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>
47 #include <trace/events/android_fs.h>
49 #define MPAGE_DA_EXTENT_TAIL 0x01
51 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
52 struct ext4_inode_info *ei)
54 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
57 int offset = offsetof(struct ext4_inode, i_checksum_lo);
58 unsigned int csum_size = sizeof(dummy_csum);
60 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
61 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
63 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
64 EXT4_GOOD_OLD_INODE_SIZE - offset);
66 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
67 offset = offsetof(struct ext4_inode, i_checksum_hi);
68 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
69 EXT4_GOOD_OLD_INODE_SIZE,
70 offset - EXT4_GOOD_OLD_INODE_SIZE);
71 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
72 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
75 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
76 EXT4_INODE_SIZE(inode->i_sb) -
84 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
85 struct ext4_inode_info *ei)
87 __u32 provided, calculated;
89 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
90 cpu_to_le32(EXT4_OS_LINUX) ||
91 !ext4_has_metadata_csum(inode->i_sb))
94 provided = le16_to_cpu(raw->i_checksum_lo);
95 calculated = ext4_inode_csum(inode, raw, ei);
96 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
97 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
98 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
100 calculated &= 0xFFFF;
102 return provided == calculated;
105 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
106 struct ext4_inode_info *ei)
110 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
111 cpu_to_le32(EXT4_OS_LINUX) ||
112 !ext4_has_metadata_csum(inode->i_sb))
115 csum = ext4_inode_csum(inode, raw, ei);
116 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
117 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
118 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
119 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
122 static inline int ext4_begin_ordered_truncate(struct inode *inode,
125 trace_ext4_begin_ordered_truncate(inode, new_size);
127 * If jinode is zero, then we never opened the file for
128 * writing, so there's no need to call
129 * jbd2_journal_begin_ordered_truncate() since there's no
130 * outstanding writes we need to flush.
132 if (!EXT4_I(inode)->jinode)
134 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
135 EXT4_I(inode)->jinode,
139 static void ext4_invalidatepage(struct page *page, unsigned int offset,
140 unsigned int length);
141 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
142 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
143 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
147 * Test whether an inode is a fast symlink.
149 int ext4_inode_is_fast_symlink(struct inode *inode)
151 int ea_blocks = EXT4_I(inode)->i_file_acl ?
152 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
154 if (ext4_has_inline_data(inode))
157 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
161 * Restart the transaction associated with *handle. This does a commit,
162 * so before we call here everything must be consistently dirtied against
165 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
171 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
172 * moment, get_block can be called only for blocks inside i_size since
173 * page cache has been already dropped and writes are blocked by
174 * i_mutex. So we can safely drop the i_data_sem here.
176 BUG_ON(EXT4_JOURNAL(inode) == NULL);
177 jbd_debug(2, "restarting handle %p\n", handle);
178 up_write(&EXT4_I(inode)->i_data_sem);
179 ret = ext4_journal_restart(handle, nblocks);
180 down_write(&EXT4_I(inode)->i_data_sem);
181 ext4_discard_preallocations(inode);
187 * Called at the last iput() if i_nlink is zero.
189 void ext4_evict_inode(struct inode *inode)
194 trace_ext4_evict_inode(inode);
196 if (inode->i_nlink) {
198 * When journalling data dirty buffers are tracked only in the
199 * journal. So although mm thinks everything is clean and
200 * ready for reaping the inode might still have some pages to
201 * write in the running transaction or waiting to be
202 * checkpointed. Thus calling jbd2_journal_invalidatepage()
203 * (via truncate_inode_pages()) to discard these buffers can
204 * cause data loss. Also even if we did not discard these
205 * buffers, we would have no way to find them after the inode
206 * is reaped and thus user could see stale data if he tries to
207 * read them before the transaction is checkpointed. So be
208 * careful and force everything to disk here... We use
209 * ei->i_datasync_tid to store the newest transaction
210 * containing inode's data.
212 * Note that directories do not have this problem because they
213 * don't use page cache.
215 if (inode->i_ino != EXT4_JOURNAL_INO &&
216 ext4_should_journal_data(inode) &&
217 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
218 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
219 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
221 jbd2_complete_transaction(journal, commit_tid);
222 filemap_write_and_wait(&inode->i_data);
224 truncate_inode_pages_final(&inode->i_data);
226 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
230 if (is_bad_inode(inode))
232 dquot_initialize(inode);
234 if (ext4_should_order_data(inode))
235 ext4_begin_ordered_truncate(inode, 0);
236 truncate_inode_pages_final(&inode->i_data);
238 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
241 * Protect us against freezing - iput() caller didn't have to have any
242 * protection against it
244 sb_start_intwrite(inode->i_sb);
245 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
246 ext4_blocks_for_truncate(inode)+3);
247 if (IS_ERR(handle)) {
248 ext4_std_error(inode->i_sb, PTR_ERR(handle));
250 * If we're going to skip the normal cleanup, we still need to
251 * make sure that the in-core orphan linked list is properly
254 ext4_orphan_del(NULL, inode);
255 sb_end_intwrite(inode->i_sb);
260 ext4_handle_sync(handle);
262 err = ext4_mark_inode_dirty(handle, inode);
264 ext4_warning(inode->i_sb,
265 "couldn't mark inode dirty (err %d)", err);
269 ext4_truncate(inode);
272 * ext4_ext_truncate() doesn't reserve any slop when it
273 * restarts journal transactions; therefore there may not be
274 * enough credits left in the handle to remove the inode from
275 * the orphan list and set the dtime field.
277 if (!ext4_handle_has_enough_credits(handle, 3)) {
278 err = ext4_journal_extend(handle, 3);
280 err = ext4_journal_restart(handle, 3);
282 ext4_warning(inode->i_sb,
283 "couldn't extend journal (err %d)", err);
285 ext4_journal_stop(handle);
286 ext4_orphan_del(NULL, inode);
287 sb_end_intwrite(inode->i_sb);
293 * Kill off the orphan record which ext4_truncate created.
294 * AKPM: I think this can be inside the above `if'.
295 * Note that ext4_orphan_del() has to be able to cope with the
296 * deletion of a non-existent orphan - this is because we don't
297 * know if ext4_truncate() actually created an orphan record.
298 * (Well, we could do this if we need to, but heck - it works)
300 ext4_orphan_del(handle, inode);
301 EXT4_I(inode)->i_dtime = get_seconds();
304 * One subtle ordering requirement: if anything has gone wrong
305 * (transaction abort, IO errors, whatever), then we can still
306 * do these next steps (the fs will already have been marked as
307 * having errors), but we can't free the inode if the mark_dirty
310 if (ext4_mark_inode_dirty(handle, inode))
311 /* If that failed, just do the required in-core inode clear. */
312 ext4_clear_inode(inode);
314 ext4_free_inode(handle, inode);
315 ext4_journal_stop(handle);
316 sb_end_intwrite(inode->i_sb);
319 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
323 qsize_t *ext4_get_reserved_space(struct inode *inode)
325 return &EXT4_I(inode)->i_reserved_quota;
330 * Called with i_data_sem down, which is important since we can call
331 * ext4_discard_preallocations() from here.
333 void ext4_da_update_reserve_space(struct inode *inode,
334 int used, int quota_claim)
336 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
337 struct ext4_inode_info *ei = EXT4_I(inode);
339 spin_lock(&ei->i_block_reservation_lock);
340 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
341 if (unlikely(used > ei->i_reserved_data_blocks)) {
342 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
343 "with only %d reserved data blocks",
344 __func__, inode->i_ino, used,
345 ei->i_reserved_data_blocks);
347 used = ei->i_reserved_data_blocks;
350 /* Update per-inode reservations */
351 ei->i_reserved_data_blocks -= used;
352 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
354 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
356 /* Update quota subsystem for data blocks */
358 dquot_claim_block(inode, EXT4_C2B(sbi, used));
361 * We did fallocate with an offset that is already delayed
362 * allocated. So on delayed allocated writeback we should
363 * not re-claim the quota for fallocated blocks.
365 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
369 * If we have done all the pending block allocations and if
370 * there aren't any writers on the inode, we can discard the
371 * inode's preallocations.
373 if ((ei->i_reserved_data_blocks == 0) &&
374 (atomic_read(&inode->i_writecount) == 0))
375 ext4_discard_preallocations(inode);
378 static int __check_block_validity(struct inode *inode, const char *func,
380 struct ext4_map_blocks *map)
382 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
384 ext4_error_inode(inode, func, line, map->m_pblk,
385 "lblock %lu mapped to illegal pblock "
386 "(length %d)", (unsigned long) map->m_lblk,
388 return -EFSCORRUPTED;
393 #define check_block_validity(inode, map) \
394 __check_block_validity((inode), __func__, __LINE__, (map))
396 #ifdef ES_AGGRESSIVE_TEST
397 static void ext4_map_blocks_es_recheck(handle_t *handle,
399 struct ext4_map_blocks *es_map,
400 struct ext4_map_blocks *map,
407 * There is a race window that the result is not the same.
408 * e.g. xfstests #223 when dioread_nolock enables. The reason
409 * is that we lookup a block mapping in extent status tree with
410 * out taking i_data_sem. So at the time the unwritten extent
411 * could be converted.
413 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
414 down_read(&EXT4_I(inode)->i_data_sem);
415 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
416 retval = ext4_ext_map_blocks(handle, inode, map, flags &
417 EXT4_GET_BLOCKS_KEEP_SIZE);
419 retval = ext4_ind_map_blocks(handle, inode, map, flags &
420 EXT4_GET_BLOCKS_KEEP_SIZE);
422 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
423 up_read((&EXT4_I(inode)->i_data_sem));
426 * We don't check m_len because extent will be collpased in status
427 * tree. So the m_len might not equal.
429 if (es_map->m_lblk != map->m_lblk ||
430 es_map->m_flags != map->m_flags ||
431 es_map->m_pblk != map->m_pblk) {
432 printk("ES cache assertion failed for inode: %lu "
433 "es_cached ex [%d/%d/%llu/%x] != "
434 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
435 inode->i_ino, es_map->m_lblk, es_map->m_len,
436 es_map->m_pblk, es_map->m_flags, map->m_lblk,
437 map->m_len, map->m_pblk, map->m_flags,
441 #endif /* ES_AGGRESSIVE_TEST */
444 * The ext4_map_blocks() function tries to look up the requested blocks,
445 * and returns if the blocks are already mapped.
447 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
448 * and store the allocated blocks in the result buffer head and mark it
451 * If file type is extents based, it will call ext4_ext_map_blocks(),
452 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
455 * On success, it returns the number of blocks being mapped or allocated.
456 * if create==0 and the blocks are pre-allocated and unwritten block,
457 * the result buffer head is unmapped. If the create ==1, it will make sure
458 * the buffer head is mapped.
460 * It returns 0 if plain look up failed (blocks have not been allocated), in
461 * that case, buffer head is unmapped
463 * It returns the error in case of allocation failure.
465 int ext4_map_blocks(handle_t *handle, struct inode *inode,
466 struct ext4_map_blocks *map, int flags)
468 struct extent_status es;
471 #ifdef ES_AGGRESSIVE_TEST
472 struct ext4_map_blocks orig_map;
474 memcpy(&orig_map, map, sizeof(*map));
478 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
479 "logical block %lu\n", inode->i_ino, flags, map->m_len,
480 (unsigned long) map->m_lblk);
483 * ext4_map_blocks returns an int, and m_len is an unsigned int
485 if (unlikely(map->m_len > INT_MAX))
486 map->m_len = INT_MAX;
488 /* We can handle the block number less than EXT_MAX_BLOCKS */
489 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
490 return -EFSCORRUPTED;
492 /* Lookup extent status tree firstly */
493 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
494 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
495 map->m_pblk = ext4_es_pblock(&es) +
496 map->m_lblk - es.es_lblk;
497 map->m_flags |= ext4_es_is_written(&es) ?
498 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
499 retval = es.es_len - (map->m_lblk - es.es_lblk);
500 if (retval > map->m_len)
503 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
508 #ifdef ES_AGGRESSIVE_TEST
509 ext4_map_blocks_es_recheck(handle, inode, map,
516 * Try to see if we can get the block without requesting a new
519 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
520 down_read(&EXT4_I(inode)->i_data_sem);
521 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
522 retval = ext4_ext_map_blocks(handle, inode, map, flags &
523 EXT4_GET_BLOCKS_KEEP_SIZE);
525 retval = ext4_ind_map_blocks(handle, inode, map, flags &
526 EXT4_GET_BLOCKS_KEEP_SIZE);
531 if (unlikely(retval != map->m_len)) {
532 ext4_warning(inode->i_sb,
533 "ES len assertion failed for inode "
534 "%lu: retval %d != map->m_len %d",
535 inode->i_ino, retval, map->m_len);
539 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
540 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
541 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
542 !(status & EXTENT_STATUS_WRITTEN) &&
543 ext4_find_delalloc_range(inode, map->m_lblk,
544 map->m_lblk + map->m_len - 1))
545 status |= EXTENT_STATUS_DELAYED;
546 ret = ext4_es_insert_extent(inode, map->m_lblk,
547 map->m_len, map->m_pblk, status);
551 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
552 up_read((&EXT4_I(inode)->i_data_sem));
555 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
556 ret = check_block_validity(inode, map);
561 /* If it is only a block(s) look up */
562 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
566 * Returns if the blocks have already allocated
568 * Note that if blocks have been preallocated
569 * ext4_ext_get_block() returns the create = 0
570 * with buffer head unmapped.
572 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
574 * If we need to convert extent to unwritten
575 * we continue and do the actual work in
576 * ext4_ext_map_blocks()
578 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
582 * Here we clear m_flags because after allocating an new extent,
583 * it will be set again.
585 map->m_flags &= ~EXT4_MAP_FLAGS;
588 * New blocks allocate and/or writing to unwritten extent
589 * will possibly result in updating i_data, so we take
590 * the write lock of i_data_sem, and call get_block()
591 * with create == 1 flag.
593 down_write(&EXT4_I(inode)->i_data_sem);
596 * We need to check for EXT4 here because migrate
597 * could have changed the inode type in between
599 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
600 retval = ext4_ext_map_blocks(handle, inode, map, flags);
602 retval = ext4_ind_map_blocks(handle, inode, map, flags);
604 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
606 * We allocated new blocks which will result in
607 * i_data's format changing. Force the migrate
608 * to fail by clearing migrate flags
610 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
614 * Update reserved blocks/metadata blocks after successful
615 * block allocation which had been deferred till now. We don't
616 * support fallocate for non extent files. So we can update
617 * reserve space here.
620 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
621 ext4_da_update_reserve_space(inode, retval, 1);
627 if (unlikely(retval != map->m_len)) {
628 ext4_warning(inode->i_sb,
629 "ES len assertion failed for inode "
630 "%lu: retval %d != map->m_len %d",
631 inode->i_ino, retval, map->m_len);
636 * If the extent has been zeroed out, we don't need to update
637 * extent status tree.
639 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
640 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
641 if (ext4_es_is_written(&es))
644 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
645 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
646 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
647 !(status & EXTENT_STATUS_WRITTEN) &&
648 ext4_find_delalloc_range(inode, map->m_lblk,
649 map->m_lblk + map->m_len - 1))
650 status |= EXTENT_STATUS_DELAYED;
651 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
652 map->m_pblk, status);
658 up_write((&EXT4_I(inode)->i_data_sem));
659 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
660 ret = check_block_validity(inode, map);
668 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
669 * we have to be careful as someone else may be manipulating b_state as well.
671 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
673 unsigned long old_state;
674 unsigned long new_state;
676 flags &= EXT4_MAP_FLAGS;
678 /* Dummy buffer_head? Set non-atomically. */
680 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
684 * Someone else may be modifying b_state. Be careful! This is ugly but
685 * once we get rid of using bh as a container for mapping information
686 * to pass to / from get_block functions, this can go away.
689 old_state = READ_ONCE(bh->b_state);
690 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
692 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
695 /* Maximum number of blocks we map for direct IO at once. */
696 #define DIO_MAX_BLOCKS 4096
698 static int _ext4_get_block(struct inode *inode, sector_t iblock,
699 struct buffer_head *bh, int flags)
701 handle_t *handle = ext4_journal_current_handle();
702 struct ext4_map_blocks map;
703 int ret = 0, started = 0;
706 if (ext4_has_inline_data(inode))
710 map.m_len = bh->b_size >> inode->i_blkbits;
712 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
713 /* Direct IO write... */
714 if (map.m_len > DIO_MAX_BLOCKS)
715 map.m_len = DIO_MAX_BLOCKS;
716 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
717 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
719 if (IS_ERR(handle)) {
720 ret = PTR_ERR(handle);
726 ret = ext4_map_blocks(handle, inode, &map, flags);
728 ext4_io_end_t *io_end = ext4_inode_aio(inode);
730 map_bh(bh, inode->i_sb, map.m_pblk);
731 ext4_update_bh_state(bh, map.m_flags);
732 if (IS_DAX(inode) && buffer_unwritten(bh)) {
734 * dgc: I suspect unwritten conversion on ext4+DAX is
735 * fundamentally broken here when there are concurrent
736 * read/write in progress on this inode.
