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 static 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);
889 static int ext4_write_begin(struct file *file, struct address_space *mapping,
890 loff_t pos, unsigned len, unsigned flags,
891 struct page **pagep, void **fsdata)
893 struct inode *inode = mapping->host;
894 int ret, needed_blocks;
901 trace_ext4_write_begin(inode, pos, len, flags);
903 * Reserve one block more for addition to orphan list in case
904 * we allocate blocks but write fails for some reason
906 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
907 index = pos >> PAGE_CACHE_SHIFT;
908 from = pos & (PAGE_CACHE_SIZE - 1);
911 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
912 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
921 * grab_cache_page_write_begin() can take a long time if the
922 * system is thrashing due to memory pressure, or if the page
923 * is being written back. So grab it first before we start
924 * the transaction handle. This also allows us to allocate
925 * the page (if needed) without using GFP_NOFS.
928 page = grab_cache_page_write_begin(mapping, index, flags);
934 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
935 if (IS_ERR(handle)) {
936 page_cache_release(page);
937 return PTR_ERR(handle);
941 if (page->mapping != mapping) {
942 /* The page got truncated from under us */
944 page_cache_release(page);
945 ext4_journal_stop(handle);
948 /* In case writeback began while the page was unlocked */
949 wait_for_stable_page(page);
951 if (ext4_should_dioread_nolock(inode))
952 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
954 ret = __block_write_begin(page, pos, len, ext4_get_block);
956 if (!ret && ext4_should_journal_data(inode)) {
957 ret = ext4_walk_page_buffers(handle, page_buffers(page),
959 do_journal_get_write_access);
965 * __block_write_begin may have instantiated a few blocks
966 * outside i_size. Trim these off again. Don't need
967 * i_size_read because we hold i_mutex.
969 * Add inode to orphan list in case we crash before
972 if (pos + len > inode->i_size && ext4_can_truncate(inode))
973 ext4_orphan_add(handle, inode);
975 ext4_journal_stop(handle);
976 if (pos + len > inode->i_size) {
977 ext4_truncate_failed_write(inode);
979 * If truncate failed early the inode might
980 * still be on the orphan list; we need to
981 * make sure the inode is removed from the
982 * orphan list in that case.
985 ext4_orphan_del(NULL, inode);
988 if (ret == -ENOSPC &&
989 ext4_should_retry_alloc(inode->i_sb, &retries))
991 page_cache_release(page);
998 /* For write_end() in data=journal mode */
999 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1002 if (!buffer_mapped(bh) || buffer_freed(bh))
1004 set_buffer_uptodate(bh);
1005 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1006 clear_buffer_meta(bh);
1007 clear_buffer_prio(bh);
1012 * We need to pick up the new inode size which generic_commit_write gave us
1013 * `file' can be NULL - eg, when called from page_symlink().
1015 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1016 * buffers are managed internally.
1018 static int ext4_write_end(struct file *file,
1019 struct address_space *mapping,
1020 loff_t pos, unsigned len, unsigned copied,
1021 struct page *page, void *fsdata)
1023 handle_t *handle = ext4_journal_current_handle();
1024 struct inode *inode = mapping->host;
1025 loff_t old_size = inode->i_size;
1027 int i_size_changed = 0;
1029 trace_ext4_write_end(inode, pos, len, copied);
1030 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1031 ret = ext4_jbd2_file_inode(handle, inode);
1034 page_cache_release(page);
1039 if (ext4_has_inline_data(inode)) {
1040 ret = ext4_write_inline_data_end(inode, pos, len,
1046 copied = block_write_end(file, mapping, pos,
1047 len, copied, page, fsdata);
1049 * it's important to update i_size while still holding page lock:
1050 * page writeout could otherwise come in and zero beyond i_size.
1052 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1054 page_cache_release(page);
1057 pagecache_isize_extended(inode, old_size, pos);
1059 * Don't mark the inode dirty under page lock. First, it unnecessarily
1060 * makes the holding time of page lock longer. Second, it forces lock
1061 * ordering of page lock and transaction start for journaling
1065 ext4_mark_inode_dirty(handle, inode);
1067 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1068 /* if we have allocated more blocks and copied
1069 * less. We will have blocks allocated outside
1070 * inode->i_size. So truncate them
1072 ext4_orphan_add(handle, inode);
1074 ret2 = ext4_journal_stop(handle);
1078 if (pos + len > inode->i_size) {
1079 ext4_truncate_failed_write(inode);
1081 * If truncate failed early the inode might still be
1082 * on the orphan list; we need to make sure the inode
1083 * is removed from the orphan list in that case.
1086 ext4_orphan_del(NULL, inode);
1089 return ret ? ret : copied;
1092 static int ext4_journalled_write_end(struct file *file,
1093 struct address_space *mapping,
1094 loff_t pos, unsigned len, unsigned copied,
1095 struct page *page, void *fsdata)
1097 handle_t *handle = ext4_journal_current_handle();
1098 struct inode *inode = mapping->host;
1099 loff_t old_size = inode->i_size;
1103 int size_changed = 0;
1105 trace_ext4_journalled_write_end(inode, pos, len, copied);
1106 from = pos & (PAGE_CACHE_SIZE - 1);
1109 BUG_ON(!ext4_handle_valid(handle));
1111 if (ext4_has_inline_data(inode))
1112 copied = ext4_write_inline_data_end(inode, pos, len,
1116 if (!PageUptodate(page))
1118 page_zero_new_buffers(page, from+copied, to);
1121 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1122 to, &partial, write_end_fn);
1124 SetPageUptodate(page);
1126 size_changed = ext4_update_inode_size(inode, pos + copied);
1127 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1128 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1130 page_cache_release(page);
1133 pagecache_isize_extended(inode, old_size, pos);
1136 ret2 = ext4_mark_inode_dirty(handle, inode);
1141 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1142 /* if we have allocated more blocks and copied
1143 * less. We will have blocks allocated outside
1144 * inode->i_size. So truncate them
1146 ext4_orphan_add(handle, inode);
1148 ret2 = ext4_journal_stop(handle);
1151 if (pos + len > inode->i_size) {
1152 ext4_truncate_failed_write(inode);
1154 * If truncate failed early the inode might still be
1155 * on the orphan list; we need to make sure the inode
1156 * is removed from the orphan list in that case.
1159 ext4_orphan_del(NULL, inode);
1162 return ret ? ret : copied;
1166 * Reserve a single cluster located at lblock
1168 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1170 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1171 struct ext4_inode_info *ei = EXT4_I(inode);
1172 unsigned int md_needed;
1176 * We will charge metadata quota at writeout time; this saves
1177 * us from metadata over-estimation, though we may go over by
1178 * a small amount in the end. Here we just reserve for data.
1180 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1185 * recalculate the amount of metadata blocks to reserve
1186 * in order to allocate nrblocks
1187 * worse case is one extent per block
1189 spin_lock(&ei->i_block_reservation_lock);
1191 * ext4_calc_metadata_amount() has side effects, which we have
1192 * to be prepared undo if we fail to claim space.
1195 trace_ext4_da_reserve_space(inode, 0);
1197 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1198 spin_unlock(&ei->i_block_reservation_lock);
1199 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1202 ei->i_reserved_data_blocks++;
1203 spin_unlock(&ei->i_block_reservation_lock);
1205 return 0; /* success */
1208 static void ext4_da_release_space(struct inode *inode, int to_free)
1210 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1211 struct ext4_inode_info *ei = EXT4_I(inode);
1214 return; /* Nothing to release, exit */
1216 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1218 trace_ext4_da_release_space(inode, to_free);
1219 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1221 * if there aren't enough reserved blocks, then the
1222 * counter is messed up somewhere. Since this
1223 * function is called from invalidate page, it's
1224 * harmless to return without any action.
1226 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1227 "ino %lu, to_free %d with only %d reserved "
1228 "data blocks", inode->i_ino, to_free,
1229 ei->i_reserved_data_blocks);
1231 to_free = ei->i_reserved_data_blocks;
1233 ei->i_reserved_data_blocks -= to_free;
1235 /* update fs dirty data blocks counter */
1236 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1238 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1240 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1243 static void ext4_da_page_release_reservation(struct page *page,
1244 unsigned int offset,
1245 unsigned int length)
1248 struct buffer_head *head, *bh;
1249 unsigned int curr_off = 0;
1250 struct inode *inode = page->mapping->host;
1251 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1252 unsigned int stop = offset + length;
1256 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1258 head = page_buffers(page);
1261 unsigned int next_off = curr_off + bh->b_size;
1263 if (next_off > stop)
1266 if ((offset <= curr_off) && (buffer_delay(bh))) {
1268 clear_buffer_delay(bh);
1270 curr_off = next_off;
1271 } while ((bh = bh->b_this_page) != head);
1274 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1275 ext4_es_remove_extent(inode, lblk, to_release);
1278 /* If we have released all the blocks belonging to a cluster, then we
1279 * need to release the reserved space for that cluster. */
1280 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1281 while (num_clusters > 0) {
1282 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1283 ((num_clusters - 1) << sbi->s_cluster_bits);
1284 if (sbi->s_cluster_ratio == 1 ||
1285 !ext4_find_delalloc_cluster(inode, lblk))
1286 ext4_da_release_space(inode, 1);
1293 * Delayed allocation stuff
1296 struct mpage_da_data {
1297 struct inode *inode;
1298 struct writeback_control *wbc;
1300 pgoff_t first_page; /* The first page to write */
1301 pgoff_t next_page; /* Current page to examine */
1302 pgoff_t last_page; /* Last page to examine */
1304 * Extent to map - this can be after first_page because that can be
1305 * fully mapped. We somewhat abuse m_flags to store whether the extent
1306 * is delalloc or unwritten.
1308 struct ext4_map_blocks map;
1309 struct ext4_io_submit io_submit; /* IO submission data */
1312 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1317 struct pagevec pvec;
1318 struct inode *inode = mpd->inode;
1319 struct address_space *mapping = inode->i_mapping;
1321 /* This is necessary when next_page == 0. */
1322 if (mpd->first_page >= mpd->next_page)
1325 index = mpd->first_page;
1326 end = mpd->next_page - 1;
1328 ext4_lblk_t start, last;
1329 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1330 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1331 ext4_es_remove_extent(inode, start, last - start + 1);
1334 pagevec_init(&pvec, 0);
1335 while (index <= end) {
1336 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1339 for (i = 0; i < nr_pages; i++) {
1340 struct page *page = pvec.pages[i];
1341 if (page->index > end)
1343 BUG_ON(!PageLocked(page));
1344 BUG_ON(PageWriteback(page));
1346 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1347 ClearPageUptodate(page);
1351 index = pvec.pages[nr_pages - 1]->index + 1;
1352 pagevec_release(&pvec);
1356 static void ext4_print_free_blocks(struct inode *inode)
1358 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1359 struct super_block *sb = inode->i_sb;
1360 struct ext4_inode_info *ei = EXT4_I(inode);
1362 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1363 EXT4_C2B(EXT4_SB(inode->i_sb),
1364 ext4_count_free_clusters(sb)));
1365 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1366 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1367 (long long) EXT4_C2B(EXT4_SB(sb),
1368 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1369 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1370 (long long) EXT4_C2B(EXT4_SB(sb),
1371 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1372 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1373 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1374 ei->i_reserved_data_blocks);
1378 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1380 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1384 * This function is grabs code from the very beginning of
1385 * ext4_map_blocks, but assumes that the caller is from delayed write
1386 * time. This function looks up the requested blocks and sets the
1387 * buffer delay bit under the protection of i_data_sem.
1389 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1390 struct ext4_map_blocks *map,
1391 struct buffer_head *bh)
1393 struct extent_status es;
1395 sector_t invalid_block = ~((sector_t) 0xffff);
1396 #ifdef ES_AGGRESSIVE_TEST
1397 struct ext4_map_blocks orig_map;
1399 memcpy(&orig_map, map, sizeof(*map));
1402 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1406 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1407 "logical block %lu\n", inode->i_ino, map->m_len,
1408 (unsigned long) map->m_lblk);
1410 /* Lookup extent status tree firstly */
1411 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1412 if (ext4_es_is_hole(&es)) {
1414 down_read(&EXT4_I(inode)->i_data_sem);
1419 * Delayed extent could be allocated by fallocate.
1420 * So we need to check it.
1422 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1423 map_bh(bh, inode->i_sb, invalid_block);
1425 set_buffer_delay(bh);
1429 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1430 retval = es.es_len - (iblock - es.es_lblk);
1431 if (retval > map->m_len)
1432 retval = map->m_len;
1433 map->m_len = retval;
1434 if (ext4_es_is_written(&es))
1435 map->m_flags |= EXT4_MAP_MAPPED;
1436 else if (ext4_es_is_unwritten(&es))
1437 map->m_flags |= EXT4_MAP_UNWRITTEN;
1441 #ifdef ES_AGGRESSIVE_TEST
1442 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1448 * Try to see if we can get the block without requesting a new
1449 * file system block.
1451 down_read(&EXT4_I(inode)->i_data_sem);
1452 if (ext4_has_inline_data(inode))
1454 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1455 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1457 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1463 * XXX: __block_prepare_write() unmaps passed block,
1467 * If the block was allocated from previously allocated cluster,
1468 * then we don't need to reserve it again. However we still need
1469 * to reserve metadata for every block we're going to write.
