2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/dax.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/bitops.h>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51 struct ext4_inode_info *ei)
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 csum_lo = le16_to_cpu(raw->i_checksum_lo);
59 raw->i_checksum_lo = 0;
60 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
61 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
62 csum_hi = le16_to_cpu(raw->i_checksum_hi);
63 raw->i_checksum_hi = 0;
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
67 EXT4_INODE_SIZE(inode->i_sb));
69 raw->i_checksum_lo = cpu_to_le16(csum_lo);
70 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
71 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
72 raw->i_checksum_hi = cpu_to_le16(csum_hi);
77 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
78 struct ext4_inode_info *ei)
80 __u32 provided, calculated;
82 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
83 cpu_to_le32(EXT4_OS_LINUX) ||
84 !ext4_has_metadata_csum(inode->i_sb))
87 provided = le16_to_cpu(raw->i_checksum_lo);
88 calculated = ext4_inode_csum(inode, raw, ei);
89 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
90 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
91 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
95 return provided == calculated;
98 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
99 struct ext4_inode_info *ei)
103 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
104 cpu_to_le32(EXT4_OS_LINUX) ||
105 !ext4_has_metadata_csum(inode->i_sb))
108 csum = ext4_inode_csum(inode, raw, ei);
109 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
110 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
111 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
112 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
115 static inline int ext4_begin_ordered_truncate(struct inode *inode,
118 trace_ext4_begin_ordered_truncate(inode, new_size);
120 * If jinode is zero, then we never opened the file for
121 * writing, so there's no need to call
122 * jbd2_journal_begin_ordered_truncate() since there's no
123 * outstanding writes we need to flush.
125 if (!EXT4_I(inode)->jinode)
127 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
128 EXT4_I(inode)->jinode,
132 static void ext4_invalidatepage(struct page *page, unsigned int offset,
133 unsigned int length);
134 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
135 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
136 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
140 * Test whether an inode is a fast symlink.
142 int ext4_inode_is_fast_symlink(struct inode *inode)
144 int ea_blocks = EXT4_I(inode)->i_file_acl ?
145 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
147 if (ext4_has_inline_data(inode))
150 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
154 * Restart the transaction associated with *handle. This does a commit,
155 * so before we call here everything must be consistently dirtied against
158 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
164 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
165 * moment, get_block can be called only for blocks inside i_size since
166 * page cache has been already dropped and writes are blocked by
167 * i_mutex. So we can safely drop the i_data_sem here.
169 BUG_ON(EXT4_JOURNAL(inode) == NULL);
170 jbd_debug(2, "restarting handle %p\n", handle);
171 up_write(&EXT4_I(inode)->i_data_sem);
172 ret = ext4_journal_restart(handle, nblocks);
173 down_write(&EXT4_I(inode)->i_data_sem);
174 ext4_discard_preallocations(inode);
180 * Called at the last iput() if i_nlink is zero.
182 void ext4_evict_inode(struct inode *inode)
187 trace_ext4_evict_inode(inode);
189 if (inode->i_nlink) {
191 * When journalling data dirty buffers are tracked only in the
192 * journal. So although mm thinks everything is clean and
193 * ready for reaping the inode might still have some pages to
194 * write in the running transaction or waiting to be
195 * checkpointed. Thus calling jbd2_journal_invalidatepage()
196 * (via truncate_inode_pages()) to discard these buffers can
197 * cause data loss. Also even if we did not discard these
198 * buffers, we would have no way to find them after the inode
199 * is reaped and thus user could see stale data if he tries to
200 * read them before the transaction is checkpointed. So be
201 * careful and force everything to disk here... We use
202 * ei->i_datasync_tid to store the newest transaction
203 * containing inode's data.
205 * Note that directories do not have this problem because they
206 * don't use page cache.
208 if (ext4_should_journal_data(inode) &&
209 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
210 inode->i_ino != EXT4_JOURNAL_INO) {
211 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
212 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
214 jbd2_complete_transaction(journal, commit_tid);
215 filemap_write_and_wait(&inode->i_data);
217 truncate_inode_pages_final(&inode->i_data);
219 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
223 if (is_bad_inode(inode))
225 dquot_initialize(inode);
227 if (ext4_should_order_data(inode))
228 ext4_begin_ordered_truncate(inode, 0);
229 truncate_inode_pages_final(&inode->i_data);
231 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
234 * Protect us against freezing - iput() caller didn't have to have any
235 * protection against it
237 sb_start_intwrite(inode->i_sb);
238 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
239 ext4_blocks_for_truncate(inode)+3);
240 if (IS_ERR(handle)) {
241 ext4_std_error(inode->i_sb, PTR_ERR(handle));
243 * If we're going to skip the normal cleanup, we still need to
244 * make sure that the in-core orphan linked list is properly
247 ext4_orphan_del(NULL, inode);
248 sb_end_intwrite(inode->i_sb);
253 ext4_handle_sync(handle);
255 err = ext4_mark_inode_dirty(handle, inode);
257 ext4_warning(inode->i_sb,
258 "couldn't mark inode dirty (err %d)", err);
262 ext4_truncate(inode);
265 * ext4_ext_truncate() doesn't reserve any slop when it
266 * restarts journal transactions; therefore there may not be
267 * enough credits left in the handle to remove the inode from
268 * the orphan list and set the dtime field.
270 if (!ext4_handle_has_enough_credits(handle, 3)) {
271 err = ext4_journal_extend(handle, 3);
273 err = ext4_journal_restart(handle, 3);
275 ext4_warning(inode->i_sb,
276 "couldn't extend journal (err %d)", err);
278 ext4_journal_stop(handle);
279 ext4_orphan_del(NULL, inode);
280 sb_end_intwrite(inode->i_sb);
286 * Kill off the orphan record which ext4_truncate created.
287 * AKPM: I think this can be inside the above `if'.
288 * Note that ext4_orphan_del() has to be able to cope with the
289 * deletion of a non-existent orphan - this is because we don't
290 * know if ext4_truncate() actually created an orphan record.
291 * (Well, we could do this if we need to, but heck - it works)
293 ext4_orphan_del(handle, inode);
294 EXT4_I(inode)->i_dtime = get_seconds();
297 * One subtle ordering requirement: if anything has gone wrong
298 * (transaction abort, IO errors, whatever), then we can still
299 * do these next steps (the fs will already have been marked as
300 * having errors), but we can't free the inode if the mark_dirty
303 if (ext4_mark_inode_dirty(handle, inode))
304 /* If that failed, just do the required in-core inode clear. */
305 ext4_clear_inode(inode);
307 ext4_free_inode(handle, inode);
308 ext4_journal_stop(handle);
309 sb_end_intwrite(inode->i_sb);
312 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
316 qsize_t *ext4_get_reserved_space(struct inode *inode)
318 return &EXT4_I(inode)->i_reserved_quota;
323 * Called with i_data_sem down, which is important since we can call
324 * ext4_discard_preallocations() from here.
326 void ext4_da_update_reserve_space(struct inode *inode,
327 int used, int quota_claim)
329 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
330 struct ext4_inode_info *ei = EXT4_I(inode);
332 spin_lock(&ei->i_block_reservation_lock);
333 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
334 if (unlikely(used > ei->i_reserved_data_blocks)) {
335 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
336 "with only %d reserved data blocks",
337 __func__, inode->i_ino, used,
338 ei->i_reserved_data_blocks);
340 used = ei->i_reserved_data_blocks;
343 /* Update per-inode reservations */
344 ei->i_reserved_data_blocks -= used;
345 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
347 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
349 /* Update quota subsystem for data blocks */
351 dquot_claim_block(inode, EXT4_C2B(sbi, used));
354 * We did fallocate with an offset that is already delayed
355 * allocated. So on delayed allocated writeback we should
356 * not re-claim the quota for fallocated blocks.
358 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
362 * If we have done all the pending block allocations and if
363 * there aren't any writers on the inode, we can discard the
364 * inode's preallocations.
366 if ((ei->i_reserved_data_blocks == 0) &&
367 (atomic_read(&inode->i_writecount) == 0))
368 ext4_discard_preallocations(inode);
371 static int __check_block_validity(struct inode *inode, const char *func,
373 struct ext4_map_blocks *map)
375 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
377 ext4_error_inode(inode, func, line, map->m_pblk,
378 "lblock %lu mapped to illegal pblock "
379 "(length %d)", (unsigned long) map->m_lblk,
381 return -EFSCORRUPTED;
386 #define check_block_validity(inode, map) \
387 __check_block_validity((inode), __func__, __LINE__, (map))
389 #ifdef ES_AGGRESSIVE_TEST
390 static void ext4_map_blocks_es_recheck(handle_t *handle,
392 struct ext4_map_blocks *es_map,
393 struct ext4_map_blocks *map,
400 * There is a race window that the result is not the same.
401 * e.g. xfstests #223 when dioread_nolock enables. The reason
402 * is that we lookup a block mapping in extent status tree with
403 * out taking i_data_sem. So at the time the unwritten extent
404 * could be converted.
406 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
407 down_read(&EXT4_I(inode)->i_data_sem);
408 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
409 retval = ext4_ext_map_blocks(handle, inode, map, flags &
410 EXT4_GET_BLOCKS_KEEP_SIZE);
412 retval = ext4_ind_map_blocks(handle, inode, map, flags &
413 EXT4_GET_BLOCKS_KEEP_SIZE);
415 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
416 up_read((&EXT4_I(inode)->i_data_sem));
419 * We don't check m_len because extent will be collpased in status
420 * tree. So the m_len might not equal.
422 if (es_map->m_lblk != map->m_lblk ||
423 es_map->m_flags != map->m_flags ||
424 es_map->m_pblk != map->m_pblk) {
425 printk("ES cache assertion failed for inode: %lu "
426 "es_cached ex [%d/%d/%llu/%x] != "
427 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
428 inode->i_ino, es_map->m_lblk, es_map->m_len,
429 es_map->m_pblk, es_map->m_flags, map->m_lblk,
430 map->m_len, map->m_pblk, map->m_flags,
434 #endif /* ES_AGGRESSIVE_TEST */
437 * The ext4_map_blocks() function tries to look up the requested blocks,
438 * and returns if the blocks are already mapped.
440 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
441 * and store the allocated blocks in the result buffer head and mark it
444 * If file type is extents based, it will call ext4_ext_map_blocks(),
445 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
448 * On success, it returns the number of blocks being mapped or allocated.
449 * if create==0 and the blocks are pre-allocated and unwritten block,
450 * the result buffer head is unmapped. If the create ==1, it will make sure
451 * the buffer head is mapped.
453 * It returns 0 if plain look up failed (blocks have not been allocated), in
454 * that case, buffer head is unmapped
456 * It returns the error in case of allocation failure.
458 int ext4_map_blocks(handle_t *handle, struct inode *inode,
459 struct ext4_map_blocks *map, int flags)
461 struct extent_status es;
464 #ifdef ES_AGGRESSIVE_TEST
465 struct ext4_map_blocks orig_map;
467 memcpy(&orig_map, map, sizeof(*map));
471 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
472 "logical block %lu\n", inode->i_ino, flags, map->m_len,
473 (unsigned long) map->m_lblk);
476 * ext4_map_blocks returns an int, and m_len is an unsigned int
478 if (unlikely(map->m_len > INT_MAX))
479 map->m_len = INT_MAX;
481 /* We can handle the block number less than EXT_MAX_BLOCKS */
482 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
483 return -EFSCORRUPTED;
485 /* Lookup extent status tree firstly */
486 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
487 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
488 map->m_pblk = ext4_es_pblock(&es) +
489 map->m_lblk - es.es_lblk;
490 map->m_flags |= ext4_es_is_written(&es) ?
491 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
492 retval = es.es_len - (map->m_lblk - es.es_lblk);
493 if (retval > map->m_len)
496 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
501 #ifdef ES_AGGRESSIVE_TEST
502 ext4_map_blocks_es_recheck(handle, inode, map,
509 * Try to see if we can get the block without requesting a new
512 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
513 down_read(&EXT4_I(inode)->i_data_sem);
514 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
515 retval = ext4_ext_map_blocks(handle, inode, map, flags &
516 EXT4_GET_BLOCKS_KEEP_SIZE);
518 retval = ext4_ind_map_blocks(handle, inode, map, flags &
519 EXT4_GET_BLOCKS_KEEP_SIZE);
524 if (unlikely(retval != map->m_len)) {
525 ext4_warning(inode->i_sb,
526 "ES len assertion failed for inode "
527 "%lu: retval %d != map->m_len %d",
528 inode->i_ino, retval, map->m_len);
532 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
533 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
534 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
535 !(status & EXTENT_STATUS_WRITTEN) &&
536 ext4_find_delalloc_range(inode, map->m_lblk,
537 map->m_lblk + map->m_len - 1))
538 status |= EXTENT_STATUS_DELAYED;
539 ret = ext4_es_insert_extent(inode, map->m_lblk,
540 map->m_len, map->m_pblk, status);
544 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
545 up_read((&EXT4_I(inode)->i_data_sem));
548 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
549 ret = check_block_validity(inode, map);
554 /* If it is only a block(s) look up */
555 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
559 * Returns if the blocks have already allocated
561 * Note that if blocks have been preallocated
562 * ext4_ext_get_block() returns the create = 0
563 * with buffer head unmapped.
565 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
567 * If we need to convert extent to unwritten
568 * we continue and do the actual work in
569 * ext4_ext_map_blocks()
571 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
575 * Here we clear m_flags because after allocating an new extent,
576 * it will be set again.
578 map->m_flags &= ~EXT4_MAP_FLAGS;
581 * New blocks allocate and/or writing to unwritten extent
582 * will possibly result in updating i_data, so we take
583 * the write lock of i_data_sem, and call get_block()
584 * with create == 1 flag.
586 down_write(&EXT4_I(inode)->i_data_sem);
589 * We need to check for EXT4 here because migrate
590 * could have changed the inode type in between
592 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
593 retval = ext4_ext_map_blocks(handle, inode, map, flags);
595 retval = ext4_ind_map_blocks(handle, inode, map, flags);
597 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
599 * We allocated new blocks which will result in
600 * i_data's format changing. Force the migrate
601 * to fail by clearing migrate flags
603 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
607 * Update reserved blocks/metadata blocks after successful
608 * block allocation which had been deferred till now. We don't
609 * support fallocate for non extent files. So we can update
610 * reserve space here.
613 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
614 ext4_da_update_reserve_space(inode, retval, 1);
620 if (unlikely(retval != map->m_len)) {
621 ext4_warning(inode->i_sb,
622 "ES len assertion failed for inode "
623 "%lu: retval %d != map->m_len %d",
624 inode->i_ino, retval, map->m_len);
629 * If the extent has been zeroed out, we don't need to update
630 * extent status tree.
632 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
633 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
634 if (ext4_es_is_written(&es))
637 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
638 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
639 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
640 !(status & EXTENT_STATUS_WRITTEN) &&
641 ext4_find_delalloc_range(inode, map->m_lblk,
642 map->m_lblk + map->m_len - 1))
643 status |= EXTENT_STATUS_DELAYED;
644 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
645 map->m_pblk, status);
651 up_write((&EXT4_I(inode)->i_data_sem));
652 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
653 ret = check_block_validity(inode, map);
661 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
662 * we have to be careful as someone else may be manipulating b_state as well.
664 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
666 unsigned long old_state;
667 unsigned long new_state;
669 flags &= EXT4_MAP_FLAGS;
671 /* Dummy buffer_head? Set non-atomically. */
673 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
677 * Someone else may be modifying b_state. Be careful! This is ugly but
678 * once we get rid of using bh as a container for mapping information
679 * to pass to / from get_block functions, this can go away.
682 old_state = READ_ONCE(bh->b_state);
683 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
685 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
688 /* Maximum number of blocks we map for direct IO at once. */
689 #define DIO_MAX_BLOCKS 4096
691 static int _ext4_get_block(struct inode *inode, sector_t iblock,
692 struct buffer_head *bh, int flags)
694 handle_t *handle = ext4_journal_current_handle();
695 struct ext4_map_blocks map;
696 int ret = 0, started = 0;
699 if (ext4_has_inline_data(inode))
703 map.m_len = bh->b_size >> inode->i_blkbits;
705 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
706 /* Direct IO write... */
707 if (map.m_len > DIO_MAX_BLOCKS)
708 map.m_len = DIO_MAX_BLOCKS;
709 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
710 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
712 if (IS_ERR(handle)) {
713 ret = PTR_ERR(handle);
719 ret = ext4_map_blocks(handle, inode, &map, flags);
721 ext4_io_end_t *io_end = ext4_inode_aio(inode);
723 map_bh(bh, inode->i_sb, map.m_pblk);
724 ext4_update_bh_state(bh, map.m_flags);
725 if (IS_DAX(inode) && buffer_unwritten(bh)) {
727 * dgc: I suspect unwritten conversion on ext4+DAX is
728 * fundamentally broken here when there are concurrent
729 * read/write in progress on this inode.