738 WARN_ON_ONCE(io_end);
739 bh->b_assoc_map = inode->i_mapping;
740 bh->b_private = (void *)(unsigned long)iblock;
742 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
743 set_buffer_defer_completion(bh);
744 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
748 ext4_journal_stop(handle);
752 int ext4_get_block(struct inode *inode, sector_t iblock,
753 struct buffer_head *bh, int create)
755 return _ext4_get_block(inode, iblock, bh,
756 create ? EXT4_GET_BLOCKS_CREATE : 0);
760 * `handle' can be NULL if create is zero
762 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
763 ext4_lblk_t block, int map_flags)
765 struct ext4_map_blocks map;
766 struct buffer_head *bh;
767 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
770 J_ASSERT(handle != NULL || create == 0);
774 err = ext4_map_blocks(handle, inode, &map, map_flags);
777 return create ? ERR_PTR(-ENOSPC) : NULL;
781 bh = sb_getblk(inode->i_sb, map.m_pblk);
783 return ERR_PTR(-ENOMEM);
784 if (map.m_flags & EXT4_MAP_NEW) {
785 J_ASSERT(create != 0);
786 J_ASSERT(handle != NULL);
789 * Now that we do not always journal data, we should
790 * keep in mind whether this should always journal the
791 * new buffer as metadata. For now, regular file
792 * writes use ext4_get_block instead, so it's not a
796 BUFFER_TRACE(bh, "call get_create_access");
797 err = ext4_journal_get_create_access(handle, bh);
802 if (!buffer_uptodate(bh)) {
803 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
804 set_buffer_uptodate(bh);
807 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
808 err = ext4_handle_dirty_metadata(handle, inode, bh);
812 BUFFER_TRACE(bh, "not a new buffer");
819 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
820 ext4_lblk_t block, int map_flags)
822 struct buffer_head *bh;
824 bh = ext4_getblk(handle, inode, block, map_flags);
827 if (!bh || buffer_uptodate(bh))
829 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
831 if (buffer_uptodate(bh))
834 return ERR_PTR(-EIO);
837 int ext4_walk_page_buffers(handle_t *handle,
838 struct buffer_head *head,
842 int (*fn)(handle_t *handle,
843 struct buffer_head *bh))
845 struct buffer_head *bh;
846 unsigned block_start, block_end;
847 unsigned blocksize = head->b_size;
849 struct buffer_head *next;
851 for (bh = head, block_start = 0;
852 ret == 0 && (bh != head || !block_start);
853 block_start = block_end, bh = next) {
854 next = bh->b_this_page;
855 block_end = block_start + blocksize;
856 if (block_end <= from || block_start >= to) {
857 if (partial && !buffer_uptodate(bh))
861 err = (*fn)(handle, bh);
869 * To preserve ordering, it is essential that the hole instantiation and
870 * the data write be encapsulated in a single transaction. We cannot
871 * close off a transaction and start a new one between the ext4_get_block()
872 * and the commit_write(). So doing the jbd2_journal_start at the start of
873 * prepare_write() is the right place.
875 * Also, this function can nest inside ext4_writepage(). In that case, we
876 * *know* that ext4_writepage() has generated enough buffer credits to do the
877 * whole page. So we won't block on the journal in that case, which is good,
878 * because the caller may be PF_MEMALLOC.
880 * By accident, ext4 can be reentered when a transaction is open via
881 * quota file writes. If we were to commit the transaction while thus
882 * reentered, there can be a deadlock - we would be holding a quota
883 * lock, and the commit would never complete if another thread had a
884 * transaction open and was blocking on the quota lock - a ranking
887 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
888 * will _not_ run commit under these circumstances because handle->h_ref
889 * is elevated. We'll still have enough credits for the tiny quotafile
892 int do_journal_get_write_access(handle_t *handle,
893 struct buffer_head *bh)
895 int dirty = buffer_dirty(bh);
898 if (!buffer_mapped(bh) || buffer_freed(bh))
901 * __block_write_begin() could have dirtied some buffers. Clean
902 * the dirty bit as jbd2_journal_get_write_access() could complain
903 * otherwise about fs integrity issues. Setting of the dirty bit
904 * by __block_write_begin() isn't a real problem here as we clear
905 * the bit before releasing a page lock and thus writeback cannot
906 * ever write the buffer.
909 clear_buffer_dirty(bh);
910 BUFFER_TRACE(bh, "get write access");
911 ret = ext4_journal_get_write_access(handle, bh);
913 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
917 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
918 struct buffer_head *bh_result, int create);
920 #ifdef CONFIG_EXT4_FS_ENCRYPTION
921 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
922 get_block_t *get_block)
924 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
925 unsigned to = from + len;
926 struct inode *inode = page->mapping->host;
927 unsigned block_start, block_end;
930 unsigned blocksize = inode->i_sb->s_blocksize;
932 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
933 bool decrypt = false;
935 BUG_ON(!PageLocked(page));
936 BUG_ON(from > PAGE_CACHE_SIZE);
937 BUG_ON(to > PAGE_CACHE_SIZE);
940 if (!page_has_buffers(page))
941 create_empty_buffers(page, blocksize, 0);
942 head = page_buffers(page);
943 bbits = ilog2(blocksize);
944 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
946 for (bh = head, block_start = 0; bh != head || !block_start;
947 block++, block_start = block_end, bh = bh->b_this_page) {
948 block_end = block_start + blocksize;
949 if (block_end <= from || block_start >= to) {
950 if (PageUptodate(page)) {
951 if (!buffer_uptodate(bh))
952 set_buffer_uptodate(bh);
957 clear_buffer_new(bh);
958 if (!buffer_mapped(bh)) {
959 WARN_ON(bh->b_size != blocksize);
960 err = get_block(inode, block, bh, 1);
963 if (buffer_new(bh)) {
964 unmap_underlying_metadata(bh->b_bdev,
966 if (PageUptodate(page)) {
967 clear_buffer_new(bh);
968 set_buffer_uptodate(bh);
969 mark_buffer_dirty(bh);
972 if (block_end > to || block_start < from)
973 zero_user_segments(page, to, block_end,
978 if (PageUptodate(page)) {
979 if (!buffer_uptodate(bh))
980 set_buffer_uptodate(bh);
983 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
984 !buffer_unwritten(bh) &&
985 (block_start < from || block_end > to)) {
986 ll_rw_block(READ, 1, &bh);
988 decrypt = ext4_encrypted_inode(inode) &&
989 S_ISREG(inode->i_mode);
993 * If we issued read requests, let them complete.
995 while (wait_bh > wait) {
996 wait_on_buffer(*--wait_bh);
997 if (!buffer_uptodate(*wait_bh))
1001 page_zero_new_buffers(page, from, to);
1003 err = ext4_decrypt(page);
1008 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1009 loff_t pos, unsigned len, unsigned flags,
1010 struct page **pagep, void **fsdata)
1012 struct inode *inode = mapping->host;
1013 int ret, needed_blocks;
1020 trace_android_fs_datawrite_start(inode, pos, len,
1021 current->pid, current->comm);
1022 trace_ext4_write_begin(inode, pos, len, flags);
1024 * Reserve one block more for addition to orphan list in case
1025 * we allocate blocks but write fails for some reason
1027 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1028 index = pos >> PAGE_CACHE_SHIFT;
1029 from = pos & (PAGE_CACHE_SIZE - 1);
1032 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1033 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1042 * grab_cache_page_write_begin() can take a long time if the
1043 * system is thrashing due to memory pressure, or if the page
1044 * is being written back. So grab it first before we start
1045 * the transaction handle. This also allows us to allocate
1046 * the page (if needed) without using GFP_NOFS.
1049 page = grab_cache_page_write_begin(mapping, index, flags);
1055 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1056 if (IS_ERR(handle)) {
1057 page_cache_release(page);
1058 return PTR_ERR(handle);
1062 if (page->mapping != mapping) {
1063 /* The page got truncated from under us */
1065 page_cache_release(page);
1066 ext4_journal_stop(handle);
1069 /* In case writeback began while the page was unlocked */
1070 wait_for_stable_page(page);
1072 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1073 if (ext4_should_dioread_nolock(inode))
1074 ret = ext4_block_write_begin(page, pos, len,
1075 ext4_get_block_write);
1077 ret = ext4_block_write_begin(page, pos, len,
1080 if (ext4_should_dioread_nolock(inode))
1081 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1083 ret = __block_write_begin(page, pos, len, ext4_get_block);
1085 if (!ret && ext4_should_journal_data(inode)) {
1086 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1088 do_journal_get_write_access);
1094 * __block_write_begin may have instantiated a few blocks
1095 * outside i_size. Trim these off again. Don't need
1096 * i_size_read because we hold i_mutex.
1098 * Add inode to orphan list in case we crash before
1101 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1102 ext4_orphan_add(handle, inode);
1104 ext4_journal_stop(handle);
1105 if (pos + len > inode->i_size) {
1106 ext4_truncate_failed_write(inode);
1108 * If truncate failed early the inode might
1109 * still be on the orphan list; we need to
1110 * make sure the inode is removed from the
1111 * orphan list in that case.
1114 ext4_orphan_del(NULL, inode);
1117 if (ret == -ENOSPC &&
1118 ext4_should_retry_alloc(inode->i_sb, &retries))
1120 page_cache_release(page);
1127 /* For write_end() in data=journal mode */
1128 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1131 if (!buffer_mapped(bh) || buffer_freed(bh))
1133 set_buffer_uptodate(bh);
1134 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1135 clear_buffer_meta(bh);
1136 clear_buffer_prio(bh);
1141 * We need to pick up the new inode size which generic_commit_write gave us
1142 * `file' can be NULL - eg, when called from page_symlink().
1144 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1145 * buffers are managed internally.
1147 static int ext4_write_end(struct file *file,
1148 struct address_space *mapping,
1149 loff_t pos, unsigned len, unsigned copied,
1150 struct page *page, void *fsdata)
1152 handle_t *handle = ext4_journal_current_handle();
1153 struct inode *inode = mapping->host;
1154 loff_t old_size = inode->i_size;
1156 int i_size_changed = 0;
1158 trace_android_fs_datawrite_end(inode, pos, len);
1159 trace_ext4_write_end(inode, pos, len, copied);
1160 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1161 ret = ext4_jbd2_file_inode(handle, inode);
1164 page_cache_release(page);
1169 if (ext4_has_inline_data(inode)) {
1170 ret = ext4_write_inline_data_end(inode, pos, len,
1176 copied = block_write_end(file, mapping, pos,
1177 len, copied, page, fsdata);
1179 * it's important to update i_size while still holding page lock:
1180 * page writeout could otherwise come in and zero beyond i_size.
1182 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1184 page_cache_release(page);
1187 pagecache_isize_extended(inode, old_size, pos);
1189 * Don't mark the inode dirty under page lock. First, it unnecessarily
1190 * makes the holding time of page lock longer. Second, it forces lock
1191 * ordering of page lock and transaction start for journaling
1195 ext4_mark_inode_dirty(handle, inode);
1197 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1198 /* if we have allocated more blocks and copied
1199 * less. We will have blocks allocated outside
1200 * inode->i_size. So truncate them
1202 ext4_orphan_add(handle, inode);
1204 ret2 = ext4_journal_stop(handle);
1208 if (pos + len > inode->i_size) {
1209 ext4_truncate_failed_write(inode);
1211 * If truncate failed early the inode might still be
1212 * on the orphan list; we need to make sure the inode
1213 * is removed from the orphan list in that case.
1216 ext4_orphan_del(NULL, inode);
1219 return ret ? ret : copied;
1223 * This is a private version of page_zero_new_buffers() which doesn't
1224 * set the buffer to be dirty, since in data=journalled mode we need
1225 * to call ext4_handle_dirty_metadata() instead.
1227 static void zero_new_buffers(struct page *page, unsigned from, unsigned to)
1229 unsigned int block_start = 0, block_end;
1230 struct buffer_head *head, *bh;
1232 bh = head = page_buffers(page);
1234 block_end = block_start + bh->b_size;
1235 if (buffer_new(bh)) {
1236 if (block_end > from && block_start < to) {
1237 if (!PageUptodate(page)) {
1238 unsigned start, size;
1240 start = max(from, block_start);
1241 size = min(to, block_end) - start;
1243 zero_user(page, start, size);
1244 set_buffer_uptodate(bh);
1246 clear_buffer_new(bh);
1249 block_start = block_end;
1250 bh = bh->b_this_page;
1251 } while (bh != head);
1254 static int ext4_journalled_write_end(struct file *file,
1255 struct address_space *mapping,
1256 loff_t pos, unsigned len, unsigned copied,
1257 struct page *page, void *fsdata)
1259 handle_t *handle = ext4_journal_current_handle();
1260 struct inode *inode = mapping->host;
1261 loff_t old_size = inode->i_size;
1265 int size_changed = 0;
1267 trace_android_fs_datawrite_end(inode, pos, len);
1268 trace_ext4_journalled_write_end(inode, pos, len, copied);
1269 from = pos & (PAGE_CACHE_SIZE - 1);
1272 BUG_ON(!ext4_handle_valid(handle));
1274 if (ext4_has_inline_data(inode))
1275 copied = ext4_write_inline_data_end(inode, pos, len,
1279 if (!PageUptodate(page))
1281 zero_new_buffers(page, from+copied, to);
1284 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1285 to, &partial, write_end_fn);
1287 SetPageUptodate(page);
1289 size_changed = ext4_update_inode_size(inode, pos + copied);
1290 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1291 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1293 page_cache_release(page);
1296 pagecache_isize_extended(inode, old_size, pos);
1299 ret2 = ext4_mark_inode_dirty(handle, inode);
1304 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1305 /* if we have allocated more blocks and copied
1306 * less. We will have blocks allocated outside
1307 * inode->i_size. So truncate them
1309 ext4_orphan_add(handle, inode);
1311 ret2 = ext4_journal_stop(handle);
1314 if (pos + len > inode->i_size) {
1315 ext4_truncate_failed_write(inode);
1317 * If truncate failed early the inode might still be
1318 * on the orphan list; we need to make sure the inode
1319 * is removed from the orphan list in that case.
1322 ext4_orphan_del(NULL, inode);
1325 return ret ? ret : copied;
1329 * Reserve space for a single cluster
1331 static int ext4_da_reserve_space(struct inode *inode)
1333 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1334 struct ext4_inode_info *ei = EXT4_I(inode);
1338 * We will charge metadata quota at writeout time; this saves
1339 * us from metadata over-estimation, though we may go over by
1340 * a small amount in the end. Here we just reserve for data.
1342 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1346 spin_lock(&ei->i_block_reservation_lock);
1347 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1348 spin_unlock(&ei->i_block_reservation_lock);
1349 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1352 ei->i_reserved_data_blocks++;
1353 trace_ext4_da_reserve_space(inode);
1354 spin_unlock(&ei->i_block_reservation_lock);
1356 return 0; /* success */
1359 static void ext4_da_release_space(struct inode *inode, int to_free)
1361 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1362 struct ext4_inode_info *ei = EXT4_I(inode);
1365 return; /* Nothing to release, exit */
1367 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1369 trace_ext4_da_release_space(inode, to_free);
1370 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1372 * if there aren't enough reserved blocks, then the
1373 * counter is messed up somewhere. Since this
1374 * function is called from invalidate page, it's
1375 * harmless to return without any action.
1377 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1378 "ino %lu, to_free %d with only %d reserved "
1379 "data blocks", inode->i_ino, to_free,
1380 ei->i_reserved_data_blocks);
1382 to_free = ei->i_reserved_data_blocks;
1384 ei->i_reserved_data_blocks -= to_free;
1386 /* update fs dirty data blocks counter */
1387 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1389 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1391 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1394 static void ext4_da_page_release_reservation(struct page *page,
1395 unsigned int offset,
1396 unsigned int length)
1398 int to_release = 0, contiguous_blks = 0;
1399 struct buffer_head *head, *bh;
1400 unsigned int curr_off = 0;
1401 struct inode *inode = page->mapping->host;
1402 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1403 unsigned int stop = offset + length;
1407 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1409 head = page_buffers(page);
1412 unsigned int next_off = curr_off + bh->b_size;
1414 if (next_off > stop)
1417 if ((offset <= curr_off) && (buffer_delay(bh))) {
1420 clear_buffer_delay(bh);
1421 } else if (contiguous_blks) {
1422 lblk = page->index <<
1423 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1424 lblk += (curr_off >> inode->i_blkbits) -
1426 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1427 contiguous_blks = 0;
1429 curr_off = next_off;
1430 } while ((bh = bh->b_this_page) != head);
1432 if (contiguous_blks) {
1433 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1434 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1435 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1438 /* If we have released all the blocks belonging to a cluster, then we
1439 * need to release the reserved space for that cluster. */
1440 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1441 while (num_clusters > 0) {
1442 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1443 ((num_clusters - 1) << sbi->s_cluster_bits);
1444 if (sbi->s_cluster_ratio == 1 ||
1445 !ext4_find_delalloc_cluster(inode, lblk))
1446 ext4_da_release_space(inode, 1);
1453 * Delayed allocation stuff
1456 struct mpage_da_data {
1457 struct inode *inode;
1458 struct writeback_control *wbc;
1460 pgoff_t first_page; /* The first page to write */
1461 pgoff_t next_page; /* Current page to examine */
1462 pgoff_t last_page; /* Last page to examine */
1464 * Extent to map - this can be after first_page because that can be
1465 * fully mapped. We somewhat abuse m_flags to store whether the extent
1466 * is delalloc or unwritten.