1471 if (EXT4_SB(inode->i_sb)->s_cluster_ratio <= 1 ||
1472 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1473 ret = ext4_da_reserve_space(inode, iblock);
1475 /* not enough space to reserve */
1481 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1482 ~0, EXTENT_STATUS_DELAYED);
1488 map_bh(bh, inode->i_sb, invalid_block);
1490 set_buffer_delay(bh);
1491 } else if (retval > 0) {
1493 unsigned int status;
1495 if (unlikely(retval != map->m_len)) {
1496 ext4_warning(inode->i_sb,
1497 "ES len assertion failed for inode "
1498 "%lu: retval %d != map->m_len %d",
1499 inode->i_ino, retval, map->m_len);
1503 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1504 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1505 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1506 map->m_pblk, status);
1512 up_read((&EXT4_I(inode)->i_data_sem));
1518 * This is a special get_block_t callback which is used by
1519 * ext4_da_write_begin(). It will either return mapped block or
1520 * reserve space for a single block.
1522 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1523 * We also have b_blocknr = -1 and b_bdev initialized properly
1525 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1526 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1527 * initialized properly.
1529 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1530 struct buffer_head *bh, int create)
1532 struct ext4_map_blocks map;
1535 BUG_ON(create == 0);
1536 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1538 map.m_lblk = iblock;
1542 * first, we need to know whether the block is allocated already
1543 * preallocated blocks are unmapped but should treated
1544 * the same as allocated blocks.
1546 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1550 map_bh(bh, inode->i_sb, map.m_pblk);
1551 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1553 if (buffer_unwritten(bh)) {
1554 /* A delayed write to unwritten bh should be marked
1555 * new and mapped. Mapped ensures that we don't do
1556 * get_block multiple times when we write to the same
1557 * offset and new ensures that we do proper zero out
1558 * for partial write.
1561 set_buffer_mapped(bh);
1566 static int bget_one(handle_t *handle, struct buffer_head *bh)
1572 static int bput_one(handle_t *handle, struct buffer_head *bh)
1578 static int __ext4_journalled_writepage(struct page *page,
1581 struct address_space *mapping = page->mapping;
1582 struct inode *inode = mapping->host;
1583 struct buffer_head *page_bufs = NULL;
1584 handle_t *handle = NULL;
1585 int ret = 0, err = 0;
1586 int inline_data = ext4_has_inline_data(inode);
1587 struct buffer_head *inode_bh = NULL;
1589 ClearPageChecked(page);
1592 BUG_ON(page->index != 0);
1593 BUG_ON(len > ext4_get_max_inline_size(inode));
1594 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1595 if (inode_bh == NULL)
1598 page_bufs = page_buffers(page);
1603 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1606 /* As soon as we unlock the page, it can go away, but we have
1607 * references to buffers so we are safe */
1610 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1611 ext4_writepage_trans_blocks(inode));
1612 if (IS_ERR(handle)) {
1613 ret = PTR_ERR(handle);
1617 BUG_ON(!ext4_handle_valid(handle));
1620 BUFFER_TRACE(inode_bh, "get write access");
1621 ret = ext4_journal_get_write_access(handle, inode_bh);
1623 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1626 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1627 do_journal_get_write_access);
1629 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1634 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1635 err = ext4_journal_stop(handle);
1639 if (!ext4_has_inline_data(inode))
1640 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1642 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1649 * Note that we don't need to start a transaction unless we're journaling data
1650 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1651 * need to file the inode to the transaction's list in ordered mode because if
1652 * we are writing back data added by write(), the inode is already there and if
1653 * we are writing back data modified via mmap(), no one guarantees in which
1654 * transaction the data will hit the disk. In case we are journaling data, we
1655 * cannot start transaction directly because transaction start ranks above page
1656 * lock so we have to do some magic.
1658 * This function can get called via...
1659 * - ext4_writepages after taking page lock (have journal handle)
1660 * - journal_submit_inode_data_buffers (no journal handle)
1661 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1662 * - grab_page_cache when doing write_begin (have journal handle)
1664 * We don't do any block allocation in this function. If we have page with
1665 * multiple blocks we need to write those buffer_heads that are mapped. This
1666 * is important for mmaped based write. So if we do with blocksize 1K
1667 * truncate(f, 1024);
1668 * a = mmap(f, 0, 4096);
1670 * truncate(f, 4096);
1671 * we have in the page first buffer_head mapped via page_mkwrite call back
1672 * but other buffer_heads would be unmapped but dirty (dirty done via the
1673 * do_wp_page). So writepage should write the first block. If we modify
1674 * the mmap area beyond 1024 we will again get a page_fault and the
1675 * page_mkwrite callback will do the block allocation and mark the
1676 * buffer_heads mapped.
1678 * We redirty the page if we have any buffer_heads that is either delay or
1679 * unwritten in the page.
1681 * We can get recursively called as show below.
1683 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1686 * But since we don't do any block allocation we should not deadlock.
1687 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1689 static int ext4_writepage(struct page *page,
1690 struct writeback_control *wbc)
1695 struct buffer_head *page_bufs = NULL;
1696 struct inode *inode = page->mapping->host;
1697 struct ext4_io_submit io_submit;
1698 bool keep_towrite = false;
1700 trace_ext4_writepage(page);
1701 size = i_size_read(inode);
1702 if (page->index == size >> PAGE_CACHE_SHIFT)
1703 len = size & ~PAGE_CACHE_MASK;
1705 len = PAGE_CACHE_SIZE;
1707 page_bufs = page_buffers(page);
1709 * We cannot do block allocation or other extent handling in this
1710 * function. If there are buffers needing that, we have to redirty
1711 * the page. But we may reach here when we do a journal commit via
1712 * journal_submit_inode_data_buffers() and in that case we must write
1713 * allocated buffers to achieve data=ordered mode guarantees.
1715 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1716 ext4_bh_delay_or_unwritten)) {
1717 redirty_page_for_writepage(wbc, page);
1718 if (current->flags & PF_MEMALLOC) {
1720 * For memory cleaning there's no point in writing only
1721 * some buffers. So just bail out. Warn if we came here
1722 * from direct reclaim.
1724 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1729 keep_towrite = true;
1732 if (PageChecked(page) && ext4_should_journal_data(inode))
1734 * It's mmapped pagecache. Add buffers and journal it. There
1735 * doesn't seem much point in redirtying the page here.
1737 return __ext4_journalled_writepage(page, len);
1739 ext4_io_submit_init(&io_submit, wbc);
1740 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1741 if (!io_submit.io_end) {
1742 redirty_page_for_writepage(wbc, page);
1746 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1747 ext4_io_submit(&io_submit);
1748 /* Drop io_end reference we got from init */
1749 ext4_put_io_end_defer(io_submit.io_end);
1753 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1756 loff_t size = i_size_read(mpd->inode);
1759 BUG_ON(page->index != mpd->first_page);
1760 if (page->index == size >> PAGE_CACHE_SHIFT)
1761 len = size & ~PAGE_CACHE_MASK;
1763 len = PAGE_CACHE_SIZE;
1764 clear_page_dirty_for_io(page);
1765 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1767 mpd->wbc->nr_to_write--;
1773 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1776 * mballoc gives us at most this number of blocks...
1777 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1778 * The rest of mballoc seems to handle chunks up to full group size.
1780 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1783 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1785 * @mpd - extent of blocks
1786 * @lblk - logical number of the block in the file
1787 * @bh - buffer head we want to add to the extent
1789 * The function is used to collect contig. blocks in the same state. If the
1790 * buffer doesn't require mapping for writeback and we haven't started the
1791 * extent of buffers to map yet, the function returns 'true' immediately - the
1792 * caller can write the buffer right away. Otherwise the function returns true
1793 * if the block has been added to the extent, false if the block couldn't be
1796 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1797 struct buffer_head *bh)
1799 struct ext4_map_blocks *map = &mpd->map;
1801 /* Buffer that doesn't need mapping for writeback? */
1802 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1803 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1804 /* So far no extent to map => we write the buffer right away */
1805 if (map->m_len == 0)
1810 /* First block in the extent? */
1811 if (map->m_len == 0) {
1814 map->m_flags = bh->b_state & BH_FLAGS;
1818 /* Don't go larger than mballoc is willing to allocate */
1819 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1822 /* Can we merge the block to our big extent? */
1823 if (lblk == map->m_lblk + map->m_len &&
1824 (bh->b_state & BH_FLAGS) == map->m_flags) {
1832 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1834 * @mpd - extent of blocks for mapping
1835 * @head - the first buffer in the page
1836 * @bh - buffer we should start processing from
1837 * @lblk - logical number of the block in the file corresponding to @bh
1839 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1840 * the page for IO if all buffers in this page were mapped and there's no
1841 * accumulated extent of buffers to map or add buffers in the page to the
1842 * extent of buffers to map. The function returns 1 if the caller can continue
1843 * by processing the next page, 0 if it should stop adding buffers to the
1844 * extent to map because we cannot extend it anymore. It can also return value
1845 * < 0 in case of error during IO submission.
1847 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1848 struct buffer_head *head,
1849 struct buffer_head *bh,
1852 struct inode *inode = mpd->inode;
1854 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
1855 >> inode->i_blkbits;
1858 BUG_ON(buffer_locked(bh));
1860 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
1861 /* Found extent to map? */
1864 /* Everything mapped so far and we hit EOF */
1867 } while (lblk++, (bh = bh->b_this_page) != head);
1868 /* So far everything mapped? Submit the page for IO. */
1869 if (mpd->map.m_len == 0) {
1870 err = mpage_submit_page(mpd, head->b_page);
1874 return lblk < blocks;
1878 * mpage_map_buffers - update buffers corresponding to changed extent and
1879 * submit fully mapped pages for IO
1881 * @mpd - description of extent to map, on return next extent to map
1883 * Scan buffers corresponding to changed extent (we expect corresponding pages
1884 * to be already locked) and update buffer state according to new extent state.
1885 * We map delalloc buffers to their physical location, clear unwritten bits,
1886 * and mark buffers as uninit when we perform writes to unwritten extents
1887 * and do extent conversion after IO is finished. If the last page is not fully
1888 * mapped, we update @map to the next extent in the last page that needs
1889 * mapping. Otherwise we submit the page for IO.
1891 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
1893 struct pagevec pvec;
1895 struct inode *inode = mpd->inode;
1896 struct buffer_head *head, *bh;
1897 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
1903 start = mpd->map.m_lblk >> bpp_bits;
1904 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
1905 lblk = start << bpp_bits;
1906 pblock = mpd->map.m_pblk;
1908 pagevec_init(&pvec, 0);
1909 while (start <= end) {
1910 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
1914 for (i = 0; i < nr_pages; i++) {
1915 struct page *page = pvec.pages[i];
1917 if (page->index > end)
1919 /* Up to 'end' pages must be contiguous */
1920 BUG_ON(page->index != start);
1921 bh = head = page_buffers(page);
1923 if (lblk < mpd->map.m_lblk)
1925 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
1927 * Buffer after end of mapped extent.
1928 * Find next buffer in the page to map.
1931 mpd->map.m_flags = 0;
1933 * FIXME: If dioread_nolock supports
1934 * blocksize < pagesize, we need to make
1935 * sure we add size mapped so far to
1936 * io_end->size as the following call
1937 * can submit the page for IO.
1939 err = mpage_process_page_bufs(mpd, head,
1941 pagevec_release(&pvec);
1946 if (buffer_delay(bh)) {
1947 clear_buffer_delay(bh);
1948 bh->b_blocknr = pblock++;
1950 clear_buffer_unwritten(bh);
1951 } while (lblk++, (bh = bh->b_this_page) != head);
1954 * FIXME: This is going to break if dioread_nolock
1955 * supports blocksize < pagesize as we will try to
1956 * convert potentially unmapped parts of inode.
1958 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
1959 /* Page fully mapped - let IO run! */
1960 err = mpage_submit_page(mpd, page);
1962 pagevec_release(&pvec);
1967 pagevec_release(&pvec);
1969 /* Extent fully mapped and matches with page boundary. We are done. */
1971 mpd->map.m_flags = 0;
1975 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
1977 struct inode *inode = mpd->inode;
1978 struct ext4_map_blocks *map = &mpd->map;
1979 int get_blocks_flags;
1980 int err, dioread_nolock;
1982 trace_ext4_da_write_pages_extent(inode, map);
1984 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
1985 * to convert an unwritten extent to be initialized (in the case
1986 * where we have written into one or more preallocated blocks). It is
1987 * possible that we're going to need more metadata blocks than
1988 * previously reserved. However we must not fail because we're in
1989 * writeback and there is nothing we can do about it so it might result
1990 * in data loss. So use reserved blocks to allocate metadata if
1993 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
1994 * the blocks in question are delalloc blocks. This indicates
1995 * that the blocks and quotas has already been checked when
1996 * the data was copied into the page cache.
