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_metadata_csum(inode->i_sb))
89 provided = le16_to_cpu(raw->i_checksum_lo);
90 calculated = ext4_inode_csum(inode, raw, ei);
91 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
92 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
93 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
97 return provided == calculated;
100 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
101 struct ext4_inode_info *ei)
105 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
106 cpu_to_le32(EXT4_OS_LINUX) ||
107 !ext4_has_metadata_csum(inode->i_sb))
110 csum = ext4_inode_csum(inode, raw, ei);
111 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
112 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
113 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
114 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
117 static inline int ext4_begin_ordered_truncate(struct inode *inode,
120 trace_ext4_begin_ordered_truncate(inode, new_size);
122 * If jinode is zero, then we never opened the file for
123 * writing, so there's no need to call
124 * jbd2_journal_begin_ordered_truncate() since there's no
125 * outstanding writes we need to flush.
127 if (!EXT4_I(inode)->jinode)
129 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
130 EXT4_I(inode)->jinode,
134 static void ext4_invalidatepage(struct page *page, unsigned int offset,
135 unsigned int length);
136 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
137 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
138 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
142 * Test whether an inode is a fast symlink.
144 static int ext4_inode_is_fast_symlink(struct inode *inode)
146 int ea_blocks = EXT4_I(inode)->i_file_acl ?
147 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
149 if (ext4_has_inline_data(inode))
152 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
156 * Restart the transaction associated with *handle. This does a commit,
157 * so before we call here everything must be consistently dirtied against
160 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
166 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
167 * moment, get_block can be called only for blocks inside i_size since
168 * page cache has been already dropped and writes are blocked by
169 * i_mutex. So we can safely drop the i_data_sem here.
171 BUG_ON(EXT4_JOURNAL(inode) == NULL);
172 jbd_debug(2, "restarting handle %p\n", handle);
173 up_write(&EXT4_I(inode)->i_data_sem);
174 ret = ext4_journal_restart(handle, nblocks);
175 down_write(&EXT4_I(inode)->i_data_sem);
176 ext4_discard_preallocations(inode);
182 * Called at the last iput() if i_nlink is zero.
184 void ext4_evict_inode(struct inode *inode)
189 trace_ext4_evict_inode(inode);
191 if (inode->i_nlink) {
193 * When journalling data dirty buffers are tracked only in the
194 * journal. So although mm thinks everything is clean and
195 * ready for reaping the inode might still have some pages to
196 * write in the running transaction or waiting to be
197 * checkpointed. Thus calling jbd2_journal_invalidatepage()
198 * (via truncate_inode_pages()) to discard these buffers can
199 * cause data loss. Also even if we did not discard these
200 * buffers, we would have no way to find them after the inode
201 * is reaped and thus user could see stale data if he tries to
202 * read them before the transaction is checkpointed. So be
203 * careful and force everything to disk here... We use
204 * ei->i_datasync_tid to store the newest transaction
205 * containing inode's data.
207 * Note that directories do not have this problem because they
208 * don't use page cache.
210 if (ext4_should_journal_data(inode) &&
211 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
212 inode->i_ino != EXT4_JOURNAL_INO) {
213 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
214 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
216 jbd2_complete_transaction(journal, commit_tid);
217 filemap_write_and_wait(&inode->i_data);
219 truncate_inode_pages_final(&inode->i_data);
221 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
225 if (is_bad_inode(inode))
227 dquot_initialize(inode);
229 if (ext4_should_order_data(inode))
230 ext4_begin_ordered_truncate(inode, 0);
231 truncate_inode_pages_final(&inode->i_data);
233 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
236 * Protect us against freezing - iput() caller didn't have to have any
237 * protection against it
239 sb_start_intwrite(inode->i_sb);
240 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
241 ext4_blocks_for_truncate(inode)+3);
242 if (IS_ERR(handle)) {
243 ext4_std_error(inode->i_sb, PTR_ERR(handle));
245 * If we're going to skip the normal cleanup, we still need to
246 * make sure that the in-core orphan linked list is properly
249 ext4_orphan_del(NULL, inode);
250 sb_end_intwrite(inode->i_sb);
255 ext4_handle_sync(handle);
257 err = ext4_mark_inode_dirty(handle, inode);
259 ext4_warning(inode->i_sb,
260 "couldn't mark inode dirty (err %d)", err);
264 ext4_truncate(inode);
267 * ext4_ext_truncate() doesn't reserve any slop when it
268 * restarts journal transactions; therefore there may not be
269 * enough credits left in the handle to remove the inode from
270 * the orphan list and set the dtime field.
272 if (!ext4_handle_has_enough_credits(handle, 3)) {
273 err = ext4_journal_extend(handle, 3);
275 err = ext4_journal_restart(handle, 3);
277 ext4_warning(inode->i_sb,
278 "couldn't extend journal (err %d)", err);
280 ext4_journal_stop(handle);
281 ext4_orphan_del(NULL, inode);
282 sb_end_intwrite(inode->i_sb);
288 * Kill off the orphan record which ext4_truncate created.
289 * AKPM: I think this can be inside the above `if'.
290 * Note that ext4_orphan_del() has to be able to cope with the
291 * deletion of a non-existent orphan - this is because we don't
292 * know if ext4_truncate() actually created an orphan record.
293 * (Well, we could do this if we need to, but heck - it works)
295 ext4_orphan_del(handle, inode);
296 EXT4_I(inode)->i_dtime = get_seconds();
299 * One subtle ordering requirement: if anything has gone wrong
300 * (transaction abort, IO errors, whatever), then we can still
301 * do these next steps (the fs will already have been marked as
302 * having errors), but we can't free the inode if the mark_dirty
305 if (ext4_mark_inode_dirty(handle, inode))
306 /* If that failed, just do the required in-core inode clear. */
307 ext4_clear_inode(inode);
309 ext4_free_inode(handle, inode);
310 ext4_journal_stop(handle);
311 sb_end_intwrite(inode->i_sb);
314 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
318 qsize_t *ext4_get_reserved_space(struct inode *inode)
320 return &EXT4_I(inode)->i_reserved_quota;
325 * Called with i_data_sem down, which is important since we can call
326 * ext4_discard_preallocations() from here.
328 void ext4_da_update_reserve_space(struct inode *inode,
329 int used, int quota_claim)
331 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
332 struct ext4_inode_info *ei = EXT4_I(inode);
334 spin_lock(&ei->i_block_reservation_lock);
335 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
336 if (unlikely(used > ei->i_reserved_data_blocks)) {
337 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
338 "with only %d reserved data blocks",
339 __func__, inode->i_ino, used,
340 ei->i_reserved_data_blocks);
342 used = ei->i_reserved_data_blocks;
345 /* Update per-inode reservations */
346 ei->i_reserved_data_blocks -= used;
347 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
349 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
351 /* Update quota subsystem for data blocks */
353 dquot_claim_block(inode, EXT4_C2B(sbi, used));
356 * We did fallocate with an offset that is already delayed
357 * allocated. So on delayed allocated writeback we should
358 * not re-claim the quota for fallocated blocks.
360 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
364 * If we have done all the pending block allocations and if
365 * there aren't any writers on the inode, we can discard the
366 * inode's preallocations.
368 if ((ei->i_reserved_data_blocks == 0) &&
369 (atomic_read(&inode->i_writecount) == 0))
370 ext4_discard_preallocations(inode);
373 static int __check_block_validity(struct inode *inode, const char *func,
375 struct ext4_map_blocks *map)
377 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
379 ext4_error_inode(inode, func, line, map->m_pblk,
380 "lblock %lu mapped to illegal pblock "
381 "(length %d)", (unsigned long) map->m_lblk,
388 #define check_block_validity(inode, map) \
389 __check_block_validity((inode), __func__, __LINE__, (map))
391 #ifdef ES_AGGRESSIVE_TEST
392 static void ext4_map_blocks_es_recheck(handle_t *handle,
394 struct ext4_map_blocks *es_map,
395 struct ext4_map_blocks *map,
402 * There is a race window that the result is not the same.
403 * e.g. xfstests #223 when dioread_nolock enables. The reason
404 * is that we lookup a block mapping in extent status tree with
405 * out taking i_data_sem. So at the time the unwritten extent
406 * could be converted.
408 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
409 down_read(&EXT4_I(inode)->i_data_sem);
410 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
411 retval = ext4_ext_map_blocks(handle, inode, map, flags &
412 EXT4_GET_BLOCKS_KEEP_SIZE);
414 retval = ext4_ind_map_blocks(handle, inode, map, flags &
415 EXT4_GET_BLOCKS_KEEP_SIZE);
417 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
418 up_read((&EXT4_I(inode)->i_data_sem));
421 * We don't check m_len because extent will be collpased in status
422 * tree. So the m_len might not equal.
424 if (es_map->m_lblk != map->m_lblk ||
425 es_map->m_flags != map->m_flags ||
426 es_map->m_pblk != map->m_pblk) {
427 printk("ES cache assertion failed for inode: %lu "
428 "es_cached ex [%d/%d/%llu/%x] != "
429 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
430 inode->i_ino, es_map->m_lblk, es_map->m_len,
431 es_map->m_pblk, es_map->m_flags, map->m_lblk,
432 map->m_len, map->m_pblk, map->m_flags,
436 #endif /* ES_AGGRESSIVE_TEST */
439 * The ext4_map_blocks() function tries to look up the requested blocks,
440 * and returns if the blocks are already mapped.
442 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
443 * and store the allocated blocks in the result buffer head and mark it
446 * If file type is extents based, it will call ext4_ext_map_blocks(),
447 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
450 * On success, it returns the number of blocks being mapped or allocated.
451 * if create==0 and the blocks are pre-allocated and unwritten block,
452 * the result buffer head is unmapped. If the create ==1, it will make sure
453 * the buffer head is mapped.
455 * It returns 0 if plain look up failed (blocks have not been allocated), in
456 * that case, buffer head is unmapped
458 * It returns the error in case of allocation failure.
460 int ext4_map_blocks(handle_t *handle, struct inode *inode,
461 struct ext4_map_blocks *map, int flags)
463 struct extent_status es;
466 #ifdef ES_AGGRESSIVE_TEST
467 struct ext4_map_blocks orig_map;
469 memcpy(&orig_map, map, sizeof(*map));
473 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
474 "logical block %lu\n", inode->i_ino, flags, map->m_len,
475 (unsigned long) map->m_lblk);
478 * ext4_map_blocks returns an int, and m_len is an unsigned int
480 if (unlikely(map->m_len > INT_MAX))
481 map->m_len = INT_MAX;
483 /* We can handle the block number less than EXT_MAX_BLOCKS */
484 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
487 /* Lookup extent status tree firstly */
488 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
489 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
490 map->m_pblk = ext4_es_pblock(&es) +
491 map->m_lblk - es.es_lblk;
492 map->m_flags |= ext4_es_is_written(&es) ?
493 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
494 retval = es.es_len - (map->m_lblk - es.es_lblk);
495 if (retval > map->m_len)
498 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
503 #ifdef ES_AGGRESSIVE_TEST
504 ext4_map_blocks_es_recheck(handle, inode, map,
511 * Try to see if we can get the block without requesting a new
514 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
515 down_read(&EXT4_I(inode)->i_data_sem);
516 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
517 retval = ext4_ext_map_blocks(handle, inode, map, flags &
518 EXT4_GET_BLOCKS_KEEP_SIZE);
520 retval = ext4_ind_map_blocks(handle, inode, map, flags &
521 EXT4_GET_BLOCKS_KEEP_SIZE);
526 if (unlikely(retval != map->m_len)) {
527 ext4_warning(inode->i_sb,
528 "ES len assertion failed for inode "
529 "%lu: retval %d != map->m_len %d",
530 inode->i_ino, retval, map->m_len);
534 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
535 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
536 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
537 ext4_find_delalloc_range(inode, map->m_lblk,
538 map->m_lblk + map->m_len - 1))
539 status |= EXTENT_STATUS_DELAYED;
540 ret = ext4_es_insert_extent(inode, map->m_lblk,
541 map->m_len, map->m_pblk, status);
545 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
546 up_read((&EXT4_I(inode)->i_data_sem));
549 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
550 ret = check_block_validity(inode, map);
555 /* If it is only a block(s) look up */
556 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
560 * Returns if the blocks have already allocated
562 * Note that if blocks have been preallocated
563 * ext4_ext_get_block() returns the create = 0
564 * with buffer head unmapped.