731 WARN_ON_ONCE(io_end);
732 bh->b_assoc_map = inode->i_mapping;
733 bh->b_private = (void *)(unsigned long)iblock;
735 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
736 set_buffer_defer_completion(bh);
737 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
741 ext4_journal_stop(handle);
745 int ext4_get_block(struct inode *inode, sector_t iblock,
746 struct buffer_head *bh, int create)
748 return _ext4_get_block(inode, iblock, bh,
749 create ? EXT4_GET_BLOCKS_CREATE : 0);
753 * `handle' can be NULL if create is zero
755 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
756 ext4_lblk_t block, int map_flags)
758 struct ext4_map_blocks map;
759 struct buffer_head *bh;
760 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
763 J_ASSERT(handle != NULL || create == 0);
767 err = ext4_map_blocks(handle, inode, &map, map_flags);
770 return create ? ERR_PTR(-ENOSPC) : NULL;
774 bh = sb_getblk(inode->i_sb, map.m_pblk);
776 return ERR_PTR(-ENOMEM);
777 if (map.m_flags & EXT4_MAP_NEW) {
778 J_ASSERT(create != 0);
779 J_ASSERT(handle != NULL);
782 * Now that we do not always journal data, we should
783 * keep in mind whether this should always journal the
784 * new buffer as metadata. For now, regular file
785 * writes use ext4_get_block instead, so it's not a
789 BUFFER_TRACE(bh, "call get_create_access");
790 err = ext4_journal_get_create_access(handle, bh);
795 if (!buffer_uptodate(bh)) {
796 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
797 set_buffer_uptodate(bh);
800 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
801 err = ext4_handle_dirty_metadata(handle, inode, bh);
805 BUFFER_TRACE(bh, "not a new buffer");
812 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
813 ext4_lblk_t block, int map_flags)
815 struct buffer_head *bh;
817 bh = ext4_getblk(handle, inode, block, map_flags);
820 if (!bh || buffer_uptodate(bh))
822 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
824 if (buffer_uptodate(bh))
827 return ERR_PTR(-EIO);
830 int ext4_walk_page_buffers(handle_t *handle,
831 struct buffer_head *head,
835 int (*fn)(handle_t *handle,
836 struct buffer_head *bh))
838 struct buffer_head *bh;
839 unsigned block_start, block_end;
840 unsigned blocksize = head->b_size;
842 struct buffer_head *next;
844 for (bh = head, block_start = 0;
845 ret == 0 && (bh != head || !block_start);
846 block_start = block_end, bh = next) {
847 next = bh->b_this_page;
848 block_end = block_start + blocksize;
849 if (block_end <= from || block_start >= to) {
850 if (partial && !buffer_uptodate(bh))
854 err = (*fn)(handle, bh);
862 * To preserve ordering, it is essential that the hole instantiation and
863 * the data write be encapsulated in a single transaction. We cannot
864 * close off a transaction and start a new one between the ext4_get_block()
865 * and the commit_write(). So doing the jbd2_journal_start at the start of
866 * prepare_write() is the right place.
868 * Also, this function can nest inside ext4_writepage(). In that case, we
869 * *know* that ext4_writepage() has generated enough buffer credits to do the
870 * whole page. So we won't block on the journal in that case, which is good,
871 * because the caller may be PF_MEMALLOC.
873 * By accident, ext4 can be reentered when a transaction is open via
874 * quota file writes. If we were to commit the transaction while thus
875 * reentered, there can be a deadlock - we would be holding a quota
876 * lock, and the commit would never complete if another thread had a
877 * transaction open and was blocking on the quota lock - a ranking
880 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
881 * will _not_ run commit under these circumstances because handle->h_ref
882 * is elevated. We'll still have enough credits for the tiny quotafile
885 int do_journal_get_write_access(handle_t *handle,
886 struct buffer_head *bh)
888 int dirty = buffer_dirty(bh);
891 if (!buffer_mapped(bh) || buffer_freed(bh))
894 * __block_write_begin() could have dirtied some buffers. Clean
895 * the dirty bit as jbd2_journal_get_write_access() could complain
896 * otherwise about fs integrity issues. Setting of the dirty bit
897 * by __block_write_begin() isn't a real problem here as we clear
898 * the bit before releasing a page lock and thus writeback cannot
899 * ever write the buffer.
902 clear_buffer_dirty(bh);
903 BUFFER_TRACE(bh, "get write access");
904 ret = ext4_journal_get_write_access(handle, bh);
906 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
910 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
911 struct buffer_head *bh_result, int create);
913 #ifdef CONFIG_EXT4_FS_ENCRYPTION
914 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
915 get_block_t *get_block)
917 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
918 unsigned to = from + len;
919 struct inode *inode = page->mapping->host;
920 unsigned block_start, block_end;
923 unsigned blocksize = inode->i_sb->s_blocksize;
925 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
926 bool decrypt = false;
928 BUG_ON(!PageLocked(page));
929 BUG_ON(from > PAGE_CACHE_SIZE);
930 BUG_ON(to > PAGE_CACHE_SIZE);
933 if (!page_has_buffers(page))
934 create_empty_buffers(page, blocksize, 0);
935 head = page_buffers(page);
936 bbits = ilog2(blocksize);
937 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
939 for (bh = head, block_start = 0; bh != head || !block_start;
940 block++, block_start = block_end, bh = bh->b_this_page) {
941 block_end = block_start + blocksize;
942 if (block_end <= from || block_start >= to) {
943 if (PageUptodate(page)) {
944 if (!buffer_uptodate(bh))
945 set_buffer_uptodate(bh);
950 clear_buffer_new(bh);
951 if (!buffer_mapped(bh)) {
952 WARN_ON(bh->b_size != blocksize);
953 err = get_block(inode, block, bh, 1);
956 if (buffer_new(bh)) {
957 unmap_underlying_metadata(bh->b_bdev,
959 if (PageUptodate(page)) {
960 clear_buffer_new(bh);
961 set_buffer_uptodate(bh);
962 mark_buffer_dirty(bh);
965 if (block_end > to || block_start < from)
966 zero_user_segments(page, to, block_end,
971 if (PageUptodate(page)) {
972 if (!buffer_uptodate(bh))
973 set_buffer_uptodate(bh);
976 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
977 !buffer_unwritten(bh) &&
978 (block_start < from || block_end > to)) {
979 ll_rw_block(READ, 1, &bh);
981 decrypt = ext4_encrypted_inode(inode) &&
982 S_ISREG(inode->i_mode);
986 * If we issued read requests, let them complete.
988 while (wait_bh > wait) {
989 wait_on_buffer(*--wait_bh);
990 if (!buffer_uptodate(*wait_bh))
994 page_zero_new_buffers(page, from, to);
996 err = ext4_decrypt(page);
1001 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1002 loff_t pos, unsigned len, unsigned flags,
1003 struct page **pagep, void **fsdata)
1005 struct inode *inode = mapping->host;
1006 int ret, needed_blocks;
1013 trace_ext4_write_begin(inode, pos, len, flags);
1015 * Reserve one block more for addition to orphan list in case
1016 * we allocate blocks but write fails for some reason
1018 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1019 index = pos >> PAGE_CACHE_SHIFT;
1020 from = pos & (PAGE_CACHE_SIZE - 1);
1023 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1024 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1033 * grab_cache_page_write_begin() can take a long time if the
1034 * system is thrashing due to memory pressure, or if the page
1035 * is being written back. So grab it first before we start
1036 * the transaction handle. This also allows us to allocate
1037 * the page (if needed) without using GFP_NOFS.
1040 page = grab_cache_page_write_begin(mapping, index, flags);
1046 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1047 if (IS_ERR(handle)) {
1048 page_cache_release(page);
1049 return PTR_ERR(handle);
1053 if (page->mapping != mapping) {
1054 /* The page got truncated from under us */
1056 page_cache_release(page);
1057 ext4_journal_stop(handle);
1060 /* In case writeback began while the page was unlocked */
1061 wait_for_stable_page(page);
1063 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1064 if (ext4_should_dioread_nolock(inode))
1065 ret = ext4_block_write_begin(page, pos, len,
1066 ext4_get_block_write);
1068 ret = ext4_block_write_begin(page, pos, len,
1071 if (ext4_should_dioread_nolock(inode))
1072 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1074 ret = __block_write_begin(page, pos, len, ext4_get_block);
1076 if (!ret && ext4_should_journal_data(inode)) {
1077 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1079 do_journal_get_write_access);
1085 * __block_write_begin may have instantiated a few blocks
1086 * outside i_size. Trim these off again. Don't need
1087 * i_size_read because we hold i_mutex.
1089 * Add inode to orphan list in case we crash before
1092 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1093 ext4_orphan_add(handle, inode);
1095 ext4_journal_stop(handle);
1096 if (pos + len > inode->i_size) {
1097 ext4_truncate_failed_write(inode);
1099 * If truncate failed early the inode might
1100 * still be on the orphan list; we need to
1101 * make sure the inode is removed from the
1102 * orphan list in that case.
1105 ext4_orphan_del(NULL, inode);
1108 if (ret == -ENOSPC &&
1109 ext4_should_retry_alloc(inode->i_sb, &retries))
1111 page_cache_release(page);
1118 /* For write_end() in data=journal mode */
1119 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1122 if (!buffer_mapped(bh) || buffer_freed(bh))
1124 set_buffer_uptodate(bh);
1125 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1126 clear_buffer_meta(bh);
1127 clear_buffer_prio(bh);
1132 * We need to pick up the new inode size which generic_commit_write gave us
1133 * `file' can be NULL - eg, when called from page_symlink().
1135 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1136 * buffers are managed internally.
1138 static int ext4_write_end(struct file *file,
1139 struct address_space *mapping,
1140 loff_t pos, unsigned len, unsigned copied,
1141 struct page *page, void *fsdata)
1143 handle_t *handle = ext4_journal_current_handle();
1144 struct inode *inode = mapping->host;
1145 loff_t old_size = inode->i_size;
1147 int i_size_changed = 0;
1149 trace_ext4_write_end(inode, pos, len, copied);
1150 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1151 ret = ext4_jbd2_file_inode(handle, inode);
1154 page_cache_release(page);
1159 if (ext4_has_inline_data(inode)) {
1160 ret = ext4_write_inline_data_end(inode, pos, len,
1166 copied = block_write_end(file, mapping, pos,
1167 len, copied, page, fsdata);
1169 * it's important to update i_size while still holding page lock:
1170 * page writeout could otherwise come in and zero beyond i_size.
1172 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1174 page_cache_release(page);
1177 pagecache_isize_extended(inode, old_size, pos);
1179 * Don't mark the inode dirty under page lock. First, it unnecessarily
1180 * makes the holding time of page lock longer. Second, it forces lock
1181 * ordering of page lock and transaction start for journaling
1185 ext4_mark_inode_dirty(handle, inode);
1187 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1188 /* if we have allocated more blocks and copied
1189 * less. We will have blocks allocated outside
1190 * inode->i_size. So truncate them
1192 ext4_orphan_add(handle, inode);
1194 ret2 = ext4_journal_stop(handle);
1198 if (pos + len > inode->i_size) {
1199 ext4_truncate_failed_write(inode);
1201 * If truncate failed early the inode might still be
1202 * on the orphan list; we need to make sure the inode
1203 * is removed from the orphan list in that case.
1206 ext4_orphan_del(NULL, inode);
1209 return ret ? ret : copied;
1213 * This is a private version of page_zero_new_buffers() which doesn't
1214 * set the buffer to be dirty, since in data=journalled mode we need
1215 * to call ext4_handle_dirty_metadata() instead.
1217 static void zero_new_buffers(struct page *page, unsigned from, unsigned to)
1219 unsigned int block_start = 0, block_end;
1220 struct buffer_head *head, *bh;
1222 bh = head = page_buffers(page);
1224 block_end = block_start + bh->b_size;
1225 if (buffer_new(bh)) {
1226 if (block_end > from && block_start < to) {
1227 if (!PageUptodate(page)) {
1228 unsigned start, size;
1230 start = max(from, block_start);
1231 size = min(to, block_end) - start;
1233 zero_user(page, start, size);
1234 set_buffer_uptodate(bh);
1236 clear_buffer_new(bh);
1239 block_start = block_end;
1240 bh = bh->b_this_page;
1241 } while (bh != head);
1244 static int ext4_journalled_write_end(struct file *file,
1245 struct address_space *mapping,
1246 loff_t pos, unsigned len, unsigned copied,
1247 struct page *page, void *fsdata)
1249 handle_t *handle = ext4_journal_current_handle();
1250 struct inode *inode = mapping->host;
1251 loff_t old_size = inode->i_size;
1255 int size_changed = 0;
1257 trace_ext4_journalled_write_end(inode, pos, len, copied);
1258 from = pos & (PAGE_CACHE_SIZE - 1);
1261 BUG_ON(!ext4_handle_valid(handle));
1263 if (ext4_has_inline_data(inode))
1264 copied = ext4_write_inline_data_end(inode, pos, len,
1268 if (!PageUptodate(page))
1270 zero_new_buffers(page, from+copied, to);
1273 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1274 to, &partial, write_end_fn);
1276 SetPageUptodate(page);
1278 size_changed = ext4_update_inode_size(inode, pos + copied);
1279 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1280 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1282 page_cache_release(page);
1285 pagecache_isize_extended(inode, old_size, pos);
1288 ret2 = ext4_mark_inode_dirty(handle, inode);
1293 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1294 /* if we have allocated more blocks and copied
1295 * less. We will have blocks allocated outside
1296 * inode->i_size. So truncate them
1298 ext4_orphan_add(handle, inode);
1300 ret2 = ext4_journal_stop(handle);
1303 if (pos + len > inode->i_size) {
1304 ext4_truncate_failed_write(inode);
1306 * If truncate failed early the inode might still be
1307 * on the orphan list; we need to make sure the inode
1308 * is removed from the orphan list in that case.
1311 ext4_orphan_del(NULL, inode);
1314 return ret ? ret : copied;
1318 * Reserve space for a single cluster
1320 static int ext4_da_reserve_space(struct inode *inode)
1322 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1323 struct ext4_inode_info *ei = EXT4_I(inode);
1327 * We will charge metadata quota at writeout time; this saves
1328 * us from metadata over-estimation, though we may go over by
1329 * a small amount in the end. Here we just reserve for data.
1331 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1335 spin_lock(&ei->i_block_reservation_lock);
1336 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1337 spin_unlock(&ei->i_block_reservation_lock);
1338 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1341 ei->i_reserved_data_blocks++;
1342 trace_ext4_da_reserve_space(inode);
1343 spin_unlock(&ei->i_block_reservation_lock);
1345 return 0; /* success */
1348 static void ext4_da_release_space(struct inode *inode, int to_free)
1350 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1351 struct ext4_inode_info *ei = EXT4_I(inode);
1354 return; /* Nothing to release, exit */
1356 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1358 trace_ext4_da_release_space(inode, to_free);
1359 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1361 * if there aren't enough reserved blocks, then the
1362 * counter is messed up somewhere. Since this
1363 * function is called from invalidate page, it's
1364 * harmless to return without any action.
1366 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1367 "ino %lu, to_free %d with only %d reserved "
1368 "data blocks", inode->i_ino, to_free,
1369 ei->i_reserved_data_blocks);
1371 to_free = ei->i_reserved_data_blocks;
1373 ei->i_reserved_data_blocks -= to_free;
1375 /* update fs dirty data blocks counter */
1376 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1378 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1380 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1383 static void ext4_da_page_release_reservation(struct page *page,
1384 unsigned int offset,
1385 unsigned int length)
1387 int to_release = 0, contiguous_blks = 0;
1388 struct buffer_head *head, *bh;
1389 unsigned int curr_off = 0;
1390 struct inode *inode = page->mapping->host;
1391 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1392 unsigned int stop = offset + length;
1396 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1398 head = page_buffers(page);
1401 unsigned int next_off = curr_off + bh->b_size;
1403 if (next_off > stop)
1406 if ((offset <= curr_off) && (buffer_delay(bh))) {
1409 clear_buffer_delay(bh);
1410 } else if (contiguous_blks) {
1411 lblk = page->index <<
1412 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1413 lblk += (curr_off >> inode->i_blkbits) -
1415 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1416 contiguous_blks = 0;
1418 curr_off = next_off;
1419 } while ((bh = bh->b_this_page) != head);
1421 if (contiguous_blks) {
1422 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1423 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1424 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1427 /* If we have released all the blocks belonging to a cluster, then we
1428 * need to release the reserved space for that cluster. */
1429 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1430 while (num_clusters > 0) {
1431 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1432 ((num_clusters - 1) << sbi->s_cluster_bits);
1433 if (sbi->s_cluster_ratio == 1 ||
1434 !ext4_find_delalloc_cluster(inode, lblk))
1435 ext4_da_release_space(inode, 1);
1442 * Delayed allocation stuff
1445 struct mpage_da_data {
1446 struct inode *inode;
1447 struct writeback_control *wbc;
1449 pgoff_t first_page; /* The first page to write */
1450 pgoff_t next_page; /* Current page to examine */
1451 pgoff_t last_page; /* Last page to examine */
1453 * Extent to map - this can be after first_page because that can be
1454 * fully mapped. We somewhat abuse m_flags to store whether the extent
1455 * is delalloc or unwritten.