1468 struct ext4_map_blocks map;
1469 struct ext4_io_submit io_submit; /* IO submission data */
1472 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1477 struct pagevec pvec;
1478 struct inode *inode = mpd->inode;
1479 struct address_space *mapping = inode->i_mapping;
1481 /* This is necessary when next_page == 0. */
1482 if (mpd->first_page >= mpd->next_page)
1485 index = mpd->first_page;
1486 end = mpd->next_page - 1;
1488 ext4_lblk_t start, last;
1489 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1490 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1491 ext4_es_remove_extent(inode, start, last - start + 1);
1494 pagevec_init(&pvec, 0);
1495 while (index <= end) {
1496 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1499 for (i = 0; i < nr_pages; i++) {
1500 struct page *page = pvec.pages[i];
1501 if (page->index > end)
1503 BUG_ON(!PageLocked(page));
1504 BUG_ON(PageWriteback(page));
1506 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1507 ClearPageUptodate(page);
1511 index = pvec.pages[nr_pages - 1]->index + 1;
1512 pagevec_release(&pvec);
1516 static void ext4_print_free_blocks(struct inode *inode)
1518 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1519 struct super_block *sb = inode->i_sb;
1520 struct ext4_inode_info *ei = EXT4_I(inode);
1522 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1523 EXT4_C2B(EXT4_SB(inode->i_sb),
1524 ext4_count_free_clusters(sb)));
1525 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1526 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1527 (long long) EXT4_C2B(EXT4_SB(sb),
1528 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1529 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1530 (long long) EXT4_C2B(EXT4_SB(sb),
1531 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1532 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1533 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1534 ei->i_reserved_data_blocks);
1538 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1540 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1544 * This function is grabs code from the very beginning of
1545 * ext4_map_blocks, but assumes that the caller is from delayed write
1546 * time. This function looks up the requested blocks and sets the
1547 * buffer delay bit under the protection of i_data_sem.
1549 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1550 struct ext4_map_blocks *map,
1551 struct buffer_head *bh)
1553 struct extent_status es;
1555 sector_t invalid_block = ~((sector_t) 0xffff);
1556 #ifdef ES_AGGRESSIVE_TEST
1557 struct ext4_map_blocks orig_map;
1559 memcpy(&orig_map, map, sizeof(*map));
1562 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1566 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1567 "logical block %lu\n", inode->i_ino, map->m_len,
1568 (unsigned long) map->m_lblk);
1570 /* Lookup extent status tree firstly */
1571 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1572 if (ext4_es_is_hole(&es)) {
1574 down_read(&EXT4_I(inode)->i_data_sem);
1579 * Delayed extent could be allocated by fallocate.
1580 * So we need to check it.
1582 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1583 map_bh(bh, inode->i_sb, invalid_block);
1585 set_buffer_delay(bh);
1589 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1590 retval = es.es_len - (iblock - es.es_lblk);
1591 if (retval > map->m_len)
1592 retval = map->m_len;
1593 map->m_len = retval;
1594 if (ext4_es_is_written(&es))
1595 map->m_flags |= EXT4_MAP_MAPPED;
1596 else if (ext4_es_is_unwritten(&es))
1597 map->m_flags |= EXT4_MAP_UNWRITTEN;
1601 #ifdef ES_AGGRESSIVE_TEST
1602 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1608 * Try to see if we can get the block without requesting a new
1609 * file system block.
1611 down_read(&EXT4_I(inode)->i_data_sem);
1612 if (ext4_has_inline_data(inode))
1614 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1615 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1617 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1623 * XXX: __block_prepare_write() unmaps passed block,
1627 * If the block was allocated from previously allocated cluster,
1628 * then we don't need to reserve it again. However we still need
1629 * to reserve metadata for every block we're going to write.
1631 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1632 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1633 ret = ext4_da_reserve_space(inode);
1635 /* not enough space to reserve */
1641 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1642 ~0, EXTENT_STATUS_DELAYED);
1648 map_bh(bh, inode->i_sb, invalid_block);
1650 set_buffer_delay(bh);
1651 } else if (retval > 0) {
1653 unsigned int status;
1655 if (unlikely(retval != map->m_len)) {
1656 ext4_warning(inode->i_sb,
1657 "ES len assertion failed for inode "
1658 "%lu: retval %d != map->m_len %d",
1659 inode->i_ino, retval, map->m_len);
1663 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1664 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1665 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1666 map->m_pblk, status);
1672 up_read((&EXT4_I(inode)->i_data_sem));
1678 * This is a special get_block_t callback which is used by
1679 * ext4_da_write_begin(). It will either return mapped block or
1680 * reserve space for a single block.
1682 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1683 * We also have b_blocknr = -1 and b_bdev initialized properly
1685 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1686 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1687 * initialized properly.
1689 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1690 struct buffer_head *bh, int create)
1692 struct ext4_map_blocks map;
1695 BUG_ON(create == 0);
1696 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1698 map.m_lblk = iblock;
1702 * first, we need to know whether the block is allocated already
1703 * preallocated blocks are unmapped but should treated
1704 * the same as allocated blocks.
1706 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1710 map_bh(bh, inode->i_sb, map.m_pblk);
1711 ext4_update_bh_state(bh, map.m_flags);
1713 if (buffer_unwritten(bh)) {
1714 /* A delayed write to unwritten bh should be marked
1715 * new and mapped. Mapped ensures that we don't do
1716 * get_block multiple times when we write to the same
1717 * offset and new ensures that we do proper zero out
1718 * for partial write.
1721 set_buffer_mapped(bh);
1726 static int bget_one(handle_t *handle, struct buffer_head *bh)
1732 static int bput_one(handle_t *handle, struct buffer_head *bh)
1738 static int __ext4_journalled_writepage(struct page *page,
1741 struct address_space *mapping = page->mapping;
1742 struct inode *inode = mapping->host;
1743 struct buffer_head *page_bufs = NULL;
1744 handle_t *handle = NULL;
1745 int ret = 0, err = 0;
1746 int inline_data = ext4_has_inline_data(inode);
1747 struct buffer_head *inode_bh = NULL;
1749 ClearPageChecked(page);
1752 BUG_ON(page->index != 0);
1753 BUG_ON(len > ext4_get_max_inline_size(inode));
1754 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1755 if (inode_bh == NULL)
1758 page_bufs = page_buffers(page);
1763 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1767 * We need to release the page lock before we start the
1768 * journal, so grab a reference so the page won't disappear
1769 * out from under us.
1774 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1775 ext4_writepage_trans_blocks(inode));
1776 if (IS_ERR(handle)) {
1777 ret = PTR_ERR(handle);
1779 goto out_no_pagelock;
1781 BUG_ON(!ext4_handle_valid(handle));
1785 if (page->mapping != mapping) {
1786 /* The page got truncated from under us */
1787 ext4_journal_stop(handle);
1793 BUFFER_TRACE(inode_bh, "get write access");
1794 ret = ext4_journal_get_write_access(handle, inode_bh);
1796 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1799 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1800 do_journal_get_write_access);
1802 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1807 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1808 err = ext4_journal_stop(handle);
1812 if (!ext4_has_inline_data(inode))
1813 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1815 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1824 * Note that we don't need to start a transaction unless we're journaling data
1825 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1826 * need to file the inode to the transaction's list in ordered mode because if
1827 * we are writing back data added by write(), the inode is already there and if
1828 * we are writing back data modified via mmap(), no one guarantees in which
1829 * transaction the data will hit the disk. In case we are journaling data, we
1830 * cannot start transaction directly because transaction start ranks above page
1831 * lock so we have to do some magic.
1833 * This function can get called via...
1834 * - ext4_writepages after taking page lock (have journal handle)
1835 * - journal_submit_inode_data_buffers (no journal handle)
1836 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1837 * - grab_page_cache when doing write_begin (have journal handle)
1839 * We don't do any block allocation in this function. If we have page with
1840 * multiple blocks we need to write those buffer_heads that are mapped. This
1841 * is important for mmaped based write. So if we do with blocksize 1K
1842 * truncate(f, 1024);
1843 * a = mmap(f, 0, 4096);
1845 * truncate(f, 4096);
1846 * we have in the page first buffer_head mapped via page_mkwrite call back
1847 * but other buffer_heads would be unmapped but dirty (dirty done via the
1848 * do_wp_page). So writepage should write the first block. If we modify
1849 * the mmap area beyond 1024 we will again get a page_fault and the
1850 * page_mkwrite callback will do the block allocation and mark the
1851 * buffer_heads mapped.
1853 * We redirty the page if we have any buffer_heads that is either delay or
1854 * unwritten in the page.
1856 * We can get recursively called as show below.
1858 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1861 * But since we don't do any block allocation we should not deadlock.
1862 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1864 static int ext4_writepage(struct page *page,
1865 struct writeback_control *wbc)
1870 struct buffer_head *page_bufs = NULL;
1871 struct inode *inode = page->mapping->host;
1872 struct ext4_io_submit io_submit;
1873 bool keep_towrite = false;
1875 trace_ext4_writepage(page);
1876 size = i_size_read(inode);
1877 if (page->index == size >> PAGE_CACHE_SHIFT)
1878 len = size & ~PAGE_CACHE_MASK;
1880 len = PAGE_CACHE_SIZE;
1882 page_bufs = page_buffers(page);
1884 * We cannot do block allocation or other extent handling in this
1885 * function. If there are buffers needing that, we have to redirty
1886 * the page. But we may reach here when we do a journal commit via
1887 * journal_submit_inode_data_buffers() and in that case we must write
1888 * allocated buffers to achieve data=ordered mode guarantees.
1890 * Also, if there is only one buffer per page (the fs block
1891 * size == the page size), if one buffer needs block
1892 * allocation or needs to modify the extent tree to clear the
1893 * unwritten flag, we know that the page can't be written at
1894 * all, so we might as well refuse the write immediately.
1895 * Unfortunately if the block size != page size, we can't as
1896 * easily detect this case using ext4_walk_page_buffers(), but
1897 * for the extremely common case, this is an optimization that
1898 * skips a useless round trip through ext4_bio_write_page().
1900 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1901 ext4_bh_delay_or_unwritten)) {
1902 redirty_page_for_writepage(wbc, page);
1903 if ((current->flags & PF_MEMALLOC) ||
1904 (inode->i_sb->s_blocksize == PAGE_CACHE_SIZE)) {
1906 * For memory cleaning there's no point in writing only
1907 * some buffers. So just bail out. Warn if we came here
1908 * from direct reclaim.
1910 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1915 keep_towrite = true;
1918 if (PageChecked(page) && ext4_should_journal_data(inode))
1920 * It's mmapped pagecache. Add buffers and journal it. There
1921 * doesn't seem much point in redirtying the page here.
1923 return __ext4_journalled_writepage(page, len);
1925 ext4_io_submit_init(&io_submit, wbc);
1926 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1927 if (!io_submit.io_end) {
1928 redirty_page_for_writepage(wbc, page);
1932 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1933 ext4_io_submit(&io_submit);
1934 /* Drop io_end reference we got from init */
1935 ext4_put_io_end_defer(io_submit.io_end);
1939 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1942 loff_t size = i_size_read(mpd->inode);
1945 BUG_ON(page->index != mpd->first_page);
1946 if (page->index == size >> PAGE_CACHE_SHIFT)
1947 len = size & ~PAGE_CACHE_MASK;
1949 len = PAGE_CACHE_SIZE;
1950 clear_page_dirty_for_io(page);
1951 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1953 mpd->wbc->nr_to_write--;
1959 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1962 * mballoc gives us at most this number of blocks...
1963 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1964 * The rest of mballoc seems to handle chunks up to full group size.
1966 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1969 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1971 * @mpd - extent of blocks
1972 * @lblk - logical number of the block in the file
1973 * @bh - buffer head we want to add to the extent
1975 * The function is used to collect contig. blocks in the same state. If the
1976 * buffer doesn't require mapping for writeback and we haven't started the
1977 * extent of buffers to map yet, the function returns 'true' immediately - the
1978 * caller can write the buffer right away. Otherwise the function returns true
1979 * if the block has been added to the extent, false if the block couldn't be
1982 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1983 struct buffer_head *bh)
1985 struct ext4_map_blocks *map = &mpd->map;
1987 /* Buffer that doesn't need mapping for writeback? */
1988 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1989 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1990 /* So far no extent to map => we write the buffer right away */
1991 if (map->m_len == 0)
1996 /* First block in the extent? */
1997 if (map->m_len == 0) {
2000 map->m_flags = bh->b_state & BH_FLAGS;
2004 /* Don't go larger than mballoc is willing to allocate */
2005 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2008 /* Can we merge the block to our big extent? */
2009 if (lblk == map->m_lblk + map->m_len &&
2010 (bh->b_state & BH_FLAGS) == map->m_flags) {
2018 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2020 * @mpd - extent of blocks for mapping
2021 * @head - the first buffer in the page
2022 * @bh - buffer we should start processing from
2023 * @lblk - logical number of the block in the file corresponding to @bh
2025 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2026 * the page for IO if all buffers in this page were mapped and there's no
2027 * accumulated extent of buffers to map or add buffers in the page to the
2028 * extent of buffers to map. The function returns 1 if the caller can continue
2029 * by processing the next page, 0 if it should stop adding buffers to the
2030 * extent to map because we cannot extend it anymore. It can also return value
2031 * < 0 in case of error during IO submission.
2033 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2034 struct buffer_head *head,
2035 struct buffer_head *bh,
2038 struct inode *inode = mpd->inode;
2040 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2041 >> inode->i_blkbits;
2044 BUG_ON(buffer_locked(bh));
2046 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2047 /* Found extent to map? */
2050 /* Everything mapped so far and we hit EOF */
2053 } while (lblk++, (bh = bh->b_this_page) != head);
2054 /* So far everything mapped? Submit the page for IO. */
2055 if (mpd->map.m_len == 0) {
2056 err = mpage_submit_page(mpd, head->b_page);
2060 return lblk < blocks;
2064 * mpage_map_buffers - update buffers corresponding to changed extent and
2065 * submit fully mapped pages for IO
2067 * @mpd - description of extent to map, on return next extent to map
2069 * Scan buffers corresponding to changed extent (we expect corresponding pages
2070 * to be already locked) and update buffer state according to new extent state.
2071 * We map delalloc buffers to their physical location, clear unwritten bits,
2072 * and mark buffers as uninit when we perform writes to unwritten extents
2073 * and do extent conversion after IO is finished. If the last page is not fully
2074 * mapped, we update @map to the next extent in the last page that needs
2075 * mapping. Otherwise we submit the page for IO.
2077 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2079 struct pagevec pvec;
2081 struct inode *inode = mpd->inode;
2082 struct buffer_head *head, *bh;
2083 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
2089 start = mpd->map.m_lblk >> bpp_bits;
2090 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2091 lblk = start << bpp_bits;
2092 pblock = mpd->map.m_pblk;
2094 pagevec_init(&pvec, 0);
2095 while (start <= end) {
2096 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2100 for (i = 0; i < nr_pages; i++) {
2101 struct page *page = pvec.pages[i];
2103 if (page->index > end)
2105 /* Up to 'end' pages must be contiguous */
2106 BUG_ON(page->index != start);
2107 bh = head = page_buffers(page);
2109 if (lblk < mpd->map.m_lblk)
2111 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2113 * Buffer after end of mapped extent.
2114 * Find next buffer in the page to map.
2117 mpd->map.m_flags = 0;
2119 * FIXME: If dioread_nolock supports
2120 * blocksize < pagesize, we need to make
2121 * sure we add size mapped so far to
2122 * io_end->size as the following call
2123 * can submit the page for IO.
2125 err = mpage_process_page_bufs(mpd, head,
2127 pagevec_release(&pvec);
2132 if (buffer_delay(bh)) {
2133 clear_buffer_delay(bh);
2134 bh->b_blocknr = pblock++;
2136 clear_buffer_unwritten(bh);
2137 } while (lblk++, (bh = bh->b_this_page) != head);
2140 * FIXME: This is going to break if dioread_nolock
2141 * supports blocksize < pagesize as we will try to
2142 * convert potentially unmapped parts of inode.