566 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
568 * If we need to convert extent to unwritten
569 * we continue and do the actual work in
570 * ext4_ext_map_blocks()
572 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
576 * Here we clear m_flags because after allocating an new extent,
577 * it will be set again.
579 map->m_flags &= ~EXT4_MAP_FLAGS;
582 * New blocks allocate and/or writing to unwritten extent
583 * will possibly result in updating i_data, so we take
584 * the write lock of i_data_sem, and call get_block()
585 * with create == 1 flag.
587 down_write(&EXT4_I(inode)->i_data_sem);
590 * We need to check for EXT4 here because migrate
591 * could have changed the inode type in between
593 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
594 retval = ext4_ext_map_blocks(handle, inode, map, flags);
596 retval = ext4_ind_map_blocks(handle, inode, map, flags);
598 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
600 * We allocated new blocks which will result in
601 * i_data's format changing. Force the migrate
602 * to fail by clearing migrate flags
604 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
608 * Update reserved blocks/metadata blocks after successful
609 * block allocation which had been deferred till now. We don't
610 * support fallocate for non extent files. So we can update
611 * reserve space here.
614 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
615 ext4_da_update_reserve_space(inode, retval, 1);
621 if (unlikely(retval != map->m_len)) {
622 ext4_warning(inode->i_sb,
623 "ES len assertion failed for inode "
624 "%lu: retval %d != map->m_len %d",
625 inode->i_ino, retval, map->m_len);
630 * If the extent has been zeroed out, we don't need to update
631 * extent status tree.
633 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
634 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
635 if (ext4_es_is_written(&es))
638 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
639 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
640 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
641 ext4_find_delalloc_range(inode, map->m_lblk,
642 map->m_lblk + map->m_len - 1))
643 status |= EXTENT_STATUS_DELAYED;
644 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
645 map->m_pblk, status);
651 up_write((&EXT4_I(inode)->i_data_sem));
652 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
653 ret = check_block_validity(inode, map);
660 static void ext4_end_io_unwritten(struct buffer_head *bh, int uptodate)
662 struct inode *inode = bh->b_assoc_map->host;
663 /* XXX: breaks on 32-bit > 16GB. Is that even supported? */
664 loff_t offset = (loff_t)(uintptr_t)bh->b_private << inode->i_blkbits;
668 WARN_ON(!buffer_unwritten(bh));
669 err = ext4_convert_unwritten_extents(NULL, inode, offset, bh->b_size);
672 /* Maximum number of blocks we map for direct IO at once. */
673 #define DIO_MAX_BLOCKS 4096
675 static int _ext4_get_block(struct inode *inode, sector_t iblock,
676 struct buffer_head *bh, int flags)
678 handle_t *handle = ext4_journal_current_handle();
679 struct ext4_map_blocks map;
680 int ret = 0, started = 0;
683 if (ext4_has_inline_data(inode))
687 map.m_len = bh->b_size >> inode->i_blkbits;
689 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
690 /* Direct IO write... */
691 if (map.m_len > DIO_MAX_BLOCKS)
692 map.m_len = DIO_MAX_BLOCKS;
693 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
694 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
696 if (IS_ERR(handle)) {
697 ret = PTR_ERR(handle);
703 ret = ext4_map_blocks(handle, inode, &map, flags);
705 ext4_io_end_t *io_end = ext4_inode_aio(inode);
707 map_bh(bh, inode->i_sb, map.m_pblk);
708 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
709 if (IS_DAX(inode) && buffer_unwritten(bh) && !io_end) {
710 bh->b_assoc_map = inode->i_mapping;
711 bh->b_private = (void *)(unsigned long)iblock;
712 bh->b_end_io = ext4_end_io_unwritten;
714 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
715 set_buffer_defer_completion(bh);
716 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
720 ext4_journal_stop(handle);
724 int ext4_get_block(struct inode *inode, sector_t iblock,
725 struct buffer_head *bh, int create)
727 return _ext4_get_block(inode, iblock, bh,
728 create ? EXT4_GET_BLOCKS_CREATE : 0);
732 * `handle' can be NULL if create is zero
734 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
735 ext4_lblk_t block, int create)
737 struct ext4_map_blocks map;
738 struct buffer_head *bh;
741 J_ASSERT(handle != NULL || create == 0);
745 err = ext4_map_blocks(handle, inode, &map,
746 create ? EXT4_GET_BLOCKS_CREATE : 0);
749 return create ? ERR_PTR(-ENOSPC) : NULL;
753 bh = sb_getblk(inode->i_sb, map.m_pblk);
755 return ERR_PTR(-ENOMEM);
756 if (map.m_flags & EXT4_MAP_NEW) {
757 J_ASSERT(create != 0);
758 J_ASSERT(handle != NULL);
761 * Now that we do not always journal data, we should
762 * keep in mind whether this should always journal the
763 * new buffer as metadata. For now, regular file
764 * writes use ext4_get_block instead, so it's not a
768 BUFFER_TRACE(bh, "call get_create_access");
769 err = ext4_journal_get_create_access(handle, bh);
774 if (!buffer_uptodate(bh)) {
775 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
776 set_buffer_uptodate(bh);
779 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
780 err = ext4_handle_dirty_metadata(handle, inode, bh);
784 BUFFER_TRACE(bh, "not a new buffer");
791 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
792 ext4_lblk_t block, int create)
794 struct buffer_head *bh;
796 bh = ext4_getblk(handle, inode, block, create);
799 if (!bh || buffer_uptodate(bh))
801 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
803 if (buffer_uptodate(bh))
806 return ERR_PTR(-EIO);
809 int ext4_walk_page_buffers(handle_t *handle,
810 struct buffer_head *head,
814 int (*fn)(handle_t *handle,
815 struct buffer_head *bh))
817 struct buffer_head *bh;
818 unsigned block_start, block_end;
819 unsigned blocksize = head->b_size;
821 struct buffer_head *next;
823 for (bh = head, block_start = 0;
824 ret == 0 && (bh != head || !block_start);
825 block_start = block_end, bh = next) {
826 next = bh->b_this_page;
827 block_end = block_start + blocksize;
828 if (block_end <= from || block_start >= to) {
829 if (partial && !buffer_uptodate(bh))
833 err = (*fn)(handle, bh);
841 * To preserve ordering, it is essential that the hole instantiation and
842 * the data write be encapsulated in a single transaction. We cannot
843 * close off a transaction and start a new one between the ext4_get_block()
844 * and the commit_write(). So doing the jbd2_journal_start at the start of
845 * prepare_write() is the right place.
847 * Also, this function can nest inside ext4_writepage(). In that case, we
848 * *know* that ext4_writepage() has generated enough buffer credits to do the
849 * whole page. So we won't block on the journal in that case, which is good,
850 * because the caller may be PF_MEMALLOC.
852 * By accident, ext4 can be reentered when a transaction is open via
853 * quota file writes. If we were to commit the transaction while thus
854 * reentered, there can be a deadlock - we would be holding a quota
855 * lock, and the commit would never complete if another thread had a
856 * transaction open and was blocking on the quota lock - a ranking
859 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
860 * will _not_ run commit under these circumstances because handle->h_ref
861 * is elevated. We'll still have enough credits for the tiny quotafile
864 int do_journal_get_write_access(handle_t *handle,
865 struct buffer_head *bh)
867 int dirty = buffer_dirty(bh);
870 if (!buffer_mapped(bh) || buffer_freed(bh))
873 * __block_write_begin() could have dirtied some buffers. Clean
874 * the dirty bit as jbd2_journal_get_write_access() could complain
875 * otherwise about fs integrity issues. Setting of the dirty bit
876 * by __block_write_begin() isn't a real problem here as we clear
877 * the bit before releasing a page lock and thus writeback cannot
878 * ever write the buffer.
881 clear_buffer_dirty(bh);
882 BUFFER_TRACE(bh, "get write access");
883 ret = ext4_journal_get_write_access(handle, bh);
885 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
889 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
890 struct buffer_head *bh_result, int create);
891 static int ext4_write_begin(struct file *file, struct address_space *mapping,
892 loff_t pos, unsigned len, unsigned flags,
893 struct page **pagep, void **fsdata)
895 struct inode *inode = mapping->host;
896 int ret, needed_blocks;
903 trace_ext4_write_begin(inode, pos, len, flags);
905 * Reserve one block more for addition to orphan list in case
906 * we allocate blocks but write fails for some reason
908 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
909 index = pos >> PAGE_CACHE_SHIFT;
910 from = pos & (PAGE_CACHE_SIZE - 1);
913 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
914 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
923 * grab_cache_page_write_begin() can take a long time if the
924 * system is thrashing due to memory pressure, or if the page
925 * is being written back. So grab it first before we start
926 * the transaction handle. This also allows us to allocate
927 * the page (if needed) without using GFP_NOFS.
930 page = grab_cache_page_write_begin(mapping, index, flags);
936 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
937 if (IS_ERR(handle)) {
938 page_cache_release(page);
939 return PTR_ERR(handle);
943 if (page->mapping != mapping) {
944 /* The page got truncated from under us */
946 page_cache_release(page);
947 ext4_journal_stop(handle);
950 /* In case writeback began while the page was unlocked */
951 wait_for_stable_page(page);
953 if (ext4_should_dioread_nolock(inode))
954 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
956 ret = __block_write_begin(page, pos, len, ext4_get_block);
958 if (!ret && ext4_should_journal_data(inode)) {
959 ret = ext4_walk_page_buffers(handle, page_buffers(page),
961 do_journal_get_write_access);
967 * __block_write_begin may have instantiated a few blocks
968 * outside i_size. Trim these off again. Don't need
969 * i_size_read because we hold i_mutex.
971 * Add inode to orphan list in case we crash before
974 if (pos + len > inode->i_size && ext4_can_truncate(inode))
975 ext4_orphan_add(handle, inode);
977 ext4_journal_stop(handle);
978 if (pos + len > inode->i_size) {
979 ext4_truncate_failed_write(inode);
981 * If truncate failed early the inode might
982 * still be on the orphan list; we need to
983 * make sure the inode is removed from the
984 * orphan list in that case.
987 ext4_orphan_del(NULL, inode);
990 if (ret == -ENOSPC &&
991 ext4_should_retry_alloc(inode->i_sb, &retries))
993 page_cache_release(page);
1000 /* For write_end() in data=journal mode */
1001 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1004 if (!buffer_mapped(bh) || buffer_freed(bh))
1006 set_buffer_uptodate(bh);
1007 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1008 clear_buffer_meta(bh);
1009 clear_buffer_prio(bh);
1014 * We need to pick up the new inode size which generic_commit_write gave us
1015 * `file' can be NULL - eg, when called from page_symlink().
1017 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1018 * buffers are managed internally.
1020 static int ext4_write_end(struct file *file,
1021 struct address_space *mapping,
1022 loff_t pos, unsigned len, unsigned copied,
1023 struct page *page, void *fsdata)
1025 handle_t *handle = ext4_journal_current_handle();
1026 struct inode *inode = mapping->host;
1028 int i_size_changed = 0;
1030 trace_ext4_write_end(inode, pos, len, copied);
1031 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1032 ret = ext4_jbd2_file_inode(handle, inode);
1035 page_cache_release(page);
1040 if (ext4_has_inline_data(inode)) {
1041 ret = ext4_write_inline_data_end(inode, pos, len,
1047 copied = block_write_end(file, mapping, pos,
1048 len, copied, page, fsdata);
1050 * it's important to update i_size while still holding page lock:
1051 * page writeout could otherwise come in and zero beyond i_size.
