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/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/aio.h>
41 #include <linux/bitops.h>
43 #include "ext4_jbd2.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
52 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
53 struct ext4_inode_info *ei)
55 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
60 csum_lo = le16_to_cpu(raw->i_checksum_lo);
61 raw->i_checksum_lo = 0;
62 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
63 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
64 csum_hi = le16_to_cpu(raw->i_checksum_hi);
65 raw->i_checksum_hi = 0;
68 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
69 EXT4_INODE_SIZE(inode->i_sb));
71 raw->i_checksum_lo = cpu_to_le16(csum_lo);
72 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
73 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
74 raw->i_checksum_hi = cpu_to_le16(csum_hi);
79 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
80 struct ext4_inode_info *ei)
82 __u32 provided, calculated;
84 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
85 cpu_to_le32(EXT4_OS_LINUX) ||
86 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
87 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
90 provided = le16_to_cpu(raw->i_checksum_lo);
91 calculated = ext4_inode_csum(inode, raw, ei);
92 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
93 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
94 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
98 return provided == calculated;
101 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
102 struct ext4_inode_info *ei)
106 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
107 cpu_to_le32(EXT4_OS_LINUX) ||
108 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
109 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
112 csum = ext4_inode_csum(inode, raw, ei);
113 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
114 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
115 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
116 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
119 static inline int ext4_begin_ordered_truncate(struct inode *inode,
122 trace_ext4_begin_ordered_truncate(inode, new_size);
124 * If jinode is zero, then we never opened the file for
125 * writing, so there's no need to call
126 * jbd2_journal_begin_ordered_truncate() since there's no
127 * outstanding writes we need to flush.
129 if (!EXT4_I(inode)->jinode)
131 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
132 EXT4_I(inode)->jinode,
136 static void ext4_invalidatepage(struct page *page, unsigned int offset,
137 unsigned int length);
138 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
139 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
140 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
144 * Test whether an inode is a fast symlink.
146 static int ext4_inode_is_fast_symlink(struct inode *inode)
148 int ea_blocks = EXT4_I(inode)->i_file_acl ?
149 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
151 if (ext4_has_inline_data(inode))
154 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
158 * Restart the transaction associated with *handle. This does a commit,
159 * so before we call here everything must be consistently dirtied against
162 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
168 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
169 * moment, get_block can be called only for blocks inside i_size since
170 * page cache has been already dropped and writes are blocked by
171 * i_mutex. So we can safely drop the i_data_sem here.
173 BUG_ON(EXT4_JOURNAL(inode) == NULL);
174 jbd_debug(2, "restarting handle %p\n", handle);
175 up_write(&EXT4_I(inode)->i_data_sem);
176 ret = ext4_journal_restart(handle, nblocks);
177 down_write(&EXT4_I(inode)->i_data_sem);
178 ext4_discard_preallocations(inode);
184 * Called at the last iput() if i_nlink is zero.
186 void ext4_evict_inode(struct inode *inode)
191 trace_ext4_evict_inode(inode);
193 if (inode->i_nlink) {
195 * When journalling data dirty buffers are tracked only in the
196 * journal. So although mm thinks everything is clean and
197 * ready for reaping the inode might still have some pages to
198 * write in the running transaction or waiting to be
199 * checkpointed. Thus calling jbd2_journal_invalidatepage()
200 * (via truncate_inode_pages()) to discard these buffers can
201 * cause data loss. Also even if we did not discard these
202 * buffers, we would have no way to find them after the inode
203 * is reaped and thus user could see stale data if he tries to
204 * read them before the transaction is checkpointed. So be
205 * careful and force everything to disk here... We use
206 * ei->i_datasync_tid to store the newest transaction
207 * containing inode's data.
209 * Note that directories do not have this problem because they
210 * don't use page cache.
212 if (ext4_should_journal_data(inode) &&
213 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
214 inode->i_ino != EXT4_JOURNAL_INO) {
215 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
216 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
218 jbd2_complete_transaction(journal, commit_tid);
219 filemap_write_and_wait(&inode->i_data);
221 truncate_inode_pages_final(&inode->i_data);
223 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
227 if (is_bad_inode(inode))
229 dquot_initialize(inode);
231 if (ext4_should_order_data(inode))
232 ext4_begin_ordered_truncate(inode, 0);
233 truncate_inode_pages_final(&inode->i_data);
235 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
238 * Protect us against freezing - iput() caller didn't have to have any
239 * protection against it
241 sb_start_intwrite(inode->i_sb);
242 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
243 ext4_blocks_for_truncate(inode)+3);
244 if (IS_ERR(handle)) {
245 ext4_std_error(inode->i_sb, PTR_ERR(handle));
247 * If we're going to skip the normal cleanup, we still need to
248 * make sure that the in-core orphan linked list is properly
251 ext4_orphan_del(NULL, inode);
252 sb_end_intwrite(inode->i_sb);
257 ext4_handle_sync(handle);
259 err = ext4_mark_inode_dirty(handle, inode);
261 ext4_warning(inode->i_sb,
262 "couldn't mark inode dirty (err %d)", err);
266 ext4_truncate(inode);
269 * ext4_ext_truncate() doesn't reserve any slop when it
270 * restarts journal transactions; therefore there may not be
271 * enough credits left in the handle to remove the inode from
272 * the orphan list and set the dtime field.
274 if (!ext4_handle_has_enough_credits(handle, 3)) {
275 err = ext4_journal_extend(handle, 3);
277 err = ext4_journal_restart(handle, 3);
279 ext4_warning(inode->i_sb,
280 "couldn't extend journal (err %d)", err);
282 ext4_journal_stop(handle);
283 ext4_orphan_del(NULL, inode);
284 sb_end_intwrite(inode->i_sb);
290 * Kill off the orphan record which ext4_truncate created.
291 * AKPM: I think this can be inside the above `if'.
292 * Note that ext4_orphan_del() has to be able to cope with the
293 * deletion of a non-existent orphan - this is because we don't
294 * know if ext4_truncate() actually created an orphan record.
295 * (Well, we could do this if we need to, but heck - it works)
297 ext4_orphan_del(handle, inode);
298 EXT4_I(inode)->i_dtime = get_seconds();
301 * One subtle ordering requirement: if anything has gone wrong
302 * (transaction abort, IO errors, whatever), then we can still
303 * do these next steps (the fs will already have been marked as
304 * having errors), but we can't free the inode if the mark_dirty
307 if (ext4_mark_inode_dirty(handle, inode))
308 /* If that failed, just do the required in-core inode clear. */
309 ext4_clear_inode(inode);
311 ext4_free_inode(handle, inode);
312 ext4_journal_stop(handle);
313 sb_end_intwrite(inode->i_sb);
316 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
320 qsize_t *ext4_get_reserved_space(struct inode *inode)
322 return &EXT4_I(inode)->i_reserved_quota;
327 * Called with i_data_sem down, which is important since we can call
328 * ext4_discard_preallocations() from here.
330 void ext4_da_update_reserve_space(struct inode *inode,
331 int used, int quota_claim)
333 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
334 struct ext4_inode_info *ei = EXT4_I(inode);
336 spin_lock(&ei->i_block_reservation_lock);
337 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
338 if (unlikely(used > ei->i_reserved_data_blocks)) {
339 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
340 "with only %d reserved data blocks",
341 __func__, inode->i_ino, used,
342 ei->i_reserved_data_blocks);
344 used = ei->i_reserved_data_blocks;
347 /* Update per-inode reservations */
348 ei->i_reserved_data_blocks -= used;
349 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
351 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
353 /* Update quota subsystem for data blocks */
355 dquot_claim_block(inode, EXT4_C2B(sbi, used));
358 * We did fallocate with an offset that is already delayed
359 * allocated. So on delayed allocated writeback we should
360 * not re-claim the quota for fallocated blocks.
362 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
366 * If we have done all the pending block allocations and if
367 * there aren't any writers on the inode, we can discard the
368 * inode's preallocations.
370 if ((ei->i_reserved_data_blocks == 0) &&
371 (atomic_read(&inode->i_writecount) == 0))
372 ext4_discard_preallocations(inode);
375 static int __check_block_validity(struct inode *inode, const char *func,
377 struct ext4_map_blocks *map)
379 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
381 ext4_error_inode(inode, func, line, map->m_pblk,
382 "lblock %lu mapped to illegal pblock "
383 "(length %d)", (unsigned long) map->m_lblk,
390 #define check_block_validity(inode, map) \
391 __check_block_validity((inode), __func__, __LINE__, (map))
393 #ifdef ES_AGGRESSIVE_TEST
394 static void ext4_map_blocks_es_recheck(handle_t *handle,
396 struct ext4_map_blocks *es_map,
397 struct ext4_map_blocks *map,
404 * There is a race window that the result is not the same.
405 * e.g. xfstests #223 when dioread_nolock enables. The reason
406 * is that we lookup a block mapping in extent status tree with
407 * out taking i_data_sem. So at the time the unwritten extent
408 * could be converted.
410 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
411 down_read(&EXT4_I(inode)->i_data_sem);
412 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
413 retval = ext4_ext_map_blocks(handle, inode, map, flags &
414 EXT4_GET_BLOCKS_KEEP_SIZE);
416 retval = ext4_ind_map_blocks(handle, inode, map, flags &
417 EXT4_GET_BLOCKS_KEEP_SIZE);
419 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
420 up_read((&EXT4_I(inode)->i_data_sem));
422 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
423 * because it shouldn't be marked in es_map->m_flags.
425 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
428 * We don't check m_len because extent will be collpased in status
429 * tree. So the m_len might not equal.
431 if (es_map->m_lblk != map->m_lblk ||
432 es_map->m_flags != map->m_flags ||
433 es_map->m_pblk != map->m_pblk) {
434 printk("ES cache assertion failed for inode: %lu "
435 "es_cached ex [%d/%d/%llu/%x] != "
436 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
437 inode->i_ino, es_map->m_lblk, es_map->m_len,
438 es_map->m_pblk, es_map->m_flags, map->m_lblk,
439 map->m_len, map->m_pblk, map->m_flags,
443 #endif /* ES_AGGRESSIVE_TEST */
446 * The ext4_map_blocks() function tries to look up the requested blocks,
447 * and returns if the blocks are already mapped.
449 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
450 * and store the allocated blocks in the result buffer head and mark it
453 * If file type is extents based, it will call ext4_ext_map_blocks(),
454 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
457 * On success, it returns the number of blocks being mapped or allocated.
458 * if create==0 and the blocks are pre-allocated and unwritten block,
459 * the result buffer head is unmapped. If the create ==1, it will make sure
460 * the buffer head is mapped.
462 * It returns 0 if plain look up failed (blocks have not been allocated), in
463 * that case, buffer head is unmapped
465 * It returns the error in case of allocation failure.
467 int ext4_map_blocks(handle_t *handle, struct inode *inode,
468 struct ext4_map_blocks *map, int flags)
470 struct extent_status es;
473 #ifdef ES_AGGRESSIVE_TEST
474 struct ext4_map_blocks orig_map;
476 memcpy(&orig_map, map, sizeof(*map));
480 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
481 "logical block %lu\n", inode->i_ino, flags, map->m_len,
482 (unsigned long) map->m_lblk);
485 * ext4_map_blocks returns an int, and m_len is an unsigned int
487 if (unlikely(map->m_len > INT_MAX))
488 map->m_len = INT_MAX;
490 /* We can handle the block number less than EXT_MAX_BLOCKS */
491 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
494 /* Lookup extent status tree firstly */
495 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
496 ext4_es_lru_add(inode);
497 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
498 map->m_pblk = ext4_es_pblock(&es) +
499 map->m_lblk - es.es_lblk;
500 map->m_flags |= ext4_es_is_written(&es) ?
501 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
502 retval = es.es_len - (map->m_lblk - es.es_lblk);
503 if (retval > map->m_len)
506 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
511 #ifdef ES_AGGRESSIVE_TEST
512 ext4_map_blocks_es_recheck(handle, inode, map,
519 * Try to see if we can get the block without requesting a new
522 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
523 down_read(&EXT4_I(inode)->i_data_sem);
524 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
525 retval = ext4_ext_map_blocks(handle, inode, map, flags &
526 EXT4_GET_BLOCKS_KEEP_SIZE);
528 retval = ext4_ind_map_blocks(handle, inode, map, flags &
529 EXT4_GET_BLOCKS_KEEP_SIZE);
534 if (unlikely(retval != map->m_len)) {
535 ext4_warning(inode->i_sb,
536 "ES len assertion failed for inode "
537 "%lu: retval %d != map->m_len %d",
538 inode->i_ino, retval, map->m_len);
542 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
543 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
544 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
545 ext4_find_delalloc_range(inode, map->m_lblk,
546 map->m_lblk + map->m_len - 1))
547 status |= EXTENT_STATUS_DELAYED;
548 ret = ext4_es_insert_extent(inode, map->m_lblk,
549 map->m_len, map->m_pblk, status);
553 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
554 up_read((&EXT4_I(inode)->i_data_sem));
557 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
558 ret = check_block_validity(inode, map);
563 /* If it is only a block(s) look up */
564 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
568 * Returns if the blocks have already allocated
570 * Note that if blocks have been preallocated
571 * ext4_ext_get_block() returns the create = 0
572 * with buffer head unmapped.
574 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
576 * If we need to convert extent to unwritten
577 * we continue and do the actual work in
578 * ext4_ext_map_blocks()
580 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
584 * Here we clear m_flags because after allocating an new extent,
585 * it will be set again.
587 map->m_flags &= ~EXT4_MAP_FLAGS;
590 * New blocks allocate and/or writing to unwritten extent
591 * will possibly result in updating i_data, so we take
592 * the write lock of i_data_sem, and call get_block()
593 * with create == 1 flag.
595 down_write(&EXT4_I(inode)->i_data_sem);
598 * We need to check for EXT4 here because migrate
599 * could have changed the inode type in between
601 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
602 retval = ext4_ext_map_blocks(handle, inode, map, flags);
604 retval = ext4_ind_map_blocks(handle, inode, map, flags);
606 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
608 * We allocated new blocks which will result in
609 * i_data's format changing. Force the migrate
610 * to fail by clearing migrate flags
612 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
616 * Update reserved blocks/metadata blocks after successful
617 * block allocation which had been deferred till now. We don't
618 * support fallocate for non extent files. So we can update
619 * reserve space here.
622 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
623 ext4_da_update_reserve_space(inode, retval, 1);
629 if (unlikely(retval != map->m_len)) {
630 ext4_warning(inode->i_sb,
631 "ES len assertion failed for inode "
632 "%lu: retval %d != map->m_len %d",
633 inode->i_ino, retval, map->m_len);
638 * If the extent has been zeroed out, we don't need to update
639 * extent status tree.