1998 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
1999 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2000 dioread_nolock = ext4_should_dioread_nolock(inode);
2002 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2003 if (map->m_flags & (1 << BH_Delay))
2004 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2006 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2009 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2010 if (!mpd->io_submit.io_end->handle &&
2011 ext4_handle_valid(handle)) {
2012 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2013 handle->h_rsv_handle = NULL;
2015 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2018 BUG_ON(map->m_len == 0);
2019 if (map->m_flags & EXT4_MAP_NEW) {
2020 struct block_device *bdev = inode->i_sb->s_bdev;
2023 for (i = 0; i < map->m_len; i++)
2024 unmap_underlying_metadata(bdev, map->m_pblk + i);
2030 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2031 * mpd->len and submit pages underlying it for IO
2033 * @handle - handle for journal operations
2034 * @mpd - extent to map
2035 * @give_up_on_write - we set this to true iff there is a fatal error and there
2036 * is no hope of writing the data. The caller should discard
2037 * dirty pages to avoid infinite loops.
2039 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2040 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2041 * them to initialized or split the described range from larger unwritten
2042 * extent. Note that we need not map all the described range since allocation
2043 * can return less blocks or the range is covered by more unwritten extents. We
2044 * cannot map more because we are limited by reserved transaction credits. On
2045 * the other hand we always make sure that the last touched page is fully
2046 * mapped so that it can be written out (and thus forward progress is
2047 * guaranteed). After mapping we submit all mapped pages for IO.
2049 static int mpage_map_and_submit_extent(handle_t *handle,
2050 struct mpage_da_data *mpd,
2051 bool *give_up_on_write)
2053 struct inode *inode = mpd->inode;
2054 struct ext4_map_blocks *map = &mpd->map;
2059 mpd->io_submit.io_end->offset =
2060 ((loff_t)map->m_lblk) << inode->i_blkbits;
2062 err = mpage_map_one_extent(handle, mpd);
2064 struct super_block *sb = inode->i_sb;
2066 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2067 goto invalidate_dirty_pages;
2069 * Let the uper layers retry transient errors.
2070 * In the case of ENOSPC, if ext4_count_free_blocks()
2071 * is non-zero, a commit should free up blocks.
2073 if ((err == -ENOMEM) ||
2074 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2076 goto update_disksize;
2079 ext4_msg(sb, KERN_CRIT,
2080 "Delayed block allocation failed for "
2081 "inode %lu at logical offset %llu with"
2082 " max blocks %u with error %d",
2084 (unsigned long long)map->m_lblk,
2085 (unsigned)map->m_len, -err);
2086 ext4_msg(sb, KERN_CRIT,
2087 "This should not happen!! Data will "
2090 ext4_print_free_blocks(inode);
2091 invalidate_dirty_pages:
2092 *give_up_on_write = true;
2097 * Update buffer state, submit mapped pages, and get us new
2100 err = mpage_map_and_submit_buffers(mpd);
2102 goto update_disksize;
2103 } while (map->m_len);
2107 * Update on-disk size after IO is submitted. Races with
2108 * truncate are avoided by checking i_size under i_data_sem.
2110 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2111 if (disksize > EXT4_I(inode)->i_disksize) {
2115 down_write(&EXT4_I(inode)->i_data_sem);
2116 i_size = i_size_read(inode);
2117 if (disksize > i_size)
2119 if (disksize > EXT4_I(inode)->i_disksize)
2120 EXT4_I(inode)->i_disksize = disksize;
2121 err2 = ext4_mark_inode_dirty(handle, inode);
2122 up_write(&EXT4_I(inode)->i_data_sem);
2124 ext4_error(inode->i_sb,
2125 "Failed to mark inode %lu dirty",
2134 * Calculate the total number of credits to reserve for one writepages
2135 * iteration. This is called from ext4_writepages(). We map an extent of
2136 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2137 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2138 * bpp - 1 blocks in bpp different extents.
2140 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2142 int bpp = ext4_journal_blocks_per_page(inode);
2144 return ext4_meta_trans_blocks(inode,
2145 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2149 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2150 * and underlying extent to map
2152 * @mpd - where to look for pages
2154 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2155 * IO immediately. When we find a page which isn't mapped we start accumulating
2156 * extent of buffers underlying these pages that needs mapping (formed by
2157 * either delayed or unwritten buffers). We also lock the pages containing
2158 * these buffers. The extent found is returned in @mpd structure (starting at
2159 * mpd->lblk with length mpd->len blocks).
2161 * Note that this function can attach bios to one io_end structure which are
2162 * neither logically nor physically contiguous. Although it may seem as an
2163 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2164 * case as we need to track IO to all buffers underlying a page in one io_end.
2166 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2168 struct address_space *mapping = mpd->inode->i_mapping;
2169 struct pagevec pvec;
2170 unsigned int nr_pages;
2171 long left = mpd->wbc->nr_to_write;
2172 pgoff_t index = mpd->first_page;
2173 pgoff_t end = mpd->last_page;
2176 int blkbits = mpd->inode->i_blkbits;
2178 struct buffer_head *head;
2180 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2181 tag = PAGECACHE_TAG_TOWRITE;
2183 tag = PAGECACHE_TAG_DIRTY;
2185 pagevec_init(&pvec, 0);
2187 mpd->next_page = index;
2188 while (index <= end) {
2189 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2190 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2194 for (i = 0; i < nr_pages; i++) {
2195 struct page *page = pvec.pages[i];
2198 * At this point, the page may be truncated or
2199 * invalidated (changing page->mapping to NULL), or
2200 * even swizzled back from swapper_space to tmpfs file
2201 * mapping. However, page->index will not change
2202 * because we have a reference on the page.
2204 if (page->index > end)
2208 * Accumulated enough dirty pages? This doesn't apply
2209 * to WB_SYNC_ALL mode. For integrity sync we have to
2210 * keep going because someone may be concurrently
2211 * dirtying pages, and we might have synced a lot of
2212 * newly appeared dirty pages, but have not synced all
2213 * of the old dirty pages.
2215 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2218 /* If we can't merge this page, we are done. */
2219 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2224 * If the page is no longer dirty, or its mapping no
2225 * longer corresponds to inode we are writing (which
2226 * means it has been truncated or invalidated), or the
2227 * page is already under writeback and we are not doing
2228 * a data integrity writeback, skip the page
2230 if (!PageDirty(page) ||
2231 (PageWriteback(page) &&
2232 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2233 unlikely(page->mapping != mapping)) {
2238 wait_on_page_writeback(page);
2239 BUG_ON(PageWriteback(page));
2241 if (mpd->map.m_len == 0)
2242 mpd->first_page = page->index;
2243 mpd->next_page = page->index + 1;
2244 /* Add all dirty buffers to mpd */
2245 lblk = ((ext4_lblk_t)page->index) <<
2246 (PAGE_CACHE_SHIFT - blkbits);
2247 head = page_buffers(page);
2248 err = mpage_process_page_bufs(mpd, head, head, lblk);
2254 pagevec_release(&pvec);
2259 pagevec_release(&pvec);
2263 static int __writepage(struct page *page, struct writeback_control *wbc,
2266 struct address_space *mapping = data;
2267 int ret = ext4_writepage(page, wbc);
2268 mapping_set_error(mapping, ret);
2272 static int ext4_writepages(struct address_space *mapping,
2273 struct writeback_control *wbc)
2275 pgoff_t writeback_index = 0;
2276 long nr_to_write = wbc->nr_to_write;
2277 int range_whole = 0;
2279 handle_t *handle = NULL;
2280 struct mpage_da_data mpd;
2281 struct inode *inode = mapping->host;
2282 int needed_blocks, rsv_blocks = 0, ret = 0;
2283 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2285 struct blk_plug plug;
2286 bool give_up_on_write = false;
2288 trace_ext4_writepages(inode, wbc);
2291 * No pages to write? This is mainly a kludge to avoid starting
2292 * a transaction for special inodes like journal inode on last iput()
2293 * because that could violate lock ordering on umount
2295 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2296 goto out_writepages;
2298 if (ext4_should_journal_data(inode)) {
2299 struct blk_plug plug;
2301 blk_start_plug(&plug);
2302 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2303 blk_finish_plug(&plug);
2304 goto out_writepages;
2308 * If the filesystem has aborted, it is read-only, so return
2309 * right away instead of dumping stack traces later on that
2310 * will obscure the real source of the problem. We test
2311 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2312 * the latter could be true if the filesystem is mounted
2313 * read-only, and in that case, ext4_writepages should
2314 * *never* be called, so if that ever happens, we would want
2317 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2319 goto out_writepages;
2322 if (ext4_should_dioread_nolock(inode)) {
2324 * We may need to convert up to one extent per block in
2325 * the page and we may dirty the inode.
2327 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2331 * If we have inline data and arrive here, it means that
2332 * we will soon create the block for the 1st page, so
2333 * we'd better clear the inline data here.
2335 if (ext4_has_inline_data(inode)) {
2336 /* Just inode will be modified... */
2337 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2338 if (IS_ERR(handle)) {
2339 ret = PTR_ERR(handle);
2340 goto out_writepages;
2342 BUG_ON(ext4_test_inode_state(inode,
2343 EXT4_STATE_MAY_INLINE_DATA));
2344 ext4_destroy_inline_data(handle, inode);
2345 ext4_journal_stop(handle);
2348 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2351 if (wbc->range_cyclic) {
2352 writeback_index = mapping->writeback_index;
2353 if (writeback_index)
2355 mpd.first_page = writeback_index;
2358 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2359 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2364 ext4_io_submit_init(&mpd.io_submit, wbc);
2366 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2367 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2369 blk_start_plug(&plug);
2370 while (!done && mpd.first_page <= mpd.last_page) {
2371 /* For each extent of pages we use new io_end */
2372 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2373 if (!mpd.io_submit.io_end) {
2379 * We have two constraints: We find one extent to map and we
2380 * must always write out whole page (makes a difference when
2381 * blocksize < pagesize) so that we don't block on IO when we
2382 * try to write out the rest of the page. Journalled mode is
2383 * not supported by delalloc.
2385 BUG_ON(ext4_should_journal_data(inode));
2386 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2388 /* start a new transaction */
2389 handle = ext4_journal_start_with_reserve(inode,
2390 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2391 if (IS_ERR(handle)) {
2392 ret = PTR_ERR(handle);
2393 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2394 "%ld pages, ino %lu; err %d", __func__,
2395 wbc->nr_to_write, inode->i_ino, ret);
2396 /* Release allocated io_end */
2397 ext4_put_io_end(mpd.io_submit.io_end);
2401 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2402 ret = mpage_prepare_extent_to_map(&mpd);
2405 ret = mpage_map_and_submit_extent(handle, &mpd,
2409 * We scanned the whole range (or exhausted
2410 * nr_to_write), submitted what was mapped and
2411 * didn't find anything needing mapping. We are
2417 ext4_journal_stop(handle);
2418 /* Submit prepared bio */
2419 ext4_io_submit(&mpd.io_submit);
2420 /* Unlock pages we didn't use */
2421 mpage_release_unused_pages(&mpd, give_up_on_write);
2422 /* Drop our io_end reference we got from init */
2423 ext4_put_io_end(mpd.io_submit.io_end);
2425 if (ret == -ENOSPC && sbi->s_journal) {
2427 * Commit the transaction which would
2428 * free blocks released in the transaction
2431 jbd2_journal_force_commit_nested(sbi->s_journal);
2435 /* Fatal error - ENOMEM, EIO... */
2439 blk_finish_plug(&plug);
2440 if (!ret && !cycled && wbc->nr_to_write > 0) {
2442 mpd.last_page = writeback_index - 1;
2448 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2450 * Set the writeback_index so that range_cyclic
2451 * mode will write it back later
2453 mapping->writeback_index = mpd.first_page;
2456 trace_ext4_writepages_result(inode, wbc, ret,
2457 nr_to_write - wbc->nr_to_write);
2461 static int ext4_nonda_switch(struct super_block *sb)
2463 s64 free_clusters, dirty_clusters;
2464 struct ext4_sb_info *sbi = EXT4_SB(sb);
2467 * switch to non delalloc mode if we are running low
2468 * on free block. The free block accounting via percpu
2469 * counters can get slightly wrong with percpu_counter_batch getting
2470 * accumulated on each CPU without updating global counters
2471 * Delalloc need an accurate free block accounting. So switch
2472 * to non delalloc when we are near to error range.
2475 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2477 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2479 * Start pushing delalloc when 1/2 of free blocks are dirty.
2481 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2482 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2484 if (2 * free_clusters < 3 * dirty_clusters ||
2485 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2487 * free block count is less than 150% of dirty blocks
2488 * or free blocks is less than watermark
2495 /* We always reserve for an inode update; the superblock could be there too */
2496 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2498 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
2499 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2502 if (pos + len <= 0x7fffffffULL)
2505 /* We might need to update the superblock to set LARGE_FILE */
2509 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2510 loff_t pos, unsigned len, unsigned flags,
2511 struct page **pagep, void **fsdata)
2513 int ret, retries = 0;
2516 struct inode *inode = mapping->host;
2519 index = pos >> PAGE_CACHE_SHIFT;
2521 if (ext4_nonda_switch(inode->i_sb)) {
2522 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2523 return ext4_write_begin(file, mapping, pos,
2524 len, flags, pagep, fsdata);
2526 *fsdata = (void *)0;
2527 trace_ext4_da_write_begin(inode, pos, len, flags);
2529 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2530 ret = ext4_da_write_inline_data_begin(mapping, inode,
2540 * grab_cache_page_write_begin() can take a long time if the
2541 * system is thrashing due to memory pressure, or if the page
2542 * is being written back. So grab it first before we start
2543 * the transaction handle. This also allows us to allocate
2544 * the page (if needed) without using GFP_NOFS.