1053 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1055 page_cache_release(page);
1058 * Don't mark the inode dirty under page lock. First, it unnecessarily
1059 * makes the holding time of page lock longer. Second, it forces lock
1060 * ordering of page lock and transaction start for journaling
1064 ext4_mark_inode_dirty(handle, inode);
1066 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1067 /* if we have allocated more blocks and copied
1068 * less. We will have blocks allocated outside
1069 * inode->i_size. So truncate them
1071 ext4_orphan_add(handle, inode);
1073 ret2 = ext4_journal_stop(handle);
1077 if (pos + len > inode->i_size) {
1078 ext4_truncate_failed_write(inode);
1080 * If truncate failed early the inode might still be
1081 * on the orphan list; we need to make sure the inode
1082 * is removed from the orphan list in that case.
1085 ext4_orphan_del(NULL, inode);
1088 return ret ? ret : copied;
1091 static int ext4_journalled_write_end(struct file *file,
1092 struct address_space *mapping,
1093 loff_t pos, unsigned len, unsigned copied,
1094 struct page *page, void *fsdata)
1096 handle_t *handle = ext4_journal_current_handle();
1097 struct inode *inode = mapping->host;
1101 int size_changed = 0;
1103 trace_ext4_journalled_write_end(inode, pos, len, copied);
1104 from = pos & (PAGE_CACHE_SIZE - 1);
1107 BUG_ON(!ext4_handle_valid(handle));
1109 if (ext4_has_inline_data(inode))
1110 copied = ext4_write_inline_data_end(inode, pos, len,
1114 if (!PageUptodate(page))
1116 page_zero_new_buffers(page, from+copied, to);
1119 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1120 to, &partial, write_end_fn);
1122 SetPageUptodate(page);
1124 size_changed = ext4_update_inode_size(inode, pos + copied);
1125 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1126 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1128 page_cache_release(page);
1131 ret2 = ext4_mark_inode_dirty(handle, inode);
1136 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1137 /* if we have allocated more blocks and copied
1138 * less. We will have blocks allocated outside
1139 * inode->i_size. So truncate them
1141 ext4_orphan_add(handle, inode);
1143 ret2 = ext4_journal_stop(handle);
1146 if (pos + len > inode->i_size) {
1147 ext4_truncate_failed_write(inode);
1149 * If truncate failed early the inode might still be
1150 * on the orphan list; we need to make sure the inode
1151 * is removed from the orphan list in that case.
1154 ext4_orphan_del(NULL, inode);
1157 return ret ? ret : copied;
1161 * Reserve a single cluster located at lblock
1163 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1165 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1166 struct ext4_inode_info *ei = EXT4_I(inode);
1167 unsigned int md_needed;
1171 * We will charge metadata quota at writeout time; this saves
1172 * us from metadata over-estimation, though we may go over by
1173 * a small amount in the end. Here we just reserve for data.
1175 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1180 * recalculate the amount of metadata blocks to reserve
1181 * in order to allocate nrblocks
1182 * worse case is one extent per block
1184 spin_lock(&ei->i_block_reservation_lock);
1186 * ext4_calc_metadata_amount() has side effects, which we have
1187 * to be prepared undo if we fail to claim space.
1190 trace_ext4_da_reserve_space(inode, 0);
1192 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1193 spin_unlock(&ei->i_block_reservation_lock);
1194 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1197 ei->i_reserved_data_blocks++;
1198 spin_unlock(&ei->i_block_reservation_lock);
1200 return 0; /* success */
1203 static void ext4_da_release_space(struct inode *inode, int to_free)
1205 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1206 struct ext4_inode_info *ei = EXT4_I(inode);
1209 return; /* Nothing to release, exit */
1211 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1213 trace_ext4_da_release_space(inode, to_free);
1214 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1216 * if there aren't enough reserved blocks, then the
1217 * counter is messed up somewhere. Since this
1218 * function is called from invalidate page, it's
1219 * harmless to return without any action.
1221 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1222 "ino %lu, to_free %d with only %d reserved "
1223 "data blocks", inode->i_ino, to_free,
1224 ei->i_reserved_data_blocks);
1226 to_free = ei->i_reserved_data_blocks;
1228 ei->i_reserved_data_blocks -= to_free;
1230 /* update fs dirty data blocks counter */
1231 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1233 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1235 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1238 static void ext4_da_page_release_reservation(struct page *page,
1239 unsigned int offset,
1240 unsigned int length)
1243 struct buffer_head *head, *bh;
1244 unsigned int curr_off = 0;
1245 struct inode *inode = page->mapping->host;
1246 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1247 unsigned int stop = offset + length;
1251 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1253 head = page_buffers(page);
1256 unsigned int next_off = curr_off + bh->b_size;
1258 if (next_off > stop)
1261 if ((offset <= curr_off) && (buffer_delay(bh))) {
1263 clear_buffer_delay(bh);
1265 curr_off = next_off;
1266 } while ((bh = bh->b_this_page) != head);
1269 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1270 ext4_es_remove_extent(inode, lblk, to_release);
1273 /* If we have released all the blocks belonging to a cluster, then we
1274 * need to release the reserved space for that cluster. */
1275 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1276 while (num_clusters > 0) {
1277 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1278 ((num_clusters - 1) << sbi->s_cluster_bits);
1279 if (sbi->s_cluster_ratio == 1 ||
1280 !ext4_find_delalloc_cluster(inode, lblk))
1281 ext4_da_release_space(inode, 1);
1288 * Delayed allocation stuff
1291 struct mpage_da_data {
1292 struct inode *inode;
1293 struct writeback_control *wbc;
1295 pgoff_t first_page; /* The first page to write */
1296 pgoff_t next_page; /* Current page to examine */
1297 pgoff_t last_page; /* Last page to examine */
1299 * Extent to map - this can be after first_page because that can be
1300 * fully mapped. We somewhat abuse m_flags to store whether the extent
1301 * is delalloc or unwritten.
1303 struct ext4_map_blocks map;
1304 struct ext4_io_submit io_submit; /* IO submission data */
1307 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1312 struct pagevec pvec;
1313 struct inode *inode = mpd->inode;
1314 struct address_space *mapping = inode->i_mapping;
1316 /* This is necessary when next_page == 0. */
1317 if (mpd->first_page >= mpd->next_page)
1320 index = mpd->first_page;
1321 end = mpd->next_page - 1;
1323 ext4_lblk_t start, last;
1324 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1325 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1326 ext4_es_remove_extent(inode, start, last - start + 1);
1329 pagevec_init(&pvec, 0);
1330 while (index <= end) {
1331 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1334 for (i = 0; i < nr_pages; i++) {
1335 struct page *page = pvec.pages[i];
1336 if (page->index > end)
1338 BUG_ON(!PageLocked(page));
1339 BUG_ON(PageWriteback(page));
1341 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1342 ClearPageUptodate(page);
1346 index = pvec.pages[nr_pages - 1]->index + 1;
1347 pagevec_release(&pvec);
1351 static void ext4_print_free_blocks(struct inode *inode)
1353 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1354 struct super_block *sb = inode->i_sb;
1355 struct ext4_inode_info *ei = EXT4_I(inode);
1357 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1358 EXT4_C2B(EXT4_SB(inode->i_sb),
1359 ext4_count_free_clusters(sb)));
1360 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1361 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1362 (long long) EXT4_C2B(EXT4_SB(sb),
1363 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1364 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1365 (long long) EXT4_C2B(EXT4_SB(sb),
1366 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1367 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1368 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1369 ei->i_reserved_data_blocks);
1373 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1375 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1379 * This function is grabs code from the very beginning of
1380 * ext4_map_blocks, but assumes that the caller is from delayed write
1381 * time. This function looks up the requested blocks and sets the
1382 * buffer delay bit under the protection of i_data_sem.
1384 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1385 struct ext4_map_blocks *map,
1386 struct buffer_head *bh)
1388 struct extent_status es;
1390 sector_t invalid_block = ~((sector_t) 0xffff);
1391 #ifdef ES_AGGRESSIVE_TEST
1392 struct ext4_map_blocks orig_map;
1394 memcpy(&orig_map, map, sizeof(*map));
1397 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1401 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1402 "logical block %lu\n", inode->i_ino, map->m_len,
1403 (unsigned long) map->m_lblk);
1405 /* Lookup extent status tree firstly */
1406 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1407 if (ext4_es_is_hole(&es)) {
1409 down_read(&EXT4_I(inode)->i_data_sem);
1414 * Delayed extent could be allocated by fallocate.
1415 * So we need to check it.
1417 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1418 map_bh(bh, inode->i_sb, invalid_block);
1420 set_buffer_delay(bh);
1424 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1425 retval = es.es_len - (iblock - es.es_lblk);
1426 if (retval > map->m_len)
1427 retval = map->m_len;
1428 map->m_len = retval;
1429 if (ext4_es_is_written(&es))
1430 map->m_flags |= EXT4_MAP_MAPPED;
1431 else if (ext4_es_is_unwritten(&es))
1432 map->m_flags |= EXT4_MAP_UNWRITTEN;
1436 #ifdef ES_AGGRESSIVE_TEST
1437 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1443 * Try to see if we can get the block without requesting a new
1444 * file system block.
1446 down_read(&EXT4_I(inode)->i_data_sem);
1447 if (ext4_has_inline_data(inode))
1449 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1450 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1452 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1458 * XXX: __block_prepare_write() unmaps passed block,
1462 * If the block was allocated from previously allocated cluster,
1463 * then we don't need to reserve it again. However we still need
1464 * to reserve metadata for every block we're going to write.
1466 if (EXT4_SB(inode->i_sb)->s_cluster_ratio <= 1 ||
1467 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1468 ret = ext4_da_reserve_space(inode, iblock);
1470 /* not enough space to reserve */
1476 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1477 ~0, EXTENT_STATUS_DELAYED);
1483 map_bh(bh, inode->i_sb, invalid_block);
1485 set_buffer_delay(bh);
1486 } else if (retval > 0) {
1488 unsigned int status;
1490 if (unlikely(retval != map->m_len)) {
1491 ext4_warning(inode->i_sb,
1492 "ES len assertion failed for inode "
1493 "%lu: retval %d != map->m_len %d",
1494 inode->i_ino, retval, map->m_len);
1498 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1499 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1500 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1501 map->m_pblk, status);
1507 up_read((&EXT4_I(inode)->i_data_sem));
1513 * This is a special get_block_t callback which is used by
1514 * ext4_da_write_begin(). It will either return mapped block or
1515 * reserve space for a single block.
1517 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1518 * We also have b_blocknr = -1 and b_bdev initialized properly
1520 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1521 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1522 * initialized properly.
1524 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1525 struct buffer_head *bh, int create)
1527 struct ext4_map_blocks map;
1530 BUG_ON(create == 0);
1531 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1533 map.m_lblk = iblock;
1537 * first, we need to know whether the block is allocated already
1538 * preallocated blocks are unmapped but should treated
1539 * the same as allocated blocks.
1541 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1545 map_bh(bh, inode->i_sb, map.m_pblk);
1546 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1548 if (buffer_unwritten(bh)) {
1549 /* A delayed write to unwritten bh should be marked
1550 * new and mapped. Mapped ensures that we don't do
1551 * get_block multiple times when we write to the same
1552 * offset and new ensures that we do proper zero out
1553 * for partial write.
