4047 static int ext4_block_zero_page_range(handle_t *handle,
4048 struct address_space *mapping, loff_t from, loff_t length)
4050 struct inode *inode = mapping->host;
4051 unsigned offset = from & (PAGE_SIZE-1);
4052 unsigned blocksize = inode->i_sb->s_blocksize;
4053 unsigned max = blocksize - (offset & (blocksize - 1));
4056 * correct length if it does not fall between
4057 * 'from' and the end of the block
4059 if (length > max || length < 0)
4062 if (IS_DAX(inode)) {
4063 return iomap_zero_range(inode, from, length, NULL,
4066 return __ext4_block_zero_page_range(handle, mapping, from, length);
4070 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4071 * up to the end of the block which corresponds to `from'.
4072 * This required during truncate. We need to physically zero the tail end
4073 * of that block so it doesn't yield old data if the file is later grown.
4075 static int ext4_block_truncate_page(handle_t *handle,
4076 struct address_space *mapping, loff_t from)
4078 unsigned offset = from & (PAGE_SIZE-1);
4081 struct inode *inode = mapping->host;
4083 /* If we are processing an encrypted inode during orphan list handling */
4084 if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode))
4087 blocksize = inode->i_sb->s_blocksize;
4088 length = blocksize - (offset & (blocksize - 1));
4090 return ext4_block_zero_page_range(handle, mapping, from, length);
4093 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4094 loff_t lstart, loff_t length)
4096 struct super_block *sb = inode->i_sb;
4097 struct address_space *mapping = inode->i_mapping;
4098 unsigned partial_start, partial_end;
4099 ext4_fsblk_t start, end;
4100 loff_t byte_end = (lstart + length - 1);
4103 partial_start = lstart & (sb->s_blocksize - 1);
4104 partial_end = byte_end & (sb->s_blocksize - 1);
4106 start = lstart >> sb->s_blocksize_bits;
4107 end = byte_end >> sb->s_blocksize_bits;
4109 /* Handle partial zero within the single block */
4111 (partial_start || (partial_end != sb->s_blocksize - 1))) {
4112 err = ext4_block_zero_page_range(handle, mapping,
4116 /* Handle partial zero out on the start of the range */
4117 if (partial_start) {
4118 err = ext4_block_zero_page_range(handle, mapping,
4119 lstart, sb->s_blocksize);
4123 /* Handle partial zero out on the end of the range */
4124 if (partial_end != sb->s_blocksize - 1)
4125 err = ext4_block_zero_page_range(handle, mapping,
4126 byte_end - partial_end,
4131 int ext4_can_truncate(struct inode *inode)
4133 if (S_ISREG(inode->i_mode))
4135 if (S_ISDIR(inode->i_mode))
4137 if (S_ISLNK(inode->i_mode))
4138 return !ext4_inode_is_fast_symlink(inode);
4143 * We have to make sure i_disksize gets properly updated before we truncate
4144 * page cache due to hole punching or zero range. Otherwise i_disksize update
4145 * can get lost as it may have been postponed to submission of writeback but
4146 * that will never happen after we truncate page cache.
4148 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4152 loff_t size = i_size_read(inode);
4154 WARN_ON(!inode_is_locked(inode));
4155 if (offset > size || offset + len < size)
4158 if (EXT4_I(inode)->i_disksize >= size)
4161 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4163 return PTR_ERR(handle);
4164 ext4_update_i_disksize(inode, size);
4165 ext4_mark_inode_dirty(handle, inode);
4166 ext4_journal_stop(handle);
4172 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4173 * associated with the given offset and length
4175 * @inode: File inode
4176 * @offset: The offset where the hole will begin
4177 * @len: The length of the hole
4179 * Returns: 0 on success or negative on failure
4182 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4184 struct super_block *sb = inode->i_sb;
4185 ext4_lblk_t first_block, stop_block;
4186 struct address_space *mapping = inode->i_mapping;
4187 loff_t first_block_offset, last_block_offset;
4189 unsigned int credits;
4192 if (!S_ISREG(inode->i_mode))
4195 trace_ext4_punch_hole(inode, offset, length, 0);
4198 * Write out all dirty pages to avoid race conditions
4199 * Then release them.
4201 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4202 ret = filemap_write_and_wait_range(mapping, offset,
4203 offset + length - 1);
4210 /* No need to punch hole beyond i_size */
4211 if (offset >= inode->i_size)
4215 * If the hole extends beyond i_size, set the hole
4216 * to end after the page that contains i_size
4218 if (offset + length > inode->i_size) {
4219 length = inode->i_size +
4220 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4224 if (offset & (sb->s_blocksize - 1) ||
4225 (offset + length) & (sb->s_blocksize - 1)) {
4227 * Attach jinode to inode for jbd2 if we do any zeroing of
4230 ret = ext4_inode_attach_jinode(inode);
4236 /* Wait all existing dio workers, newcomers will block on i_mutex */
4237 ext4_inode_block_unlocked_dio(inode);
4238 inode_dio_wait(inode);
4241 * Prevent page faults from reinstantiating pages we have released from
4244 down_write(&EXT4_I(inode)->i_mmap_sem);
4245 first_block_offset = round_up(offset, sb->s_blocksize);
4246 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4248 /* Now release the pages and zero block aligned part of pages*/
4249 if (last_block_offset > first_block_offset) {
4250 ret = ext4_update_disksize_before_punch(inode, offset, length);
4253 truncate_pagecache_range(inode, first_block_offset,
4257 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4258 credits = ext4_writepage_trans_blocks(inode);
4260 credits = ext4_blocks_for_truncate(inode);
4261 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4262 if (IS_ERR(handle)) {
4263 ret = PTR_ERR(handle);
4264 ext4_std_error(sb, ret);
4268 ret = ext4_zero_partial_blocks(handle, inode, offset,
4273 first_block = (offset + sb->s_blocksize - 1) >>
4274 EXT4_BLOCK_SIZE_BITS(sb);
4275 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4277 /* If there are no blocks to remove, return now */
4278 if (first_block >= stop_block)
4281 down_write(&EXT4_I(inode)->i_data_sem);
4282 ext4_discard_preallocations(inode);
4284 ret = ext4_es_remove_extent(inode, first_block,
4285 stop_block - first_block);
4287 up_write(&EXT4_I(inode)->i_data_sem);
4291 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4292 ret = ext4_ext_remove_space(inode, first_block,
4295 ret = ext4_ind_remove_space(handle, inode, first_block,
4298 up_write(&EXT4_I(inode)->i_data_sem);
4300 ext4_handle_sync(handle);
4302 inode->i_mtime = inode->i_ctime = current_time(inode);
4303 ext4_mark_inode_dirty(handle, inode);
4305 ext4_update_inode_fsync_trans(handle, inode, 1);
4307 ext4_journal_stop(handle);
4309 up_write(&EXT4_I(inode)->i_mmap_sem);
4310 ext4_inode_resume_unlocked_dio(inode);
4312 inode_unlock(inode);
4316 int ext4_inode_attach_jinode(struct inode *inode)
4318 struct ext4_inode_info *ei = EXT4_I(inode);
4319 struct jbd2_inode *jinode;
4321 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4324 jinode = jbd2_alloc_inode(GFP_KERNEL);
4325 spin_lock(&inode->i_lock);
4328 spin_unlock(&inode->i_lock);
4331 ei->jinode = jinode;
4332 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4335 spin_unlock(&inode->i_lock);
4336 if (unlikely(jinode != NULL))
4337 jbd2_free_inode(jinode);
4344 * We block out ext4_get_block() block instantiations across the entire
4345 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4346 * simultaneously on behalf of the same inode.
