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/quotaops.h>
26 #include <linux/string.h>
27 #include <linux/buffer_head.h>
28 #include <linux/writeback.h>
29 #include <linux/pagevec.h>
30 #include <linux/mpage.h>
31 #include <linux/namei.h>
32 #include <linux/uio.h>
33 #include <linux/bio.h>
34 #include <linux/workqueue.h>
35 #include <linux/kernel.h>
36 #include <linux/printk.h>
37 #include <linux/slab.h>
38 #include <linux/aio.h>
39 #include <linux/bitops.h>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51 struct ext4_inode_info *ei)
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 csum_lo = le16_to_cpu(raw->i_checksum_lo);
59 raw->i_checksum_lo = 0;
60 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
61 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
62 csum_hi = le16_to_cpu(raw->i_checksum_hi);
63 raw->i_checksum_hi = 0;
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
67 EXT4_INODE_SIZE(inode->i_sb));
69 raw->i_checksum_lo = cpu_to_le16(csum_lo);
70 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
71 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
72 raw->i_checksum_hi = cpu_to_le16(csum_hi);
77 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
78 struct ext4_inode_info *ei)
80 __u32 provided, calculated;
82 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
83 cpu_to_le32(EXT4_OS_LINUX) ||
84 !ext4_has_metadata_csum(inode->i_sb))
87 provided = le16_to_cpu(raw->i_checksum_lo);
88 calculated = ext4_inode_csum(inode, raw, ei);
89 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
90 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
91 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
95 return provided == calculated;
98 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
99 struct ext4_inode_info *ei)
103 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
104 cpu_to_le32(EXT4_OS_LINUX) ||
105 !ext4_has_metadata_csum(inode->i_sb))
108 csum = ext4_inode_csum(inode, raw, ei);
109 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
110 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
111 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
112 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
115 static inline int ext4_begin_ordered_truncate(struct inode *inode,
118 trace_ext4_begin_ordered_truncate(inode, new_size);
120 * If jinode is zero, then we never opened the file for
121 * writing, so there's no need to call
122 * jbd2_journal_begin_ordered_truncate() since there's no
123 * outstanding writes we need to flush.
125 if (!EXT4_I(inode)->jinode)
127 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
128 EXT4_I(inode)->jinode,
132 static void ext4_invalidatepage(struct page *page, unsigned int offset,
133 unsigned int length);
134 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
135 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
136 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
140 * Test whether an inode is a fast symlink.
142 int ext4_inode_is_fast_symlink(struct inode *inode)
144 int ea_blocks = EXT4_I(inode)->i_file_acl ?
145 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
147 if (ext4_has_inline_data(inode))
150 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
154 * Restart the transaction associated with *handle. This does a commit,
155 * so before we call here everything must be consistently dirtied against
158 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
164 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
165 * moment, get_block can be called only for blocks inside i_size since
166 * page cache has been already dropped and writes are blocked by
167 * i_mutex. So we can safely drop the i_data_sem here.
169 BUG_ON(EXT4_JOURNAL(inode) == NULL);
170 jbd_debug(2, "restarting handle %p\n", handle);
171 up_write(&EXT4_I(inode)->i_data_sem);
172 ret = ext4_journal_restart(handle, nblocks);
173 down_write(&EXT4_I(inode)->i_data_sem);
174 ext4_discard_preallocations(inode);
180 * Called at the last iput() if i_nlink is zero.
182 void ext4_evict_inode(struct inode *inode)
187 trace_ext4_evict_inode(inode);
189 if (inode->i_nlink) {
191 * When journalling data dirty buffers are tracked only in the
192 * journal. So although mm thinks everything is clean and
193 * ready for reaping the inode might still have some pages to
194 * write in the running transaction or waiting to be
195 * checkpointed. Thus calling jbd2_journal_invalidatepage()
196 * (via truncate_inode_pages()) to discard these buffers can
197 * cause data loss. Also even if we did not discard these
198 * buffers, we would have no way to find them after the inode
199 * is reaped and thus user could see stale data if he tries to
200 * read them before the transaction is checkpointed. So be
201 * careful and force everything to disk here... We use
202 * ei->i_datasync_tid to store the newest transaction
203 * containing inode's data.
205 * Note that directories do not have this problem because they
206 * don't use page cache.
208 if (ext4_should_journal_data(inode) &&
209 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
210 inode->i_ino != EXT4_JOURNAL_INO) {
211 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
212 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
214 jbd2_complete_transaction(journal, commit_tid);
215 filemap_write_and_wait(&inode->i_data);
217 truncate_inode_pages_final(&inode->i_data);
219 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
223 if (is_bad_inode(inode))
225 dquot_initialize(inode);
227 if (ext4_should_order_data(inode))
228 ext4_begin_ordered_truncate(inode, 0);
229 truncate_inode_pages_final(&inode->i_data);
231 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
234 * Protect us against freezing - iput() caller didn't have to have any
235 * protection against it
237 sb_start_intwrite(inode->i_sb);
238 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
239 ext4_blocks_for_truncate(inode)+3);
240 if (IS_ERR(handle)) {
241 ext4_std_error(inode->i_sb, PTR_ERR(handle));
243 * If we're going to skip the normal cleanup, we still need to
244 * make sure that the in-core orphan linked list is properly
247 ext4_orphan_del(NULL, inode);
248 sb_end_intwrite(inode->i_sb);
253 ext4_handle_sync(handle);
255 err = ext4_mark_inode_dirty(handle, inode);
257 ext4_warning(inode->i_sb,
258 "couldn't mark inode dirty (err %d)", err);
262 ext4_truncate(inode);
265 * ext4_ext_truncate() doesn't reserve any slop when it
266 * restarts journal transactions; therefore there may not be
267 * enough credits left in the handle to remove the inode from
268 * the orphan list and set the dtime field.
270 if (!ext4_handle_has_enough_credits(handle, 3)) {
271 err = ext4_journal_extend(handle, 3);
273 err = ext4_journal_restart(handle, 3);
275 ext4_warning(inode->i_sb,
276 "couldn't extend journal (err %d)", err);
278 ext4_journal_stop(handle);
279 ext4_orphan_del(NULL, inode);
280 sb_end_intwrite(inode->i_sb);
286 * Kill off the orphan record which ext4_truncate created.
287 * AKPM: I think this can be inside the above `if'.
288 * Note that ext4_orphan_del() has to be able to cope with the
289 * deletion of a non-existent orphan - this is because we don't
290 * know if ext4_truncate() actually created an orphan record.
291 * (Well, we could do this if we need to, but heck - it works)
293 ext4_orphan_del(handle, inode);
294 EXT4_I(inode)->i_dtime = get_seconds();
297 * One subtle ordering requirement: if anything has gone wrong
298 * (transaction abort, IO errors, whatever), then we can still
299 * do these next steps (the fs will already have been marked as
300 * having errors), but we can't free the inode if the mark_dirty
303 if (ext4_mark_inode_dirty(handle, inode))
304 /* If that failed, just do the required in-core inode clear. */
305 ext4_clear_inode(inode);
307 ext4_free_inode(handle, inode);
308 ext4_journal_stop(handle);
309 sb_end_intwrite(inode->i_sb);
312 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
316 qsize_t *ext4_get_reserved_space(struct inode *inode)
318 return &EXT4_I(inode)->i_reserved_quota;
323 * Called with i_data_sem down, which is important since we can call
324 * ext4_discard_preallocations() from here.
326 void ext4_da_update_reserve_space(struct inode *inode,
327 int used, int quota_claim)
329 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
330 struct ext4_inode_info *ei = EXT4_I(inode);
332 spin_lock(&ei->i_block_reservation_lock);
333 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
334 if (unlikely(used > ei->i_reserved_data_blocks)) {
335 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
336 "with only %d reserved data blocks",
337 __func__, inode->i_ino, used,
338 ei->i_reserved_data_blocks);
340 used = ei->i_reserved_data_blocks;
343 /* Update per-inode reservations */
344 ei->i_reserved_data_blocks -= used;
345 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
347 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
349 /* Update quota subsystem for data blocks */
351 dquot_claim_block(inode, EXT4_C2B(sbi, used));
354 * We did fallocate with an offset that is already delayed
355 * allocated. So on delayed allocated writeback we should
356 * not re-claim the quota for fallocated blocks.
358 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
362 * If we have done all the pending block allocations and if
363 * there aren't any writers on the inode, we can discard the
364 * inode's preallocations.
366 if ((ei->i_reserved_data_blocks == 0) &&
367 (atomic_read(&inode->i_writecount) == 0))
368 ext4_discard_preallocations(inode);
371 static int __check_block_validity(struct inode *inode, const char *func,
373 struct ext4_map_blocks *map)
375 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
377 ext4_error_inode(inode, func, line, map->m_pblk,
378 "lblock %lu mapped to illegal pblock "
379 "(length %d)", (unsigned long) map->m_lblk,
386 #define check_block_validity(inode, map) \
387 __check_block_validity((inode), __func__, __LINE__, (map))
389 #ifdef ES_AGGRESSIVE_TEST
390 static void ext4_map_blocks_es_recheck(handle_t *handle,
392 struct ext4_map_blocks *es_map,
393 struct ext4_map_blocks *map,
400 * There is a race window that the result is not the same.
401 * e.g. xfstests #223 when dioread_nolock enables. The reason
402 * is that we lookup a block mapping in extent status tree with
403 * out taking i_data_sem. So at the time the unwritten extent
404 * could be converted.
406 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
407 down_read(&EXT4_I(inode)->i_data_sem);
408 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
409 retval = ext4_ext_map_blocks(handle, inode, map, flags &
410 EXT4_GET_BLOCKS_KEEP_SIZE);
412 retval = ext4_ind_map_blocks(handle, inode, map, flags &
413 EXT4_GET_BLOCKS_KEEP_SIZE);
415 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
416 up_read((&EXT4_I(inode)->i_data_sem));
419 * We don't check m_len because extent will be collpased in status
420 * tree. So the m_len might not equal.
422 if (es_map->m_lblk != map->m_lblk ||
423 es_map->m_flags != map->m_flags ||
424 es_map->m_pblk != map->m_pblk) {
425 printk("ES cache assertion failed for inode: %lu "
426 "es_cached ex [%d/%d/%llu/%x] != "
427 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
428 inode->i_ino, es_map->m_lblk, es_map->m_len,
429 es_map->m_pblk, es_map->m_flags, map->m_lblk,
430 map->m_len, map->m_pblk, map->m_flags,
434 #endif /* ES_AGGRESSIVE_TEST */
437 * The ext4_map_blocks() function tries to look up the requested blocks,
438 * and returns if the blocks are already mapped.
440 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
441 * and store the allocated blocks in the result buffer head and mark it
444 * If file type is extents based, it will call ext4_ext_map_blocks(),
445 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
448 * On success, it returns the number of blocks being mapped or allocated.
449 * if create==0 and the blocks are pre-allocated and unwritten block,
450 * the result buffer head is unmapped. If the create ==1, it will make sure
451 * the buffer head is mapped.
453 * It returns 0 if plain look up failed (blocks have not been allocated), in
454 * that case, buffer head is unmapped
456 * It returns the error in case of allocation failure.
458 int ext4_map_blocks(handle_t *handle, struct inode *inode,
459 struct ext4_map_blocks *map, int flags)
461 struct extent_status es;
464 #ifdef ES_AGGRESSIVE_TEST
465 struct ext4_map_blocks orig_map;
467 memcpy(&orig_map, map, sizeof(*map));
471 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
472 "logical block %lu\n", inode->i_ino, flags, map->m_len,
473 (unsigned long) map->m_lblk);
476 * ext4_map_blocks returns an int, and m_len is an unsigned int
478 if (unlikely(map->m_len > INT_MAX))
479 map->m_len = INT_MAX;
481 /* We can handle the block number less than EXT_MAX_BLOCKS */
482 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
485 /* Lookup extent status tree firstly */
486 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
487 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
488 map->m_pblk = ext4_es_pblock(&es) +
489 map->m_lblk - es.es_lblk;
490 map->m_flags |= ext4_es_is_written(&es) ?
491 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
492 retval = es.es_len - (map->m_lblk - es.es_lblk);
493 if (retval > map->m_len)
496 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
501 #ifdef ES_AGGRESSIVE_TEST
502 ext4_map_blocks_es_recheck(handle, inode, map,
509 * Try to see if we can get the block without requesting a new
512 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
513 down_read(&EXT4_I(inode)->i_data_sem);
514 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
515 retval = ext4_ext_map_blocks(handle, inode, map, flags &
516 EXT4_GET_BLOCKS_KEEP_SIZE);
518 retval = ext4_ind_map_blocks(handle, inode, map, flags &
519 EXT4_GET_BLOCKS_KEEP_SIZE);
524 if (unlikely(retval != map->m_len)) {
525 ext4_warning(inode->i_sb,
526 "ES len assertion failed for inode "
527 "%lu: retval %d != map->m_len %d",
528 inode->i_ino, retval, map->m_len);
532 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
533 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
534 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
535 ext4_find_delalloc_range(inode, map->m_lblk,
536 map->m_lblk + map->m_len - 1))
537 status |= EXTENT_STATUS_DELAYED;
538 ret = ext4_es_insert_extent(inode, map->m_lblk,
539 map->m_len, map->m_pblk, status);
543 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
544 up_read((&EXT4_I(inode)->i_data_sem));
547 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
548 ret = check_block_validity(inode, map);
553 /* If it is only a block(s) look up */
554 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
558 * Returns if the blocks have already allocated
560 * Note that if blocks have been preallocated
561 * ext4_ext_get_block() returns the create = 0
562 * with buffer head unmapped.
564 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
566 * If we need to convert extent to unwritten
567 * we continue and do the actual work in
568 * ext4_ext_map_blocks()
570 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
574 * Here we clear m_flags because after allocating an new extent,
575 * it will be set again.
577 map->m_flags &= ~EXT4_MAP_FLAGS;
580 * New blocks allocate and/or writing to unwritten extent
581 * will possibly result in updating i_data, so we take
582 * the write lock of i_data_sem, and call get_block()
583 * with create == 1 flag.
585 down_write(&EXT4_I(inode)->i_data_sem);
588 * We need to check for EXT4 here because migrate
589 * could have changed the inode type in between
591 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
592 retval = ext4_ext_map_blocks(handle, inode, map, flags);
594 retval = ext4_ind_map_blocks(handle, inode, map, flags);
596 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
598 * We allocated new blocks which will result in
599 * i_data's format changing. Force the migrate
600 * to fail by clearing migrate flags
602 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
606 * Update reserved blocks/metadata blocks after successful
607 * block allocation which had been deferred till now. We don't
608 * support fallocate for non extent files. So we can update
609 * reserve space here.
612 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
613 ext4_da_update_reserve_space(inode, retval, 1);
619 if (unlikely(retval != map->m_len)) {
620 ext4_warning(inode->i_sb,
621 "ES len assertion failed for inode "
622 "%lu: retval %d != map->m_len %d",
623 inode->i_ino, retval, map->m_len);
628 * If the extent has been zeroed out, we don't need to update
629 * extent status tree.