1556 set_buffer_mapped(bh);
1561 static int bget_one(handle_t *handle, struct buffer_head *bh)
1567 static int bput_one(handle_t *handle, struct buffer_head *bh)
1573 static int __ext4_journalled_writepage(struct page *page,
1576 struct address_space *mapping = page->mapping;
1577 struct inode *inode = mapping->host;
1578 struct buffer_head *page_bufs = NULL;
1579 handle_t *handle = NULL;
1580 int ret = 0, err = 0;
1581 int inline_data = ext4_has_inline_data(inode);
1582 struct buffer_head *inode_bh = NULL;
1584 ClearPageChecked(page);
1587 BUG_ON(page->index != 0);
1588 BUG_ON(len > ext4_get_max_inline_size(inode));
1589 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1590 if (inode_bh == NULL)
1593 page_bufs = page_buffers(page);
1598 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1601 /* As soon as we unlock the page, it can go away, but we have
1602 * references to buffers so we are safe */
1605 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1606 ext4_writepage_trans_blocks(inode));
1607 if (IS_ERR(handle)) {
1608 ret = PTR_ERR(handle);
1612 BUG_ON(!ext4_handle_valid(handle));
1615 BUFFER_TRACE(inode_bh, "get write access");
1616 ret = ext4_journal_get_write_access(handle, inode_bh);
1618 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1621 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1622 do_journal_get_write_access);
1624 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1629 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1630 err = ext4_journal_stop(handle);
1634 if (!ext4_has_inline_data(inode))
1635 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1637 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1644 * Note that we don't need to start a transaction unless we're journaling data
1645 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1646 * need to file the inode to the transaction's list in ordered mode because if
1647 * we are writing back data added by write(), the inode is already there and if
1648 * we are writing back data modified via mmap(), no one guarantees in which
1649 * transaction the data will hit the disk. In case we are journaling data, we
1650 * cannot start transaction directly because transaction start ranks above page
1651 * lock so we have to do some magic.
1653 * This function can get called via...
1654 * - ext4_writepages after taking page lock (have journal handle)
1655 * - journal_submit_inode_data_buffers (no journal handle)
1656 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1657 * - grab_page_cache when doing write_begin (have journal handle)
1659 * We don't do any block allocation in this function. If we have page with
1660 * multiple blocks we need to write those buffer_heads that are mapped. This
1661 * is important for mmaped based write. So if we do with blocksize 1K
1662 * truncate(f, 1024);
1663 * a = mmap(f, 0, 4096);
1665 * truncate(f, 4096);
1666 * we have in the page first buffer_head mapped via page_mkwrite call back
1667 * but other buffer_heads would be unmapped but dirty (dirty done via the
1668 * do_wp_page). So writepage should write the first block. If we modify
1669 * the mmap area beyond 1024 we will again get a page_fault and the
1670 * page_mkwrite callback will do the block allocation and mark the
1671 * buffer_heads mapped.
1673 * We redirty the page if we have any buffer_heads that is either delay or
1674 * unwritten in the page.
1676 * We can get recursively called as show below.
1678 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1681 * But since we don't do any block allocation we should not deadlock.
1682 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1684 static int ext4_writepage(struct page *page,
1685 struct writeback_control *wbc)
1690 struct buffer_head *page_bufs = NULL;
1691 struct inode *inode = page->mapping->host;
1692 struct ext4_io_submit io_submit;
1693 bool keep_towrite = false;
1695 trace_ext4_writepage(page);
1696 size = i_size_read(inode);
1697 if (page->index == size >> PAGE_CACHE_SHIFT)
1698 len = size & ~PAGE_CACHE_MASK;
1700 len = PAGE_CACHE_SIZE;
1702 page_bufs = page_buffers(page);
1704 * We cannot do block allocation or other extent handling in this
1705 * function. If there are buffers needing that, we have to redirty
1706 * the page. But we may reach here when we do a journal commit via
1707 * journal_submit_inode_data_buffers() and in that case we must write
1708 * allocated buffers to achieve data=ordered mode guarantees.
1710 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1711 ext4_bh_delay_or_unwritten)) {
1712 redirty_page_for_writepage(wbc, page);
1713 if (current->flags & PF_MEMALLOC) {
1715 * For memory cleaning there's no point in writing only
1716 * some buffers. So just bail out. Warn if we came here
1717 * from direct reclaim.
1719 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1724 keep_towrite = true;
1727 if (PageChecked(page) && ext4_should_journal_data(inode))
1729 * It's mmapped pagecache. Add buffers and journal it. There
1730 * doesn't seem much point in redirtying the page here.
1732 return __ext4_journalled_writepage(page, len);
1734 ext4_io_submit_init(&io_submit, wbc);
1735 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1736 if (!io_submit.io_end) {
1737 redirty_page_for_writepage(wbc, page);
1741 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1742 ext4_io_submit(&io_submit);
1743 /* Drop io_end reference we got from init */
1744 ext4_put_io_end_defer(io_submit.io_end);
1748 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1751 loff_t size = i_size_read(mpd->inode);
1754 BUG_ON(page->index != mpd->first_page);
1755 if (page->index == size >> PAGE_CACHE_SHIFT)
1756 len = size & ~PAGE_CACHE_MASK;
1758 len = PAGE_CACHE_SIZE;
1759 clear_page_dirty_for_io(page);
1760 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1762 mpd->wbc->nr_to_write--;
1768 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1771 * mballoc gives us at most this number of blocks...
1772 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1773 * The rest of mballoc seems to handle chunks up to full group size.
1775 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1778 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1780 * @mpd - extent of blocks
1781 * @lblk - logical number of the block in the file
1782 * @bh - buffer head we want to add to the extent
1784 * The function is used to collect contig. blocks in the same state. If the
1785 * buffer doesn't require mapping for writeback and we haven't started the
1786 * extent of buffers to map yet, the function returns 'true' immediately - the
1787 * caller can write the buffer right away. Otherwise the function returns true
1788 * if the block has been added to the extent, false if the block couldn't be
1791 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1792 struct buffer_head *bh)
1794 struct ext4_map_blocks *map = &mpd->map;
1796 /* Buffer that doesn't need mapping for writeback? */
1797 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1798 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1799 /* So far no extent to map => we write the buffer right away */
1800 if (map->m_len == 0)
1805 /* First block in the extent? */
1806 if (map->m_len == 0) {
1809 map->m_flags = bh->b_state & BH_FLAGS;
1813 /* Don't go larger than mballoc is willing to allocate */
1814 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1817 /* Can we merge the block to our big extent? */
1818 if (lblk == map->m_lblk + map->m_len &&
1819 (bh->b_state & BH_FLAGS) == map->m_flags) {
1827 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1829 * @mpd - extent of blocks for mapping
1830 * @head - the first buffer in the page
1831 * @bh - buffer we should start processing from
1832 * @lblk - logical number of the block in the file corresponding to @bh
1834 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1835 * the page for IO if all buffers in this page were mapped and there's no
1836 * accumulated extent of buffers to map or add buffers in the page to the
1837 * extent of buffers to map. The function returns 1 if the caller can continue
1838 * by processing the next page, 0 if it should stop adding buffers to the
1839 * extent to map because we cannot extend it anymore. It can also return value
1840 * < 0 in case of error during IO submission.
1842 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1843 struct buffer_head *head,
1844 struct buffer_head *bh,
1847 struct inode *inode = mpd->inode;
1849 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
1850 >> inode->i_blkbits;
1853 BUG_ON(buffer_locked(bh));
1855 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
1856 /* Found extent to map? */
1859 /* Everything mapped so far and we hit EOF */
1862 } while (lblk++, (bh = bh->b_this_page) != head);
1863 /* So far everything mapped? Submit the page for IO. */
1864 if (mpd->map.m_len == 0) {
1865 err = mpage_submit_page(mpd, head->b_page);
1869 return lblk < blocks;
1873 * mpage_map_buffers - update buffers corresponding to changed extent and
1874 * submit fully mapped pages for IO
1876 * @mpd - description of extent to map, on return next extent to map
1878 * Scan buffers corresponding to changed extent (we expect corresponding pages
1879 * to be already locked) and update buffer state according to new extent state.
1880 * We map delalloc buffers to their physical location, clear unwritten bits,
1881 * and mark buffers as uninit when we perform writes to unwritten extents
1882 * and do extent conversion after IO is finished. If the last page is not fully
1883 * mapped, we update @map to the next extent in the last page that needs
1884 * mapping. Otherwise we submit the page for IO.
1886 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
1888 struct pagevec pvec;
1890 struct inode *inode = mpd->inode;
1891 struct buffer_head *head, *bh;
1892 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
1898 start = mpd->map.m_lblk >> bpp_bits;
1899 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
1900 lblk = start << bpp_bits;
1901 pblock = mpd->map.m_pblk;
1903 pagevec_init(&pvec, 0);
1904 while (start <= end) {
1905 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
1909 for (i = 0; i < nr_pages; i++) {
1910 struct page *page = pvec.pages[i];
1912 if (page->index > end)
1914 /* Up to 'end' pages must be contiguous */
1915 BUG_ON(page->index != start);
1916 bh = head = page_buffers(page);
1918 if (lblk < mpd->map.m_lblk)
1920 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
1922 * Buffer after end of mapped extent.
1923 * Find next buffer in the page to map.
1926 mpd->map.m_flags = 0;
1928 * FIXME: If dioread_nolock supports
1929 * blocksize < pagesize, we need to make
1930 * sure we add size mapped so far to
1931 * io_end->size as the following call
1932 * can submit the page for IO.
1934 err = mpage_process_page_bufs(mpd, head,
1936 pagevec_release(&pvec);
1941 if (buffer_delay(bh)) {
1942 clear_buffer_delay(bh);
1943 bh->b_blocknr = pblock++;
1945 clear_buffer_unwritten(bh);
1946 } while (lblk++, (bh = bh->b_this_page) != head);
1949 * FIXME: This is going to break if dioread_nolock
1950 * supports blocksize < pagesize as we will try to
1951 * convert potentially unmapped parts of inode.
1953 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
1954 /* Page fully mapped - let IO run! */
1955 err = mpage_submit_page(mpd, page);
1957 pagevec_release(&pvec);
1962 pagevec_release(&pvec);
1964 /* Extent fully mapped and matches with page boundary. We are done. */
1966 mpd->map.m_flags = 0;
1970 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
1972 struct inode *inode = mpd->inode;
1973 struct ext4_map_blocks *map = &mpd->map;
1974 int get_blocks_flags;
1975 int err, dioread_nolock;
1977 trace_ext4_da_write_pages_extent(inode, map);
1979 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
1980 * to convert an unwritten extent to be initialized (in the case
1981 * where we have written into one or more preallocated blocks). It is
1982 * possible that we're going to need more metadata blocks than
1983 * previously reserved. However we must not fail because we're in
1984 * writeback and there is nothing we can do about it so it might result
1985 * in data loss. So use reserved blocks to allocate metadata if
1988 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
1989 * the blocks in question are delalloc blocks. This indicates
1990 * that the blocks and quotas has already been checked when
1991 * the data was copied into the page cache.
1993 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
1994 EXT4_GET_BLOCKS_METADATA_NOFAIL;
1995 dioread_nolock = ext4_should_dioread_nolock(inode);
1997 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1998 if (map->m_flags & (1 << BH_Delay))
1999 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2001 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2004 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2005 if (!mpd->io_submit.io_end->handle &&
2006 ext4_handle_valid(handle)) {
2007 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2008 handle->h_rsv_handle = NULL;
2010 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2013 BUG_ON(map->m_len == 0);
2014 if (map->m_flags & EXT4_MAP_NEW) {
2015 struct block_device *bdev = inode->i_sb->s_bdev;
2018 for (i = 0; i < map->m_len; i++)
2019 unmap_underlying_metadata(bdev, map->m_pblk + i);
2025 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2026 * mpd->len and submit pages underlying it for IO
2028 * @handle - handle for journal operations
2029 * @mpd - extent to map
2030 * @give_up_on_write - we set this to true iff there is a fatal error and there
2031 * is no hope of writing the data. The caller should discard
2032 * dirty pages to avoid infinite loops.