4348 * As we work through the truncate and commit bits of it to the journal there
4349 * is one core, guiding principle: the file's tree must always be consistent on
4350 * disk. We must be able to restart the truncate after a crash.
4352 * The file's tree may be transiently inconsistent in memory (although it
4353 * probably isn't), but whenever we close off and commit a journal transaction,
4354 * the contents of (the filesystem + the journal) must be consistent and
4355 * restartable. It's pretty simple, really: bottom up, right to left (although
4356 * left-to-right works OK too).
4358 * Note that at recovery time, journal replay occurs *before* the restart of
4359 * truncate against the orphan inode list.
4361 * The committed inode has the new, desired i_size (which is the same as
4362 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4363 * that this inode's truncate did not complete and it will again call
4364 * ext4_truncate() to have another go. So there will be instantiated blocks
4365 * to the right of the truncation point in a crashed ext4 filesystem. But
4366 * that's fine - as long as they are linked from the inode, the post-crash
4367 * ext4_truncate() run will find them and release them.
4369 int ext4_truncate(struct inode *inode)
4371 struct ext4_inode_info *ei = EXT4_I(inode);
4372 unsigned int credits;
4375 struct address_space *mapping = inode->i_mapping;
4378 * There is a possibility that we're either freeing the inode
4379 * or it's a completely new inode. In those cases we might not
4380 * have i_mutex locked because it's not necessary.
4382 if (!(inode->i_state & (I_NEW|I_FREEING)))
4383 WARN_ON(!inode_is_locked(inode));
4384 trace_ext4_truncate_enter(inode);
4386 if (!ext4_can_truncate(inode))
4389 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4391 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4392 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4394 if (ext4_has_inline_data(inode)) {
4397 err = ext4_inline_data_truncate(inode, &has_inline);
4404 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4405 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4406 if (ext4_inode_attach_jinode(inode) < 0)
4410 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4411 credits = ext4_writepage_trans_blocks(inode);
4413 credits = ext4_blocks_for_truncate(inode);
4415 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4417 return PTR_ERR(handle);
4419 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4420 ext4_block_truncate_page(handle, mapping, inode->i_size);
4423 * We add the inode to the orphan list, so that if this
4424 * truncate spans multiple transactions, and we crash, we will
4425 * resume the truncate when the filesystem recovers. It also
4426 * marks the inode dirty, to catch the new size.
4428 * Implication: the file must always be in a sane, consistent
4429 * truncatable state while each transaction commits.
4431 err = ext4_orphan_add(handle, inode);
4435 down_write(&EXT4_I(inode)->i_data_sem);
4437 ext4_discard_preallocations(inode);
4439 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4440 err = ext4_ext_truncate(handle, inode);
4442 ext4_ind_truncate(handle, inode);
4444 up_write(&ei->i_data_sem);
4449 ext4_handle_sync(handle);
4453 * If this was a simple ftruncate() and the file will remain alive,
4454 * then we need to clear up the orphan record which we created above.
4455 * However, if this was a real unlink then we were called by
4456 * ext4_evict_inode(), and we allow that function to clean up the
4457 * orphan info for us.
4460 ext4_orphan_del(handle, inode);
4462 inode->i_mtime = inode->i_ctime = current_time(inode);
4463 ext4_mark_inode_dirty(handle, inode);
4464 ext4_journal_stop(handle);
4466 trace_ext4_truncate_exit(inode);
4471 * ext4_get_inode_loc returns with an extra refcount against the inode's
4472 * underlying buffer_head on success. If 'in_mem' is true, we have all
4473 * data in memory that is needed to recreate the on-disk version of this
4476 static int __ext4_get_inode_loc(struct inode *inode,
4477 struct ext4_iloc *iloc, int in_mem)
4479 struct ext4_group_desc *gdp;
4480 struct buffer_head *bh;
4481 struct super_block *sb = inode->i_sb;
4483 int inodes_per_block, inode_offset;
4486 if (!ext4_valid_inum(sb, inode->i_ino))
4487 return -EFSCORRUPTED;
4489 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4490 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4495 * Figure out the offset within the block group inode table
4497 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4498 inode_offset = ((inode->i_ino - 1) %
4499 EXT4_INODES_PER_GROUP(sb));
4500 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4501 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4503 bh = sb_getblk(sb, block);
4506 if (!buffer_uptodate(bh)) {
4510 * If the buffer has the write error flag, we have failed
4511 * to write out another inode in the same block. In this
4512 * case, we don't have to read the block because we may
4513 * read the old inode data successfully.
4515 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4516 set_buffer_uptodate(bh);
4518 if (buffer_uptodate(bh)) {
4519 /* someone brought it uptodate while we waited */
4525 * If we have all information of the inode in memory and this
4526 * is the only valid inode in the block, we need not read the
4530 struct buffer_head *bitmap_bh;
4533 start = inode_offset & ~(inodes_per_block - 1);
4535 /* Is the inode bitmap in cache? */
4536 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4537 if (unlikely(!bitmap_bh))
4541 * If the inode bitmap isn't in cache then the
4542 * optimisation may end up performing two reads instead
4543 * of one, so skip it.
4545 if (!buffer_uptodate(bitmap_bh)) {
4549 for (i = start; i < start + inodes_per_block; i++) {
4550 if (i == inode_offset)
4552 if (ext4_test_bit(i, bitmap_bh->b_data))
4556 if (i == start + inodes_per_block) {
4557 /* all other inodes are free, so skip I/O */
4558 memset(bh->b_data, 0, bh->b_size);
4559 set_buffer_uptodate(bh);
4567 * If we need to do any I/O, try to pre-readahead extra
4568 * blocks from the inode table.