1457 struct ext4_map_blocks map;
1458 struct ext4_io_submit io_submit; /* IO submission data */
1461 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1466 struct pagevec pvec;
1467 struct inode *inode = mpd->inode;
1468 struct address_space *mapping = inode->i_mapping;
1470 /* This is necessary when next_page == 0. */
1471 if (mpd->first_page >= mpd->next_page)
1474 index = mpd->first_page;
1475 end = mpd->next_page - 1;
1477 ext4_lblk_t start, last;
1478 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1479 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1480 ext4_es_remove_extent(inode, start, last - start + 1);
1483 pagevec_init(&pvec, 0);
1484 while (index <= end) {
1485 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1488 for (i = 0; i < nr_pages; i++) {
1489 struct page *page = pvec.pages[i];
1490 if (page->index > end)
1492 BUG_ON(!PageLocked(page));
1493 BUG_ON(PageWriteback(page));
1495 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1496 ClearPageUptodate(page);
1500 index = pvec.pages[nr_pages - 1]->index + 1;
1501 pagevec_release(&pvec);
1505 static void ext4_print_free_blocks(struct inode *inode)
1507 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1508 struct super_block *sb = inode->i_sb;
1509 struct ext4_inode_info *ei = EXT4_I(inode);
1511 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1512 EXT4_C2B(EXT4_SB(inode->i_sb),
1513 ext4_count_free_clusters(sb)));
1514 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1515 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1516 (long long) EXT4_C2B(EXT4_SB(sb),
1517 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1518 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1519 (long long) EXT4_C2B(EXT4_SB(sb),
1520 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1521 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1522 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1523 ei->i_reserved_data_blocks);
1527 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1529 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1533 * This function is grabs code from the very beginning of
1534 * ext4_map_blocks, but assumes that the caller is from delayed write
1535 * time. This function looks up the requested blocks and sets the
1536 * buffer delay bit under the protection of i_data_sem.
1538 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1539 struct ext4_map_blocks *map,
1540 struct buffer_head *bh)
1542 struct extent_status es;
1544 sector_t invalid_block = ~((sector_t) 0xffff);
1545 #ifdef ES_AGGRESSIVE_TEST
1546 struct ext4_map_blocks orig_map;
1548 memcpy(&orig_map, map, sizeof(*map));
1551 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1555 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1556 "logical block %lu\n", inode->i_ino, map->m_len,
1557 (unsigned long) map->m_lblk);
1559 /* Lookup extent status tree firstly */
1560 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1561 if (ext4_es_is_hole(&es)) {
1563 down_read(&EXT4_I(inode)->i_data_sem);
1568 * Delayed extent could be allocated by fallocate.
1569 * So we need to check it.
1571 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1572 map_bh(bh, inode->i_sb, invalid_block);
1574 set_buffer_delay(bh);
1578 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1579 retval = es.es_len - (iblock - es.es_lblk);
1580 if (retval > map->m_len)
1581 retval = map->m_len;
1582 map->m_len = retval;
1583 if (ext4_es_is_written(&es))
1584 map->m_flags |= EXT4_MAP_MAPPED;
1585 else if (ext4_es_is_unwritten(&es))
1586 map->m_flags |= EXT4_MAP_UNWRITTEN;
1590 #ifdef ES_AGGRESSIVE_TEST
1591 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1597 * Try to see if we can get the block without requesting a new
1598 * file system block.
1600 down_read(&EXT4_I(inode)->i_data_sem);
1601 if (ext4_has_inline_data(inode))
1603 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1604 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1606 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1612 * XXX: __block_prepare_write() unmaps passed block,
1616 * If the block was allocated from previously allocated cluster,
1617 * then we don't need to reserve it again. However we still need
1618 * to reserve metadata for every block we're going to write.
1620 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1621 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1622 ret = ext4_da_reserve_space(inode);
1624 /* not enough space to reserve */
1630 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1631 ~0, EXTENT_STATUS_DELAYED);
1637 map_bh(bh, inode->i_sb, invalid_block);
1639 set_buffer_delay(bh);
1640 } else if (retval > 0) {
1642 unsigned int status;
1644 if (unlikely(retval != map->m_len)) {
1645 ext4_warning(inode->i_sb,
1646 "ES len assertion failed for inode "
1647 "%lu: retval %d != map->m_len %d",
1648 inode->i_ino, retval, map->m_len);
1652 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1653 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1654 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1655 map->m_pblk, status);
1661 up_read((&EXT4_I(inode)->i_data_sem));
1667 * This is a special get_block_t callback which is used by
1668 * ext4_da_write_begin(). It will either return mapped block or
1669 * reserve space for a single block.
1671 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1672 * We also have b_blocknr = -1 and b_bdev initialized properly
1674 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1675 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1676 * initialized properly.
1678 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1679 struct buffer_head *bh, int create)
1681 struct ext4_map_blocks map;
1684 BUG_ON(create == 0);
1685 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1687 map.m_lblk = iblock;
1691 * first, we need to know whether the block is allocated already
1692 * preallocated blocks are unmapped but should treated
1693 * the same as allocated blocks.
1695 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1699 map_bh(bh, inode->i_sb, map.m_pblk);
1700 ext4_update_bh_state(bh, map.m_flags);
1702 if (buffer_unwritten(bh)) {
1703 /* A delayed write to unwritten bh should be marked
1704 * new and mapped. Mapped ensures that we don't do
1705 * get_block multiple times when we write to the same
1706 * offset and new ensures that we do proper zero out
1707 * for partial write.
1710 set_buffer_mapped(bh);
1715 static int bget_one(handle_t *handle, struct buffer_head *bh)
1721 static int bput_one(handle_t *handle, struct buffer_head *bh)
1727 static int __ext4_journalled_writepage(struct page *page,
1730 struct address_space *mapping = page->mapping;
1731 struct inode *inode = mapping->host;
1732 struct buffer_head *page_bufs = NULL;
1733 handle_t *handle = NULL;
1734 int ret = 0, err = 0;
1735 int inline_data = ext4_has_inline_data(inode);
1736 struct buffer_head *inode_bh = NULL;
1738 ClearPageChecked(page);
1741 BUG_ON(page->index != 0);
1742 BUG_ON(len > ext4_get_max_inline_size(inode));
1743 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1744 if (inode_bh == NULL)
1747 page_bufs = page_buffers(page);
1752 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1756 * We need to release the page lock before we start the
1757 * journal, so grab a reference so the page won't disappear
1758 * out from under us.
1763 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1764 ext4_writepage_trans_blocks(inode));
1765 if (IS_ERR(handle)) {
1766 ret = PTR_ERR(handle);
1768 goto out_no_pagelock;
1770 BUG_ON(!ext4_handle_valid(handle));
1774 if (page->mapping != mapping) {
1775 /* The page got truncated from under us */
1776 ext4_journal_stop(handle);
1782 BUFFER_TRACE(inode_bh, "get write access");
1783 ret = ext4_journal_get_write_access(handle, inode_bh);
1785 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1788 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1789 do_journal_get_write_access);
1791 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1796 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1797 err = ext4_journal_stop(handle);
1801 if (!ext4_has_inline_data(inode))
1802 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1804 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1813 * Note that we don't need to start a transaction unless we're journaling data
1814 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1815 * need to file the inode to the transaction's list in ordered mode because if
1816 * we are writing back data added by write(), the inode is already there and if
1817 * we are writing back data modified via mmap(), no one guarantees in which
1818 * transaction the data will hit the disk. In case we are journaling data, we
1819 * cannot start transaction directly because transaction start ranks above page
1820 * lock so we have to do some magic.
1822 * This function can get called via...
1823 * - ext4_writepages after taking page lock (have journal handle)
1824 * - journal_submit_inode_data_buffers (no journal handle)
1825 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1826 * - grab_page_cache when doing write_begin (have journal handle)
1828 * We don't do any block allocation in this function. If we have page with
1829 * multiple blocks we need to write those buffer_heads that are mapped. This
1830 * is important for mmaped based write. So if we do with blocksize 1K
1831 * truncate(f, 1024);
1832 * a = mmap(f, 0, 4096);
1834 * truncate(f, 4096);
1835 * we have in the page first buffer_head mapped via page_mkwrite call back
1836 * but other buffer_heads would be unmapped but dirty (dirty done via the
1837 * do_wp_page). So writepage should write the first block. If we modify
1838 * the mmap area beyond 1024 we will again get a page_fault and the
1839 * page_mkwrite callback will do the block allocation and mark the
1840 * buffer_heads mapped.
1842 * We redirty the page if we have any buffer_heads that is either delay or
1843 * unwritten in the page.
1845 * We can get recursively called as show below.
1847 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1850 * But since we don't do any block allocation we should not deadlock.
1851 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1853 static int ext4_writepage(struct page *page,
1854 struct writeback_control *wbc)
1859 struct buffer_head *page_bufs = NULL;
1860 struct inode *inode = page->mapping->host;
1861 struct ext4_io_submit io_submit;
1862 bool keep_towrite = false;
1864 trace_ext4_writepage(page);
1865 size = i_size_read(inode);
1866 if (page->index == size >> PAGE_CACHE_SHIFT)
1867 len = size & ~PAGE_CACHE_MASK;
1869 len = PAGE_CACHE_SIZE;
1871 page_bufs = page_buffers(page);
1873 * We cannot do block allocation or other extent handling in this
1874 * function. If there are buffers needing that, we have to redirty
1875 * the page. But we may reach here when we do a journal commit via
1876 * journal_submit_inode_data_buffers() and in that case we must write
1877 * allocated buffers to achieve data=ordered mode guarantees.
1879 * Also, if there is only one buffer per page (the fs block
1880 * size == the page size), if one buffer needs block
1881 * allocation or needs to modify the extent tree to clear the
1882 * unwritten flag, we know that the page can't be written at
1883 * all, so we might as well refuse the write immediately.
1884 * Unfortunately if the block size != page size, we can't as
1885 * easily detect this case using ext4_walk_page_buffers(), but
1886 * for the extremely common case, this is an optimization that
1887 * skips a useless round trip through ext4_bio_write_page().
1889 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1890 ext4_bh_delay_or_unwritten)) {
1891 redirty_page_for_writepage(wbc, page);
1892 if ((current->flags & PF_MEMALLOC) ||
1893 (inode->i_sb->s_blocksize == PAGE_CACHE_SIZE)) {
1895 * For memory cleaning there's no point in writing only
1896 * some buffers. So just bail out. Warn if we came here
1897 * from direct reclaim.
1899 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1904 keep_towrite = true;
1907 if (PageChecked(page) && ext4_should_journal_data(inode))
1909 * It's mmapped pagecache. Add buffers and journal it. There
1910 * doesn't seem much point in redirtying the page here.
1912 return __ext4_journalled_writepage(page, len);
1914 ext4_io_submit_init(&io_submit, wbc);
1915 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1916 if (!io_submit.io_end) {
1917 redirty_page_for_writepage(wbc, page);
1921 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1922 ext4_io_submit(&io_submit);
1923 /* Drop io_end reference we got from init */
1924 ext4_put_io_end_defer(io_submit.io_end);
1928 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1931 loff_t size = i_size_read(mpd->inode);
1934 BUG_ON(page->index != mpd->first_page);
1935 if (page->index == size >> PAGE_CACHE_SHIFT)
1936 len = size & ~PAGE_CACHE_MASK;
1938 len = PAGE_CACHE_SIZE;
1939 clear_page_dirty_for_io(page);
1940 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1942 mpd->wbc->nr_to_write--;
1948 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1951 * mballoc gives us at most this number of blocks...
1952 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1953 * The rest of mballoc seems to handle chunks up to full group size.
1955 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1958 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1960 * @mpd - extent of blocks
1961 * @lblk - logical number of the block in the file
1962 * @bh - buffer head we want to add to the extent
1964 * The function is used to collect contig. blocks in the same state. If the
1965 * buffer doesn't require mapping for writeback and we haven't started the
1966 * extent of buffers to map yet, the function returns 'true' immediately - the
1967 * caller can write the buffer right away. Otherwise the function returns true
1968 * if the block has been added to the extent, false if the block couldn't be
1971 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1972 struct buffer_head *bh)
1974 struct ext4_map_blocks *map = &mpd->map;
1976 /* Buffer that doesn't need mapping for writeback? */
1977 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1978 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1979 /* So far no extent to map => we write the buffer right away */
1980 if (map->m_len == 0)
1985 /* First block in the extent? */
1986 if (map->m_len == 0) {
1989 map->m_flags = bh->b_state & BH_FLAGS;
1993 /* Don't go larger than mballoc is willing to allocate */
1994 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1997 /* Can we merge the block to our big extent? */
1998 if (lblk == map->m_lblk + map->m_len &&
1999 (bh->b_state & BH_FLAGS) == map->m_flags) {
2007 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2009 * @mpd - extent of blocks for mapping
2010 * @head - the first buffer in the page
2011 * @bh - buffer we should start processing from
2012 * @lblk - logical number of the block in the file corresponding to @bh
2014 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2015 * the page for IO if all buffers in this page were mapped and there's no
2016 * accumulated extent of buffers to map or add buffers in the page to the
2017 * extent of buffers to map. The function returns 1 if the caller can continue
2018 * by processing the next page, 0 if it should stop adding buffers to the
2019 * extent to map because we cannot extend it anymore. It can also return value
2020 * < 0 in case of error during IO submission.
2022 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2023 struct buffer_head *head,
2024 struct buffer_head *bh,
2027 struct inode *inode = mpd->inode;
2029 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2030 >> inode->i_blkbits;
2033 BUG_ON(buffer_locked(bh));
2035 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2036 /* Found extent to map? */
2039 /* Everything mapped so far and we hit EOF */
2042 } while (lblk++, (bh = bh->b_this_page) != head);
2043 /* So far everything mapped? Submit the page for IO. */
2044 if (mpd->map.m_len == 0) {
2045 err = mpage_submit_page(mpd, head->b_page);
2049 return lblk < blocks;
2053 * mpage_map_buffers - update buffers corresponding to changed extent and
2054 * submit fully mapped pages for IO
2056 * @mpd - description of extent to map, on return next extent to map
2058 * Scan buffers corresponding to changed extent (we expect corresponding pages
2059 * to be already locked) and update buffer state according to new extent state.
2060 * We map delalloc buffers to their physical location, clear unwritten bits,
2061 * and mark buffers as uninit when we perform writes to unwritten extents
2062 * and do extent conversion after IO is finished. If the last page is not fully
2063 * mapped, we update @map to the next extent in the last page that needs
2064 * mapping. Otherwise we submit the page for IO.
2066 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2068 struct pagevec pvec;
2070 struct inode *inode = mpd->inode;
2071 struct buffer_head *head, *bh;
2072 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
2078 start = mpd->map.m_lblk >> bpp_bits;
2079 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2080 lblk = start << bpp_bits;
2081 pblock = mpd->map.m_pblk;
2083 pagevec_init(&pvec, 0);
2084 while (start <= end) {
2085 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2089 for (i = 0; i < nr_pages; i++) {
2090 struct page *page = pvec.pages[i];
2092 if (page->index > end)
2094 /* Up to 'end' pages must be contiguous */
2095 BUG_ON(page->index != start);
2096 bh = head = page_buffers(page);
2098 if (lblk < mpd->map.m_lblk)
2100 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2102 * Buffer after end of mapped extent.
2103 * Find next buffer in the page to map.
2106 mpd->map.m_flags = 0;
2108 * FIXME: If dioread_nolock supports
2109 * blocksize < pagesize, we need to make
2110 * sure we add size mapped so far to
2111 * io_end->size as the following call
2112 * can submit the page for IO.
2114 err = mpage_process_page_bufs(mpd, head,
2116 pagevec_release(&pvec);
2121 if (buffer_delay(bh)) {
2122 clear_buffer_delay(bh);
2123 bh->b_blocknr = pblock++;
2125 clear_buffer_unwritten(bh);
2126 } while (lblk++, (bh = bh->b_this_page) != head);
2129 * FIXME: This is going to break if dioread_nolock
2130 * supports blocksize < pagesize as we will try to
2131 * convert potentially unmapped parts of inode.