641 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
642 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
643 if (ext4_es_is_written(&es))
646 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
647 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
648 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
649 ext4_find_delalloc_range(inode, map->m_lblk,
650 map->m_lblk + map->m_len - 1))
651 status |= EXTENT_STATUS_DELAYED;
652 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
653 map->m_pblk, status);
659 up_write((&EXT4_I(inode)->i_data_sem));
660 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
661 ret = check_block_validity(inode, map);
668 /* Maximum number of blocks we map for direct IO at once. */
669 #define DIO_MAX_BLOCKS 4096
671 static int _ext4_get_block(struct inode *inode, sector_t iblock,
672 struct buffer_head *bh, int flags)
674 handle_t *handle = ext4_journal_current_handle();
675 struct ext4_map_blocks map;
676 int ret = 0, started = 0;
679 if (ext4_has_inline_data(inode))
683 map.m_len = bh->b_size >> inode->i_blkbits;
685 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
686 /* Direct IO write... */
687 if (map.m_len > DIO_MAX_BLOCKS)
688 map.m_len = DIO_MAX_BLOCKS;
689 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
690 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
692 if (IS_ERR(handle)) {
693 ret = PTR_ERR(handle);
699 ret = ext4_map_blocks(handle, inode, &map, flags);
701 ext4_io_end_t *io_end = ext4_inode_aio(inode);
703 map_bh(bh, inode->i_sb, map.m_pblk);
704 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
705 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
706 set_buffer_defer_completion(bh);
707 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
711 ext4_journal_stop(handle);
715 int ext4_get_block(struct inode *inode, sector_t iblock,
716 struct buffer_head *bh, int create)
718 return _ext4_get_block(inode, iblock, bh,
719 create ? EXT4_GET_BLOCKS_CREATE : 0);
723 * `handle' can be NULL if create is zero
725 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
726 ext4_lblk_t block, int create)
728 struct ext4_map_blocks map;
729 struct buffer_head *bh;
732 J_ASSERT(handle != NULL || create == 0);
736 err = ext4_map_blocks(handle, inode, &map,
737 create ? EXT4_GET_BLOCKS_CREATE : 0);
740 return create ? ERR_PTR(-ENOSPC) : NULL;
744 bh = sb_getblk(inode->i_sb, map.m_pblk);
746 return ERR_PTR(-ENOMEM);
747 if (map.m_flags & EXT4_MAP_NEW) {
748 J_ASSERT(create != 0);
749 J_ASSERT(handle != NULL);
752 * Now that we do not always journal data, we should
753 * keep in mind whether this should always journal the
754 * new buffer as metadata. For now, regular file
755 * writes use ext4_get_block instead, so it's not a
759 BUFFER_TRACE(bh, "call get_create_access");
760 err = ext4_journal_get_create_access(handle, bh);
765 if (!buffer_uptodate(bh)) {
766 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
767 set_buffer_uptodate(bh);
770 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
771 err = ext4_handle_dirty_metadata(handle, inode, bh);
775 BUFFER_TRACE(bh, "not a new buffer");
782 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
783 ext4_lblk_t block, int create)
785 struct buffer_head *bh;
787 bh = ext4_getblk(handle, inode, block, create);
790 if (!bh || buffer_uptodate(bh))
792 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
794 if (buffer_uptodate(bh))
797 return ERR_PTR(-EIO);
800 int ext4_walk_page_buffers(handle_t *handle,
801 struct buffer_head *head,
805 int (*fn)(handle_t *handle,
806 struct buffer_head *bh))
808 struct buffer_head *bh;
809 unsigned block_start, block_end;
810 unsigned blocksize = head->b_size;
812 struct buffer_head *next;
814 for (bh = head, block_start = 0;
815 ret == 0 && (bh != head || !block_start);
816 block_start = block_end, bh = next) {
817 next = bh->b_this_page;
818 block_end = block_start + blocksize;
819 if (block_end <= from || block_start >= to) {
820 if (partial && !buffer_uptodate(bh))
824 err = (*fn)(handle, bh);
832 * To preserve ordering, it is essential that the hole instantiation and
833 * the data write be encapsulated in a single transaction. We cannot
834 * close off a transaction and start a new one between the ext4_get_block()
835 * and the commit_write(). So doing the jbd2_journal_start at the start of
836 * prepare_write() is the right place.
838 * Also, this function can nest inside ext4_writepage(). In that case, we
839 * *know* that ext4_writepage() has generated enough buffer credits to do the
840 * whole page. So we won't block on the journal in that case, which is good,
841 * because the caller may be PF_MEMALLOC.
843 * By accident, ext4 can be reentered when a transaction is open via
844 * quota file writes. If we were to commit the transaction while thus
845 * reentered, there can be a deadlock - we would be holding a quota
846 * lock, and the commit would never complete if another thread had a
847 * transaction open and was blocking on the quota lock - a ranking
850 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
851 * will _not_ run commit under these circumstances because handle->h_ref
852 * is elevated. We'll still have enough credits for the tiny quotafile
855 int do_journal_get_write_access(handle_t *handle,
856 struct buffer_head *bh)
858 int dirty = buffer_dirty(bh);
861 if (!buffer_mapped(bh) || buffer_freed(bh))
864 * __block_write_begin() could have dirtied some buffers. Clean
865 * the dirty bit as jbd2_journal_get_write_access() could complain
866 * otherwise about fs integrity issues. Setting of the dirty bit
867 * by __block_write_begin() isn't a real problem here as we clear
868 * the bit before releasing a page lock and thus writeback cannot
869 * ever write the buffer.
872 clear_buffer_dirty(bh);
873 BUFFER_TRACE(bh, "get write access");
874 ret = ext4_journal_get_write_access(handle, bh);
876 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
880 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
881 struct buffer_head *bh_result, int create);
882 static int ext4_write_begin(struct file *file, struct address_space *mapping,
883 loff_t pos, unsigned len, unsigned flags,
884 struct page **pagep, void **fsdata)
886 struct inode *inode = mapping->host;
887 int ret, needed_blocks;
894 trace_ext4_write_begin(inode, pos, len, flags);
896 * Reserve one block more for addition to orphan list in case
897 * we allocate blocks but write fails for some reason
899 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
900 index = pos >> PAGE_CACHE_SHIFT;
901 from = pos & (PAGE_CACHE_SIZE - 1);
904 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
905 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
914 * grab_cache_page_write_begin() can take a long time if the
915 * system is thrashing due to memory pressure, or if the page
916 * is being written back. So grab it first before we start
917 * the transaction handle. This also allows us to allocate
918 * the page (if needed) without using GFP_NOFS.
921 page = grab_cache_page_write_begin(mapping, index, flags);
927 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
928 if (IS_ERR(handle)) {
929 page_cache_release(page);
930 return PTR_ERR(handle);
934 if (page->mapping != mapping) {
935 /* The page got truncated from under us */
937 page_cache_release(page);
938 ext4_journal_stop(handle);
941 /* In case writeback began while the page was unlocked */
942 wait_for_stable_page(page);
944 if (ext4_should_dioread_nolock(inode))
945 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
947 ret = __block_write_begin(page, pos, len, ext4_get_block);
949 if (!ret && ext4_should_journal_data(inode)) {
950 ret = ext4_walk_page_buffers(handle, page_buffers(page),
952 do_journal_get_write_access);
958 * __block_write_begin may have instantiated a few blocks
959 * outside i_size. Trim these off again. Don't need
960 * i_size_read because we hold i_mutex.
962 * Add inode to orphan list in case we crash before
965 if (pos + len > inode->i_size && ext4_can_truncate(inode))
966 ext4_orphan_add(handle, inode);
968 ext4_journal_stop(handle);
969 if (pos + len > inode->i_size) {
970 ext4_truncate_failed_write(inode);
972 * If truncate failed early the inode might
973 * still be on the orphan list; we need to
974 * make sure the inode is removed from the
975 * orphan list in that case.
978 ext4_orphan_del(NULL, inode);
981 if (ret == -ENOSPC &&
982 ext4_should_retry_alloc(inode->i_sb, &retries))
984 page_cache_release(page);
991 /* For write_end() in data=journal mode */
992 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
995 if (!buffer_mapped(bh) || buffer_freed(bh))
997 set_buffer_uptodate(bh);
998 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
999 clear_buffer_meta(bh);
1000 clear_buffer_prio(bh);
1005 * We need to pick up the new inode size which generic_commit_write gave us
1006 * `file' can be NULL - eg, when called from page_symlink().
1008 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1009 * buffers are managed internally.
1011 static int ext4_write_end(struct file *file,
1012 struct address_space *mapping,
1013 loff_t pos, unsigned len, unsigned copied,
1014 struct page *page, void *fsdata)
1016 handle_t *handle = ext4_journal_current_handle();
1017 struct inode *inode = mapping->host;
1019 int i_size_changed = 0;
1021 trace_ext4_write_end(inode, pos, len, copied);
1022 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1023 ret = ext4_jbd2_file_inode(handle, inode);
1026 page_cache_release(page);
1031 if (ext4_has_inline_data(inode)) {
1032 ret = ext4_write_inline_data_end(inode, pos, len,
1038 copied = block_write_end(file, mapping, pos,
1039 len, copied, page, fsdata);
1041 * it's important to update i_size while still holding page lock:
1042 * page writeout could otherwise come in and zero beyond i_size.
1044 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1046 page_cache_release(page);
1049 * Don't mark the inode dirty under page lock. First, it unnecessarily
1050 * makes the holding time of page lock longer. Second, it forces lock
1051 * ordering of page lock and transaction start for journaling
1055 ext4_mark_inode_dirty(handle, inode);
1057 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1058 /* if we have allocated more blocks and copied
1059 * less. We will have blocks allocated outside
1060 * inode->i_size. So truncate them
1062 ext4_orphan_add(handle, inode);
1064 ret2 = ext4_journal_stop(handle);
1068 if (pos + len > inode->i_size) {
1069 ext4_truncate_failed_write(inode);
1071 * If truncate failed early the inode might still be
1072 * on the orphan list; we need to make sure the inode
1073 * is removed from the orphan list in that case.
1076 ext4_orphan_del(NULL, inode);
1079 return ret ? ret : copied;
1082 static int ext4_journalled_write_end(struct file *file,
1083 struct address_space *mapping,
1084 loff_t pos, unsigned len, unsigned copied,
1085 struct page *page, void *fsdata)
1087 handle_t *handle = ext4_journal_current_handle();
1088 struct inode *inode = mapping->host;
1092 int size_changed = 0;
1094 trace_ext4_journalled_write_end(inode, pos, len, copied);
1095 from = pos & (PAGE_CACHE_SIZE - 1);
1098 BUG_ON(!ext4_handle_valid(handle));
1100 if (ext4_has_inline_data(inode))
1101 copied = ext4_write_inline_data_end(inode, pos, len,
1105 if (!PageUptodate(page))
1107 page_zero_new_buffers(page, from+copied, to);
1110 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1111 to, &partial, write_end_fn);
1113 SetPageUptodate(page);
1115 size_changed = ext4_update_inode_size(inode, pos + copied);
1116 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1117 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1119 page_cache_release(page);
1122 ret2 = ext4_mark_inode_dirty(handle, inode);
1127 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1128 /* if we have allocated more blocks and copied
1129 * less. We will have blocks allocated outside
1130 * inode->i_size. So truncate them
1132 ext4_orphan_add(handle, inode);
1134 ret2 = ext4_journal_stop(handle);
1137 if (pos + len > inode->i_size) {
1138 ext4_truncate_failed_write(inode);
1140 * If truncate failed early the inode might still be
1141 * on the orphan list; we need to make sure the inode
1142 * is removed from the orphan list in that case.
1145 ext4_orphan_del(NULL, inode);
1148 return ret ? ret : copied;
1152 * Reserve a single cluster located at lblock
1154 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1156 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1157 struct ext4_inode_info *ei = EXT4_I(inode);
1158 unsigned int md_needed;
1162 * We will charge metadata quota at writeout time; this saves
1163 * us from metadata over-estimation, though we may go over by
1164 * a small amount in the end. Here we just reserve for data.
1166 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1171 * recalculate the amount of metadata blocks to reserve
1172 * in order to allocate nrblocks
1173 * worse case is one extent per block
1175 spin_lock(&ei->i_block_reservation_lock);
1177 * ext4_calc_metadata_amount() has side effects, which we have
1178 * to be prepared undo if we fail to claim space.
1181 trace_ext4_da_reserve_space(inode, 0);
1183 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1184 spin_unlock(&ei->i_block_reservation_lock);
1185 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1188 ei->i_reserved_data_blocks++;
1189 spin_unlock(&ei->i_block_reservation_lock);
1191 return 0; /* success */
1194 static void ext4_da_release_space(struct inode *inode, int to_free)
1196 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1197 struct ext4_inode_info *ei = EXT4_I(inode);
1200 return; /* Nothing to release, exit */
1202 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1204 trace_ext4_da_release_space(inode, to_free);
1205 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1207 * if there aren't enough reserved blocks, then the
1208 * counter is messed up somewhere. Since this
1209 * function is called from invalidate page, it's
1210 * harmless to return without any action.
1212 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1213 "ino %lu, to_free %d with only %d reserved "
1214 "data blocks", inode->i_ino, to_free,
1215 ei->i_reserved_data_blocks);
1217 to_free = ei->i_reserved_data_blocks;
1219 ei->i_reserved_data_blocks -= to_free;
1221 /* update fs dirty data blocks counter */
1222 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1224 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1226 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1229 static void ext4_da_page_release_reservation(struct page *page,
1230 unsigned int offset,
1231 unsigned int length)
1234 struct buffer_head *head, *bh;
1235 unsigned int curr_off = 0;
1236 struct inode *inode = page->mapping->host;
1237 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1238 unsigned int stop = offset + length;
1242 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1244 head = page_buffers(page);
1247 unsigned int next_off = curr_off + bh->b_size;
1249 if (next_off > stop)
1252 if ((offset <= curr_off) && (buffer_delay(bh))) {
1254 clear_buffer_delay(bh);
1256 curr_off = next_off;
1257 } while ((bh = bh->b_this_page) != head);
1260 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1261 ext4_es_remove_extent(inode, lblk, to_release);
1264 /* If we have released all the blocks belonging to a cluster, then we
1265 * need to release the reserved space for that cluster. */
1266 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1267 while (num_clusters > 0) {
1268 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1269 ((num_clusters - 1) << sbi->s_cluster_bits);
1270 if (sbi->s_cluster_ratio == 1 ||
1271 !ext4_find_delalloc_cluster(inode, lblk))
1272 ext4_da_release_space(inode, 1);
1279 * Delayed allocation stuff
1282 struct mpage_da_data {
1283 struct inode *inode;
1284 struct writeback_control *wbc;
1286 pgoff_t first_page; /* The first page to write */
1287 pgoff_t next_page; /* Current page to examine */
1288 pgoff_t last_page; /* Last page to examine */
1290 * Extent to map - this can be after first_page because that can be
1291 * fully mapped. We somewhat abuse m_flags to store whether the extent
1292 * is delalloc or unwritten.
1294 struct ext4_map_blocks map;
1295 struct ext4_io_submit io_submit; /* IO submission data */
1298 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1303 struct pagevec pvec;
1304 struct inode *inode = mpd->inode;
1305 struct address_space *mapping = inode->i_mapping;
1307 /* This is necessary when next_page == 0. */
1308 if (mpd->first_page >= mpd->next_page)
1311 index = mpd->first_page;
1312 end = mpd->next_page - 1;
1314 ext4_lblk_t start, last;
1315 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1316 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1317 ext4_es_remove_extent(inode, start, last - start + 1);
1320 pagevec_init(&pvec, 0);
1321 while (index <= end) {
1322 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1325 for (i = 0; i < nr_pages; i++) {
1326 struct page *page = pvec.pages[i];
1327 if (page->index > end)
1329 BUG_ON(!PageLocked(page));
1330 BUG_ON(PageWriteback(page));
1332 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1333 ClearPageUptodate(page);
1337 index = pvec.pages[nr_pages - 1]->index + 1;
1338 pagevec_release(&pvec);
1342 static void ext4_print_free_blocks(struct inode *inode)
1344 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1345 struct super_block *sb = inode->i_sb;
1346 struct ext4_inode_info *ei = EXT4_I(inode);
1348 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1349 EXT4_C2B(EXT4_SB(inode->i_sb),
1350 ext4_count_free_clusters(sb)));
1351 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1352 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1353 (long long) EXT4_C2B(EXT4_SB(sb),
1354 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1355 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1356 (long long) EXT4_C2B(EXT4_SB(sb),
1357 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1358 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1359 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1360 ei->i_reserved_data_blocks);
1364 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1366 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1370 * This function is grabs code from the very beginning of
1371 * ext4_map_blocks, but assumes that the caller is from delayed write
1372 * time. This function looks up the requested blocks and sets the
1373 * buffer delay bit under the protection of i_data_sem.