4570 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4571 ext4_fsblk_t b, end, table;
4573 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4575 table = ext4_inode_table(sb, gdp);
4576 /* s_inode_readahead_blks is always a power of 2 */
4577 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4581 num = EXT4_INODES_PER_GROUP(sb);
4582 if (ext4_has_group_desc_csum(sb))
4583 num -= ext4_itable_unused_count(sb, gdp);
4584 table += num / inodes_per_block;
4588 sb_breadahead(sb, b++);
4592 * There are other valid inodes in the buffer, this inode
4593 * has in-inode xattrs, or we don't have this inode in memory.
4594 * Read the block from disk.
4596 trace_ext4_load_inode(inode);
4598 bh->b_end_io = end_buffer_read_sync;
4599 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4601 if (!buffer_uptodate(bh)) {
4602 EXT4_ERROR_INODE_BLOCK(inode, block,
4603 "unable to read itable block");
4613 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4615 /* We have all inode data except xattrs in memory here. */
4616 return __ext4_get_inode_loc(inode, iloc,
4617 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4620 static bool ext4_should_use_dax(struct inode *inode)
4622 if (!test_opt(inode->i_sb, DAX))
4624 if (!S_ISREG(inode->i_mode))
4626 if (ext4_should_journal_data(inode))
4628 if (ext4_has_inline_data(inode))
4630 if (ext4_encrypted_inode(inode))
4635 void ext4_set_inode_flags(struct inode *inode)
4637 unsigned int flags = EXT4_I(inode)->i_flags;
4638 unsigned int new_fl = 0;
4640 if (flags & EXT4_SYNC_FL)
4642 if (flags & EXT4_APPEND_FL)
4644 if (flags & EXT4_IMMUTABLE_FL)
4645 new_fl |= S_IMMUTABLE;
4646 if (flags & EXT4_NOATIME_FL)
4647 new_fl |= S_NOATIME;
4648 if (flags & EXT4_DIRSYNC_FL)
4649 new_fl |= S_DIRSYNC;
4650 if (ext4_should_use_dax(inode))
4652 if (flags & EXT4_ENCRYPT_FL)
4653 new_fl |= S_ENCRYPTED;
4654 inode_set_flags(inode, new_fl,
4655 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4659 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4660 struct ext4_inode_info *ei)
4663 struct inode *inode = &(ei->vfs_inode);
4664 struct super_block *sb = inode->i_sb;
4666 if (ext4_has_feature_huge_file(sb)) {
4667 /* we are using combined 48 bit field */
4668 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4669 le32_to_cpu(raw_inode->i_blocks_lo);
4670 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4671 /* i_blocks represent file system block size */
4672 return i_blocks << (inode->i_blkbits - 9);
4677 return le32_to_cpu(raw_inode->i_blocks_lo);
4681 static inline void ext4_iget_extra_inode(struct inode *inode,
4682 struct ext4_inode *raw_inode,
4683 struct ext4_inode_info *ei)
4685 __le32 *magic = (void *)raw_inode +
4686 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4687 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4688 EXT4_INODE_SIZE(inode->i_sb) &&
4689 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4690 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4691 ext4_find_inline_data_nolock(inode);
4693 EXT4_I(inode)->i_inline_off = 0;
4696 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4698 if (!ext4_has_feature_project(inode->i_sb))
4700 *projid = EXT4_I(inode)->i_projid;
4704 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4706 struct ext4_iloc iloc;
4707 struct ext4_inode *raw_inode;
4708 struct ext4_inode_info *ei;
4709 struct inode *inode;
4710 journal_t *journal = EXT4_SB(sb)->s_journal;
4718 inode = iget_locked(sb, ino);
4720 return ERR_PTR(-ENOMEM);
4721 if (!(inode->i_state & I_NEW))
4727 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4730 raw_inode = ext4_raw_inode(&iloc);
4732 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4733 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4734 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4735 EXT4_INODE_SIZE(inode->i_sb) ||
4736 (ei->i_extra_isize & 3)) {
4737 EXT4_ERROR_INODE(inode,
4738 "bad extra_isize %u (inode size %u)",
4740 EXT4_INODE_SIZE(inode->i_sb));
4741 ret = -EFSCORRUPTED;
4745 ei->i_extra_isize = 0;
4747 /* Precompute checksum seed for inode metadata */
4748 if (ext4_has_metadata_csum(sb)) {
4749 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4751 __le32 inum = cpu_to_le32(inode->i_ino);
4752 __le32 gen = raw_inode->i_generation;
4753 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4755 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4759 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4760 EXT4_ERROR_INODE(inode, "checksum invalid");
4765 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4766 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4767 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4768 if (ext4_has_feature_project(sb) &&
4769 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4770 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4771 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4773 i_projid = EXT4_DEF_PROJID;
4775 if (!(test_opt(inode->i_sb, NO_UID32))) {
4776 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4777 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4779 i_uid_write(inode, i_uid);
4780 i_gid_write(inode, i_gid);
4781 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4782 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4784 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4785 ei->i_inline_off = 0;
4786 ei->i_dir_start_lookup = 0;
4787 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4788 /* We now have enough fields to check if the inode was active or not.
4789 * This is needed because nfsd might try to access dead inodes
4790 * the test is that same one that e2fsck uses
4791 * NeilBrown 1999oct15
4793 if (inode->i_nlink == 0) {
4794 if ((inode->i_mode == 0 ||
4795 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4796 ino != EXT4_BOOT_LOADER_INO) {
4797 /* this inode is deleted */
4801 /* The only unlinked inodes we let through here have
4802 * valid i_mode and are being read by the orphan
4803 * recovery code: that's fine, we're about to complete
4804 * the process of deleting those.
4805 * OR it is the EXT4_BOOT_LOADER_INO which is
4806 * not initialized on a new filesystem. */
4808 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4809 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4810 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4811 if (ext4_has_feature_64bit(sb))
4813 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4814 inode->i_size = ext4_isize(sb, raw_inode);
4815 if ((size = i_size_read(inode)) < 0) {
4816 EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size);
4817 ret = -EFSCORRUPTED;
4820 ei->i_disksize = inode->i_size;
4822 ei->i_reserved_quota = 0;
4824 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4825 ei->i_block_group = iloc.block_group;
4826 ei->i_last_alloc_group = ~0;
4828 * NOTE! The in-memory inode i_data array is in little-endian order
4829 * even on big-endian machines: we do NOT byteswap the block numbers!