2144 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2145 /* Page fully mapped - let IO run! */
2146 err = mpage_submit_page(mpd, page);
2148 pagevec_release(&pvec);
2153 pagevec_release(&pvec);
2155 /* Extent fully mapped and matches with page boundary. We are done. */
2157 mpd->map.m_flags = 0;
2161 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2163 struct inode *inode = mpd->inode;
2164 struct ext4_map_blocks *map = &mpd->map;
2165 int get_blocks_flags;
2166 int err, dioread_nolock;
2168 trace_ext4_da_write_pages_extent(inode, map);
2170 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2171 * to convert an unwritten extent to be initialized (in the case
2172 * where we have written into one or more preallocated blocks). It is
2173 * possible that we're going to need more metadata blocks than
2174 * previously reserved. However we must not fail because we're in
2175 * writeback and there is nothing we can do about it so it might result
2176 * in data loss. So use reserved blocks to allocate metadata if
2179 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2180 * the blocks in question are delalloc blocks. This indicates
2181 * that the blocks and quotas has already been checked when
2182 * the data was copied into the page cache.
2184 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2185 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2186 dioread_nolock = ext4_should_dioread_nolock(inode);
2188 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2189 if (map->m_flags & (1 << BH_Delay))
2190 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2192 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2195 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2196 if (!mpd->io_submit.io_end->handle &&
2197 ext4_handle_valid(handle)) {
2198 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2199 handle->h_rsv_handle = NULL;
2201 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2204 BUG_ON(map->m_len == 0);
2205 if (map->m_flags & EXT4_MAP_NEW) {
2206 struct block_device *bdev = inode->i_sb->s_bdev;
2209 for (i = 0; i < map->m_len; i++)
2210 unmap_underlying_metadata(bdev, map->m_pblk + i);
2216 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2217 * mpd->len and submit pages underlying it for IO
2219 * @handle - handle for journal operations
2220 * @mpd - extent to map
2221 * @give_up_on_write - we set this to true iff there is a fatal error and there
2222 * is no hope of writing the data. The caller should discard
2223 * dirty pages to avoid infinite loops.
2225 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2226 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2227 * them to initialized or split the described range from larger unwritten
2228 * extent. Note that we need not map all the described range since allocation
2229 * can return less blocks or the range is covered by more unwritten extents. We
2230 * cannot map more because we are limited by reserved transaction credits. On
2231 * the other hand we always make sure that the last touched page is fully
2232 * mapped so that it can be written out (and thus forward progress is
2233 * guaranteed). After mapping we submit all mapped pages for IO.
2235 static int mpage_map_and_submit_extent(handle_t *handle,
2236 struct mpage_da_data *mpd,
2237 bool *give_up_on_write)
2239 struct inode *inode = mpd->inode;
2240 struct ext4_map_blocks *map = &mpd->map;
2245 mpd->io_submit.io_end->offset =
2246 ((loff_t)map->m_lblk) << inode->i_blkbits;
2248 err = mpage_map_one_extent(handle, mpd);
2250 struct super_block *sb = inode->i_sb;
2252 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2253 goto invalidate_dirty_pages;
2255 * Let the uper layers retry transient errors.
2256 * In the case of ENOSPC, if ext4_count_free_blocks()
2257 * is non-zero, a commit should free up blocks.
2259 if ((err == -ENOMEM) ||
2260 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2262 goto update_disksize;
2265 ext4_msg(sb, KERN_CRIT,
2266 "Delayed block allocation failed for "
2267 "inode %lu at logical offset %llu with"
2268 " max blocks %u with error %d",
2270 (unsigned long long)map->m_lblk,
2271 (unsigned)map->m_len, -err);
2272 ext4_msg(sb, KERN_CRIT,
2273 "This should not happen!! Data will "
2276 ext4_print_free_blocks(inode);
2277 invalidate_dirty_pages:
2278 *give_up_on_write = true;
2283 * Update buffer state, submit mapped pages, and get us new
2286 err = mpage_map_and_submit_buffers(mpd);
2288 goto update_disksize;
2289 } while (map->m_len);
2293 * Update on-disk size after IO is submitted. Races with
2294 * truncate are avoided by checking i_size under i_data_sem.
2296 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2297 if (disksize > EXT4_I(inode)->i_disksize) {
2301 down_write(&EXT4_I(inode)->i_data_sem);
2302 i_size = i_size_read(inode);
2303 if (disksize > i_size)
2305 if (disksize > EXT4_I(inode)->i_disksize)
2306 EXT4_I(inode)->i_disksize = disksize;
2307 err2 = ext4_mark_inode_dirty(handle, inode);
2308 up_write(&EXT4_I(inode)->i_data_sem);
2310 ext4_error(inode->i_sb,
2311 "Failed to mark inode %lu dirty",
2320 * Calculate the total number of credits to reserve for one writepages
2321 * iteration. This is called from ext4_writepages(). We map an extent of
2322 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2323 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2324 * bpp - 1 blocks in bpp different extents.
2326 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2328 int bpp = ext4_journal_blocks_per_page(inode);
2330 return ext4_meta_trans_blocks(inode,
2331 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2335 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2336 * and underlying extent to map
2338 * @mpd - where to look for pages
2340 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2341 * IO immediately. When we find a page which isn't mapped we start accumulating
2342 * extent of buffers underlying these pages that needs mapping (formed by
2343 * either delayed or unwritten buffers). We also lock the pages containing
2344 * these buffers. The extent found is returned in @mpd structure (starting at
2345 * mpd->lblk with length mpd->len blocks).
2347 * Note that this function can attach bios to one io_end structure which are
2348 * neither logically nor physically contiguous. Although it may seem as an
2349 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2350 * case as we need to track IO to all buffers underlying a page in one io_end.
2352 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2354 struct address_space *mapping = mpd->inode->i_mapping;
2355 struct pagevec pvec;
2356 unsigned int nr_pages;
2357 long left = mpd->wbc->nr_to_write;
2358 pgoff_t index = mpd->first_page;
2359 pgoff_t end = mpd->last_page;
2362 int blkbits = mpd->inode->i_blkbits;
2364 struct buffer_head *head;
2366 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2367 tag = PAGECACHE_TAG_TOWRITE;
2369 tag = PAGECACHE_TAG_DIRTY;
2371 pagevec_init(&pvec, 0);
2373 mpd->next_page = index;
2374 while (index <= end) {
2375 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2376 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2380 for (i = 0; i < nr_pages; i++) {
2381 struct page *page = pvec.pages[i];
2384 * At this point, the page may be truncated or
2385 * invalidated (changing page->mapping to NULL), or
2386 * even swizzled back from swapper_space to tmpfs file
2387 * mapping. However, page->index will not change
2388 * because we have a reference on the page.
2390 if (page->index > end)
2394 * Accumulated enough dirty pages? This doesn't apply
2395 * to WB_SYNC_ALL mode. For integrity sync we have to
2396 * keep going because someone may be concurrently
2397 * dirtying pages, and we might have synced a lot of
2398 * newly appeared dirty pages, but have not synced all
2399 * of the old dirty pages.
2401 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2404 /* If we can't merge this page, we are done. */
2405 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2410 * If the page is no longer dirty, or its mapping no
2411 * longer corresponds to inode we are writing (which
2412 * means it has been truncated or invalidated), or the
2413 * page is already under writeback and we are not doing
2414 * a data integrity writeback, skip the page
2416 if (!PageDirty(page) ||
2417 (PageWriteback(page) &&
2418 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2419 unlikely(page->mapping != mapping)) {
2424 wait_on_page_writeback(page);
2425 BUG_ON(PageWriteback(page));
2427 if (mpd->map.m_len == 0)
2428 mpd->first_page = page->index;
2429 mpd->next_page = page->index + 1;
2430 /* Add all dirty buffers to mpd */
2431 lblk = ((ext4_lblk_t)page->index) <<
2432 (PAGE_CACHE_SHIFT - blkbits);
2433 head = page_buffers(page);
2434 err = mpage_process_page_bufs(mpd, head, head, lblk);
2440 pagevec_release(&pvec);
2445 pagevec_release(&pvec);
2449 static int __writepage(struct page *page, struct writeback_control *wbc,
2452 struct address_space *mapping = data;
2453 int ret = ext4_writepage(page, wbc);
2454 mapping_set_error(mapping, ret);
2458 static int ext4_writepages(struct address_space *mapping,
2459 struct writeback_control *wbc)
2461 pgoff_t writeback_index = 0;
2462 long nr_to_write = wbc->nr_to_write;
2463 int range_whole = 0;
2465 handle_t *handle = NULL;
2466 struct mpage_da_data mpd;
2467 struct inode *inode = mapping->host;
2468 int needed_blocks, rsv_blocks = 0, ret = 0;
2469 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2471 struct blk_plug plug;
2472 bool give_up_on_write = false;
2474 trace_ext4_writepages(inode, wbc);
2477 * No pages to write? This is mainly a kludge to avoid starting
2478 * a transaction for special inodes like journal inode on last iput()
2479 * because that could violate lock ordering on umount
2481 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2482 goto out_writepages;
2484 if (ext4_should_journal_data(inode)) {
2485 struct blk_plug plug;
2487 blk_start_plug(&plug);
2488 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2489 blk_finish_plug(&plug);
2490 goto out_writepages;
2494 * If the filesystem has aborted, it is read-only, so return
2495 * right away instead of dumping stack traces later on that
2496 * will obscure the real source of the problem. We test
2497 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2498 * the latter could be true if the filesystem is mounted
2499 * read-only, and in that case, ext4_writepages should
2500 * *never* be called, so if that ever happens, we would want
2503 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2505 goto out_writepages;
2508 if (ext4_should_dioread_nolock(inode)) {
2510 * We may need to convert up to one extent per block in
2511 * the page and we may dirty the inode.
2513 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2517 * If we have inline data and arrive here, it means that
2518 * we will soon create the block for the 1st page, so
2519 * we'd better clear the inline data here.
2521 if (ext4_has_inline_data(inode)) {
2522 /* Just inode will be modified... */
2523 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2524 if (IS_ERR(handle)) {
2525 ret = PTR_ERR(handle);
2526 goto out_writepages;
2528 BUG_ON(ext4_test_inode_state(inode,
2529 EXT4_STATE_MAY_INLINE_DATA));
2530 ext4_destroy_inline_data(handle, inode);
2531 ext4_journal_stop(handle);
2534 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2537 if (wbc->range_cyclic) {
2538 writeback_index = mapping->writeback_index;
2539 if (writeback_index)
2541 mpd.first_page = writeback_index;
2544 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2545 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2550 ext4_io_submit_init(&mpd.io_submit, wbc);
2552 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2553 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2555 blk_start_plug(&plug);
2556 while (!done && mpd.first_page <= mpd.last_page) {
2557 /* For each extent of pages we use new io_end */
2558 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2559 if (!mpd.io_submit.io_end) {
2565 * We have two constraints: We find one extent to map and we
2566 * must always write out whole page (makes a difference when
2567 * blocksize < pagesize) so that we don't block on IO when we
2568 * try to write out the rest of the page. Journalled mode is
2569 * not supported by delalloc.
2571 BUG_ON(ext4_should_journal_data(inode));
2572 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2574 /* start a new transaction */
2575 handle = ext4_journal_start_with_reserve(inode,
2576 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2577 if (IS_ERR(handle)) {
2578 ret = PTR_ERR(handle);
2579 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2580 "%ld pages, ino %lu; err %d", __func__,
2581 wbc->nr_to_write, inode->i_ino, ret);
2582 /* Release allocated io_end */
2583 ext4_put_io_end(mpd.io_submit.io_end);
2587 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2588 ret = mpage_prepare_extent_to_map(&mpd);
2591 ret = mpage_map_and_submit_extent(handle, &mpd,
2595 * We scanned the whole range (or exhausted
2596 * nr_to_write), submitted what was mapped and
2597 * didn't find anything needing mapping. We are
2604 * Caution: If the handle is synchronous,
2605 * ext4_journal_stop() can wait for transaction commit
2606 * to finish which may depend on writeback of pages to
2607 * complete or on page lock to be released. In that
2608 * case, we have to wait until after after we have
2609 * submitted all the IO, released page locks we hold,
2610 * and dropped io_end reference (for extent conversion
2611 * to be able to complete) before stopping the handle.
2613 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2614 ext4_journal_stop(handle);
2617 /* Submit prepared bio */
2618 ext4_io_submit(&mpd.io_submit);
2619 /* Unlock pages we didn't use */
2620 mpage_release_unused_pages(&mpd, give_up_on_write);
2622 * Drop our io_end reference we got from init. We have
2623 * to be careful and use deferred io_end finishing if
2624 * we are still holding the transaction as we can
2625 * release the last reference to io_end which may end
2626 * up doing unwritten extent conversion.
2629 ext4_put_io_end_defer(mpd.io_submit.io_end);
2630 ext4_journal_stop(handle);
2632 ext4_put_io_end(mpd.io_submit.io_end);
2634 if (ret == -ENOSPC && sbi->s_journal) {
2636 * Commit the transaction which would
2637 * free blocks released in the transaction
2640 jbd2_journal_force_commit_nested(sbi->s_journal);
2644 /* Fatal error - ENOMEM, EIO... */
2648 blk_finish_plug(&plug);
2649 if (!ret && !cycled && wbc->nr_to_write > 0) {
2651 mpd.last_page = writeback_index - 1;
2657 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2659 * Set the writeback_index so that range_cyclic
2660 * mode will write it back later
2662 mapping->writeback_index = mpd.first_page;
2665 trace_ext4_writepages_result(inode, wbc, ret,
2666 nr_to_write - wbc->nr_to_write);
2670 static int ext4_nonda_switch(struct super_block *sb)
2672 s64 free_clusters, dirty_clusters;
2673 struct ext4_sb_info *sbi = EXT4_SB(sb);
2676 * switch to non delalloc mode if we are running low
2677 * on free block. The free block accounting via percpu
2678 * counters can get slightly wrong with percpu_counter_batch getting
2679 * accumulated on each CPU without updating global counters
2680 * Delalloc need an accurate free block accounting. So switch
2681 * to non delalloc when we are near to error range.
2684 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2686 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2688 * Start pushing delalloc when 1/2 of free blocks are dirty.
2690 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2691 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2693 if (2 * free_clusters < 3 * dirty_clusters ||
2694 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2696 * free block count is less than 150% of dirty blocks
2697 * or free blocks is less than watermark
2704 /* We always reserve for an inode update; the superblock could be there too */
2705 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2707 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2710 if (pos + len <= 0x7fffffffULL)
2713 /* We might need to update the superblock to set LARGE_FILE */
2717 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2718 loff_t pos, unsigned len, unsigned flags,
2719 struct page **pagep, void **fsdata)
2721 int ret, retries = 0;
2724 struct inode *inode = mapping->host;
2727 index = pos >> PAGE_CACHE_SHIFT;
2729 if (ext4_nonda_switch(inode->i_sb)) {
2730 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2731 return ext4_write_begin(file, mapping, pos,
2732 len, flags, pagep, fsdata);
2734 *fsdata = (void *)0;
2735 trace_android_fs_datawrite_start(inode, pos, len,
2736 current->pid, current->comm);
2737 trace_ext4_da_write_begin(inode, pos, len, flags);
2739 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2740 ret = ext4_da_write_inline_data_begin(mapping, inode,
2750 * grab_cache_page_write_begin() can take a long time if the
2751 * system is thrashing due to memory pressure, or if the page
2752 * is being written back. So grab it first before we start
2753 * the transaction handle. This also allows us to allocate
2754 * the page (if needed) without using GFP_NOFS.
2757 page = grab_cache_page_write_begin(mapping, index, flags);
2763 * With delayed allocation, we don't log the i_disksize update
2764 * if there is delayed block allocation. But we still need
2765 * to journalling the i_disksize update if writes to the end
2766 * of file which has an already mapped buffer.
2769 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2770 ext4_da_write_credits(inode, pos, len));
2771 if (IS_ERR(handle)) {
2772 page_cache_release(page);
2773 return PTR_ERR(handle);
2777 if (page->mapping != mapping) {
2778 /* The page got truncated from under us */
2780 page_cache_release(page);
2781 ext4_journal_stop(handle);
2784 /* In case writeback began while the page was unlocked */
2785 wait_for_stable_page(page);
2787 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2788 ret = ext4_block_write_begin(page, pos, len,
2789 ext4_da_get_block_prep);
2791 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2795 ext4_journal_stop(handle);
2797 * block_write_begin may have instantiated a few blocks
2798 * outside i_size. Trim these off again. Don't need
2799 * i_size_read because we hold i_mutex.