1375 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1376 struct ext4_map_blocks *map,
1377 struct buffer_head *bh)
1379 struct extent_status es;
1381 sector_t invalid_block = ~((sector_t) 0xffff);
1382 #ifdef ES_AGGRESSIVE_TEST
1383 struct ext4_map_blocks orig_map;
1385 memcpy(&orig_map, map, sizeof(*map));
1388 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1392 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1393 "logical block %lu\n", inode->i_ino, map->m_len,
1394 (unsigned long) map->m_lblk);
1396 /* Lookup extent status tree firstly */
1397 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1398 ext4_es_lru_add(inode);
1399 if (ext4_es_is_hole(&es)) {
1401 down_read(&EXT4_I(inode)->i_data_sem);
1406 * Delayed extent could be allocated by fallocate.
1407 * So we need to check it.
1409 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1410 map_bh(bh, inode->i_sb, invalid_block);
1412 set_buffer_delay(bh);
1416 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1417 retval = es.es_len - (iblock - es.es_lblk);
1418 if (retval > map->m_len)
1419 retval = map->m_len;
1420 map->m_len = retval;
1421 if (ext4_es_is_written(&es))
1422 map->m_flags |= EXT4_MAP_MAPPED;
1423 else if (ext4_es_is_unwritten(&es))
1424 map->m_flags |= EXT4_MAP_UNWRITTEN;
1428 #ifdef ES_AGGRESSIVE_TEST
1429 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1435 * Try to see if we can get the block without requesting a new
1436 * file system block.
1438 down_read(&EXT4_I(inode)->i_data_sem);
1439 if (ext4_has_inline_data(inode)) {
1441 * We will soon create blocks for this page, and let
1442 * us pretend as if the blocks aren't allocated yet.
1443 * In case of clusters, we have to handle the work
1444 * of mapping from cluster so that the reserved space
1445 * is calculated properly.
1447 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1448 ext4_find_delalloc_cluster(inode, map->m_lblk))
1449 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1451 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1452 retval = ext4_ext_map_blocks(NULL, inode, map,
1453 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1455 retval = ext4_ind_map_blocks(NULL, inode, map,
1456 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1462 * XXX: __block_prepare_write() unmaps passed block,
1466 * If the block was allocated from previously allocated cluster,
1467 * then we don't need to reserve it again. However we still need
1468 * to reserve metadata for every block we're going to write.
1470 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1471 ret = ext4_da_reserve_space(inode, iblock);
1473 /* not enough space to reserve */
1479 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1480 ~0, EXTENT_STATUS_DELAYED);
1486 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1487 * and it should not appear on the bh->b_state.
1489 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1491 map_bh(bh, inode->i_sb, invalid_block);
1493 set_buffer_delay(bh);
1494 } else if (retval > 0) {
1496 unsigned int status;
1498 if (unlikely(retval != map->m_len)) {
1499 ext4_warning(inode->i_sb,
1500 "ES len assertion failed for inode "
1501 "%lu: retval %d != map->m_len %d",
1502 inode->i_ino, retval, map->m_len);
1506 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1507 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1508 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1509 map->m_pblk, status);
1515 up_read((&EXT4_I(inode)->i_data_sem));
1521 * This is a special get_block_t callback which is used by
1522 * ext4_da_write_begin(). It will either return mapped block or
1523 * reserve space for a single block.
1525 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1526 * We also have b_blocknr = -1 and b_bdev initialized properly
1528 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1529 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1530 * initialized properly.
1532 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1533 struct buffer_head *bh, int create)
1535 struct ext4_map_blocks map;
1538 BUG_ON(create == 0);
1539 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1541 map.m_lblk = iblock;
1545 * first, we need to know whether the block is allocated already
1546 * preallocated blocks are unmapped but should treated
1547 * the same as allocated blocks.
1549 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1553 map_bh(bh, inode->i_sb, map.m_pblk);
1554 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1556 if (buffer_unwritten(bh)) {
1557 /* A delayed write to unwritten bh should be marked
1558 * new and mapped. Mapped ensures that we don't do
1559 * get_block multiple times when we write to the same
1560 * offset and new ensures that we do proper zero out
1561 * for partial write.
1564 set_buffer_mapped(bh);
1569 static int bget_one(handle_t *handle, struct buffer_head *bh)
1575 static int bput_one(handle_t *handle, struct buffer_head *bh)
1581 static int __ext4_journalled_writepage(struct page *page,
1584 struct address_space *mapping = page->mapping;
1585 struct inode *inode = mapping->host;
1586 struct buffer_head *page_bufs = NULL;
1587 handle_t *handle = NULL;
1588 int ret = 0, err = 0;
1589 int inline_data = ext4_has_inline_data(inode);
1590 struct buffer_head *inode_bh = NULL;
1592 ClearPageChecked(page);
1595 BUG_ON(page->index != 0);
1596 BUG_ON(len > ext4_get_max_inline_size(inode));
1597 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1598 if (inode_bh == NULL)
1601 page_bufs = page_buffers(page);
1606 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1609 /* As soon as we unlock the page, it can go away, but we have
1610 * references to buffers so we are safe */
1613 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1614 ext4_writepage_trans_blocks(inode));
1615 if (IS_ERR(handle)) {
1616 ret = PTR_ERR(handle);
1620 BUG_ON(!ext4_handle_valid(handle));
1623 BUFFER_TRACE(inode_bh, "get write access");
1624 ret = ext4_journal_get_write_access(handle, inode_bh);
1626 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1629 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1630 do_journal_get_write_access);
1632 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1637 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1638 err = ext4_journal_stop(handle);
1642 if (!ext4_has_inline_data(inode))
1643 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1645 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1652 * Note that we don't need to start a transaction unless we're journaling data
1653 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1654 * need to file the inode to the transaction's list in ordered mode because if
1655 * we are writing back data added by write(), the inode is already there and if
1656 * we are writing back data modified via mmap(), no one guarantees in which
1657 * transaction the data will hit the disk. In case we are journaling data, we
1658 * cannot start transaction directly because transaction start ranks above page
1659 * lock so we have to do some magic.
1661 * This function can get called via...
1662 * - ext4_writepages after taking page lock (have journal handle)
1663 * - journal_submit_inode_data_buffers (no journal handle)
1664 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1665 * - grab_page_cache when doing write_begin (have journal handle)
1667 * We don't do any block allocation in this function. If we have page with
1668 * multiple blocks we need to write those buffer_heads that are mapped. This
1669 * is important for mmaped based write. So if we do with blocksize 1K
1670 * truncate(f, 1024);
1671 * a = mmap(f, 0, 4096);
1673 * truncate(f, 4096);
1674 * we have in the page first buffer_head mapped via page_mkwrite call back
1675 * but other buffer_heads would be unmapped but dirty (dirty done via the
1676 * do_wp_page). So writepage should write the first block. If we modify
1677 * the mmap area beyond 1024 we will again get a page_fault and the
1678 * page_mkwrite callback will do the block allocation and mark the
1679 * buffer_heads mapped.
1681 * We redirty the page if we have any buffer_heads that is either delay or
1682 * unwritten in the page.
1684 * We can get recursively called as show below.
1686 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1689 * But since we don't do any block allocation we should not deadlock.
1690 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1692 static int ext4_writepage(struct page *page,
1693 struct writeback_control *wbc)
1698 struct buffer_head *page_bufs = NULL;
1699 struct inode *inode = page->mapping->host;
1700 struct ext4_io_submit io_submit;
1701 bool keep_towrite = false;
1703 trace_ext4_writepage(page);
1704 size = i_size_read(inode);
1705 if (page->index == size >> PAGE_CACHE_SHIFT)
1706 len = size & ~PAGE_CACHE_MASK;
1708 len = PAGE_CACHE_SIZE;
1710 page_bufs = page_buffers(page);
1712 * We cannot do block allocation or other extent handling in this
1713 * function. If there are buffers needing that, we have to redirty
1714 * the page. But we may reach here when we do a journal commit via
1715 * journal_submit_inode_data_buffers() and in that case we must write
1716 * allocated buffers to achieve data=ordered mode guarantees.
1718 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1719 ext4_bh_delay_or_unwritten)) {
1720 redirty_page_for_writepage(wbc, page);
1721 if (current->flags & PF_MEMALLOC) {
1723 * For memory cleaning there's no point in writing only
1724 * some buffers. So just bail out. Warn if we came here
1725 * from direct reclaim.
1727 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1732 keep_towrite = true;
1735 if (PageChecked(page) && ext4_should_journal_data(inode))
1737 * It's mmapped pagecache. Add buffers and journal it. There
1738 * doesn't seem much point in redirtying the page here.
1740 return __ext4_journalled_writepage(page, len);
1742 ext4_io_submit_init(&io_submit, wbc);
1743 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1744 if (!io_submit.io_end) {
1745 redirty_page_for_writepage(wbc, page);
1749 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1750 ext4_io_submit(&io_submit);
1751 /* Drop io_end reference we got from init */
1752 ext4_put_io_end_defer(io_submit.io_end);
1756 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1759 loff_t size = i_size_read(mpd->inode);
1762 BUG_ON(page->index != mpd->first_page);
1763 if (page->index == size >> PAGE_CACHE_SHIFT)
1764 len = size & ~PAGE_CACHE_MASK;
1766 len = PAGE_CACHE_SIZE;
1767 clear_page_dirty_for_io(page);
1768 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1770 mpd->wbc->nr_to_write--;
1776 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1779 * mballoc gives us at most this number of blocks...
1780 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1781 * The rest of mballoc seems to handle chunks up to full group size.
1783 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1786 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1788 * @mpd - extent of blocks
1789 * @lblk - logical number of the block in the file
1790 * @bh - buffer head we want to add to the extent
1792 * The function is used to collect contig. blocks in the same state. If the
1793 * buffer doesn't require mapping for writeback and we haven't started the
1794 * extent of buffers to map yet, the function returns 'true' immediately - the
1795 * caller can write the buffer right away. Otherwise the function returns true
1796 * if the block has been added to the extent, false if the block couldn't be
1799 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1800 struct buffer_head *bh)
1802 struct ext4_map_blocks *map = &mpd->map;
1804 /* Buffer that doesn't need mapping for writeback? */
1805 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1806 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1807 /* So far no extent to map => we write the buffer right away */
1808 if (map->m_len == 0)
1813 /* First block in the extent? */
1814 if (map->m_len == 0) {
1817 map->m_flags = bh->b_state & BH_FLAGS;
1821 /* Don't go larger than mballoc is willing to allocate */
1822 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1825 /* Can we merge the block to our big extent? */
1826 if (lblk == map->m_lblk + map->m_len &&
1827 (bh->b_state & BH_FLAGS) == map->m_flags) {
1835 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1837 * @mpd - extent of blocks for mapping
1838 * @head - the first buffer in the page
1839 * @bh - buffer we should start processing from
1840 * @lblk - logical number of the block in the file corresponding to @bh
1842 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1843 * the page for IO if all buffers in this page were mapped and there's no
1844 * accumulated extent of buffers to map or add buffers in the page to the
1845 * extent of buffers to map. The function returns 1 if the caller can continue
1846 * by processing the next page, 0 if it should stop adding buffers to the
1847 * extent to map because we cannot extend it anymore. It can also return value
1848 * < 0 in case of error during IO submission.
1850 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1851 struct buffer_head *head,
1852 struct buffer_head *bh,
1855 struct inode *inode = mpd->inode;
1857 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
1858 >> inode->i_blkbits;
1861 BUG_ON(buffer_locked(bh));
1863 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
1864 /* Found extent to map? */
1867 /* Everything mapped so far and we hit EOF */
1870 } while (lblk++, (bh = bh->b_this_page) != head);
1871 /* So far everything mapped? Submit the page for IO. */
1872 if (mpd->map.m_len == 0) {
1873 err = mpage_submit_page(mpd, head->b_page);
1877 return lblk < blocks;
1881 * mpage_map_buffers - update buffers corresponding to changed extent and
1882 * submit fully mapped pages for IO
1884 * @mpd - description of extent to map, on return next extent to map
1886 * Scan buffers corresponding to changed extent (we expect corresponding pages
1887 * to be already locked) and update buffer state according to new extent state.
1888 * We map delalloc buffers to their physical location, clear unwritten bits,
1889 * and mark buffers as uninit when we perform writes to unwritten extents
1890 * and do extent conversion after IO is finished. If the last page is not fully
1891 * mapped, we update @map to the next extent in the last page that needs
1892 * mapping. Otherwise we submit the page for IO.
1894 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
1896 struct pagevec pvec;
1898 struct inode *inode = mpd->inode;
1899 struct buffer_head *head, *bh;
1900 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
1906 start = mpd->map.m_lblk >> bpp_bits;
1907 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
1908 lblk = start << bpp_bits;
1909 pblock = mpd->map.m_pblk;
1911 pagevec_init(&pvec, 0);
1912 while (start <= end) {
1913 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
1917 for (i = 0; i < nr_pages; i++) {
1918 struct page *page = pvec.pages[i];
1920 if (page->index > end)
1922 /* Up to 'end' pages must be contiguous */
1923 BUG_ON(page->index != start);
1924 bh = head = page_buffers(page);
1926 if (lblk < mpd->map.m_lblk)
1928 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
1930 * Buffer after end of mapped extent.
1931 * Find next buffer in the page to map.
1934 mpd->map.m_flags = 0;
1936 * FIXME: If dioread_nolock supports
1937 * blocksize < pagesize, we need to make
1938 * sure we add size mapped so far to
1939 * io_end->size as the following call
1940 * can submit the page for IO.
1942 err = mpage_process_page_bufs(mpd, head,
1944 pagevec_release(&pvec);
1949 if (buffer_delay(bh)) {
1950 clear_buffer_delay(bh);
1951 bh->b_blocknr = pblock++;
1953 clear_buffer_unwritten(bh);
1954 } while (lblk++, (bh = bh->b_this_page) != head);
1957 * FIXME: This is going to break if dioread_nolock
1958 * supports blocksize < pagesize as we will try to
1959 * convert potentially unmapped parts of inode.
1961 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
1962 /* Page fully mapped - let IO run! */
1963 err = mpage_submit_page(mpd, page);
1965 pagevec_release(&pvec);
1970 pagevec_release(&pvec);
1972 /* Extent fully mapped and matches with page boundary. We are done. */
1974 mpd->map.m_flags = 0;
1978 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
1980 struct inode *inode = mpd->inode;
1981 struct ext4_map_blocks *map = &mpd->map;
1982 int get_blocks_flags;
1983 int err, dioread_nolock;
1985 trace_ext4_da_write_pages_extent(inode, map);
1987 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
1988 * to convert an unwritten extent to be initialized (in the case
1989 * where we have written into one or more preallocated blocks). It is
1990 * possible that we're going to need more metadata blocks than
1991 * previously reserved. However we must not fail because we're in
1992 * writeback and there is nothing we can do about it so it might result
1993 * in data loss. So use reserved blocks to allocate metadata if
1996 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
1997 * the blocks in question are delalloc blocks. This indicates
1998 * that the blocks and quotas has already been checked when
1999 * the data was copied into the page cache.