4831 for (block = 0; block < EXT4_N_BLOCKS; block++)
4832 ei->i_data[block] = raw_inode->i_block[block];
4833 INIT_LIST_HEAD(&ei->i_orphan);
4836 * Set transaction id's of transactions that have to be committed
4837 * to finish f[data]sync. We set them to currently running transaction
4838 * as we cannot be sure that the inode or some of its metadata isn't
4839 * part of the transaction - the inode could have been reclaimed and
4840 * now it is reread from disk.
4843 transaction_t *transaction;
4846 read_lock(&journal->j_state_lock);
4847 if (journal->j_running_transaction)
4848 transaction = journal->j_running_transaction;
4850 transaction = journal->j_committing_transaction;
4852 tid = transaction->t_tid;
4854 tid = journal->j_commit_sequence;
4855 read_unlock(&journal->j_state_lock);
4856 ei->i_sync_tid = tid;
4857 ei->i_datasync_tid = tid;
4860 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4861 if (ei->i_extra_isize == 0) {
4862 /* The extra space is currently unused. Use it. */
4863 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4864 ei->i_extra_isize = sizeof(struct ext4_inode) -
4865 EXT4_GOOD_OLD_INODE_SIZE;
4867 ext4_iget_extra_inode(inode, raw_inode, ei);
4871 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4872 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4873 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4874 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4876 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4877 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4878 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4879 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4881 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4886 if (ei->i_file_acl &&
4887 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4888 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4890 ret = -EFSCORRUPTED;
4892 } else if (!ext4_has_inline_data(inode)) {
4893 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4894 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4895 (S_ISLNK(inode->i_mode) &&
4896 !ext4_inode_is_fast_symlink(inode))))
4897 /* Validate extent which is part of inode */
4898 ret = ext4_ext_check_inode(inode);
4899 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4900 (S_ISLNK(inode->i_mode) &&
4901 !ext4_inode_is_fast_symlink(inode))) {
4902 /* Validate block references which are part of inode */
4903 ret = ext4_ind_check_inode(inode);
4909 if (S_ISREG(inode->i_mode)) {
4910 inode->i_op = &ext4_file_inode_operations;
4911 inode->i_fop = &ext4_file_operations;
4912 ext4_set_aops(inode);
4913 } else if (S_ISDIR(inode->i_mode)) {
4914 inode->i_op = &ext4_dir_inode_operations;
4915 inode->i_fop = &ext4_dir_operations;
4916 } else if (S_ISLNK(inode->i_mode)) {
4917 if (ext4_encrypted_inode(inode)) {
4918 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4919 ext4_set_aops(inode);
4920 } else if (ext4_inode_is_fast_symlink(inode)) {
4921 inode->i_link = (char *)ei->i_data;
4922 inode->i_op = &ext4_fast_symlink_inode_operations;
4923 nd_terminate_link(ei->i_data, inode->i_size,
4924 sizeof(ei->i_data) - 1);
4926 inode->i_op = &ext4_symlink_inode_operations;
4927 ext4_set_aops(inode);
4929 inode_nohighmem(inode);
4930 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4931 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4932 inode->i_op = &ext4_special_inode_operations;
4933 if (raw_inode->i_block[0])
4934 init_special_inode(inode, inode->i_mode,
4935 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4937 init_special_inode(inode, inode->i_mode,
4938 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4939 } else if (ino == EXT4_BOOT_LOADER_INO) {
4940 make_bad_inode(inode);
4942 ret = -EFSCORRUPTED;
4943 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4947 ext4_set_inode_flags(inode);
4949 unlock_new_inode(inode);
4955 return ERR_PTR(ret);
4958 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4960 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4961 return ERR_PTR(-EFSCORRUPTED);
4962 return ext4_iget(sb, ino);
4965 static int ext4_inode_blocks_set(handle_t *handle,
4966 struct ext4_inode *raw_inode,
4967 struct ext4_inode_info *ei)
4969 struct inode *inode = &(ei->vfs_inode);
4970 u64 i_blocks = inode->i_blocks;
4971 struct super_block *sb = inode->i_sb;
4973 if (i_blocks <= ~0U) {
4975 * i_blocks can be represented in a 32 bit variable
4976 * as multiple of 512 bytes
4978 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4979 raw_inode->i_blocks_high = 0;
4980 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4983 if (!ext4_has_feature_huge_file(sb))
4986 if (i_blocks <= 0xffffffffffffULL) {
4988 * i_blocks can be represented in a 48 bit variable
4989 * as multiple of 512 bytes
4991 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4992 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4993 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4995 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4996 /* i_block is stored in file system block size */
4997 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4998 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4999 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5004 struct other_inode {
5005 unsigned long orig_ino;
5006 struct ext4_inode *raw_inode;
5009 static int other_inode_match(struct inode * inode, unsigned long ino,
5012 struct other_inode *oi = (struct other_inode *) data;
5014 if ((inode->i_ino != ino) ||
5015 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5016 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
5017 ((inode->i_state & I_DIRTY_TIME) == 0))
5019 spin_lock(&inode->i_lock);
5020 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5021 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
5022 (inode->i_state & I_DIRTY_TIME)) {
5023 struct ext4_inode_info *ei = EXT4_I(inode);
5025 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
5026 spin_unlock(&inode->i_lock);
5028 spin_lock(&ei->i_raw_lock);
5029 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
5030 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
5031 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
5032 ext4_inode_csum_set(inode, oi->raw_inode, ei);
5033 spin_unlock(&ei->i_raw_lock);
5034 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
5037 spin_unlock(&inode->i_lock);
5042 * Opportunistically update the other time fields for other inodes in
5043 * the same inode table block.
5045 static void ext4_update_other_inodes_time(struct super_block *sb,
5046 unsigned long orig_ino, char *buf)
5048 struct other_inode oi;
5050 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5051 int inode_size = EXT4_INODE_SIZE(sb);
5053 oi.orig_ino = orig_ino;
5055 * Calculate the first inode in the inode table block. Inode
5056 * numbers are one-based. That is, the first inode in a block
5057 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5059 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5060 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5061 if (ino == orig_ino)
5063 oi.raw_inode = (struct ext4_inode *) buf;
5064 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
5069 * Post the struct inode info into an on-disk inode location in the
5070 * buffer-cache. This gobbles the caller's reference to the
5071 * buffer_head in the inode location struct.
5073 * The caller must have write access to iloc->bh.