2547 page = grab_cache_page_write_begin(mapping, index, flags);
2553 * With delayed allocation, we don't log the i_disksize update
2554 * if there is delayed block allocation. But we still need
2555 * to journalling the i_disksize update if writes to the end
2556 * of file which has an already mapped buffer.
2559 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2560 ext4_da_write_credits(inode, pos, len));
2561 if (IS_ERR(handle)) {
2562 page_cache_release(page);
2563 return PTR_ERR(handle);
2567 if (page->mapping != mapping) {
2568 /* The page got truncated from under us */
2570 page_cache_release(page);
2571 ext4_journal_stop(handle);
2574 /* In case writeback began while the page was unlocked */
2575 wait_for_stable_page(page);
2577 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2580 ext4_journal_stop(handle);
2582 * block_write_begin may have instantiated a few blocks
2583 * outside i_size. Trim these off again. Don't need
2584 * i_size_read because we hold i_mutex.
2586 if (pos + len > inode->i_size)
2587 ext4_truncate_failed_write(inode);
2589 if (ret == -ENOSPC &&
2590 ext4_should_retry_alloc(inode->i_sb, &retries))
2593 page_cache_release(page);
2602 * Check if we should update i_disksize
2603 * when write to the end of file but not require block allocation
2605 static int ext4_da_should_update_i_disksize(struct page *page,
2606 unsigned long offset)
2608 struct buffer_head *bh;
2609 struct inode *inode = page->mapping->host;
2613 bh = page_buffers(page);
2614 idx = offset >> inode->i_blkbits;
2616 for (i = 0; i < idx; i++)
2617 bh = bh->b_this_page;
2619 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2624 static int ext4_da_write_end(struct file *file,
2625 struct address_space *mapping,
2626 loff_t pos, unsigned len, unsigned copied,
2627 struct page *page, void *fsdata)
2629 struct inode *inode = mapping->host;
2631 handle_t *handle = ext4_journal_current_handle();
2633 unsigned long start, end;
2634 int write_mode = (int)(unsigned long)fsdata;
2636 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2637 return ext4_write_end(file, mapping, pos,
2638 len, copied, page, fsdata);
2640 trace_ext4_da_write_end(inode, pos, len, copied);
2641 start = pos & (PAGE_CACHE_SIZE - 1);
2642 end = start + copied - 1;
2645 * generic_write_end() will run mark_inode_dirty() if i_size
2646 * changes. So let's piggyback the i_disksize mark_inode_dirty
2649 new_i_size = pos + copied;
2650 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2651 if (ext4_has_inline_data(inode) ||
2652 ext4_da_should_update_i_disksize(page, end)) {
2653 ext4_update_i_disksize(inode, new_i_size);
2654 /* We need to mark inode dirty even if
2655 * new_i_size is less that inode->i_size
2656 * bu greater than i_disksize.(hint delalloc)
2658 ext4_mark_inode_dirty(handle, inode);
2662 if (write_mode != CONVERT_INLINE_DATA &&
2663 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2664 ext4_has_inline_data(inode))
2665 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2668 ret2 = generic_write_end(file, mapping, pos, len, copied,
2674 ret2 = ext4_journal_stop(handle);
2678 return ret ? ret : copied;
2681 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2682 unsigned int length)
2685 * Drop reserved blocks
2687 BUG_ON(!PageLocked(page));
2688 if (!page_has_buffers(page))
2691 ext4_da_page_release_reservation(page, offset, length);
2694 ext4_invalidatepage(page, offset, length);
2700 * Force all delayed allocation blocks to be allocated for a given inode.
2702 int ext4_alloc_da_blocks(struct inode *inode)
2704 trace_ext4_alloc_da_blocks(inode);
2706 if (!EXT4_I(inode)->i_reserved_data_blocks)
2710 * We do something simple for now. The filemap_flush() will
2711 * also start triggering a write of the data blocks, which is
2712 * not strictly speaking necessary (and for users of
2713 * laptop_mode, not even desirable). However, to do otherwise
2714 * would require replicating code paths in:
2716 * ext4_writepages() ->
2717 * write_cache_pages() ---> (via passed in callback function)
2718 * __mpage_da_writepage() -->
2719 * mpage_add_bh_to_extent()
2720 * mpage_da_map_blocks()
2722 * The problem is that write_cache_pages(), located in
2723 * mm/page-writeback.c, marks pages clean in preparation for
2724 * doing I/O, which is not desirable if we're not planning on
2727 * We could call write_cache_pages(), and then redirty all of
2728 * the pages by calling redirty_page_for_writepage() but that
2729 * would be ugly in the extreme. So instead we would need to
2730 * replicate parts of the code in the above functions,
2731 * simplifying them because we wouldn't actually intend to
2732 * write out the pages, but rather only collect contiguous
2733 * logical block extents, call the multi-block allocator, and
2734 * then update the buffer heads with the block allocations.
2736 * For now, though, we'll cheat by calling filemap_flush(),
2737 * which will map the blocks, and start the I/O, but not
2738 * actually wait for the I/O to complete.
2740 return filemap_flush(inode->i_mapping);
2744 * bmap() is special. It gets used by applications such as lilo and by
2745 * the swapper to find the on-disk block of a specific piece of data.
2747 * Naturally, this is dangerous if the block concerned is still in the
2748 * journal. If somebody makes a swapfile on an ext4 data-journaling
2749 * filesystem and enables swap, then they may get a nasty shock when the
2750 * data getting swapped to that swapfile suddenly gets overwritten by
2751 * the original zero's written out previously to the journal and
2752 * awaiting writeback in the kernel's buffer cache.
2754 * So, if we see any bmap calls here on a modified, data-journaled file,
2755 * take extra steps to flush any blocks which might be in the cache.
2757 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2759 struct inode *inode = mapping->host;
2764 * We can get here for an inline file via the FIBMAP ioctl
2766 if (ext4_has_inline_data(inode))
2769 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2770 test_opt(inode->i_sb, DELALLOC)) {
2772 * With delalloc we want to sync the file
2773 * so that we can make sure we allocate
2776 filemap_write_and_wait(mapping);
2779 if (EXT4_JOURNAL(inode) &&
2780 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2782 * This is a REALLY heavyweight approach, but the use of
2783 * bmap on dirty files is expected to be extremely rare:
2784 * only if we run lilo or swapon on a freshly made file
2785 * do we expect this to happen.
2787 * (bmap requires CAP_SYS_RAWIO so this does not
2788 * represent an unprivileged user DOS attack --- we'd be
2789 * in trouble if mortal users could trigger this path at
2792 * NB. EXT4_STATE_JDATA is not set on files other than
2793 * regular files. If somebody wants to bmap a directory
2794 * or symlink and gets confused because the buffer
2795 * hasn't yet been flushed to disk, they deserve
2796 * everything they get.
2799 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2800 journal = EXT4_JOURNAL(inode);
2801 jbd2_journal_lock_updates(journal);
2802 err = jbd2_journal_flush(journal);
2803 jbd2_journal_unlock_updates(journal);
2809 return generic_block_bmap(mapping, block, ext4_get_block);
2812 static int ext4_readpage(struct file *file, struct page *page)
2815 struct inode *inode = page->mapping->host;
2817 trace_ext4_readpage(page);
2819 if (ext4_has_inline_data(inode))
2820 ret = ext4_readpage_inline(inode, page);
2823 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
2829 ext4_readpages(struct file *file, struct address_space *mapping,
2830 struct list_head *pages, unsigned nr_pages)
2832 struct inode *inode = mapping->host;
2834 /* If the file has inline data, no need to do readpages. */
2835 if (ext4_has_inline_data(inode))
2838 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
2841 static void ext4_invalidatepage(struct page *page, unsigned int offset,
2842 unsigned int length)
2844 trace_ext4_invalidatepage(page, offset, length);
2846 /* No journalling happens on data buffers when this function is used */
2847 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2849 block_invalidatepage(page, offset, length);
2852 static int __ext4_journalled_invalidatepage(struct page *page,
2853 unsigned int offset,
2854 unsigned int length)
2856 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2858 trace_ext4_journalled_invalidatepage(page, offset, length);
2861 * If it's a full truncate we just forget about the pending dirtying
2863 if (offset == 0 && length == PAGE_CACHE_SIZE)
2864 ClearPageChecked(page);
2866 return jbd2_journal_invalidatepage(journal, page, offset, length);
2869 /* Wrapper for aops... */
2870 static void ext4_journalled_invalidatepage(struct page *page,
2871 unsigned int offset,
2872 unsigned int length)
2874 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
2877 static int ext4_releasepage(struct page *page, gfp_t wait)
2879 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2881 trace_ext4_releasepage(page);
2883 /* Page has dirty journalled data -> cannot release */
2884 if (PageChecked(page))
2887 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2889 return try_to_free_buffers(page);
2893 * ext4_get_block used when preparing for a DIO write or buffer write.
2894 * We allocate an uinitialized extent if blocks haven't been allocated.
2895 * The extent will be converted to initialized after the IO is complete.
2897 int ext4_get_block_write(struct inode *inode, sector_t iblock,
2898 struct buffer_head *bh_result, int create)
2900 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2901 inode->i_ino, create);
2902 return _ext4_get_block(inode, iblock, bh_result,
2903 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2906 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2907 struct buffer_head *bh_result, int create)
2909 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2910 inode->i_ino, create);
2911 return _ext4_get_block(inode, iblock, bh_result,
2912 EXT4_GET_BLOCKS_NO_LOCK);
2915 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2916 ssize_t size, void *private)
2918 ext4_io_end_t *io_end = iocb->private;
2920 /* if not async direct IO just return */
2924 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2925 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2926 iocb->private, io_end->inode->i_ino, iocb, offset,
2929 iocb->private = NULL;
2930 io_end->offset = offset;
2931 io_end->size = size;
2932 ext4_put_io_end(io_end);
2936 * For ext4 extent files, ext4 will do direct-io write to holes,
2937 * preallocated extents, and those write extend the file, no need to
2938 * fall back to buffered IO.
2940 * For holes, we fallocate those blocks, mark them as unwritten
2941 * If those blocks were preallocated, we mark sure they are split, but
2942 * still keep the range to write as unwritten.
2944 * The unwritten extents will be converted to written when DIO is completed.
2945 * For async direct IO, since the IO may still pending when return, we
2946 * set up an end_io call back function, which will do the conversion
2947 * when async direct IO completed.
2949 * If the O_DIRECT write will extend the file then add this inode to the
2950 * orphan list. So recovery will truncate it back to the original size
2951 * if the machine crashes during the write.
2954 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2955 struct iov_iter *iter, loff_t offset)
2957 struct file *file = iocb->ki_filp;
2958 struct inode *inode = file->f_mapping->host;
2960 size_t count = iov_iter_count(iter);
2962 get_block_t *get_block_func = NULL;
2964 loff_t final_size = offset + count;
2965 ext4_io_end_t *io_end = NULL;
2967 /* Use the old path for reads and writes beyond i_size. */
2968 if (rw != WRITE || final_size > inode->i_size)
2969 return ext4_ind_direct_IO(rw, iocb, iter, offset);
2971 BUG_ON(iocb->private == NULL);
2974 * Make all waiters for direct IO properly wait also for extent
2975 * conversion. This also disallows race between truncate() and
2976 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
2979 atomic_inc(&inode->i_dio_count);
2981 /* If we do a overwrite dio, i_mutex locking can be released */
2982 overwrite = *((int *)iocb->private);
2985 down_read(&EXT4_I(inode)->i_data_sem);
2986 mutex_unlock(&inode->i_mutex);
2990 * We could direct write to holes and fallocate.
2992 * Allocated blocks to fill the hole are marked as
2993 * unwritten to prevent parallel buffered read to expose
2994 * the stale data before DIO complete the data IO.
2996 * As to previously fallocated extents, ext4 get_block will
2997 * just simply mark the buffer mapped but still keep the
2998 * extents unwritten.
3000 * For non AIO case, we will convert those unwritten extents
3001 * to written after return back from blockdev_direct_IO.
3003 * For async DIO, the conversion needs to be deferred when the
3004 * IO is completed. The ext4 end_io callback function will be
3005 * called to take care of the conversion work. Here for async
3006 * case, we allocate an io_end structure to hook to the iocb.
3008 iocb->private = NULL;
3009 ext4_inode_aio_set(inode, NULL);
3010 if (!is_sync_kiocb(iocb)) {
3011 io_end = ext4_init_io_end(inode, GFP_NOFS);
3017 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3019 iocb->private = ext4_get_io_end(io_end);
3021 * we save the io structure for current async direct
3022 * IO, so that later ext4_map_blocks() could flag the
3023 * io structure whether there is a unwritten extents
3024 * needs to be converted when IO is completed.
3026 ext4_inode_aio_set(inode, io_end);
3030 get_block_func = ext4_get_block_write_nolock;
3032 get_block_func = ext4_get_block_write;
3033 dio_flags = DIO_LOCKING;
3036 ret = dax_do_io(rw, iocb, inode, iter, offset, get_block_func,
3037 ext4_end_io_dio, dio_flags);
3039 ret = __blockdev_direct_IO(rw, iocb, inode,
3040 inode->i_sb->s_bdev, iter, offset,
3042 ext4_end_io_dio, NULL, dio_flags);
3045 * Put our reference to io_end. This can free the io_end structure e.g.