2801 if (pos + len > inode->i_size)
2802 ext4_truncate_failed_write(inode);
2804 if (ret == -ENOSPC &&
2805 ext4_should_retry_alloc(inode->i_sb, &retries))
2808 page_cache_release(page);
2817 * Check if we should update i_disksize
2818 * when write to the end of file but not require block allocation
2820 static int ext4_da_should_update_i_disksize(struct page *page,
2821 unsigned long offset)
2823 struct buffer_head *bh;
2824 struct inode *inode = page->mapping->host;
2828 bh = page_buffers(page);
2829 idx = offset >> inode->i_blkbits;
2831 for (i = 0; i < idx; i++)
2832 bh = bh->b_this_page;
2834 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2839 static int ext4_da_write_end(struct file *file,
2840 struct address_space *mapping,
2841 loff_t pos, unsigned len, unsigned copied,
2842 struct page *page, void *fsdata)
2844 struct inode *inode = mapping->host;
2846 handle_t *handle = ext4_journal_current_handle();
2848 unsigned long start, end;
2849 int write_mode = (int)(unsigned long)fsdata;
2851 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2852 return ext4_write_end(file, mapping, pos,
2853 len, copied, page, fsdata);
2855 trace_android_fs_datawrite_end(inode, pos, len);
2856 trace_ext4_da_write_end(inode, pos, len, copied);
2857 start = pos & (PAGE_CACHE_SIZE - 1);
2858 end = start + copied - 1;
2861 * generic_write_end() will run mark_inode_dirty() if i_size
2862 * changes. So let's piggyback the i_disksize mark_inode_dirty
2865 new_i_size = pos + copied;
2866 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2867 if (ext4_has_inline_data(inode) ||
2868 ext4_da_should_update_i_disksize(page, end)) {
2869 ext4_update_i_disksize(inode, new_i_size);
2870 /* We need to mark inode dirty even if
2871 * new_i_size is less that inode->i_size
2872 * bu greater than i_disksize.(hint delalloc)
2874 ext4_mark_inode_dirty(handle, inode);
2878 if (write_mode != CONVERT_INLINE_DATA &&
2879 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2880 ext4_has_inline_data(inode))
2881 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2884 ret2 = generic_write_end(file, mapping, pos, len, copied,
2890 ret2 = ext4_journal_stop(handle);
2894 return ret ? ret : copied;
2897 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2898 unsigned int length)
2901 * Drop reserved blocks
2903 BUG_ON(!PageLocked(page));
2904 if (!page_has_buffers(page))
2907 ext4_da_page_release_reservation(page, offset, length);
2910 ext4_invalidatepage(page, offset, length);
2916 * Force all delayed allocation blocks to be allocated for a given inode.
2918 int ext4_alloc_da_blocks(struct inode *inode)
2920 trace_ext4_alloc_da_blocks(inode);
2922 if (!EXT4_I(inode)->i_reserved_data_blocks)
2926 * We do something simple for now. The filemap_flush() will
2927 * also start triggering a write of the data blocks, which is
2928 * not strictly speaking necessary (and for users of
2929 * laptop_mode, not even desirable). However, to do otherwise
2930 * would require replicating code paths in:
2932 * ext4_writepages() ->
2933 * write_cache_pages() ---> (via passed in callback function)
2934 * __mpage_da_writepage() -->
2935 * mpage_add_bh_to_extent()
2936 * mpage_da_map_blocks()
2938 * The problem is that write_cache_pages(), located in
2939 * mm/page-writeback.c, marks pages clean in preparation for
2940 * doing I/O, which is not desirable if we're not planning on
2943 * We could call write_cache_pages(), and then redirty all of
2944 * the pages by calling redirty_page_for_writepage() but that
2945 * would be ugly in the extreme. So instead we would need to
2946 * replicate parts of the code in the above functions,
2947 * simplifying them because we wouldn't actually intend to
2948 * write out the pages, but rather only collect contiguous
2949 * logical block extents, call the multi-block allocator, and
2950 * then update the buffer heads with the block allocations.
2952 * For now, though, we'll cheat by calling filemap_flush(),
2953 * which will map the blocks, and start the I/O, but not
2954 * actually wait for the I/O to complete.
2956 return filemap_flush(inode->i_mapping);
2960 * bmap() is special. It gets used by applications such as lilo and by
2961 * the swapper to find the on-disk block of a specific piece of data.
2963 * Naturally, this is dangerous if the block concerned is still in the
2964 * journal. If somebody makes a swapfile on an ext4 data-journaling
2965 * filesystem and enables swap, then they may get a nasty shock when the
2966 * data getting swapped to that swapfile suddenly gets overwritten by
2967 * the original zero's written out previously to the journal and
2968 * awaiting writeback in the kernel's buffer cache.
2970 * So, if we see any bmap calls here on a modified, data-journaled file,
2971 * take extra steps to flush any blocks which might be in the cache.
2973 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2975 struct inode *inode = mapping->host;
2980 * We can get here for an inline file via the FIBMAP ioctl
2982 if (ext4_has_inline_data(inode))
2985 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2986 test_opt(inode->i_sb, DELALLOC)) {
2988 * With delalloc we want to sync the file
2989 * so that we can make sure we allocate
2992 filemap_write_and_wait(mapping);
2995 if (EXT4_JOURNAL(inode) &&
2996 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2998 * This is a REALLY heavyweight approach, but the use of
2999 * bmap on dirty files is expected to be extremely rare:
3000 * only if we run lilo or swapon on a freshly made file
3001 * do we expect this to happen.
3003 * (bmap requires CAP_SYS_RAWIO so this does not
3004 * represent an unprivileged user DOS attack --- we'd be
3005 * in trouble if mortal users could trigger this path at
3008 * NB. EXT4_STATE_JDATA is not set on files other than
3009 * regular files. If somebody wants to bmap a directory
3010 * or symlink and gets confused because the buffer
3011 * hasn't yet been flushed to disk, they deserve
3012 * everything they get.
3015 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3016 journal = EXT4_JOURNAL(inode);
3017 jbd2_journal_lock_updates(journal);
3018 err = jbd2_journal_flush(journal);
3019 jbd2_journal_unlock_updates(journal);
3025 return generic_block_bmap(mapping, block, ext4_get_block);
3028 static int ext4_readpage(struct file *file, struct page *page)
3031 struct inode *inode = page->mapping->host;
3033 trace_ext4_readpage(page);
3035 if (ext4_has_inline_data(inode))
3036 ret = ext4_readpage_inline(inode, page);
3039 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3045 ext4_readpages(struct file *file, struct address_space *mapping,
3046 struct list_head *pages, unsigned nr_pages)
3048 struct inode *inode = mapping->host;
3050 /* If the file has inline data, no need to do readpages. */
3051 if (ext4_has_inline_data(inode))
3054 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3057 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3058 unsigned int length)
3060 trace_ext4_invalidatepage(page, offset, length);
3062 /* No journalling happens on data buffers when this function is used */
3063 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3065 block_invalidatepage(page, offset, length);
3068 static int __ext4_journalled_invalidatepage(struct page *page,
3069 unsigned int offset,
3070 unsigned int length)
3072 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3074 trace_ext4_journalled_invalidatepage(page, offset, length);
3077 * If it's a full truncate we just forget about the pending dirtying
3079 if (offset == 0 && length == PAGE_CACHE_SIZE)
3080 ClearPageChecked(page);
3082 return jbd2_journal_invalidatepage(journal, page, offset, length);
3085 /* Wrapper for aops... */
3086 static void ext4_journalled_invalidatepage(struct page *page,
3087 unsigned int offset,
3088 unsigned int length)
3090 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3093 static int ext4_releasepage(struct page *page, gfp_t wait)
3095 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3097 trace_ext4_releasepage(page);
3099 /* Page has dirty journalled data -> cannot release */
3100 if (PageChecked(page))
3103 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3105 return try_to_free_buffers(page);
3109 * ext4_get_block used when preparing for a DIO write or buffer write.
3110 * We allocate an uinitialized extent if blocks haven't been allocated.
3111 * The extent will be converted to initialized after the IO is complete.
3113 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3114 struct buffer_head *bh_result, int create)
3116 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3117 inode->i_ino, create);
3118 return _ext4_get_block(inode, iblock, bh_result,
3119 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3122 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3123 struct buffer_head *bh_result, int create)
3125 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3126 inode->i_ino, create);
3127 return _ext4_get_block(inode, iblock, bh_result,
3128 EXT4_GET_BLOCKS_NO_LOCK);
3131 int ext4_get_block_dax(struct inode *inode, sector_t iblock,
3132 struct buffer_head *bh_result, int create)
3134 int flags = EXT4_GET_BLOCKS_PRE_IO | EXT4_GET_BLOCKS_UNWRIT_EXT;
3136 flags |= EXT4_GET_BLOCKS_CREATE;
3137 ext4_debug("ext4_get_block_dax: inode %lu, create flag %d\n",
3138 inode->i_ino, create);
3139 return _ext4_get_block(inode, iblock, bh_result, flags);
3142 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3143 ssize_t size, void *private)
3145 ext4_io_end_t *io_end = iocb->private;
3147 /* if not async direct IO just return */
3151 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3152 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3153 iocb->private, io_end->inode->i_ino, iocb, offset,
3156 iocb->private = NULL;
3157 io_end->offset = offset;
3158 io_end->size = size;
3159 ext4_put_io_end(io_end);
3163 * For ext4 extent files, ext4 will do direct-io write to holes,
3164 * preallocated extents, and those write extend the file, no need to
3165 * fall back to buffered IO.
3167 * For holes, we fallocate those blocks, mark them as unwritten
3168 * If those blocks were preallocated, we mark sure they are split, but
3169 * still keep the range to write as unwritten.
3171 * The unwritten extents will be converted to written when DIO is completed.
3172 * For async direct IO, since the IO may still pending when return, we
3173 * set up an end_io call back function, which will do the conversion
3174 * when async direct IO completed.
3176 * If the O_DIRECT write will extend the file then add this inode to the
3177 * orphan list. So recovery will truncate it back to the original size
3178 * if the machine crashes during the write.
3181 static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3184 struct file *file = iocb->ki_filp;
3185 struct inode *inode = file->f_mapping->host;
3187 size_t count = iov_iter_count(iter);
3189 get_block_t *get_block_func = NULL;
3191 loff_t final_size = offset + count;
3192 ext4_io_end_t *io_end = NULL;
3194 /* Use the old path for reads and writes beyond i_size. */
3195 if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size)
3196 return ext4_ind_direct_IO(iocb, iter, offset);
3198 BUG_ON(iocb->private == NULL);
3201 * Make all waiters for direct IO properly wait also for extent
3202 * conversion. This also disallows race between truncate() and
3203 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3205 if (iov_iter_rw(iter) == WRITE)
3206 inode_dio_begin(inode);
3208 /* If we do a overwrite dio, i_mutex locking can be released */
3209 overwrite = *((int *)iocb->private);
3212 down_read(&EXT4_I(inode)->i_data_sem);
3213 mutex_unlock(&inode->i_mutex);
3217 * We could direct write to holes and fallocate.
3219 * Allocated blocks to fill the hole are marked as
3220 * unwritten to prevent parallel buffered read to expose
3221 * the stale data before DIO complete the data IO.
3223 * As to previously fallocated extents, ext4 get_block will
3224 * just simply mark the buffer mapped but still keep the
3225 * extents unwritten.
3227 * For non AIO case, we will convert those unwritten extents
3228 * to written after return back from blockdev_direct_IO.
3230 * For async DIO, the conversion needs to be deferred when the
3231 * IO is completed. The ext4 end_io callback function will be
3232 * called to take care of the conversion work. Here for async
3233 * case, we allocate an io_end structure to hook to the iocb.
3235 iocb->private = NULL;
3236 ext4_inode_aio_set(inode, NULL);
3237 if (!is_sync_kiocb(iocb)) {
3238 io_end = ext4_init_io_end(inode, GFP_NOFS);
3244 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3246 iocb->private = ext4_get_io_end(io_end);
3248 * we save the io structure for current async direct
3249 * IO, so that later ext4_map_blocks() could flag the
3250 * io structure whether there is a unwritten extents
3251 * needs to be converted when IO is completed.
3253 ext4_inode_aio_set(inode, io_end);
3257 get_block_func = ext4_get_block_write_nolock;
3259 get_block_func = ext4_get_block_write;
3260 dio_flags = DIO_LOCKING;
3262 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3263 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3266 ret = dax_do_io(iocb, inode, iter, offset, get_block_func,
3267 ext4_end_io_dio, dio_flags);
3269 ret = __blockdev_direct_IO(iocb, inode,
3270 inode->i_sb->s_bdev, iter, offset,
3272 ext4_end_io_dio, NULL, dio_flags);
3275 * Put our reference to io_end. This can free the io_end structure e.g.
3276 * in sync IO case or in case of error. It can even perform extent
3277 * conversion if all bios we submitted finished before we got here.
3278 * Note that in that case iocb->private can be already set to NULL
3282 ext4_inode_aio_set(inode, NULL);
3283 ext4_put_io_end(io_end);
3285 * When no IO was submitted ext4_end_io_dio() was not
3286 * called so we have to put iocb's reference.
3288 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3289 WARN_ON(iocb->private != io_end);
3290 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3291 ext4_put_io_end(io_end);
3292 iocb->private = NULL;
3295 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3296 EXT4_STATE_DIO_UNWRITTEN)) {
3299 * for non AIO case, since the IO is already
3300 * completed, we could do the conversion right here
3302 err = ext4_convert_unwritten_extents(NULL, inode,
3306 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3310 if (iov_iter_rw(iter) == WRITE)
3311 inode_dio_end(inode);
3312 /* take i_mutex locking again if we do a ovewrite dio */
3314 up_read(&EXT4_I(inode)->i_data_sem);
3315 mutex_lock(&inode->i_mutex);
3321 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3324 struct file *file = iocb->ki_filp;
3325 struct inode *inode = file->f_mapping->host;
3326 size_t count = iov_iter_count(iter);
3329 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3330 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3335 * If we are doing data journalling we don't support O_DIRECT
3337 if (ext4_should_journal_data(inode))
3340 /* Let buffer I/O handle the inline data case. */
3341 if (ext4_has_inline_data(inode))
3344 if (trace_android_fs_dataread_start_enabled() &&
3345 (iov_iter_rw(iter) == READ))
3346 trace_android_fs_dataread_start(inode, offset, count,
3349 if (trace_android_fs_datawrite_start_enabled() &&
3350 (iov_iter_rw(iter) == WRITE))
3351 trace_android_fs_datawrite_start(inode, offset, count,
3355 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3356 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3357 ret = ext4_ext_direct_IO(iocb, iter, offset);
3359 ret = ext4_ind_direct_IO(iocb, iter, offset);
3360 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3362 if (trace_android_fs_dataread_start_enabled() &&
3363 (iov_iter_rw(iter) == READ))
3364 trace_android_fs_dataread_end(inode, offset, count);
3365 if (trace_android_fs_datawrite_start_enabled() &&
3366 (iov_iter_rw(iter) == WRITE))
3367 trace_android_fs_datawrite_end(inode, offset, count);
3373 * Pages can be marked dirty completely asynchronously from ext4's journalling
3374 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3375 * much here because ->set_page_dirty is called under VFS locks. The page is
3376 * not necessarily locked.
3378 * We cannot just dirty the page and leave attached buffers clean, because the
3379 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3380 * or jbddirty because all the journalling code will explode.
3382 * So what we do is to mark the page "pending dirty" and next time writepage
3383 * is called, propagate that into the buffers appropriately.