631 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
632 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
633 if (ext4_es_is_written(&es))
636 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
637 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
638 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
639 ext4_find_delalloc_range(inode, map->m_lblk,
640 map->m_lblk + map->m_len - 1))
641 status |= EXTENT_STATUS_DELAYED;
642 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
643 map->m_pblk, status);
649 up_write((&EXT4_I(inode)->i_data_sem));
650 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
651 ret = check_block_validity(inode, map);
658 static void ext4_end_io_unwritten(struct buffer_head *bh, int uptodate)
660 struct inode *inode = bh->b_assoc_map->host;
661 /* XXX: breaks on 32-bit > 16GB. Is that even supported? */
662 loff_t offset = (loff_t)(uintptr_t)bh->b_private << inode->i_blkbits;
666 WARN_ON(!buffer_unwritten(bh));
667 err = ext4_convert_unwritten_extents(NULL, inode, offset, bh->b_size);
670 /* Maximum number of blocks we map for direct IO at once. */
671 #define DIO_MAX_BLOCKS 4096
673 static int _ext4_get_block(struct inode *inode, sector_t iblock,
674 struct buffer_head *bh, int flags)
676 handle_t *handle = ext4_journal_current_handle();
677 struct ext4_map_blocks map;
678 int ret = 0, started = 0;
681 if (ext4_has_inline_data(inode))
685 map.m_len = bh->b_size >> inode->i_blkbits;
687 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
688 /* Direct IO write... */
689 if (map.m_len > DIO_MAX_BLOCKS)
690 map.m_len = DIO_MAX_BLOCKS;
691 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
692 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
694 if (IS_ERR(handle)) {
695 ret = PTR_ERR(handle);
701 ret = ext4_map_blocks(handle, inode, &map, flags);
703 ext4_io_end_t *io_end = ext4_inode_aio(inode);
705 map_bh(bh, inode->i_sb, map.m_pblk);
706 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
707 if (IS_DAX(inode) && buffer_unwritten(bh) && !io_end) {
708 bh->b_assoc_map = inode->i_mapping;
709 bh->b_private = (void *)(unsigned long)iblock;
710 bh->b_end_io = ext4_end_io_unwritten;
712 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
713 set_buffer_defer_completion(bh);
714 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
718 ext4_journal_stop(handle);
722 int ext4_get_block(struct inode *inode, sector_t iblock,
723 struct buffer_head *bh, int create)
725 return _ext4_get_block(inode, iblock, bh,
726 create ? EXT4_GET_BLOCKS_CREATE : 0);
730 * `handle' can be NULL if create is zero
732 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
733 ext4_lblk_t block, int create)
735 struct ext4_map_blocks map;
736 struct buffer_head *bh;
739 J_ASSERT(handle != NULL || create == 0);
743 err = ext4_map_blocks(handle, inode, &map,
744 create ? EXT4_GET_BLOCKS_CREATE : 0);
747 return create ? ERR_PTR(-ENOSPC) : NULL;
751 bh = sb_getblk(inode->i_sb, map.m_pblk);
753 return ERR_PTR(-ENOMEM);
754 if (map.m_flags & EXT4_MAP_NEW) {
755 J_ASSERT(create != 0);
756 J_ASSERT(handle != NULL);
759 * Now that we do not always journal data, we should
760 * keep in mind whether this should always journal the
761 * new buffer as metadata. For now, regular file
762 * writes use ext4_get_block instead, so it's not a
766 BUFFER_TRACE(bh, "call get_create_access");
767 err = ext4_journal_get_create_access(handle, bh);
772 if (!buffer_uptodate(bh)) {
773 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
774 set_buffer_uptodate(bh);
777 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
778 err = ext4_handle_dirty_metadata(handle, inode, bh);
782 BUFFER_TRACE(bh, "not a new buffer");
789 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
790 ext4_lblk_t block, int create)
792 struct buffer_head *bh;
794 bh = ext4_getblk(handle, inode, block, create);
797 if (!bh || buffer_uptodate(bh))
799 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
801 if (buffer_uptodate(bh))
804 return ERR_PTR(-EIO);
807 int ext4_walk_page_buffers(handle_t *handle,
808 struct buffer_head *head,
812 int (*fn)(handle_t *handle,
813 struct buffer_head *bh))
815 struct buffer_head *bh;
816 unsigned block_start, block_end;
817 unsigned blocksize = head->b_size;
819 struct buffer_head *next;
821 for (bh = head, block_start = 0;
822 ret == 0 && (bh != head || !block_start);
823 block_start = block_end, bh = next) {
824 next = bh->b_this_page;
825 block_end = block_start + blocksize;
826 if (block_end <= from || block_start >= to) {
827 if (partial && !buffer_uptodate(bh))
831 err = (*fn)(handle, bh);
839 * To preserve ordering, it is essential that the hole instantiation and
840 * the data write be encapsulated in a single transaction. We cannot
841 * close off a transaction and start a new one between the ext4_get_block()
842 * and the commit_write(). So doing the jbd2_journal_start at the start of
843 * prepare_write() is the right place.
845 * Also, this function can nest inside ext4_writepage(). In that case, we
846 * *know* that ext4_writepage() has generated enough buffer credits to do the
847 * whole page. So we won't block on the journal in that case, which is good,
848 * because the caller may be PF_MEMALLOC.
850 * By accident, ext4 can be reentered when a transaction is open via
851 * quota file writes. If we were to commit the transaction while thus
852 * reentered, there can be a deadlock - we would be holding a quota
853 * lock, and the commit would never complete if another thread had a
854 * transaction open and was blocking on the quota lock - a ranking
857 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
858 * will _not_ run commit under these circumstances because handle->h_ref
859 * is elevated. We'll still have enough credits for the tiny quotafile
862 int do_journal_get_write_access(handle_t *handle,
863 struct buffer_head *bh)
865 int dirty = buffer_dirty(bh);
868 if (!buffer_mapped(bh) || buffer_freed(bh))
871 * __block_write_begin() could have dirtied some buffers. Clean
872 * the dirty bit as jbd2_journal_get_write_access() could complain
873 * otherwise about fs integrity issues. Setting of the dirty bit
874 * by __block_write_begin() isn't a real problem here as we clear
875 * the bit before releasing a page lock and thus writeback cannot
876 * ever write the buffer.
879 clear_buffer_dirty(bh);
880 BUFFER_TRACE(bh, "get write access");
881 ret = ext4_journal_get_write_access(handle, bh);
883 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
887 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
888 struct buffer_head *bh_result, int create);
890 #ifdef CONFIG_EXT4_FS_ENCRYPTION
891 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
892 get_block_t *get_block)
894 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
895 unsigned to = from + len;
896 struct inode *inode = page->mapping->host;
897 unsigned block_start, block_end;
900 unsigned blocksize = inode->i_sb->s_blocksize;
902 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
903 bool decrypt = false;
905 BUG_ON(!PageLocked(page));
906 BUG_ON(from > PAGE_CACHE_SIZE);
907 BUG_ON(to > PAGE_CACHE_SIZE);
910 if (!page_has_buffers(page))
911 create_empty_buffers(page, blocksize, 0);
912 head = page_buffers(page);
913 bbits = ilog2(blocksize);
914 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
916 for (bh = head, block_start = 0; bh != head || !block_start;
917 block++, block_start = block_end, bh = bh->b_this_page) {
918 block_end = block_start + blocksize;
919 if (block_end <= from || block_start >= to) {
920 if (PageUptodate(page)) {
921 if (!buffer_uptodate(bh))
922 set_buffer_uptodate(bh);
927 clear_buffer_new(bh);
928 if (!buffer_mapped(bh)) {
929 WARN_ON(bh->b_size != blocksize);
930 err = get_block(inode, block, bh, 1);
933 if (buffer_new(bh)) {
934 unmap_underlying_metadata(bh->b_bdev,
936 if (PageUptodate(page)) {
937 clear_buffer_new(bh);
938 set_buffer_uptodate(bh);
939 mark_buffer_dirty(bh);
942 if (block_end > to || block_start < from)
943 zero_user_segments(page, to, block_end,
948 if (PageUptodate(page)) {
949 if (!buffer_uptodate(bh))
950 set_buffer_uptodate(bh);
953 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
954 !buffer_unwritten(bh) &&
955 (block_start < from || block_end > to)) {
956 ll_rw_block(READ, 1, &bh);
958 decrypt = ext4_encrypted_inode(inode) &&
959 S_ISREG(inode->i_mode);
963 * If we issued read requests, let them complete.
965 while (wait_bh > wait) {
966 wait_on_buffer(*--wait_bh);
967 if (!buffer_uptodate(*wait_bh))
971 page_zero_new_buffers(page, from, to);
973 err = ext4_decrypt_one(inode, page);
978 static int ext4_write_begin(struct file *file, struct address_space *mapping,
979 loff_t pos, unsigned len, unsigned flags,
980 struct page **pagep, void **fsdata)
982 struct inode *inode = mapping->host;
983 int ret, needed_blocks;
990 trace_ext4_write_begin(inode, pos, len, flags);
992 * Reserve one block more for addition to orphan list in case
993 * we allocate blocks but write fails for some reason
995 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
996 index = pos >> PAGE_CACHE_SHIFT;
997 from = pos & (PAGE_CACHE_SIZE - 1);
1000 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1001 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1010 * grab_cache_page_write_begin() can take a long time if the
1011 * system is thrashing due to memory pressure, or if the page
1012 * is being written back. So grab it first before we start
1013 * the transaction handle. This also allows us to allocate
1014 * the page (if needed) without using GFP_NOFS.
1017 page = grab_cache_page_write_begin(mapping, index, flags);
1023 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1024 if (IS_ERR(handle)) {
1025 page_cache_release(page);
1026 return PTR_ERR(handle);
1030 if (page->mapping != mapping) {
1031 /* The page got truncated from under us */
1033 page_cache_release(page);
1034 ext4_journal_stop(handle);
1037 /* In case writeback began while the page was unlocked */
1038 wait_for_stable_page(page);
1040 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1041 if (ext4_should_dioread_nolock(inode))
1042 ret = ext4_block_write_begin(page, pos, len,
1043 ext4_get_block_write);
1045 ret = ext4_block_write_begin(page, pos, len,
1048 if (ext4_should_dioread_nolock(inode))
1049 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1051 ret = __block_write_begin(page, pos, len, ext4_get_block);
1053 if (!ret && ext4_should_journal_data(inode)) {
1054 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1056 do_journal_get_write_access);
1062 * __block_write_begin may have instantiated a few blocks
1063 * outside i_size. Trim these off again. Don't need
1064 * i_size_read because we hold i_mutex.
1066 * Add inode to orphan list in case we crash before
1069 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1070 ext4_orphan_add(handle, inode);
1072 ext4_journal_stop(handle);
1073 if (pos + len > inode->i_size) {
1074 ext4_truncate_failed_write(inode);
1076 * If truncate failed early the inode might
1077 * still be on the orphan list; we need to
1078 * make sure the inode is removed from the
1079 * orphan list in that case.
1082 ext4_orphan_del(NULL, inode);
1085 if (ret == -ENOSPC &&
1086 ext4_should_retry_alloc(inode->i_sb, &retries))
1088 page_cache_release(page);
1095 /* For write_end() in data=journal mode */
1096 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1099 if (!buffer_mapped(bh) || buffer_freed(bh))
1101 set_buffer_uptodate(bh);
1102 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1103 clear_buffer_meta(bh);
1104 clear_buffer_prio(bh);
1109 * We need to pick up the new inode size which generic_commit_write gave us
1110 * `file' can be NULL - eg, when called from page_symlink().
1112 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1113 * buffers are managed internally.
1115 static int ext4_write_end(struct file *file,
1116 struct address_space *mapping,
1117 loff_t pos, unsigned len, unsigned copied,
1118 struct page *page, void *fsdata)
1120 handle_t *handle = ext4_journal_current_handle();
1121 struct inode *inode = mapping->host;
1122 loff_t old_size = inode->i_size;
1124 int i_size_changed = 0;
1126 trace_ext4_write_end(inode, pos, len, copied);
1127 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1128 ret = ext4_jbd2_file_inode(handle, inode);
1131 page_cache_release(page);
1136 if (ext4_has_inline_data(inode)) {
1137 ret = ext4_write_inline_data_end(inode, pos, len,
1143 copied = block_write_end(file, mapping, pos,
1144 len, copied, page, fsdata);
1146 * it's important to update i_size while still holding page lock:
1147 * page writeout could otherwise come in and zero beyond i_size.
1149 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1151 page_cache_release(page);
1154 pagecache_isize_extended(inode, old_size, pos);
1156 * Don't mark the inode dirty under page lock. First, it unnecessarily
1157 * makes the holding time of page lock longer. Second, it forces lock
1158 * ordering of page lock and transaction start for journaling
1162 ext4_mark_inode_dirty(handle, inode);
1164 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1165 /* if we have allocated more blocks and copied
1166 * less. We will have blocks allocated outside
1167 * inode->i_size. So truncate them
1169 ext4_orphan_add(handle, inode);
1171 ret2 = ext4_journal_stop(handle);
1175 if (pos + len > inode->i_size) {
1176 ext4_truncate_failed_write(inode);
1178 * If truncate failed early the inode might still be
1179 * on the orphan list; we need to make sure the inode
1180 * is removed from the orphan list in that case.
1183 ext4_orphan_del(NULL, inode);
1186 return ret ? ret : copied;
1189 static int ext4_journalled_write_end(struct file *file,
1190 struct address_space *mapping,
1191 loff_t pos, unsigned len, unsigned copied,
1192 struct page *page, void *fsdata)
1194 handle_t *handle = ext4_journal_current_handle();
1195 struct inode *inode = mapping->host;
1196 loff_t old_size = inode->i_size;
1200 int size_changed = 0;
1202 trace_ext4_journalled_write_end(inode, pos, len, copied);
1203 from = pos & (PAGE_CACHE_SIZE - 1);
1206 BUG_ON(!ext4_handle_valid(handle));
1208 if (ext4_has_inline_data(inode))
1209 copied = ext4_write_inline_data_end(inode, pos, len,
1213 if (!PageUptodate(page))
1215 page_zero_new_buffers(page, from+copied, to);
1218 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1219 to, &partial, write_end_fn);
1221 SetPageUptodate(page);
1223 size_changed = ext4_update_inode_size(inode, pos + copied);
1224 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1225 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1227 page_cache_release(page);
1230 pagecache_isize_extended(inode, old_size, pos);
1233 ret2 = ext4_mark_inode_dirty(handle, inode);
1238 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1239 /* if we have allocated more blocks and copied
1240 * less. We will have blocks allocated outside
1241 * inode->i_size. So truncate them
1243 ext4_orphan_add(handle, inode);
1245 ret2 = ext4_journal_stop(handle);
1248 if (pos + len > inode->i_size) {
1249 ext4_truncate_failed_write(inode);
1251 * If truncate failed early the inode might still be
1252 * on the orphan list; we need to make sure the inode
1253 * is removed from the orphan list in that case.
1256 ext4_orphan_del(NULL, inode);
1259 return ret ? ret : copied;
1263 * Reserve a single cluster located at lblock
1265 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1267 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1268 struct ext4_inode_info *ei = EXT4_I(inode);
1269 unsigned int md_needed;
1273 * We will charge metadata quota at writeout time; this saves
1274 * us from metadata over-estimation, though we may go over by
1275 * a small amount in the end. Here we just reserve for data.
1277 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1282 * recalculate the amount of metadata blocks to reserve
1283 * in order to allocate nrblocks
1284 * worse case is one extent per block
1286 spin_lock(&ei->i_block_reservation_lock);
1288 * ext4_calc_metadata_amount() has side effects, which we have
1289 * to be prepared undo if we fail to claim space.
1292 trace_ext4_da_reserve_space(inode, 0);
1294 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1295 spin_unlock(&ei->i_block_reservation_lock);
1296 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1299 ei->i_reserved_data_blocks++;
1300 spin_unlock(&ei->i_block_reservation_lock);
1302 return 0; /* success */
1305 static void ext4_da_release_space(struct inode *inode, int to_free)
1307 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1308 struct ext4_inode_info *ei = EXT4_I(inode);
1311 return; /* Nothing to release, exit */
1313 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1315 trace_ext4_da_release_space(inode, to_free);
1316 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1318 * if there aren't enough reserved blocks, then the
1319 * counter is messed up somewhere. Since this
1320 * function is called from invalidate page, it's
1321 * harmless to return without any action.
1323 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1324 "ino %lu, to_free %d with only %d reserved "
1325 "data blocks", inode->i_ino, to_free,
1326 ei->i_reserved_data_blocks);
1328 to_free = ei->i_reserved_data_blocks;
1330 ei->i_reserved_data_blocks -= to_free;
1332 /* update fs dirty data blocks counter */
1333 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1335 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1337 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1340 static void ext4_da_page_release_reservation(struct page *page,
1341 unsigned int offset,
1342 unsigned int length)
1345 struct buffer_head *head, *bh;
1346 unsigned int curr_off = 0;
1347 struct inode *inode = page->mapping->host;
1348 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1349 unsigned int stop = offset + length;
1353 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1355 head = page_buffers(page);
1358 unsigned int next_off = curr_off + bh->b_size;
1360 if (next_off > stop)
1363 if ((offset <= curr_off) && (buffer_delay(bh))) {
1365 clear_buffer_delay(bh);
1367 curr_off = next_off;
1368 } while ((bh = bh->b_this_page) != head);
1371 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1372 ext4_es_remove_extent(inode, lblk, to_release);
1375 /* If we have released all the blocks belonging to a cluster, then we
1376 * need to release the reserved space for that cluster. */
1377 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1378 while (num_clusters > 0) {
1379 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1380 ((num_clusters - 1) << sbi->s_cluster_bits);
1381 if (sbi->s_cluster_ratio == 1 ||
1382 !ext4_find_delalloc_cluster(inode, lblk))
1383 ext4_da_release_space(inode, 1);
1390 * Delayed allocation stuff
1393 struct mpage_da_data {
1394 struct inode *inode;
1395 struct writeback_control *wbc;
1397 pgoff_t first_page; /* The first page to write */
1398 pgoff_t next_page; /* Current page to examine */
1399 pgoff_t last_page; /* Last page to examine */
1401 * Extent to map - this can be after first_page because that can be
1402 * fully mapped. We somewhat abuse m_flags to store whether the extent
1403 * is delalloc or unwritten.