3046 * in sync IO case or in case of error. It can even perform extent
3047 * conversion if all bios we submitted finished before we got here.
3048 * Note that in that case iocb->private can be already set to NULL
3052 ext4_inode_aio_set(inode, NULL);
3053 ext4_put_io_end(io_end);
3055 * When no IO was submitted ext4_end_io_dio() was not
3056 * called so we have to put iocb's reference.
3058 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3059 WARN_ON(iocb->private != io_end);
3060 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3061 ext4_put_io_end(io_end);
3062 iocb->private = NULL;
3065 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3066 EXT4_STATE_DIO_UNWRITTEN)) {
3069 * for non AIO case, since the IO is already
3070 * completed, we could do the conversion right here
3072 err = ext4_convert_unwritten_extents(NULL, inode,
3076 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3081 inode_dio_done(inode);
3082 /* take i_mutex locking again if we do a ovewrite dio */
3084 up_read(&EXT4_I(inode)->i_data_sem);
3085 mutex_lock(&inode->i_mutex);
3091 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3092 struct iov_iter *iter, loff_t offset)
3094 struct file *file = iocb->ki_filp;
3095 struct inode *inode = file->f_mapping->host;
3096 size_t count = iov_iter_count(iter);
3100 * If we are doing data journalling we don't support O_DIRECT
3102 if (ext4_should_journal_data(inode))
3105 /* Let buffer I/O handle the inline data case. */
3106 if (ext4_has_inline_data(inode))
3109 trace_ext4_direct_IO_enter(inode, offset, count, rw);
3110 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3111 ret = ext4_ext_direct_IO(rw, iocb, iter, offset);
3113 ret = ext4_ind_direct_IO(rw, iocb, iter, offset);
3114 trace_ext4_direct_IO_exit(inode, offset, count, rw, ret);
3119 * Pages can be marked dirty completely asynchronously from ext4's journalling
3120 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3121 * much here because ->set_page_dirty is called under VFS locks. The page is
3122 * not necessarily locked.
3124 * We cannot just dirty the page and leave attached buffers clean, because the
3125 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3126 * or jbddirty because all the journalling code will explode.
3128 * So what we do is to mark the page "pending dirty" and next time writepage
3129 * is called, propagate that into the buffers appropriately.
3131 static int ext4_journalled_set_page_dirty(struct page *page)
3133 SetPageChecked(page);
3134 return __set_page_dirty_nobuffers(page);
3137 static const struct address_space_operations ext4_aops = {
3138 .readpage = ext4_readpage,
3139 .readpages = ext4_readpages,
3140 .writepage = ext4_writepage,
3141 .writepages = ext4_writepages,
3142 .write_begin = ext4_write_begin,
3143 .write_end = ext4_write_end,
3145 .invalidatepage = ext4_invalidatepage,
3146 .releasepage = ext4_releasepage,
3147 .direct_IO = ext4_direct_IO,
3148 .migratepage = buffer_migrate_page,
3149 .is_partially_uptodate = block_is_partially_uptodate,
3150 .error_remove_page = generic_error_remove_page,
3153 static const struct address_space_operations ext4_journalled_aops = {
3154 .readpage = ext4_readpage,
3155 .readpages = ext4_readpages,
3156 .writepage = ext4_writepage,
3157 .writepages = ext4_writepages,
3158 .write_begin = ext4_write_begin,
3159 .write_end = ext4_journalled_write_end,
3160 .set_page_dirty = ext4_journalled_set_page_dirty,
3162 .invalidatepage = ext4_journalled_invalidatepage,
3163 .releasepage = ext4_releasepage,
3164 .direct_IO = ext4_direct_IO,
3165 .is_partially_uptodate = block_is_partially_uptodate,
3166 .error_remove_page = generic_error_remove_page,
3169 static const struct address_space_operations ext4_da_aops = {
3170 .readpage = ext4_readpage,
3171 .readpages = ext4_readpages,
3172 .writepage = ext4_writepage,
3173 .writepages = ext4_writepages,
3174 .write_begin = ext4_da_write_begin,
3175 .write_end = ext4_da_write_end,
3177 .invalidatepage = ext4_da_invalidatepage,
3178 .releasepage = ext4_releasepage,
3179 .direct_IO = ext4_direct_IO,
3180 .migratepage = buffer_migrate_page,
3181 .is_partially_uptodate = block_is_partially_uptodate,
3182 .error_remove_page = generic_error_remove_page,
3185 void ext4_set_aops(struct inode *inode)
3187 switch (ext4_inode_journal_mode(inode)) {
3188 case EXT4_INODE_ORDERED_DATA_MODE:
3189 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3191 case EXT4_INODE_WRITEBACK_DATA_MODE:
3192 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3194 case EXT4_INODE_JOURNAL_DATA_MODE:
3195 inode->i_mapping->a_ops = &ext4_journalled_aops;
3200 if (test_opt(inode->i_sb, DELALLOC))
3201 inode->i_mapping->a_ops = &ext4_da_aops;
3203 inode->i_mapping->a_ops = &ext4_aops;
3206 static int __ext4_block_zero_page_range(handle_t *handle,
3207 struct address_space *mapping, loff_t from, loff_t length)
3209 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3210 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3211 unsigned blocksize, pos;
3213 struct inode *inode = mapping->host;
3214 struct buffer_head *bh;
3218 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3219 mapping_gfp_mask(mapping) & ~__GFP_FS);
3223 blocksize = inode->i_sb->s_blocksize;
3225 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3227 if (!page_has_buffers(page))
3228 create_empty_buffers(page, blocksize, 0);
3230 /* Find the buffer that contains "offset" */
3231 bh = page_buffers(page);
3233 while (offset >= pos) {
3234 bh = bh->b_this_page;
3238 if (buffer_freed(bh)) {
3239 BUFFER_TRACE(bh, "freed: skip");
3242 if (!buffer_mapped(bh)) {
3243 BUFFER_TRACE(bh, "unmapped");
3244 ext4_get_block(inode, iblock, bh, 0);
3245 /* unmapped? It's a hole - nothing to do */
3246 if (!buffer_mapped(bh)) {
3247 BUFFER_TRACE(bh, "still unmapped");
3252 /* Ok, it's mapped. Make sure it's up-to-date */
3253 if (PageUptodate(page))
3254 set_buffer_uptodate(bh);
3256 if (!buffer_uptodate(bh)) {
3258 ll_rw_block(READ, 1, &bh);
3260 /* Uhhuh. Read error. Complain and punt. */
3261 if (!buffer_uptodate(bh))
3264 if (ext4_should_journal_data(inode)) {
3265 BUFFER_TRACE(bh, "get write access");
3266 err = ext4_journal_get_write_access(handle, bh);
3270 zero_user(page, offset, length);
3271 BUFFER_TRACE(bh, "zeroed end of block");
3273 if (ext4_should_journal_data(inode)) {
3274 err = ext4_handle_dirty_metadata(handle, inode, bh);
3277 mark_buffer_dirty(bh);
3278 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3279 err = ext4_jbd2_file_inode(handle, inode);
3284 page_cache_release(page);
3289 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3290 * starting from file offset 'from'. The range to be zero'd must
3291 * be contained with in one block. If the specified range exceeds
3292 * the end of the block it will be shortened to end of the block
3293 * that cooresponds to 'from'
3295 static int ext4_block_zero_page_range(handle_t *handle,
3296 struct address_space *mapping, loff_t from, loff_t length)
3298 struct inode *inode = mapping->host;
3299 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3300 unsigned blocksize = inode->i_sb->s_blocksize;
3301 unsigned max = blocksize - (offset & (blocksize - 1));
3304 * correct length if it does not fall between
3305 * 'from' and the end of the block
3307 if (length > max || length < 0)
3311 return dax_zero_page_range(inode, from, length, ext4_get_block);
3312 return __ext4_block_zero_page_range(handle, mapping, from, length);
3316 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3317 * up to the end of the block which corresponds to `from'.
3318 * This required during truncate. We need to physically zero the tail end
3319 * of that block so it doesn't yield old data if the file is later grown.
3321 static int ext4_block_truncate_page(handle_t *handle,
3322 struct address_space *mapping, loff_t from)
3324 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3327 struct inode *inode = mapping->host;
3329 blocksize = inode->i_sb->s_blocksize;
3330 length = blocksize - (offset & (blocksize - 1));
3332 return ext4_block_zero_page_range(handle, mapping, from, length);
3335 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3336 loff_t lstart, loff_t length)
3338 struct super_block *sb = inode->i_sb;
3339 struct address_space *mapping = inode->i_mapping;
3340 unsigned partial_start, partial_end;
3341 ext4_fsblk_t start, end;
3342 loff_t byte_end = (lstart + length - 1);
3345 partial_start = lstart & (sb->s_blocksize - 1);
3346 partial_end = byte_end & (sb->s_blocksize - 1);
3348 start = lstart >> sb->s_blocksize_bits;
3349 end = byte_end >> sb->s_blocksize_bits;
3351 /* Handle partial zero within the single block */
3353 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3354 err = ext4_block_zero_page_range(handle, mapping,
3358 /* Handle partial zero out on the start of the range */
3359 if (partial_start) {
3360 err = ext4_block_zero_page_range(handle, mapping,
3361 lstart, sb->s_blocksize);
3365 /* Handle partial zero out on the end of the range */
3366 if (partial_end != sb->s_blocksize - 1)
3367 err = ext4_block_zero_page_range(handle, mapping,
3368 byte_end - partial_end,
3373 int ext4_can_truncate(struct inode *inode)
3375 if (S_ISREG(inode->i_mode))
3377 if (S_ISDIR(inode->i_mode))
3379 if (S_ISLNK(inode->i_mode))
3380 return !ext4_inode_is_fast_symlink(inode);
3385 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3386 * associated with the given offset and length
3388 * @inode: File inode
3389 * @offset: The offset where the hole will begin
3390 * @len: The length of the hole
3392 * Returns: 0 on success or negative on failure
3395 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3397 struct super_block *sb = inode->i_sb;
3398 ext4_lblk_t first_block, stop_block;
3399 struct address_space *mapping = inode->i_mapping;
3400 loff_t first_block_offset, last_block_offset;
3402 unsigned int credits;
3405 if (!S_ISREG(inode->i_mode))
3408 trace_ext4_punch_hole(inode, offset, length, 0);
3411 * Write out all dirty pages to avoid race conditions
3412 * Then release them.
3414 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3415 ret = filemap_write_and_wait_range(mapping, offset,
3416 offset + length - 1);
3421 mutex_lock(&inode->i_mutex);
3423 /* No need to punch hole beyond i_size */
3424 if (offset >= inode->i_size)
3428 * If the hole extends beyond i_size, set the hole
3429 * to end after the page that contains i_size
3431 if (offset + length > inode->i_size) {
3432 length = inode->i_size +
3433 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3437 if (offset & (sb->s_blocksize - 1) ||
3438 (offset + length) & (sb->s_blocksize - 1)) {
3440 * Attach jinode to inode for jbd2 if we do any zeroing of
3443 ret = ext4_inode_attach_jinode(inode);
3449 first_block_offset = round_up(offset, sb->s_blocksize);
3450 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3452 /* Now release the pages and zero block aligned part of pages*/
3453 if (last_block_offset > first_block_offset)
3454 truncate_pagecache_range(inode, first_block_offset,
3457 /* Wait all existing dio workers, newcomers will block on i_mutex */
3458 ext4_inode_block_unlocked_dio(inode);
3459 inode_dio_wait(inode);
3461 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3462 credits = ext4_writepage_trans_blocks(inode);
3464 credits = ext4_blocks_for_truncate(inode);
3465 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3466 if (IS_ERR(handle)) {
3467 ret = PTR_ERR(handle);
3468 ext4_std_error(sb, ret);
3472 ret = ext4_zero_partial_blocks(handle, inode, offset,
3477 first_block = (offset + sb->s_blocksize - 1) >>
3478 EXT4_BLOCK_SIZE_BITS(sb);
3479 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3481 /* If there are no blocks to remove, return now */
3482 if (first_block >= stop_block)
3485 down_write(&EXT4_I(inode)->i_data_sem);
3486 ext4_discard_preallocations(inode);
3488 ret = ext4_es_remove_extent(inode, first_block,
3489 stop_block - first_block);
3491 up_write(&EXT4_I(inode)->i_data_sem);
3495 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3496 ret = ext4_ext_remove_space(inode, first_block,
3499 ret = ext4_ind_remove_space(handle, inode, first_block,
3502 up_write(&EXT4_I(inode)->i_data_sem);
3504 ext4_handle_sync(handle);
3506 /* Now release the pages again to reduce race window */
3507 if (last_block_offset > first_block_offset)
3508 truncate_pagecache_range(inode, first_block_offset,
3511 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3512 ext4_mark_inode_dirty(handle, inode);
3514 ext4_journal_stop(handle);
3516 ext4_inode_resume_unlocked_dio(inode);
3518 mutex_unlock(&inode->i_mutex);
3522 int ext4_inode_attach_jinode(struct inode *inode)
3524 struct ext4_inode_info *ei = EXT4_I(inode);
3525 struct jbd2_inode *jinode;
3527 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3530 jinode = jbd2_alloc_inode(GFP_KERNEL);
3531 spin_lock(&inode->i_lock);
3534 spin_unlock(&inode->i_lock);
3537 ei->jinode = jinode;
3538 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3541 spin_unlock(&inode->i_lock);
3542 if (unlikely(jinode != NULL))
3543 jbd2_free_inode(jinode);
3550 * We block out ext4_get_block() block instantiations across the entire
3551 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3552 * simultaneously on behalf of the same inode.