2001 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2002 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2003 dioread_nolock = ext4_should_dioread_nolock(inode);
2005 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2006 if (map->m_flags & (1 << BH_Delay))
2007 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2009 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2012 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2013 if (!mpd->io_submit.io_end->handle &&
2014 ext4_handle_valid(handle)) {
2015 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2016 handle->h_rsv_handle = NULL;
2018 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2021 BUG_ON(map->m_len == 0);
2022 if (map->m_flags & EXT4_MAP_NEW) {
2023 struct block_device *bdev = inode->i_sb->s_bdev;
2026 for (i = 0; i < map->m_len; i++)
2027 unmap_underlying_metadata(bdev, map->m_pblk + i);
2033 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2034 * mpd->len and submit pages underlying it for IO
2036 * @handle - handle for journal operations
2037 * @mpd - extent to map
2038 * @give_up_on_write - we set this to true iff there is a fatal error and there
2039 * is no hope of writing the data. The caller should discard
2040 * dirty pages to avoid infinite loops.
2042 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2043 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2044 * them to initialized or split the described range from larger unwritten
2045 * extent. Note that we need not map all the described range since allocation
2046 * can return less blocks or the range is covered by more unwritten extents. We
2047 * cannot map more because we are limited by reserved transaction credits. On
2048 * the other hand we always make sure that the last touched page is fully
2049 * mapped so that it can be written out (and thus forward progress is
2050 * guaranteed). After mapping we submit all mapped pages for IO.
2052 static int mpage_map_and_submit_extent(handle_t *handle,
2053 struct mpage_da_data *mpd,
2054 bool *give_up_on_write)
2056 struct inode *inode = mpd->inode;
2057 struct ext4_map_blocks *map = &mpd->map;
2062 mpd->io_submit.io_end->offset =
2063 ((loff_t)map->m_lblk) << inode->i_blkbits;
2065 err = mpage_map_one_extent(handle, mpd);
2067 struct super_block *sb = inode->i_sb;
2069 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2070 goto invalidate_dirty_pages;
2072 * Let the uper layers retry transient errors.
2073 * In the case of ENOSPC, if ext4_count_free_blocks()
2074 * is non-zero, a commit should free up blocks.
2076 if ((err == -ENOMEM) ||
2077 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2079 goto update_disksize;
2082 ext4_msg(sb, KERN_CRIT,
2083 "Delayed block allocation failed for "
2084 "inode %lu at logical offset %llu with"
2085 " max blocks %u with error %d",
2087 (unsigned long long)map->m_lblk,
2088 (unsigned)map->m_len, -err);
2089 ext4_msg(sb, KERN_CRIT,
2090 "This should not happen!! Data will "
2093 ext4_print_free_blocks(inode);
2094 invalidate_dirty_pages:
2095 *give_up_on_write = true;
2100 * Update buffer state, submit mapped pages, and get us new
2103 err = mpage_map_and_submit_buffers(mpd);
2105 goto update_disksize;
2106 } while (map->m_len);
2110 * Update on-disk size after IO is submitted. Races with
2111 * truncate are avoided by checking i_size under i_data_sem.
2113 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2114 if (disksize > EXT4_I(inode)->i_disksize) {
2118 down_write(&EXT4_I(inode)->i_data_sem);
2119 i_size = i_size_read(inode);
2120 if (disksize > i_size)
2122 if (disksize > EXT4_I(inode)->i_disksize)
2123 EXT4_I(inode)->i_disksize = disksize;
2124 err2 = ext4_mark_inode_dirty(handle, inode);
2125 up_write(&EXT4_I(inode)->i_data_sem);
2127 ext4_error(inode->i_sb,
2128 "Failed to mark inode %lu dirty",
2137 * Calculate the total number of credits to reserve for one writepages
2138 * iteration. This is called from ext4_writepages(). We map an extent of
2139 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2140 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2141 * bpp - 1 blocks in bpp different extents.
2143 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2145 int bpp = ext4_journal_blocks_per_page(inode);
2147 return ext4_meta_trans_blocks(inode,
2148 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2152 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2153 * and underlying extent to map
2155 * @mpd - where to look for pages
2157 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2158 * IO immediately. When we find a page which isn't mapped we start accumulating
2159 * extent of buffers underlying these pages that needs mapping (formed by
2160 * either delayed or unwritten buffers). We also lock the pages containing
2161 * these buffers. The extent found is returned in @mpd structure (starting at
2162 * mpd->lblk with length mpd->len blocks).
2164 * Note that this function can attach bios to one io_end structure which are
2165 * neither logically nor physically contiguous. Although it may seem as an
2166 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2167 * case as we need to track IO to all buffers underlying a page in one io_end.
2169 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2171 struct address_space *mapping = mpd->inode->i_mapping;
2172 struct pagevec pvec;
2173 unsigned int nr_pages;
2174 long left = mpd->wbc->nr_to_write;
2175 pgoff_t index = mpd->first_page;
2176 pgoff_t end = mpd->last_page;
2179 int blkbits = mpd->inode->i_blkbits;
2181 struct buffer_head *head;
2183 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2184 tag = PAGECACHE_TAG_TOWRITE;
2186 tag = PAGECACHE_TAG_DIRTY;
2188 pagevec_init(&pvec, 0);
2190 mpd->next_page = index;
2191 while (index <= end) {
2192 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2193 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2197 for (i = 0; i < nr_pages; i++) {
2198 struct page *page = pvec.pages[i];
2201 * At this point, the page may be truncated or
2202 * invalidated (changing page->mapping to NULL), or
2203 * even swizzled back from swapper_space to tmpfs file
2204 * mapping. However, page->index will not change
2205 * because we have a reference on the page.
2207 if (page->index > end)
2211 * Accumulated enough dirty pages? This doesn't apply
2212 * to WB_SYNC_ALL mode. For integrity sync we have to
2213 * keep going because someone may be concurrently
2214 * dirtying pages, and we might have synced a lot of
2215 * newly appeared dirty pages, but have not synced all
2216 * of the old dirty pages.
2218 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2221 /* If we can't merge this page, we are done. */
2222 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2227 * If the page is no longer dirty, or its mapping no
2228 * longer corresponds to inode we are writing (which
2229 * means it has been truncated or invalidated), or the
2230 * page is already under writeback and we are not doing
2231 * a data integrity writeback, skip the page
2233 if (!PageDirty(page) ||
2234 (PageWriteback(page) &&
2235 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2236 unlikely(page->mapping != mapping)) {
2241 wait_on_page_writeback(page);
2242 BUG_ON(PageWriteback(page));
2244 if (mpd->map.m_len == 0)
2245 mpd->first_page = page->index;
2246 mpd->next_page = page->index + 1;
2247 /* Add all dirty buffers to mpd */
2248 lblk = ((ext4_lblk_t)page->index) <<
2249 (PAGE_CACHE_SHIFT - blkbits);
2250 head = page_buffers(page);
2251 err = mpage_process_page_bufs(mpd, head, head, lblk);
2257 pagevec_release(&pvec);
2262 pagevec_release(&pvec);
2266 static int __writepage(struct page *page, struct writeback_control *wbc,
2269 struct address_space *mapping = data;
2270 int ret = ext4_writepage(page, wbc);
2271 mapping_set_error(mapping, ret);
2275 static int ext4_writepages(struct address_space *mapping,
2276 struct writeback_control *wbc)
2278 pgoff_t writeback_index = 0;
2279 long nr_to_write = wbc->nr_to_write;
2280 int range_whole = 0;
2282 handle_t *handle = NULL;
2283 struct mpage_da_data mpd;
2284 struct inode *inode = mapping->host;
2285 int needed_blocks, rsv_blocks = 0, ret = 0;
2286 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2288 struct blk_plug plug;
2289 bool give_up_on_write = false;
2291 trace_ext4_writepages(inode, wbc);
2294 * No pages to write? This is mainly a kludge to avoid starting
2295 * a transaction for special inodes like journal inode on last iput()
2296 * because that could violate lock ordering on umount
2298 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2299 goto out_writepages;
2301 if (ext4_should_journal_data(inode)) {
2302 struct blk_plug plug;
2304 blk_start_plug(&plug);
2305 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2306 blk_finish_plug(&plug);
2307 goto out_writepages;
2311 * If the filesystem has aborted, it is read-only, so return
2312 * right away instead of dumping stack traces later on that
2313 * will obscure the real source of the problem. We test
2314 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2315 * the latter could be true if the filesystem is mounted
2316 * read-only, and in that case, ext4_writepages should
2317 * *never* be called, so if that ever happens, we would want
2320 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2322 goto out_writepages;
2325 if (ext4_should_dioread_nolock(inode)) {
2327 * We may need to convert up to one extent per block in
2328 * the page and we may dirty the inode.
2330 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2334 * If we have inline data and arrive here, it means that
2335 * we will soon create the block for the 1st page, so
2336 * we'd better clear the inline data here.
2338 if (ext4_has_inline_data(inode)) {
2339 /* Just inode will be modified... */
2340 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2341 if (IS_ERR(handle)) {
2342 ret = PTR_ERR(handle);
2343 goto out_writepages;
2345 BUG_ON(ext4_test_inode_state(inode,
2346 EXT4_STATE_MAY_INLINE_DATA));
2347 ext4_destroy_inline_data(handle, inode);
2348 ext4_journal_stop(handle);
2351 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2354 if (wbc->range_cyclic) {
2355 writeback_index = mapping->writeback_index;
2356 if (writeback_index)
2358 mpd.first_page = writeback_index;
2361 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2362 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2367 ext4_io_submit_init(&mpd.io_submit, wbc);
2369 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2370 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2372 blk_start_plug(&plug);
2373 while (!done && mpd.first_page <= mpd.last_page) {
2374 /* For each extent of pages we use new io_end */
2375 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2376 if (!mpd.io_submit.io_end) {
2382 * We have two constraints: We find one extent to map and we
2383 * must always write out whole page (makes a difference when
2384 * blocksize < pagesize) so that we don't block on IO when we
2385 * try to write out the rest of the page. Journalled mode is
2386 * not supported by delalloc.
2388 BUG_ON(ext4_should_journal_data(inode));
2389 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2391 /* start a new transaction */
2392 handle = ext4_journal_start_with_reserve(inode,
2393 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2394 if (IS_ERR(handle)) {
2395 ret = PTR_ERR(handle);
2396 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2397 "%ld pages, ino %lu; err %d", __func__,
2398 wbc->nr_to_write, inode->i_ino, ret);
2399 /* Release allocated io_end */
2400 ext4_put_io_end(mpd.io_submit.io_end);
2404 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2405 ret = mpage_prepare_extent_to_map(&mpd);
2408 ret = mpage_map_and_submit_extent(handle, &mpd,
2412 * We scanned the whole range (or exhausted
2413 * nr_to_write), submitted what was mapped and
2414 * didn't find anything needing mapping. We are
2420 ext4_journal_stop(handle);
2421 /* Submit prepared bio */
2422 ext4_io_submit(&mpd.io_submit);
2423 /* Unlock pages we didn't use */
2424 mpage_release_unused_pages(&mpd, give_up_on_write);
2425 /* Drop our io_end reference we got from init */
2426 ext4_put_io_end(mpd.io_submit.io_end);
2428 if (ret == -ENOSPC && sbi->s_journal) {
2430 * Commit the transaction which would
2431 * free blocks released in the transaction
2434 jbd2_journal_force_commit_nested(sbi->s_journal);
2438 /* Fatal error - ENOMEM, EIO... */
2442 blk_finish_plug(&plug);
2443 if (!ret && !cycled && wbc->nr_to_write > 0) {
2445 mpd.last_page = writeback_index - 1;
2451 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2453 * Set the writeback_index so that range_cyclic
2454 * mode will write it back later
2456 mapping->writeback_index = mpd.first_page;
2459 trace_ext4_writepages_result(inode, wbc, ret,
2460 nr_to_write - wbc->nr_to_write);
2464 static int ext4_nonda_switch(struct super_block *sb)
2466 s64 free_clusters, dirty_clusters;
2467 struct ext4_sb_info *sbi = EXT4_SB(sb);
2470 * switch to non delalloc mode if we are running low
2471 * on free block. The free block accounting via percpu
2472 * counters can get slightly wrong with percpu_counter_batch getting
2473 * accumulated on each CPU without updating global counters
2474 * Delalloc need an accurate free block accounting. So switch
2475 * to non delalloc when we are near to error range.
2478 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2480 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2482 * Start pushing delalloc when 1/2 of free blocks are dirty.
2484 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2485 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2487 if (2 * free_clusters < 3 * dirty_clusters ||
2488 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2490 * free block count is less than 150% of dirty blocks
2491 * or free blocks is less than watermark
2498 /* We always reserve for an inode update; the superblock could be there too */
2499 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2501 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
2502 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2505 if (pos + len <= 0x7fffffffULL)
2508 /* We might need to update the superblock to set LARGE_FILE */
2512 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2513 loff_t pos, unsigned len, unsigned flags,
2514 struct page **pagep, void **fsdata)
2516 int ret, retries = 0;
2519 struct inode *inode = mapping->host;
2522 index = pos >> PAGE_CACHE_SHIFT;
2524 if (ext4_nonda_switch(inode->i_sb)) {
2525 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2526 return ext4_write_begin(file, mapping, pos,
2527 len, flags, pagep, fsdata);
2529 *fsdata = (void *)0;
2530 trace_ext4_da_write_begin(inode, pos, len, flags);
2532 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2533 ret = ext4_da_write_inline_data_begin(mapping, inode,
2543 * grab_cache_page_write_begin() can take a long time if the
2544 * system is thrashing due to memory pressure, or if the page
2545 * is being written back. So grab it first before we start
2546 * the transaction handle. This also allows us to allocate
2547 * the page (if needed) without using GFP_NOFS.
2550 page = grab_cache_page_write_begin(mapping, index, flags);
2556 * With delayed allocation, we don't log the i_disksize update
2557 * if there is delayed block allocation. But we still need
2558 * to journalling the i_disksize update if writes to the end
2559 * of file which has an already mapped buffer.
2562 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2563 ext4_da_write_credits(inode, pos, len));
2564 if (IS_ERR(handle)) {
2565 page_cache_release(page);
2566 return PTR_ERR(handle);
2570 if (page->mapping != mapping) {
2571 /* The page got truncated from under us */
2573 page_cache_release(page);
2574 ext4_journal_stop(handle);
2577 /* In case writeback began while the page was unlocked */
2578 wait_for_stable_page(page);
2580 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2583 ext4_journal_stop(handle);
2585 * block_write_begin may have instantiated a few blocks
2586 * outside i_size. Trim these off again. Don't need
2587 * i_size_read because we hold i_mutex.