2034 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2035 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2036 * them to initialized or split the described range from larger unwritten
2037 * extent. Note that we need not map all the described range since allocation
2038 * can return less blocks or the range is covered by more unwritten extents. We
2039 * cannot map more because we are limited by reserved transaction credits. On
2040 * the other hand we always make sure that the last touched page is fully
2041 * mapped so that it can be written out (and thus forward progress is
2042 * guaranteed). After mapping we submit all mapped pages for IO.
2044 static int mpage_map_and_submit_extent(handle_t *handle,
2045 struct mpage_da_data *mpd,
2046 bool *give_up_on_write)
2048 struct inode *inode = mpd->inode;
2049 struct ext4_map_blocks *map = &mpd->map;
2054 mpd->io_submit.io_end->offset =
2055 ((loff_t)map->m_lblk) << inode->i_blkbits;
2057 err = mpage_map_one_extent(handle, mpd);
2059 struct super_block *sb = inode->i_sb;
2061 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2062 goto invalidate_dirty_pages;
2064 * Let the uper layers retry transient errors.
2065 * In the case of ENOSPC, if ext4_count_free_blocks()
2066 * is non-zero, a commit should free up blocks.
2068 if ((err == -ENOMEM) ||
2069 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2071 goto update_disksize;
2074 ext4_msg(sb, KERN_CRIT,
2075 "Delayed block allocation failed for "
2076 "inode %lu at logical offset %llu with"
2077 " max blocks %u with error %d",
2079 (unsigned long long)map->m_lblk,
2080 (unsigned)map->m_len, -err);
2081 ext4_msg(sb, KERN_CRIT,
2082 "This should not happen!! Data will "
2085 ext4_print_free_blocks(inode);
2086 invalidate_dirty_pages:
2087 *give_up_on_write = true;
2092 * Update buffer state, submit mapped pages, and get us new
2095 err = mpage_map_and_submit_buffers(mpd);
2097 goto update_disksize;
2098 } while (map->m_len);
2102 * Update on-disk size after IO is submitted. Races with
2103 * truncate are avoided by checking i_size under i_data_sem.
2105 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2106 if (disksize > EXT4_I(inode)->i_disksize) {
2110 down_write(&EXT4_I(inode)->i_data_sem);
2111 i_size = i_size_read(inode);
2112 if (disksize > i_size)
2114 if (disksize > EXT4_I(inode)->i_disksize)
2115 EXT4_I(inode)->i_disksize = disksize;
2116 err2 = ext4_mark_inode_dirty(handle, inode);
2117 up_write(&EXT4_I(inode)->i_data_sem);
2119 ext4_error(inode->i_sb,
2120 "Failed to mark inode %lu dirty",
2129 * Calculate the total number of credits to reserve for one writepages
2130 * iteration. This is called from ext4_writepages(). We map an extent of
2131 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2132 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2133 * bpp - 1 blocks in bpp different extents.
2135 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2137 int bpp = ext4_journal_blocks_per_page(inode);
2139 return ext4_meta_trans_blocks(inode,
2140 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2144 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2145 * and underlying extent to map
2147 * @mpd - where to look for pages
2149 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2150 * IO immediately. When we find a page which isn't mapped we start accumulating
2151 * extent of buffers underlying these pages that needs mapping (formed by
2152 * either delayed or unwritten buffers). We also lock the pages containing
2153 * these buffers. The extent found is returned in @mpd structure (starting at
2154 * mpd->lblk with length mpd->len blocks).
2156 * Note that this function can attach bios to one io_end structure which are
2157 * neither logically nor physically contiguous. Although it may seem as an
2158 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2159 * case as we need to track IO to all buffers underlying a page in one io_end.
2161 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2163 struct address_space *mapping = mpd->inode->i_mapping;
2164 struct pagevec pvec;
2165 unsigned int nr_pages;
2166 long left = mpd->wbc->nr_to_write;
2167 pgoff_t index = mpd->first_page;
2168 pgoff_t end = mpd->last_page;
2171 int blkbits = mpd->inode->i_blkbits;
2173 struct buffer_head *head;
2175 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2176 tag = PAGECACHE_TAG_TOWRITE;
2178 tag = PAGECACHE_TAG_DIRTY;
2180 pagevec_init(&pvec, 0);
2182 mpd->next_page = index;
2183 while (index <= end) {
2184 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2185 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2189 for (i = 0; i < nr_pages; i++) {
2190 struct page *page = pvec.pages[i];
2193 * At this point, the page may be truncated or
2194 * invalidated (changing page->mapping to NULL), or
2195 * even swizzled back from swapper_space to tmpfs file
2196 * mapping. However, page->index will not change
2197 * because we have a reference on the page.
2199 if (page->index > end)
2203 * Accumulated enough dirty pages? This doesn't apply
2204 * to WB_SYNC_ALL mode. For integrity sync we have to
2205 * keep going because someone may be concurrently
2206 * dirtying pages, and we might have synced a lot of
2207 * newly appeared dirty pages, but have not synced all
2208 * of the old dirty pages.
2210 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2213 /* If we can't merge this page, we are done. */
2214 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2219 * If the page is no longer dirty, or its mapping no
2220 * longer corresponds to inode we are writing (which
2221 * means it has been truncated or invalidated), or the
2222 * page is already under writeback and we are not doing
2223 * a data integrity writeback, skip the page
2225 if (!PageDirty(page) ||
2226 (PageWriteback(page) &&
2227 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2228 unlikely(page->mapping != mapping)) {
2233 wait_on_page_writeback(page);
2234 BUG_ON(PageWriteback(page));
2236 if (mpd->map.m_len == 0)
2237 mpd->first_page = page->index;
2238 mpd->next_page = page->index + 1;
2239 /* Add all dirty buffers to mpd */
2240 lblk = ((ext4_lblk_t)page->index) <<
2241 (PAGE_CACHE_SHIFT - blkbits);
2242 head = page_buffers(page);
2243 err = mpage_process_page_bufs(mpd, head, head, lblk);
2249 pagevec_release(&pvec);
2254 pagevec_release(&pvec);
2258 static int __writepage(struct page *page, struct writeback_control *wbc,
2261 struct address_space *mapping = data;
2262 int ret = ext4_writepage(page, wbc);
2263 mapping_set_error(mapping, ret);
2267 static int ext4_writepages(struct address_space *mapping,
2268 struct writeback_control *wbc)
2270 pgoff_t writeback_index = 0;
2271 long nr_to_write = wbc->nr_to_write;
2272 int range_whole = 0;
2274 handle_t *handle = NULL;
2275 struct mpage_da_data mpd;
2276 struct inode *inode = mapping->host;
2277 int needed_blocks, rsv_blocks = 0, ret = 0;
2278 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2280 struct blk_plug plug;
2281 bool give_up_on_write = false;
2283 trace_ext4_writepages(inode, wbc);
2286 * No pages to write? This is mainly a kludge to avoid starting
2287 * a transaction for special inodes like journal inode on last iput()
2288 * because that could violate lock ordering on umount
2290 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2291 goto out_writepages;
2293 if (ext4_should_journal_data(inode)) {
2294 struct blk_plug plug;
2296 blk_start_plug(&plug);
2297 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2298 blk_finish_plug(&plug);
2299 goto out_writepages;
2303 * If the filesystem has aborted, it is read-only, so return
2304 * right away instead of dumping stack traces later on that
2305 * will obscure the real source of the problem. We test
2306 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2307 * the latter could be true if the filesystem is mounted
2308 * read-only, and in that case, ext4_writepages should
2309 * *never* be called, so if that ever happens, we would want
2312 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2314 goto out_writepages;
2317 if (ext4_should_dioread_nolock(inode)) {
2319 * We may need to convert up to one extent per block in
2320 * the page and we may dirty the inode.
2322 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2326 * If we have inline data and arrive here, it means that
2327 * we will soon create the block for the 1st page, so
2328 * we'd better clear the inline data here.
2330 if (ext4_has_inline_data(inode)) {
2331 /* Just inode will be modified... */
2332 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2333 if (IS_ERR(handle)) {
2334 ret = PTR_ERR(handle);
2335 goto out_writepages;
2337 BUG_ON(ext4_test_inode_state(inode,
2338 EXT4_STATE_MAY_INLINE_DATA));
2339 ext4_destroy_inline_data(handle, inode);
2340 ext4_journal_stop(handle);
2343 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2346 if (wbc->range_cyclic) {
2347 writeback_index = mapping->writeback_index;
2348 if (writeback_index)
2350 mpd.first_page = writeback_index;
2353 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2354 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2359 ext4_io_submit_init(&mpd.io_submit, wbc);
2361 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2362 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2364 blk_start_plug(&plug);
2365 while (!done && mpd.first_page <= mpd.last_page) {
2366 /* For each extent of pages we use new io_end */
2367 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2368 if (!mpd.io_submit.io_end) {
2374 * We have two constraints: We find one extent to map and we
2375 * must always write out whole page (makes a difference when
2376 * blocksize < pagesize) so that we don't block on IO when we
2377 * try to write out the rest of the page. Journalled mode is
2378 * not supported by delalloc.
2380 BUG_ON(ext4_should_journal_data(inode));
2381 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2383 /* start a new transaction */
2384 handle = ext4_journal_start_with_reserve(inode,
2385 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2386 if (IS_ERR(handle)) {
2387 ret = PTR_ERR(handle);
2388 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2389 "%ld pages, ino %lu; err %d", __func__,
2390 wbc->nr_to_write, inode->i_ino, ret);
2391 /* Release allocated io_end */
2392 ext4_put_io_end(mpd.io_submit.io_end);
2396 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2397 ret = mpage_prepare_extent_to_map(&mpd);
2400 ret = mpage_map_and_submit_extent(handle, &mpd,
2404 * We scanned the whole range (or exhausted
2405 * nr_to_write), submitted what was mapped and
2406 * didn't find anything needing mapping. We are
2412 ext4_journal_stop(handle);
2413 /* Submit prepared bio */
2414 ext4_io_submit(&mpd.io_submit);
2415 /* Unlock pages we didn't use */
2416 mpage_release_unused_pages(&mpd, give_up_on_write);
2417 /* Drop our io_end reference we got from init */
2418 ext4_put_io_end(mpd.io_submit.io_end);
2420 if (ret == -ENOSPC && sbi->s_journal) {
2422 * Commit the transaction which would
2423 * free blocks released in the transaction
2426 jbd2_journal_force_commit_nested(sbi->s_journal);
2430 /* Fatal error - ENOMEM, EIO... */
2434 blk_finish_plug(&plug);
2435 if (!ret && !cycled && wbc->nr_to_write > 0) {
2437 mpd.last_page = writeback_index - 1;
2443 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2445 * Set the writeback_index so that range_cyclic
2446 * mode will write it back later
2448 mapping->writeback_index = mpd.first_page;
2451 trace_ext4_writepages_result(inode, wbc, ret,
2452 nr_to_write - wbc->nr_to_write);
2456 static int ext4_nonda_switch(struct super_block *sb)
2458 s64 free_clusters, dirty_clusters;
2459 struct ext4_sb_info *sbi = EXT4_SB(sb);
2462 * switch to non delalloc mode if we are running low
2463 * on free block. The free block accounting via percpu
2464 * counters can get slightly wrong with percpu_counter_batch getting
2465 * accumulated on each CPU without updating global counters
2466 * Delalloc need an accurate free block accounting. So switch
2467 * to non delalloc when we are near to error range.
2470 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2472 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2474 * Start pushing delalloc when 1/2 of free blocks are dirty.