2133 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2134 /* Page fully mapped - let IO run! */
2135 err = mpage_submit_page(mpd, page);
2137 pagevec_release(&pvec);
2142 pagevec_release(&pvec);
2144 /* Extent fully mapped and matches with page boundary. We are done. */
2146 mpd->map.m_flags = 0;
2150 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2152 struct inode *inode = mpd->inode;
2153 struct ext4_map_blocks *map = &mpd->map;
2154 int get_blocks_flags;
2155 int err, dioread_nolock;
2157 trace_ext4_da_write_pages_extent(inode, map);
2159 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2160 * to convert an unwritten extent to be initialized (in the case
2161 * where we have written into one or more preallocated blocks). It is
2162 * possible that we're going to need more metadata blocks than
2163 * previously reserved. However we must not fail because we're in
2164 * writeback and there is nothing we can do about it so it might result
2165 * in data loss. So use reserved blocks to allocate metadata if
2168 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2169 * the blocks in question are delalloc blocks. This indicates
2170 * that the blocks and quotas has already been checked when
2171 * the data was copied into the page cache.
2173 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2174 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2175 dioread_nolock = ext4_should_dioread_nolock(inode);
2177 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2178 if (map->m_flags & (1 << BH_Delay))
2179 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2181 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2184 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2185 if (!mpd->io_submit.io_end->handle &&
2186 ext4_handle_valid(handle)) {
2187 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2188 handle->h_rsv_handle = NULL;
2190 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2193 BUG_ON(map->m_len == 0);
2194 if (map->m_flags & EXT4_MAP_NEW) {
2195 struct block_device *bdev = inode->i_sb->s_bdev;
2198 for (i = 0; i < map->m_len; i++)
2199 unmap_underlying_metadata(bdev, map->m_pblk + i);
2205 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2206 * mpd->len and submit pages underlying it for IO
2208 * @handle - handle for journal operations
2209 * @mpd - extent to map
2210 * @give_up_on_write - we set this to true iff there is a fatal error and there
2211 * is no hope of writing the data. The caller should discard
2212 * dirty pages to avoid infinite loops.
2214 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2215 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2216 * them to initialized or split the described range from larger unwritten
2217 * extent. Note that we need not map all the described range since allocation
2218 * can return less blocks or the range is covered by more unwritten extents. We
2219 * cannot map more because we are limited by reserved transaction credits. On
2220 * the other hand we always make sure that the last touched page is fully
2221 * mapped so that it can be written out (and thus forward progress is
2222 * guaranteed). After mapping we submit all mapped pages for IO.
2224 static int mpage_map_and_submit_extent(handle_t *handle,
2225 struct mpage_da_data *mpd,
2226 bool *give_up_on_write)
2228 struct inode *inode = mpd->inode;
2229 struct ext4_map_blocks *map = &mpd->map;
2234 mpd->io_submit.io_end->offset =
2235 ((loff_t)map->m_lblk) << inode->i_blkbits;
2237 err = mpage_map_one_extent(handle, mpd);
2239 struct super_block *sb = inode->i_sb;
2241 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2242 goto invalidate_dirty_pages;
2244 * Let the uper layers retry transient errors.
2245 * In the case of ENOSPC, if ext4_count_free_blocks()
2246 * is non-zero, a commit should free up blocks.
2248 if ((err == -ENOMEM) ||
2249 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2251 goto update_disksize;
2254 ext4_msg(sb, KERN_CRIT,
2255 "Delayed block allocation failed for "
2256 "inode %lu at logical offset %llu with"
2257 " max blocks %u with error %d",
2259 (unsigned long long)map->m_lblk,
2260 (unsigned)map->m_len, -err);
2261 ext4_msg(sb, KERN_CRIT,
2262 "This should not happen!! Data will "
2265 ext4_print_free_blocks(inode);
2266 invalidate_dirty_pages:
2267 *give_up_on_write = true;
2272 * Update buffer state, submit mapped pages, and get us new
2275 err = mpage_map_and_submit_buffers(mpd);
2277 goto update_disksize;
2278 } while (map->m_len);
2282 * Update on-disk size after IO is submitted. Races with
2283 * truncate are avoided by checking i_size under i_data_sem.
2285 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2286 if (disksize > EXT4_I(inode)->i_disksize) {
2290 down_write(&EXT4_I(inode)->i_data_sem);
2291 i_size = i_size_read(inode);
2292 if (disksize > i_size)
2294 if (disksize > EXT4_I(inode)->i_disksize)
2295 EXT4_I(inode)->i_disksize = disksize;
2296 err2 = ext4_mark_inode_dirty(handle, inode);
2297 up_write(&EXT4_I(inode)->i_data_sem);
2299 ext4_error(inode->i_sb,
2300 "Failed to mark inode %lu dirty",
2309 * Calculate the total number of credits to reserve for one writepages
2310 * iteration. This is called from ext4_writepages(). We map an extent of
2311 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2312 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2313 * bpp - 1 blocks in bpp different extents.
2315 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2317 int bpp = ext4_journal_blocks_per_page(inode);
2319 return ext4_meta_trans_blocks(inode,
2320 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2324 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2325 * and underlying extent to map
2327 * @mpd - where to look for pages
2329 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2330 * IO immediately. When we find a page which isn't mapped we start accumulating
2331 * extent of buffers underlying these pages that needs mapping (formed by
2332 * either delayed or unwritten buffers). We also lock the pages containing
2333 * these buffers. The extent found is returned in @mpd structure (starting at
2334 * mpd->lblk with length mpd->len blocks).
2336 * Note that this function can attach bios to one io_end structure which are
2337 * neither logically nor physically contiguous. Although it may seem as an
2338 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2339 * case as we need to track IO to all buffers underlying a page in one io_end.
2341 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2343 struct address_space *mapping = mpd->inode->i_mapping;
2344 struct pagevec pvec;
2345 unsigned int nr_pages;
2346 long left = mpd->wbc->nr_to_write;
2347 pgoff_t index = mpd->first_page;
2348 pgoff_t end = mpd->last_page;
2351 int blkbits = mpd->inode->i_blkbits;
2353 struct buffer_head *head;
2355 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2356 tag = PAGECACHE_TAG_TOWRITE;
2358 tag = PAGECACHE_TAG_DIRTY;
2360 pagevec_init(&pvec, 0);
2362 mpd->next_page = index;
2363 while (index <= end) {
2364 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2365 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2369 for (i = 0; i < nr_pages; i++) {
2370 struct page *page = pvec.pages[i];
2373 * At this point, the page may be truncated or
2374 * invalidated (changing page->mapping to NULL), or
2375 * even swizzled back from swapper_space to tmpfs file
2376 * mapping. However, page->index will not change
2377 * because we have a reference on the page.
2379 if (page->index > end)
2383 * Accumulated enough dirty pages? This doesn't apply
2384 * to WB_SYNC_ALL mode. For integrity sync we have to
2385 * keep going because someone may be concurrently
2386 * dirtying pages, and we might have synced a lot of
2387 * newly appeared dirty pages, but have not synced all
2388 * of the old dirty pages.
2390 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2393 /* If we can't merge this page, we are done. */
2394 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2399 * If the page is no longer dirty, or its mapping no
2400 * longer corresponds to inode we are writing (which
2401 * means it has been truncated or invalidated), or the
2402 * page is already under writeback and we are not doing
2403 * a data integrity writeback, skip the page
2405 if (!PageDirty(page) ||
2406 (PageWriteback(page) &&
2407 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2408 unlikely(page->mapping != mapping)) {
2413 wait_on_page_writeback(page);
2414 BUG_ON(PageWriteback(page));
2416 if (mpd->map.m_len == 0)
2417 mpd->first_page = page->index;
2418 mpd->next_page = page->index + 1;
2419 /* Add all dirty buffers to mpd */
2420 lblk = ((ext4_lblk_t)page->index) <<
2421 (PAGE_CACHE_SHIFT - blkbits);
2422 head = page_buffers(page);
2423 err = mpage_process_page_bufs(mpd, head, head, lblk);
2429 pagevec_release(&pvec);
2434 pagevec_release(&pvec);
2438 static int __writepage(struct page *page, struct writeback_control *wbc,
2441 struct address_space *mapping = data;
2442 int ret = ext4_writepage(page, wbc);
2443 mapping_set_error(mapping, ret);
2447 static int ext4_writepages(struct address_space *mapping,
2448 struct writeback_control *wbc)
2450 pgoff_t writeback_index = 0;
2451 long nr_to_write = wbc->nr_to_write;
2452 int range_whole = 0;
2454 handle_t *handle = NULL;
2455 struct mpage_da_data mpd;
2456 struct inode *inode = mapping->host;
2457 int needed_blocks, rsv_blocks = 0, ret = 0;
2458 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2460 struct blk_plug plug;
2461 bool give_up_on_write = false;
2463 trace_ext4_writepages(inode, wbc);
2466 * No pages to write? This is mainly a kludge to avoid starting
2467 * a transaction for special inodes like journal inode on last iput()
2468 * because that could violate lock ordering on umount
2470 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2471 goto out_writepages;
2473 if (ext4_should_journal_data(inode)) {
2474 struct blk_plug plug;
2476 blk_start_plug(&plug);
2477 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2478 blk_finish_plug(&plug);
2479 goto out_writepages;
2483 * If the filesystem has aborted, it is read-only, so return
2484 * right away instead of dumping stack traces later on that
2485 * will obscure the real source of the problem. We test
2486 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2487 * the latter could be true if the filesystem is mounted
2488 * read-only, and in that case, ext4_writepages should
2489 * *never* be called, so if that ever happens, we would want
2492 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2494 goto out_writepages;
2497 if (ext4_should_dioread_nolock(inode)) {
2499 * We may need to convert up to one extent per block in
2500 * the page and we may dirty the inode.
2502 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2506 * If we have inline data and arrive here, it means that
2507 * we will soon create the block for the 1st page, so
2508 * we'd better clear the inline data here.
2510 if (ext4_has_inline_data(inode)) {
2511 /* Just inode will be modified... */
2512 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2513 if (IS_ERR(handle)) {
2514 ret = PTR_ERR(handle);
2515 goto out_writepages;
2517 BUG_ON(ext4_test_inode_state(inode,
2518 EXT4_STATE_MAY_INLINE_DATA));
2519 ext4_destroy_inline_data(handle, inode);
2520 ext4_journal_stop(handle);
2523 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2526 if (wbc->range_cyclic) {
2527 writeback_index = mapping->writeback_index;
2528 if (writeback_index)
2530 mpd.first_page = writeback_index;
2533 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2534 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2539 ext4_io_submit_init(&mpd.io_submit, wbc);
2541 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2542 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2544 blk_start_plug(&plug);
2545 while (!done && mpd.first_page <= mpd.last_page) {
2546 /* For each extent of pages we use new io_end */
2547 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2548 if (!mpd.io_submit.io_end) {
2554 * We have two constraints: We find one extent to map and we
2555 * must always write out whole page (makes a difference when
2556 * blocksize < pagesize) so that we don't block on IO when we
2557 * try to write out the rest of the page. Journalled mode is
2558 * not supported by delalloc.
2560 BUG_ON(ext4_should_journal_data(inode));
2561 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2563 /* start a new transaction */
2564 handle = ext4_journal_start_with_reserve(inode,
2565 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2566 if (IS_ERR(handle)) {
2567 ret = PTR_ERR(handle);
2568 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2569 "%ld pages, ino %lu; err %d", __func__,
2570 wbc->nr_to_write, inode->i_ino, ret);
2571 /* Release allocated io_end */
2572 ext4_put_io_end(mpd.io_submit.io_end);
2576 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2577 ret = mpage_prepare_extent_to_map(&mpd);
2580 ret = mpage_map_and_submit_extent(handle, &mpd,
2584 * We scanned the whole range (or exhausted
2585 * nr_to_write), submitted what was mapped and
2586 * didn't find anything needing mapping. We are
2593 * Caution: If the handle is synchronous,
2594 * ext4_journal_stop() can wait for transaction commit
2595 * to finish which may depend on writeback of pages to
2596 * complete or on page lock to be released. In that
2597 * case, we have to wait until after after we have
2598 * submitted all the IO, released page locks we hold,
2599 * and dropped io_end reference (for extent conversion
2600 * to be able to complete) before stopping the handle.
2602 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2603 ext4_journal_stop(handle);
2606 /* Submit prepared bio */
2607 ext4_io_submit(&mpd.io_submit);
2608 /* Unlock pages we didn't use */
2609 mpage_release_unused_pages(&mpd, give_up_on_write);
2611 * Drop our io_end reference we got from init. We have
2612 * to be careful and use deferred io_end finishing if
2613 * we are still holding the transaction as we can
2614 * release the last reference to io_end which may end
2615 * up doing unwritten extent conversion.
2618 ext4_put_io_end_defer(mpd.io_submit.io_end);
2619 ext4_journal_stop(handle);
2621 ext4_put_io_end(mpd.io_submit.io_end);
2623 if (ret == -ENOSPC && sbi->s_journal) {
2625 * Commit the transaction which would
2626 * free blocks released in the transaction
2629 jbd2_journal_force_commit_nested(sbi->s_journal);
2633 /* Fatal error - ENOMEM, EIO... */
2637 blk_finish_plug(&plug);
2638 if (!ret && !cycled && wbc->nr_to_write > 0) {
2640 mpd.last_page = writeback_index - 1;
2646 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2648 * Set the writeback_index so that range_cyclic
2649 * mode will write it back later
2651 mapping->writeback_index = mpd.first_page;
2654 trace_ext4_writepages_result(inode, wbc, ret,
2655 nr_to_write - wbc->nr_to_write);
2659 static int ext4_nonda_switch(struct super_block *sb)
2661 s64 free_clusters, dirty_clusters;
2662 struct ext4_sb_info *sbi = EXT4_SB(sb);
2665 * switch to non delalloc mode if we are running low
2666 * on free block. The free block accounting via percpu
2667 * counters can get slightly wrong with percpu_counter_batch getting
2668 * accumulated on each CPU without updating global counters
2669 * Delalloc need an accurate free block accounting. So switch
2670 * to non delalloc when we are near to error range.
2673 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2675 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2677 * Start pushing delalloc when 1/2 of free blocks are dirty.
2679 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2680 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2682 if (2 * free_clusters < 3 * dirty_clusters ||
2683 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2685 * free block count is less than 150% of dirty blocks
2686 * or free blocks is less than watermark
2693 /* We always reserve for an inode update; the superblock could be there too */
2694 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2696 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2699 if (pos + len <= 0x7fffffffULL)
2702 /* We might need to update the superblock to set LARGE_FILE */
2706 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2707 loff_t pos, unsigned len, unsigned flags,
2708 struct page **pagep, void **fsdata)
2710 int ret, retries = 0;
2713 struct inode *inode = mapping->host;
2716 index = pos >> PAGE_CACHE_SHIFT;
2718 if (ext4_nonda_switch(inode->i_sb)) {
2719 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2720 return ext4_write_begin(file, mapping, pos,
2721 len, flags, pagep, fsdata);
2723 *fsdata = (void *)0;
2724 trace_ext4_da_write_begin(inode, pos, len, flags);
2726 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2727 ret = ext4_da_write_inline_data_begin(mapping, inode,
2737 * grab_cache_page_write_begin() can take a long time if the
2738 * system is thrashing due to memory pressure, or if the page
2739 * is being written back. So grab it first before we start
2740 * the transaction handle. This also allows us to allocate
2741 * the page (if needed) without using GFP_NOFS.
2744 page = grab_cache_page_write_begin(mapping, index, flags);
2750 * With delayed allocation, we don't log the i_disksize update
2751 * if there is delayed block allocation. But we still need
2752 * to journalling the i_disksize update if writes to the end
2753 * of file which has an already mapped buffer.
2756 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2757 ext4_da_write_credits(inode, pos, len));
2758 if (IS_ERR(handle)) {
2759 page_cache_release(page);
2760 return PTR_ERR(handle);
2764 if (page->mapping != mapping) {
2765 /* The page got truncated from under us */
2767 page_cache_release(page);
2768 ext4_journal_stop(handle);
2771 /* In case writeback began while the page was unlocked */
2772 wait_for_stable_page(page);
2774 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2775 ret = ext4_block_write_begin(page, pos, len,
2776 ext4_da_get_block_prep);
2778 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2782 ext4_journal_stop(handle);
2784 * block_write_begin may have instantiated a few blocks
2785 * outside i_size. Trim these off again. Don't need
2786 * i_size_read because we hold i_mutex.