5075 static int ext4_do_update_inode(handle_t *handle,
5076 struct inode *inode,
5077 struct ext4_iloc *iloc)
5079 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5080 struct ext4_inode_info *ei = EXT4_I(inode);
5081 struct buffer_head *bh = iloc->bh;
5082 struct super_block *sb = inode->i_sb;
5083 int err = 0, rc, block;
5084 int need_datasync = 0, set_large_file = 0;
5089 spin_lock(&ei->i_raw_lock);
5091 /* For fields not tracked in the in-memory inode,
5092 * initialise them to zero for new inodes. */
5093 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5094 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5096 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5097 i_uid = i_uid_read(inode);
5098 i_gid = i_gid_read(inode);
5099 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5100 if (!(test_opt(inode->i_sb, NO_UID32))) {
5101 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5102 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5104 * Fix up interoperability with old kernels. Otherwise, old inodes get
5105 * re-used with the upper 16 bits of the uid/gid intact
5107 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5108 raw_inode->i_uid_high = 0;
5109 raw_inode->i_gid_high = 0;
5111 raw_inode->i_uid_high =
5112 cpu_to_le16(high_16_bits(i_uid));
5113 raw_inode->i_gid_high =
5114 cpu_to_le16(high_16_bits(i_gid));
5117 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5118 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5119 raw_inode->i_uid_high = 0;
5120 raw_inode->i_gid_high = 0;
5122 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5124 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5125 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5126 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5127 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5129 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5131 spin_unlock(&ei->i_raw_lock);
5134 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5135 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5136 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5137 raw_inode->i_file_acl_high =
5138 cpu_to_le16(ei->i_file_acl >> 32);
5139 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5140 if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
5141 ext4_isize_set(raw_inode, ei->i_disksize);
5144 if (ei->i_disksize > 0x7fffffffULL) {
5145 if (!ext4_has_feature_large_file(sb) ||
5146 EXT4_SB(sb)->s_es->s_rev_level ==
5147 cpu_to_le32(EXT4_GOOD_OLD_REV))
5150 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5151 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5152 if (old_valid_dev(inode->i_rdev)) {
5153 raw_inode->i_block[0] =
5154 cpu_to_le32(old_encode_dev(inode->i_rdev));
5155 raw_inode->i_block[1] = 0;
5157 raw_inode->i_block[0] = 0;
5158 raw_inode->i_block[1] =
5159 cpu_to_le32(new_encode_dev(inode->i_rdev));
5160 raw_inode->i_block[2] = 0;
5162 } else if (!ext4_has_inline_data(inode)) {
5163 for (block = 0; block < EXT4_N_BLOCKS; block++)
5164 raw_inode->i_block[block] = ei->i_data[block];
5167 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5168 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5169 if (ei->i_extra_isize) {
5170 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5171 raw_inode->i_version_hi =
5172 cpu_to_le32(inode->i_version >> 32);
5173 raw_inode->i_extra_isize =
5174 cpu_to_le16(ei->i_extra_isize);
5178 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5179 i_projid != EXT4_DEF_PROJID);
5181 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5182 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5183 raw_inode->i_projid = cpu_to_le32(i_projid);
5185 ext4_inode_csum_set(inode, raw_inode, ei);
5186 spin_unlock(&ei->i_raw_lock);
5187 if (inode->i_sb->s_flags & SB_LAZYTIME)
5188 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5191 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5192 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5195 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5196 if (set_large_file) {
5197 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5198 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5201 ext4_update_dynamic_rev(sb);
5202 ext4_set_feature_large_file(sb);
5203 ext4_handle_sync(handle);
5204 err = ext4_handle_dirty_super(handle, sb);
5206 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5209 ext4_std_error(inode->i_sb, err);
5214 * ext4_write_inode()
5216 * We are called from a few places:
5218 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5219 * Here, there will be no transaction running. We wait for any running
5220 * transaction to commit.
5222 * - Within flush work (sys_sync(), kupdate and such).
5223 * We wait on commit, if told to.
5225 * - Within iput_final() -> write_inode_now()
5226 * We wait on commit, if told to.
5228 * In all cases it is actually safe for us to return without doing anything,
5229 * because the inode has been copied into a raw inode buffer in
5230 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5233 * Note that we are absolutely dependent upon all inode dirtiers doing the
5234 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5235 * which we are interested.
5237 * It would be a bug for them to not do this. The code:
5239 * mark_inode_dirty(inode)
5241 * inode->i_size = expr;
5243 * is in error because write_inode() could occur while `stuff()' is running,
5244 * and the new i_size will be lost. Plus the inode will no longer be on the
5245 * superblock's dirty inode list.
5247 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5251 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5254 if (EXT4_SB(inode->i_sb)->s_journal) {
5255 if (ext4_journal_current_handle()) {
5256 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5262 * No need to force transaction in WB_SYNC_NONE mode. Also
5263 * ext4_sync_fs() will force the commit after everything is
5266 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5269 err = ext4_force_commit(inode->i_sb);
5271 struct ext4_iloc iloc;
5273 err = __ext4_get_inode_loc(inode, &iloc, 0);
5277 * sync(2) will flush the whole buffer cache. No need to do
5278 * it here separately for each inode.
5280 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5281 sync_dirty_buffer(iloc.bh);
5282 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5283 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5284 "IO error syncing inode");
5293 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5294 * buffers that are attached to a page stradding i_size and are undergoing
5295 * commit. In that case we have to wait for commit to finish and try again.
5297 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5301 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5302 tid_t commit_tid = 0;
5305 offset = inode->i_size & (PAGE_SIZE - 1);
5307 * All buffers in the last page remain valid? Then there's nothing to
5308 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5311 if (offset > PAGE_SIZE - i_blocksize(inode))
5314 page = find_lock_page(inode->i_mapping,
5315 inode->i_size >> PAGE_SHIFT);
5318 ret = __ext4_journalled_invalidatepage(page, offset,
5319 PAGE_SIZE - offset);
5325 read_lock(&journal->j_state_lock);
5326 if (journal->j_committing_transaction)
5327 commit_tid = journal->j_committing_transaction->t_tid;
5328 read_unlock(&journal->j_state_lock);
5330 jbd2_log_wait_commit(journal, commit_tid);
5337 * Called from notify_change.
5339 * We want to trap VFS attempts to truncate the file as soon as
5340 * possible. In particular, we want to make sure that when the VFS
5341 * shrinks i_size, we put the inode on the orphan list and modify
5342 * i_disksize immediately, so that during the subsequent flushing of
5343 * dirty pages and freeing of disk blocks, we can guarantee that any
5344 * commit will leave the blocks being flushed in an unused state on
5345 * disk. (On recovery, the inode will get truncated and the blocks will
5346 * be freed, so we have a strong guarantee that no future commit will
5347 * leave these blocks visible to the user.)