2476 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2477 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2479 if (2 * free_clusters < 3 * dirty_clusters ||
2480 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2482 * free block count is less than 150% of dirty blocks
2483 * or free blocks is less than watermark
2490 /* We always reserve for an inode update; the superblock could be there too */
2491 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2493 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
2494 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2497 if (pos + len <= 0x7fffffffULL)
2500 /* We might need to update the superblock to set LARGE_FILE */
2504 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2505 loff_t pos, unsigned len, unsigned flags,
2506 struct page **pagep, void **fsdata)
2508 int ret, retries = 0;
2511 struct inode *inode = mapping->host;
2514 index = pos >> PAGE_CACHE_SHIFT;
2516 if (ext4_nonda_switch(inode->i_sb)) {
2517 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2518 return ext4_write_begin(file, mapping, pos,
2519 len, flags, pagep, fsdata);
2521 *fsdata = (void *)0;
2522 trace_ext4_da_write_begin(inode, pos, len, flags);
2524 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2525 ret = ext4_da_write_inline_data_begin(mapping, inode,
2535 * grab_cache_page_write_begin() can take a long time if the
2536 * system is thrashing due to memory pressure, or if the page
2537 * is being written back. So grab it first before we start
2538 * the transaction handle. This also allows us to allocate
2539 * the page (if needed) without using GFP_NOFS.
2542 page = grab_cache_page_write_begin(mapping, index, flags);
2548 * With delayed allocation, we don't log the i_disksize update
2549 * if there is delayed block allocation. But we still need
2550 * to journalling the i_disksize update if writes to the end
2551 * of file which has an already mapped buffer.
2554 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2555 ext4_da_write_credits(inode, pos, len));
2556 if (IS_ERR(handle)) {
2557 page_cache_release(page);
2558 return PTR_ERR(handle);
2562 if (page->mapping != mapping) {
2563 /* The page got truncated from under us */
2565 page_cache_release(page);
2566 ext4_journal_stop(handle);
2569 /* In case writeback began while the page was unlocked */
2570 wait_for_stable_page(page);
2572 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2575 ext4_journal_stop(handle);
2577 * block_write_begin may have instantiated a few blocks
2578 * outside i_size. Trim these off again. Don't need
2579 * i_size_read because we hold i_mutex.
2581 if (pos + len > inode->i_size)
2582 ext4_truncate_failed_write(inode);
2584 if (ret == -ENOSPC &&
2585 ext4_should_retry_alloc(inode->i_sb, &retries))
2588 page_cache_release(page);
2597 * Check if we should update i_disksize
2598 * when write to the end of file but not require block allocation
2600 static int ext4_da_should_update_i_disksize(struct page *page,
2601 unsigned long offset)
2603 struct buffer_head *bh;
2604 struct inode *inode = page->mapping->host;
2608 bh = page_buffers(page);
2609 idx = offset >> inode->i_blkbits;
2611 for (i = 0; i < idx; i++)
2612 bh = bh->b_this_page;
2614 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2619 static int ext4_da_write_end(struct file *file,
2620 struct address_space *mapping,
2621 loff_t pos, unsigned len, unsigned copied,
2622 struct page *page, void *fsdata)
2624 struct inode *inode = mapping->host;
2626 handle_t *handle = ext4_journal_current_handle();
2628 unsigned long start, end;
2629 int write_mode = (int)(unsigned long)fsdata;
2631 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2632 return ext4_write_end(file, mapping, pos,
2633 len, copied, page, fsdata);
2635 trace_ext4_da_write_end(inode, pos, len, copied);
2636 start = pos & (PAGE_CACHE_SIZE - 1);
2637 end = start + copied - 1;
2640 * generic_write_end() will run mark_inode_dirty() if i_size
2641 * changes. So let's piggyback the i_disksize mark_inode_dirty
2644 new_i_size = pos + copied;
2645 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2646 if (ext4_has_inline_data(inode) ||
2647 ext4_da_should_update_i_disksize(page, end)) {
2648 ext4_update_i_disksize(inode, new_i_size);
2649 /* We need to mark inode dirty even if
2650 * new_i_size is less that inode->i_size
2651 * bu greater than i_disksize.(hint delalloc)
2653 ext4_mark_inode_dirty(handle, inode);
2657 if (write_mode != CONVERT_INLINE_DATA &&
2658 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2659 ext4_has_inline_data(inode))
2660 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2663 ret2 = generic_write_end(file, mapping, pos, len, copied,
2669 ret2 = ext4_journal_stop(handle);
2673 return ret ? ret : copied;
2676 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2677 unsigned int length)
2680 * Drop reserved blocks
2682 BUG_ON(!PageLocked(page));
2683 if (!page_has_buffers(page))
2686 ext4_da_page_release_reservation(page, offset, length);
2689 ext4_invalidatepage(page, offset, length);
2695 * Force all delayed allocation blocks to be allocated for a given inode.
2697 int ext4_alloc_da_blocks(struct inode *inode)
2699 trace_ext4_alloc_da_blocks(inode);
2701 if (!EXT4_I(inode)->i_reserved_data_blocks)
2705 * We do something simple for now. The filemap_flush() will
2706 * also start triggering a write of the data blocks, which is
2707 * not strictly speaking necessary (and for users of
2708 * laptop_mode, not even desirable). However, to do otherwise
2709 * would require replicating code paths in:
2711 * ext4_writepages() ->
2712 * write_cache_pages() ---> (via passed in callback function)
2713 * __mpage_da_writepage() -->
2714 * mpage_add_bh_to_extent()
2715 * mpage_da_map_blocks()
2717 * The problem is that write_cache_pages(), located in
2718 * mm/page-writeback.c, marks pages clean in preparation for
2719 * doing I/O, which is not desirable if we're not planning on
2722 * We could call write_cache_pages(), and then redirty all of
2723 * the pages by calling redirty_page_for_writepage() but that
2724 * would be ugly in the extreme. So instead we would need to
2725 * replicate parts of the code in the above functions,
2726 * simplifying them because we wouldn't actually intend to
2727 * write out the pages, but rather only collect contiguous
2728 * logical block extents, call the multi-block allocator, and
2729 * then update the buffer heads with the block allocations.
2731 * For now, though, we'll cheat by calling filemap_flush(),
2732 * which will map the blocks, and start the I/O, but not
2733 * actually wait for the I/O to complete.
2735 return filemap_flush(inode->i_mapping);
2739 * bmap() is special. It gets used by applications such as lilo and by
2740 * the swapper to find the on-disk block of a specific piece of data.
2742 * Naturally, this is dangerous if the block concerned is still in the
2743 * journal. If somebody makes a swapfile on an ext4 data-journaling
2744 * filesystem and enables swap, then they may get a nasty shock when the
2745 * data getting swapped to that swapfile suddenly gets overwritten by
2746 * the original zero's written out previously to the journal and
2747 * awaiting writeback in the kernel's buffer cache.
2749 * So, if we see any bmap calls here on a modified, data-journaled file,
2750 * take extra steps to flush any blocks which might be in the cache.
2752 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2754 struct inode *inode = mapping->host;
2759 * We can get here for an inline file via the FIBMAP ioctl
2761 if (ext4_has_inline_data(inode))
2764 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2765 test_opt(inode->i_sb, DELALLOC)) {
2767 * With delalloc we want to sync the file
2768 * so that we can make sure we allocate
2771 filemap_write_and_wait(mapping);
2774 if (EXT4_JOURNAL(inode) &&
2775 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2777 * This is a REALLY heavyweight approach, but the use of
2778 * bmap on dirty files is expected to be extremely rare:
2779 * only if we run lilo or swapon on a freshly made file
2780 * do we expect this to happen.
2782 * (bmap requires CAP_SYS_RAWIO so this does not
2783 * represent an unprivileged user DOS attack --- we'd be
2784 * in trouble if mortal users could trigger this path at
2787 * NB. EXT4_STATE_JDATA is not set on files other than
2788 * regular files. If somebody wants to bmap a directory
2789 * or symlink and gets confused because the buffer
2790 * hasn't yet been flushed to disk, they deserve
2791 * everything they get.
2794 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2795 journal = EXT4_JOURNAL(inode);
2796 jbd2_journal_lock_updates(journal);
2797 err = jbd2_journal_flush(journal);
2798 jbd2_journal_unlock_updates(journal);
2804 return generic_block_bmap(mapping, block, ext4_get_block);
2807 static int ext4_readpage(struct file *file, struct page *page)
2810 struct inode *inode = page->mapping->host;
2812 trace_ext4_readpage(page);
2814 if (ext4_has_inline_data(inode))
2815 ret = ext4_readpage_inline(inode, page);
2818 return mpage_readpage(page, ext4_get_block);
2824 ext4_readpages(struct file *file, struct address_space *mapping,
2825 struct list_head *pages, unsigned nr_pages)
2827 struct inode *inode = mapping->host;
2829 /* If the file has inline data, no need to do readpages. */
2830 if (ext4_has_inline_data(inode))
2833 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2836 static void ext4_invalidatepage(struct page *page, unsigned int offset,
2837 unsigned int length)
2839 trace_ext4_invalidatepage(page, offset, length);
2841 /* No journalling happens on data buffers when this function is used */
2842 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2844 block_invalidatepage(page, offset, length);
2847 static int __ext4_journalled_invalidatepage(struct page *page,
2848 unsigned int offset,
2849 unsigned int length)
2851 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2853 trace_ext4_journalled_invalidatepage(page, offset, length);
2856 * If it's a full truncate we just forget about the pending dirtying
2858 if (offset == 0 && length == PAGE_CACHE_SIZE)
2859 ClearPageChecked(page);
2861 return jbd2_journal_invalidatepage(journal, page, offset, length);
2864 /* Wrapper for aops... */
2865 static void ext4_journalled_invalidatepage(struct page *page,
2866 unsigned int offset,
2867 unsigned int length)
2869 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
2872 static int ext4_releasepage(struct page *page, gfp_t wait)
2874 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2876 trace_ext4_releasepage(page);
2878 /* Page has dirty journalled data -> cannot release */
2879 if (PageChecked(page))
2882 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2884 return try_to_free_buffers(page);
2888 * ext4_get_block used when preparing for a DIO write or buffer write.
2889 * We allocate an uinitialized extent if blocks haven't been allocated.
2890 * The extent will be converted to initialized after the IO is complete.
2892 int ext4_get_block_write(struct inode *inode, sector_t iblock,
2893 struct buffer_head *bh_result, int create)
2895 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2896 inode->i_ino, create);
2897 return _ext4_get_block(inode, iblock, bh_result,
2898 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2901 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2902 struct buffer_head *bh_result, int create)
2904 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2905 inode->i_ino, create);
2906 return _ext4_get_block(inode, iblock, bh_result,
2907 EXT4_GET_BLOCKS_NO_LOCK);
2910 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2911 ssize_t size, void *private)
2913 ext4_io_end_t *io_end = iocb->private;
2915 /* if not async direct IO just return */
2919 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2920 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2921 iocb->private, io_end->inode->i_ino, iocb, offset,
2924 iocb->private = NULL;
2925 io_end->offset = offset;
2926 io_end->size = size;
2927 ext4_put_io_end(io_end);
2931 * For ext4 extent files, ext4 will do direct-io write to holes,
2932 * preallocated extents, and those write extend the file, no need to
2933 * fall back to buffered IO.
2935 * For holes, we fallocate those blocks, mark them as unwritten
2936 * If those blocks were preallocated, we mark sure they are split, but
2937 * still keep the range to write as unwritten.
2939 * The unwritten extents will be converted to written when DIO is completed.
2940 * For async direct IO, since the IO may still pending when return, we
2941 * set up an end_io call back function, which will do the conversion
2942 * when async direct IO completed.