2788 if (pos + len > inode->i_size)
2789 ext4_truncate_failed_write(inode);
2791 if (ret == -ENOSPC &&
2792 ext4_should_retry_alloc(inode->i_sb, &retries))
2795 page_cache_release(page);
2804 * Check if we should update i_disksize
2805 * when write to the end of file but not require block allocation
2807 static int ext4_da_should_update_i_disksize(struct page *page,
2808 unsigned long offset)
2810 struct buffer_head *bh;
2811 struct inode *inode = page->mapping->host;
2815 bh = page_buffers(page);
2816 idx = offset >> inode->i_blkbits;
2818 for (i = 0; i < idx; i++)
2819 bh = bh->b_this_page;
2821 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2826 static int ext4_da_write_end(struct file *file,
2827 struct address_space *mapping,
2828 loff_t pos, unsigned len, unsigned copied,
2829 struct page *page, void *fsdata)
2831 struct inode *inode = mapping->host;
2833 handle_t *handle = ext4_journal_current_handle();
2835 unsigned long start, end;
2836 int write_mode = (int)(unsigned long)fsdata;
2838 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2839 return ext4_write_end(file, mapping, pos,
2840 len, copied, page, fsdata);
2842 trace_ext4_da_write_end(inode, pos, len, copied);
2843 start = pos & (PAGE_CACHE_SIZE - 1);
2844 end = start + copied - 1;
2847 * generic_write_end() will run mark_inode_dirty() if i_size
2848 * changes. So let's piggyback the i_disksize mark_inode_dirty
2851 new_i_size = pos + copied;
2852 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2853 if (ext4_has_inline_data(inode) ||
2854 ext4_da_should_update_i_disksize(page, end)) {
2855 ext4_update_i_disksize(inode, new_i_size);
2856 /* We need to mark inode dirty even if
2857 * new_i_size is less that inode->i_size
2858 * bu greater than i_disksize.(hint delalloc)
2860 ext4_mark_inode_dirty(handle, inode);
2864 if (write_mode != CONVERT_INLINE_DATA &&
2865 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2866 ext4_has_inline_data(inode))
2867 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2870 ret2 = generic_write_end(file, mapping, pos, len, copied,
2876 ret2 = ext4_journal_stop(handle);
2880 return ret ? ret : copied;
2883 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2884 unsigned int length)
2887 * Drop reserved blocks
2889 BUG_ON(!PageLocked(page));
2890 if (!page_has_buffers(page))
2893 ext4_da_page_release_reservation(page, offset, length);
2896 ext4_invalidatepage(page, offset, length);
2902 * Force all delayed allocation blocks to be allocated for a given inode.
2904 int ext4_alloc_da_blocks(struct inode *inode)
2906 trace_ext4_alloc_da_blocks(inode);
2908 if (!EXT4_I(inode)->i_reserved_data_blocks)
2912 * We do something simple for now. The filemap_flush() will
2913 * also start triggering a write of the data blocks, which is
2914 * not strictly speaking necessary (and for users of
2915 * laptop_mode, not even desirable). However, to do otherwise
2916 * would require replicating code paths in:
2918 * ext4_writepages() ->
2919 * write_cache_pages() ---> (via passed in callback function)
2920 * __mpage_da_writepage() -->
2921 * mpage_add_bh_to_extent()
2922 * mpage_da_map_blocks()
2924 * The problem is that write_cache_pages(), located in
2925 * mm/page-writeback.c, marks pages clean in preparation for
2926 * doing I/O, which is not desirable if we're not planning on
2929 * We could call write_cache_pages(), and then redirty all of
2930 * the pages by calling redirty_page_for_writepage() but that
2931 * would be ugly in the extreme. So instead we would need to
2932 * replicate parts of the code in the above functions,
2933 * simplifying them because we wouldn't actually intend to
2934 * write out the pages, but rather only collect contiguous
2935 * logical block extents, call the multi-block allocator, and
2936 * then update the buffer heads with the block allocations.
2938 * For now, though, we'll cheat by calling filemap_flush(),
2939 * which will map the blocks, and start the I/O, but not
2940 * actually wait for the I/O to complete.
2942 return filemap_flush(inode->i_mapping);
2946 * bmap() is special. It gets used by applications such as lilo and by
2947 * the swapper to find the on-disk block of a specific piece of data.
2949 * Naturally, this is dangerous if the block concerned is still in the
2950 * journal. If somebody makes a swapfile on an ext4 data-journaling
2951 * filesystem and enables swap, then they may get a nasty shock when the
2952 * data getting swapped to that swapfile suddenly gets overwritten by
2953 * the original zero's written out previously to the journal and
2954 * awaiting writeback in the kernel's buffer cache.
2956 * So, if we see any bmap calls here on a modified, data-journaled file,
2957 * take extra steps to flush any blocks which might be in the cache.
2959 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2961 struct inode *inode = mapping->host;
2966 * We can get here for an inline file via the FIBMAP ioctl
2968 if (ext4_has_inline_data(inode))
2971 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2972 test_opt(inode->i_sb, DELALLOC)) {
2974 * With delalloc we want to sync the file
2975 * so that we can make sure we allocate
2978 filemap_write_and_wait(mapping);
2981 if (EXT4_JOURNAL(inode) &&
2982 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2984 * This is a REALLY heavyweight approach, but the use of
2985 * bmap on dirty files is expected to be extremely rare:
2986 * only if we run lilo or swapon on a freshly made file
2987 * do we expect this to happen.
2989 * (bmap requires CAP_SYS_RAWIO so this does not
2990 * represent an unprivileged user DOS attack --- we'd be
2991 * in trouble if mortal users could trigger this path at
2994 * NB. EXT4_STATE_JDATA is not set on files other than
2995 * regular files. If somebody wants to bmap a directory
2996 * or symlink and gets confused because the buffer
2997 * hasn't yet been flushed to disk, they deserve
2998 * everything they get.
3001 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3002 journal = EXT4_JOURNAL(inode);
3003 jbd2_journal_lock_updates(journal);
3004 err = jbd2_journal_flush(journal);
3005 jbd2_journal_unlock_updates(journal);
3011 return generic_block_bmap(mapping, block, ext4_get_block);
3014 static int ext4_readpage(struct file *file, struct page *page)
3017 struct inode *inode = page->mapping->host;
3019 trace_ext4_readpage(page);
3021 if (ext4_has_inline_data(inode))
3022 ret = ext4_readpage_inline(inode, page);
3025 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3031 ext4_readpages(struct file *file, struct address_space *mapping,
3032 struct list_head *pages, unsigned nr_pages)
3034 struct inode *inode = mapping->host;
3036 /* If the file has inline data, no need to do readpages. */
3037 if (ext4_has_inline_data(inode))
3040 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3043 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3044 unsigned int length)
3046 trace_ext4_invalidatepage(page, offset, length);
3048 /* No journalling happens on data buffers when this function is used */
3049 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3051 block_invalidatepage(page, offset, length);
3054 static int __ext4_journalled_invalidatepage(struct page *page,
3055 unsigned int offset,
3056 unsigned int length)
3058 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3060 trace_ext4_journalled_invalidatepage(page, offset, length);
3063 * If it's a full truncate we just forget about the pending dirtying
3065 if (offset == 0 && length == PAGE_CACHE_SIZE)
3066 ClearPageChecked(page);
3068 return jbd2_journal_invalidatepage(journal, page, offset, length);
3071 /* Wrapper for aops... */
3072 static void ext4_journalled_invalidatepage(struct page *page,
3073 unsigned int offset,
3074 unsigned int length)
3076 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3079 static int ext4_releasepage(struct page *page, gfp_t wait)
3081 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3083 trace_ext4_releasepage(page);
3085 /* Page has dirty journalled data -> cannot release */
3086 if (PageChecked(page))
3089 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3091 return try_to_free_buffers(page);
3095 * ext4_get_block used when preparing for a DIO write or buffer write.
3096 * We allocate an uinitialized extent if blocks haven't been allocated.
3097 * The extent will be converted to initialized after the IO is complete.
3099 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3100 struct buffer_head *bh_result, int create)
3102 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3103 inode->i_ino, create);
3104 return _ext4_get_block(inode, iblock, bh_result,
3105 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3108 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3109 struct buffer_head *bh_result, int create)
3111 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3112 inode->i_ino, create);
3113 return _ext4_get_block(inode, iblock, bh_result,
3114 EXT4_GET_BLOCKS_NO_LOCK);
3117 int ext4_get_block_dax(struct inode *inode, sector_t iblock,
3118 struct buffer_head *bh_result, int create)
3120 int flags = EXT4_GET_BLOCKS_PRE_IO | EXT4_GET_BLOCKS_UNWRIT_EXT;
3122 flags |= EXT4_GET_BLOCKS_CREATE;
3123 ext4_debug("ext4_get_block_dax: inode %lu, create flag %d\n",
3124 inode->i_ino, create);
3125 return _ext4_get_block(inode, iblock, bh_result, flags);
3128 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3129 ssize_t size, void *private)
3131 ext4_io_end_t *io_end = iocb->private;
3133 /* if not async direct IO just return */
3137 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3138 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3139 iocb->private, io_end->inode->i_ino, iocb, offset,
3142 iocb->private = NULL;
3143 io_end->offset = offset;
3144 io_end->size = size;
3145 ext4_put_io_end(io_end);
3149 * For ext4 extent files, ext4 will do direct-io write to holes,
3150 * preallocated extents, and those write extend the file, no need to
3151 * fall back to buffered IO.
3153 * For holes, we fallocate those blocks, mark them as unwritten
3154 * If those blocks were preallocated, we mark sure they are split, but
3155 * still keep the range to write as unwritten.
3157 * The unwritten extents will be converted to written when DIO is completed.
3158 * For async direct IO, since the IO may still pending when return, we
3159 * set up an end_io call back function, which will do the conversion
3160 * when async direct IO completed.
3162 * If the O_DIRECT write will extend the file then add this inode to the
3163 * orphan list. So recovery will truncate it back to the original size
3164 * if the machine crashes during the write.
3167 static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3170 struct file *file = iocb->ki_filp;
3171 struct inode *inode = file->f_mapping->host;
3173 size_t count = iov_iter_count(iter);
3175 get_block_t *get_block_func = NULL;
3177 loff_t final_size = offset + count;
3178 ext4_io_end_t *io_end = NULL;
3180 /* Use the old path for reads and writes beyond i_size. */
3181 if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size)
3182 return ext4_ind_direct_IO(iocb, iter, offset);
3184 BUG_ON(iocb->private == NULL);
3187 * Make all waiters for direct IO properly wait also for extent
3188 * conversion. This also disallows race between truncate() and
3189 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3191 if (iov_iter_rw(iter) == WRITE)
3192 inode_dio_begin(inode);
3194 /* If we do a overwrite dio, i_mutex locking can be released */
3195 overwrite = *((int *)iocb->private);
3198 down_read(&EXT4_I(inode)->i_data_sem);
3199 mutex_unlock(&inode->i_mutex);
3203 * We could direct write to holes and fallocate.
3205 * Allocated blocks to fill the hole are marked as
3206 * unwritten to prevent parallel buffered read to expose
3207 * the stale data before DIO complete the data IO.
3209 * As to previously fallocated extents, ext4 get_block will
3210 * just simply mark the buffer mapped but still keep the
3211 * extents unwritten.
3213 * For non AIO case, we will convert those unwritten extents
3214 * to written after return back from blockdev_direct_IO.
3216 * For async DIO, the conversion needs to be deferred when the
3217 * IO is completed. The ext4 end_io callback function will be
3218 * called to take care of the conversion work. Here for async
3219 * case, we allocate an io_end structure to hook to the iocb.
3221 iocb->private = NULL;
3222 ext4_inode_aio_set(inode, NULL);
3223 if (!is_sync_kiocb(iocb)) {
3224 io_end = ext4_init_io_end(inode, GFP_NOFS);
3230 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3232 iocb->private = ext4_get_io_end(io_end);
3234 * we save the io structure for current async direct
3235 * IO, so that later ext4_map_blocks() could flag the
3236 * io structure whether there is a unwritten extents
3237 * needs to be converted when IO is completed.
3239 ext4_inode_aio_set(inode, io_end);
3243 get_block_func = ext4_get_block_write_nolock;
3245 get_block_func = ext4_get_block_write;
3246 dio_flags = DIO_LOCKING;
3248 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3249 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3252 ret = dax_do_io(iocb, inode, iter, offset, get_block_func,
3253 ext4_end_io_dio, dio_flags);
3255 ret = __blockdev_direct_IO(iocb, inode,
3256 inode->i_sb->s_bdev, iter, offset,
3258 ext4_end_io_dio, NULL, dio_flags);
3261 * Put our reference to io_end. This can free the io_end structure e.g.
3262 * in sync IO case or in case of error. It can even perform extent
3263 * conversion if all bios we submitted finished before we got here.
3264 * Note that in that case iocb->private can be already set to NULL
3268 ext4_inode_aio_set(inode, NULL);
3269 ext4_put_io_end(io_end);
3271 * When no IO was submitted ext4_end_io_dio() was not
3272 * called so we have to put iocb's reference.
3274 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3275 WARN_ON(iocb->private != io_end);
3276 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3277 ext4_put_io_end(io_end);
3278 iocb->private = NULL;
3281 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3282 EXT4_STATE_DIO_UNWRITTEN)) {
3285 * for non AIO case, since the IO is already
3286 * completed, we could do the conversion right here
3288 err = ext4_convert_unwritten_extents(NULL, inode,
3292 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3296 if (iov_iter_rw(iter) == WRITE)
3297 inode_dio_end(inode);
3298 /* take i_mutex locking again if we do a ovewrite dio */
3300 up_read(&EXT4_I(inode)->i_data_sem);
3301 mutex_lock(&inode->i_mutex);
3307 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3310 struct file *file = iocb->ki_filp;
3311 struct inode *inode = file->f_mapping->host;
3312 size_t count = iov_iter_count(iter);
3315 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3316 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3321 * If we are doing data journalling we don't support O_DIRECT
3323 if (ext4_should_journal_data(inode))
3326 /* Let buffer I/O handle the inline data case. */
3327 if (ext4_has_inline_data(inode))
3330 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3331 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3332 ret = ext4_ext_direct_IO(iocb, iter, offset);
3334 ret = ext4_ind_direct_IO(iocb, iter, offset);
3335 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3340 * Pages can be marked dirty completely asynchronously from ext4's journalling
3341 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3342 * much here because ->set_page_dirty is called under VFS locks. The page is
3343 * not necessarily locked.
3345 * We cannot just dirty the page and leave attached buffers clean, because the
3346 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3347 * or jbddirty because all the journalling code will explode.
3349 * So what we do is to mark the page "pending dirty" and next time writepage
3350 * is called, propagate that into the buffers appropriately.