3554 * As we work through the truncate and commit bits of it to the journal there
3555 * is one core, guiding principle: the file's tree must always be consistent on
3556 * disk. We must be able to restart the truncate after a crash.
3558 * The file's tree may be transiently inconsistent in memory (although it
3559 * probably isn't), but whenever we close off and commit a journal transaction,
3560 * the contents of (the filesystem + the journal) must be consistent and
3561 * restartable. It's pretty simple, really: bottom up, right to left (although
3562 * left-to-right works OK too).
3564 * Note that at recovery time, journal replay occurs *before* the restart of
3565 * truncate against the orphan inode list.
3567 * The committed inode has the new, desired i_size (which is the same as
3568 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3569 * that this inode's truncate did not complete and it will again call
3570 * ext4_truncate() to have another go. So there will be instantiated blocks
3571 * to the right of the truncation point in a crashed ext4 filesystem. But
3572 * that's fine - as long as they are linked from the inode, the post-crash
3573 * ext4_truncate() run will find them and release them.
3575 void ext4_truncate(struct inode *inode)
3577 struct ext4_inode_info *ei = EXT4_I(inode);
3578 unsigned int credits;
3580 struct address_space *mapping = inode->i_mapping;
3583 * There is a possibility that we're either freeing the inode
3584 * or it's a completely new inode. In those cases we might not
3585 * have i_mutex locked because it's not necessary.
3587 if (!(inode->i_state & (I_NEW|I_FREEING)))
3588 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3589 trace_ext4_truncate_enter(inode);
3591 if (!ext4_can_truncate(inode))
3594 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3596 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3597 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3599 if (ext4_has_inline_data(inode)) {
3602 ext4_inline_data_truncate(inode, &has_inline);
3607 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3608 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3609 if (ext4_inode_attach_jinode(inode) < 0)
3613 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3614 credits = ext4_writepage_trans_blocks(inode);
3616 credits = ext4_blocks_for_truncate(inode);
3618 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3619 if (IS_ERR(handle)) {
3620 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3624 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3625 ext4_block_truncate_page(handle, mapping, inode->i_size);
3628 * We add the inode to the orphan list, so that if this
3629 * truncate spans multiple transactions, and we crash, we will
3630 * resume the truncate when the filesystem recovers. It also
3631 * marks the inode dirty, to catch the new size.
3633 * Implication: the file must always be in a sane, consistent
3634 * truncatable state while each transaction commits.
3636 if (ext4_orphan_add(handle, inode))
3639 down_write(&EXT4_I(inode)->i_data_sem);
3641 ext4_discard_preallocations(inode);
3643 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3644 ext4_ext_truncate(handle, inode);
3646 ext4_ind_truncate(handle, inode);
3648 up_write(&ei->i_data_sem);
3651 ext4_handle_sync(handle);
3655 * If this was a simple ftruncate() and the file will remain alive,
3656 * then we need to clear up the orphan record which we created above.
3657 * However, if this was a real unlink then we were called by
3658 * ext4_evict_inode(), and we allow that function to clean up the
3659 * orphan info for us.
3662 ext4_orphan_del(handle, inode);
3664 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3665 ext4_mark_inode_dirty(handle, inode);
3666 ext4_journal_stop(handle);
3668 trace_ext4_truncate_exit(inode);
3672 * ext4_get_inode_loc returns with an extra refcount against the inode's
3673 * underlying buffer_head on success. If 'in_mem' is true, we have all
3674 * data in memory that is needed to recreate the on-disk version of this
3677 static int __ext4_get_inode_loc(struct inode *inode,
3678 struct ext4_iloc *iloc, int in_mem)
3680 struct ext4_group_desc *gdp;
3681 struct buffer_head *bh;
3682 struct super_block *sb = inode->i_sb;
3684 int inodes_per_block, inode_offset;
3687 if (!ext4_valid_inum(sb, inode->i_ino))
3690 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3691 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3696 * Figure out the offset within the block group inode table
3698 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3699 inode_offset = ((inode->i_ino - 1) %
3700 EXT4_INODES_PER_GROUP(sb));
3701 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3702 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3704 bh = sb_getblk(sb, block);
3707 if (!buffer_uptodate(bh)) {
3711 * If the buffer has the write error flag, we have failed
3712 * to write out another inode in the same block. In this
3713 * case, we don't have to read the block because we may
3714 * read the old inode data successfully.
3716 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3717 set_buffer_uptodate(bh);
3719 if (buffer_uptodate(bh)) {
3720 /* someone brought it uptodate while we waited */
3726 * If we have all information of the inode in memory and this
3727 * is the only valid inode in the block, we need not read the
3731 struct buffer_head *bitmap_bh;
3734 start = inode_offset & ~(inodes_per_block - 1);
3736 /* Is the inode bitmap in cache? */
3737 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3738 if (unlikely(!bitmap_bh))
3742 * If the inode bitmap isn't in cache then the
3743 * optimisation may end up performing two reads instead
3744 * of one, so skip it.
3746 if (!buffer_uptodate(bitmap_bh)) {
3750 for (i = start; i < start + inodes_per_block; i++) {
3751 if (i == inode_offset)
3753 if (ext4_test_bit(i, bitmap_bh->b_data))
3757 if (i == start + inodes_per_block) {
3758 /* all other inodes are free, so skip I/O */
3759 memset(bh->b_data, 0, bh->b_size);
3760 set_buffer_uptodate(bh);
3768 * If we need to do any I/O, try to pre-readahead extra
3769 * blocks from the inode table.
3771 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3772 ext4_fsblk_t b, end, table;
3774 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
3776 table = ext4_inode_table(sb, gdp);
3777 /* s_inode_readahead_blks is always a power of 2 */
3778 b = block & ~((ext4_fsblk_t) ra_blks - 1);
3782 num = EXT4_INODES_PER_GROUP(sb);
3783 if (ext4_has_group_desc_csum(sb))
3784 num -= ext4_itable_unused_count(sb, gdp);
3785 table += num / inodes_per_block;
3789 sb_breadahead(sb, b++);
3793 * There are other valid inodes in the buffer, this inode
3794 * has in-inode xattrs, or we don't have this inode in memory.
3795 * Read the block from disk.
3797 trace_ext4_load_inode(inode);
3799 bh->b_end_io = end_buffer_read_sync;
3800 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3802 if (!buffer_uptodate(bh)) {
3803 EXT4_ERROR_INODE_BLOCK(inode, block,
3804 "unable to read itable block");
3814 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3816 /* We have all inode data except xattrs in memory here. */
3817 return __ext4_get_inode_loc(inode, iloc,
3818 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3821 void ext4_set_inode_flags(struct inode *inode)
3823 unsigned int flags = EXT4_I(inode)->i_flags;
3824 unsigned int new_fl = 0;
3826 if (flags & EXT4_SYNC_FL)
3828 if (flags & EXT4_APPEND_FL)
3830 if (flags & EXT4_IMMUTABLE_FL)
3831 new_fl |= S_IMMUTABLE;
3832 if (flags & EXT4_NOATIME_FL)
3833 new_fl |= S_NOATIME;
3834 if (flags & EXT4_DIRSYNC_FL)
3835 new_fl |= S_DIRSYNC;
3836 if (test_opt(inode->i_sb, DAX))
3838 inode_set_flags(inode, new_fl,
3839 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
3842 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3843 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3845 unsigned int vfs_fl;
3846 unsigned long old_fl, new_fl;
3849 vfs_fl = ei->vfs_inode.i_flags;
3850 old_fl = ei->i_flags;
3851 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3852 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3854 if (vfs_fl & S_SYNC)
3855 new_fl |= EXT4_SYNC_FL;
3856 if (vfs_fl & S_APPEND)
3857 new_fl |= EXT4_APPEND_FL;
3858 if (vfs_fl & S_IMMUTABLE)
3859 new_fl |= EXT4_IMMUTABLE_FL;
3860 if (vfs_fl & S_NOATIME)
3861 new_fl |= EXT4_NOATIME_FL;
3862 if (vfs_fl & S_DIRSYNC)
3863 new_fl |= EXT4_DIRSYNC_FL;
3864 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3867 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3868 struct ext4_inode_info *ei)
3871 struct inode *inode = &(ei->vfs_inode);
3872 struct super_block *sb = inode->i_sb;
3874 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3875 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3876 /* we are using combined 48 bit field */
3877 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3878 le32_to_cpu(raw_inode->i_blocks_lo);
3879 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3880 /* i_blocks represent file system block size */
3881 return i_blocks << (inode->i_blkbits - 9);
3886 return le32_to_cpu(raw_inode->i_blocks_lo);
3890 static inline void ext4_iget_extra_inode(struct inode *inode,
3891 struct ext4_inode *raw_inode,
3892 struct ext4_inode_info *ei)
3894 __le32 *magic = (void *)raw_inode +
3895 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3896 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3897 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3898 ext4_find_inline_data_nolock(inode);
3900 EXT4_I(inode)->i_inline_off = 0;
3903 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3905 struct ext4_iloc iloc;
3906 struct ext4_inode *raw_inode;
3907 struct ext4_inode_info *ei;
3908 struct inode *inode;
3909 journal_t *journal = EXT4_SB(sb)->s_journal;
3915 inode = iget_locked(sb, ino);
3917 return ERR_PTR(-ENOMEM);
3918 if (!(inode->i_state & I_NEW))
3924 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3927 raw_inode = ext4_raw_inode(&iloc);
3929 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3930 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3931 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3932 EXT4_INODE_SIZE(inode->i_sb)) {
3933 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3934 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3935 EXT4_INODE_SIZE(inode->i_sb));
3940 ei->i_extra_isize = 0;
3942 /* Precompute checksum seed for inode metadata */
3943 if (ext4_has_metadata_csum(sb)) {
3944 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3946 __le32 inum = cpu_to_le32(inode->i_ino);
3947 __le32 gen = raw_inode->i_generation;
3948 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3950 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3954 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3955 EXT4_ERROR_INODE(inode, "checksum invalid");
3960 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3961 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3962 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3963 if (!(test_opt(inode->i_sb, NO_UID32))) {
3964 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3965 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3967 i_uid_write(inode, i_uid);
3968 i_gid_write(inode, i_gid);
3969 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3971 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3972 ei->i_inline_off = 0;
3973 ei->i_dir_start_lookup = 0;
3974 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3975 /* We now have enough fields to check if the inode was active or not.
3976 * This is needed because nfsd might try to access dead inodes
3977 * the test is that same one that e2fsck uses
3978 * NeilBrown 1999oct15
3980 if (inode->i_nlink == 0) {
3981 if ((inode->i_mode == 0 ||
3982 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
3983 ino != EXT4_BOOT_LOADER_INO) {
3984 /* this inode is deleted */
3988 /* The only unlinked inodes we let through here have
3989 * valid i_mode and are being read by the orphan
3990 * recovery code: that's fine, we're about to complete
3991 * the process of deleting those.
3992 * OR it is the EXT4_BOOT_LOADER_INO which is
3993 * not initialized on a new filesystem. */
3995 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3996 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3997 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3998 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4000 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4001 inode->i_size = ext4_isize(raw_inode);
4002 ei->i_disksize = inode->i_size;
4004 ei->i_reserved_quota = 0;
4006 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4007 ei->i_block_group = iloc.block_group;
4008 ei->i_last_alloc_group = ~0;
4010 * NOTE! The in-memory inode i_data array is in little-endian order
4011 * even on big-endian machines: we do NOT byteswap the block numbers!
4013 for (block = 0; block < EXT4_N_BLOCKS; block++)
4014 ei->i_data[block] = raw_inode->i_block[block];
4015 INIT_LIST_HEAD(&ei->i_orphan);
4018 * Set transaction id's of transactions that have to be committed
4019 * to finish f[data]sync. We set them to currently running transaction
4020 * as we cannot be sure that the inode or some of its metadata isn't
4021 * part of the transaction - the inode could have been reclaimed and
4022 * now it is reread from disk.