3385 static int ext4_journalled_set_page_dirty(struct page *page)
3387 SetPageChecked(page);
3388 return __set_page_dirty_nobuffers(page);
3391 static const struct address_space_operations ext4_aops = {
3392 .readpage = ext4_readpage,
3393 .readpages = ext4_readpages,
3394 .writepage = ext4_writepage,
3395 .writepages = ext4_writepages,
3396 .write_begin = ext4_write_begin,
3397 .write_end = ext4_write_end,
3399 .invalidatepage = ext4_invalidatepage,
3400 .releasepage = ext4_releasepage,
3401 .direct_IO = ext4_direct_IO,
3402 .migratepage = buffer_migrate_page,
3403 .is_partially_uptodate = block_is_partially_uptodate,
3404 .error_remove_page = generic_error_remove_page,
3407 static const struct address_space_operations ext4_journalled_aops = {
3408 .readpage = ext4_readpage,
3409 .readpages = ext4_readpages,
3410 .writepage = ext4_writepage,
3411 .writepages = ext4_writepages,
3412 .write_begin = ext4_write_begin,
3413 .write_end = ext4_journalled_write_end,
3414 .set_page_dirty = ext4_journalled_set_page_dirty,
3416 .invalidatepage = ext4_journalled_invalidatepage,
3417 .releasepage = ext4_releasepage,
3418 .direct_IO = ext4_direct_IO,
3419 .is_partially_uptodate = block_is_partially_uptodate,
3420 .error_remove_page = generic_error_remove_page,
3423 static const struct address_space_operations ext4_da_aops = {
3424 .readpage = ext4_readpage,
3425 .readpages = ext4_readpages,
3426 .writepage = ext4_writepage,
3427 .writepages = ext4_writepages,
3428 .write_begin = ext4_da_write_begin,
3429 .write_end = ext4_da_write_end,
3431 .invalidatepage = ext4_da_invalidatepage,
3432 .releasepage = ext4_releasepage,
3433 .direct_IO = ext4_direct_IO,
3434 .migratepage = buffer_migrate_page,
3435 .is_partially_uptodate = block_is_partially_uptodate,
3436 .error_remove_page = generic_error_remove_page,
3439 void ext4_set_aops(struct inode *inode)
3441 switch (ext4_inode_journal_mode(inode)) {
3442 case EXT4_INODE_ORDERED_DATA_MODE:
3443 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3445 case EXT4_INODE_WRITEBACK_DATA_MODE:
3446 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3448 case EXT4_INODE_JOURNAL_DATA_MODE:
3449 inode->i_mapping->a_ops = &ext4_journalled_aops;
3454 if (test_opt(inode->i_sb, DELALLOC))
3455 inode->i_mapping->a_ops = &ext4_da_aops;
3457 inode->i_mapping->a_ops = &ext4_aops;
3460 static int __ext4_block_zero_page_range(handle_t *handle,
3461 struct address_space *mapping, loff_t from, loff_t length)
3463 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3464 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3465 unsigned blocksize, pos;
3467 struct inode *inode = mapping->host;
3468 struct buffer_head *bh;
3472 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3473 mapping_gfp_constraint(mapping, ~__GFP_FS));
3477 blocksize = inode->i_sb->s_blocksize;
3479 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3481 if (!page_has_buffers(page))
3482 create_empty_buffers(page, blocksize, 0);
3484 /* Find the buffer that contains "offset" */
3485 bh = page_buffers(page);
3487 while (offset >= pos) {
3488 bh = bh->b_this_page;
3492 if (buffer_freed(bh)) {
3493 BUFFER_TRACE(bh, "freed: skip");
3496 if (!buffer_mapped(bh)) {
3497 BUFFER_TRACE(bh, "unmapped");
3498 ext4_get_block(inode, iblock, bh, 0);
3499 /* unmapped? It's a hole - nothing to do */
3500 if (!buffer_mapped(bh)) {
3501 BUFFER_TRACE(bh, "still unmapped");
3506 /* Ok, it's mapped. Make sure it's up-to-date */
3507 if (PageUptodate(page))
3508 set_buffer_uptodate(bh);
3510 if (!buffer_uptodate(bh)) {
3512 ll_rw_block(READ, 1, &bh);
3514 /* Uhhuh. Read error. Complain and punt. */
3515 if (!buffer_uptodate(bh))
3517 if (S_ISREG(inode->i_mode) &&
3518 ext4_encrypted_inode(inode)) {
3519 /* We expect the key to be set. */
3520 BUG_ON(!ext4_has_encryption_key(inode));
3521 BUG_ON(blocksize != PAGE_CACHE_SIZE);
3522 WARN_ON_ONCE(ext4_decrypt(page));
3525 if (ext4_should_journal_data(inode)) {
3526 BUFFER_TRACE(bh, "get write access");
3527 err = ext4_journal_get_write_access(handle, bh);
3531 zero_user(page, offset, length);
3532 BUFFER_TRACE(bh, "zeroed end of block");
3534 if (ext4_should_journal_data(inode)) {
3535 err = ext4_handle_dirty_metadata(handle, inode, bh);
3538 mark_buffer_dirty(bh);
3539 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3540 err = ext4_jbd2_file_inode(handle, inode);
3545 page_cache_release(page);
3550 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3551 * starting from file offset 'from'. The range to be zero'd must
3552 * be contained with in one block. If the specified range exceeds
3553 * the end of the block it will be shortened to end of the block
3554 * that cooresponds to 'from'
3556 static int ext4_block_zero_page_range(handle_t *handle,
3557 struct address_space *mapping, loff_t from, loff_t length)
3559 struct inode *inode = mapping->host;
3560 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3561 unsigned blocksize = inode->i_sb->s_blocksize;
3562 unsigned max = blocksize - (offset & (blocksize - 1));
3565 * correct length if it does not fall between
3566 * 'from' and the end of the block
3568 if (length > max || length < 0)
3572 return dax_zero_page_range(inode, from, length, ext4_get_block);
3573 return __ext4_block_zero_page_range(handle, mapping, from, length);
3577 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3578 * up to the end of the block which corresponds to `from'.
3579 * This required during truncate. We need to physically zero the tail end
3580 * of that block so it doesn't yield old data if the file is later grown.
3582 static int ext4_block_truncate_page(handle_t *handle,
3583 struct address_space *mapping, loff_t from)
3585 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3588 struct inode *inode = mapping->host;
3590 blocksize = inode->i_sb->s_blocksize;
3591 length = blocksize - (offset & (blocksize - 1));
3593 return ext4_block_zero_page_range(handle, mapping, from, length);
3596 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3597 loff_t lstart, loff_t length)
3599 struct super_block *sb = inode->i_sb;
3600 struct address_space *mapping = inode->i_mapping;
3601 unsigned partial_start, partial_end;
3602 ext4_fsblk_t start, end;
3603 loff_t byte_end = (lstart + length - 1);
3606 partial_start = lstart & (sb->s_blocksize - 1);
3607 partial_end = byte_end & (sb->s_blocksize - 1);
3609 start = lstart >> sb->s_blocksize_bits;
3610 end = byte_end >> sb->s_blocksize_bits;
3612 /* Handle partial zero within the single block */
3614 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3615 err = ext4_block_zero_page_range(handle, mapping,
3619 /* Handle partial zero out on the start of the range */
3620 if (partial_start) {
3621 err = ext4_block_zero_page_range(handle, mapping,
3622 lstart, sb->s_blocksize);
3626 /* Handle partial zero out on the end of the range */
3627 if (partial_end != sb->s_blocksize - 1)
3628 err = ext4_block_zero_page_range(handle, mapping,
3629 byte_end - partial_end,
3634 int ext4_can_truncate(struct inode *inode)
3636 if (S_ISREG(inode->i_mode))
3638 if (S_ISDIR(inode->i_mode))
3640 if (S_ISLNK(inode->i_mode))
3641 return !ext4_inode_is_fast_symlink(inode);
3646 * We have to make sure i_disksize gets properly updated before we truncate
3647 * page cache due to hole punching or zero range. Otherwise i_disksize update
3648 * can get lost as it may have been postponed to submission of writeback but
3649 * that will never happen after we truncate page cache.
3651 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3655 loff_t size = i_size_read(inode);
3657 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3658 if (offset > size || offset + len < size)
3661 if (EXT4_I(inode)->i_disksize >= size)
3664 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3666 return PTR_ERR(handle);
3667 ext4_update_i_disksize(inode, size);
3668 ext4_mark_inode_dirty(handle, inode);
3669 ext4_journal_stop(handle);
3675 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3676 * associated with the given offset and length
3678 * @inode: File inode
3679 * @offset: The offset where the hole will begin
3680 * @len: The length of the hole
3682 * Returns: 0 on success or negative on failure
3685 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3687 struct super_block *sb = inode->i_sb;
3688 ext4_lblk_t first_block, stop_block;
3689 struct address_space *mapping = inode->i_mapping;
3690 loff_t first_block_offset, last_block_offset;
3692 unsigned int credits;
3695 if (!S_ISREG(inode->i_mode))
3698 trace_ext4_punch_hole(inode, offset, length, 0);
3701 * Write out all dirty pages to avoid race conditions
3702 * Then release them.
3704 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3705 ret = filemap_write_and_wait_range(mapping, offset,
3706 offset + length - 1);
3711 mutex_lock(&inode->i_mutex);
3713 /* No need to punch hole beyond i_size */
3714 if (offset >= inode->i_size)
3718 * If the hole extends beyond i_size, set the hole
3719 * to end after the page that contains i_size
3721 if (offset + length > inode->i_size) {
3722 length = inode->i_size +
3723 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3727 if (offset & (sb->s_blocksize - 1) ||
3728 (offset + length) & (sb->s_blocksize - 1)) {
3730 * Attach jinode to inode for jbd2 if we do any zeroing of
3733 ret = ext4_inode_attach_jinode(inode);
3739 /* Wait all existing dio workers, newcomers will block on i_mutex */
3740 ext4_inode_block_unlocked_dio(inode);
3741 inode_dio_wait(inode);
3744 * Prevent page faults from reinstantiating pages we have released from
3747 down_write(&EXT4_I(inode)->i_mmap_sem);
3748 first_block_offset = round_up(offset, sb->s_blocksize);
3749 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3751 /* Now release the pages and zero block aligned part of pages*/
3752 if (last_block_offset > first_block_offset) {
3753 ret = ext4_update_disksize_before_punch(inode, offset, length);
3756 truncate_pagecache_range(inode, first_block_offset,
3760 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3761 credits = ext4_writepage_trans_blocks(inode);
3763 credits = ext4_blocks_for_truncate(inode);
3764 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3765 if (IS_ERR(handle)) {
3766 ret = PTR_ERR(handle);
3767 ext4_std_error(sb, ret);
3771 ret = ext4_zero_partial_blocks(handle, inode, offset,
3776 first_block = (offset + sb->s_blocksize - 1) >>
3777 EXT4_BLOCK_SIZE_BITS(sb);
3778 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3780 /* If there are no blocks to remove, return now */
3781 if (first_block >= stop_block)
3784 down_write(&EXT4_I(inode)->i_data_sem);
3785 ext4_discard_preallocations(inode);
3787 ret = ext4_es_remove_extent(inode, first_block,
3788 stop_block - first_block);
3790 up_write(&EXT4_I(inode)->i_data_sem);
3794 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3795 ret = ext4_ext_remove_space(inode, first_block,
3798 ret = ext4_ind_remove_space(handle, inode, first_block,
3801 up_write(&EXT4_I(inode)->i_data_sem);
3803 ext4_handle_sync(handle);
3805 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3806 ext4_mark_inode_dirty(handle, inode);
3808 ext4_journal_stop(handle);
3810 up_write(&EXT4_I(inode)->i_mmap_sem);
3811 ext4_inode_resume_unlocked_dio(inode);
3813 mutex_unlock(&inode->i_mutex);
3817 int ext4_inode_attach_jinode(struct inode *inode)
3819 struct ext4_inode_info *ei = EXT4_I(inode);
3820 struct jbd2_inode *jinode;
3822 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3825 jinode = jbd2_alloc_inode(GFP_KERNEL);
3826 spin_lock(&inode->i_lock);
3829 spin_unlock(&inode->i_lock);
3832 ei->jinode = jinode;
3833 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3836 spin_unlock(&inode->i_lock);
3837 if (unlikely(jinode != NULL))
3838 jbd2_free_inode(jinode);
3845 * We block out ext4_get_block() block instantiations across the entire
3846 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3847 * simultaneously on behalf of the same inode.
3849 * As we work through the truncate and commit bits of it to the journal there
3850 * is one core, guiding principle: the file's tree must always be consistent on
3851 * disk. We must be able to restart the truncate after a crash.
3853 * The file's tree may be transiently inconsistent in memory (although it
3854 * probably isn't), but whenever we close off and commit a journal transaction,
3855 * the contents of (the filesystem + the journal) must be consistent and
3856 * restartable. It's pretty simple, really: bottom up, right to left (although
3857 * left-to-right works OK too).
3859 * Note that at recovery time, journal replay occurs *before* the restart of
3860 * truncate against the orphan inode list.
3862 * The committed inode has the new, desired i_size (which is the same as
3863 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3864 * that this inode's truncate did not complete and it will again call
3865 * ext4_truncate() to have another go. So there will be instantiated blocks
3866 * to the right of the truncation point in a crashed ext4 filesystem. But
3867 * that's fine - as long as they are linked from the inode, the post-crash
3868 * ext4_truncate() run will find them and release them.
3870 void ext4_truncate(struct inode *inode)
3872 struct ext4_inode_info *ei = EXT4_I(inode);
3873 unsigned int credits;
3875 struct address_space *mapping = inode->i_mapping;
3878 * There is a possibility that we're either freeing the inode
3879 * or it's a completely new inode. In those cases we might not
3880 * have i_mutex locked because it's not necessary.
3882 if (!(inode->i_state & (I_NEW|I_FREEING)))
3883 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3884 trace_ext4_truncate_enter(inode);
3886 if (!ext4_can_truncate(inode))
3889 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3891 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3892 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3894 if (ext4_has_inline_data(inode)) {
3897 ext4_inline_data_truncate(inode, &has_inline);
3902 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3903 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3904 if (ext4_inode_attach_jinode(inode) < 0)
3908 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3909 credits = ext4_writepage_trans_blocks(inode);
3911 credits = ext4_blocks_for_truncate(inode);
3913 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3914 if (IS_ERR(handle)) {
3915 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3919 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3920 ext4_block_truncate_page(handle, mapping, inode->i_size);
3923 * We add the inode to the orphan list, so that if this
3924 * truncate spans multiple transactions, and we crash, we will
3925 * resume the truncate when the filesystem recovers. It also
3926 * marks the inode dirty, to catch the new size.
3928 * Implication: the file must always be in a sane, consistent
3929 * truncatable state while each transaction commits.
3931 if (ext4_orphan_add(handle, inode))
3934 down_write(&EXT4_I(inode)->i_data_sem);
3936 ext4_discard_preallocations(inode);
3938 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3939 ext4_ext_truncate(handle, inode);
3941 ext4_ind_truncate(handle, inode);
3943 up_write(&ei->i_data_sem);
3946 ext4_handle_sync(handle);
3950 * If this was a simple ftruncate() and the file will remain alive,
3951 * then we need to clear up the orphan record which we created above.
3952 * However, if this was a real unlink then we were called by
3953 * ext4_evict_inode(), and we allow that function to clean up the
3954 * orphan info for us.
3957 ext4_orphan_del(handle, inode);
3959 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3960 ext4_mark_inode_dirty(handle, inode);
3961 ext4_journal_stop(handle);
3963 trace_ext4_truncate_exit(inode);
3967 * ext4_get_inode_loc returns with an extra refcount against the inode's
3968 * underlying buffer_head on success. If 'in_mem' is true, we have all
3969 * data in memory that is needed to recreate the on-disk version of this
3972 static int __ext4_get_inode_loc(struct inode *inode,
3973 struct ext4_iloc *iloc, int in_mem)
3975 struct ext4_group_desc *gdp;
3976 struct buffer_head *bh;
3977 struct super_block *sb = inode->i_sb;
3979 int inodes_per_block, inode_offset;
3982 if (!ext4_valid_inum(sb, inode->i_ino))
3983 return -EFSCORRUPTED;
3985 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3986 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3991 * Figure out the offset within the block group inode table
3993 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3994 inode_offset = ((inode->i_ino - 1) %
3995 EXT4_INODES_PER_GROUP(sb));
3996 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3997 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3999 bh = sb_getblk(sb, block);
4002 if (!buffer_uptodate(bh)) {
4006 * If the buffer has the write error flag, we have failed
4007 * to write out another inode in the same block. In this
4008 * case, we don't have to read the block because we may
4009 * read the old inode data successfully.
4011 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4012 set_buffer_uptodate(bh);
4014 if (buffer_uptodate(bh)) {
4015 /* someone brought it uptodate while we waited */
4021 * If we have all information of the inode in memory and this
4022 * is the only valid inode in the block, we need not read the
4026 struct buffer_head *bitmap_bh;
4029 start = inode_offset & ~(inodes_per_block - 1);
4031 /* Is the inode bitmap in cache? */
4032 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4033 if (unlikely(!bitmap_bh))
4037 * If the inode bitmap isn't in cache then the
4038 * optimisation may end up performing two reads instead
4039 * of one, so skip it.
4041 if (!buffer_uptodate(bitmap_bh)) {
4045 for (i = start; i < start + inodes_per_block; i++) {
4046 if (i == inode_offset)
4048 if (ext4_test_bit(i, bitmap_bh->b_data))
4052 if (i == start + inodes_per_block) {
4053 /* all other inodes are free, so skip I/O */
4054 memset(bh->b_data, 0, bh->b_size);
4055 set_buffer_uptodate(bh);
4063 * If we need to do any I/O, try to pre-readahead extra
4064 * blocks from the inode table.
4066 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4067 ext4_fsblk_t b, end, table;
4069 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4071 table = ext4_inode_table(sb, gdp);
4072 /* s_inode_readahead_blks is always a power of 2 */
4073 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4077 num = EXT4_INODES_PER_GROUP(sb);
4078 if (ext4_has_group_desc_csum(sb))
4079 num -= ext4_itable_unused_count(sb, gdp);
4080 table += num / inodes_per_block;
4084 sb_breadahead(sb, b++);
4088 * There are other valid inodes in the buffer, this inode
4089 * has in-inode xattrs, or we don't have this inode in memory.
4090 * Read the block from disk.