3352 static int ext4_journalled_set_page_dirty(struct page *page)
3354 SetPageChecked(page);
3355 return __set_page_dirty_nobuffers(page);
3358 static const struct address_space_operations ext4_aops = {
3359 .readpage = ext4_readpage,
3360 .readpages = ext4_readpages,
3361 .writepage = ext4_writepage,
3362 .writepages = ext4_writepages,
3363 .write_begin = ext4_write_begin,
3364 .write_end = ext4_write_end,
3366 .invalidatepage = ext4_invalidatepage,
3367 .releasepage = ext4_releasepage,
3368 .direct_IO = ext4_direct_IO,
3369 .migratepage = buffer_migrate_page,
3370 .is_partially_uptodate = block_is_partially_uptodate,
3371 .error_remove_page = generic_error_remove_page,
3374 static const struct address_space_operations ext4_journalled_aops = {
3375 .readpage = ext4_readpage,
3376 .readpages = ext4_readpages,
3377 .writepage = ext4_writepage,
3378 .writepages = ext4_writepages,
3379 .write_begin = ext4_write_begin,
3380 .write_end = ext4_journalled_write_end,
3381 .set_page_dirty = ext4_journalled_set_page_dirty,
3383 .invalidatepage = ext4_journalled_invalidatepage,
3384 .releasepage = ext4_releasepage,
3385 .direct_IO = ext4_direct_IO,
3386 .is_partially_uptodate = block_is_partially_uptodate,
3387 .error_remove_page = generic_error_remove_page,
3390 static const struct address_space_operations ext4_da_aops = {
3391 .readpage = ext4_readpage,
3392 .readpages = ext4_readpages,
3393 .writepage = ext4_writepage,
3394 .writepages = ext4_writepages,
3395 .write_begin = ext4_da_write_begin,
3396 .write_end = ext4_da_write_end,
3398 .invalidatepage = ext4_da_invalidatepage,
3399 .releasepage = ext4_releasepage,
3400 .direct_IO = ext4_direct_IO,
3401 .migratepage = buffer_migrate_page,
3402 .is_partially_uptodate = block_is_partially_uptodate,
3403 .error_remove_page = generic_error_remove_page,
3406 void ext4_set_aops(struct inode *inode)
3408 switch (ext4_inode_journal_mode(inode)) {
3409 case EXT4_INODE_ORDERED_DATA_MODE:
3410 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3412 case EXT4_INODE_WRITEBACK_DATA_MODE:
3413 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3415 case EXT4_INODE_JOURNAL_DATA_MODE:
3416 inode->i_mapping->a_ops = &ext4_journalled_aops;
3421 if (test_opt(inode->i_sb, DELALLOC))
3422 inode->i_mapping->a_ops = &ext4_da_aops;
3424 inode->i_mapping->a_ops = &ext4_aops;
3427 static int __ext4_block_zero_page_range(handle_t *handle,
3428 struct address_space *mapping, loff_t from, loff_t length)
3430 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3431 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3432 unsigned blocksize, pos;
3434 struct inode *inode = mapping->host;
3435 struct buffer_head *bh;
3439 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3440 mapping_gfp_constraint(mapping, ~__GFP_FS));
3444 blocksize = inode->i_sb->s_blocksize;
3446 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3448 if (!page_has_buffers(page))
3449 create_empty_buffers(page, blocksize, 0);
3451 /* Find the buffer that contains "offset" */
3452 bh = page_buffers(page);
3454 while (offset >= pos) {
3455 bh = bh->b_this_page;
3459 if (buffer_freed(bh)) {
3460 BUFFER_TRACE(bh, "freed: skip");
3463 if (!buffer_mapped(bh)) {
3464 BUFFER_TRACE(bh, "unmapped");
3465 ext4_get_block(inode, iblock, bh, 0);
3466 /* unmapped? It's a hole - nothing to do */
3467 if (!buffer_mapped(bh)) {
3468 BUFFER_TRACE(bh, "still unmapped");
3473 /* Ok, it's mapped. Make sure it's up-to-date */
3474 if (PageUptodate(page))
3475 set_buffer_uptodate(bh);
3477 if (!buffer_uptodate(bh)) {
3479 ll_rw_block(READ, 1, &bh);
3481 /* Uhhuh. Read error. Complain and punt. */
3482 if (!buffer_uptodate(bh))
3484 if (S_ISREG(inode->i_mode) &&
3485 ext4_encrypted_inode(inode)) {
3486 /* We expect the key to be set. */
3487 BUG_ON(!ext4_has_encryption_key(inode));
3488 BUG_ON(blocksize != PAGE_CACHE_SIZE);
3489 WARN_ON_ONCE(ext4_decrypt(page));
3492 if (ext4_should_journal_data(inode)) {
3493 BUFFER_TRACE(bh, "get write access");
3494 err = ext4_journal_get_write_access(handle, bh);
3498 zero_user(page, offset, length);
3499 BUFFER_TRACE(bh, "zeroed end of block");
3501 if (ext4_should_journal_data(inode)) {
3502 err = ext4_handle_dirty_metadata(handle, inode, bh);
3505 mark_buffer_dirty(bh);
3506 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3507 err = ext4_jbd2_file_inode(handle, inode);
3512 page_cache_release(page);
3517 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3518 * starting from file offset 'from'. The range to be zero'd must
3519 * be contained with in one block. If the specified range exceeds
3520 * the end of the block it will be shortened to end of the block
3521 * that cooresponds to 'from'
3523 static int ext4_block_zero_page_range(handle_t *handle,
3524 struct address_space *mapping, loff_t from, loff_t length)
3526 struct inode *inode = mapping->host;
3527 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3528 unsigned blocksize = inode->i_sb->s_blocksize;
3529 unsigned max = blocksize - (offset & (blocksize - 1));
3532 * correct length if it does not fall between
3533 * 'from' and the end of the block
3535 if (length > max || length < 0)
3539 return dax_zero_page_range(inode, from, length, ext4_get_block);
3540 return __ext4_block_zero_page_range(handle, mapping, from, length);
3544 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3545 * up to the end of the block which corresponds to `from'.
3546 * This required during truncate. We need to physically zero the tail end
3547 * of that block so it doesn't yield old data if the file is later grown.
3549 static int ext4_block_truncate_page(handle_t *handle,
3550 struct address_space *mapping, loff_t from)
3552 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3555 struct inode *inode = mapping->host;
3557 blocksize = inode->i_sb->s_blocksize;
3558 length = blocksize - (offset & (blocksize - 1));
3560 return ext4_block_zero_page_range(handle, mapping, from, length);
3563 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3564 loff_t lstart, loff_t length)
3566 struct super_block *sb = inode->i_sb;
3567 struct address_space *mapping = inode->i_mapping;
3568 unsigned partial_start, partial_end;
3569 ext4_fsblk_t start, end;
3570 loff_t byte_end = (lstart + length - 1);
3573 partial_start = lstart & (sb->s_blocksize - 1);
3574 partial_end = byte_end & (sb->s_blocksize - 1);
3576 start = lstart >> sb->s_blocksize_bits;
3577 end = byte_end >> sb->s_blocksize_bits;
3579 /* Handle partial zero within the single block */
3581 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3582 err = ext4_block_zero_page_range(handle, mapping,
3586 /* Handle partial zero out on the start of the range */
3587 if (partial_start) {
3588 err = ext4_block_zero_page_range(handle, mapping,
3589 lstart, sb->s_blocksize);
3593 /* Handle partial zero out on the end of the range */
3594 if (partial_end != sb->s_blocksize - 1)
3595 err = ext4_block_zero_page_range(handle, mapping,
3596 byte_end - partial_end,
3601 int ext4_can_truncate(struct inode *inode)
3603 if (S_ISREG(inode->i_mode))
3605 if (S_ISDIR(inode->i_mode))
3607 if (S_ISLNK(inode->i_mode))
3608 return !ext4_inode_is_fast_symlink(inode);
3613 * We have to make sure i_disksize gets properly updated before we truncate
3614 * page cache due to hole punching or zero range. Otherwise i_disksize update
3615 * can get lost as it may have been postponed to submission of writeback but
3616 * that will never happen after we truncate page cache.
3618 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3622 loff_t size = i_size_read(inode);
3624 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3625 if (offset > size || offset + len < size)
3628 if (EXT4_I(inode)->i_disksize >= size)
3631 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3633 return PTR_ERR(handle);
3634 ext4_update_i_disksize(inode, size);
3635 ext4_mark_inode_dirty(handle, inode);
3636 ext4_journal_stop(handle);
3642 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3643 * associated with the given offset and length
3645 * @inode: File inode
3646 * @offset: The offset where the hole will begin
3647 * @len: The length of the hole
3649 * Returns: 0 on success or negative on failure
3652 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3654 struct super_block *sb = inode->i_sb;
3655 ext4_lblk_t first_block, stop_block;
3656 struct address_space *mapping = inode->i_mapping;
3657 loff_t first_block_offset, last_block_offset;
3659 unsigned int credits;
3662 if (!S_ISREG(inode->i_mode))
3665 trace_ext4_punch_hole(inode, offset, length, 0);
3668 * Write out all dirty pages to avoid race conditions
3669 * Then release them.
3671 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3672 ret = filemap_write_and_wait_range(mapping, offset,
3673 offset + length - 1);
3678 mutex_lock(&inode->i_mutex);
3680 /* No need to punch hole beyond i_size */
3681 if (offset >= inode->i_size)
3685 * If the hole extends beyond i_size, set the hole
3686 * to end after the page that contains i_size
3688 if (offset + length > inode->i_size) {
3689 length = inode->i_size +
3690 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3694 if (offset & (sb->s_blocksize - 1) ||
3695 (offset + length) & (sb->s_blocksize - 1)) {
3697 * Attach jinode to inode for jbd2 if we do any zeroing of
3700 ret = ext4_inode_attach_jinode(inode);
3706 /* Wait all existing dio workers, newcomers will block on i_mutex */
3707 ext4_inode_block_unlocked_dio(inode);
3708 inode_dio_wait(inode);
3711 * Prevent page faults from reinstantiating pages we have released from
3714 down_write(&EXT4_I(inode)->i_mmap_sem);
3715 first_block_offset = round_up(offset, sb->s_blocksize);
3716 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3718 /* Now release the pages and zero block aligned part of pages*/
3719 if (last_block_offset > first_block_offset) {
3720 ret = ext4_update_disksize_before_punch(inode, offset, length);
3723 truncate_pagecache_range(inode, first_block_offset,
3727 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3728 credits = ext4_writepage_trans_blocks(inode);
3730 credits = ext4_blocks_for_truncate(inode);
3731 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3732 if (IS_ERR(handle)) {
3733 ret = PTR_ERR(handle);
3734 ext4_std_error(sb, ret);
3738 ret = ext4_zero_partial_blocks(handle, inode, offset,
3743 first_block = (offset + sb->s_blocksize - 1) >>
3744 EXT4_BLOCK_SIZE_BITS(sb);
3745 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3747 /* If there are no blocks to remove, return now */
3748 if (first_block >= stop_block)
3751 down_write(&EXT4_I(inode)->i_data_sem);
3752 ext4_discard_preallocations(inode);
3754 ret = ext4_es_remove_extent(inode, first_block,
3755 stop_block - first_block);
3757 up_write(&EXT4_I(inode)->i_data_sem);
3761 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3762 ret = ext4_ext_remove_space(inode, first_block,
3765 ret = ext4_ind_remove_space(handle, inode, first_block,
3768 up_write(&EXT4_I(inode)->i_data_sem);
3770 ext4_handle_sync(handle);
3772 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3773 ext4_mark_inode_dirty(handle, inode);
3775 ext4_journal_stop(handle);
3777 up_write(&EXT4_I(inode)->i_mmap_sem);
3778 ext4_inode_resume_unlocked_dio(inode);
3780 mutex_unlock(&inode->i_mutex);
3784 int ext4_inode_attach_jinode(struct inode *inode)
3786 struct ext4_inode_info *ei = EXT4_I(inode);
3787 struct jbd2_inode *jinode;
3789 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3792 jinode = jbd2_alloc_inode(GFP_KERNEL);
3793 spin_lock(&inode->i_lock);
3796 spin_unlock(&inode->i_lock);
3799 ei->jinode = jinode;
3800 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3803 spin_unlock(&inode->i_lock);
3804 if (unlikely(jinode != NULL))
3805 jbd2_free_inode(jinode);
3812 * We block out ext4_get_block() block instantiations across the entire
3813 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3814 * simultaneously on behalf of the same inode.
3816 * As we work through the truncate and commit bits of it to the journal there
3817 * is one core, guiding principle: the file's tree must always be consistent on
3818 * disk. We must be able to restart the truncate after a crash.
3820 * The file's tree may be transiently inconsistent in memory (although it
3821 * probably isn't), but whenever we close off and commit a journal transaction,
3822 * the contents of (the filesystem + the journal) must be consistent and
3823 * restartable. It's pretty simple, really: bottom up, right to left (although
3824 * left-to-right works OK too).
3826 * Note that at recovery time, journal replay occurs *before* the restart of
3827 * truncate against the orphan inode list.
3829 * The committed inode has the new, desired i_size (which is the same as
3830 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3831 * that this inode's truncate did not complete and it will again call
3832 * ext4_truncate() to have another go. So there will be instantiated blocks
3833 * to the right of the truncation point in a crashed ext4 filesystem. But
3834 * that's fine - as long as they are linked from the inode, the post-crash
3835 * ext4_truncate() run will find them and release them.
3837 void ext4_truncate(struct inode *inode)
3839 struct ext4_inode_info *ei = EXT4_I(inode);
3840 unsigned int credits;
3842 struct address_space *mapping = inode->i_mapping;
3845 * There is a possibility that we're either freeing the inode
3846 * or it's a completely new inode. In those cases we might not
3847 * have i_mutex locked because it's not necessary.
3849 if (!(inode->i_state & (I_NEW|I_FREEING)))
3850 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3851 trace_ext4_truncate_enter(inode);
3853 if (!ext4_can_truncate(inode))
3856 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3858 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3859 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3861 if (ext4_has_inline_data(inode)) {
3864 ext4_inline_data_truncate(inode, &has_inline);
3869 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3870 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3871 if (ext4_inode_attach_jinode(inode) < 0)
3875 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3876 credits = ext4_writepage_trans_blocks(inode);
3878 credits = ext4_blocks_for_truncate(inode);
3880 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3881 if (IS_ERR(handle)) {
3882 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3886 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3887 ext4_block_truncate_page(handle, mapping, inode->i_size);
3890 * We add the inode to the orphan list, so that if this
3891 * truncate spans multiple transactions, and we crash, we will
3892 * resume the truncate when the filesystem recovers. It also
3893 * marks the inode dirty, to catch the new size.
3895 * Implication: the file must always be in a sane, consistent
3896 * truncatable state while each transaction commits.
3898 if (ext4_orphan_add(handle, inode))
3901 down_write(&EXT4_I(inode)->i_data_sem);
3903 ext4_discard_preallocations(inode);
3905 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3906 ext4_ext_truncate(handle, inode);
3908 ext4_ind_truncate(handle, inode);
3910 up_write(&ei->i_data_sem);
3913 ext4_handle_sync(handle);
3917 * If this was a simple ftruncate() and the file will remain alive,
3918 * then we need to clear up the orphan record which we created above.
3919 * However, if this was a real unlink then we were called by
3920 * ext4_evict_inode(), and we allow that function to clean up the
3921 * orphan info for us.
3924 ext4_orphan_del(handle, inode);
3926 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3927 ext4_mark_inode_dirty(handle, inode);
3928 ext4_journal_stop(handle);
3930 trace_ext4_truncate_exit(inode);
3934 * ext4_get_inode_loc returns with an extra refcount against the inode's
3935 * underlying buffer_head on success. If 'in_mem' is true, we have all
3936 * data in memory that is needed to recreate the on-disk version of this
3939 static int __ext4_get_inode_loc(struct inode *inode,
3940 struct ext4_iloc *iloc, int in_mem)
3942 struct ext4_group_desc *gdp;
3943 struct buffer_head *bh;
3944 struct super_block *sb = inode->i_sb;
3946 int inodes_per_block, inode_offset;
3949 if (!ext4_valid_inum(sb, inode->i_ino))
3950 return -EFSCORRUPTED;
3952 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3953 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3958 * Figure out the offset within the block group inode table
3960 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3961 inode_offset = ((inode->i_ino - 1) %
3962 EXT4_INODES_PER_GROUP(sb));
3963 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3964 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3966 bh = sb_getblk(sb, block);
3969 if (!buffer_uptodate(bh)) {
3973 * If the buffer has the write error flag, we have failed
3974 * to write out another inode in the same block. In this
3975 * case, we don't have to read the block because we may
3976 * read the old inode data successfully.
3978 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3979 set_buffer_uptodate(bh);
3981 if (buffer_uptodate(bh)) {
3982 /* someone brought it uptodate while we waited */
3988 * If we have all information of the inode in memory and this
3989 * is the only valid inode in the block, we need not read the
3993 struct buffer_head *bitmap_bh;
3996 start = inode_offset & ~(inodes_per_block - 1);
3998 /* Is the inode bitmap in cache? */
3999 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4000 if (unlikely(!bitmap_bh))
4004 * If the inode bitmap isn't in cache then the
4005 * optimisation may end up performing two reads instead
4006 * of one, so skip it.
4008 if (!buffer_uptodate(bitmap_bh)) {
4012 for (i = start; i < start + inodes_per_block; i++) {
4013 if (i == inode_offset)
4015 if (ext4_test_bit(i, bitmap_bh->b_data))
4019 if (i == start + inodes_per_block) {
4020 /* all other inodes are free, so skip I/O */
4021 memset(bh->b_data, 0, bh->b_size);
4022 set_buffer_uptodate(bh);
4030 * If we need to do any I/O, try to pre-readahead extra
4031 * blocks from the inode table.
4033 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4034 ext4_fsblk_t b, end, table;
4036 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4038 table = ext4_inode_table(sb, gdp);
4039 /* s_inode_readahead_blks is always a power of 2 */
4040 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4044 num = EXT4_INODES_PER_GROUP(sb);
4045 if (ext4_has_group_desc_csum(sb))
4046 num -= ext4_itable_unused_count(sb, gdp);
4047 table += num / inodes_per_block;
4051 sb_breadahead(sb, b++);
4055 * There are other valid inodes in the buffer, this inode
4056 * has in-inode xattrs, or we don't have this inode in memory.
4057 * Read the block from disk.