2944 * If the O_DIRECT write will extend the file then add this inode to the
2945 * orphan list. So recovery will truncate it back to the original size
2946 * if the machine crashes during the write.
2949 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2950 struct iov_iter *iter, loff_t offset)
2952 struct file *file = iocb->ki_filp;
2953 struct inode *inode = file->f_mapping->host;
2955 size_t count = iov_iter_count(iter);
2957 get_block_t *get_block_func = NULL;
2959 loff_t final_size = offset + count;
2960 ext4_io_end_t *io_end = NULL;
2962 /* Use the old path for reads and writes beyond i_size. */
2963 if (rw != WRITE || final_size > inode->i_size)
2964 return ext4_ind_direct_IO(rw, iocb, iter, offset);
2966 BUG_ON(iocb->private == NULL);
2969 * Make all waiters for direct IO properly wait also for extent
2970 * conversion. This also disallows race between truncate() and
2971 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
2974 atomic_inc(&inode->i_dio_count);
2976 /* If we do a overwrite dio, i_mutex locking can be released */
2977 overwrite = *((int *)iocb->private);
2980 down_read(&EXT4_I(inode)->i_data_sem);
2981 mutex_unlock(&inode->i_mutex);
2985 * We could direct write to holes and fallocate.
2987 * Allocated blocks to fill the hole are marked as
2988 * unwritten to prevent parallel buffered read to expose
2989 * the stale data before DIO complete the data IO.
2991 * As to previously fallocated extents, ext4 get_block will
2992 * just simply mark the buffer mapped but still keep the
2993 * extents unwritten.
2995 * For non AIO case, we will convert those unwritten extents
2996 * to written after return back from blockdev_direct_IO.
2998 * For async DIO, the conversion needs to be deferred when the
2999 * IO is completed. The ext4 end_io callback function will be
3000 * called to take care of the conversion work. Here for async
3001 * case, we allocate an io_end structure to hook to the iocb.
3003 iocb->private = NULL;
3004 ext4_inode_aio_set(inode, NULL);
3005 if (!is_sync_kiocb(iocb)) {
3006 io_end = ext4_init_io_end(inode, GFP_NOFS);
3012 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3014 iocb->private = ext4_get_io_end(io_end);
3016 * we save the io structure for current async direct
3017 * IO, so that later ext4_map_blocks() could flag the
3018 * io structure whether there is a unwritten extents
3019 * needs to be converted when IO is completed.
3021 ext4_inode_aio_set(inode, io_end);
3025 get_block_func = ext4_get_block_write_nolock;
3027 get_block_func = ext4_get_block_write;
3028 dio_flags = DIO_LOCKING;
3031 ret = dax_do_io(rw, iocb, inode, iter, offset, get_block_func,
3032 ext4_end_io_dio, dio_flags);
3034 ret = __blockdev_direct_IO(rw, iocb, inode,
3035 inode->i_sb->s_bdev, iter, offset,
3037 ext4_end_io_dio, NULL, dio_flags);
3040 * Put our reference to io_end. This can free the io_end structure e.g.
3041 * in sync IO case or in case of error. It can even perform extent
3042 * conversion if all bios we submitted finished before we got here.
3043 * Note that in that case iocb->private can be already set to NULL
3047 ext4_inode_aio_set(inode, NULL);
3048 ext4_put_io_end(io_end);
3050 * When no IO was submitted ext4_end_io_dio() was not
3051 * called so we have to put iocb's reference.
3053 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3054 WARN_ON(iocb->private != io_end);
3055 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3056 ext4_put_io_end(io_end);
3057 iocb->private = NULL;
3060 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3061 EXT4_STATE_DIO_UNWRITTEN)) {
3064 * for non AIO case, since the IO is already
3065 * completed, we could do the conversion right here
3067 err = ext4_convert_unwritten_extents(NULL, inode,
3071 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3076 inode_dio_done(inode);
3077 /* take i_mutex locking again if we do a ovewrite dio */
3079 up_read(&EXT4_I(inode)->i_data_sem);
3080 mutex_lock(&inode->i_mutex);
3086 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3087 struct iov_iter *iter, loff_t offset)
3089 struct file *file = iocb->ki_filp;
3090 struct inode *inode = file->f_mapping->host;
3091 size_t count = iov_iter_count(iter);
3095 * If we are doing data journalling we don't support O_DIRECT
3097 if (ext4_should_journal_data(inode))
3100 /* Let buffer I/O handle the inline data case. */
3101 if (ext4_has_inline_data(inode))
3104 trace_ext4_direct_IO_enter(inode, offset, count, rw);
3105 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3106 ret = ext4_ext_direct_IO(rw, iocb, iter, offset);
3108 ret = ext4_ind_direct_IO(rw, iocb, iter, offset);
3109 trace_ext4_direct_IO_exit(inode, offset, count, rw, ret);
3114 * Pages can be marked dirty completely asynchronously from ext4's journalling
3115 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3116 * much here because ->set_page_dirty is called under VFS locks. The page is
3117 * not necessarily locked.
3119 * We cannot just dirty the page and leave attached buffers clean, because the
3120 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3121 * or jbddirty because all the journalling code will explode.
3123 * So what we do is to mark the page "pending dirty" and next time writepage
3124 * is called, propagate that into the buffers appropriately.
3126 static int ext4_journalled_set_page_dirty(struct page *page)
3128 SetPageChecked(page);
3129 return __set_page_dirty_nobuffers(page);
3132 static const struct address_space_operations ext4_aops = {
3133 .readpage = ext4_readpage,
3134 .readpages = ext4_readpages,
3135 .writepage = ext4_writepage,
3136 .writepages = ext4_writepages,
3137 .write_begin = ext4_write_begin,
3138 .write_end = ext4_write_end,
3140 .invalidatepage = ext4_invalidatepage,
3141 .releasepage = ext4_releasepage,
3142 .direct_IO = ext4_direct_IO,
3143 .migratepage = buffer_migrate_page,
3144 .is_partially_uptodate = block_is_partially_uptodate,
3145 .error_remove_page = generic_error_remove_page,
3148 static const struct address_space_operations ext4_journalled_aops = {
3149 .readpage = ext4_readpage,
3150 .readpages = ext4_readpages,
3151 .writepage = ext4_writepage,
3152 .writepages = ext4_writepages,
3153 .write_begin = ext4_write_begin,
3154 .write_end = ext4_journalled_write_end,
3155 .set_page_dirty = ext4_journalled_set_page_dirty,
3157 .invalidatepage = ext4_journalled_invalidatepage,
3158 .releasepage = ext4_releasepage,
3159 .direct_IO = ext4_direct_IO,
3160 .is_partially_uptodate = block_is_partially_uptodate,
3161 .error_remove_page = generic_error_remove_page,
3164 static const struct address_space_operations ext4_da_aops = {
3165 .readpage = ext4_readpage,
3166 .readpages = ext4_readpages,
3167 .writepage = ext4_writepage,
3168 .writepages = ext4_writepages,
3169 .write_begin = ext4_da_write_begin,
3170 .write_end = ext4_da_write_end,
3172 .invalidatepage = ext4_da_invalidatepage,
3173 .releasepage = ext4_releasepage,
3174 .direct_IO = ext4_direct_IO,
3175 .migratepage = buffer_migrate_page,
3176 .is_partially_uptodate = block_is_partially_uptodate,
3177 .error_remove_page = generic_error_remove_page,
3180 void ext4_set_aops(struct inode *inode)
3182 switch (ext4_inode_journal_mode(inode)) {
3183 case EXT4_INODE_ORDERED_DATA_MODE:
3184 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3186 case EXT4_INODE_WRITEBACK_DATA_MODE:
3187 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3189 case EXT4_INODE_JOURNAL_DATA_MODE:
3190 inode->i_mapping->a_ops = &ext4_journalled_aops;
3195 if (test_opt(inode->i_sb, DELALLOC))
3196 inode->i_mapping->a_ops = &ext4_da_aops;
3198 inode->i_mapping->a_ops = &ext4_aops;
3201 static int __ext4_block_zero_page_range(handle_t *handle,
3202 struct address_space *mapping, loff_t from, loff_t length)
3204 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3205 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3206 unsigned blocksize, pos;
3208 struct inode *inode = mapping->host;
3209 struct buffer_head *bh;
3213 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3214 mapping_gfp_mask(mapping) & ~__GFP_FS);
3218 blocksize = inode->i_sb->s_blocksize;
3220 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3222 if (!page_has_buffers(page))
3223 create_empty_buffers(page, blocksize, 0);
3225 /* Find the buffer that contains "offset" */
3226 bh = page_buffers(page);
3228 while (offset >= pos) {
3229 bh = bh->b_this_page;
3233 if (buffer_freed(bh)) {
3234 BUFFER_TRACE(bh, "freed: skip");
3237 if (!buffer_mapped(bh)) {
3238 BUFFER_TRACE(bh, "unmapped");
3239 ext4_get_block(inode, iblock, bh, 0);
3240 /* unmapped? It's a hole - nothing to do */
3241 if (!buffer_mapped(bh)) {
3242 BUFFER_TRACE(bh, "still unmapped");
3247 /* Ok, it's mapped. Make sure it's up-to-date */
3248 if (PageUptodate(page))
3249 set_buffer_uptodate(bh);
3251 if (!buffer_uptodate(bh)) {
3253 ll_rw_block(READ, 1, &bh);
3255 /* Uhhuh. Read error. Complain and punt. */
3256 if (!buffer_uptodate(bh))
3259 if (ext4_should_journal_data(inode)) {
3260 BUFFER_TRACE(bh, "get write access");
3261 err = ext4_journal_get_write_access(handle, bh);
3265 zero_user(page, offset, length);
3266 BUFFER_TRACE(bh, "zeroed end of block");
3268 if (ext4_should_journal_data(inode)) {
3269 err = ext4_handle_dirty_metadata(handle, inode, bh);
3272 mark_buffer_dirty(bh);
3273 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3274 err = ext4_jbd2_file_inode(handle, inode);
3279 page_cache_release(page);
3284 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3285 * starting from file offset 'from'. The range to be zero'd must
3286 * be contained with in one block. If the specified range exceeds
3287 * the end of the block it will be shortened to end of the block
3288 * that cooresponds to 'from'
3290 static int ext4_block_zero_page_range(handle_t *handle,
3291 struct address_space *mapping, loff_t from, loff_t length)
3293 struct inode *inode = mapping->host;
3294 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3295 unsigned blocksize = inode->i_sb->s_blocksize;
3296 unsigned max = blocksize - (offset & (blocksize - 1));
3299 * correct length if it does not fall between
3300 * 'from' and the end of the block
3302 if (length > max || length < 0)
3306 return dax_zero_page_range(inode, from, length, ext4_get_block);
3307 return __ext4_block_zero_page_range(handle, mapping, from, length);
3311 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3312 * up to the end of the block which corresponds to `from'.
3313 * This required during truncate. We need to physically zero the tail end
3314 * of that block so it doesn't yield old data if the file is later grown.