4092 trace_ext4_load_inode(inode);
4094 bh->b_end_io = end_buffer_read_sync;
4095 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4097 if (!buffer_uptodate(bh)) {
4098 EXT4_ERROR_INODE_BLOCK(inode, block,
4099 "unable to read itable block");
4109 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4111 /* We have all inode data except xattrs in memory here. */
4112 return __ext4_get_inode_loc(inode, iloc,
4113 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4116 void ext4_set_inode_flags(struct inode *inode)
4118 unsigned int flags = EXT4_I(inode)->i_flags;
4119 unsigned int new_fl = 0;
4121 if (flags & EXT4_SYNC_FL)
4123 if (flags & EXT4_APPEND_FL)
4125 if (flags & EXT4_IMMUTABLE_FL)
4126 new_fl |= S_IMMUTABLE;
4127 if (flags & EXT4_NOATIME_FL)
4128 new_fl |= S_NOATIME;
4129 if (flags & EXT4_DIRSYNC_FL)
4130 new_fl |= S_DIRSYNC;
4131 if (test_opt(inode->i_sb, DAX))
4133 inode_set_flags(inode, new_fl,
4134 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4137 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4138 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4140 unsigned int vfs_fl;
4141 unsigned long old_fl, new_fl;
4144 vfs_fl = ei->vfs_inode.i_flags;
4145 old_fl = ei->i_flags;
4146 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4147 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4149 if (vfs_fl & S_SYNC)
4150 new_fl |= EXT4_SYNC_FL;
4151 if (vfs_fl & S_APPEND)
4152 new_fl |= EXT4_APPEND_FL;
4153 if (vfs_fl & S_IMMUTABLE)
4154 new_fl |= EXT4_IMMUTABLE_FL;
4155 if (vfs_fl & S_NOATIME)
4156 new_fl |= EXT4_NOATIME_FL;
4157 if (vfs_fl & S_DIRSYNC)
4158 new_fl |= EXT4_DIRSYNC_FL;
4159 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4162 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4163 struct ext4_inode_info *ei)
4166 struct inode *inode = &(ei->vfs_inode);
4167 struct super_block *sb = inode->i_sb;
4169 if (ext4_has_feature_huge_file(sb)) {
4170 /* we are using combined 48 bit field */
4171 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4172 le32_to_cpu(raw_inode->i_blocks_lo);
4173 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4174 /* i_blocks represent file system block size */
4175 return i_blocks << (inode->i_blkbits - 9);
4180 return le32_to_cpu(raw_inode->i_blocks_lo);
4184 static inline void ext4_iget_extra_inode(struct inode *inode,
4185 struct ext4_inode *raw_inode,
4186 struct ext4_inode_info *ei)
4188 __le32 *magic = (void *)raw_inode +
4189 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4190 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4191 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4192 ext4_find_inline_data_nolock(inode);
4194 EXT4_I(inode)->i_inline_off = 0;
4197 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4199 struct ext4_iloc iloc;
4200 struct ext4_inode *raw_inode;
4201 struct ext4_inode_info *ei;
4202 struct inode *inode;
4203 journal_t *journal = EXT4_SB(sb)->s_journal;
4209 inode = iget_locked(sb, ino);
4211 return ERR_PTR(-ENOMEM);
4212 if (!(inode->i_state & I_NEW))
4218 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4221 raw_inode = ext4_raw_inode(&iloc);
4223 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4224 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4225 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4226 EXT4_INODE_SIZE(inode->i_sb)) {
4227 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4228 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4229 EXT4_INODE_SIZE(inode->i_sb));
4230 ret = -EFSCORRUPTED;
4234 ei->i_extra_isize = 0;
4236 /* Precompute checksum seed for inode metadata */
4237 if (ext4_has_metadata_csum(sb)) {
4238 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4240 __le32 inum = cpu_to_le32(inode->i_ino);
4241 __le32 gen = raw_inode->i_generation;
4242 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4244 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4248 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4249 EXT4_ERROR_INODE(inode, "checksum invalid");
4254 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4255 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4256 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4257 if (!(test_opt(inode->i_sb, NO_UID32))) {
4258 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4259 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4261 i_uid_write(inode, i_uid);
4262 i_gid_write(inode, i_gid);
4263 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4265 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4266 ei->i_inline_off = 0;
4267 ei->i_dir_start_lookup = 0;
4268 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4269 /* We now have enough fields to check if the inode was active or not.
4270 * This is needed because nfsd might try to access dead inodes
4271 * the test is that same one that e2fsck uses
4272 * NeilBrown 1999oct15
4274 if (inode->i_nlink == 0) {
4275 if ((inode->i_mode == 0 ||
4276 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4277 ino != EXT4_BOOT_LOADER_INO) {
4278 /* this inode is deleted */
4282 /* The only unlinked inodes we let through here have
4283 * valid i_mode and are being read by the orphan
4284 * recovery code: that's fine, we're about to complete
4285 * the process of deleting those.
4286 * OR it is the EXT4_BOOT_LOADER_INO which is
4287 * not initialized on a new filesystem. */
4289 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4290 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4291 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4292 if (ext4_has_feature_64bit(sb))
4294 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4295 inode->i_size = ext4_isize(raw_inode);
4296 ei->i_disksize = inode->i_size;
4298 ei->i_reserved_quota = 0;
4300 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4301 ei->i_block_group = iloc.block_group;
4302 ei->i_last_alloc_group = ~0;
4304 * NOTE! The in-memory inode i_data array is in little-endian order
4305 * even on big-endian machines: we do NOT byteswap the block numbers!
4307 for (block = 0; block < EXT4_N_BLOCKS; block++)
4308 ei->i_data[block] = raw_inode->i_block[block];
4309 INIT_LIST_HEAD(&ei->i_orphan);
4312 * Set transaction id's of transactions that have to be committed
4313 * to finish f[data]sync. We set them to currently running transaction
4314 * as we cannot be sure that the inode or some of its metadata isn't
4315 * part of the transaction - the inode could have been reclaimed and
4316 * now it is reread from disk.
4319 transaction_t *transaction;
4322 read_lock(&journal->j_state_lock);
4323 if (journal->j_running_transaction)
4324 transaction = journal->j_running_transaction;
4326 transaction = journal->j_committing_transaction;
4328 tid = transaction->t_tid;
4330 tid = journal->j_commit_sequence;
4331 read_unlock(&journal->j_state_lock);
4332 ei->i_sync_tid = tid;
4333 ei->i_datasync_tid = tid;
4336 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4337 if (ei->i_extra_isize == 0) {
4338 /* The extra space is currently unused. Use it. */
4339 ei->i_extra_isize = sizeof(struct ext4_inode) -
4340 EXT4_GOOD_OLD_INODE_SIZE;
4342 ext4_iget_extra_inode(inode, raw_inode, ei);
4346 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4347 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4348 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4349 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4351 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4352 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4353 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4354 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4356 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4361 if (ei->i_file_acl &&
4362 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4363 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4365 ret = -EFSCORRUPTED;
4367 } else if (!ext4_has_inline_data(inode)) {
4368 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4369 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4370 (S_ISLNK(inode->i_mode) &&
4371 !ext4_inode_is_fast_symlink(inode))))
4372 /* Validate extent which is part of inode */
4373 ret = ext4_ext_check_inode(inode);
4374 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4375 (S_ISLNK(inode->i_mode) &&
4376 !ext4_inode_is_fast_symlink(inode))) {
4377 /* Validate block references which are part of inode */
4378 ret = ext4_ind_check_inode(inode);
4384 if (S_ISREG(inode->i_mode)) {
4385 inode->i_op = &ext4_file_inode_operations;
4386 inode->i_fop = &ext4_file_operations;
4387 ext4_set_aops(inode);
4388 } else if (S_ISDIR(inode->i_mode)) {
4389 inode->i_op = &ext4_dir_inode_operations;
4390 inode->i_fop = &ext4_dir_operations;
4391 } else if (S_ISLNK(inode->i_mode)) {
4392 if (ext4_encrypted_inode(inode)) {
4393 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4394 ext4_set_aops(inode);
4395 } else if (ext4_inode_is_fast_symlink(inode)) {
4396 inode->i_link = (char *)ei->i_data;
4397 inode->i_op = &ext4_fast_symlink_inode_operations;
4398 nd_terminate_link(ei->i_data, inode->i_size,
4399 sizeof(ei->i_data) - 1);
4401 inode->i_op = &ext4_symlink_inode_operations;
4402 ext4_set_aops(inode);
4404 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4405 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4406 inode->i_op = &ext4_special_inode_operations;
4407 if (raw_inode->i_block[0])
4408 init_special_inode(inode, inode->i_mode,
4409 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4411 init_special_inode(inode, inode->i_mode,
4412 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4413 } else if (ino == EXT4_BOOT_LOADER_INO) {
4414 make_bad_inode(inode);
4416 ret = -EFSCORRUPTED;
4417 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4421 ext4_set_inode_flags(inode);
4422 unlock_new_inode(inode);
4428 return ERR_PTR(ret);
4431 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4433 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4434 return ERR_PTR(-EFSCORRUPTED);
4435 return ext4_iget(sb, ino);
4438 static int ext4_inode_blocks_set(handle_t *handle,
4439 struct ext4_inode *raw_inode,
4440 struct ext4_inode_info *ei)
4442 struct inode *inode = &(ei->vfs_inode);
4443 u64 i_blocks = inode->i_blocks;
4444 struct super_block *sb = inode->i_sb;
4446 if (i_blocks <= ~0U) {
4448 * i_blocks can be represented in a 32 bit variable
4449 * as multiple of 512 bytes
4451 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4452 raw_inode->i_blocks_high = 0;
4453 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4456 if (!ext4_has_feature_huge_file(sb))
4459 if (i_blocks <= 0xffffffffffffULL) {
4461 * i_blocks can be represented in a 48 bit variable
4462 * as multiple of 512 bytes
4464 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4465 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4466 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4468 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4469 /* i_block is stored in file system block size */
4470 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4471 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4472 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4477 struct other_inode {
4478 unsigned long orig_ino;
4479 struct ext4_inode *raw_inode;
4482 static int other_inode_match(struct inode * inode, unsigned long ino,
4485 struct other_inode *oi = (struct other_inode *) data;
4487 if ((inode->i_ino != ino) ||
4488 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4489 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4490 ((inode->i_state & I_DIRTY_TIME) == 0))
4492 spin_lock(&inode->i_lock);
4493 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4494 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4495 (inode->i_state & I_DIRTY_TIME)) {
4496 struct ext4_inode_info *ei = EXT4_I(inode);
4498 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4499 spin_unlock(&inode->i_lock);
4501 spin_lock(&ei->i_raw_lock);
4502 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4503 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4504 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4505 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4506 spin_unlock(&ei->i_raw_lock);
4507 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4510 spin_unlock(&inode->i_lock);
4515 * Opportunistically update the other time fields for other inodes in
4516 * the same inode table block.
4518 static void ext4_update_other_inodes_time(struct super_block *sb,
4519 unsigned long orig_ino, char *buf)
4521 struct other_inode oi;
4523 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4524 int inode_size = EXT4_INODE_SIZE(sb);
4526 oi.orig_ino = orig_ino;
4528 * Calculate the first inode in the inode table block. Inode
4529 * numbers are one-based. That is, the first inode in a block
4530 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4532 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4533 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4534 if (ino == orig_ino)
4536 oi.raw_inode = (struct ext4_inode *) buf;
4537 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4542 * Post the struct inode info into an on-disk inode location in the
4543 * buffer-cache. This gobbles the caller's reference to the
4544 * buffer_head in the inode location struct.
4546 * The caller must have write access to iloc->bh.
4548 static int ext4_do_update_inode(handle_t *handle,
4549 struct inode *inode,
4550 struct ext4_iloc *iloc)
4552 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4553 struct ext4_inode_info *ei = EXT4_I(inode);
4554 struct buffer_head *bh = iloc->bh;
4555 struct super_block *sb = inode->i_sb;
4556 int err = 0, rc, block;
4557 int need_datasync = 0, set_large_file = 0;
4561 spin_lock(&ei->i_raw_lock);
4563 /* For fields not tracked in the in-memory inode,
4564 * initialise them to zero for new inodes. */
4565 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4566 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4568 ext4_get_inode_flags(ei);
4569 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4570 i_uid = i_uid_read(inode);
4571 i_gid = i_gid_read(inode);
4572 if (!(test_opt(inode->i_sb, NO_UID32))) {
4573 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4574 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4576 * Fix up interoperability with old kernels. Otherwise, old inodes get
4577 * re-used with the upper 16 bits of the uid/gid intact
4579 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4580 raw_inode->i_uid_high = 0;
4581 raw_inode->i_gid_high = 0;
4583 raw_inode->i_uid_high =
4584 cpu_to_le16(high_16_bits(i_uid));
4585 raw_inode->i_gid_high =
4586 cpu_to_le16(high_16_bits(i_gid));
4589 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4590 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4591 raw_inode->i_uid_high = 0;
4592 raw_inode->i_gid_high = 0;
4594 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4596 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4597 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4598 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4599 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4601 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4603 spin_unlock(&ei->i_raw_lock);
4606 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4607 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4608 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4609 raw_inode->i_file_acl_high =
4610 cpu_to_le16(ei->i_file_acl >> 32);
4611 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4612 if (ei->i_disksize != ext4_isize(raw_inode)) {
4613 ext4_isize_set(raw_inode, ei->i_disksize);
4616 if (ei->i_disksize > 0x7fffffffULL) {
4617 if (!ext4_has_feature_large_file(sb) ||
4618 EXT4_SB(sb)->s_es->s_rev_level ==
4619 cpu_to_le32(EXT4_GOOD_OLD_REV))
4622 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4623 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4624 if (old_valid_dev(inode->i_rdev)) {
4625 raw_inode->i_block[0] =
4626 cpu_to_le32(old_encode_dev(inode->i_rdev));
4627 raw_inode->i_block[1] = 0;
4629 raw_inode->i_block[0] = 0;
4630 raw_inode->i_block[1] =
4631 cpu_to_le32(new_encode_dev(inode->i_rdev));
4632 raw_inode->i_block[2] = 0;
4634 } else if (!ext4_has_inline_data(inode)) {
4635 for (block = 0; block < EXT4_N_BLOCKS; block++)
4636 raw_inode->i_block[block] = ei->i_data[block];
4639 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4640 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4641 if (ei->i_extra_isize) {
4642 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4643 raw_inode->i_version_hi =
4644 cpu_to_le32(inode->i_version >> 32);
4645 raw_inode->i_extra_isize =
4646 cpu_to_le16(ei->i_extra_isize);
4649 ext4_inode_csum_set(inode, raw_inode, ei);
4650 spin_unlock(&ei->i_raw_lock);
4651 if (inode->i_sb->s_flags & MS_LAZYTIME)
4652 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4655 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4656 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4659 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4660 if (set_large_file) {
4661 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4662 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4665 ext4_update_dynamic_rev(sb);
4666 ext4_set_feature_large_file(sb);
4667 ext4_handle_sync(handle);
4668 err = ext4_handle_dirty_super(handle, sb);
4670 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4673 ext4_std_error(inode->i_sb, err);
4678 * ext4_write_inode()
4680 * We are called from a few places:
4682 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4683 * Here, there will be no transaction running. We wait for any running
4684 * transaction to commit.
4686 * - Within flush work (sys_sync(), kupdate and such).
4687 * We wait on commit, if told to.
4689 * - Within iput_final() -> write_inode_now()
4690 * We wait on commit, if told to.
4692 * In all cases it is actually safe for us to return without doing anything,
4693 * because the inode has been copied into a raw inode buffer in
4694 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4697 * Note that we are absolutely dependent upon all inode dirtiers doing the
4698 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4699 * which we are interested.
4701 * It would be a bug for them to not do this. The code:
4703 * mark_inode_dirty(inode)
4705 * inode->i_size = expr;
4707 * is in error because write_inode() could occur while `stuff()' is running,
4708 * and the new i_size will be lost. Plus the inode will no longer be on the
4709 * superblock's dirty inode list.
4711 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4715 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4718 if (EXT4_SB(inode->i_sb)->s_journal) {
4719 if (ext4_journal_current_handle()) {
4720 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4726 * No need to force transaction in WB_SYNC_NONE mode. Also
4727 * ext4_sync_fs() will force the commit after everything is
4730 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4733 err = ext4_force_commit(inode->i_sb);
4735 struct ext4_iloc iloc;
4737 err = __ext4_get_inode_loc(inode, &iloc, 0);
4741 * sync(2) will flush the whole buffer cache. No need to do
4742 * it here separately for each inode.
4744 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4745 sync_dirty_buffer(iloc.bh);
4746 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4747 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4748 "IO error syncing inode");
4757 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4758 * buffers that are attached to a page stradding i_size and are undergoing
4759 * commit. In that case we have to wait for commit to finish and try again.
4761 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4765 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4766 tid_t commit_tid = 0;
4769 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4771 * All buffers in the last page remain valid? Then there's nothing to
4772 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4775 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4778 page = find_lock_page(inode->i_mapping,
4779 inode->i_size >> PAGE_CACHE_SHIFT);
4782 ret = __ext4_journalled_invalidatepage(page, offset,
4783 PAGE_CACHE_SIZE - offset);
4785 page_cache_release(page);
4789 read_lock(&journal->j_state_lock);
4790 if (journal->j_committing_transaction)
4791 commit_tid = journal->j_committing_transaction->t_tid;
4792 read_unlock(&journal->j_state_lock);
4794 jbd2_log_wait_commit(journal, commit_tid);
4801 * Called from notify_change.