4059 trace_ext4_load_inode(inode);
4061 bh->b_end_io = end_buffer_read_sync;
4062 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4064 if (!buffer_uptodate(bh)) {
4065 EXT4_ERROR_INODE_BLOCK(inode, block,
4066 "unable to read itable block");
4076 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4078 /* We have all inode data except xattrs in memory here. */
4079 return __ext4_get_inode_loc(inode, iloc,
4080 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4083 void ext4_set_inode_flags(struct inode *inode)
4085 unsigned int flags = EXT4_I(inode)->i_flags;
4086 unsigned int new_fl = 0;
4088 if (flags & EXT4_SYNC_FL)
4090 if (flags & EXT4_APPEND_FL)
4092 if (flags & EXT4_IMMUTABLE_FL)
4093 new_fl |= S_IMMUTABLE;
4094 if (flags & EXT4_NOATIME_FL)
4095 new_fl |= S_NOATIME;
4096 if (flags & EXT4_DIRSYNC_FL)
4097 new_fl |= S_DIRSYNC;
4098 if (test_opt(inode->i_sb, DAX))
4100 inode_set_flags(inode, new_fl,
4101 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4104 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4105 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4107 unsigned int vfs_fl;
4108 unsigned long old_fl, new_fl;
4111 vfs_fl = ei->vfs_inode.i_flags;
4112 old_fl = ei->i_flags;
4113 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4114 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4116 if (vfs_fl & S_SYNC)
4117 new_fl |= EXT4_SYNC_FL;
4118 if (vfs_fl & S_APPEND)
4119 new_fl |= EXT4_APPEND_FL;
4120 if (vfs_fl & S_IMMUTABLE)
4121 new_fl |= EXT4_IMMUTABLE_FL;
4122 if (vfs_fl & S_NOATIME)
4123 new_fl |= EXT4_NOATIME_FL;
4124 if (vfs_fl & S_DIRSYNC)
4125 new_fl |= EXT4_DIRSYNC_FL;
4126 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4129 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4130 struct ext4_inode_info *ei)
4133 struct inode *inode = &(ei->vfs_inode);
4134 struct super_block *sb = inode->i_sb;
4136 if (ext4_has_feature_huge_file(sb)) {
4137 /* we are using combined 48 bit field */
4138 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4139 le32_to_cpu(raw_inode->i_blocks_lo);
4140 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4141 /* i_blocks represent file system block size */
4142 return i_blocks << (inode->i_blkbits - 9);
4147 return le32_to_cpu(raw_inode->i_blocks_lo);
4151 static inline void ext4_iget_extra_inode(struct inode *inode,
4152 struct ext4_inode *raw_inode,
4153 struct ext4_inode_info *ei)
4155 __le32 *magic = (void *)raw_inode +
4156 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4157 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4158 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4159 ext4_find_inline_data_nolock(inode);
4161 EXT4_I(inode)->i_inline_off = 0;
4164 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4166 struct ext4_iloc iloc;
4167 struct ext4_inode *raw_inode;
4168 struct ext4_inode_info *ei;
4169 struct inode *inode;
4170 journal_t *journal = EXT4_SB(sb)->s_journal;
4176 inode = iget_locked(sb, ino);
4178 return ERR_PTR(-ENOMEM);
4179 if (!(inode->i_state & I_NEW))
4185 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4188 raw_inode = ext4_raw_inode(&iloc);
4190 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4191 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4192 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4193 EXT4_INODE_SIZE(inode->i_sb)) {
4194 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4195 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4196 EXT4_INODE_SIZE(inode->i_sb));
4197 ret = -EFSCORRUPTED;
4201 ei->i_extra_isize = 0;
4203 /* Precompute checksum seed for inode metadata */
4204 if (ext4_has_metadata_csum(sb)) {
4205 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4207 __le32 inum = cpu_to_le32(inode->i_ino);
4208 __le32 gen = raw_inode->i_generation;
4209 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4211 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4215 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4216 EXT4_ERROR_INODE(inode, "checksum invalid");
4221 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4222 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4223 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4224 if (!(test_opt(inode->i_sb, NO_UID32))) {
4225 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4226 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4228 i_uid_write(inode, i_uid);
4229 i_gid_write(inode, i_gid);
4230 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4232 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4233 ei->i_inline_off = 0;
4234 ei->i_dir_start_lookup = 0;
4235 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4236 /* We now have enough fields to check if the inode was active or not.
4237 * This is needed because nfsd might try to access dead inodes
4238 * the test is that same one that e2fsck uses
4239 * NeilBrown 1999oct15
4241 if (inode->i_nlink == 0) {
4242 if ((inode->i_mode == 0 ||
4243 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4244 ino != EXT4_BOOT_LOADER_INO) {
4245 /* this inode is deleted */
4249 /* The only unlinked inodes we let through here have
4250 * valid i_mode and are being read by the orphan
4251 * recovery code: that's fine, we're about to complete
4252 * the process of deleting those.
4253 * OR it is the EXT4_BOOT_LOADER_INO which is
4254 * not initialized on a new filesystem. */
4256 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4257 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4258 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4259 if (ext4_has_feature_64bit(sb))
4261 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4262 inode->i_size = ext4_isize(raw_inode);
4263 ei->i_disksize = inode->i_size;
4265 ei->i_reserved_quota = 0;
4267 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4268 ei->i_block_group = iloc.block_group;
4269 ei->i_last_alloc_group = ~0;
4271 * NOTE! The in-memory inode i_data array is in little-endian order
4272 * even on big-endian machines: we do NOT byteswap the block numbers!
4274 for (block = 0; block < EXT4_N_BLOCKS; block++)
4275 ei->i_data[block] = raw_inode->i_block[block];
4276 INIT_LIST_HEAD(&ei->i_orphan);
4279 * Set transaction id's of transactions that have to be committed
4280 * to finish f[data]sync. We set them to currently running transaction
4281 * as we cannot be sure that the inode or some of its metadata isn't
4282 * part of the transaction - the inode could have been reclaimed and
4283 * now it is reread from disk.
4286 transaction_t *transaction;
4289 read_lock(&journal->j_state_lock);
4290 if (journal->j_running_transaction)
4291 transaction = journal->j_running_transaction;
4293 transaction = journal->j_committing_transaction;
4295 tid = transaction->t_tid;
4297 tid = journal->j_commit_sequence;
4298 read_unlock(&journal->j_state_lock);
4299 ei->i_sync_tid = tid;
4300 ei->i_datasync_tid = tid;
4303 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4304 if (ei->i_extra_isize == 0) {
4305 /* The extra space is currently unused. Use it. */
4306 ei->i_extra_isize = sizeof(struct ext4_inode) -
4307 EXT4_GOOD_OLD_INODE_SIZE;
4309 ext4_iget_extra_inode(inode, raw_inode, ei);
4313 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4314 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4315 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4316 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4318 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4319 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4320 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4321 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4323 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4328 if (ei->i_file_acl &&
4329 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4330 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4332 ret = -EFSCORRUPTED;
4334 } else if (!ext4_has_inline_data(inode)) {
4335 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4336 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4337 (S_ISLNK(inode->i_mode) &&
4338 !ext4_inode_is_fast_symlink(inode))))
4339 /* Validate extent which is part of inode */
4340 ret = ext4_ext_check_inode(inode);
4341 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4342 (S_ISLNK(inode->i_mode) &&
4343 !ext4_inode_is_fast_symlink(inode))) {
4344 /* Validate block references which are part of inode */
4345 ret = ext4_ind_check_inode(inode);
4351 if (S_ISREG(inode->i_mode)) {
4352 inode->i_op = &ext4_file_inode_operations;
4353 inode->i_fop = &ext4_file_operations;
4354 ext4_set_aops(inode);
4355 } else if (S_ISDIR(inode->i_mode)) {
4356 inode->i_op = &ext4_dir_inode_operations;
4357 inode->i_fop = &ext4_dir_operations;
4358 } else if (S_ISLNK(inode->i_mode)) {
4359 if (ext4_encrypted_inode(inode)) {
4360 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4361 ext4_set_aops(inode);
4362 } else if (ext4_inode_is_fast_symlink(inode)) {
4363 inode->i_link = (char *)ei->i_data;
4364 inode->i_op = &ext4_fast_symlink_inode_operations;
4365 nd_terminate_link(ei->i_data, inode->i_size,
4366 sizeof(ei->i_data) - 1);
4368 inode->i_op = &ext4_symlink_inode_operations;
4369 ext4_set_aops(inode);
4371 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4372 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4373 inode->i_op = &ext4_special_inode_operations;
4374 if (raw_inode->i_block[0])
4375 init_special_inode(inode, inode->i_mode,
4376 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4378 init_special_inode(inode, inode->i_mode,
4379 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4380 } else if (ino == EXT4_BOOT_LOADER_INO) {
4381 make_bad_inode(inode);
4383 ret = -EFSCORRUPTED;
4384 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4388 ext4_set_inode_flags(inode);
4389 unlock_new_inode(inode);
4395 return ERR_PTR(ret);
4398 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4400 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4401 return ERR_PTR(-EFSCORRUPTED);
4402 return ext4_iget(sb, ino);
4405 static int ext4_inode_blocks_set(handle_t *handle,
4406 struct ext4_inode *raw_inode,
4407 struct ext4_inode_info *ei)
4409 struct inode *inode = &(ei->vfs_inode);
4410 u64 i_blocks = inode->i_blocks;
4411 struct super_block *sb = inode->i_sb;
4413 if (i_blocks <= ~0U) {
4415 * i_blocks can be represented in a 32 bit variable
4416 * as multiple of 512 bytes
4418 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4419 raw_inode->i_blocks_high = 0;
4420 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4423 if (!ext4_has_feature_huge_file(sb))
4426 if (i_blocks <= 0xffffffffffffULL) {
4428 * i_blocks can be represented in a 48 bit variable
4429 * as multiple of 512 bytes
4431 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4432 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4433 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4435 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4436 /* i_block is stored in file system block size */
4437 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4438 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4439 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4444 struct other_inode {
4445 unsigned long orig_ino;
4446 struct ext4_inode *raw_inode;
4449 static int other_inode_match(struct inode * inode, unsigned long ino,
4452 struct other_inode *oi = (struct other_inode *) data;
4454 if ((inode->i_ino != ino) ||
4455 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4456 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4457 ((inode->i_state & I_DIRTY_TIME) == 0))
4459 spin_lock(&inode->i_lock);
4460 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4461 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4462 (inode->i_state & I_DIRTY_TIME)) {
4463 struct ext4_inode_info *ei = EXT4_I(inode);
4465 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4466 spin_unlock(&inode->i_lock);
4468 spin_lock(&ei->i_raw_lock);
4469 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4470 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4471 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4472 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4473 spin_unlock(&ei->i_raw_lock);
4474 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4477 spin_unlock(&inode->i_lock);
4482 * Opportunistically update the other time fields for other inodes in
4483 * the same inode table block.
4485 static void ext4_update_other_inodes_time(struct super_block *sb,
4486 unsigned long orig_ino, char *buf)
4488 struct other_inode oi;
4490 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4491 int inode_size = EXT4_INODE_SIZE(sb);
4493 oi.orig_ino = orig_ino;
4495 * Calculate the first inode in the inode table block. Inode
4496 * numbers are one-based. That is, the first inode in a block
4497 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4499 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4500 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4501 if (ino == orig_ino)
4503 oi.raw_inode = (struct ext4_inode *) buf;
4504 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4509 * Post the struct inode info into an on-disk inode location in the
4510 * buffer-cache. This gobbles the caller's reference to the
4511 * buffer_head in the inode location struct.
4513 * The caller must have write access to iloc->bh.
4515 static int ext4_do_update_inode(handle_t *handle,
4516 struct inode *inode,
4517 struct ext4_iloc *iloc)
4519 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4520 struct ext4_inode_info *ei = EXT4_I(inode);
4521 struct buffer_head *bh = iloc->bh;
4522 struct super_block *sb = inode->i_sb;
4523 int err = 0, rc, block;
4524 int need_datasync = 0, set_large_file = 0;
4528 spin_lock(&ei->i_raw_lock);
4530 /* For fields not tracked in the in-memory inode,
4531 * initialise them to zero for new inodes. */
4532 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4533 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4535 ext4_get_inode_flags(ei);
4536 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4537 i_uid = i_uid_read(inode);
4538 i_gid = i_gid_read(inode);
4539 if (!(test_opt(inode->i_sb, NO_UID32))) {
4540 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4541 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4543 * Fix up interoperability with old kernels. Otherwise, old inodes get
4544 * re-used with the upper 16 bits of the uid/gid intact
4547 raw_inode->i_uid_high =
4548 cpu_to_le16(high_16_bits(i_uid));
4549 raw_inode->i_gid_high =
4550 cpu_to_le16(high_16_bits(i_gid));
4552 raw_inode->i_uid_high = 0;
4553 raw_inode->i_gid_high = 0;
4556 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4557 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4558 raw_inode->i_uid_high = 0;
4559 raw_inode->i_gid_high = 0;
4561 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4563 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4564 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4565 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4566 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4568 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4570 spin_unlock(&ei->i_raw_lock);
4573 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4574 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4575 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4576 raw_inode->i_file_acl_high =
4577 cpu_to_le16(ei->i_file_acl >> 32);
4578 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4579 if (ei->i_disksize != ext4_isize(raw_inode)) {
4580 ext4_isize_set(raw_inode, ei->i_disksize);
4583 if (ei->i_disksize > 0x7fffffffULL) {
4584 if (!ext4_has_feature_large_file(sb) ||
4585 EXT4_SB(sb)->s_es->s_rev_level ==
4586 cpu_to_le32(EXT4_GOOD_OLD_REV))
4589 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4590 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4591 if (old_valid_dev(inode->i_rdev)) {
4592 raw_inode->i_block[0] =
4593 cpu_to_le32(old_encode_dev(inode->i_rdev));
4594 raw_inode->i_block[1] = 0;
4596 raw_inode->i_block[0] = 0;
4597 raw_inode->i_block[1] =
4598 cpu_to_le32(new_encode_dev(inode->i_rdev));
4599 raw_inode->i_block[2] = 0;
4601 } else if (!ext4_has_inline_data(inode)) {
4602 for (block = 0; block < EXT4_N_BLOCKS; block++)
4603 raw_inode->i_block[block] = ei->i_data[block];
4606 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4607 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4608 if (ei->i_extra_isize) {
4609 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4610 raw_inode->i_version_hi =
4611 cpu_to_le32(inode->i_version >> 32);
4612 raw_inode->i_extra_isize =
4613 cpu_to_le16(ei->i_extra_isize);
4616 ext4_inode_csum_set(inode, raw_inode, ei);
4617 spin_unlock(&ei->i_raw_lock);
4618 if (inode->i_sb->s_flags & MS_LAZYTIME)
4619 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4622 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4623 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4626 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4627 if (set_large_file) {
4628 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4629 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4632 ext4_update_dynamic_rev(sb);
4633 ext4_set_feature_large_file(sb);
4634 ext4_handle_sync(handle);
4635 err = ext4_handle_dirty_super(handle, sb);
4637 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4640 ext4_std_error(inode->i_sb, err);
4645 * ext4_write_inode()
4647 * We are called from a few places:
4649 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4650 * Here, there will be no transaction running. We wait for any running
4651 * transaction to commit.
4653 * - Within flush work (sys_sync(), kupdate and such).
4654 * We wait on commit, if told to.
4656 * - Within iput_final() -> write_inode_now()
4657 * We wait on commit, if told to.
4659 * In all cases it is actually safe for us to return without doing anything,
4660 * because the inode has been copied into a raw inode buffer in
4661 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4664 * Note that we are absolutely dependent upon all inode dirtiers doing the
4665 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4666 * which we are interested.
4668 * It would be a bug for them to not do this. The code:
4670 * mark_inode_dirty(inode)
4672 * inode->i_size = expr;
4674 * is in error because write_inode() could occur while `stuff()' is running,
4675 * and the new i_size will be lost. Plus the inode will no longer be on the
4676 * superblock's dirty inode list.
4678 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4682 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4685 if (EXT4_SB(inode->i_sb)->s_journal) {
4686 if (ext4_journal_current_handle()) {
4687 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4693 * No need to force transaction in WB_SYNC_NONE mode. Also
4694 * ext4_sync_fs() will force the commit after everything is
4697 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4700 err = ext4_force_commit(inode->i_sb);
4702 struct ext4_iloc iloc;
4704 err = __ext4_get_inode_loc(inode, &iloc, 0);
4708 * sync(2) will flush the whole buffer cache. No need to do
4709 * it here separately for each inode.
4711 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4712 sync_dirty_buffer(iloc.bh);
4713 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4714 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4715 "IO error syncing inode");
4724 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4725 * buffers that are attached to a page stradding i_size and are undergoing
4726 * commit. In that case we have to wait for commit to finish and try again.
4728 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4732 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4733 tid_t commit_tid = 0;
4736 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4738 * All buffers in the last page remain valid? Then there's nothing to
4739 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4742 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4745 page = find_lock_page(inode->i_mapping,
4746 inode->i_size >> PAGE_CACHE_SHIFT);
4749 ret = __ext4_journalled_invalidatepage(page, offset,
4750 PAGE_CACHE_SIZE - offset);
4752 page_cache_release(page);
4756 read_lock(&journal->j_state_lock);
4757 if (journal->j_committing_transaction)
4758 commit_tid = journal->j_committing_transaction->t_tid;
4759 read_unlock(&journal->j_state_lock);
4761 jbd2_log_wait_commit(journal, commit_tid);
4768 * Called from notify_change.