1405 struct ext4_map_blocks map;
1406 struct ext4_io_submit io_submit; /* IO submission data */
1409 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1414 struct pagevec pvec;
1415 struct inode *inode = mpd->inode;
1416 struct address_space *mapping = inode->i_mapping;
1418 /* This is necessary when next_page == 0. */
1419 if (mpd->first_page >= mpd->next_page)
1422 index = mpd->first_page;
1423 end = mpd->next_page - 1;
1425 ext4_lblk_t start, last;
1426 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1427 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1428 ext4_es_remove_extent(inode, start, last - start + 1);
1431 pagevec_init(&pvec, 0);
1432 while (index <= end) {
1433 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1436 for (i = 0; i < nr_pages; i++) {
1437 struct page *page = pvec.pages[i];
1438 if (page->index > end)
1440 BUG_ON(!PageLocked(page));
1441 BUG_ON(PageWriteback(page));
1443 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1444 ClearPageUptodate(page);
1448 index = pvec.pages[nr_pages - 1]->index + 1;
1449 pagevec_release(&pvec);
1453 static void ext4_print_free_blocks(struct inode *inode)
1455 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1456 struct super_block *sb = inode->i_sb;
1457 struct ext4_inode_info *ei = EXT4_I(inode);
1459 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1460 EXT4_C2B(EXT4_SB(inode->i_sb),
1461 ext4_count_free_clusters(sb)));
1462 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1463 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1464 (long long) EXT4_C2B(EXT4_SB(sb),
1465 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1466 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1467 (long long) EXT4_C2B(EXT4_SB(sb),
1468 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1469 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1470 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1471 ei->i_reserved_data_blocks);
1475 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1477 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1481 * This function is grabs code from the very beginning of
1482 * ext4_map_blocks, but assumes that the caller is from delayed write
1483 * time. This function looks up the requested blocks and sets the
1484 * buffer delay bit under the protection of i_data_sem.
1486 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1487 struct ext4_map_blocks *map,
1488 struct buffer_head *bh)
1490 struct extent_status es;
1492 sector_t invalid_block = ~((sector_t) 0xffff);
1493 #ifdef ES_AGGRESSIVE_TEST
1494 struct ext4_map_blocks orig_map;
1496 memcpy(&orig_map, map, sizeof(*map));
1499 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1503 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1504 "logical block %lu\n", inode->i_ino, map->m_len,
1505 (unsigned long) map->m_lblk);
1507 /* Lookup extent status tree firstly */
1508 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1509 if (ext4_es_is_hole(&es)) {
1511 down_read(&EXT4_I(inode)->i_data_sem);
1516 * Delayed extent could be allocated by fallocate.
1517 * So we need to check it.
1519 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1520 map_bh(bh, inode->i_sb, invalid_block);
1522 set_buffer_delay(bh);
1526 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1527 retval = es.es_len - (iblock - es.es_lblk);
1528 if (retval > map->m_len)
1529 retval = map->m_len;
1530 map->m_len = retval;
1531 if (ext4_es_is_written(&es))
1532 map->m_flags |= EXT4_MAP_MAPPED;
1533 else if (ext4_es_is_unwritten(&es))
1534 map->m_flags |= EXT4_MAP_UNWRITTEN;
1538 #ifdef ES_AGGRESSIVE_TEST
1539 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1545 * Try to see if we can get the block without requesting a new
1546 * file system block.
1548 down_read(&EXT4_I(inode)->i_data_sem);
1549 if (ext4_has_inline_data(inode))
1551 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1552 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1554 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1560 * XXX: __block_prepare_write() unmaps passed block,
1564 * If the block was allocated from previously allocated cluster,
1565 * then we don't need to reserve it again. However we still need
1566 * to reserve metadata for every block we're going to write.
1568 if (EXT4_SB(inode->i_sb)->s_cluster_ratio <= 1 ||
1569 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1570 ret = ext4_da_reserve_space(inode, iblock);
1572 /* not enough space to reserve */
1578 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1579 ~0, EXTENT_STATUS_DELAYED);
1585 map_bh(bh, inode->i_sb, invalid_block);
1587 set_buffer_delay(bh);
1588 } else if (retval > 0) {
1590 unsigned int status;
1592 if (unlikely(retval != map->m_len)) {
1593 ext4_warning(inode->i_sb,
1594 "ES len assertion failed for inode "
1595 "%lu: retval %d != map->m_len %d",
1596 inode->i_ino, retval, map->m_len);
1600 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1601 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1602 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1603 map->m_pblk, status);
1609 up_read((&EXT4_I(inode)->i_data_sem));
1615 * This is a special get_block_t callback which is used by
1616 * ext4_da_write_begin(). It will either return mapped block or
1617 * reserve space for a single block.
1619 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1620 * We also have b_blocknr = -1 and b_bdev initialized properly
1622 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1623 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1624 * initialized properly.
1626 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1627 struct buffer_head *bh, int create)
1629 struct ext4_map_blocks map;
1632 BUG_ON(create == 0);
1633 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1635 map.m_lblk = iblock;
1639 * first, we need to know whether the block is allocated already
1640 * preallocated blocks are unmapped but should treated
1641 * the same as allocated blocks.
1643 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1647 map_bh(bh, inode->i_sb, map.m_pblk);
1648 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1650 if (buffer_unwritten(bh)) {
1651 /* A delayed write to unwritten bh should be marked
1652 * new and mapped. Mapped ensures that we don't do
1653 * get_block multiple times when we write to the same
1654 * offset and new ensures that we do proper zero out
1655 * for partial write.
1658 set_buffer_mapped(bh);
1663 static int bget_one(handle_t *handle, struct buffer_head *bh)
1669 static int bput_one(handle_t *handle, struct buffer_head *bh)
1675 static int __ext4_journalled_writepage(struct page *page,
1678 struct address_space *mapping = page->mapping;
1679 struct inode *inode = mapping->host;
1680 struct buffer_head *page_bufs = NULL;
1681 handle_t *handle = NULL;
1682 int ret = 0, err = 0;
1683 int inline_data = ext4_has_inline_data(inode);
1684 struct buffer_head *inode_bh = NULL;
1686 ClearPageChecked(page);
1689 BUG_ON(page->index != 0);
1690 BUG_ON(len > ext4_get_max_inline_size(inode));
1691 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1692 if (inode_bh == NULL)
1695 page_bufs = page_buffers(page);
1700 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1703 /* As soon as we unlock the page, it can go away, but we have
1704 * references to buffers so we are safe */
1707 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1708 ext4_writepage_trans_blocks(inode));
1709 if (IS_ERR(handle)) {
1710 ret = PTR_ERR(handle);
1714 BUG_ON(!ext4_handle_valid(handle));
1717 BUFFER_TRACE(inode_bh, "get write access");
1718 ret = ext4_journal_get_write_access(handle, inode_bh);
1720 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1723 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1724 do_journal_get_write_access);
1726 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1731 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1732 err = ext4_journal_stop(handle);
1736 if (!ext4_has_inline_data(inode))
1737 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1739 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1746 * Note that we don't need to start a transaction unless we're journaling data
1747 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1748 * need to file the inode to the transaction's list in ordered mode because if
1749 * we are writing back data added by write(), the inode is already there and if
1750 * we are writing back data modified via mmap(), no one guarantees in which
1751 * transaction the data will hit the disk. In case we are journaling data, we
1752 * cannot start transaction directly because transaction start ranks above page
1753 * lock so we have to do some magic.
1755 * This function can get called via...
1756 * - ext4_writepages after taking page lock (have journal handle)
1757 * - journal_submit_inode_data_buffers (no journal handle)
1758 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1759 * - grab_page_cache when doing write_begin (have journal handle)
1761 * We don't do any block allocation in this function. If we have page with
1762 * multiple blocks we need to write those buffer_heads that are mapped. This
1763 * is important for mmaped based write. So if we do with blocksize 1K
1764 * truncate(f, 1024);
1765 * a = mmap(f, 0, 4096);
1767 * truncate(f, 4096);
1768 * we have in the page first buffer_head mapped via page_mkwrite call back
1769 * but other buffer_heads would be unmapped but dirty (dirty done via the
1770 * do_wp_page). So writepage should write the first block. If we modify
1771 * the mmap area beyond 1024 we will again get a page_fault and the
1772 * page_mkwrite callback will do the block allocation and mark the
1773 * buffer_heads mapped.
1775 * We redirty the page if we have any buffer_heads that is either delay or
1776 * unwritten in the page.
1778 * We can get recursively called as show below.
1780 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1783 * But since we don't do any block allocation we should not deadlock.
1784 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1786 static int ext4_writepage(struct page *page,
1787 struct writeback_control *wbc)
1792 struct buffer_head *page_bufs = NULL;
1793 struct inode *inode = page->mapping->host;
1794 struct ext4_io_submit io_submit;
1795 bool keep_towrite = false;
1797 trace_ext4_writepage(page);
1798 size = i_size_read(inode);
1799 if (page->index == size >> PAGE_CACHE_SHIFT)
1800 len = size & ~PAGE_CACHE_MASK;
1802 len = PAGE_CACHE_SIZE;
1804 page_bufs = page_buffers(page);
1806 * We cannot do block allocation or other extent handling in this
1807 * function. If there are buffers needing that, we have to redirty
1808 * the page. But we may reach here when we do a journal commit via
1809 * journal_submit_inode_data_buffers() and in that case we must write
1810 * allocated buffers to achieve data=ordered mode guarantees.
1812 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1813 ext4_bh_delay_or_unwritten)) {
1814 redirty_page_for_writepage(wbc, page);
1815 if (current->flags & PF_MEMALLOC) {
1817 * For memory cleaning there's no point in writing only
1818 * some buffers. So just bail out. Warn if we came here
1819 * from direct reclaim.
1821 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1826 keep_towrite = true;
1829 if (PageChecked(page) && ext4_should_journal_data(inode))
1831 * It's mmapped pagecache. Add buffers and journal it. There
1832 * doesn't seem much point in redirtying the page here.
1834 return __ext4_journalled_writepage(page, len);
1836 ext4_io_submit_init(&io_submit, wbc);
1837 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1838 if (!io_submit.io_end) {
1839 redirty_page_for_writepage(wbc, page);
1843 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1844 ext4_io_submit(&io_submit);
1845 /* Drop io_end reference we got from init */
1846 ext4_put_io_end_defer(io_submit.io_end);
1850 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1853 loff_t size = i_size_read(mpd->inode);
1856 BUG_ON(page->index != mpd->first_page);
1857 if (page->index == size >> PAGE_CACHE_SHIFT)
1858 len = size & ~PAGE_CACHE_MASK;
1860 len = PAGE_CACHE_SIZE;
1861 clear_page_dirty_for_io(page);
1862 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1864 mpd->wbc->nr_to_write--;
1870 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1873 * mballoc gives us at most this number of blocks...
1874 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1875 * The rest of mballoc seems to handle chunks up to full group size.
1877 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1880 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1882 * @mpd - extent of blocks
1883 * @lblk - logical number of the block in the file
1884 * @bh - buffer head we want to add to the extent
1886 * The function is used to collect contig. blocks in the same state. If the
1887 * buffer doesn't require mapping for writeback and we haven't started the
1888 * extent of buffers to map yet, the function returns 'true' immediately - the
1889 * caller can write the buffer right away. Otherwise the function returns true
1890 * if the block has been added to the extent, false if the block couldn't be
1893 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1894 struct buffer_head *bh)
1896 struct ext4_map_blocks *map = &mpd->map;
1898 /* Buffer that doesn't need mapping for writeback? */
1899 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1900 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1901 /* So far no extent to map => we write the buffer right away */
1902 if (map->m_len == 0)
1907 /* First block in the extent? */
1908 if (map->m_len == 0) {
1911 map->m_flags = bh->b_state & BH_FLAGS;
1915 /* Don't go larger than mballoc is willing to allocate */
1916 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1919 /* Can we merge the block to our big extent? */
1920 if (lblk == map->m_lblk + map->m_len &&
1921 (bh->b_state & BH_FLAGS) == map->m_flags) {
1929 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1931 * @mpd - extent of blocks for mapping
1932 * @head - the first buffer in the page
1933 * @bh - buffer we should start processing from
1934 * @lblk - logical number of the block in the file corresponding to @bh
1936 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1937 * the page for IO if all buffers in this page were mapped and there's no
1938 * accumulated extent of buffers to map or add buffers in the page to the
1939 * extent of buffers to map. The function returns 1 if the caller can continue
1940 * by processing the next page, 0 if it should stop adding buffers to the
1941 * extent to map because we cannot extend it anymore. It can also return value
1942 * < 0 in case of error during IO submission.
1944 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1945 struct buffer_head *head,
1946 struct buffer_head *bh,
1949 struct inode *inode = mpd->inode;
1951 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
1952 >> inode->i_blkbits;
1955 BUG_ON(buffer_locked(bh));
1957 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
1958 /* Found extent to map? */
1961 /* Everything mapped so far and we hit EOF */
1964 } while (lblk++, (bh = bh->b_this_page) != head);
1965 /* So far everything mapped? Submit the page for IO. */
1966 if (mpd->map.m_len == 0) {
1967 err = mpage_submit_page(mpd, head->b_page);
1971 return lblk < blocks;
1975 * mpage_map_buffers - update buffers corresponding to changed extent and
1976 * submit fully mapped pages for IO
1978 * @mpd - description of extent to map, on return next extent to map
1980 * Scan buffers corresponding to changed extent (we expect corresponding pages
1981 * to be already locked) and update buffer state according to new extent state.
1982 * We map delalloc buffers to their physical location, clear unwritten bits,
1983 * and mark buffers as uninit when we perform writes to unwritten extents
1984 * and do extent conversion after IO is finished. If the last page is not fully
1985 * mapped, we update @map to the next extent in the last page that needs
1986 * mapping. Otherwise we submit the page for IO.
1988 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
1990 struct pagevec pvec;
1992 struct inode *inode = mpd->inode;
1993 struct buffer_head *head, *bh;
1994 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
2000 start = mpd->map.m_lblk >> bpp_bits;
2001 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2002 lblk = start << bpp_bits;
2003 pblock = mpd->map.m_pblk;
2005 pagevec_init(&pvec, 0);
2006 while (start <= end) {
2007 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2011 for (i = 0; i < nr_pages; i++) {
2012 struct page *page = pvec.pages[i];
2014 if (page->index > end)
2016 /* Up to 'end' pages must be contiguous */
2017 BUG_ON(page->index != start);
2018 bh = head = page_buffers(page);
2020 if (lblk < mpd->map.m_lblk)
2022 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2024 * Buffer after end of mapped extent.
2025 * Find next buffer in the page to map.
2028 mpd->map.m_flags = 0;
2030 * FIXME: If dioread_nolock supports
2031 * blocksize < pagesize, we need to make
2032 * sure we add size mapped so far to
2033 * io_end->size as the following call
2034 * can submit the page for IO.
2036 err = mpage_process_page_bufs(mpd, head,
2038 pagevec_release(&pvec);
2043 if (buffer_delay(bh)) {
2044 clear_buffer_delay(bh);
2045 bh->b_blocknr = pblock++;
2047 clear_buffer_unwritten(bh);
2048 } while (lblk++, (bh = bh->b_this_page) != head);
2051 * FIXME: This is going to break if dioread_nolock
2052 * supports blocksize < pagesize as we will try to
2053 * convert potentially unmapped parts of inode.