2589 if (pos + len > inode->i_size)
2590 ext4_truncate_failed_write(inode);
2592 if (ret == -ENOSPC &&
2593 ext4_should_retry_alloc(inode->i_sb, &retries))
2596 page_cache_release(page);
2605 * Check if we should update i_disksize
2606 * when write to the end of file but not require block allocation
2608 static int ext4_da_should_update_i_disksize(struct page *page,
2609 unsigned long offset)
2611 struct buffer_head *bh;
2612 struct inode *inode = page->mapping->host;
2616 bh = page_buffers(page);
2617 idx = offset >> inode->i_blkbits;
2619 for (i = 0; i < idx; i++)
2620 bh = bh->b_this_page;
2622 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2627 static int ext4_da_write_end(struct file *file,
2628 struct address_space *mapping,
2629 loff_t pos, unsigned len, unsigned copied,
2630 struct page *page, void *fsdata)
2632 struct inode *inode = mapping->host;
2634 handle_t *handle = ext4_journal_current_handle();
2636 unsigned long start, end;
2637 int write_mode = (int)(unsigned long)fsdata;
2639 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2640 return ext4_write_end(file, mapping, pos,
2641 len, copied, page, fsdata);
2643 trace_ext4_da_write_end(inode, pos, len, copied);
2644 start = pos & (PAGE_CACHE_SIZE - 1);
2645 end = start + copied - 1;
2648 * generic_write_end() will run mark_inode_dirty() if i_size
2649 * changes. So let's piggyback the i_disksize mark_inode_dirty
2652 new_i_size = pos + copied;
2653 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2654 if (ext4_has_inline_data(inode) ||
2655 ext4_da_should_update_i_disksize(page, end)) {
2656 ext4_update_i_disksize(inode, new_i_size);
2657 /* We need to mark inode dirty even if
2658 * new_i_size is less that inode->i_size
2659 * bu greater than i_disksize.(hint delalloc)
2661 ext4_mark_inode_dirty(handle, inode);
2665 if (write_mode != CONVERT_INLINE_DATA &&
2666 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2667 ext4_has_inline_data(inode))
2668 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2671 ret2 = generic_write_end(file, mapping, pos, len, copied,
2677 ret2 = ext4_journal_stop(handle);
2681 return ret ? ret : copied;
2684 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2685 unsigned int length)
2688 * Drop reserved blocks
2690 BUG_ON(!PageLocked(page));
2691 if (!page_has_buffers(page))
2694 ext4_da_page_release_reservation(page, offset, length);
2697 ext4_invalidatepage(page, offset, length);
2703 * Force all delayed allocation blocks to be allocated for a given inode.
2705 int ext4_alloc_da_blocks(struct inode *inode)
2707 trace_ext4_alloc_da_blocks(inode);
2709 if (!EXT4_I(inode)->i_reserved_data_blocks)
2713 * We do something simple for now. The filemap_flush() will
2714 * also start triggering a write of the data blocks, which is
2715 * not strictly speaking necessary (and for users of
2716 * laptop_mode, not even desirable). However, to do otherwise
2717 * would require replicating code paths in:
2719 * ext4_writepages() ->
2720 * write_cache_pages() ---> (via passed in callback function)
2721 * __mpage_da_writepage() -->
2722 * mpage_add_bh_to_extent()
2723 * mpage_da_map_blocks()
2725 * The problem is that write_cache_pages(), located in
2726 * mm/page-writeback.c, marks pages clean in preparation for
2727 * doing I/O, which is not desirable if we're not planning on
2730 * We could call write_cache_pages(), and then redirty all of
2731 * the pages by calling redirty_page_for_writepage() but that
2732 * would be ugly in the extreme. So instead we would need to
2733 * replicate parts of the code in the above functions,
2734 * simplifying them because we wouldn't actually intend to
2735 * write out the pages, but rather only collect contiguous
2736 * logical block extents, call the multi-block allocator, and
2737 * then update the buffer heads with the block allocations.
2739 * For now, though, we'll cheat by calling filemap_flush(),
2740 * which will map the blocks, and start the I/O, but not
2741 * actually wait for the I/O to complete.
2743 return filemap_flush(inode->i_mapping);
2747 * bmap() is special. It gets used by applications such as lilo and by
2748 * the swapper to find the on-disk block of a specific piece of data.
2750 * Naturally, this is dangerous if the block concerned is still in the
2751 * journal. If somebody makes a swapfile on an ext4 data-journaling
2752 * filesystem and enables swap, then they may get a nasty shock when the
2753 * data getting swapped to that swapfile suddenly gets overwritten by
2754 * the original zero's written out previously to the journal and
2755 * awaiting writeback in the kernel's buffer cache.
2757 * So, if we see any bmap calls here on a modified, data-journaled file,
2758 * take extra steps to flush any blocks which might be in the cache.
2760 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2762 struct inode *inode = mapping->host;
2767 * We can get here for an inline file via the FIBMAP ioctl
2769 if (ext4_has_inline_data(inode))
2772 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2773 test_opt(inode->i_sb, DELALLOC)) {
2775 * With delalloc we want to sync the file
2776 * so that we can make sure we allocate
2779 filemap_write_and_wait(mapping);
2782 if (EXT4_JOURNAL(inode) &&
2783 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2785 * This is a REALLY heavyweight approach, but the use of
2786 * bmap on dirty files is expected to be extremely rare:
2787 * only if we run lilo or swapon on a freshly made file
2788 * do we expect this to happen.
2790 * (bmap requires CAP_SYS_RAWIO so this does not
2791 * represent an unprivileged user DOS attack --- we'd be
2792 * in trouble if mortal users could trigger this path at
2795 * NB. EXT4_STATE_JDATA is not set on files other than
2796 * regular files. If somebody wants to bmap a directory
2797 * or symlink and gets confused because the buffer
2798 * hasn't yet been flushed to disk, they deserve
2799 * everything they get.
2802 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2803 journal = EXT4_JOURNAL(inode);
2804 jbd2_journal_lock_updates(journal);
2805 err = jbd2_journal_flush(journal);
2806 jbd2_journal_unlock_updates(journal);
2812 return generic_block_bmap(mapping, block, ext4_get_block);
2815 static int ext4_readpage(struct file *file, struct page *page)
2818 struct inode *inode = page->mapping->host;
2820 trace_ext4_readpage(page);
2822 if (ext4_has_inline_data(inode))
2823 ret = ext4_readpage_inline(inode, page);
2826 return mpage_readpage(page, ext4_get_block);
2832 ext4_readpages(struct file *file, struct address_space *mapping,
2833 struct list_head *pages, unsigned nr_pages)
2835 struct inode *inode = mapping->host;
2837 /* If the file has inline data, no need to do readpages. */
2838 if (ext4_has_inline_data(inode))
2841 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2844 static void ext4_invalidatepage(struct page *page, unsigned int offset,
2845 unsigned int length)
2847 trace_ext4_invalidatepage(page, offset, length);
2849 /* No journalling happens on data buffers when this function is used */
2850 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2852 block_invalidatepage(page, offset, length);
2855 static int __ext4_journalled_invalidatepage(struct page *page,
2856 unsigned int offset,
2857 unsigned int length)
2859 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2861 trace_ext4_journalled_invalidatepage(page, offset, length);
2864 * If it's a full truncate we just forget about the pending dirtying
2866 if (offset == 0 && length == PAGE_CACHE_SIZE)
2867 ClearPageChecked(page);
2869 return jbd2_journal_invalidatepage(journal, page, offset, length);
2872 /* Wrapper for aops... */
2873 static void ext4_journalled_invalidatepage(struct page *page,
2874 unsigned int offset,
2875 unsigned int length)
2877 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
2880 static int ext4_releasepage(struct page *page, gfp_t wait)
2882 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2884 trace_ext4_releasepage(page);
2886 /* Page has dirty journalled data -> cannot release */
2887 if (PageChecked(page))
2890 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2892 return try_to_free_buffers(page);
2896 * ext4_get_block used when preparing for a DIO write or buffer write.
2897 * We allocate an uinitialized extent if blocks haven't been allocated.
2898 * The extent will be converted to initialized after the IO is complete.
2900 int ext4_get_block_write(struct inode *inode, sector_t iblock,
2901 struct buffer_head *bh_result, int create)
2903 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2904 inode->i_ino, create);
2905 return _ext4_get_block(inode, iblock, bh_result,
2906 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2909 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2910 struct buffer_head *bh_result, int create)
2912 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2913 inode->i_ino, create);
2914 return _ext4_get_block(inode, iblock, bh_result,
2915 EXT4_GET_BLOCKS_NO_LOCK);
2918 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2919 ssize_t size, void *private)
2921 ext4_io_end_t *io_end = iocb->private;
2923 /* if not async direct IO just return */
2927 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2928 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2929 iocb->private, io_end->inode->i_ino, iocb, offset,
2932 iocb->private = NULL;
2933 io_end->offset = offset;
2934 io_end->size = size;
2935 ext4_put_io_end(io_end);
2939 * For ext4 extent files, ext4 will do direct-io write to holes,
2940 * preallocated extents, and those write extend the file, no need to
2941 * fall back to buffered IO.
2943 * For holes, we fallocate those blocks, mark them as unwritten
2944 * If those blocks were preallocated, we mark sure they are split, but
2945 * still keep the range to write as unwritten.
2947 * The unwritten extents will be converted to written when DIO is completed.
2948 * For async direct IO, since the IO may still pending when return, we
2949 * set up an end_io call back function, which will do the conversion
2950 * when async direct IO completed.
2952 * If the O_DIRECT write will extend the file then add this inode to the
2953 * orphan list. So recovery will truncate it back to the original size
2954 * if the machine crashes during the write.
2957 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2958 struct iov_iter *iter, loff_t offset)
2960 struct file *file = iocb->ki_filp;
2961 struct inode *inode = file->f_mapping->host;
2963 size_t count = iov_iter_count(iter);
2965 get_block_t *get_block_func = NULL;
2967 loff_t final_size = offset + count;
2968 ext4_io_end_t *io_end = NULL;
2970 /* Use the old path for reads and writes beyond i_size. */
2971 if (rw != WRITE || final_size > inode->i_size)
2972 return ext4_ind_direct_IO(rw, iocb, iter, offset);
2974 BUG_ON(iocb->private == NULL);
2977 * Make all waiters for direct IO properly wait also for extent
2978 * conversion. This also disallows race between truncate() and
2979 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
2982 atomic_inc(&inode->i_dio_count);
2984 /* If we do a overwrite dio, i_mutex locking can be released */
2985 overwrite = *((int *)iocb->private);
2988 down_read(&EXT4_I(inode)->i_data_sem);
2989 mutex_unlock(&inode->i_mutex);
2993 * We could direct write to holes and fallocate.
2995 * Allocated blocks to fill the hole are marked as
2996 * unwritten to prevent parallel buffered read to expose
2997 * the stale data before DIO complete the data IO.
2999 * As to previously fallocated extents, ext4 get_block will
3000 * just simply mark the buffer mapped but still keep the
3001 * extents unwritten.
3003 * For non AIO case, we will convert those unwritten extents
3004 * to written after return back from blockdev_direct_IO.
3006 * For async DIO, the conversion needs to be deferred when the
3007 * IO is completed. The ext4 end_io callback function will be
3008 * called to take care of the conversion work. Here for async
3009 * case, we allocate an io_end structure to hook to the iocb.
3011 iocb->private = NULL;
3012 ext4_inode_aio_set(inode, NULL);
3013 if (!is_sync_kiocb(iocb)) {
3014 io_end = ext4_init_io_end(inode, GFP_NOFS);
3020 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3022 iocb->private = ext4_get_io_end(io_end);
3024 * we save the io structure for current async direct
3025 * IO, so that later ext4_map_blocks() could flag the
3026 * io structure whether there is a unwritten extents
3027 * needs to be converted when IO is completed.
3029 ext4_inode_aio_set(inode, io_end);
3033 get_block_func = ext4_get_block_write_nolock;
3035 get_block_func = ext4_get_block_write;
3036 dio_flags = DIO_LOCKING;
3038 ret = __blockdev_direct_IO(rw, iocb, inode,
3039 inode->i_sb->s_bdev, iter,
3047 * Put our reference to io_end. This can free the io_end structure e.g.
3048 * in sync IO case or in case of error. It can even perform extent
3049 * conversion if all bios we submitted finished before we got here.
3050 * Note that in that case iocb->private can be already set to NULL
3054 ext4_inode_aio_set(inode, NULL);
3055 ext4_put_io_end(io_end);
3057 * When no IO was submitted ext4_end_io_dio() was not
3058 * called so we have to put iocb's reference.
3060 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3061 WARN_ON(iocb->private != io_end);
3062 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3063 ext4_put_io_end(io_end);
3064 iocb->private = NULL;
3067 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3068 EXT4_STATE_DIO_UNWRITTEN)) {
3071 * for non AIO case, since the IO is already
3072 * completed, we could do the conversion right here
3074 err = ext4_convert_unwritten_extents(NULL, inode,
3078 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3083 inode_dio_done(inode);
3084 /* take i_mutex locking again if we do a ovewrite dio */
3086 up_read(&EXT4_I(inode)->i_data_sem);
3087 mutex_lock(&inode->i_mutex);
3093 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3094 struct iov_iter *iter, loff_t offset)
3096 struct file *file = iocb->ki_filp;
3097 struct inode *inode = file->f_mapping->host;
3098 size_t count = iov_iter_count(iter);
3102 * If we are doing data journalling we don't support O_DIRECT
3104 if (ext4_should_journal_data(inode))
3107 /* Let buffer I/O handle the inline data case. */
3108 if (ext4_has_inline_data(inode))
3111 trace_ext4_direct_IO_enter(inode, offset, count, rw);
3112 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3113 ret = ext4_ext_direct_IO(rw, iocb, iter, offset);
3115 ret = ext4_ind_direct_IO(rw, iocb, iter, offset);
3116 trace_ext4_direct_IO_exit(inode, offset, count, rw, ret);
3121 * Pages can be marked dirty completely asynchronously from ext4's journalling
3122 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3123 * much here because ->set_page_dirty is called under VFS locks. The page is
3124 * not necessarily locked.
3126 * We cannot just dirty the page and leave attached buffers clean, because the
3127 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3128 * or jbddirty because all the journalling code will explode.
3130 * So what we do is to mark the page "pending dirty" and next time writepage
3131 * is called, propagate that into the buffers appropriately.