4803 * We want to trap VFS attempts to truncate the file as soon as
4804 * possible. In particular, we want to make sure that when the VFS
4805 * shrinks i_size, we put the inode on the orphan list and modify
4806 * i_disksize immediately, so that during the subsequent flushing of
4807 * dirty pages and freeing of disk blocks, we can guarantee that any
4808 * commit will leave the blocks being flushed in an unused state on
4809 * disk. (On recovery, the inode will get truncated and the blocks will
4810 * be freed, so we have a strong guarantee that no future commit will
4811 * leave these blocks visible to the user.)
4813 * Another thing we have to assure is that if we are in ordered mode
4814 * and inode is still attached to the committing transaction, we must
4815 * we start writeout of all the dirty pages which are being truncated.
4816 * This way we are sure that all the data written in the previous
4817 * transaction are already on disk (truncate waits for pages under
4820 * Called with inode->i_mutex down.
4822 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4824 struct inode *inode = d_inode(dentry);
4827 const unsigned int ia_valid = attr->ia_valid;
4829 error = inode_change_ok(inode, attr);
4833 if (is_quota_modification(inode, attr)) {
4834 error = dquot_initialize(inode);
4838 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4839 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4842 /* (user+group)*(old+new) structure, inode write (sb,
4843 * inode block, ? - but truncate inode update has it) */
4844 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4845 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4846 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4847 if (IS_ERR(handle)) {
4848 error = PTR_ERR(handle);
4851 error = dquot_transfer(inode, attr);
4853 ext4_journal_stop(handle);
4856 /* Update corresponding info in inode so that everything is in
4857 * one transaction */
4858 if (attr->ia_valid & ATTR_UID)
4859 inode->i_uid = attr->ia_uid;
4860 if (attr->ia_valid & ATTR_GID)
4861 inode->i_gid = attr->ia_gid;
4862 error = ext4_mark_inode_dirty(handle, inode);
4863 ext4_journal_stop(handle);
4866 if (attr->ia_valid & ATTR_SIZE) {
4868 loff_t oldsize = inode->i_size;
4869 int shrink = (attr->ia_size <= inode->i_size);
4871 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4872 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4874 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4877 if (!S_ISREG(inode->i_mode))
4880 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4881 inode_inc_iversion(inode);
4883 if (ext4_should_order_data(inode) &&
4884 (attr->ia_size < inode->i_size)) {
4885 error = ext4_begin_ordered_truncate(inode,
4890 if (attr->ia_size != inode->i_size) {
4891 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4892 if (IS_ERR(handle)) {
4893 error = PTR_ERR(handle);
4896 if (ext4_handle_valid(handle) && shrink) {
4897 error = ext4_orphan_add(handle, inode);
4901 * Update c/mtime on truncate up, ext4_truncate() will
4902 * update c/mtime in shrink case below
4905 inode->i_mtime = ext4_current_time(inode);
4906 inode->i_ctime = inode->i_mtime;
4908 down_write(&EXT4_I(inode)->i_data_sem);
4909 EXT4_I(inode)->i_disksize = attr->ia_size;
4910 rc = ext4_mark_inode_dirty(handle, inode);
4914 * We have to update i_size under i_data_sem together
4915 * with i_disksize to avoid races with writeback code
4916 * running ext4_wb_update_i_disksize().
4919 i_size_write(inode, attr->ia_size);
4920 up_write(&EXT4_I(inode)->i_data_sem);
4921 ext4_journal_stop(handle);
4924 ext4_orphan_del(NULL, inode);
4929 pagecache_isize_extended(inode, oldsize, inode->i_size);
4932 * Blocks are going to be removed from the inode. Wait
4933 * for dio in flight. Temporarily disable
4934 * dioread_nolock to prevent livelock.
4937 if (!ext4_should_journal_data(inode)) {
4938 ext4_inode_block_unlocked_dio(inode);
4939 inode_dio_wait(inode);
4940 ext4_inode_resume_unlocked_dio(inode);
4942 ext4_wait_for_tail_page_commit(inode);
4944 down_write(&EXT4_I(inode)->i_mmap_sem);
4946 * Truncate pagecache after we've waited for commit
4947 * in data=journal mode to make pages freeable.
4949 truncate_pagecache(inode, inode->i_size);
4951 ext4_truncate(inode);
4952 up_write(&EXT4_I(inode)->i_mmap_sem);
4956 setattr_copy(inode, attr);
4957 mark_inode_dirty(inode);
4961 * If the call to ext4_truncate failed to get a transaction handle at
4962 * all, we need to clean up the in-core orphan list manually.
4964 if (orphan && inode->i_nlink)
4965 ext4_orphan_del(NULL, inode);
4967 if (!rc && (ia_valid & ATTR_MODE))
4968 rc = posix_acl_chmod(inode, inode->i_mode);
4971 ext4_std_error(inode->i_sb, error);
4977 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4980 struct inode *inode;
4981 unsigned long long delalloc_blocks;
4983 inode = d_inode(dentry);
4984 generic_fillattr(inode, stat);
4987 * If there is inline data in the inode, the inode will normally not
4988 * have data blocks allocated (it may have an external xattr block).
4989 * Report at least one sector for such files, so tools like tar, rsync,
4990 * others doen't incorrectly think the file is completely sparse.
4992 if (unlikely(ext4_has_inline_data(inode)))
4993 stat->blocks += (stat->size + 511) >> 9;
4996 * We can't update i_blocks if the block allocation is delayed
4997 * otherwise in the case of system crash before the real block
4998 * allocation is done, we will have i_blocks inconsistent with
4999 * on-disk file blocks.
5000 * We always keep i_blocks updated together with real
5001 * allocation. But to not confuse with user, stat
5002 * will return the blocks that include the delayed allocation
5003 * blocks for this file.
5005 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5006 EXT4_I(inode)->i_reserved_data_blocks);
5007 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5011 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5014 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5015 return ext4_ind_trans_blocks(inode, lblocks);
5016 return ext4_ext_index_trans_blocks(inode, pextents);
5020 * Account for index blocks, block groups bitmaps and block group
5021 * descriptor blocks if modify datablocks and index blocks
5022 * worse case, the indexs blocks spread over different block groups
5024 * If datablocks are discontiguous, they are possible to spread over
5025 * different block groups too. If they are contiguous, with flexbg,
5026 * they could still across block group boundary.
5028 * Also account for superblock, inode, quota and xattr blocks
5030 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5033 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5039 * How many index blocks need to touch to map @lblocks logical blocks
5040 * to @pextents physical extents?
5042 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5047 * Now let's see how many group bitmaps and group descriptors need
5050 groups = idxblocks + pextents;
5052 if (groups > ngroups)
5054 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5055 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5057 /* bitmaps and block group descriptor blocks */
5058 ret += groups + gdpblocks;
5060 /* Blocks for super block, inode, quota and xattr blocks */
5061 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5067 * Calculate the total number of credits to reserve to fit
5068 * the modification of a single pages into a single transaction,
5069 * which may include multiple chunks of block allocations.
5071 * This could be called via ext4_write_begin()
5073 * We need to consider the worse case, when
5074 * one new block per extent.
5076 int ext4_writepage_trans_blocks(struct inode *inode)
5078 int bpp = ext4_journal_blocks_per_page(inode);
5081 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5083 /* Account for data blocks for journalled mode */
5084 if (ext4_should_journal_data(inode))
5090 * Calculate the journal credits for a chunk of data modification.
5092 * This is called from DIO, fallocate or whoever calling
5093 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5095 * journal buffers for data blocks are not included here, as DIO
5096 * and fallocate do no need to journal data buffers.
5098 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5100 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5104 * The caller must have previously called ext4_reserve_inode_write().
5105 * Give this, we know that the caller already has write access to iloc->bh.
5107 int ext4_mark_iloc_dirty(handle_t *handle,
5108 struct inode *inode, struct ext4_iloc *iloc)
5112 if (IS_I_VERSION(inode))
5113 inode_inc_iversion(inode);
5115 /* the do_update_inode consumes one bh->b_count */
5118 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5119 err = ext4_do_update_inode(handle, inode, iloc);
5125 * On success, We end up with an outstanding reference count against
5126 * iloc->bh. This _must_ be cleaned up later.
5130 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5131 struct ext4_iloc *iloc)
5135 err = ext4_get_inode_loc(inode, iloc);
5137 BUFFER_TRACE(iloc->bh, "get_write_access");
5138 err = ext4_journal_get_write_access(handle, iloc->bh);
5144 ext4_std_error(inode->i_sb, err);
5149 * Expand an inode by new_extra_isize bytes.
5150 * Returns 0 on success or negative error number on failure.
5152 static int ext4_expand_extra_isize(struct inode *inode,
5153 unsigned int new_extra_isize,
5154 struct ext4_iloc iloc,
5157 struct ext4_inode *raw_inode;
5158 struct ext4_xattr_ibody_header *header;
5160 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5163 raw_inode = ext4_raw_inode(&iloc);
5165 header = IHDR(inode, raw_inode);
5167 /* No extended attributes present */
5168 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5169 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5170 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5172 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5176 /* try to expand with EAs present */
5177 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5182 * What we do here is to mark the in-core inode as clean with respect to inode
5183 * dirtiness (it may still be data-dirty).
5184 * This means that the in-core inode may be reaped by prune_icache
5185 * without having to perform any I/O. This is a very good thing,
5186 * because *any* task may call prune_icache - even ones which
5187 * have a transaction open against a different journal.
5189 * Is this cheating? Not really. Sure, we haven't written the
5190 * inode out, but prune_icache isn't a user-visible syncing function.
5191 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5192 * we start and wait on commits.
5194 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5196 struct ext4_iloc iloc;
5197 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5198 static unsigned int mnt_count;
5202 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5203 err = ext4_reserve_inode_write(handle, inode, &iloc);
5206 if (ext4_handle_valid(handle) &&
5207 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5208 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5210 * We need extra buffer credits since we may write into EA block
5211 * with this same handle. If journal_extend fails, then it will
5212 * only result in a minor loss of functionality for that inode.
5213 * If this is felt to be critical, then e2fsck should be run to
5214 * force a large enough s_min_extra_isize.
5216 if ((jbd2_journal_extend(handle,
5217 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5218 ret = ext4_expand_extra_isize(inode,
5219 sbi->s_want_extra_isize,
5223 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5224 ext4_warning(inode->i_sb,
5225 "Unable to expand inode %lu. Delete"
5226 " some EAs or run e2fsck.",
5229 le16_to_cpu(sbi->s_es->s_mnt_count);
5234 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5238 * ext4_dirty_inode() is called from __mark_inode_dirty()
5240 * We're really interested in the case where a file is being extended.
5241 * i_size has been changed by generic_commit_write() and we thus need
5242 * to include the updated inode in the current transaction.
5244 * Also, dquot_alloc_block() will always dirty the inode when blocks
5245 * are allocated to the file.
5247 * If the inode is marked synchronous, we don't honour that here - doing
5248 * so would cause a commit on atime updates, which we don't bother doing.
5249 * We handle synchronous inodes at the highest possible level.
5251 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5252 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5253 * to copy into the on-disk inode structure are the timestamp files.
5255 void ext4_dirty_inode(struct inode *inode, int flags)
5259 if (flags == I_DIRTY_TIME)
5261 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5265 ext4_mark_inode_dirty(handle, inode);
5267 ext4_journal_stop(handle);
5274 * Bind an inode's backing buffer_head into this transaction, to prevent
5275 * it from being flushed to disk early. Unlike
5276 * ext4_reserve_inode_write, this leaves behind no bh reference and
5277 * returns no iloc structure, so the caller needs to repeat the iloc
5278 * lookup to mark the inode dirty later.
5280 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5282 struct ext4_iloc iloc;
5286 err = ext4_get_inode_loc(inode, &iloc);
5288 BUFFER_TRACE(iloc.bh, "get_write_access");
5289 err = jbd2_journal_get_write_access(handle, iloc.bh);
5291 err = ext4_handle_dirty_metadata(handle,
5297 ext4_std_error(inode->i_sb, err);
5302 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5309 * We have to be very careful here: changing a data block's
5310 * journaling status dynamically is dangerous. If we write a
5311 * data block to the journal, change the status and then delete
5312 * that block, we risk forgetting to revoke the old log record
5313 * from the journal and so a subsequent replay can corrupt data.
5314 * So, first we make sure that the journal is empty and that
5315 * nobody is changing anything.
5318 journal = EXT4_JOURNAL(inode);
5321 if (is_journal_aborted(journal))
5323 /* We have to allocate physical blocks for delalloc blocks
5324 * before flushing journal. otherwise delalloc blocks can not
5325 * be allocated any more. even more truncate on delalloc blocks
5326 * could trigger BUG by flushing delalloc blocks in journal.
5327 * There is no delalloc block in non-journal data mode.
5329 if (val && test_opt(inode->i_sb, DELALLOC)) {
5330 err = ext4_alloc_da_blocks(inode);
5335 /* Wait for all existing dio workers */
5336 ext4_inode_block_unlocked_dio(inode);
5337 inode_dio_wait(inode);
5339 jbd2_journal_lock_updates(journal);
5342 * OK, there are no updates running now, and all cached data is
5343 * synced to disk. We are now in a completely consistent state
5344 * which doesn't have anything in the journal, and we know that
5345 * no filesystem updates are running, so it is safe to modify
5346 * the inode's in-core data-journaling state flag now.
5350 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5352 err = jbd2_journal_flush(journal);
5354 jbd2_journal_unlock_updates(journal);
5355 ext4_inode_resume_unlocked_dio(inode);
5358 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5360 ext4_set_aops(inode);
5362 jbd2_journal_unlock_updates(journal);
5363 ext4_inode_resume_unlocked_dio(inode);
5365 /* Finally we can mark the inode as dirty. */
5367 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5369 return PTR_ERR(handle);
5371 err = ext4_mark_inode_dirty(handle, inode);
5372 ext4_handle_sync(handle);
5373 ext4_journal_stop(handle);
5374 ext4_std_error(inode->i_sb, err);
5379 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5381 return !buffer_mapped(bh);
5384 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5386 struct page *page = vmf->page;
5390 struct file *file = vma->vm_file;
5391 struct inode *inode = file_inode(file);
5392 struct address_space *mapping = inode->i_mapping;
5394 get_block_t *get_block;
5397 sb_start_pagefault(inode->i_sb);
5398 file_update_time(vma->vm_file);
5400 down_read(&EXT4_I(inode)->i_mmap_sem);
5401 /* Delalloc case is easy... */
5402 if (test_opt(inode->i_sb, DELALLOC) &&
5403 !ext4_should_journal_data(inode) &&
5404 !ext4_nonda_switch(inode->i_sb)) {
5406 ret = block_page_mkwrite(vma, vmf,
5407 ext4_da_get_block_prep);
5408 } while (ret == -ENOSPC &&
5409 ext4_should_retry_alloc(inode->i_sb, &retries));
5414 size = i_size_read(inode);
5415 /* Page got truncated from under us? */
5416 if (page->mapping != mapping || page_offset(page) > size) {
5418 ret = VM_FAULT_NOPAGE;
5422 if (page->index == size >> PAGE_CACHE_SHIFT)
5423 len = size & ~PAGE_CACHE_MASK;
5425 len = PAGE_CACHE_SIZE;
5427 * Return if we have all the buffers mapped. This avoids the need to do
5428 * journal_start/journal_stop which can block and take a long time
5430 if (page_has_buffers(page)) {
5431 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5433 ext4_bh_unmapped)) {
5434 /* Wait so that we don't change page under IO */
5435 wait_for_stable_page(page);
5436 ret = VM_FAULT_LOCKED;
5441 /* OK, we need to fill the hole... */
5442 if (ext4_should_dioread_nolock(inode))
5443 get_block = ext4_get_block_write;
5445 get_block = ext4_get_block;
5447 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5448 ext4_writepage_trans_blocks(inode));
5449 if (IS_ERR(handle)) {
5450 ret = VM_FAULT_SIGBUS;
5453 ret = block_page_mkwrite(vma, vmf, get_block);
5454 if (!ret && ext4_should_journal_data(inode)) {
5455 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5456 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5458 ret = VM_FAULT_SIGBUS;
5459 ext4_journal_stop(handle);
5462 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5464 ext4_journal_stop(handle);
5465 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5468 ret = block_page_mkwrite_return(ret);
5470 up_read(&EXT4_I(inode)->i_mmap_sem);
5471 sb_end_pagefault(inode->i_sb);
5475 int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
5477 struct inode *inode = file_inode(vma->vm_file);
5480 down_read(&EXT4_I(inode)->i_mmap_sem);
5481 err = filemap_fault(vma, vmf);
5482 up_read(&EXT4_I(inode)->i_mmap_sem);