2055 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2056 /* Page fully mapped - let IO run! */
2057 err = mpage_submit_page(mpd, page);
2059 pagevec_release(&pvec);
2064 pagevec_release(&pvec);
2066 /* Extent fully mapped and matches with page boundary. We are done. */
2068 mpd->map.m_flags = 0;
2072 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2074 struct inode *inode = mpd->inode;
2075 struct ext4_map_blocks *map = &mpd->map;
2076 int get_blocks_flags;
2077 int err, dioread_nolock;
2079 trace_ext4_da_write_pages_extent(inode, map);
2081 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2082 * to convert an unwritten extent to be initialized (in the case
2083 * where we have written into one or more preallocated blocks). It is
2084 * possible that we're going to need more metadata blocks than
2085 * previously reserved. However we must not fail because we're in
2086 * writeback and there is nothing we can do about it so it might result
2087 * in data loss. So use reserved blocks to allocate metadata if
2090 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2091 * the blocks in question are delalloc blocks. This indicates
2092 * that the blocks and quotas has already been checked when
2093 * the data was copied into the page cache.
2095 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2096 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2097 dioread_nolock = ext4_should_dioread_nolock(inode);
2099 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2100 if (map->m_flags & (1 << BH_Delay))
2101 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2103 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2106 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2107 if (!mpd->io_submit.io_end->handle &&
2108 ext4_handle_valid(handle)) {
2109 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2110 handle->h_rsv_handle = NULL;
2112 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2115 BUG_ON(map->m_len == 0);
2116 if (map->m_flags & EXT4_MAP_NEW) {
2117 struct block_device *bdev = inode->i_sb->s_bdev;
2120 for (i = 0; i < map->m_len; i++)
2121 unmap_underlying_metadata(bdev, map->m_pblk + i);
2127 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2128 * mpd->len and submit pages underlying it for IO
2130 * @handle - handle for journal operations
2131 * @mpd - extent to map
2132 * @give_up_on_write - we set this to true iff there is a fatal error and there
2133 * is no hope of writing the data. The caller should discard
2134 * dirty pages to avoid infinite loops.
2136 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2137 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2138 * them to initialized or split the described range from larger unwritten
2139 * extent. Note that we need not map all the described range since allocation
2140 * can return less blocks or the range is covered by more unwritten extents. We
2141 * cannot map more because we are limited by reserved transaction credits. On
2142 * the other hand we always make sure that the last touched page is fully
2143 * mapped so that it can be written out (and thus forward progress is
2144 * guaranteed). After mapping we submit all mapped pages for IO.
2146 static int mpage_map_and_submit_extent(handle_t *handle,
2147 struct mpage_da_data *mpd,
2148 bool *give_up_on_write)
2150 struct inode *inode = mpd->inode;
2151 struct ext4_map_blocks *map = &mpd->map;
2156 mpd->io_submit.io_end->offset =
2157 ((loff_t)map->m_lblk) << inode->i_blkbits;
2159 err = mpage_map_one_extent(handle, mpd);
2161 struct super_block *sb = inode->i_sb;
2163 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2164 goto invalidate_dirty_pages;
2166 * Let the uper layers retry transient errors.
2167 * In the case of ENOSPC, if ext4_count_free_blocks()
2168 * is non-zero, a commit should free up blocks.
2170 if ((err == -ENOMEM) ||
2171 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2173 goto update_disksize;
2176 ext4_msg(sb, KERN_CRIT,
2177 "Delayed block allocation failed for "
2178 "inode %lu at logical offset %llu with"
2179 " max blocks %u with error %d",
2181 (unsigned long long)map->m_lblk,
2182 (unsigned)map->m_len, -err);
2183 ext4_msg(sb, KERN_CRIT,
2184 "This should not happen!! Data will "
2187 ext4_print_free_blocks(inode);
2188 invalidate_dirty_pages:
2189 *give_up_on_write = true;
2194 * Update buffer state, submit mapped pages, and get us new
2197 err = mpage_map_and_submit_buffers(mpd);
2199 goto update_disksize;
2200 } while (map->m_len);
2204 * Update on-disk size after IO is submitted. Races with
2205 * truncate are avoided by checking i_size under i_data_sem.
2207 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2208 if (disksize > EXT4_I(inode)->i_disksize) {
2212 down_write(&EXT4_I(inode)->i_data_sem);
2213 i_size = i_size_read(inode);
2214 if (disksize > i_size)
2216 if (disksize > EXT4_I(inode)->i_disksize)
2217 EXT4_I(inode)->i_disksize = disksize;
2218 err2 = ext4_mark_inode_dirty(handle, inode);
2219 up_write(&EXT4_I(inode)->i_data_sem);
2221 ext4_error(inode->i_sb,
2222 "Failed to mark inode %lu dirty",
2231 * Calculate the total number of credits to reserve for one writepages
2232 * iteration. This is called from ext4_writepages(). We map an extent of
2233 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2234 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2235 * bpp - 1 blocks in bpp different extents.
2237 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2239 int bpp = ext4_journal_blocks_per_page(inode);
2241 return ext4_meta_trans_blocks(inode,
2242 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2246 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2247 * and underlying extent to map
2249 * @mpd - where to look for pages
2251 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2252 * IO immediately. When we find a page which isn't mapped we start accumulating
2253 * extent of buffers underlying these pages that needs mapping (formed by
2254 * either delayed or unwritten buffers). We also lock the pages containing
2255 * these buffers. The extent found is returned in @mpd structure (starting at
2256 * mpd->lblk with length mpd->len blocks).
2258 * Note that this function can attach bios to one io_end structure which are
2259 * neither logically nor physically contiguous. Although it may seem as an
2260 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2261 * case as we need to track IO to all buffers underlying a page in one io_end.
2263 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2265 struct address_space *mapping = mpd->inode->i_mapping;
2266 struct pagevec pvec;
2267 unsigned int nr_pages;
2268 long left = mpd->wbc->nr_to_write;
2269 pgoff_t index = mpd->first_page;
2270 pgoff_t end = mpd->last_page;
2273 int blkbits = mpd->inode->i_blkbits;
2275 struct buffer_head *head;
2277 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2278 tag = PAGECACHE_TAG_TOWRITE;
2280 tag = PAGECACHE_TAG_DIRTY;
2282 pagevec_init(&pvec, 0);
2284 mpd->next_page = index;
2285 while (index <= end) {
2286 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2287 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2291 for (i = 0; i < nr_pages; i++) {
2292 struct page *page = pvec.pages[i];
2295 * At this point, the page may be truncated or
2296 * invalidated (changing page->mapping to NULL), or
2297 * even swizzled back from swapper_space to tmpfs file
2298 * mapping. However, page->index will not change
2299 * because we have a reference on the page.
2301 if (page->index > end)
2305 * Accumulated enough dirty pages? This doesn't apply
2306 * to WB_SYNC_ALL mode. For integrity sync we have to
2307 * keep going because someone may be concurrently
2308 * dirtying pages, and we might have synced a lot of
2309 * newly appeared dirty pages, but have not synced all
2310 * of the old dirty pages.
2312 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2315 /* If we can't merge this page, we are done. */
2316 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2321 * If the page is no longer dirty, or its mapping no
2322 * longer corresponds to inode we are writing (which
2323 * means it has been truncated or invalidated), or the
2324 * page is already under writeback and we are not doing
2325 * a data integrity writeback, skip the page
2327 if (!PageDirty(page) ||
2328 (PageWriteback(page) &&
2329 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2330 unlikely(page->mapping != mapping)) {
2335 wait_on_page_writeback(page);
2336 BUG_ON(PageWriteback(page));
2338 if (mpd->map.m_len == 0)
2339 mpd->first_page = page->index;
2340 mpd->next_page = page->index + 1;
2341 /* Add all dirty buffers to mpd */
2342 lblk = ((ext4_lblk_t)page->index) <<
2343 (PAGE_CACHE_SHIFT - blkbits);
2344 head = page_buffers(page);
2345 err = mpage_process_page_bufs(mpd, head, head, lblk);
2351 pagevec_release(&pvec);
2356 pagevec_release(&pvec);
2360 static int __writepage(struct page *page, struct writeback_control *wbc,
2363 struct address_space *mapping = data;
2364 int ret = ext4_writepage(page, wbc);
2365 mapping_set_error(mapping, ret);
2369 static int ext4_writepages(struct address_space *mapping,
2370 struct writeback_control *wbc)
2372 pgoff_t writeback_index = 0;
2373 long nr_to_write = wbc->nr_to_write;
2374 int range_whole = 0;
2376 handle_t *handle = NULL;
2377 struct mpage_da_data mpd;
2378 struct inode *inode = mapping->host;
2379 int needed_blocks, rsv_blocks = 0, ret = 0;
2380 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2382 struct blk_plug plug;
2383 bool give_up_on_write = false;
2385 trace_ext4_writepages(inode, wbc);
2388 * No pages to write? This is mainly a kludge to avoid starting
2389 * a transaction for special inodes like journal inode on last iput()
2390 * because that could violate lock ordering on umount
2392 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2393 goto out_writepages;
2395 if (ext4_should_journal_data(inode)) {
2396 struct blk_plug plug;
2398 blk_start_plug(&plug);
2399 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2400 blk_finish_plug(&plug);
2401 goto out_writepages;
2405 * If the filesystem has aborted, it is read-only, so return
2406 * right away instead of dumping stack traces later on that
2407 * will obscure the real source of the problem. We test
2408 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2409 * the latter could be true if the filesystem is mounted
2410 * read-only, and in that case, ext4_writepages should
2411 * *never* be called, so if that ever happens, we would want
2414 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2416 goto out_writepages;
2419 if (ext4_should_dioread_nolock(inode)) {
2421 * We may need to convert up to one extent per block in
2422 * the page and we may dirty the inode.
2424 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2428 * If we have inline data and arrive here, it means that
2429 * we will soon create the block for the 1st page, so
2430 * we'd better clear the inline data here.
2432 if (ext4_has_inline_data(inode)) {
2433 /* Just inode will be modified... */
2434 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2435 if (IS_ERR(handle)) {
2436 ret = PTR_ERR(handle);
2437 goto out_writepages;
2439 BUG_ON(ext4_test_inode_state(inode,
2440 EXT4_STATE_MAY_INLINE_DATA));
2441 ext4_destroy_inline_data(handle, inode);
2442 ext4_journal_stop(handle);
2445 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2448 if (wbc->range_cyclic) {
2449 writeback_index = mapping->writeback_index;
2450 if (writeback_index)
2452 mpd.first_page = writeback_index;
2455 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2456 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2461 ext4_io_submit_init(&mpd.io_submit, wbc);
2463 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2464 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2466 blk_start_plug(&plug);
2467 while (!done && mpd.first_page <= mpd.last_page) {
2468 /* For each extent of pages we use new io_end */
2469 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2470 if (!mpd.io_submit.io_end) {
2476 * We have two constraints: We find one extent to map and we
2477 * must always write out whole page (makes a difference when
2478 * blocksize < pagesize) so that we don't block on IO when we
2479 * try to write out the rest of the page. Journalled mode is
2480 * not supported by delalloc.
2482 BUG_ON(ext4_should_journal_data(inode));
2483 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2485 /* start a new transaction */
2486 handle = ext4_journal_start_with_reserve(inode,
2487 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2488 if (IS_ERR(handle)) {
2489 ret = PTR_ERR(handle);
2490 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2491 "%ld pages, ino %lu; err %d", __func__,
2492 wbc->nr_to_write, inode->i_ino, ret);
2493 /* Release allocated io_end */
2494 ext4_put_io_end(mpd.io_submit.io_end);
2498 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2499 ret = mpage_prepare_extent_to_map(&mpd);
2502 ret = mpage_map_and_submit_extent(handle, &mpd,
2506 * We scanned the whole range (or exhausted
2507 * nr_to_write), submitted what was mapped and
2508 * didn't find anything needing mapping. We are
2514 ext4_journal_stop(handle);
2515 /* Submit prepared bio */
2516 ext4_io_submit(&mpd.io_submit);
2517 /* Unlock pages we didn't use */
2518 mpage_release_unused_pages(&mpd, give_up_on_write);
2519 /* Drop our io_end reference we got from init */
2520 ext4_put_io_end(mpd.io_submit.io_end);
2522 if (ret == -ENOSPC && sbi->s_journal) {
2524 * Commit the transaction which would
2525 * free blocks released in the transaction
2528 jbd2_journal_force_commit_nested(sbi->s_journal);
2532 /* Fatal error - ENOMEM, EIO... */
2536 blk_finish_plug(&plug);
2537 if (!ret && !cycled && wbc->nr_to_write > 0) {
2539 mpd.last_page = writeback_index - 1;
2545 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2547 * Set the writeback_index so that range_cyclic
2548 * mode will write it back later
2550 mapping->writeback_index = mpd.first_page;
2553 trace_ext4_writepages_result(inode, wbc, ret,
2554 nr_to_write - wbc->nr_to_write);
2558 static int ext4_nonda_switch(struct super_block *sb)
2560 s64 free_clusters, dirty_clusters;
2561 struct ext4_sb_info *sbi = EXT4_SB(sb);
2564 * switch to non delalloc mode if we are running low
2565 * on free block. The free block accounting via percpu
2566 * counters can get slightly wrong with percpu_counter_batch getting
2567 * accumulated on each CPU without updating global counters
2568 * Delalloc need an accurate free block accounting. So switch
2569 * to non delalloc when we are near to error range.
2572 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2574 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2576 * Start pushing delalloc when 1/2 of free blocks are dirty.
2578 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2579 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2581 if (2 * free_clusters < 3 * dirty_clusters ||
2582 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2584 * free block count is less than 150% of dirty blocks
2585 * or free blocks is less than watermark
2592 /* We always reserve for an inode update; the superblock could be there too */
2593 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2595 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
2596 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2599 if (pos + len <= 0x7fffffffULL)
2602 /* We might need to update the superblock to set LARGE_FILE */
2606 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2607 loff_t pos, unsigned len, unsigned flags,
2608 struct page **pagep, void **fsdata)
2610 int ret, retries = 0;
2613 struct inode *inode = mapping->host;
2616 index = pos >> PAGE_CACHE_SHIFT;
2618 if (ext4_nonda_switch(inode->i_sb)) {
2619 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2620 return ext4_write_begin(file, mapping, pos,
2621 len, flags, pagep, fsdata);
2623 *fsdata = (void *)0;
2624 trace_ext4_da_write_begin(inode, pos, len, flags);
2626 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2627 ret = ext4_da_write_inline_data_begin(mapping, inode,
2637 * grab_cache_page_write_begin() can take a long time if the
2638 * system is thrashing due to memory pressure, or if the page
2639 * is being written back. So grab it first before we start
2640 * the transaction handle. This also allows us to allocate
2641 * the page (if needed) without using GFP_NOFS.
2644 page = grab_cache_page_write_begin(mapping, index, flags);
2650 * With delayed allocation, we don't log the i_disksize update
2651 * if there is delayed block allocation. But we still need
2652 * to journalling the i_disksize update if writes to the end
2653 * of file which has an already mapped buffer.
2656 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2657 ext4_da_write_credits(inode, pos, len));
2658 if (IS_ERR(handle)) {
2659 page_cache_release(page);
2660 return PTR_ERR(handle);
2664 if (page->mapping != mapping) {
2665 /* The page got truncated from under us */
2667 page_cache_release(page);
2668 ext4_journal_stop(handle);
2671 /* In case writeback began while the page was unlocked */
2672 wait_for_stable_page(page);
2674 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2675 ret = ext4_block_write_begin(page, pos, len,
2676 ext4_da_get_block_prep);
2678 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2682 ext4_journal_stop(handle);
2684 * block_write_begin may have instantiated a few blocks
2685 * outside i_size. Trim these off again. Don't need
2686 * i_size_read because we hold i_mutex.