5349 * Another thing we have to assure is that if we are in ordered mode
5350 * and inode is still attached to the committing transaction, we must
5351 * we start writeout of all the dirty pages which are being truncated.
5352 * This way we are sure that all the data written in the previous
5353 * transaction are already on disk (truncate waits for pages under
5356 * Called with inode->i_mutex down.
5358 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5360 struct inode *inode = d_inode(dentry);
5363 const unsigned int ia_valid = attr->ia_valid;
5365 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5368 error = setattr_prepare(dentry, attr);
5372 error = fscrypt_prepare_setattr(dentry, attr);
5376 if (is_quota_modification(inode, attr)) {
5377 error = dquot_initialize(inode);
5381 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5382 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5385 /* (user+group)*(old+new) structure, inode write (sb,
5386 * inode block, ? - but truncate inode update has it) */
5387 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5388 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5389 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5390 if (IS_ERR(handle)) {
5391 error = PTR_ERR(handle);
5395 /* dquot_transfer() calls back ext4_get_inode_usage() which
5396 * counts xattr inode references.
5398 down_read(&EXT4_I(inode)->xattr_sem);
5399 error = dquot_transfer(inode, attr);
5400 up_read(&EXT4_I(inode)->xattr_sem);
5403 ext4_journal_stop(handle);
5406 /* Update corresponding info in inode so that everything is in
5407 * one transaction */
5408 if (attr->ia_valid & ATTR_UID)
5409 inode->i_uid = attr->ia_uid;
5410 if (attr->ia_valid & ATTR_GID)
5411 inode->i_gid = attr->ia_gid;
5412 error = ext4_mark_inode_dirty(handle, inode);
5413 ext4_journal_stop(handle);
5416 if (attr->ia_valid & ATTR_SIZE) {
5418 loff_t oldsize = inode->i_size;
5419 int shrink = (attr->ia_size <= inode->i_size);
5421 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5422 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5424 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5427 if (!S_ISREG(inode->i_mode))
5430 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5431 inode_inc_iversion(inode);
5433 if (ext4_should_order_data(inode) &&
5434 (attr->ia_size < inode->i_size)) {
5435 error = ext4_begin_ordered_truncate(inode,
5440 if (attr->ia_size != inode->i_size) {
5441 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5442 if (IS_ERR(handle)) {
5443 error = PTR_ERR(handle);
5446 if (ext4_handle_valid(handle) && shrink) {
5447 error = ext4_orphan_add(handle, inode);
5451 * Update c/mtime on truncate up, ext4_truncate() will
5452 * update c/mtime in shrink case below
5455 inode->i_mtime = current_time(inode);
5456 inode->i_ctime = inode->i_mtime;
5458 down_write(&EXT4_I(inode)->i_data_sem);
5459 EXT4_I(inode)->i_disksize = attr->ia_size;
5460 rc = ext4_mark_inode_dirty(handle, inode);
5464 * We have to update i_size under i_data_sem together
5465 * with i_disksize to avoid races with writeback code
5466 * running ext4_wb_update_i_disksize().
5469 i_size_write(inode, attr->ia_size);
5470 up_write(&EXT4_I(inode)->i_data_sem);
5471 ext4_journal_stop(handle);
5474 ext4_orphan_del(NULL, inode);
5479 pagecache_isize_extended(inode, oldsize, inode->i_size);
5482 * Blocks are going to be removed from the inode. Wait
5483 * for dio in flight. Temporarily disable
5484 * dioread_nolock to prevent livelock.
5487 if (!ext4_should_journal_data(inode)) {
5488 ext4_inode_block_unlocked_dio(inode);
5489 inode_dio_wait(inode);
5490 ext4_inode_resume_unlocked_dio(inode);
5492 ext4_wait_for_tail_page_commit(inode);
5494 down_write(&EXT4_I(inode)->i_mmap_sem);
5496 * Truncate pagecache after we've waited for commit
5497 * in data=journal mode to make pages freeable.
5499 truncate_pagecache(inode, inode->i_size);
5501 rc = ext4_truncate(inode);
5505 up_write(&EXT4_I(inode)->i_mmap_sem);
5509 setattr_copy(inode, attr);
5510 mark_inode_dirty(inode);
5514 * If the call to ext4_truncate failed to get a transaction handle at
5515 * all, we need to clean up the in-core orphan list manually.
5517 if (orphan && inode->i_nlink)
5518 ext4_orphan_del(NULL, inode);
5520 if (!error && (ia_valid & ATTR_MODE))
5521 rc = posix_acl_chmod(inode, inode->i_mode);
5524 ext4_std_error(inode->i_sb, error);
5530 int ext4_getattr(const struct path *path, struct kstat *stat,
5531 u32 request_mask, unsigned int query_flags)
5533 struct inode *inode = d_inode(path->dentry);
5534 struct ext4_inode *raw_inode;
5535 struct ext4_inode_info *ei = EXT4_I(inode);
5538 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5539 stat->result_mask |= STATX_BTIME;
5540 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5541 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5544 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5545 if (flags & EXT4_APPEND_FL)
5546 stat->attributes |= STATX_ATTR_APPEND;
5547 if (flags & EXT4_COMPR_FL)
5548 stat->attributes |= STATX_ATTR_COMPRESSED;
5549 if (flags & EXT4_ENCRYPT_FL)
5550 stat->attributes |= STATX_ATTR_ENCRYPTED;
5551 if (flags & EXT4_IMMUTABLE_FL)
5552 stat->attributes |= STATX_ATTR_IMMUTABLE;
5553 if (flags & EXT4_NODUMP_FL)
5554 stat->attributes |= STATX_ATTR_NODUMP;
5556 stat->attributes_mask |= (STATX_ATTR_APPEND |
5557 STATX_ATTR_COMPRESSED |
5558 STATX_ATTR_ENCRYPTED |
5559 STATX_ATTR_IMMUTABLE |
5562 generic_fillattr(inode, stat);
5566 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5567 u32 request_mask, unsigned int query_flags)
5569 struct inode *inode = d_inode(path->dentry);
5570 u64 delalloc_blocks;
5572 ext4_getattr(path, stat, request_mask, query_flags);
5575 * If there is inline data in the inode, the inode will normally not
5576 * have data blocks allocated (it may have an external xattr block).
5577 * Report at least one sector for such files, so tools like tar, rsync,
5578 * others don't incorrectly think the file is completely sparse.