4025 transaction_t *transaction;
4028 read_lock(&journal->j_state_lock);
4029 if (journal->j_running_transaction)
4030 transaction = journal->j_running_transaction;
4032 transaction = journal->j_committing_transaction;
4034 tid = transaction->t_tid;
4036 tid = journal->j_commit_sequence;
4037 read_unlock(&journal->j_state_lock);
4038 ei->i_sync_tid = tid;
4039 ei->i_datasync_tid = tid;
4042 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4043 if (ei->i_extra_isize == 0) {
4044 /* The extra space is currently unused. Use it. */
4045 ei->i_extra_isize = sizeof(struct ext4_inode) -
4046 EXT4_GOOD_OLD_INODE_SIZE;
4048 ext4_iget_extra_inode(inode, raw_inode, ei);
4052 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4053 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4054 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4055 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4057 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4058 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4059 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4060 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4062 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4067 if (ei->i_file_acl &&
4068 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4069 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4073 } else if (!ext4_has_inline_data(inode)) {
4074 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4075 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4076 (S_ISLNK(inode->i_mode) &&
4077 !ext4_inode_is_fast_symlink(inode))))
4078 /* Validate extent which is part of inode */
4079 ret = ext4_ext_check_inode(inode);
4080 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4081 (S_ISLNK(inode->i_mode) &&
4082 !ext4_inode_is_fast_symlink(inode))) {
4083 /* Validate block references which are part of inode */
4084 ret = ext4_ind_check_inode(inode);
4090 if (S_ISREG(inode->i_mode)) {
4091 inode->i_op = &ext4_file_inode_operations;
4092 if (test_opt(inode->i_sb, DAX))
4093 inode->i_fop = &ext4_dax_file_operations;
4095 inode->i_fop = &ext4_file_operations;
4096 ext4_set_aops(inode);
4097 } else if (S_ISDIR(inode->i_mode)) {
4098 inode->i_op = &ext4_dir_inode_operations;
4099 inode->i_fop = &ext4_dir_operations;
4100 } else if (S_ISLNK(inode->i_mode)) {
4101 if (ext4_inode_is_fast_symlink(inode)) {
4102 inode->i_op = &ext4_fast_symlink_inode_operations;
4103 nd_terminate_link(ei->i_data, inode->i_size,
4104 sizeof(ei->i_data) - 1);
4106 inode->i_op = &ext4_symlink_inode_operations;
4107 ext4_set_aops(inode);
4109 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4110 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4111 inode->i_op = &ext4_special_inode_operations;
4112 if (raw_inode->i_block[0])
4113 init_special_inode(inode, inode->i_mode,
4114 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4116 init_special_inode(inode, inode->i_mode,
4117 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4118 } else if (ino == EXT4_BOOT_LOADER_INO) {
4119 make_bad_inode(inode);
4122 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4126 ext4_set_inode_flags(inode);
4127 unlock_new_inode(inode);
4133 return ERR_PTR(ret);
4136 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4138 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4139 return ERR_PTR(-EIO);
4140 return ext4_iget(sb, ino);
4143 static int ext4_inode_blocks_set(handle_t *handle,
4144 struct ext4_inode *raw_inode,
4145 struct ext4_inode_info *ei)
4147 struct inode *inode = &(ei->vfs_inode);
4148 u64 i_blocks = inode->i_blocks;
4149 struct super_block *sb = inode->i_sb;
4151 if (i_blocks <= ~0U) {
4153 * i_blocks can be represented in a 32 bit variable
4154 * as multiple of 512 bytes
4156 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4157 raw_inode->i_blocks_high = 0;
4158 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4161 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4164 if (i_blocks <= 0xffffffffffffULL) {
4166 * i_blocks can be represented in a 48 bit variable
4167 * as multiple of 512 bytes
4169 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4170 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4171 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4173 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4174 /* i_block is stored in file system block size */
4175 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4176 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4177 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4182 struct other_inode {
4183 unsigned long orig_ino;
4184 struct ext4_inode *raw_inode;
4187 static int other_inode_match(struct inode * inode, unsigned long ino,
4190 struct other_inode *oi = (struct other_inode *) data;
4192 if ((inode->i_ino != ino) ||
4193 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4194 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4195 ((inode->i_state & I_DIRTY_TIME) == 0))
4197 spin_lock(&inode->i_lock);
4198 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4199 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4200 (inode->i_state & I_DIRTY_TIME)) {
4201 struct ext4_inode_info *ei = EXT4_I(inode);
4203 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4204 spin_unlock(&inode->i_lock);
4206 spin_lock(&ei->i_raw_lock);
4207 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4208 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4209 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4210 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4211 spin_unlock(&ei->i_raw_lock);
4212 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4215 spin_unlock(&inode->i_lock);
4220 * Opportunistically update the other time fields for other inodes in
4221 * the same inode table block.
4223 static void ext4_update_other_inodes_time(struct super_block *sb,
4224 unsigned long orig_ino, char *buf)
4226 struct other_inode oi;
4228 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4229 int inode_size = EXT4_INODE_SIZE(sb);
4231 oi.orig_ino = orig_ino;
4232 ino = orig_ino & ~(inodes_per_block - 1);
4233 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4234 if (ino == orig_ino)
4236 oi.raw_inode = (struct ext4_inode *) buf;
4237 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4242 * Post the struct inode info into an on-disk inode location in the
4243 * buffer-cache. This gobbles the caller's reference to the
4244 * buffer_head in the inode location struct.
4246 * The caller must have write access to iloc->bh.
4248 static int ext4_do_update_inode(handle_t *handle,
4249 struct inode *inode,
4250 struct ext4_iloc *iloc)
4252 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4253 struct ext4_inode_info *ei = EXT4_I(inode);
4254 struct buffer_head *bh = iloc->bh;
4255 struct super_block *sb = inode->i_sb;
4256 int err = 0, rc, block;
4257 int need_datasync = 0, set_large_file = 0;
4261 spin_lock(&ei->i_raw_lock);
4263 /* For fields not tracked in the in-memory inode,
4264 * initialise them to zero for new inodes. */
4265 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4266 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4268 ext4_get_inode_flags(ei);
4269 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4270 i_uid = i_uid_read(inode);
4271 i_gid = i_gid_read(inode);
4272 if (!(test_opt(inode->i_sb, NO_UID32))) {
4273 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4274 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4276 * Fix up interoperability with old kernels. Otherwise, old inodes get
4277 * re-used with the upper 16 bits of the uid/gid intact
4280 raw_inode->i_uid_high =
4281 cpu_to_le16(high_16_bits(i_uid));
4282 raw_inode->i_gid_high =
4283 cpu_to_le16(high_16_bits(i_gid));
4285 raw_inode->i_uid_high = 0;
4286 raw_inode->i_gid_high = 0;
4289 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4290 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4291 raw_inode->i_uid_high = 0;
4292 raw_inode->i_gid_high = 0;
4294 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4296 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4297 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4298 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4299 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4301 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4303 spin_unlock(&ei->i_raw_lock);
4306 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4307 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4308 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4309 raw_inode->i_file_acl_high =
4310 cpu_to_le16(ei->i_file_acl >> 32);
4311 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4312 if (ei->i_disksize != ext4_isize(raw_inode)) {
4313 ext4_isize_set(raw_inode, ei->i_disksize);
4316 if (ei->i_disksize > 0x7fffffffULL) {
4317 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4318 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4319 EXT4_SB(sb)->s_es->s_rev_level ==
4320 cpu_to_le32(EXT4_GOOD_OLD_REV))
4323 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4324 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4325 if (old_valid_dev(inode->i_rdev)) {
4326 raw_inode->i_block[0] =
4327 cpu_to_le32(old_encode_dev(inode->i_rdev));
4328 raw_inode->i_block[1] = 0;
4330 raw_inode->i_block[0] = 0;
4331 raw_inode->i_block[1] =
4332 cpu_to_le32(new_encode_dev(inode->i_rdev));
4333 raw_inode->i_block[2] = 0;
4335 } else if (!ext4_has_inline_data(inode)) {
4336 for (block = 0; block < EXT4_N_BLOCKS; block++)
4337 raw_inode->i_block[block] = ei->i_data[block];
4340 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4341 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4342 if (ei->i_extra_isize) {
4343 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4344 raw_inode->i_version_hi =
4345 cpu_to_le32(inode->i_version >> 32);
4346 raw_inode->i_extra_isize =
4347 cpu_to_le16(ei->i_extra_isize);
4350 ext4_inode_csum_set(inode, raw_inode, ei);
4351 spin_unlock(&ei->i_raw_lock);
4352 if (inode->i_sb->s_flags & MS_LAZYTIME)
4353 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4356 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4357 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4360 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4361 if (set_large_file) {
4362 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4363 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4366 ext4_update_dynamic_rev(sb);
4367 EXT4_SET_RO_COMPAT_FEATURE(sb,
4368 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4369 ext4_handle_sync(handle);
4370 err = ext4_handle_dirty_super(handle, sb);
4372 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4375 ext4_std_error(inode->i_sb, err);
4380 * ext4_write_inode()
4382 * We are called from a few places:
4384 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4385 * Here, there will be no transaction running. We wait for any running
4386 * transaction to commit.
4388 * - Within flush work (sys_sync(), kupdate and such).
4389 * We wait on commit, if told to.
4391 * - Within iput_final() -> write_inode_now()
4392 * We wait on commit, if told to.
4394 * In all cases it is actually safe for us to return without doing anything,
4395 * because the inode has been copied into a raw inode buffer in
4396 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4399 * Note that we are absolutely dependent upon all inode dirtiers doing the
4400 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4401 * which we are interested.
4403 * It would be a bug for them to not do this. The code:
4405 * mark_inode_dirty(inode)
4407 * inode->i_size = expr;
4409 * is in error because write_inode() could occur while `stuff()' is running,
4410 * and the new i_size will be lost. Plus the inode will no longer be on the
4411 * superblock's dirty inode list.
4413 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4417 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4420 if (EXT4_SB(inode->i_sb)->s_journal) {
4421 if (ext4_journal_current_handle()) {
4422 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4428 * No need to force transaction in WB_SYNC_NONE mode. Also
4429 * ext4_sync_fs() will force the commit after everything is
4432 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4435 err = ext4_force_commit(inode->i_sb);
4437 struct ext4_iloc iloc;
4439 err = __ext4_get_inode_loc(inode, &iloc, 0);
4443 * sync(2) will flush the whole buffer cache. No need to do
4444 * it here separately for each inode.
4446 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4447 sync_dirty_buffer(iloc.bh);
4448 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4449 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4450 "IO error syncing inode");
4459 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4460 * buffers that are attached to a page stradding i_size and are undergoing
4461 * commit. In that case we have to wait for commit to finish and try again.
4463 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4467 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4468 tid_t commit_tid = 0;
4471 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4473 * All buffers in the last page remain valid? Then there's nothing to
4474 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4477 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4480 page = find_lock_page(inode->i_mapping,
4481 inode->i_size >> PAGE_CACHE_SHIFT);
4484 ret = __ext4_journalled_invalidatepage(page, offset,
4485 PAGE_CACHE_SIZE - offset);
4487 page_cache_release(page);
4491 read_lock(&journal->j_state_lock);
4492 if (journal->j_committing_transaction)
4493 commit_tid = journal->j_committing_transaction->t_tid;
4494 read_unlock(&journal->j_state_lock);
4496 jbd2_log_wait_commit(journal, commit_tid);
4503 * Called from notify_change.
4505 * We want to trap VFS attempts to truncate the file as soon as
4506 * possible. In particular, we want to make sure that when the VFS
4507 * shrinks i_size, we put the inode on the orphan list and modify
4508 * i_disksize immediately, so that during the subsequent flushing of
4509 * dirty pages and freeing of disk blocks, we can guarantee that any
4510 * commit will leave the blocks being flushed in an unused state on
4511 * disk. (On recovery, the inode will get truncated and the blocks will
4512 * be freed, so we have a strong guarantee that no future commit will
4513 * leave these blocks visible to the user.)
4515 * Another thing we have to assure is that if we are in ordered mode
4516 * and inode is still attached to the committing transaction, we must
4517 * we start writeout of all the dirty pages which are being truncated.
4518 * This way we are sure that all the data written in the previous
4519 * transaction are already on disk (truncate waits for pages under
4522 * Called with inode->i_mutex down.
4524 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4526 struct inode *inode = dentry->d_inode;
4529 const unsigned int ia_valid = attr->ia_valid;
4531 error = inode_change_ok(inode, attr);
4535 if (is_quota_modification(inode, attr))
4536 dquot_initialize(inode);
4537 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4538 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4541 /* (user+group)*(old+new) structure, inode write (sb,
4542 * inode block, ? - but truncate inode update has it) */
4543 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4544 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4545 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4546 if (IS_ERR(handle)) {
4547 error = PTR_ERR(handle);
4550 error = dquot_transfer(inode, attr);
4552 ext4_journal_stop(handle);
4555 /* Update corresponding info in inode so that everything is in
4556 * one transaction */
4557 if (attr->ia_valid & ATTR_UID)
4558 inode->i_uid = attr->ia_uid;
4559 if (attr->ia_valid & ATTR_GID)
4560 inode->i_gid = attr->ia_gid;
4561 error = ext4_mark_inode_dirty(handle, inode);
4562 ext4_journal_stop(handle);
4565 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4568 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4569 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4571 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4575 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4576 inode_inc_iversion(inode);
4578 if (S_ISREG(inode->i_mode) &&
4579 (attr->ia_size < inode->i_size)) {
4580 if (ext4_should_order_data(inode)) {
4581 error = ext4_begin_ordered_truncate(inode,
4586 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4587 if (IS_ERR(handle)) {
4588 error = PTR_ERR(handle);
4591 if (ext4_handle_valid(handle)) {
4592 error = ext4_orphan_add(handle, inode);
4595 down_write(&EXT4_I(inode)->i_data_sem);
4596 EXT4_I(inode)->i_disksize = attr->ia_size;
4597 rc = ext4_mark_inode_dirty(handle, inode);
4601 * We have to update i_size under i_data_sem together
4602 * with i_disksize to avoid races with writeback code
4603 * running ext4_wb_update_i_disksize().