2688 if (pos + len > inode->i_size)
2689 ext4_truncate_failed_write(inode);
2691 if (ret == -ENOSPC &&
2692 ext4_should_retry_alloc(inode->i_sb, &retries))
2695 page_cache_release(page);
2704 * Check if we should update i_disksize
2705 * when write to the end of file but not require block allocation
2707 static int ext4_da_should_update_i_disksize(struct page *page,
2708 unsigned long offset)
2710 struct buffer_head *bh;
2711 struct inode *inode = page->mapping->host;
2715 bh = page_buffers(page);
2716 idx = offset >> inode->i_blkbits;
2718 for (i = 0; i < idx; i++)
2719 bh = bh->b_this_page;
2721 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2726 static int ext4_da_write_end(struct file *file,
2727 struct address_space *mapping,
2728 loff_t pos, unsigned len, unsigned copied,
2729 struct page *page, void *fsdata)
2731 struct inode *inode = mapping->host;
2733 handle_t *handle = ext4_journal_current_handle();
2735 unsigned long start, end;
2736 int write_mode = (int)(unsigned long)fsdata;
2738 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2739 return ext4_write_end(file, mapping, pos,
2740 len, copied, page, fsdata);
2742 trace_ext4_da_write_end(inode, pos, len, copied);
2743 start = pos & (PAGE_CACHE_SIZE - 1);
2744 end = start + copied - 1;
2747 * generic_write_end() will run mark_inode_dirty() if i_size
2748 * changes. So let's piggyback the i_disksize mark_inode_dirty
2751 new_i_size = pos + copied;
2752 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2753 if (ext4_has_inline_data(inode) ||
2754 ext4_da_should_update_i_disksize(page, end)) {
2755 ext4_update_i_disksize(inode, new_i_size);
2756 /* We need to mark inode dirty even if
2757 * new_i_size is less that inode->i_size
2758 * bu greater than i_disksize.(hint delalloc)
2760 ext4_mark_inode_dirty(handle, inode);
2764 if (write_mode != CONVERT_INLINE_DATA &&
2765 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2766 ext4_has_inline_data(inode))
2767 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2770 ret2 = generic_write_end(file, mapping, pos, len, copied,
2776 ret2 = ext4_journal_stop(handle);
2780 return ret ? ret : copied;
2783 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2784 unsigned int length)
2787 * Drop reserved blocks
2789 BUG_ON(!PageLocked(page));
2790 if (!page_has_buffers(page))
2793 ext4_da_page_release_reservation(page, offset, length);
2796 ext4_invalidatepage(page, offset, length);
2802 * Force all delayed allocation blocks to be allocated for a given inode.
2804 int ext4_alloc_da_blocks(struct inode *inode)
2806 trace_ext4_alloc_da_blocks(inode);
2808 if (!EXT4_I(inode)->i_reserved_data_blocks)
2812 * We do something simple for now. The filemap_flush() will
2813 * also start triggering a write of the data blocks, which is
2814 * not strictly speaking necessary (and for users of
2815 * laptop_mode, not even desirable). However, to do otherwise
2816 * would require replicating code paths in:
2818 * ext4_writepages() ->
2819 * write_cache_pages() ---> (via passed in callback function)
2820 * __mpage_da_writepage() -->
2821 * mpage_add_bh_to_extent()
2822 * mpage_da_map_blocks()
2824 * The problem is that write_cache_pages(), located in
2825 * mm/page-writeback.c, marks pages clean in preparation for
2826 * doing I/O, which is not desirable if we're not planning on
2829 * We could call write_cache_pages(), and then redirty all of
2830 * the pages by calling redirty_page_for_writepage() but that
2831 * would be ugly in the extreme. So instead we would need to
2832 * replicate parts of the code in the above functions,
2833 * simplifying them because we wouldn't actually intend to
2834 * write out the pages, but rather only collect contiguous
2835 * logical block extents, call the multi-block allocator, and
2836 * then update the buffer heads with the block allocations.
2838 * For now, though, we'll cheat by calling filemap_flush(),
2839 * which will map the blocks, and start the I/O, but not
2840 * actually wait for the I/O to complete.
2842 return filemap_flush(inode->i_mapping);
2846 * bmap() is special. It gets used by applications such as lilo and by
2847 * the swapper to find the on-disk block of a specific piece of data.
2849 * Naturally, this is dangerous if the block concerned is still in the
2850 * journal. If somebody makes a swapfile on an ext4 data-journaling
2851 * filesystem and enables swap, then they may get a nasty shock when the
2852 * data getting swapped to that swapfile suddenly gets overwritten by
2853 * the original zero's written out previously to the journal and
2854 * awaiting writeback in the kernel's buffer cache.
2856 * So, if we see any bmap calls here on a modified, data-journaled file,
2857 * take extra steps to flush any blocks which might be in the cache.
2859 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2861 struct inode *inode = mapping->host;
2866 * We can get here for an inline file via the FIBMAP ioctl
2868 if (ext4_has_inline_data(inode))
2871 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2872 test_opt(inode->i_sb, DELALLOC)) {
2874 * With delalloc we want to sync the file
2875 * so that we can make sure we allocate
2878 filemap_write_and_wait(mapping);
2881 if (EXT4_JOURNAL(inode) &&
2882 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2884 * This is a REALLY heavyweight approach, but the use of
2885 * bmap on dirty files is expected to be extremely rare:
2886 * only if we run lilo or swapon on a freshly made file
2887 * do we expect this to happen.
2889 * (bmap requires CAP_SYS_RAWIO so this does not
2890 * represent an unprivileged user DOS attack --- we'd be
2891 * in trouble if mortal users could trigger this path at
2894 * NB. EXT4_STATE_JDATA is not set on files other than
2895 * regular files. If somebody wants to bmap a directory
2896 * or symlink and gets confused because the buffer
2897 * hasn't yet been flushed to disk, they deserve
2898 * everything they get.
2901 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2902 journal = EXT4_JOURNAL(inode);
2903 jbd2_journal_lock_updates(journal);
2904 err = jbd2_journal_flush(journal);
2905 jbd2_journal_unlock_updates(journal);
2911 return generic_block_bmap(mapping, block, ext4_get_block);
2914 static int ext4_readpage(struct file *file, struct page *page)
2917 struct inode *inode = page->mapping->host;
2919 trace_ext4_readpage(page);
2921 if (ext4_has_inline_data(inode))
2922 ret = ext4_readpage_inline(inode, page);
2925 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
2931 ext4_readpages(struct file *file, struct address_space *mapping,
2932 struct list_head *pages, unsigned nr_pages)
2934 struct inode *inode = mapping->host;
2936 /* If the file has inline data, no need to do readpages. */
2937 if (ext4_has_inline_data(inode))
2940 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
2943 static void ext4_invalidatepage(struct page *page, unsigned int offset,
2944 unsigned int length)
2946 trace_ext4_invalidatepage(page, offset, length);
2948 /* No journalling happens on data buffers when this function is used */
2949 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2951 block_invalidatepage(page, offset, length);
2954 static int __ext4_journalled_invalidatepage(struct page *page,
2955 unsigned int offset,
2956 unsigned int length)
2958 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2960 trace_ext4_journalled_invalidatepage(page, offset, length);
2963 * If it's a full truncate we just forget about the pending dirtying
2965 if (offset == 0 && length == PAGE_CACHE_SIZE)
2966 ClearPageChecked(page);
2968 return jbd2_journal_invalidatepage(journal, page, offset, length);
2971 /* Wrapper for aops... */
2972 static void ext4_journalled_invalidatepage(struct page *page,
2973 unsigned int offset,
2974 unsigned int length)
2976 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
2979 static int ext4_releasepage(struct page *page, gfp_t wait)
2981 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2983 trace_ext4_releasepage(page);
2985 /* Page has dirty journalled data -> cannot release */
2986 if (PageChecked(page))
2989 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2991 return try_to_free_buffers(page);
2995 * ext4_get_block used when preparing for a DIO write or buffer write.
2996 * We allocate an uinitialized extent if blocks haven't been allocated.
2997 * The extent will be converted to initialized after the IO is complete.
2999 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3000 struct buffer_head *bh_result, int create)
3002 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3003 inode->i_ino, create);
3004 return _ext4_get_block(inode, iblock, bh_result,
3005 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3008 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3009 struct buffer_head *bh_result, int create)
3011 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3012 inode->i_ino, create);
3013 return _ext4_get_block(inode, iblock, bh_result,
3014 EXT4_GET_BLOCKS_NO_LOCK);
3017 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3018 ssize_t size, void *private)
3020 ext4_io_end_t *io_end = iocb->private;
3022 /* if not async direct IO just return */
3026 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3027 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3028 iocb->private, io_end->inode->i_ino, iocb, offset,
3031 iocb->private = NULL;
3032 io_end->offset = offset;
3033 io_end->size = size;
3034 ext4_put_io_end(io_end);
3038 * For ext4 extent files, ext4 will do direct-io write to holes,
3039 * preallocated extents, and those write extend the file, no need to
3040 * fall back to buffered IO.
3042 * For holes, we fallocate those blocks, mark them as unwritten
3043 * If those blocks were preallocated, we mark sure they are split, but
3044 * still keep the range to write as unwritten.
3046 * The unwritten extents will be converted to written when DIO is completed.
3047 * For async direct IO, since the IO may still pending when return, we
3048 * set up an end_io call back function, which will do the conversion
3049 * when async direct IO completed.
3051 * If the O_DIRECT write will extend the file then add this inode to the
3052 * orphan list. So recovery will truncate it back to the original size
3053 * if the machine crashes during the write.
3056 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3057 struct iov_iter *iter, loff_t offset)
3059 struct file *file = iocb->ki_filp;
3060 struct inode *inode = file->f_mapping->host;
3062 size_t count = iov_iter_count(iter);
3064 get_block_t *get_block_func = NULL;
3066 loff_t final_size = offset + count;
3067 ext4_io_end_t *io_end = NULL;
3069 /* Use the old path for reads and writes beyond i_size. */
3070 if (rw != WRITE || final_size > inode->i_size)
3071 return ext4_ind_direct_IO(rw, iocb, iter, offset);
3073 BUG_ON(iocb->private == NULL);
3076 * Make all waiters for direct IO properly wait also for extent
3077 * conversion. This also disallows race between truncate() and
3078 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3081 atomic_inc(&inode->i_dio_count);
3083 /* If we do a overwrite dio, i_mutex locking can be released */
3084 overwrite = *((int *)iocb->private);
3087 down_read(&EXT4_I(inode)->i_data_sem);
3088 mutex_unlock(&inode->i_mutex);
3092 * We could direct write to holes and fallocate.
3094 * Allocated blocks to fill the hole are marked as
3095 * unwritten to prevent parallel buffered read to expose
3096 * the stale data before DIO complete the data IO.
3098 * As to previously fallocated extents, ext4 get_block will
3099 * just simply mark the buffer mapped but still keep the
3100 * extents unwritten.
3102 * For non AIO case, we will convert those unwritten extents
3103 * to written after return back from blockdev_direct_IO.
3105 * For async DIO, the conversion needs to be deferred when the
3106 * IO is completed. The ext4 end_io callback function will be
3107 * called to take care of the conversion work. Here for async
3108 * case, we allocate an io_end structure to hook to the iocb.
3110 iocb->private = NULL;
3111 ext4_inode_aio_set(inode, NULL);
3112 if (!is_sync_kiocb(iocb)) {
3113 io_end = ext4_init_io_end(inode, GFP_NOFS);
3119 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3121 iocb->private = ext4_get_io_end(io_end);
3123 * we save the io structure for current async direct
3124 * IO, so that later ext4_map_blocks() could flag the
3125 * io structure whether there is a unwritten extents
3126 * needs to be converted when IO is completed.
3128 ext4_inode_aio_set(inode, io_end);
3132 get_block_func = ext4_get_block_write_nolock;
3134 get_block_func = ext4_get_block_write;
3135 dio_flags = DIO_LOCKING;
3137 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3138 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3141 ret = dax_do_io(rw, iocb, inode, iter, offset, get_block_func,
3142 ext4_end_io_dio, dio_flags);
3144 ret = __blockdev_direct_IO(rw, iocb, inode,
3145 inode->i_sb->s_bdev, iter, offset,
3147 ext4_end_io_dio, NULL, dio_flags);
3150 * Put our reference to io_end. This can free the io_end structure e.g.
3151 * in sync IO case or in case of error. It can even perform extent
3152 * conversion if all bios we submitted finished before we got here.
3153 * Note that in that case iocb->private can be already set to NULL
3157 ext4_inode_aio_set(inode, NULL);
3158 ext4_put_io_end(io_end);
3160 * When no IO was submitted ext4_end_io_dio() was not
3161 * called so we have to put iocb's reference.
3163 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3164 WARN_ON(iocb->private != io_end);
3165 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3166 ext4_put_io_end(io_end);
3167 iocb->private = NULL;
3170 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3171 EXT4_STATE_DIO_UNWRITTEN)) {
3174 * for non AIO case, since the IO is already
3175 * completed, we could do the conversion right here
3177 err = ext4_convert_unwritten_extents(NULL, inode,
3181 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3186 inode_dio_done(inode);
3187 /* take i_mutex locking again if we do a ovewrite dio */
3189 up_read(&EXT4_I(inode)->i_data_sem);
3190 mutex_lock(&inode->i_mutex);
3196 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3197 struct iov_iter *iter, loff_t offset)
3199 struct file *file = iocb->ki_filp;
3200 struct inode *inode = file->f_mapping->host;
3201 size_t count = iov_iter_count(iter);
3204 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3205 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3210 * If we are doing data journalling we don't support O_DIRECT
3212 if (ext4_should_journal_data(inode))
3215 /* Let buffer I/O handle the inline data case. */
3216 if (ext4_has_inline_data(inode))
3219 trace_ext4_direct_IO_enter(inode, offset, count, rw);
3220 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3221 ret = ext4_ext_direct_IO(rw, iocb, iter, offset);
3223 ret = ext4_ind_direct_IO(rw, iocb, iter, offset);
3224 trace_ext4_direct_IO_exit(inode, offset, count, rw, ret);
3229 * Pages can be marked dirty completely asynchronously from ext4's journalling
3230 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3231 * much here because ->set_page_dirty is called under VFS locks. The page is
3232 * not necessarily locked.
3234 * We cannot just dirty the page and leave attached buffers clean, because the
3235 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3236 * or jbddirty because all the journalling code will explode.
3238 * So what we do is to mark the page "pending dirty" and next time writepage
3239 * is called, propagate that into the buffers appropriately.