3133 static int ext4_journalled_set_page_dirty(struct page *page)
3135 SetPageChecked(page);
3136 return __set_page_dirty_nobuffers(page);
3139 static const struct address_space_operations ext4_aops = {
3140 .readpage = ext4_readpage,
3141 .readpages = ext4_readpages,
3142 .writepage = ext4_writepage,
3143 .writepages = ext4_writepages,
3144 .write_begin = ext4_write_begin,
3145 .write_end = ext4_write_end,
3147 .invalidatepage = ext4_invalidatepage,
3148 .releasepage = ext4_releasepage,
3149 .direct_IO = ext4_direct_IO,
3150 .migratepage = buffer_migrate_page,
3151 .is_partially_uptodate = block_is_partially_uptodate,
3152 .error_remove_page = generic_error_remove_page,
3155 static const struct address_space_operations ext4_journalled_aops = {
3156 .readpage = ext4_readpage,
3157 .readpages = ext4_readpages,
3158 .writepage = ext4_writepage,
3159 .writepages = ext4_writepages,
3160 .write_begin = ext4_write_begin,
3161 .write_end = ext4_journalled_write_end,
3162 .set_page_dirty = ext4_journalled_set_page_dirty,
3164 .invalidatepage = ext4_journalled_invalidatepage,
3165 .releasepage = ext4_releasepage,
3166 .direct_IO = ext4_direct_IO,
3167 .is_partially_uptodate = block_is_partially_uptodate,
3168 .error_remove_page = generic_error_remove_page,
3171 static const struct address_space_operations ext4_da_aops = {
3172 .readpage = ext4_readpage,
3173 .readpages = ext4_readpages,
3174 .writepage = ext4_writepage,
3175 .writepages = ext4_writepages,
3176 .write_begin = ext4_da_write_begin,
3177 .write_end = ext4_da_write_end,
3179 .invalidatepage = ext4_da_invalidatepage,
3180 .releasepage = ext4_releasepage,
3181 .direct_IO = ext4_direct_IO,
3182 .migratepage = buffer_migrate_page,
3183 .is_partially_uptodate = block_is_partially_uptodate,
3184 .error_remove_page = generic_error_remove_page,
3187 void ext4_set_aops(struct inode *inode)
3189 switch (ext4_inode_journal_mode(inode)) {
3190 case EXT4_INODE_ORDERED_DATA_MODE:
3191 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3193 case EXT4_INODE_WRITEBACK_DATA_MODE:
3194 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3196 case EXT4_INODE_JOURNAL_DATA_MODE:
3197 inode->i_mapping->a_ops = &ext4_journalled_aops;
3202 if (test_opt(inode->i_sb, DELALLOC))
3203 inode->i_mapping->a_ops = &ext4_da_aops;
3205 inode->i_mapping->a_ops = &ext4_aops;
3209 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3210 * starting from file offset 'from'. The range to be zero'd must
3211 * be contained with in one block. If the specified range exceeds
3212 * the end of the block it will be shortened to end of the block
3213 * that cooresponds to 'from'
3215 static int ext4_block_zero_page_range(handle_t *handle,
3216 struct address_space *mapping, loff_t from, loff_t length)
3218 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3219 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3220 unsigned blocksize, max, pos;
3222 struct inode *inode = mapping->host;
3223 struct buffer_head *bh;
3227 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3228 mapping_gfp_mask(mapping) & ~__GFP_FS);
3232 blocksize = inode->i_sb->s_blocksize;
3233 max = blocksize - (offset & (blocksize - 1));
3236 * correct length if it does not fall between
3237 * 'from' and the end of the block
3239 if (length > max || length < 0)
3242 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3244 if (!page_has_buffers(page))
3245 create_empty_buffers(page, blocksize, 0);
3247 /* Find the buffer that contains "offset" */
3248 bh = page_buffers(page);
3250 while (offset >= pos) {
3251 bh = bh->b_this_page;
3255 if (buffer_freed(bh)) {
3256 BUFFER_TRACE(bh, "freed: skip");
3259 if (!buffer_mapped(bh)) {
3260 BUFFER_TRACE(bh, "unmapped");
3261 ext4_get_block(inode, iblock, bh, 0);
3262 /* unmapped? It's a hole - nothing to do */
3263 if (!buffer_mapped(bh)) {
3264 BUFFER_TRACE(bh, "still unmapped");
3269 /* Ok, it's mapped. Make sure it's up-to-date */
3270 if (PageUptodate(page))
3271 set_buffer_uptodate(bh);
3273 if (!buffer_uptodate(bh)) {
3275 ll_rw_block(READ, 1, &bh);
3277 /* Uhhuh. Read error. Complain and punt. */
3278 if (!buffer_uptodate(bh))
3281 if (ext4_should_journal_data(inode)) {
3282 BUFFER_TRACE(bh, "get write access");
3283 err = ext4_journal_get_write_access(handle, bh);
3287 zero_user(page, offset, length);
3288 BUFFER_TRACE(bh, "zeroed end of block");
3290 if (ext4_should_journal_data(inode)) {
3291 err = ext4_handle_dirty_metadata(handle, inode, bh);
3294 mark_buffer_dirty(bh);
3295 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3296 err = ext4_jbd2_file_inode(handle, inode);
3301 page_cache_release(page);
3306 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3307 * up to the end of the block which corresponds to `from'.
3308 * This required during truncate. We need to physically zero the tail end
3309 * of that block so it doesn't yield old data if the file is later grown.
3311 static int ext4_block_truncate_page(handle_t *handle,
3312 struct address_space *mapping, loff_t from)
3314 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3317 struct inode *inode = mapping->host;
3319 blocksize = inode->i_sb->s_blocksize;
3320 length = blocksize - (offset & (blocksize - 1));
3322 return ext4_block_zero_page_range(handle, mapping, from, length);
3325 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3326 loff_t lstart, loff_t length)
3328 struct super_block *sb = inode->i_sb;
3329 struct address_space *mapping = inode->i_mapping;
3330 unsigned partial_start, partial_end;
3331 ext4_fsblk_t start, end;
3332 loff_t byte_end = (lstart + length - 1);
3335 partial_start = lstart & (sb->s_blocksize - 1);
3336 partial_end = byte_end & (sb->s_blocksize - 1);
3338 start = lstart >> sb->s_blocksize_bits;
3339 end = byte_end >> sb->s_blocksize_bits;
3341 /* Handle partial zero within the single block */
3343 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3344 err = ext4_block_zero_page_range(handle, mapping,
3348 /* Handle partial zero out on the start of the range */
3349 if (partial_start) {
3350 err = ext4_block_zero_page_range(handle, mapping,
3351 lstart, sb->s_blocksize);
3355 /* Handle partial zero out on the end of the range */
3356 if (partial_end != sb->s_blocksize - 1)
3357 err = ext4_block_zero_page_range(handle, mapping,
3358 byte_end - partial_end,
3363 int ext4_can_truncate(struct inode *inode)
3365 if (S_ISREG(inode->i_mode))
3367 if (S_ISDIR(inode->i_mode))
3369 if (S_ISLNK(inode->i_mode))
3370 return !ext4_inode_is_fast_symlink(inode);
3375 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3376 * associated with the given offset and length
3378 * @inode: File inode
3379 * @offset: The offset where the hole will begin
3380 * @len: The length of the hole
3382 * Returns: 0 on success or negative on failure
3385 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3387 struct super_block *sb = inode->i_sb;
3388 ext4_lblk_t first_block, stop_block;
3389 struct address_space *mapping = inode->i_mapping;
3390 loff_t first_block_offset, last_block_offset;
3392 unsigned int credits;
3395 if (!S_ISREG(inode->i_mode))
3398 trace_ext4_punch_hole(inode, offset, length, 0);
3401 * Write out all dirty pages to avoid race conditions
3402 * Then release them.
3404 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3405 ret = filemap_write_and_wait_range(mapping, offset,
3406 offset + length - 1);
3411 mutex_lock(&inode->i_mutex);
3413 /* No need to punch hole beyond i_size */
3414 if (offset >= inode->i_size)
3418 * If the hole extends beyond i_size, set the hole
3419 * to end after the page that contains i_size
3421 if (offset + length > inode->i_size) {
3422 length = inode->i_size +
3423 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3427 if (offset & (sb->s_blocksize - 1) ||
3428 (offset + length) & (sb->s_blocksize - 1)) {
3430 * Attach jinode to inode for jbd2 if we do any zeroing of
3433 ret = ext4_inode_attach_jinode(inode);
3439 first_block_offset = round_up(offset, sb->s_blocksize);
3440 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3442 /* Now release the pages and zero block aligned part of pages*/
3443 if (last_block_offset > first_block_offset)
3444 truncate_pagecache_range(inode, first_block_offset,
3447 /* Wait all existing dio workers, newcomers will block on i_mutex */
3448 ext4_inode_block_unlocked_dio(inode);
3449 inode_dio_wait(inode);
3451 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3452 credits = ext4_writepage_trans_blocks(inode);
3454 credits = ext4_blocks_for_truncate(inode);
3455 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3456 if (IS_ERR(handle)) {
3457 ret = PTR_ERR(handle);
3458 ext4_std_error(sb, ret);
3462 ret = ext4_zero_partial_blocks(handle, inode, offset,
3467 first_block = (offset + sb->s_blocksize - 1) >>
3468 EXT4_BLOCK_SIZE_BITS(sb);
3469 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3471 /* If there are no blocks to remove, return now */
3472 if (first_block >= stop_block)
3475 down_write(&EXT4_I(inode)->i_data_sem);
3476 ext4_discard_preallocations(inode);
3478 ret = ext4_es_remove_extent(inode, first_block,
3479 stop_block - first_block);
3481 up_write(&EXT4_I(inode)->i_data_sem);
3485 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3486 ret = ext4_ext_remove_space(inode, first_block,
3489 ret = ext4_ind_remove_space(handle, inode, first_block,
3492 up_write(&EXT4_I(inode)->i_data_sem);
3494 ext4_handle_sync(handle);
3496 /* Now release the pages again to reduce race window */
3497 if (last_block_offset > first_block_offset)
3498 truncate_pagecache_range(inode, first_block_offset,
3501 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3502 ext4_mark_inode_dirty(handle, inode);
3504 ext4_journal_stop(handle);
3506 ext4_inode_resume_unlocked_dio(inode);
3508 mutex_unlock(&inode->i_mutex);
3512 int ext4_inode_attach_jinode(struct inode *inode)
3514 struct ext4_inode_info *ei = EXT4_I(inode);
3515 struct jbd2_inode *jinode;
3517 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3520 jinode = jbd2_alloc_inode(GFP_KERNEL);
3521 spin_lock(&inode->i_lock);
3524 spin_unlock(&inode->i_lock);
3527 ei->jinode = jinode;
3528 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3531 spin_unlock(&inode->i_lock);
3532 if (unlikely(jinode != NULL))
3533 jbd2_free_inode(jinode);
3540 * We block out ext4_get_block() block instantiations across the entire
3541 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3542 * simultaneously on behalf of the same inode.
3544 * As we work through the truncate and commit bits of it to the journal there
3545 * is one core, guiding principle: the file's tree must always be consistent on
3546 * disk. We must be able to restart the truncate after a crash.
3548 * The file's tree may be transiently inconsistent in memory (although it
3549 * probably isn't), but whenever we close off and commit a journal transaction,
3550 * the contents of (the filesystem + the journal) must be consistent and
3551 * restartable. It's pretty simple, really: bottom up, right to left (although
3552 * left-to-right works OK too).
3554 * Note that at recovery time, journal replay occurs *before* the restart of
3555 * truncate against the orphan inode list.
3557 * The committed inode has the new, desired i_size (which is the same as
3558 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3559 * that this inode's truncate did not complete and it will again call
3560 * ext4_truncate() to have another go. So there will be instantiated blocks
3561 * to the right of the truncation point in a crashed ext4 filesystem. But
3562 * that's fine - as long as they are linked from the inode, the post-crash
3563 * ext4_truncate() run will find them and release them.
3565 void ext4_truncate(struct inode *inode)
3567 struct ext4_inode_info *ei = EXT4_I(inode);
3568 unsigned int credits;
3570 struct address_space *mapping = inode->i_mapping;
3573 * There is a possibility that we're either freeing the inode
3574 * or it's a completely new inode. In those cases we might not
3575 * have i_mutex locked because it's not necessary.
3577 if (!(inode->i_state & (I_NEW|I_FREEING)))
3578 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3579 trace_ext4_truncate_enter(inode);
3581 if (!ext4_can_truncate(inode))
3584 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3586 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3587 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3589 if (ext4_has_inline_data(inode)) {
3592 ext4_inline_data_truncate(inode, &has_inline);
3597 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3598 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3599 if (ext4_inode_attach_jinode(inode) < 0)
3603 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3604 credits = ext4_writepage_trans_blocks(inode);
3606 credits = ext4_blocks_for_truncate(inode);
3608 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3609 if (IS_ERR(handle)) {
3610 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3614 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3615 ext4_block_truncate_page(handle, mapping, inode->i_size);
3618 * We add the inode to the orphan list, so that if this
3619 * truncate spans multiple transactions, and we crash, we will
3620 * resume the truncate when the filesystem recovers. It also
3621 * marks the inode dirty, to catch the new size.
3623 * Implication: the file must always be in a sane, consistent
3624 * truncatable state while each transaction commits.
3626 if (ext4_orphan_add(handle, inode))
3629 down_write(&EXT4_I(inode)->i_data_sem);
3631 ext4_discard_preallocations(inode);
3633 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3634 ext4_ext_truncate(handle, inode);
3636 ext4_ind_truncate(handle, inode);
3638 up_write(&ei->i_data_sem);
3641 ext4_handle_sync(handle);
3645 * If this was a simple ftruncate() and the file will remain alive,
3646 * then we need to clear up the orphan record which we created above.
3647 * However, if this was a real unlink then we were called by
3648 * ext4_delete_inode(), and we allow that function to clean up the
3649 * orphan info for us.
3652 ext4_orphan_del(handle, inode);
3654 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3655 ext4_mark_inode_dirty(handle, inode);
3656 ext4_journal_stop(handle);
3658 trace_ext4_truncate_exit(inode);
3662 * ext4_get_inode_loc returns with an extra refcount against the inode's
3663 * underlying buffer_head on success. If 'in_mem' is true, we have all
3664 * data in memory that is needed to recreate the on-disk version of this
3667 static int __ext4_get_inode_loc(struct inode *inode,
3668 struct ext4_iloc *iloc, int in_mem)
3670 struct ext4_group_desc *gdp;
3671 struct buffer_head *bh;
3672 struct super_block *sb = inode->i_sb;
3674 int inodes_per_block, inode_offset;
3677 if (!ext4_valid_inum(sb, inode->i_ino))
3680 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3681 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3686 * Figure out the offset within the block group inode table
3688 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3689 inode_offset = ((inode->i_ino - 1) %
3690 EXT4_INODES_PER_GROUP(sb));
3691 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3692 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3694 bh = sb_getblk(sb, block);
3697 if (!buffer_uptodate(bh)) {
3701 * If the buffer has the write error flag, we have failed
3702 * to write out another inode in the same block. In this
3703 * case, we don't have to read the block because we may
3704 * read the old inode data successfully.
3706 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3707 set_buffer_uptodate(bh);
3709 if (buffer_uptodate(bh)) {
3710 /* someone brought it uptodate while we waited */
3716 * If we have all information of the inode in memory and this
3717 * is the only valid inode in the block, we need not read the
3721 struct buffer_head *bitmap_bh;
3724 start = inode_offset & ~(inodes_per_block - 1);
3726 /* Is the inode bitmap in cache? */
3727 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3728 if (unlikely(!bitmap_bh))
3732 * If the inode bitmap isn't in cache then the
3733 * optimisation may end up performing two reads instead
3734 * of one, so skip it.
3736 if (!buffer_uptodate(bitmap_bh)) {
3740 for (i = start; i < start + inodes_per_block; i++) {
3741 if (i == inode_offset)
3743 if (ext4_test_bit(i, bitmap_bh->b_data))
3747 if (i == start + inodes_per_block) {
3748 /* all other inodes are free, so skip I/O */
3749 memset(bh->b_data, 0, bh->b_size);
3750 set_buffer_uptodate(bh);
3758 * If we need to do any I/O, try to pre-readahead extra
3759 * blocks from the inode table.
3761 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3762 ext4_fsblk_t b, end, table;
3764 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
3766 table = ext4_inode_table(sb, gdp);
3767 /* s_inode_readahead_blks is always a power of 2 */
3768 b = block & ~((ext4_fsblk_t) ra_blks - 1);
3772 num = EXT4_INODES_PER_GROUP(sb);
3773 if (ext4_has_group_desc_csum(sb))
3774 num -= ext4_itable_unused_count(sb, gdp);
3775 table += num / inodes_per_block;
3779 sb_breadahead(sb, b++);
3783 * There are other valid inodes in the buffer, this inode
3784 * has in-inode xattrs, or we don't have this inode in memory.
3785 * Read the block from disk.