4606 i_size_write(inode, attr->ia_size);
4607 up_write(&EXT4_I(inode)->i_data_sem);
4608 ext4_journal_stop(handle);
4610 ext4_orphan_del(NULL, inode);
4614 loff_t oldsize = inode->i_size;
4616 i_size_write(inode, attr->ia_size);
4617 pagecache_isize_extended(inode, oldsize, inode->i_size);
4621 * Blocks are going to be removed from the inode. Wait
4622 * for dio in flight. Temporarily disable
4623 * dioread_nolock to prevent livelock.
4626 if (!ext4_should_journal_data(inode)) {
4627 ext4_inode_block_unlocked_dio(inode);
4628 inode_dio_wait(inode);
4629 ext4_inode_resume_unlocked_dio(inode);
4631 ext4_wait_for_tail_page_commit(inode);
4634 * Truncate pagecache after we've waited for commit
4635 * in data=journal mode to make pages freeable.
4637 truncate_pagecache(inode, inode->i_size);
4640 * We want to call ext4_truncate() even if attr->ia_size ==
4641 * inode->i_size for cases like truncation of fallocated space
4643 if (attr->ia_valid & ATTR_SIZE)
4644 ext4_truncate(inode);
4647 setattr_copy(inode, attr);
4648 mark_inode_dirty(inode);
4652 * If the call to ext4_truncate failed to get a transaction handle at
4653 * all, we need to clean up the in-core orphan list manually.
4655 if (orphan && inode->i_nlink)
4656 ext4_orphan_del(NULL, inode);
4658 if (!rc && (ia_valid & ATTR_MODE))
4659 rc = posix_acl_chmod(inode, inode->i_mode);
4662 ext4_std_error(inode->i_sb, error);
4668 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4671 struct inode *inode;
4672 unsigned long long delalloc_blocks;
4674 inode = dentry->d_inode;
4675 generic_fillattr(inode, stat);
4678 * If there is inline data in the inode, the inode will normally not
4679 * have data blocks allocated (it may have an external xattr block).
4680 * Report at least one sector for such files, so tools like tar, rsync,
4681 * others doen't incorrectly think the file is completely sparse.
4683 if (unlikely(ext4_has_inline_data(inode)))
4684 stat->blocks += (stat->size + 511) >> 9;
4687 * We can't update i_blocks if the block allocation is delayed
4688 * otherwise in the case of system crash before the real block
4689 * allocation is done, we will have i_blocks inconsistent with
4690 * on-disk file blocks.
4691 * We always keep i_blocks updated together with real
4692 * allocation. But to not confuse with user, stat
4693 * will return the blocks that include the delayed allocation
4694 * blocks for this file.
4696 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4697 EXT4_I(inode)->i_reserved_data_blocks);
4698 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
4702 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4705 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4706 return ext4_ind_trans_blocks(inode, lblocks);
4707 return ext4_ext_index_trans_blocks(inode, pextents);
4711 * Account for index blocks, block groups bitmaps and block group
4712 * descriptor blocks if modify datablocks and index blocks
4713 * worse case, the indexs blocks spread over different block groups
4715 * If datablocks are discontiguous, they are possible to spread over
4716 * different block groups too. If they are contiguous, with flexbg,
4717 * they could still across block group boundary.
4719 * Also account for superblock, inode, quota and xattr blocks
4721 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4724 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4730 * How many index blocks need to touch to map @lblocks logical blocks
4731 * to @pextents physical extents?
4733 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
4738 * Now let's see how many group bitmaps and group descriptors need
4741 groups = idxblocks + pextents;
4743 if (groups > ngroups)
4745 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4746 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4748 /* bitmaps and block group descriptor blocks */
4749 ret += groups + gdpblocks;
4751 /* Blocks for super block, inode, quota and xattr blocks */
4752 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4758 * Calculate the total number of credits to reserve to fit
4759 * the modification of a single pages into a single transaction,
4760 * which may include multiple chunks of block allocations.
4762 * This could be called via ext4_write_begin()
4764 * We need to consider the worse case, when
4765 * one new block per extent.
4767 int ext4_writepage_trans_blocks(struct inode *inode)
4769 int bpp = ext4_journal_blocks_per_page(inode);
4772 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
4774 /* Account for data blocks for journalled mode */
4775 if (ext4_should_journal_data(inode))
4781 * Calculate the journal credits for a chunk of data modification.
4783 * This is called from DIO, fallocate or whoever calling
4784 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4786 * journal buffers for data blocks are not included here, as DIO
4787 * and fallocate do no need to journal data buffers.
4789 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4791 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4795 * The caller must have previously called ext4_reserve_inode_write().
4796 * Give this, we know that the caller already has write access to iloc->bh.
4798 int ext4_mark_iloc_dirty(handle_t *handle,
4799 struct inode *inode, struct ext4_iloc *iloc)
4803 if (IS_I_VERSION(inode))
4804 inode_inc_iversion(inode);
4806 /* the do_update_inode consumes one bh->b_count */
4809 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4810 err = ext4_do_update_inode(handle, inode, iloc);
4816 * On success, We end up with an outstanding reference count against
4817 * iloc->bh. This _must_ be cleaned up later.
4821 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4822 struct ext4_iloc *iloc)
4826 err = ext4_get_inode_loc(inode, iloc);
4828 BUFFER_TRACE(iloc->bh, "get_write_access");
4829 err = ext4_journal_get_write_access(handle, iloc->bh);
4835 ext4_std_error(inode->i_sb, err);
4840 * Expand an inode by new_extra_isize bytes.
4841 * Returns 0 on success or negative error number on failure.
4843 static int ext4_expand_extra_isize(struct inode *inode,
4844 unsigned int new_extra_isize,
4845 struct ext4_iloc iloc,
4848 struct ext4_inode *raw_inode;
4849 struct ext4_xattr_ibody_header *header;
4851 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4854 raw_inode = ext4_raw_inode(&iloc);
4856 header = IHDR(inode, raw_inode);
4858 /* No extended attributes present */
4859 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4860 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4861 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4863 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4867 /* try to expand with EAs present */
4868 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4873 * What we do here is to mark the in-core inode as clean with respect to inode
4874 * dirtiness (it may still be data-dirty).
4875 * This means that the in-core inode may be reaped by prune_icache
4876 * without having to perform any I/O. This is a very good thing,
4877 * because *any* task may call prune_icache - even ones which
4878 * have a transaction open against a different journal.
4880 * Is this cheating? Not really. Sure, we haven't written the
4881 * inode out, but prune_icache isn't a user-visible syncing function.
4882 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4883 * we start and wait on commits.
4885 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4887 struct ext4_iloc iloc;
4888 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4889 static unsigned int mnt_count;
4893 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4894 err = ext4_reserve_inode_write(handle, inode, &iloc);
4895 if (ext4_handle_valid(handle) &&
4896 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4897 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4899 * We need extra buffer credits since we may write into EA block
4900 * with this same handle. If journal_extend fails, then it will
4901 * only result in a minor loss of functionality for that inode.
4902 * If this is felt to be critical, then e2fsck should be run to
4903 * force a large enough s_min_extra_isize.
4905 if ((jbd2_journal_extend(handle,
4906 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4907 ret = ext4_expand_extra_isize(inode,
4908 sbi->s_want_extra_isize,
4911 ext4_set_inode_state(inode,
4912 EXT4_STATE_NO_EXPAND);
4914 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4915 ext4_warning(inode->i_sb,
4916 "Unable to expand inode %lu. Delete"
4917 " some EAs or run e2fsck.",
4920 le16_to_cpu(sbi->s_es->s_mnt_count);
4926 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4931 * ext4_dirty_inode() is called from __mark_inode_dirty()
4933 * We're really interested in the case where a file is being extended.
4934 * i_size has been changed by generic_commit_write() and we thus need
4935 * to include the updated inode in the current transaction.
4937 * Also, dquot_alloc_block() will always dirty the inode when blocks
4938 * are allocated to the file.
4940 * If the inode is marked synchronous, we don't honour that here - doing
4941 * so would cause a commit on atime updates, which we don't bother doing.
4942 * We handle synchronous inodes at the highest possible level.
4944 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
4945 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
4946 * to copy into the on-disk inode structure are the timestamp files.
4948 void ext4_dirty_inode(struct inode *inode, int flags)
4952 if (flags == I_DIRTY_TIME)
4954 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
4958 ext4_mark_inode_dirty(handle, inode);
4960 ext4_journal_stop(handle);
4967 * Bind an inode's backing buffer_head into this transaction, to prevent
4968 * it from being flushed to disk early. Unlike
4969 * ext4_reserve_inode_write, this leaves behind no bh reference and
4970 * returns no iloc structure, so the caller needs to repeat the iloc
4971 * lookup to mark the inode dirty later.
4973 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4975 struct ext4_iloc iloc;
4979 err = ext4_get_inode_loc(inode, &iloc);
4981 BUFFER_TRACE(iloc.bh, "get_write_access");
4982 err = jbd2_journal_get_write_access(handle, iloc.bh);
4984 err = ext4_handle_dirty_metadata(handle,
4990 ext4_std_error(inode->i_sb, err);
4995 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5002 * We have to be very careful here: changing a data block's
5003 * journaling status dynamically is dangerous. If we write a
5004 * data block to the journal, change the status and then delete
5005 * that block, we risk forgetting to revoke the old log record
5006 * from the journal and so a subsequent replay can corrupt data.
5007 * So, first we make sure that the journal is empty and that
5008 * nobody is changing anything.
5011 journal = EXT4_JOURNAL(inode);
5014 if (is_journal_aborted(journal))
5016 /* We have to allocate physical blocks for delalloc blocks
5017 * before flushing journal. otherwise delalloc blocks can not
5018 * be allocated any more. even more truncate on delalloc blocks
5019 * could trigger BUG by flushing delalloc blocks in journal.
5020 * There is no delalloc block in non-journal data mode.
5022 if (val && test_opt(inode->i_sb, DELALLOC)) {
5023 err = ext4_alloc_da_blocks(inode);
5028 /* Wait for all existing dio workers */
5029 ext4_inode_block_unlocked_dio(inode);
5030 inode_dio_wait(inode);
5032 jbd2_journal_lock_updates(journal);
5035 * OK, there are no updates running now, and all cached data is
5036 * synced to disk. We are now in a completely consistent state
5037 * which doesn't have anything in the journal, and we know that
5038 * no filesystem updates are running, so it is safe to modify
5039 * the inode's in-core data-journaling state flag now.
5043 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5045 err = jbd2_journal_flush(journal);
5047 jbd2_journal_unlock_updates(journal);
5048 ext4_inode_resume_unlocked_dio(inode);
5051 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5053 ext4_set_aops(inode);
5055 jbd2_journal_unlock_updates(journal);
5056 ext4_inode_resume_unlocked_dio(inode);
5058 /* Finally we can mark the inode as dirty. */
5060 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5062 return PTR_ERR(handle);
5064 err = ext4_mark_inode_dirty(handle, inode);
5065 ext4_handle_sync(handle);
5066 ext4_journal_stop(handle);
5067 ext4_std_error(inode->i_sb, err);
5072 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5074 return !buffer_mapped(bh);
5077 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5079 struct page *page = vmf->page;
5083 struct file *file = vma->vm_file;
5084 struct inode *inode = file_inode(file);
5085 struct address_space *mapping = inode->i_mapping;
5087 get_block_t *get_block;
5090 sb_start_pagefault(inode->i_sb);
5091 file_update_time(vma->vm_file);
5092 /* Delalloc case is easy... */
5093 if (test_opt(inode->i_sb, DELALLOC) &&
5094 !ext4_should_journal_data(inode) &&
5095 !ext4_nonda_switch(inode->i_sb)) {
5097 ret = __block_page_mkwrite(vma, vmf,
5098 ext4_da_get_block_prep);
5099 } while (ret == -ENOSPC &&
5100 ext4_should_retry_alloc(inode->i_sb, &retries));
5105 size = i_size_read(inode);
5106 /* Page got truncated from under us? */
5107 if (page->mapping != mapping || page_offset(page) > size) {
5109 ret = VM_FAULT_NOPAGE;
5113 if (page->index == size >> PAGE_CACHE_SHIFT)
5114 len = size & ~PAGE_CACHE_MASK;
5116 len = PAGE_CACHE_SIZE;
5118 * Return if we have all the buffers mapped. This avoids the need to do
5119 * journal_start/journal_stop which can block and take a long time
5121 if (page_has_buffers(page)) {
5122 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5124 ext4_bh_unmapped)) {
5125 /* Wait so that we don't change page under IO */
5126 wait_for_stable_page(page);
5127 ret = VM_FAULT_LOCKED;
5132 /* OK, we need to fill the hole... */
5133 if (ext4_should_dioread_nolock(inode))
5134 get_block = ext4_get_block_write;
5136 get_block = ext4_get_block;
5138 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5139 ext4_writepage_trans_blocks(inode));
5140 if (IS_ERR(handle)) {
5141 ret = VM_FAULT_SIGBUS;
5144 ret = __block_page_mkwrite(vma, vmf, get_block);
5145 if (!ret && ext4_should_journal_data(inode)) {
5146 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5147 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5149 ret = VM_FAULT_SIGBUS;
5150 ext4_journal_stop(handle);
5153 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5155 ext4_journal_stop(handle);
5156 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5159 ret = block_page_mkwrite_return(ret);
5161 sb_end_pagefault(inode->i_sb);