3241 static int ext4_journalled_set_page_dirty(struct page *page)
3243 SetPageChecked(page);
3244 return __set_page_dirty_nobuffers(page);
3247 static const struct address_space_operations ext4_aops = {
3248 .readpage = ext4_readpage,
3249 .readpages = ext4_readpages,
3250 .writepage = ext4_writepage,
3251 .writepages = ext4_writepages,
3252 .write_begin = ext4_write_begin,
3253 .write_end = ext4_write_end,
3255 .invalidatepage = ext4_invalidatepage,
3256 .releasepage = ext4_releasepage,
3257 .direct_IO = ext4_direct_IO,
3258 .migratepage = buffer_migrate_page,
3259 .is_partially_uptodate = block_is_partially_uptodate,
3260 .error_remove_page = generic_error_remove_page,
3263 static const struct address_space_operations ext4_journalled_aops = {
3264 .readpage = ext4_readpage,
3265 .readpages = ext4_readpages,
3266 .writepage = ext4_writepage,
3267 .writepages = ext4_writepages,
3268 .write_begin = ext4_write_begin,
3269 .write_end = ext4_journalled_write_end,
3270 .set_page_dirty = ext4_journalled_set_page_dirty,
3272 .invalidatepage = ext4_journalled_invalidatepage,
3273 .releasepage = ext4_releasepage,
3274 .direct_IO = ext4_direct_IO,
3275 .is_partially_uptodate = block_is_partially_uptodate,
3276 .error_remove_page = generic_error_remove_page,
3279 static const struct address_space_operations ext4_da_aops = {
3280 .readpage = ext4_readpage,
3281 .readpages = ext4_readpages,
3282 .writepage = ext4_writepage,
3283 .writepages = ext4_writepages,
3284 .write_begin = ext4_da_write_begin,
3285 .write_end = ext4_da_write_end,
3287 .invalidatepage = ext4_da_invalidatepage,
3288 .releasepage = ext4_releasepage,
3289 .direct_IO = ext4_direct_IO,
3290 .migratepage = buffer_migrate_page,
3291 .is_partially_uptodate = block_is_partially_uptodate,
3292 .error_remove_page = generic_error_remove_page,
3295 void ext4_set_aops(struct inode *inode)
3297 switch (ext4_inode_journal_mode(inode)) {
3298 case EXT4_INODE_ORDERED_DATA_MODE:
3299 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3301 case EXT4_INODE_WRITEBACK_DATA_MODE:
3302 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3304 case EXT4_INODE_JOURNAL_DATA_MODE:
3305 inode->i_mapping->a_ops = &ext4_journalled_aops;
3310 if (test_opt(inode->i_sb, DELALLOC))
3311 inode->i_mapping->a_ops = &ext4_da_aops;
3313 inode->i_mapping->a_ops = &ext4_aops;
3316 static int __ext4_block_zero_page_range(handle_t *handle,
3317 struct address_space *mapping, loff_t from, loff_t length)
3319 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3320 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3321 unsigned blocksize, pos;
3323 struct inode *inode = mapping->host;
3324 struct buffer_head *bh;
3328 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3329 mapping_gfp_mask(mapping) & ~__GFP_FS);
3333 blocksize = inode->i_sb->s_blocksize;
3335 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3337 if (!page_has_buffers(page))
3338 create_empty_buffers(page, blocksize, 0);
3340 /* Find the buffer that contains "offset" */
3341 bh = page_buffers(page);
3343 while (offset >= pos) {
3344 bh = bh->b_this_page;
3348 if (buffer_freed(bh)) {
3349 BUFFER_TRACE(bh, "freed: skip");
3352 if (!buffer_mapped(bh)) {
3353 BUFFER_TRACE(bh, "unmapped");
3354 ext4_get_block(inode, iblock, bh, 0);
3355 /* unmapped? It's a hole - nothing to do */
3356 if (!buffer_mapped(bh)) {
3357 BUFFER_TRACE(bh, "still unmapped");
3362 /* Ok, it's mapped. Make sure it's up-to-date */
3363 if (PageUptodate(page))
3364 set_buffer_uptodate(bh);
3366 if (!buffer_uptodate(bh)) {
3368 ll_rw_block(READ, 1, &bh);
3370 /* Uhhuh. Read error. Complain and punt. */
3371 if (!buffer_uptodate(bh))
3373 if (S_ISREG(inode->i_mode) &&
3374 ext4_encrypted_inode(inode)) {
3375 /* We expect the key to be set. */
3376 BUG_ON(!ext4_has_encryption_key(inode));
3377 BUG_ON(blocksize != PAGE_CACHE_SIZE);
3378 WARN_ON_ONCE(ext4_decrypt_one(inode, page));
3381 if (ext4_should_journal_data(inode)) {
3382 BUFFER_TRACE(bh, "get write access");
3383 err = ext4_journal_get_write_access(handle, bh);
3387 zero_user(page, offset, length);
3388 BUFFER_TRACE(bh, "zeroed end of block");
3390 if (ext4_should_journal_data(inode)) {
3391 err = ext4_handle_dirty_metadata(handle, inode, bh);
3394 mark_buffer_dirty(bh);
3395 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3396 err = ext4_jbd2_file_inode(handle, inode);
3401 page_cache_release(page);
3406 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3407 * starting from file offset 'from'. The range to be zero'd must
3408 * be contained with in one block. If the specified range exceeds
3409 * the end of the block it will be shortened to end of the block
3410 * that cooresponds to 'from'
3412 static int ext4_block_zero_page_range(handle_t *handle,
3413 struct address_space *mapping, loff_t from, loff_t length)
3415 struct inode *inode = mapping->host;
3416 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3417 unsigned blocksize = inode->i_sb->s_blocksize;
3418 unsigned max = blocksize - (offset & (blocksize - 1));
3421 * correct length if it does not fall between
3422 * 'from' and the end of the block
3424 if (length > max || length < 0)
3428 return dax_zero_page_range(inode, from, length, ext4_get_block);
3429 return __ext4_block_zero_page_range(handle, mapping, from, length);
3433 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3434 * up to the end of the block which corresponds to `from'.
3435 * This required during truncate. We need to physically zero the tail end
3436 * of that block so it doesn't yield old data if the file is later grown.
3438 static int ext4_block_truncate_page(handle_t *handle,
3439 struct address_space *mapping, loff_t from)
3441 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3444 struct inode *inode = mapping->host;
3446 blocksize = inode->i_sb->s_blocksize;
3447 length = blocksize - (offset & (blocksize - 1));
3449 return ext4_block_zero_page_range(handle, mapping, from, length);
3452 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3453 loff_t lstart, loff_t length)
3455 struct super_block *sb = inode->i_sb;
3456 struct address_space *mapping = inode->i_mapping;
3457 unsigned partial_start, partial_end;
3458 ext4_fsblk_t start, end;
3459 loff_t byte_end = (lstart + length - 1);
3462 partial_start = lstart & (sb->s_blocksize - 1);
3463 partial_end = byte_end & (sb->s_blocksize - 1);
3465 start = lstart >> sb->s_blocksize_bits;
3466 end = byte_end >> sb->s_blocksize_bits;
3468 /* Handle partial zero within the single block */
3470 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3471 err = ext4_block_zero_page_range(handle, mapping,
3475 /* Handle partial zero out on the start of the range */
3476 if (partial_start) {
3477 err = ext4_block_zero_page_range(handle, mapping,
3478 lstart, sb->s_blocksize);
3482 /* Handle partial zero out on the end of the range */
3483 if (partial_end != sb->s_blocksize - 1)
3484 err = ext4_block_zero_page_range(handle, mapping,
3485 byte_end - partial_end,
3490 int ext4_can_truncate(struct inode *inode)
3492 if (S_ISREG(inode->i_mode))
3494 if (S_ISDIR(inode->i_mode))
3496 if (S_ISLNK(inode->i_mode))
3497 return !ext4_inode_is_fast_symlink(inode);
3502 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3503 * associated with the given offset and length
3505 * @inode: File inode
3506 * @offset: The offset where the hole will begin
3507 * @len: The length of the hole
3509 * Returns: 0 on success or negative on failure
3512 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3514 struct super_block *sb = inode->i_sb;
3515 ext4_lblk_t first_block, stop_block;
3516 struct address_space *mapping = inode->i_mapping;
3517 loff_t first_block_offset, last_block_offset;
3519 unsigned int credits;
3522 if (!S_ISREG(inode->i_mode))
3525 trace_ext4_punch_hole(inode, offset, length, 0);
3528 * Write out all dirty pages to avoid race conditions
3529 * Then release them.
3531 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3532 ret = filemap_write_and_wait_range(mapping, offset,
3533 offset + length - 1);
3538 mutex_lock(&inode->i_mutex);
3540 /* No need to punch hole beyond i_size */
3541 if (offset >= inode->i_size)
3545 * If the hole extends beyond i_size, set the hole
3546 * to end after the page that contains i_size
3548 if (offset + length > inode->i_size) {
3549 length = inode->i_size +
3550 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3554 if (offset & (sb->s_blocksize - 1) ||
3555 (offset + length) & (sb->s_blocksize - 1)) {
3557 * Attach jinode to inode for jbd2 if we do any zeroing of
3560 ret = ext4_inode_attach_jinode(inode);
3566 first_block_offset = round_up(offset, sb->s_blocksize);
3567 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3569 /* Now release the pages and zero block aligned part of pages*/
3570 if (last_block_offset > first_block_offset)
3571 truncate_pagecache_range(inode, first_block_offset,
3574 /* Wait all existing dio workers, newcomers will block on i_mutex */
3575 ext4_inode_block_unlocked_dio(inode);
3576 inode_dio_wait(inode);
3578 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3579 credits = ext4_writepage_trans_blocks(inode);
3581 credits = ext4_blocks_for_truncate(inode);
3582 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3583 if (IS_ERR(handle)) {
3584 ret = PTR_ERR(handle);
3585 ext4_std_error(sb, ret);
3589 ret = ext4_zero_partial_blocks(handle, inode, offset,
3594 first_block = (offset + sb->s_blocksize - 1) >>
3595 EXT4_BLOCK_SIZE_BITS(sb);
3596 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3598 /* If there are no blocks to remove, return now */
3599 if (first_block >= stop_block)
3602 down_write(&EXT4_I(inode)->i_data_sem);
3603 ext4_discard_preallocations(inode);
3605 ret = ext4_es_remove_extent(inode, first_block,
3606 stop_block - first_block);
3608 up_write(&EXT4_I(inode)->i_data_sem);
3612 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3613 ret = ext4_ext_remove_space(inode, first_block,
3616 ret = ext4_ind_remove_space(handle, inode, first_block,
3619 up_write(&EXT4_I(inode)->i_data_sem);
3621 ext4_handle_sync(handle);
3623 /* Now release the pages again to reduce race window */
3624 if (last_block_offset > first_block_offset)
3625 truncate_pagecache_range(inode, first_block_offset,
3628 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3629 ext4_mark_inode_dirty(handle, inode);
3631 ext4_journal_stop(handle);
3633 ext4_inode_resume_unlocked_dio(inode);
3635 mutex_unlock(&inode->i_mutex);
3639 int ext4_inode_attach_jinode(struct inode *inode)
3641 struct ext4_inode_info *ei = EXT4_I(inode);
3642 struct jbd2_inode *jinode;
3644 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3647 jinode = jbd2_alloc_inode(GFP_KERNEL);
3648 spin_lock(&inode->i_lock);
3651 spin_unlock(&inode->i_lock);
3654 ei->jinode = jinode;
3655 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3658 spin_unlock(&inode->i_lock);
3659 if (unlikely(jinode != NULL))
3660 jbd2_free_inode(jinode);
3667 * We block out ext4_get_block() block instantiations across the entire
3668 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3669 * simultaneously on behalf of the same inode.
3671 * As we work through the truncate and commit bits of it to the journal there
3672 * is one core, guiding principle: the file's tree must always be consistent on
3673 * disk. We must be able to restart the truncate after a crash.
3675 * The file's tree may be transiently inconsistent in memory (although it
3676 * probably isn't), but whenever we close off and commit a journal transaction,
3677 * the contents of (the filesystem + the journal) must be consistent and
3678 * restartable. It's pretty simple, really: bottom up, right to left (although
3679 * left-to-right works OK too).
3681 * Note that at recovery time, journal replay occurs *before* the restart of
3682 * truncate against the orphan inode list.
3684 * The committed inode has the new, desired i_size (which is the same as
3685 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3686 * that this inode's truncate did not complete and it will again call
3687 * ext4_truncate() to have another go. So there will be instantiated blocks
3688 * to the right of the truncation point in a crashed ext4 filesystem. But
3689 * that's fine - as long as they are linked from the inode, the post-crash
3690 * ext4_truncate() run will find them and release them.
3692 void ext4_truncate(struct inode *inode)
3694 struct ext4_inode_info *ei = EXT4_I(inode);
3695 unsigned int credits;
3697 struct address_space *mapping = inode->i_mapping;
3700 * There is a possibility that we're either freeing the inode
3701 * or it's a completely new inode. In those cases we might not
3702 * have i_mutex locked because it's not necessary.
3704 if (!(inode->i_state & (I_NEW|I_FREEING)))
3705 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3706 trace_ext4_truncate_enter(inode);
3708 if (!ext4_can_truncate(inode))
3711 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3713 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3714 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3716 if (ext4_has_inline_data(inode)) {
3719 ext4_inline_data_truncate(inode, &has_inline);
3724 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3725 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3726 if (ext4_inode_attach_jinode(inode) < 0)
3730 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3731 credits = ext4_writepage_trans_blocks(inode);
3733 credits = ext4_blocks_for_truncate(inode);
3735 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3736 if (IS_ERR(handle)) {
3737 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3741 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3742 ext4_block_truncate_page(handle, mapping, inode->i_size);
3745 * We add the inode to the orphan list, so that if this
3746 * truncate spans multiple transactions, and we crash, we will
3747 * resume the truncate when the filesystem recovers. It also
3748 * marks the inode dirty, to catch the new size.
3750 * Implication: the file must always be in a sane, consistent
3751 * truncatable state while each transaction commits.
3753 if (ext4_orphan_add(handle, inode))
3756 down_write(&EXT4_I(inode)->i_data_sem);
3758 ext4_discard_preallocations(inode);
3760 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3761 ext4_ext_truncate(handle, inode);
3763 ext4_ind_truncate(handle, inode);
3765 up_write(&ei->i_data_sem);
3768 ext4_handle_sync(handle);
3772 * If this was a simple ftruncate() and the file will remain alive,
3773 * then we need to clear up the orphan record which we created above.
3774 * However, if this was a real unlink then we were called by
3775 * ext4_evict_inode(), and we allow that function to clean up the
3776 * orphan info for us.
3779 ext4_orphan_del(handle, inode);
3781 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3782 ext4_mark_inode_dirty(handle, inode);
3783 ext4_journal_stop(handle);
3785 trace_ext4_truncate_exit(inode);
3789 * ext4_get_inode_loc returns with an extra refcount against the inode's
3790 * underlying buffer_head on success. If 'in_mem' is true, we have all
3791 * data in memory that is needed to recreate the on-disk version of this
3794 static int __ext4_get_inode_loc(struct inode *inode,
3795 struct ext4_iloc *iloc, int in_mem)
3797 struct ext4_group_desc *gdp;
3798 struct buffer_head *bh;
3799 struct super_block *sb = inode->i_sb;
3801 int inodes_per_block, inode_offset;
3804 if (!ext4_valid_inum(sb, inode->i_ino))
3807 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3808 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3813 * Figure out the offset within the block group inode table
3815 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3816 inode_offset = ((inode->i_ino - 1) %
3817 EXT4_INODES_PER_GROUP(sb));
3818 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3819 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3821 bh = sb_getblk(sb, block);
3824 if (!buffer_uptodate(bh)) {
3828 * If the buffer has the write error flag, we have failed
3829 * to write out another inode in the same block. In this
3830 * case, we don't have to read the block because we may
3831 * read the old inode data successfully.
3833 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3834 set_buffer_uptodate(bh);
3836 if (buffer_uptodate(bh)) {
3837 /* someone brought it uptodate while we waited */
3843 * If we have all information of the inode in memory and this
3844 * is the only valid inode in the block, we need not read the
3848 struct buffer_head *bitmap_bh;
3851 start = inode_offset & ~(inodes_per_block - 1);
3853 /* Is the inode bitmap in cache? */
3854 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3855 if (unlikely(!bitmap_bh))
3859 * If the inode bitmap isn't in cache then the
3860 * optimisation may end up performing two reads instead
3861 * of one, so skip it.
3863 if (!buffer_uptodate(bitmap_bh)) {
3867 for (i = start; i < start + inodes_per_block; i++) {
3868 if (i == inode_offset)
3870 if (ext4_test_bit(i, bitmap_bh->b_data))
3874 if (i == start + inodes_per_block) {
3875 /* all other inodes are free, so skip I/O */
3876 memset(bh->b_data, 0, bh->b_size);
3877 set_buffer_uptodate(bh);
3885 * If we need to do any I/O, try to pre-readahead extra
3886 * blocks from the inode table.
3888 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3889 ext4_fsblk_t b, end, table;
3891 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
3893 table = ext4_inode_table(sb, gdp);
3894 /* s_inode_readahead_blks is always a power of 2 */
3895 b = block & ~((ext4_fsblk_t) ra_blks - 1);
3899 num = EXT4_INODES_PER_GROUP(sb);
3900 if (ext4_has_group_desc_csum(sb))
3901 num -= ext4_itable_unused_count(sb, gdp);
3902 table += num / inodes_per_block;
3906 sb_breadahead(sb, b++);
3910 * There are other valid inodes in the buffer, this inode
3911 * has in-inode xattrs, or we don't have this inode in memory.
3912 * Read the block from disk.