3787 trace_ext4_load_inode(inode);
3789 bh->b_end_io = end_buffer_read_sync;
3790 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3792 if (!buffer_uptodate(bh)) {
3793 EXT4_ERROR_INODE_BLOCK(inode, block,
3794 "unable to read itable block");
3804 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3806 /* We have all inode data except xattrs in memory here. */
3807 return __ext4_get_inode_loc(inode, iloc,
3808 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3811 void ext4_set_inode_flags(struct inode *inode)
3813 unsigned int flags = EXT4_I(inode)->i_flags;
3814 unsigned int new_fl = 0;
3816 if (flags & EXT4_SYNC_FL)
3818 if (flags & EXT4_APPEND_FL)
3820 if (flags & EXT4_IMMUTABLE_FL)
3821 new_fl |= S_IMMUTABLE;
3822 if (flags & EXT4_NOATIME_FL)
3823 new_fl |= S_NOATIME;
3824 if (flags & EXT4_DIRSYNC_FL)
3825 new_fl |= S_DIRSYNC;
3826 inode_set_flags(inode, new_fl,
3827 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3830 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3831 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3833 unsigned int vfs_fl;
3834 unsigned long old_fl, new_fl;
3837 vfs_fl = ei->vfs_inode.i_flags;
3838 old_fl = ei->i_flags;
3839 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3840 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3842 if (vfs_fl & S_SYNC)
3843 new_fl |= EXT4_SYNC_FL;
3844 if (vfs_fl & S_APPEND)
3845 new_fl |= EXT4_APPEND_FL;
3846 if (vfs_fl & S_IMMUTABLE)
3847 new_fl |= EXT4_IMMUTABLE_FL;
3848 if (vfs_fl & S_NOATIME)
3849 new_fl |= EXT4_NOATIME_FL;
3850 if (vfs_fl & S_DIRSYNC)
3851 new_fl |= EXT4_DIRSYNC_FL;
3852 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3855 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3856 struct ext4_inode_info *ei)
3859 struct inode *inode = &(ei->vfs_inode);
3860 struct super_block *sb = inode->i_sb;
3862 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3863 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3864 /* we are using combined 48 bit field */
3865 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3866 le32_to_cpu(raw_inode->i_blocks_lo);
3867 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3868 /* i_blocks represent file system block size */
3869 return i_blocks << (inode->i_blkbits - 9);
3874 return le32_to_cpu(raw_inode->i_blocks_lo);
3878 static inline void ext4_iget_extra_inode(struct inode *inode,
3879 struct ext4_inode *raw_inode,
3880 struct ext4_inode_info *ei)
3882 __le32 *magic = (void *)raw_inode +
3883 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3884 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3885 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3886 ext4_find_inline_data_nolock(inode);
3888 EXT4_I(inode)->i_inline_off = 0;
3891 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3893 struct ext4_iloc iloc;
3894 struct ext4_inode *raw_inode;
3895 struct ext4_inode_info *ei;
3896 struct inode *inode;
3897 journal_t *journal = EXT4_SB(sb)->s_journal;
3903 inode = iget_locked(sb, ino);
3905 return ERR_PTR(-ENOMEM);
3906 if (!(inode->i_state & I_NEW))
3912 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3915 raw_inode = ext4_raw_inode(&iloc);
3917 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3918 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3919 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3920 EXT4_INODE_SIZE(inode->i_sb)) {
3921 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3922 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3923 EXT4_INODE_SIZE(inode->i_sb));
3928 ei->i_extra_isize = 0;
3930 /* Precompute checksum seed for inode metadata */
3931 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3932 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3933 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3935 __le32 inum = cpu_to_le32(inode->i_ino);
3936 __le32 gen = raw_inode->i_generation;
3937 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3939 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3943 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3944 EXT4_ERROR_INODE(inode, "checksum invalid");
3949 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3950 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3951 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3952 if (!(test_opt(inode->i_sb, NO_UID32))) {
3953 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3954 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3956 i_uid_write(inode, i_uid);
3957 i_gid_write(inode, i_gid);
3958 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3960 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3961 ei->i_inline_off = 0;
3962 ei->i_dir_start_lookup = 0;
3963 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3964 /* We now have enough fields to check if the inode was active or not.
3965 * This is needed because nfsd might try to access dead inodes
3966 * the test is that same one that e2fsck uses
3967 * NeilBrown 1999oct15
3969 if (inode->i_nlink == 0) {
3970 if ((inode->i_mode == 0 ||
3971 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
3972 ino != EXT4_BOOT_LOADER_INO) {
3973 /* this inode is deleted */
3977 /* The only unlinked inodes we let through here have
3978 * valid i_mode and are being read by the orphan
3979 * recovery code: that's fine, we're about to complete
3980 * the process of deleting those.
3981 * OR it is the EXT4_BOOT_LOADER_INO which is
3982 * not initialized on a new filesystem. */
3984 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3985 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3986 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3987 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3989 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3990 inode->i_size = ext4_isize(raw_inode);
3991 ei->i_disksize = inode->i_size;
3993 ei->i_reserved_quota = 0;
3995 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3996 ei->i_block_group = iloc.block_group;
3997 ei->i_last_alloc_group = ~0;
3999 * NOTE! The in-memory inode i_data array is in little-endian order
4000 * even on big-endian machines: we do NOT byteswap the block numbers!
4002 for (block = 0; block < EXT4_N_BLOCKS; block++)
4003 ei->i_data[block] = raw_inode->i_block[block];
4004 INIT_LIST_HEAD(&ei->i_orphan);
4007 * Set transaction id's of transactions that have to be committed
4008 * to finish f[data]sync. We set them to currently running transaction
4009 * as we cannot be sure that the inode or some of its metadata isn't
4010 * part of the transaction - the inode could have been reclaimed and
4011 * now it is reread from disk.
4014 transaction_t *transaction;
4017 read_lock(&journal->j_state_lock);
4018 if (journal->j_running_transaction)
4019 transaction = journal->j_running_transaction;
4021 transaction = journal->j_committing_transaction;
4023 tid = transaction->t_tid;
4025 tid = journal->j_commit_sequence;
4026 read_unlock(&journal->j_state_lock);
4027 ei->i_sync_tid = tid;
4028 ei->i_datasync_tid = tid;
4031 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4032 if (ei->i_extra_isize == 0) {
4033 /* The extra space is currently unused. Use it. */
4034 ei->i_extra_isize = sizeof(struct ext4_inode) -
4035 EXT4_GOOD_OLD_INODE_SIZE;
4037 ext4_iget_extra_inode(inode, raw_inode, ei);
4041 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4042 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4043 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4044 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4046 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4047 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4048 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4049 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4051 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4056 if (ei->i_file_acl &&
4057 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4058 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4062 } else if (!ext4_has_inline_data(inode)) {
4063 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4064 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4065 (S_ISLNK(inode->i_mode) &&
4066 !ext4_inode_is_fast_symlink(inode))))
4067 /* Validate extent which is part of inode */
4068 ret = ext4_ext_check_inode(inode);
4069 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4070 (S_ISLNK(inode->i_mode) &&
4071 !ext4_inode_is_fast_symlink(inode))) {
4072 /* Validate block references which are part of inode */
4073 ret = ext4_ind_check_inode(inode);
4079 if (S_ISREG(inode->i_mode)) {
4080 inode->i_op = &ext4_file_inode_operations;
4081 inode->i_fop = &ext4_file_operations;
4082 ext4_set_aops(inode);
4083 } else if (S_ISDIR(inode->i_mode)) {
4084 inode->i_op = &ext4_dir_inode_operations;
4085 inode->i_fop = &ext4_dir_operations;
4086 } else if (S_ISLNK(inode->i_mode)) {
4087 if (ext4_inode_is_fast_symlink(inode)) {
4088 inode->i_op = &ext4_fast_symlink_inode_operations;
4089 nd_terminate_link(ei->i_data, inode->i_size,
4090 sizeof(ei->i_data) - 1);
4092 inode->i_op = &ext4_symlink_inode_operations;
4093 ext4_set_aops(inode);
4095 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4096 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4097 inode->i_op = &ext4_special_inode_operations;
4098 if (raw_inode->i_block[0])
4099 init_special_inode(inode, inode->i_mode,
4100 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4102 init_special_inode(inode, inode->i_mode,
4103 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4104 } else if (ino == EXT4_BOOT_LOADER_INO) {
4105 make_bad_inode(inode);
4108 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4112 ext4_set_inode_flags(inode);
4113 unlock_new_inode(inode);
4119 return ERR_PTR(ret);
4122 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4124 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4125 return ERR_PTR(-EIO);
4126 return ext4_iget(sb, ino);
4129 static int ext4_inode_blocks_set(handle_t *handle,
4130 struct ext4_inode *raw_inode,
4131 struct ext4_inode_info *ei)
4133 struct inode *inode = &(ei->vfs_inode);
4134 u64 i_blocks = inode->i_blocks;
4135 struct super_block *sb = inode->i_sb;
4137 if (i_blocks <= ~0U) {
4139 * i_blocks can be represented in a 32 bit variable
4140 * as multiple of 512 bytes
4142 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4143 raw_inode->i_blocks_high = 0;
4144 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4147 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4150 if (i_blocks <= 0xffffffffffffULL) {
4152 * i_blocks can be represented in a 48 bit variable
4153 * as multiple of 512 bytes
4155 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4156 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4157 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4159 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4160 /* i_block is stored in file system block size */
4161 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4162 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4163 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4169 * Post the struct inode info into an on-disk inode location in the
4170 * buffer-cache. This gobbles the caller's reference to the
4171 * buffer_head in the inode location struct.
4173 * The caller must have write access to iloc->bh.
4175 static int ext4_do_update_inode(handle_t *handle,
4176 struct inode *inode,
4177 struct ext4_iloc *iloc)
4179 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4180 struct ext4_inode_info *ei = EXT4_I(inode);
4181 struct buffer_head *bh = iloc->bh;
4182 struct super_block *sb = inode->i_sb;
4183 int err = 0, rc, block;
4184 int need_datasync = 0, set_large_file = 0;
4188 spin_lock(&ei->i_raw_lock);
4190 /* For fields not tracked in the in-memory inode,
4191 * initialise them to zero for new inodes. */
4192 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4193 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4195 ext4_get_inode_flags(ei);
4196 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4197 i_uid = i_uid_read(inode);
4198 i_gid = i_gid_read(inode);
4199 if (!(test_opt(inode->i_sb, NO_UID32))) {
4200 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4201 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4203 * Fix up interoperability with old kernels. Otherwise, old inodes get
4204 * re-used with the upper 16 bits of the uid/gid intact
4207 raw_inode->i_uid_high =
4208 cpu_to_le16(high_16_bits(i_uid));
4209 raw_inode->i_gid_high =
4210 cpu_to_le16(high_16_bits(i_gid));
4212 raw_inode->i_uid_high = 0;
4213 raw_inode->i_gid_high = 0;
4216 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4217 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4218 raw_inode->i_uid_high = 0;
4219 raw_inode->i_gid_high = 0;
4221 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4223 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4224 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4225 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4226 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4228 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4230 spin_unlock(&ei->i_raw_lock);
4233 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4234 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4235 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4236 raw_inode->i_file_acl_high =
4237 cpu_to_le16(ei->i_file_acl >> 32);
4238 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4239 if (ei->i_disksize != ext4_isize(raw_inode)) {
4240 ext4_isize_set(raw_inode, ei->i_disksize);
4243 if (ei->i_disksize > 0x7fffffffULL) {
4244 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4245 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4246 EXT4_SB(sb)->s_es->s_rev_level ==
4247 cpu_to_le32(EXT4_GOOD_OLD_REV))
4250 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4251 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4252 if (old_valid_dev(inode->i_rdev)) {
4253 raw_inode->i_block[0] =
4254 cpu_to_le32(old_encode_dev(inode->i_rdev));
4255 raw_inode->i_block[1] = 0;
4257 raw_inode->i_block[0] = 0;
4258 raw_inode->i_block[1] =
4259 cpu_to_le32(new_encode_dev(inode->i_rdev));
4260 raw_inode->i_block[2] = 0;
4262 } else if (!ext4_has_inline_data(inode)) {
4263 for (block = 0; block < EXT4_N_BLOCKS; block++)
4264 raw_inode->i_block[block] = ei->i_data[block];
4267 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4268 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4269 if (ei->i_extra_isize) {
4270 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4271 raw_inode->i_version_hi =
4272 cpu_to_le32(inode->i_version >> 32);
4273 raw_inode->i_extra_isize =
4274 cpu_to_le16(ei->i_extra_isize);
4278 ext4_inode_csum_set(inode, raw_inode, ei);
4280 spin_unlock(&ei->i_raw_lock);
4282 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4283 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4286 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4287 if (set_large_file) {
4288 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4289 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4292 ext4_update_dynamic_rev(sb);
4293 EXT4_SET_RO_COMPAT_FEATURE(sb,
4294 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4295 ext4_handle_sync(handle);
4296 err = ext4_handle_dirty_super(handle, sb);
4298 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4301 ext4_std_error(inode->i_sb, err);
4306 * ext4_write_inode()
4308 * We are called from a few places:
4310 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4311 * Here, there will be no transaction running. We wait for any running
4312 * transaction to commit.
4314 * - Within flush work (sys_sync(), kupdate and such).
4315 * We wait on commit, if told to.
4317 * - Within iput_final() -> write_inode_now()
4318 * We wait on commit, if told to.
4320 * In all cases it is actually safe for us to return without doing anything,
4321 * because the inode has been copied into a raw inode buffer in
4322 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4325 * Note that we are absolutely dependent upon all inode dirtiers doing the
4326 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4327 * which we are interested.
4329 * It would be a bug for them to not do this. The code:
4331 * mark_inode_dirty(inode)
4333 * inode->i_size = expr;
4335 * is in error because write_inode() could occur while `stuff()' is running,
4336 * and the new i_size will be lost. Plus the inode will no longer be on the
4337 * superblock's dirty inode list.
4339 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4343 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4346 if (EXT4_SB(inode->i_sb)->s_journal) {
4347 if (ext4_journal_current_handle()) {
4348 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4354 * No need to force transaction in WB_SYNC_NONE mode. Also
4355 * ext4_sync_fs() will force the commit after everything is
4358 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4361 err = ext4_force_commit(inode->i_sb);
4363 struct ext4_iloc iloc;
4365 err = __ext4_get_inode_loc(inode, &iloc, 0);
4369 * sync(2) will flush the whole buffer cache. No need to do
4370 * it here separately for each inode.
4372 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4373 sync_dirty_buffer(iloc.bh);
4374 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4375 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4376 "IO error syncing inode");
4385 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4386 * buffers that are attached to a page stradding i_size and are undergoing
4387 * commit. In that case we have to wait for commit to finish and try again.
4389 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4393 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4394 tid_t commit_tid = 0;
4397 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4399 * All buffers in the last page remain valid? Then there's nothing to
4400 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4403 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4406 page = find_lock_page(inode->i_mapping,
4407 inode->i_size >> PAGE_CACHE_SHIFT);
4410 ret = __ext4_journalled_invalidatepage(page, offset,
4411 PAGE_CACHE_SIZE - offset);
4413 page_cache_release(page);
4417 read_lock(&journal->j_state_lock);
4418 if (journal->j_committing_transaction)
4419 commit_tid = journal->j_committing_transaction->t_tid;
4420 read_unlock(&journal->j_state_lock);
4422 jbd2_log_wait_commit(journal, commit_tid);
4429 * Called from notify_change.
4431 * We want to trap VFS attempts to truncate the file as soon as
4432 * possible. In particular, we want to make sure that when the VFS
4433 * shrinks i_size, we put the inode on the orphan list and modify
4434 * i_disksize immediately, so that during the subsequent flushing of
4435 * dirty pages and freeing of disk blocks, we can guarantee that any
4436 * commit will leave the blocks being flushed in an unused state on
4437 * disk. (On recovery, the inode will get truncated and the blocks will
4438 * be freed, so we have a strong guarantee that no future commit will
4439 * leave these blocks visible to the user.)