5580 if (unlikely(ext4_has_inline_data(inode)))
5581 stat->blocks += (stat->size + 511) >> 9;
5584 * We can't update i_blocks if the block allocation is delayed
5585 * otherwise in the case of system crash before the real block
5586 * allocation is done, we will have i_blocks inconsistent with
5587 * on-disk file blocks.
5588 * We always keep i_blocks updated together with real
5589 * allocation. But to not confuse with user, stat
5590 * will return the blocks that include the delayed allocation
5591 * blocks for this file.
5593 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5594 EXT4_I(inode)->i_reserved_data_blocks);
5595 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5599 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5602 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5603 return ext4_ind_trans_blocks(inode, lblocks);
5604 return ext4_ext_index_trans_blocks(inode, pextents);
5608 * Account for index blocks, block groups bitmaps and block group
5609 * descriptor blocks if modify datablocks and index blocks
5610 * worse case, the indexs blocks spread over different block groups
5612 * If datablocks are discontiguous, they are possible to spread over
5613 * different block groups too. If they are contiguous, with flexbg,
5614 * they could still across block group boundary.
5616 * Also account for superblock, inode, quota and xattr blocks
5618 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5621 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5627 * How many index blocks need to touch to map @lblocks logical blocks
5628 * to @pextents physical extents?
5630 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5635 * Now let's see how many group bitmaps and group descriptors need
5638 groups = idxblocks + pextents;
5640 if (groups > ngroups)
5642 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5643 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5645 /* bitmaps and block group descriptor blocks */
5646 ret += groups + gdpblocks;
5648 /* Blocks for super block, inode, quota and xattr blocks */
5649 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5655 * Calculate the total number of credits to reserve to fit
5656 * the modification of a single pages into a single transaction,
5657 * which may include multiple chunks of block allocations.
5659 * This could be called via ext4_write_begin()
5661 * We need to consider the worse case, when
5662 * one new block per extent.
5664 int ext4_writepage_trans_blocks(struct inode *inode)
5666 int bpp = ext4_journal_blocks_per_page(inode);
5669 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5671 /* Account for data blocks for journalled mode */
5672 if (ext4_should_journal_data(inode))
5678 * Calculate the journal credits for a chunk of data modification.
5680 * This is called from DIO, fallocate or whoever calling
5681 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5683 * journal buffers for data blocks are not included here, as DIO
5684 * and fallocate do no need to journal data buffers.
5686 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5688 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5692 * The caller must have previously called ext4_reserve_inode_write().
5693 * Give this, we know that the caller already has write access to iloc->bh.
5695 int ext4_mark_iloc_dirty(handle_t *handle,
5696 struct inode *inode, struct ext4_iloc *iloc)
5700 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5703 if (IS_I_VERSION(inode))
5704 inode_inc_iversion(inode);
5706 /* the do_update_inode consumes one bh->b_count */
5709 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5710 err = ext4_do_update_inode(handle, inode, iloc);
5716 * On success, We end up with an outstanding reference count against
5717 * iloc->bh. This _must_ be cleaned up later.
5721 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5722 struct ext4_iloc *iloc)
5726 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5729 err = ext4_get_inode_loc(inode, iloc);
5731 BUFFER_TRACE(iloc->bh, "get_write_access");
5732 err = ext4_journal_get_write_access(handle, iloc->bh);
5738 ext4_std_error(inode->i_sb, err);
5742 static int __ext4_expand_extra_isize(struct inode *inode,
5743 unsigned int new_extra_isize,
5744 struct ext4_iloc *iloc,
5745 handle_t *handle, int *no_expand)
5747 struct ext4_inode *raw_inode;
5748 struct ext4_xattr_ibody_header *header;
5751 raw_inode = ext4_raw_inode(iloc);
5753 header = IHDR(inode, raw_inode);
5755 /* No extended attributes present */
5756 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5757 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5758 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5759 EXT4_I(inode)->i_extra_isize, 0,
5760 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5761 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5765 /* try to expand with EAs present */
5766 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5770 * Inode size expansion failed; don't try again
5779 * Expand an inode by new_extra_isize bytes.
5780 * Returns 0 on success or negative error number on failure.
5782 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5783 unsigned int new_extra_isize,
5784 struct ext4_iloc iloc,
5790 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5794 * In nojournal mode, we can immediately attempt to expand
5795 * the inode. When journaled, we first need to obtain extra
5796 * buffer credits since we may write into the EA block
5797 * with this same handle. If journal_extend fails, then it will
5798 * only result in a minor loss of functionality for that inode.
5799 * If this is felt to be critical, then e2fsck should be run to
5800 * force a large enough s_min_extra_isize.
5802 if (ext4_handle_valid(handle) &&
5803 jbd2_journal_extend(handle,
5804 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0)
5807 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5810 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5811 handle, &no_expand);
5812 ext4_write_unlock_xattr(inode, &no_expand);
5817 int ext4_expand_extra_isize(struct inode *inode,
5818 unsigned int new_extra_isize,
5819 struct ext4_iloc *iloc)
5825 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5830 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5831 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5832 if (IS_ERR(handle)) {
5833 error = PTR_ERR(handle);
5838 ext4_write_lock_xattr(inode, &no_expand);
5840 BUFFER_TRACE(iloc.bh, "get_write_access");
5841 error = ext4_journal_get_write_access(handle, iloc->bh);
5847 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5848 handle, &no_expand);
5850 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5854 ext4_write_unlock_xattr(inode, &no_expand);
5856 ext4_journal_stop(handle);
5861 * What we do here is to mark the in-core inode as clean with respect to inode
5862 * dirtiness (it may still be data-dirty).
5863 * This means that the in-core inode may be reaped by prune_icache
5864 * without having to perform any I/O. This is a very good thing,
5865 * because *any* task may call prune_icache - even ones which
5866 * have a transaction open against a different journal.
5868 * Is this cheating? Not really. Sure, we haven't written the
5869 * inode out, but prune_icache isn't a user-visible syncing function.
5870 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5871 * we start and wait on commits.
5873 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5875 struct ext4_iloc iloc;
5876 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5880 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5881 err = ext4_reserve_inode_write(handle, inode, &iloc);
5885 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5886 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5889 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5893 * ext4_dirty_inode() is called from __mark_inode_dirty()
5895 * We're really interested in the case where a file is being extended.
5896 * i_size has been changed by generic_commit_write() and we thus need
5897 * to include the updated inode in the current transaction.
5899 * Also, dquot_alloc_block() will always dirty the inode when blocks
5900 * are allocated to the file.
5902 * If the inode is marked synchronous, we don't honour that here - doing
5903 * so would cause a commit on atime updates, which we don't bother doing.
5904 * We handle synchronous inodes at the highest possible level.