3914 trace_ext4_load_inode(inode);
3916 bh->b_end_io = end_buffer_read_sync;
3917 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3919 if (!buffer_uptodate(bh)) {
3920 EXT4_ERROR_INODE_BLOCK(inode, block,
3921 "unable to read itable block");
3931 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3933 /* We have all inode data except xattrs in memory here. */
3934 return __ext4_get_inode_loc(inode, iloc,
3935 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3938 void ext4_set_inode_flags(struct inode *inode)
3940 unsigned int flags = EXT4_I(inode)->i_flags;
3941 unsigned int new_fl = 0;
3943 if (flags & EXT4_SYNC_FL)
3945 if (flags & EXT4_APPEND_FL)
3947 if (flags & EXT4_IMMUTABLE_FL)
3948 new_fl |= S_IMMUTABLE;
3949 if (flags & EXT4_NOATIME_FL)
3950 new_fl |= S_NOATIME;
3951 if (flags & EXT4_DIRSYNC_FL)
3952 new_fl |= S_DIRSYNC;
3953 if (test_opt(inode->i_sb, DAX))
3955 inode_set_flags(inode, new_fl,
3956 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
3959 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3960 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3962 unsigned int vfs_fl;
3963 unsigned long old_fl, new_fl;
3966 vfs_fl = ei->vfs_inode.i_flags;
3967 old_fl = ei->i_flags;
3968 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3969 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3971 if (vfs_fl & S_SYNC)
3972 new_fl |= EXT4_SYNC_FL;
3973 if (vfs_fl & S_APPEND)
3974 new_fl |= EXT4_APPEND_FL;
3975 if (vfs_fl & S_IMMUTABLE)
3976 new_fl |= EXT4_IMMUTABLE_FL;
3977 if (vfs_fl & S_NOATIME)
3978 new_fl |= EXT4_NOATIME_FL;
3979 if (vfs_fl & S_DIRSYNC)
3980 new_fl |= EXT4_DIRSYNC_FL;
3981 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3984 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3985 struct ext4_inode_info *ei)
3988 struct inode *inode = &(ei->vfs_inode);
3989 struct super_block *sb = inode->i_sb;
3991 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3992 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3993 /* we are using combined 48 bit field */
3994 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3995 le32_to_cpu(raw_inode->i_blocks_lo);
3996 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3997 /* i_blocks represent file system block size */
3998 return i_blocks << (inode->i_blkbits - 9);
4003 return le32_to_cpu(raw_inode->i_blocks_lo);
4007 static inline void ext4_iget_extra_inode(struct inode *inode,
4008 struct ext4_inode *raw_inode,
4009 struct ext4_inode_info *ei)
4011 __le32 *magic = (void *)raw_inode +
4012 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4013 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4014 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4015 ext4_find_inline_data_nolock(inode);
4017 EXT4_I(inode)->i_inline_off = 0;
4020 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4022 struct ext4_iloc iloc;
4023 struct ext4_inode *raw_inode;
4024 struct ext4_inode_info *ei;
4025 struct inode *inode;
4026 journal_t *journal = EXT4_SB(sb)->s_journal;
4032 inode = iget_locked(sb, ino);
4034 return ERR_PTR(-ENOMEM);
4035 if (!(inode->i_state & I_NEW))
4041 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4044 raw_inode = ext4_raw_inode(&iloc);
4046 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4047 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4048 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4049 EXT4_INODE_SIZE(inode->i_sb)) {
4050 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4051 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4052 EXT4_INODE_SIZE(inode->i_sb));
4057 ei->i_extra_isize = 0;
4059 /* Precompute checksum seed for inode metadata */
4060 if (ext4_has_metadata_csum(sb)) {
4061 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4063 __le32 inum = cpu_to_le32(inode->i_ino);
4064 __le32 gen = raw_inode->i_generation;
4065 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4067 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4071 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4072 EXT4_ERROR_INODE(inode, "checksum invalid");
4077 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4078 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4079 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4080 if (!(test_opt(inode->i_sb, NO_UID32))) {
4081 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4082 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4084 i_uid_write(inode, i_uid);
4085 i_gid_write(inode, i_gid);
4086 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4088 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4089 ei->i_inline_off = 0;
4090 ei->i_dir_start_lookup = 0;
4091 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4092 /* We now have enough fields to check if the inode was active or not.
4093 * This is needed because nfsd might try to access dead inodes
4094 * the test is that same one that e2fsck uses
4095 * NeilBrown 1999oct15
4097 if (inode->i_nlink == 0) {
4098 if ((inode->i_mode == 0 ||
4099 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4100 ino != EXT4_BOOT_LOADER_INO) {
4101 /* this inode is deleted */
4105 /* The only unlinked inodes we let through here have
4106 * valid i_mode and are being read by the orphan
4107 * recovery code: that's fine, we're about to complete
4108 * the process of deleting those.
4109 * OR it is the EXT4_BOOT_LOADER_INO which is
4110 * not initialized on a new filesystem. */
4112 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4113 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4114 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4115 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4117 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4118 inode->i_size = ext4_isize(raw_inode);
4119 ei->i_disksize = inode->i_size;
4121 ei->i_reserved_quota = 0;
4123 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4124 ei->i_block_group = iloc.block_group;
4125 ei->i_last_alloc_group = ~0;
4127 * NOTE! The in-memory inode i_data array is in little-endian order
4128 * even on big-endian machines: we do NOT byteswap the block numbers!
4130 for (block = 0; block < EXT4_N_BLOCKS; block++)
4131 ei->i_data[block] = raw_inode->i_block[block];
4132 INIT_LIST_HEAD(&ei->i_orphan);
4135 * Set transaction id's of transactions that have to be committed
4136 * to finish f[data]sync. We set them to currently running transaction
4137 * as we cannot be sure that the inode or some of its metadata isn't
4138 * part of the transaction - the inode could have been reclaimed and
4139 * now it is reread from disk.
4142 transaction_t *transaction;
4145 read_lock(&journal->j_state_lock);
4146 if (journal->j_running_transaction)
4147 transaction = journal->j_running_transaction;
4149 transaction = journal->j_committing_transaction;
4151 tid = transaction->t_tid;
4153 tid = journal->j_commit_sequence;
4154 read_unlock(&journal->j_state_lock);
4155 ei->i_sync_tid = tid;
4156 ei->i_datasync_tid = tid;
4159 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4160 if (ei->i_extra_isize == 0) {
4161 /* The extra space is currently unused. Use it. */
4162 ei->i_extra_isize = sizeof(struct ext4_inode) -
4163 EXT4_GOOD_OLD_INODE_SIZE;
4165 ext4_iget_extra_inode(inode, raw_inode, ei);
4169 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4170 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4171 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4172 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4174 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4175 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4176 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4177 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4179 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4184 if (ei->i_file_acl &&
4185 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4186 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4190 } else if (!ext4_has_inline_data(inode)) {
4191 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4192 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4193 (S_ISLNK(inode->i_mode) &&
4194 !ext4_inode_is_fast_symlink(inode))))
4195 /* Validate extent which is part of inode */
4196 ret = ext4_ext_check_inode(inode);
4197 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4198 (S_ISLNK(inode->i_mode) &&
4199 !ext4_inode_is_fast_symlink(inode))) {
4200 /* Validate block references which are part of inode */
4201 ret = ext4_ind_check_inode(inode);
4207 if (S_ISREG(inode->i_mode)) {
4208 inode->i_op = &ext4_file_inode_operations;
4209 if (test_opt(inode->i_sb, DAX))
4210 inode->i_fop = &ext4_dax_file_operations;
4212 inode->i_fop = &ext4_file_operations;
4213 ext4_set_aops(inode);
4214 } else if (S_ISDIR(inode->i_mode)) {
4215 inode->i_op = &ext4_dir_inode_operations;
4216 inode->i_fop = &ext4_dir_operations;
4217 } else if (S_ISLNK(inode->i_mode)) {
4218 if (ext4_inode_is_fast_symlink(inode) &&
4219 !ext4_encrypted_inode(inode)) {
4220 inode->i_op = &ext4_fast_symlink_inode_operations;
4221 nd_terminate_link(ei->i_data, inode->i_size,
4222 sizeof(ei->i_data) - 1);
4224 inode->i_op = &ext4_symlink_inode_operations;
4225 ext4_set_aops(inode);
4227 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4228 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4229 inode->i_op = &ext4_special_inode_operations;
4230 if (raw_inode->i_block[0])
4231 init_special_inode(inode, inode->i_mode,
4232 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4234 init_special_inode(inode, inode->i_mode,
4235 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4236 } else if (ino == EXT4_BOOT_LOADER_INO) {
4237 make_bad_inode(inode);
4240 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4244 ext4_set_inode_flags(inode);
4245 unlock_new_inode(inode);
4251 return ERR_PTR(ret);
4254 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4256 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4257 return ERR_PTR(-EIO);
4258 return ext4_iget(sb, ino);
4261 static int ext4_inode_blocks_set(handle_t *handle,
4262 struct ext4_inode *raw_inode,
4263 struct ext4_inode_info *ei)
4265 struct inode *inode = &(ei->vfs_inode);
4266 u64 i_blocks = inode->i_blocks;
4267 struct super_block *sb = inode->i_sb;
4269 if (i_blocks <= ~0U) {
4271 * i_blocks can be represented in a 32 bit variable
4272 * as multiple of 512 bytes
4274 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4275 raw_inode->i_blocks_high = 0;
4276 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4279 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4282 if (i_blocks <= 0xffffffffffffULL) {
4284 * i_blocks can be represented in a 48 bit variable
4285 * as multiple of 512 bytes
4287 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4288 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4289 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4291 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4292 /* i_block is stored in file system block size */
4293 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4294 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4295 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4300 struct other_inode {
4301 unsigned long orig_ino;
4302 struct ext4_inode *raw_inode;
4305 static int other_inode_match(struct inode * inode, unsigned long ino,
4308 struct other_inode *oi = (struct other_inode *) data;
4310 if ((inode->i_ino != ino) ||
4311 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4312 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4313 ((inode->i_state & I_DIRTY_TIME) == 0))
4315 spin_lock(&inode->i_lock);
4316 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4317 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4318 (inode->i_state & I_DIRTY_TIME)) {
4319 struct ext4_inode_info *ei = EXT4_I(inode);
4321 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4322 spin_unlock(&inode->i_lock);
4324 spin_lock(&ei->i_raw_lock);
4325 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4326 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4327 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4328 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4329 spin_unlock(&ei->i_raw_lock);
4330 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4333 spin_unlock(&inode->i_lock);
4338 * Opportunistically update the other time fields for other inodes in
4339 * the same inode table block.
4341 static void ext4_update_other_inodes_time(struct super_block *sb,
4342 unsigned long orig_ino, char *buf)
4344 struct other_inode oi;
4346 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4347 int inode_size = EXT4_INODE_SIZE(sb);
4349 oi.orig_ino = orig_ino;
4350 ino = orig_ino & ~(inodes_per_block - 1);
4351 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4352 if (ino == orig_ino)
4354 oi.raw_inode = (struct ext4_inode *) buf;
4355 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4360 * Post the struct inode info into an on-disk inode location in the
4361 * buffer-cache. This gobbles the caller's reference to the
4362 * buffer_head in the inode location struct.
4364 * The caller must have write access to iloc->bh.
4366 static int ext4_do_update_inode(handle_t *handle,
4367 struct inode *inode,
4368 struct ext4_iloc *iloc)
4370 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4371 struct ext4_inode_info *ei = EXT4_I(inode);
4372 struct buffer_head *bh = iloc->bh;
4373 struct super_block *sb = inode->i_sb;
4374 int err = 0, rc, block;
4375 int need_datasync = 0, set_large_file = 0;
4379 spin_lock(&ei->i_raw_lock);
4381 /* For fields not tracked in the in-memory inode,
4382 * initialise them to zero for new inodes. */
4383 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4384 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4386 ext4_get_inode_flags(ei);
4387 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4388 i_uid = i_uid_read(inode);
4389 i_gid = i_gid_read(inode);
4390 if (!(test_opt(inode->i_sb, NO_UID32))) {
4391 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4392 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4394 * Fix up interoperability with old kernels. Otherwise, old inodes get
4395 * re-used with the upper 16 bits of the uid/gid intact
4398 raw_inode->i_uid_high =
4399 cpu_to_le16(high_16_bits(i_uid));
4400 raw_inode->i_gid_high =
4401 cpu_to_le16(high_16_bits(i_gid));
4403 raw_inode->i_uid_high = 0;
4404 raw_inode->i_gid_high = 0;
4407 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4408 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4409 raw_inode->i_uid_high = 0;
4410 raw_inode->i_gid_high = 0;
4412 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4414 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4415 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4416 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4417 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4419 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4421 spin_unlock(&ei->i_raw_lock);
4424 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4425 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4426 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4427 raw_inode->i_file_acl_high =
4428 cpu_to_le16(ei->i_file_acl >> 32);
4429 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4430 if (ei->i_disksize != ext4_isize(raw_inode)) {
4431 ext4_isize_set(raw_inode, ei->i_disksize);
4434 if (ei->i_disksize > 0x7fffffffULL) {
4435 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4436 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4437 EXT4_SB(sb)->s_es->s_rev_level ==
4438 cpu_to_le32(EXT4_GOOD_OLD_REV))
4441 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4442 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4443 if (old_valid_dev(inode->i_rdev)) {
4444 raw_inode->i_block[0] =
4445 cpu_to_le32(old_encode_dev(inode->i_rdev));
4446 raw_inode->i_block[1] = 0;
4448 raw_inode->i_block[0] = 0;
4449 raw_inode->i_block[1] =
4450 cpu_to_le32(new_encode_dev(inode->i_rdev));
4451 raw_inode->i_block[2] = 0;
4453 } else if (!ext4_has_inline_data(inode)) {
4454 for (block = 0; block < EXT4_N_BLOCKS; block++)
4455 raw_inode->i_block[block] = ei->i_data[block];
4458 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4459 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4460 if (ei->i_extra_isize) {
4461 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4462 raw_inode->i_version_hi =
4463 cpu_to_le32(inode->i_version >> 32);
4464 raw_inode->i_extra_isize =
4465 cpu_to_le16(ei->i_extra_isize);
4468 ext4_inode_csum_set(inode, raw_inode, ei);
4469 spin_unlock(&ei->i_raw_lock);
4470 if (inode->i_sb->s_flags & MS_LAZYTIME)
4471 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4474 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4475 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4478 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4479 if (set_large_file) {
4480 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4481 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4484 ext4_update_dynamic_rev(sb);
4485 EXT4_SET_RO_COMPAT_FEATURE(sb,
4486 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4487 ext4_handle_sync(handle);
4488 err = ext4_handle_dirty_super(handle, sb);
4490 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4493 ext4_std_error(inode->i_sb, err);
4498 * ext4_write_inode()
4500 * We are called from a few places:
4502 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4503 * Here, there will be no transaction running. We wait for any running
4504 * transaction to commit.
4506 * - Within flush work (sys_sync(), kupdate and such).
4507 * We wait on commit, if told to.
4509 * - Within iput_final() -> write_inode_now()
4510 * We wait on commit, if told to.
4512 * In all cases it is actually safe for us to return without doing anything,
4513 * because the inode has been copied into a raw inode buffer in
4514 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4517 * Note that we are absolutely dependent upon all inode dirtiers doing the
4518 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4519 * which we are interested.
4521 * It would be a bug for them to not do this. The code:
4523 * mark_inode_dirty(inode)
4525 * inode->i_size = expr;
4527 * is in error because write_inode() could occur while `stuff()' is running,
4528 * and the new i_size will be lost. Plus the inode will no longer be on the
4529 * superblock's dirty inode list.
4531 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4535 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4538 if (EXT4_SB(inode->i_sb)->s_journal) {
4539 if (ext4_journal_current_handle()) {
4540 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4546 * No need to force transaction in WB_SYNC_NONE mode. Also
4547 * ext4_sync_fs() will force the commit after everything is
4550 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4553 err = ext4_force_commit(inode->i_sb);
4555 struct ext4_iloc iloc;
4557 err = __ext4_get_inode_loc(inode, &iloc, 0);
4561 * sync(2) will flush the whole buffer cache. No need to do
4562 * it here separately for each inode.
4564 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4565 sync_dirty_buffer(iloc.bh);
4566 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4567 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4568 "IO error syncing inode");
4577 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4578 * buffers that are attached to a page stradding i_size and are undergoing
4579 * commit. In that case we have to wait for commit to finish and try again.
4581 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4585 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4586 tid_t commit_tid = 0;
4589 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4591 * All buffers in the last page remain valid? Then there's nothing to
4592 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4595 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4598 page = find_lock_page(inode->i_mapping,
4599 inode->i_size >> PAGE_CACHE_SHIFT);
4602 ret = __ext4_journalled_invalidatepage(page, offset,
4603 PAGE_CACHE_SIZE - offset);
4605 page_cache_release(page);
4609 read_lock(&journal->j_state_lock);
4610 if (journal->j_committing_transaction)
4611 commit_tid = journal->j_committing_transaction->t_tid;
4612 read_unlock(&journal->j_state_lock);
4614 jbd2_log_wait_commit(journal, commit_tid);
4621 * Called from notify_change.