4770 * We want to trap VFS attempts to truncate the file as soon as
4771 * possible. In particular, we want to make sure that when the VFS
4772 * shrinks i_size, we put the inode on the orphan list and modify
4773 * i_disksize immediately, so that during the subsequent flushing of
4774 * dirty pages and freeing of disk blocks, we can guarantee that any
4775 * commit will leave the blocks being flushed in an unused state on
4776 * disk. (On recovery, the inode will get truncated and the blocks will
4777 * be freed, so we have a strong guarantee that no future commit will
4778 * leave these blocks visible to the user.)
4780 * Another thing we have to assure is that if we are in ordered mode
4781 * and inode is still attached to the committing transaction, we must
4782 * we start writeout of all the dirty pages which are being truncated.
4783 * This way we are sure that all the data written in the previous
4784 * transaction are already on disk (truncate waits for pages under
4787 * Called with inode->i_mutex down.
4789 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4791 struct inode *inode = d_inode(dentry);
4794 const unsigned int ia_valid = attr->ia_valid;
4796 error = inode_change_ok(inode, attr);
4800 if (is_quota_modification(inode, attr)) {
4801 error = dquot_initialize(inode);
4805 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4806 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4809 /* (user+group)*(old+new) structure, inode write (sb,
4810 * inode block, ? - but truncate inode update has it) */
4811 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4812 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4813 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4814 if (IS_ERR(handle)) {
4815 error = PTR_ERR(handle);
4818 error = dquot_transfer(inode, attr);
4820 ext4_journal_stop(handle);
4823 /* Update corresponding info in inode so that everything is in
4824 * one transaction */
4825 if (attr->ia_valid & ATTR_UID)
4826 inode->i_uid = attr->ia_uid;
4827 if (attr->ia_valid & ATTR_GID)
4828 inode->i_gid = attr->ia_gid;
4829 error = ext4_mark_inode_dirty(handle, inode);
4830 ext4_journal_stop(handle);
4833 if (attr->ia_valid & ATTR_SIZE) {
4835 loff_t oldsize = inode->i_size;
4836 int shrink = (attr->ia_size <= inode->i_size);
4838 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4839 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4841 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4844 if (!S_ISREG(inode->i_mode))
4847 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4848 inode_inc_iversion(inode);
4850 if (ext4_should_order_data(inode) &&
4851 (attr->ia_size < inode->i_size)) {
4852 error = ext4_begin_ordered_truncate(inode,
4857 if (attr->ia_size != inode->i_size) {
4858 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4859 if (IS_ERR(handle)) {
4860 error = PTR_ERR(handle);
4863 if (ext4_handle_valid(handle) && shrink) {
4864 error = ext4_orphan_add(handle, inode);
4868 * Update c/mtime on truncate up, ext4_truncate() will
4869 * update c/mtime in shrink case below
4872 inode->i_mtime = ext4_current_time(inode);
4873 inode->i_ctime = inode->i_mtime;
4875 down_write(&EXT4_I(inode)->i_data_sem);
4876 EXT4_I(inode)->i_disksize = attr->ia_size;
4877 rc = ext4_mark_inode_dirty(handle, inode);
4881 * We have to update i_size under i_data_sem together
4882 * with i_disksize to avoid races with writeback code
4883 * running ext4_wb_update_i_disksize().
4886 i_size_write(inode, attr->ia_size);
4887 up_write(&EXT4_I(inode)->i_data_sem);
4888 ext4_journal_stop(handle);
4891 ext4_orphan_del(NULL, inode);
4896 pagecache_isize_extended(inode, oldsize, inode->i_size);
4899 * Blocks are going to be removed from the inode. Wait
4900 * for dio in flight. Temporarily disable
4901 * dioread_nolock to prevent livelock.
4904 if (!ext4_should_journal_data(inode)) {
4905 ext4_inode_block_unlocked_dio(inode);
4906 inode_dio_wait(inode);
4907 ext4_inode_resume_unlocked_dio(inode);
4909 ext4_wait_for_tail_page_commit(inode);
4911 down_write(&EXT4_I(inode)->i_mmap_sem);
4913 * Truncate pagecache after we've waited for commit
4914 * in data=journal mode to make pages freeable.
4916 truncate_pagecache(inode, inode->i_size);
4918 ext4_truncate(inode);
4919 up_write(&EXT4_I(inode)->i_mmap_sem);
4923 setattr_copy(inode, attr);
4924 mark_inode_dirty(inode);
4928 * If the call to ext4_truncate failed to get a transaction handle at
4929 * all, we need to clean up the in-core orphan list manually.
4931 if (orphan && inode->i_nlink)
4932 ext4_orphan_del(NULL, inode);
4934 if (!rc && (ia_valid & ATTR_MODE))
4935 rc = posix_acl_chmod(inode, inode->i_mode);
4938 ext4_std_error(inode->i_sb, error);
4944 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4947 struct inode *inode;
4948 unsigned long long delalloc_blocks;
4950 inode = d_inode(dentry);
4951 generic_fillattr(inode, stat);
4954 * If there is inline data in the inode, the inode will normally not
4955 * have data blocks allocated (it may have an external xattr block).
4956 * Report at least one sector for such files, so tools like tar, rsync,
4957 * others doen't incorrectly think the file is completely sparse.
4959 if (unlikely(ext4_has_inline_data(inode)))
4960 stat->blocks += (stat->size + 511) >> 9;
4963 * We can't update i_blocks if the block allocation is delayed
4964 * otherwise in the case of system crash before the real block
4965 * allocation is done, we will have i_blocks inconsistent with
4966 * on-disk file blocks.
4967 * We always keep i_blocks updated together with real
4968 * allocation. But to not confuse with user, stat
4969 * will return the blocks that include the delayed allocation
4970 * blocks for this file.
4972 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4973 EXT4_I(inode)->i_reserved_data_blocks);
4974 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
4978 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4981 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4982 return ext4_ind_trans_blocks(inode, lblocks);
4983 return ext4_ext_index_trans_blocks(inode, pextents);
4987 * Account for index blocks, block groups bitmaps and block group
4988 * descriptor blocks if modify datablocks and index blocks
4989 * worse case, the indexs blocks spread over different block groups
4991 * If datablocks are discontiguous, they are possible to spread over
4992 * different block groups too. If they are contiguous, with flexbg,
4993 * they could still across block group boundary.
4995 * Also account for superblock, inode, quota and xattr blocks
4997 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5000 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5006 * How many index blocks need to touch to map @lblocks logical blocks
5007 * to @pextents physical extents?
5009 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5014 * Now let's see how many group bitmaps and group descriptors need
5017 groups = idxblocks + pextents;
5019 if (groups > ngroups)
5021 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5022 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5024 /* bitmaps and block group descriptor blocks */
5025 ret += groups + gdpblocks;
5027 /* Blocks for super block, inode, quota and xattr blocks */
5028 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5034 * Calculate the total number of credits to reserve to fit
5035 * the modification of a single pages into a single transaction,
5036 * which may include multiple chunks of block allocations.
5038 * This could be called via ext4_write_begin()
5040 * We need to consider the worse case, when
5041 * one new block per extent.
5043 int ext4_writepage_trans_blocks(struct inode *inode)
5045 int bpp = ext4_journal_blocks_per_page(inode);
5048 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5050 /* Account for data blocks for journalled mode */
5051 if (ext4_should_journal_data(inode))
5057 * Calculate the journal credits for a chunk of data modification.
5059 * This is called from DIO, fallocate or whoever calling
5060 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5062 * journal buffers for data blocks are not included here, as DIO
5063 * and fallocate do no need to journal data buffers.
5065 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5067 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5071 * The caller must have previously called ext4_reserve_inode_write().
5072 * Give this, we know that the caller already has write access to iloc->bh.
5074 int ext4_mark_iloc_dirty(handle_t *handle,
5075 struct inode *inode, struct ext4_iloc *iloc)
5079 if (IS_I_VERSION(inode))
5080 inode_inc_iversion(inode);
5082 /* the do_update_inode consumes one bh->b_count */
5085 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5086 err = ext4_do_update_inode(handle, inode, iloc);
5092 * On success, We end up with an outstanding reference count against
5093 * iloc->bh. This _must_ be cleaned up later.
5097 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5098 struct ext4_iloc *iloc)
5102 err = ext4_get_inode_loc(inode, iloc);
5104 BUFFER_TRACE(iloc->bh, "get_write_access");
5105 err = ext4_journal_get_write_access(handle, iloc->bh);
5111 ext4_std_error(inode->i_sb, err);
5116 * Expand an inode by new_extra_isize bytes.
5117 * Returns 0 on success or negative error number on failure.
5119 static int ext4_expand_extra_isize(struct inode *inode,
5120 unsigned int new_extra_isize,
5121 struct ext4_iloc iloc,
5124 struct ext4_inode *raw_inode;
5125 struct ext4_xattr_ibody_header *header;
5127 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5130 raw_inode = ext4_raw_inode(&iloc);
5132 header = IHDR(inode, raw_inode);
5134 /* No extended attributes present */
5135 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5136 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5137 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5139 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5143 /* try to expand with EAs present */
5144 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5149 * What we do here is to mark the in-core inode as clean with respect to inode
5150 * dirtiness (it may still be data-dirty).
5151 * This means that the in-core inode may be reaped by prune_icache
5152 * without having to perform any I/O. This is a very good thing,
5153 * because *any* task may call prune_icache - even ones which
5154 * have a transaction open against a different journal.
5156 * Is this cheating? Not really. Sure, we haven't written the
5157 * inode out, but prune_icache isn't a user-visible syncing function.
5158 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5159 * we start and wait on commits.
5161 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5163 struct ext4_iloc iloc;
5164 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5165 static unsigned int mnt_count;
5169 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5170 err = ext4_reserve_inode_write(handle, inode, &iloc);
5173 if (ext4_handle_valid(handle) &&
5174 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5175 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5177 * We need extra buffer credits since we may write into EA block
5178 * with this same handle. If journal_extend fails, then it will
5179 * only result in a minor loss of functionality for that inode.
5180 * If this is felt to be critical, then e2fsck should be run to
5181 * force a large enough s_min_extra_isize.
5183 if ((jbd2_journal_extend(handle,
5184 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5185 ret = ext4_expand_extra_isize(inode,
5186 sbi->s_want_extra_isize,
5189 ext4_set_inode_state(inode,
5190 EXT4_STATE_NO_EXPAND);
5192 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5193 ext4_warning(inode->i_sb,
5194 "Unable to expand inode %lu. Delete"
5195 " some EAs or run e2fsck.",
5198 le16_to_cpu(sbi->s_es->s_mnt_count);
5203 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5207 * ext4_dirty_inode() is called from __mark_inode_dirty()
5209 * We're really interested in the case where a file is being extended.
5210 * i_size has been changed by generic_commit_write() and we thus need
5211 * to include the updated inode in the current transaction.
5213 * Also, dquot_alloc_block() will always dirty the inode when blocks
5214 * are allocated to the file.
5216 * If the inode is marked synchronous, we don't honour that here - doing
5217 * so would cause a commit on atime updates, which we don't bother doing.
5218 * We handle synchronous inodes at the highest possible level.
5220 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5221 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5222 * to copy into the on-disk inode structure are the timestamp files.
5224 void ext4_dirty_inode(struct inode *inode, int flags)
5228 if (flags == I_DIRTY_TIME)
5230 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5234 ext4_mark_inode_dirty(handle, inode);
5236 ext4_journal_stop(handle);
5243 * Bind an inode's backing buffer_head into this transaction, to prevent
5244 * it from being flushed to disk early. Unlike
5245 * ext4_reserve_inode_write, this leaves behind no bh reference and
5246 * returns no iloc structure, so the caller needs to repeat the iloc
5247 * lookup to mark the inode dirty later.
5249 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5251 struct ext4_iloc iloc;
5255 err = ext4_get_inode_loc(inode, &iloc);
5257 BUFFER_TRACE(iloc.bh, "get_write_access");
5258 err = jbd2_journal_get_write_access(handle, iloc.bh);
5260 err = ext4_handle_dirty_metadata(handle,
5266 ext4_std_error(inode->i_sb, err);
5271 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5278 * We have to be very careful here: changing a data block's
5279 * journaling status dynamically is dangerous. If we write a
5280 * data block to the journal, change the status and then delete
5281 * that block, we risk forgetting to revoke the old log record
5282 * from the journal and so a subsequent replay can corrupt data.
5283 * So, first we make sure that the journal is empty and that
5284 * nobody is changing anything.
5287 journal = EXT4_JOURNAL(inode);
5290 if (is_journal_aborted(journal))
5292 /* We have to allocate physical blocks for delalloc blocks
5293 * before flushing journal. otherwise delalloc blocks can not
5294 * be allocated any more. even more truncate on delalloc blocks
5295 * could trigger BUG by flushing delalloc blocks in journal.
5296 * There is no delalloc block in non-journal data mode.
5298 if (val && test_opt(inode->i_sb, DELALLOC)) {
5299 err = ext4_alloc_da_blocks(inode);
5304 /* Wait for all existing dio workers */
5305 ext4_inode_block_unlocked_dio(inode);
5306 inode_dio_wait(inode);
5308 jbd2_journal_lock_updates(journal);
5311 * OK, there are no updates running now, and all cached data is
5312 * synced to disk. We are now in a completely consistent state
5313 * which doesn't have anything in the journal, and we know that
5314 * no filesystem updates are running, so it is safe to modify
5315 * the inode's in-core data-journaling state flag now.
5319 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5321 err = jbd2_journal_flush(journal);
5323 jbd2_journal_unlock_updates(journal);
5324 ext4_inode_resume_unlocked_dio(inode);
5327 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5329 ext4_set_aops(inode);
5331 jbd2_journal_unlock_updates(journal);
5332 ext4_inode_resume_unlocked_dio(inode);
5334 /* Finally we can mark the inode as dirty. */
5336 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5338 return PTR_ERR(handle);
5340 err = ext4_mark_inode_dirty(handle, inode);
5341 ext4_handle_sync(handle);
5342 ext4_journal_stop(handle);
5343 ext4_std_error(inode->i_sb, err);
5348 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5350 return !buffer_mapped(bh);
5353 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5355 struct page *page = vmf->page;
5359 struct file *file = vma->vm_file;
5360 struct inode *inode = file_inode(file);
5361 struct address_space *mapping = inode->i_mapping;
5363 get_block_t *get_block;
5366 sb_start_pagefault(inode->i_sb);
5367 file_update_time(vma->vm_file);
5369 down_read(&EXT4_I(inode)->i_mmap_sem);
5370 /* Delalloc case is easy... */
5371 if (test_opt(inode->i_sb, DELALLOC) &&
5372 !ext4_should_journal_data(inode) &&
5373 !ext4_nonda_switch(inode->i_sb)) {
5375 ret = block_page_mkwrite(vma, vmf,
5376 ext4_da_get_block_prep);
5377 } while (ret == -ENOSPC &&
5378 ext4_should_retry_alloc(inode->i_sb, &retries));
5383 size = i_size_read(inode);
5384 /* Page got truncated from under us? */
5385 if (page->mapping != mapping || page_offset(page) > size) {
5387 ret = VM_FAULT_NOPAGE;
5391 if (page->index == size >> PAGE_CACHE_SHIFT)
5392 len = size & ~PAGE_CACHE_MASK;
5394 len = PAGE_CACHE_SIZE;
5396 * Return if we have all the buffers mapped. This avoids the need to do
5397 * journal_start/journal_stop which can block and take a long time
5399 if (page_has_buffers(page)) {
5400 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5402 ext4_bh_unmapped)) {
5403 /* Wait so that we don't change page under IO */
5404 wait_for_stable_page(page);
5405 ret = VM_FAULT_LOCKED;
5410 /* OK, we need to fill the hole... */
5411 if (ext4_should_dioread_nolock(inode))
5412 get_block = ext4_get_block_write;
5414 get_block = ext4_get_block;
5416 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5417 ext4_writepage_trans_blocks(inode));
5418 if (IS_ERR(handle)) {
5419 ret = VM_FAULT_SIGBUS;
5422 ret = block_page_mkwrite(vma, vmf, get_block);
5423 if (!ret && ext4_should_journal_data(inode)) {
5424 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5425 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5427 ret = VM_FAULT_SIGBUS;
5428 ext4_journal_stop(handle);
5431 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5433 ext4_journal_stop(handle);
5434 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5437 ret = block_page_mkwrite_return(ret);
5439 up_read(&EXT4_I(inode)->i_mmap_sem);
5440 sb_end_pagefault(inode->i_sb);
5444 int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
5446 struct inode *inode = file_inode(vma->vm_file);
5449 down_read(&EXT4_I(inode)->i_mmap_sem);
5450 err = filemap_fault(vma, vmf);
5451 up_read(&EXT4_I(inode)->i_mmap_sem);