5906 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5907 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5908 * to copy into the on-disk inode structure are the timestamp files.
5910 void ext4_dirty_inode(struct inode *inode, int flags)
5914 if (flags == I_DIRTY_TIME)
5916 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5920 ext4_mark_inode_dirty(handle, inode);
5922 ext4_journal_stop(handle);
5929 * Bind an inode's backing buffer_head into this transaction, to prevent
5930 * it from being flushed to disk early. Unlike
5931 * ext4_reserve_inode_write, this leaves behind no bh reference and
5932 * returns no iloc structure, so the caller needs to repeat the iloc
5933 * lookup to mark the inode dirty later.
5935 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5937 struct ext4_iloc iloc;
5941 err = ext4_get_inode_loc(inode, &iloc);
5943 BUFFER_TRACE(iloc.bh, "get_write_access");
5944 err = jbd2_journal_get_write_access(handle, iloc.bh);
5946 err = ext4_handle_dirty_metadata(handle,
5952 ext4_std_error(inode->i_sb, err);
5957 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5962 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5965 * We have to be very careful here: changing a data block's
5966 * journaling status dynamically is dangerous. If we write a
5967 * data block to the journal, change the status and then delete
5968 * that block, we risk forgetting to revoke the old log record
5969 * from the journal and so a subsequent replay can corrupt data.
5970 * So, first we make sure that the journal is empty and that
5971 * nobody is changing anything.
5974 journal = EXT4_JOURNAL(inode);
5977 if (is_journal_aborted(journal))
5980 /* Wait for all existing dio workers */
5981 ext4_inode_block_unlocked_dio(inode);
5982 inode_dio_wait(inode);
5985 * Before flushing the journal and switching inode's aops, we have
5986 * to flush all dirty data the inode has. There can be outstanding
5987 * delayed allocations, there can be unwritten extents created by
5988 * fallocate or buffered writes in dioread_nolock mode covered by
5989 * dirty data which can be converted only after flushing the dirty
5990 * data (and journalled aops don't know how to handle these cases).
5993 down_write(&EXT4_I(inode)->i_mmap_sem);
5994 err = filemap_write_and_wait(inode->i_mapping);
5996 up_write(&EXT4_I(inode)->i_mmap_sem);
5997 ext4_inode_resume_unlocked_dio(inode);
6002 percpu_down_write(&sbi->s_journal_flag_rwsem);
6003 jbd2_journal_lock_updates(journal);
6006 * OK, there are no updates running now, and all cached data is
6007 * synced to disk. We are now in a completely consistent state
6008 * which doesn't have anything in the journal, and we know that
6009 * no filesystem updates are running, so it is safe to modify
6010 * the inode's in-core data-journaling state flag now.
6014 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6016 err = jbd2_journal_flush(journal);
6018 jbd2_journal_unlock_updates(journal);
6019 percpu_up_write(&sbi->s_journal_flag_rwsem);
6020 ext4_inode_resume_unlocked_dio(inode);
6023 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6025 ext4_set_aops(inode);
6027 jbd2_journal_unlock_updates(journal);
6028 percpu_up_write(&sbi->s_journal_flag_rwsem);
6031 up_write(&EXT4_I(inode)->i_mmap_sem);
6032 ext4_inode_resume_unlocked_dio(inode);
6034 /* Finally we can mark the inode as dirty. */
6036 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6038 return PTR_ERR(handle);
6040 err = ext4_mark_inode_dirty(handle, inode);
6041 ext4_handle_sync(handle);
6042 ext4_journal_stop(handle);
6043 ext4_std_error(inode->i_sb, err);
6048 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6050 return !buffer_mapped(bh);
6053 int ext4_page_mkwrite(struct vm_fault *vmf)
6055 struct vm_area_struct *vma = vmf->vma;
6056 struct page *page = vmf->page;
6060 struct file *file = vma->vm_file;
6061 struct inode *inode = file_inode(file);
6062 struct address_space *mapping = inode->i_mapping;
6064 get_block_t *get_block;
6067 sb_start_pagefault(inode->i_sb);
6068 file_update_time(vma->vm_file);
6070 down_read(&EXT4_I(inode)->i_mmap_sem);
6072 ret = ext4_convert_inline_data(inode);
6076 /* Delalloc case is easy... */
6077 if (test_opt(inode->i_sb, DELALLOC) &&
6078 !ext4_should_journal_data(inode) &&
6079 !ext4_nonda_switch(inode->i_sb)) {
6081 ret = block_page_mkwrite(vma, vmf,
6082 ext4_da_get_block_prep);
6083 } while (ret == -ENOSPC &&
6084 ext4_should_retry_alloc(inode->i_sb, &retries));
6089 size = i_size_read(inode);
6090 /* Page got truncated from under us? */
6091 if (page->mapping != mapping || page_offset(page) > size) {
6093 ret = VM_FAULT_NOPAGE;
6097 if (page->index == size >> PAGE_SHIFT)
6098 len = size & ~PAGE_MASK;
6102 * Return if we have all the buffers mapped. This avoids the need to do
6103 * journal_start/journal_stop which can block and take a long time
6105 if (page_has_buffers(page)) {
6106 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6108 ext4_bh_unmapped)) {
6109 /* Wait so that we don't change page under IO */
6110 wait_for_stable_page(page);
6111 ret = VM_FAULT_LOCKED;
6116 /* OK, we need to fill the hole... */
6117 if (ext4_should_dioread_nolock(inode))
6118 get_block = ext4_get_block_unwritten;
6120 get_block = ext4_get_block;
6122 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6123 ext4_writepage_trans_blocks(inode));
6124 if (IS_ERR(handle)) {
6125 ret = VM_FAULT_SIGBUS;
6128 ret = block_page_mkwrite(vma, vmf, get_block);
6129 if (!ret && ext4_should_journal_data(inode)) {
6130 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
6131 PAGE_SIZE, NULL, do_journal_get_write_access)) {
6133 ret = VM_FAULT_SIGBUS;
6134 ext4_journal_stop(handle);
6137 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6139 ext4_journal_stop(handle);
6140 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6143 ret = block_page_mkwrite_return(ret);
6145 up_read(&EXT4_I(inode)->i_mmap_sem);
6146 sb_end_pagefault(inode->i_sb);
6150 int ext4_filemap_fault(struct vm_fault *vmf)
6152 struct inode *inode = file_inode(vmf->vma->vm_file);
6155 down_read(&EXT4_I(inode)->i_mmap_sem);
6156 err = filemap_fault(vmf);
6157 up_read(&EXT4_I(inode)->i_mmap_sem);