4623 * We want to trap VFS attempts to truncate the file as soon as
4624 * possible. In particular, we want to make sure that when the VFS
4625 * shrinks i_size, we put the inode on the orphan list and modify
4626 * i_disksize immediately, so that during the subsequent flushing of
4627 * dirty pages and freeing of disk blocks, we can guarantee that any
4628 * commit will leave the blocks being flushed in an unused state on
4629 * disk. (On recovery, the inode will get truncated and the blocks will
4630 * be freed, so we have a strong guarantee that no future commit will
4631 * leave these blocks visible to the user.)
4633 * Another thing we have to assure is that if we are in ordered mode
4634 * and inode is still attached to the committing transaction, we must
4635 * we start writeout of all the dirty pages which are being truncated.
4636 * This way we are sure that all the data written in the previous
4637 * transaction are already on disk (truncate waits for pages under
4640 * Called with inode->i_mutex down.
4642 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4644 struct inode *inode = dentry->d_inode;
4647 const unsigned int ia_valid = attr->ia_valid;
4649 error = inode_change_ok(inode, attr);
4653 if (is_quota_modification(inode, attr))
4654 dquot_initialize(inode);
4655 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4656 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4659 /* (user+group)*(old+new) structure, inode write (sb,
4660 * inode block, ? - but truncate inode update has it) */
4661 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4662 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4663 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4664 if (IS_ERR(handle)) {
4665 error = PTR_ERR(handle);
4668 error = dquot_transfer(inode, attr);
4670 ext4_journal_stop(handle);
4673 /* Update corresponding info in inode so that everything is in
4674 * one transaction */
4675 if (attr->ia_valid & ATTR_UID)
4676 inode->i_uid = attr->ia_uid;
4677 if (attr->ia_valid & ATTR_GID)
4678 inode->i_gid = attr->ia_gid;
4679 error = ext4_mark_inode_dirty(handle, inode);
4680 ext4_journal_stop(handle);
4683 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4686 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4687 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4689 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4693 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4694 inode_inc_iversion(inode);
4696 if (S_ISREG(inode->i_mode) &&
4697 (attr->ia_size < inode->i_size)) {
4698 if (ext4_should_order_data(inode)) {
4699 error = ext4_begin_ordered_truncate(inode,
4704 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4705 if (IS_ERR(handle)) {
4706 error = PTR_ERR(handle);
4709 if (ext4_handle_valid(handle)) {
4710 error = ext4_orphan_add(handle, inode);
4713 down_write(&EXT4_I(inode)->i_data_sem);
4714 EXT4_I(inode)->i_disksize = attr->ia_size;
4715 rc = ext4_mark_inode_dirty(handle, inode);
4719 * We have to update i_size under i_data_sem together
4720 * with i_disksize to avoid races with writeback code
4721 * running ext4_wb_update_i_disksize().
4724 i_size_write(inode, attr->ia_size);
4725 up_write(&EXT4_I(inode)->i_data_sem);
4726 ext4_journal_stop(handle);
4728 ext4_orphan_del(NULL, inode);
4732 loff_t oldsize = inode->i_size;
4734 i_size_write(inode, attr->ia_size);
4735 pagecache_isize_extended(inode, oldsize, inode->i_size);
4739 * Blocks are going to be removed from the inode. Wait
4740 * for dio in flight. Temporarily disable
4741 * dioread_nolock to prevent livelock.
4744 if (!ext4_should_journal_data(inode)) {
4745 ext4_inode_block_unlocked_dio(inode);
4746 inode_dio_wait(inode);
4747 ext4_inode_resume_unlocked_dio(inode);
4749 ext4_wait_for_tail_page_commit(inode);
4752 * Truncate pagecache after we've waited for commit
4753 * in data=journal mode to make pages freeable.
4755 truncate_pagecache(inode, inode->i_size);
4758 * We want to call ext4_truncate() even if attr->ia_size ==
4759 * inode->i_size for cases like truncation of fallocated space
4761 if (attr->ia_valid & ATTR_SIZE)
4762 ext4_truncate(inode);
4765 setattr_copy(inode, attr);
4766 mark_inode_dirty(inode);
4770 * If the call to ext4_truncate failed to get a transaction handle at
4771 * all, we need to clean up the in-core orphan list manually.
4773 if (orphan && inode->i_nlink)
4774 ext4_orphan_del(NULL, inode);
4776 if (!rc && (ia_valid & ATTR_MODE))
4777 rc = posix_acl_chmod(inode, inode->i_mode);
4780 ext4_std_error(inode->i_sb, error);
4786 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4789 struct inode *inode;
4790 unsigned long long delalloc_blocks;
4792 inode = dentry->d_inode;
4793 generic_fillattr(inode, stat);
4796 * If there is inline data in the inode, the inode will normally not
4797 * have data blocks allocated (it may have an external xattr block).
4798 * Report at least one sector for such files, so tools like tar, rsync,
4799 * others doen't incorrectly think the file is completely sparse.
4801 if (unlikely(ext4_has_inline_data(inode)))
4802 stat->blocks += (stat->size + 511) >> 9;
4805 * We can't update i_blocks if the block allocation is delayed
4806 * otherwise in the case of system crash before the real block
4807 * allocation is done, we will have i_blocks inconsistent with
4808 * on-disk file blocks.
4809 * We always keep i_blocks updated together with real
4810 * allocation. But to not confuse with user, stat
4811 * will return the blocks that include the delayed allocation
4812 * blocks for this file.
4814 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4815 EXT4_I(inode)->i_reserved_data_blocks);
4816 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
4820 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4823 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4824 return ext4_ind_trans_blocks(inode, lblocks);
4825 return ext4_ext_index_trans_blocks(inode, pextents);
4829 * Account for index blocks, block groups bitmaps and block group
4830 * descriptor blocks if modify datablocks and index blocks
4831 * worse case, the indexs blocks spread over different block groups
4833 * If datablocks are discontiguous, they are possible to spread over
4834 * different block groups too. If they are contiguous, with flexbg,
4835 * they could still across block group boundary.
4837 * Also account for superblock, inode, quota and xattr blocks
4839 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4842 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4848 * How many index blocks need to touch to map @lblocks logical blocks
4849 * to @pextents physical extents?
4851 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
4856 * Now let's see how many group bitmaps and group descriptors need
4859 groups = idxblocks + pextents;
4861 if (groups > ngroups)
4863 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4864 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4866 /* bitmaps and block group descriptor blocks */
4867 ret += groups + gdpblocks;
4869 /* Blocks for super block, inode, quota and xattr blocks */
4870 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4876 * Calculate the total number of credits to reserve to fit
4877 * the modification of a single pages into a single transaction,
4878 * which may include multiple chunks of block allocations.
4880 * This could be called via ext4_write_begin()
4882 * We need to consider the worse case, when
4883 * one new block per extent.
4885 int ext4_writepage_trans_blocks(struct inode *inode)
4887 int bpp = ext4_journal_blocks_per_page(inode);
4890 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
4892 /* Account for data blocks for journalled mode */
4893 if (ext4_should_journal_data(inode))
4899 * Calculate the journal credits for a chunk of data modification.
4901 * This is called from DIO, fallocate or whoever calling
4902 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4904 * journal buffers for data blocks are not included here, as DIO
4905 * and fallocate do no need to journal data buffers.
4907 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4909 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4913 * The caller must have previously called ext4_reserve_inode_write().
4914 * Give this, we know that the caller already has write access to iloc->bh.
4916 int ext4_mark_iloc_dirty(handle_t *handle,
4917 struct inode *inode, struct ext4_iloc *iloc)
4921 if (IS_I_VERSION(inode))
4922 inode_inc_iversion(inode);
4924 /* the do_update_inode consumes one bh->b_count */
4927 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4928 err = ext4_do_update_inode(handle, inode, iloc);
4934 * On success, We end up with an outstanding reference count against
4935 * iloc->bh. This _must_ be cleaned up later.
4939 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4940 struct ext4_iloc *iloc)
4944 err = ext4_get_inode_loc(inode, iloc);
4946 BUFFER_TRACE(iloc->bh, "get_write_access");
4947 err = ext4_journal_get_write_access(handle, iloc->bh);
4953 ext4_std_error(inode->i_sb, err);
4958 * Expand an inode by new_extra_isize bytes.
4959 * Returns 0 on success or negative error number on failure.
4961 static int ext4_expand_extra_isize(struct inode *inode,
4962 unsigned int new_extra_isize,
4963 struct ext4_iloc iloc,
4966 struct ext4_inode *raw_inode;
4967 struct ext4_xattr_ibody_header *header;
4969 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4972 raw_inode = ext4_raw_inode(&iloc);
4974 header = IHDR(inode, raw_inode);
4976 /* No extended attributes present */
4977 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4978 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4979 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4981 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4985 /* try to expand with EAs present */
4986 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4991 * What we do here is to mark the in-core inode as clean with respect to inode
4992 * dirtiness (it may still be data-dirty).
4993 * This means that the in-core inode may be reaped by prune_icache
4994 * without having to perform any I/O. This is a very good thing,
4995 * because *any* task may call prune_icache - even ones which
4996 * have a transaction open against a different journal.
4998 * Is this cheating? Not really. Sure, we haven't written the
4999 * inode out, but prune_icache isn't a user-visible syncing function.
5000 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5001 * we start and wait on commits.
5003 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5005 struct ext4_iloc iloc;
5006 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5007 static unsigned int mnt_count;
5011 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5012 err = ext4_reserve_inode_write(handle, inode, &iloc);
5013 if (ext4_handle_valid(handle) &&
5014 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5015 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5017 * We need extra buffer credits since we may write into EA block
5018 * with this same handle. If journal_extend fails, then it will
5019 * only result in a minor loss of functionality for that inode.
5020 * If this is felt to be critical, then e2fsck should be run to
5021 * force a large enough s_min_extra_isize.
5023 if ((jbd2_journal_extend(handle,
5024 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5025 ret = ext4_expand_extra_isize(inode,
5026 sbi->s_want_extra_isize,
5029 ext4_set_inode_state(inode,
5030 EXT4_STATE_NO_EXPAND);
5032 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5033 ext4_warning(inode->i_sb,
5034 "Unable to expand inode %lu. Delete"
5035 " some EAs or run e2fsck.",
5038 le16_to_cpu(sbi->s_es->s_mnt_count);
5044 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5049 * ext4_dirty_inode() is called from __mark_inode_dirty()
5051 * We're really interested in the case where a file is being extended.
5052 * i_size has been changed by generic_commit_write() and we thus need
5053 * to include the updated inode in the current transaction.
5055 * Also, dquot_alloc_block() will always dirty the inode when blocks
5056 * are allocated to the file.
5058 * If the inode is marked synchronous, we don't honour that here - doing
5059 * so would cause a commit on atime updates, which we don't bother doing.
5060 * We handle synchronous inodes at the highest possible level.
5062 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5063 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5064 * to copy into the on-disk inode structure are the timestamp files.
5066 void ext4_dirty_inode(struct inode *inode, int flags)
5070 if (flags == I_DIRTY_TIME)
5072 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5076 ext4_mark_inode_dirty(handle, inode);
5078 ext4_journal_stop(handle);
5085 * Bind an inode's backing buffer_head into this transaction, to prevent
5086 * it from being flushed to disk early. Unlike
5087 * ext4_reserve_inode_write, this leaves behind no bh reference and
5088 * returns no iloc structure, so the caller needs to repeat the iloc
5089 * lookup to mark the inode dirty later.
5091 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5093 struct ext4_iloc iloc;
5097 err = ext4_get_inode_loc(inode, &iloc);
5099 BUFFER_TRACE(iloc.bh, "get_write_access");
5100 err = jbd2_journal_get_write_access(handle, iloc.bh);
5102 err = ext4_handle_dirty_metadata(handle,
5108 ext4_std_error(inode->i_sb, err);
5113 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5120 * We have to be very careful here: changing a data block's
5121 * journaling status dynamically is dangerous. If we write a
5122 * data block to the journal, change the status and then delete
5123 * that block, we risk forgetting to revoke the old log record
5124 * from the journal and so a subsequent replay can corrupt data.
5125 * So, first we make sure that the journal is empty and that
5126 * nobody is changing anything.
5129 journal = EXT4_JOURNAL(inode);
5132 if (is_journal_aborted(journal))
5134 /* We have to allocate physical blocks for delalloc blocks
5135 * before flushing journal. otherwise delalloc blocks can not
5136 * be allocated any more. even more truncate on delalloc blocks
5137 * could trigger BUG by flushing delalloc blocks in journal.
5138 * There is no delalloc block in non-journal data mode.
5140 if (val && test_opt(inode->i_sb, DELALLOC)) {
5141 err = ext4_alloc_da_blocks(inode);
5146 /* Wait for all existing dio workers */
5147 ext4_inode_block_unlocked_dio(inode);
5148 inode_dio_wait(inode);
5150 jbd2_journal_lock_updates(journal);
5153 * OK, there are no updates running now, and all cached data is
5154 * synced to disk. We are now in a completely consistent state
5155 * which doesn't have anything in the journal, and we know that
5156 * no filesystem updates are running, so it is safe to modify
5157 * the inode's in-core data-journaling state flag now.
5161 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5163 err = jbd2_journal_flush(journal);
5165 jbd2_journal_unlock_updates(journal);
5166 ext4_inode_resume_unlocked_dio(inode);
5169 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5171 ext4_set_aops(inode);
5173 jbd2_journal_unlock_updates(journal);
5174 ext4_inode_resume_unlocked_dio(inode);
5176 /* Finally we can mark the inode as dirty. */
5178 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5180 return PTR_ERR(handle);
5182 err = ext4_mark_inode_dirty(handle, inode);
5183 ext4_handle_sync(handle);
5184 ext4_journal_stop(handle);
5185 ext4_std_error(inode->i_sb, err);
5190 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5192 return !buffer_mapped(bh);
5195 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5197 struct page *page = vmf->page;
5201 struct file *file = vma->vm_file;
5202 struct inode *inode = file_inode(file);
5203 struct address_space *mapping = inode->i_mapping;
5205 get_block_t *get_block;
5208 sb_start_pagefault(inode->i_sb);
5209 file_update_time(vma->vm_file);
5210 /* Delalloc case is easy... */
5211 if (test_opt(inode->i_sb, DELALLOC) &&
5212 !ext4_should_journal_data(inode) &&
5213 !ext4_nonda_switch(inode->i_sb)) {
5215 ret = __block_page_mkwrite(vma, vmf,
5216 ext4_da_get_block_prep);
5217 } while (ret == -ENOSPC &&
5218 ext4_should_retry_alloc(inode->i_sb, &retries));
5223 size = i_size_read(inode);
5224 /* Page got truncated from under us? */
5225 if (page->mapping != mapping || page_offset(page) > size) {
5227 ret = VM_FAULT_NOPAGE;
5231 if (page->index == size >> PAGE_CACHE_SHIFT)
5232 len = size & ~PAGE_CACHE_MASK;
5234 len = PAGE_CACHE_SIZE;
5236 * Return if we have all the buffers mapped. This avoids the need to do
5237 * journal_start/journal_stop which can block and take a long time
5239 if (page_has_buffers(page)) {
5240 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5242 ext4_bh_unmapped)) {
5243 /* Wait so that we don't change page under IO */
5244 wait_for_stable_page(page);
5245 ret = VM_FAULT_LOCKED;
5250 /* OK, we need to fill the hole... */
5251 if (ext4_should_dioread_nolock(inode))
5252 get_block = ext4_get_block_write;
5254 get_block = ext4_get_block;
5256 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5257 ext4_writepage_trans_blocks(inode));
5258 if (IS_ERR(handle)) {
5259 ret = VM_FAULT_SIGBUS;
5262 ret = __block_page_mkwrite(vma, vmf, get_block);
5263 if (!ret && ext4_should_journal_data(inode)) {
5264 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5265 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5267 ret = VM_FAULT_SIGBUS;
5268 ext4_journal_stop(handle);
5271 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5273 ext4_journal_stop(handle);
5274 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5277 ret = block_page_mkwrite_return(ret);
5279 sb_end_pagefault(inode->i_sb);