4441 * Another thing we have to assure is that if we are in ordered mode
4442 * and inode is still attached to the committing transaction, we must
4443 * we start writeout of all the dirty pages which are being truncated.
4444 * This way we are sure that all the data written in the previous
4445 * transaction are already on disk (truncate waits for pages under
4448 * Called with inode->i_mutex down.
4450 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4452 struct inode *inode = dentry->d_inode;
4455 const unsigned int ia_valid = attr->ia_valid;
4457 error = inode_change_ok(inode, attr);
4461 if (is_quota_modification(inode, attr))
4462 dquot_initialize(inode);
4463 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4464 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4467 /* (user+group)*(old+new) structure, inode write (sb,
4468 * inode block, ? - but truncate inode update has it) */
4469 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4470 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4471 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4472 if (IS_ERR(handle)) {
4473 error = PTR_ERR(handle);
4476 error = dquot_transfer(inode, attr);
4478 ext4_journal_stop(handle);
4481 /* Update corresponding info in inode so that everything is in
4482 * one transaction */
4483 if (attr->ia_valid & ATTR_UID)
4484 inode->i_uid = attr->ia_uid;
4485 if (attr->ia_valid & ATTR_GID)
4486 inode->i_gid = attr->ia_gid;
4487 error = ext4_mark_inode_dirty(handle, inode);
4488 ext4_journal_stop(handle);
4491 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4494 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4495 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4497 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4501 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4502 inode_inc_iversion(inode);
4504 if (S_ISREG(inode->i_mode) &&
4505 (attr->ia_size < inode->i_size)) {
4506 if (ext4_should_order_data(inode)) {
4507 error = ext4_begin_ordered_truncate(inode,
4512 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4513 if (IS_ERR(handle)) {
4514 error = PTR_ERR(handle);
4517 if (ext4_handle_valid(handle)) {
4518 error = ext4_orphan_add(handle, inode);
4521 down_write(&EXT4_I(inode)->i_data_sem);
4522 EXT4_I(inode)->i_disksize = attr->ia_size;
4523 rc = ext4_mark_inode_dirty(handle, inode);
4527 * We have to update i_size under i_data_sem together
4528 * with i_disksize to avoid races with writeback code
4529 * running ext4_wb_update_i_disksize().
4532 i_size_write(inode, attr->ia_size);
4533 up_write(&EXT4_I(inode)->i_data_sem);
4534 ext4_journal_stop(handle);
4536 ext4_orphan_del(NULL, inode);
4540 loff_t oldsize = inode->i_size;
4542 i_size_write(inode, attr->ia_size);
4543 pagecache_isize_extended(inode, oldsize, inode->i_size);
4547 * Blocks are going to be removed from the inode. Wait
4548 * for dio in flight. Temporarily disable
4549 * dioread_nolock to prevent livelock.
4552 if (!ext4_should_journal_data(inode)) {
4553 ext4_inode_block_unlocked_dio(inode);
4554 inode_dio_wait(inode);
4555 ext4_inode_resume_unlocked_dio(inode);
4557 ext4_wait_for_tail_page_commit(inode);
4560 * Truncate pagecache after we've waited for commit
4561 * in data=journal mode to make pages freeable.
4563 truncate_pagecache(inode, inode->i_size);
4566 * We want to call ext4_truncate() even if attr->ia_size ==
4567 * inode->i_size for cases like truncation of fallocated space
4569 if (attr->ia_valid & ATTR_SIZE)
4570 ext4_truncate(inode);
4573 setattr_copy(inode, attr);
4574 mark_inode_dirty(inode);
4578 * If the call to ext4_truncate failed to get a transaction handle at
4579 * all, we need to clean up the in-core orphan list manually.
4581 if (orphan && inode->i_nlink)
4582 ext4_orphan_del(NULL, inode);
4584 if (!rc && (ia_valid & ATTR_MODE))
4585 rc = posix_acl_chmod(inode, inode->i_mode);
4588 ext4_std_error(inode->i_sb, error);
4594 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4597 struct inode *inode;
4598 unsigned long long delalloc_blocks;
4600 inode = dentry->d_inode;
4601 generic_fillattr(inode, stat);
4604 * If there is inline data in the inode, the inode will normally not
4605 * have data blocks allocated (it may have an external xattr block).
4606 * Report at least one sector for such files, so tools like tar, rsync,
4607 * others doen't incorrectly think the file is completely sparse.
4609 if (unlikely(ext4_has_inline_data(inode)))
4610 stat->blocks += (stat->size + 511) >> 9;
4613 * We can't update i_blocks if the block allocation is delayed
4614 * otherwise in the case of system crash before the real block
4615 * allocation is done, we will have i_blocks inconsistent with
4616 * on-disk file blocks.
4617 * We always keep i_blocks updated together with real
4618 * allocation. But to not confuse with user, stat
4619 * will return the blocks that include the delayed allocation
4620 * blocks for this file.
4622 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4623 EXT4_I(inode)->i_reserved_data_blocks);
4624 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
4628 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4631 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4632 return ext4_ind_trans_blocks(inode, lblocks);
4633 return ext4_ext_index_trans_blocks(inode, pextents);
4637 * Account for index blocks, block groups bitmaps and block group
4638 * descriptor blocks if modify datablocks and index blocks
4639 * worse case, the indexs blocks spread over different block groups
4641 * If datablocks are discontiguous, they are possible to spread over
4642 * different block groups too. If they are contiguous, with flexbg,
4643 * they could still across block group boundary.
4645 * Also account for superblock, inode, quota and xattr blocks
4647 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4650 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4656 * How many index blocks need to touch to map @lblocks logical blocks
4657 * to @pextents physical extents?
4659 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
4664 * Now let's see how many group bitmaps and group descriptors need
4667 groups = idxblocks + pextents;
4669 if (groups > ngroups)
4671 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4672 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4674 /* bitmaps and block group descriptor blocks */
4675 ret += groups + gdpblocks;
4677 /* Blocks for super block, inode, quota and xattr blocks */
4678 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4684 * Calculate the total number of credits to reserve to fit
4685 * the modification of a single pages into a single transaction,
4686 * which may include multiple chunks of block allocations.
4688 * This could be called via ext4_write_begin()
4690 * We need to consider the worse case, when
4691 * one new block per extent.
4693 int ext4_writepage_trans_blocks(struct inode *inode)
4695 int bpp = ext4_journal_blocks_per_page(inode);
4698 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
4700 /* Account for data blocks for journalled mode */
4701 if (ext4_should_journal_data(inode))
4707 * Calculate the journal credits for a chunk of data modification.
4709 * This is called from DIO, fallocate or whoever calling
4710 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4712 * journal buffers for data blocks are not included here, as DIO
4713 * and fallocate do no need to journal data buffers.
4715 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4717 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4721 * The caller must have previously called ext4_reserve_inode_write().
4722 * Give this, we know that the caller already has write access to iloc->bh.
4724 int ext4_mark_iloc_dirty(handle_t *handle,
4725 struct inode *inode, struct ext4_iloc *iloc)
4729 if (IS_I_VERSION(inode))
4730 inode_inc_iversion(inode);
4732 /* the do_update_inode consumes one bh->b_count */
4735 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4736 err = ext4_do_update_inode(handle, inode, iloc);
4742 * On success, We end up with an outstanding reference count against
4743 * iloc->bh. This _must_ be cleaned up later.
4747 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4748 struct ext4_iloc *iloc)
4752 err = ext4_get_inode_loc(inode, iloc);
4754 BUFFER_TRACE(iloc->bh, "get_write_access");
4755 err = ext4_journal_get_write_access(handle, iloc->bh);
4761 ext4_std_error(inode->i_sb, err);
4766 * Expand an inode by new_extra_isize bytes.
4767 * Returns 0 on success or negative error number on failure.
4769 static int ext4_expand_extra_isize(struct inode *inode,
4770 unsigned int new_extra_isize,
4771 struct ext4_iloc iloc,
4774 struct ext4_inode *raw_inode;
4775 struct ext4_xattr_ibody_header *header;
4777 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4780 raw_inode = ext4_raw_inode(&iloc);
4782 header = IHDR(inode, raw_inode);
4784 /* No extended attributes present */
4785 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4786 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4787 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4789 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4793 /* try to expand with EAs present */
4794 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4799 * What we do here is to mark the in-core inode as clean with respect to inode
4800 * dirtiness (it may still be data-dirty).
4801 * This means that the in-core inode may be reaped by prune_icache
4802 * without having to perform any I/O. This is a very good thing,
4803 * because *any* task may call prune_icache - even ones which
4804 * have a transaction open against a different journal.
4806 * Is this cheating? Not really. Sure, we haven't written the
4807 * inode out, but prune_icache isn't a user-visible syncing function.
4808 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4809 * we start and wait on commits.
4811 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4813 struct ext4_iloc iloc;
4814 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4815 static unsigned int mnt_count;
4819 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4820 err = ext4_reserve_inode_write(handle, inode, &iloc);
4821 if (ext4_handle_valid(handle) &&
4822 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4823 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4825 * We need extra buffer credits since we may write into EA block
4826 * with this same handle. If journal_extend fails, then it will
4827 * only result in a minor loss of functionality for that inode.
4828 * If this is felt to be critical, then e2fsck should be run to
4829 * force a large enough s_min_extra_isize.
4831 if ((jbd2_journal_extend(handle,
4832 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4833 ret = ext4_expand_extra_isize(inode,
4834 sbi->s_want_extra_isize,
4837 ext4_set_inode_state(inode,
4838 EXT4_STATE_NO_EXPAND);
4840 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4841 ext4_warning(inode->i_sb,
4842 "Unable to expand inode %lu. Delete"
4843 " some EAs or run e2fsck.",
4846 le16_to_cpu(sbi->s_es->s_mnt_count);
4852 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4857 * ext4_dirty_inode() is called from __mark_inode_dirty()
4859 * We're really interested in the case where a file is being extended.
4860 * i_size has been changed by generic_commit_write() and we thus need
4861 * to include the updated inode in the current transaction.
4863 * Also, dquot_alloc_block() will always dirty the inode when blocks
4864 * are allocated to the file.
4866 * If the inode is marked synchronous, we don't honour that here - doing
4867 * so would cause a commit on atime updates, which we don't bother doing.
4868 * We handle synchronous inodes at the highest possible level.
4870 void ext4_dirty_inode(struct inode *inode, int flags)
4874 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
4878 ext4_mark_inode_dirty(handle, inode);
4880 ext4_journal_stop(handle);
4887 * Bind an inode's backing buffer_head into this transaction, to prevent
4888 * it from being flushed to disk early. Unlike
4889 * ext4_reserve_inode_write, this leaves behind no bh reference and
4890 * returns no iloc structure, so the caller needs to repeat the iloc
4891 * lookup to mark the inode dirty later.
4893 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4895 struct ext4_iloc iloc;
4899 err = ext4_get_inode_loc(inode, &iloc);
4901 BUFFER_TRACE(iloc.bh, "get_write_access");
4902 err = jbd2_journal_get_write_access(handle, iloc.bh);
4904 err = ext4_handle_dirty_metadata(handle,
4910 ext4_std_error(inode->i_sb, err);
4915 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4922 * We have to be very careful here: changing a data block's
4923 * journaling status dynamically is dangerous. If we write a
4924 * data block to the journal, change the status and then delete
4925 * that block, we risk forgetting to revoke the old log record
4926 * from the journal and so a subsequent replay can corrupt data.
4927 * So, first we make sure that the journal is empty and that
4928 * nobody is changing anything.
4931 journal = EXT4_JOURNAL(inode);
4934 if (is_journal_aborted(journal))
4936 /* We have to allocate physical blocks for delalloc blocks
4937 * before flushing journal. otherwise delalloc blocks can not
4938 * be allocated any more. even more truncate on delalloc blocks
4939 * could trigger BUG by flushing delalloc blocks in journal.
4940 * There is no delalloc block in non-journal data mode.
4942 if (val && test_opt(inode->i_sb, DELALLOC)) {
4943 err = ext4_alloc_da_blocks(inode);
4948 /* Wait for all existing dio workers */
4949 ext4_inode_block_unlocked_dio(inode);
4950 inode_dio_wait(inode);
4952 jbd2_journal_lock_updates(journal);
4955 * OK, there are no updates running now, and all cached data is
4956 * synced to disk. We are now in a completely consistent state
4957 * which doesn't have anything in the journal, and we know that
4958 * no filesystem updates are running, so it is safe to modify
4959 * the inode's in-core data-journaling state flag now.
4963 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4965 jbd2_journal_flush(journal);
4966 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4968 ext4_set_aops(inode);
4970 jbd2_journal_unlock_updates(journal);
4971 ext4_inode_resume_unlocked_dio(inode);
4973 /* Finally we can mark the inode as dirty. */
4975 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
4977 return PTR_ERR(handle);
4979 err = ext4_mark_inode_dirty(handle, inode);
4980 ext4_handle_sync(handle);
4981 ext4_journal_stop(handle);
4982 ext4_std_error(inode->i_sb, err);
4987 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4989 return !buffer_mapped(bh);
4992 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4994 struct page *page = vmf->page;
4998 struct file *file = vma->vm_file;
4999 struct inode *inode = file_inode(file);
5000 struct address_space *mapping = inode->i_mapping;
5002 get_block_t *get_block;
5005 sb_start_pagefault(inode->i_sb);
5006 file_update_time(vma->vm_file);
5007 /* Delalloc case is easy... */
5008 if (test_opt(inode->i_sb, DELALLOC) &&
5009 !ext4_should_journal_data(inode) &&
5010 !ext4_nonda_switch(inode->i_sb)) {
5012 ret = __block_page_mkwrite(vma, vmf,
5013 ext4_da_get_block_prep);
5014 } while (ret == -ENOSPC &&
5015 ext4_should_retry_alloc(inode->i_sb, &retries));
5020 size = i_size_read(inode);
5021 /* Page got truncated from under us? */
5022 if (page->mapping != mapping || page_offset(page) > size) {
5024 ret = VM_FAULT_NOPAGE;
5028 if (page->index == size >> PAGE_CACHE_SHIFT)
5029 len = size & ~PAGE_CACHE_MASK;
5031 len = PAGE_CACHE_SIZE;
5033 * Return if we have all the buffers mapped. This avoids the need to do
5034 * journal_start/journal_stop which can block and take a long time
5036 if (page_has_buffers(page)) {
5037 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5039 ext4_bh_unmapped)) {
5040 /* Wait so that we don't change page under IO */
5041 wait_for_stable_page(page);
5042 ret = VM_FAULT_LOCKED;
5047 /* OK, we need to fill the hole... */
5048 if (ext4_should_dioread_nolock(inode))
5049 get_block = ext4_get_block_write;
5051 get_block = ext4_get_block;
5053 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5054 ext4_writepage_trans_blocks(inode));
5055 if (IS_ERR(handle)) {
5056 ret = VM_FAULT_SIGBUS;
5059 ret = __block_page_mkwrite(vma, vmf, get_block);
5060 if (!ret && ext4_should_journal_data(inode)) {
5061 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5062 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5064 ret = VM_FAULT_SIGBUS;
5065 ext4_journal_stop(handle);
5068 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5070 ext4_journal_stop(handle);
5071 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5074 ret = block_page_mkwrite_return(ret);
5076 sb_end_pagefault(inode->i_sb);