3316 static int ext4_block_truncate_page(handle_t *handle,
3317 struct address_space *mapping, loff_t from)
3319 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3322 struct inode *inode = mapping->host;
3324 blocksize = inode->i_sb->s_blocksize;
3325 length = blocksize - (offset & (blocksize - 1));
3327 return ext4_block_zero_page_range(handle, mapping, from, length);
3330 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3331 loff_t lstart, loff_t length)
3333 struct super_block *sb = inode->i_sb;
3334 struct address_space *mapping = inode->i_mapping;
3335 unsigned partial_start, partial_end;
3336 ext4_fsblk_t start, end;
3337 loff_t byte_end = (lstart + length - 1);
3340 partial_start = lstart & (sb->s_blocksize - 1);
3341 partial_end = byte_end & (sb->s_blocksize - 1);
3343 start = lstart >> sb->s_blocksize_bits;
3344 end = byte_end >> sb->s_blocksize_bits;
3346 /* Handle partial zero within the single block */
3348 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3349 err = ext4_block_zero_page_range(handle, mapping,
3353 /* Handle partial zero out on the start of the range */
3354 if (partial_start) {
3355 err = ext4_block_zero_page_range(handle, mapping,
3356 lstart, sb->s_blocksize);
3360 /* Handle partial zero out on the end of the range */
3361 if (partial_end != sb->s_blocksize - 1)
3362 err = ext4_block_zero_page_range(handle, mapping,
3363 byte_end - partial_end,
3368 int ext4_can_truncate(struct inode *inode)
3370 if (S_ISREG(inode->i_mode))
3372 if (S_ISDIR(inode->i_mode))
3374 if (S_ISLNK(inode->i_mode))
3375 return !ext4_inode_is_fast_symlink(inode);
3380 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3381 * associated with the given offset and length
3383 * @inode: File inode
3384 * @offset: The offset where the hole will begin
3385 * @len: The length of the hole
3387 * Returns: 0 on success or negative on failure
3390 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3392 struct super_block *sb = inode->i_sb;
3393 ext4_lblk_t first_block, stop_block;
3394 struct address_space *mapping = inode->i_mapping;
3395 loff_t first_block_offset, last_block_offset;
3397 unsigned int credits;
3400 if (!S_ISREG(inode->i_mode))
3403 trace_ext4_punch_hole(inode, offset, length, 0);
3406 * Write out all dirty pages to avoid race conditions
3407 * Then release them.
3409 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3410 ret = filemap_write_and_wait_range(mapping, offset,
3411 offset + length - 1);
3416 mutex_lock(&inode->i_mutex);
3418 /* No need to punch hole beyond i_size */
3419 if (offset >= inode->i_size)
3423 * If the hole extends beyond i_size, set the hole
3424 * to end after the page that contains i_size
3426 if (offset + length > inode->i_size) {
3427 length = inode->i_size +
3428 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3432 if (offset & (sb->s_blocksize - 1) ||
3433 (offset + length) & (sb->s_blocksize - 1)) {
3435 * Attach jinode to inode for jbd2 if we do any zeroing of
3438 ret = ext4_inode_attach_jinode(inode);
3444 first_block_offset = round_up(offset, sb->s_blocksize);
3445 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3447 /* Now release the pages and zero block aligned part of pages*/
3448 if (last_block_offset > first_block_offset)
3449 truncate_pagecache_range(inode, first_block_offset,
3452 /* Wait all existing dio workers, newcomers will block on i_mutex */
3453 ext4_inode_block_unlocked_dio(inode);
3454 inode_dio_wait(inode);
3456 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3457 credits = ext4_writepage_trans_blocks(inode);
3459 credits = ext4_blocks_for_truncate(inode);
3460 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3461 if (IS_ERR(handle)) {
3462 ret = PTR_ERR(handle);
3463 ext4_std_error(sb, ret);
3467 ret = ext4_zero_partial_blocks(handle, inode, offset,
3472 first_block = (offset + sb->s_blocksize - 1) >>
3473 EXT4_BLOCK_SIZE_BITS(sb);
3474 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3476 /* If there are no blocks to remove, return now */
3477 if (first_block >= stop_block)
3480 down_write(&EXT4_I(inode)->i_data_sem);
3481 ext4_discard_preallocations(inode);
3483 ret = ext4_es_remove_extent(inode, first_block,
3484 stop_block - first_block);
3486 up_write(&EXT4_I(inode)->i_data_sem);
3490 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3491 ret = ext4_ext_remove_space(inode, first_block,
3494 ret = ext4_ind_remove_space(handle, inode, first_block,
3497 up_write(&EXT4_I(inode)->i_data_sem);
3499 ext4_handle_sync(handle);
3501 /* Now release the pages again to reduce race window */
3502 if (last_block_offset > first_block_offset)
3503 truncate_pagecache_range(inode, first_block_offset,
3506 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3507 ext4_mark_inode_dirty(handle, inode);
3509 ext4_journal_stop(handle);
3511 ext4_inode_resume_unlocked_dio(inode);
3513 mutex_unlock(&inode->i_mutex);
3517 int ext4_inode_attach_jinode(struct inode *inode)
3519 struct ext4_inode_info *ei = EXT4_I(inode);
3520 struct jbd2_inode *jinode;
3522 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3525 jinode = jbd2_alloc_inode(GFP_KERNEL);
3526 spin_lock(&inode->i_lock);
3529 spin_unlock(&inode->i_lock);
3532 ei->jinode = jinode;
3533 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3536 spin_unlock(&inode->i_lock);
3537 if (unlikely(jinode != NULL))
3538 jbd2_free_inode(jinode);
3545 * We block out ext4_get_block() block instantiations across the entire
3546 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3547 * simultaneously on behalf of the same inode.
3549 * As we work through the truncate and commit bits of it to the journal there
3550 * is one core, guiding principle: the file's tree must always be consistent on
3551 * disk. We must be able to restart the truncate after a crash.
3553 * The file's tree may be transiently inconsistent in memory (although it
3554 * probably isn't), but whenever we close off and commit a journal transaction,
3555 * the contents of (the filesystem + the journal) must be consistent and
3556 * restartable. It's pretty simple, really: bottom up, right to left (although
3557 * left-to-right works OK too).
3559 * Note that at recovery time, journal replay occurs *before* the restart of
3560 * truncate against the orphan inode list.
3562 * The committed inode has the new, desired i_size (which is the same as
3563 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3564 * that this inode's truncate did not complete and it will again call
3565 * ext4_truncate() to have another go. So there will be instantiated blocks
3566 * to the right of the truncation point in a crashed ext4 filesystem. But
3567 * that's fine - as long as they are linked from the inode, the post-crash
3568 * ext4_truncate() run will find them and release them.
3570 void ext4_truncate(struct inode *inode)
3572 struct ext4_inode_info *ei = EXT4_I(inode);
3573 unsigned int credits;
3575 struct address_space *mapping = inode->i_mapping;
3578 * There is a possibility that we're either freeing the inode
3579 * or it's a completely new inode. In those cases we might not
3580 * have i_mutex locked because it's not necessary.
3582 if (!(inode->i_state & (I_NEW|I_FREEING)))
3583 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3584 trace_ext4_truncate_enter(inode);
3586 if (!ext4_can_truncate(inode))
3589 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3591 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3592 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3594 if (ext4_has_inline_data(inode)) {
3597 ext4_inline_data_truncate(inode, &has_inline);
3602 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3603 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3604 if (ext4_inode_attach_jinode(inode) < 0)
3608 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3609 credits = ext4_writepage_trans_blocks(inode);
3611 credits = ext4_blocks_for_truncate(inode);
3613 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3614 if (IS_ERR(handle)) {
3615 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3619 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3620 ext4_block_truncate_page(handle, mapping, inode->i_size);
3623 * We add the inode to the orphan list, so that if this
3624 * truncate spans multiple transactions, and we crash, we will
3625 * resume the truncate when the filesystem recovers. It also
3626 * marks the inode dirty, to catch the new size.
3628 * Implication: the file must always be in a sane, consistent
3629 * truncatable state while each transaction commits.
3631 if (ext4_orphan_add(handle, inode))
3634 down_write(&EXT4_I(inode)->i_data_sem);
3636 ext4_discard_preallocations(inode);
3638 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3639 ext4_ext_truncate(handle, inode);
3641 ext4_ind_truncate(handle, inode);
3643 up_write(&ei->i_data_sem);
3646 ext4_handle_sync(handle);
3650 * If this was a simple ftruncate() and the file will remain alive,
3651 * then we need to clear up the orphan record which we created above.
3652 * However, if this was a real unlink then we were called by
3653 * ext4_evict_inode(), and we allow that function to clean up the
3654 * orphan info for us.
3657 ext4_orphan_del(handle, inode);
3659 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3660 ext4_mark_inode_dirty(handle, inode);
3661 ext4_journal_stop(handle);
3663 trace_ext4_truncate_exit(inode);
3667 * ext4_get_inode_loc returns with an extra refcount against the inode's
3668 * underlying buffer_head on success. If 'in_mem' is true, we have all
3669 * data in memory that is needed to recreate the on-disk version of this
3672 static int __ext4_get_inode_loc(struct inode *inode,
3673 struct ext4_iloc *iloc, int in_mem)
3675 struct ext4_group_desc *gdp;
3676 struct buffer_head *bh;
3677 struct super_block *sb = inode->i_sb;
3679 int inodes_per_block, inode_offset;
3682 if (!ext4_valid_inum(sb, inode->i_ino))
3685 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3686 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3691 * Figure out the offset within the block group inode table
3693 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3694 inode_offset = ((inode->i_ino - 1) %
3695 EXT4_INODES_PER_GROUP(sb));
3696 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3697 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3699 bh = sb_getblk(sb, block);
3702 if (!buffer_uptodate(bh)) {
3706 * If the buffer has the write error flag, we have failed
3707 * to write out another inode in the same block. In this
3708 * case, we don't have to read the block because we may
3709 * read the old inode data successfully.
3711 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3712 set_buffer_uptodate(bh);
3714 if (buffer_uptodate(bh)) {
3715 /* someone brought it uptodate while we waited */
3721 * If we have all information of the inode in memory and this
3722 * is the only valid inode in the block, we need not read the
3726 struct buffer_head *bitmap_bh;
3729 start = inode_offset & ~(inodes_per_block - 1);
3731 /* Is the inode bitmap in cache? */
3732 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3733 if (unlikely(!bitmap_bh))
3737 * If the inode bitmap isn't in cache then the
3738 * optimisation may end up performing two reads instead
3739 * of one, so skip it.
3741 if (!buffer_uptodate(bitmap_bh)) {
3745 for (i = start; i < start + inodes_per_block; i++) {
3746 if (i == inode_offset)
3748 if (ext4_test_bit(i, bitmap_bh->b_data))
3752 if (i == start + inodes_per_block) {
3753 /* all other inodes are free, so skip I/O */
3754 memset(bh->b_data, 0, bh->b_size);
3755 set_buffer_uptodate(bh);
3763 * If we need to do any I/O, try to pre-readahead extra
3764 * blocks from the inode table.
3766 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3767 ext4_fsblk_t b, end, table;
3769 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
3771 table = ext4_inode_table(sb, gdp);
3772 /* s_inode_readahead_blks is always a power of 2 */
3773 b = block & ~((ext4_fsblk_t) ra_blks - 1);
3777 num = EXT4_INODES_PER_GROUP(sb);
3778 if (ext4_has_group_desc_csum(sb))
3779 num -= ext4_itable_unused_count(sb, gdp);
3780 table += num / inodes_per_block;
3784 sb_breadahead(sb, b++);
3788 * There are other valid inodes in the buffer, this inode
3789 * has in-inode xattrs, or we don't have this inode in memory.
3790 * Read the block from disk.
3792 trace_ext4_load_inode(inode);
3794 bh->b_end_io = end_buffer_read_sync;
3795 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3797 if (!buffer_uptodate(bh)) {
3798 EXT4_ERROR_INODE_BLOCK(inode, block,
3799 "unable to read itable block");
3809 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3811 /* We have all inode data except xattrs in memory here. */
3812 return __ext4_get_inode_loc(inode, iloc,
3813 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3816 void ext4_set_inode_flags(struct inode *inode)
3818 unsigned int flags = EXT4_I(inode)->i_flags;
3819 unsigned int new_fl = 0;
3821 if (flags & EXT4_SYNC_FL)
3823 if (flags & EXT4_APPEND_FL)
3825 if (flags & EXT4_IMMUTABLE_FL)
3826 new_fl |= S_IMMUTABLE;
3827 if (flags & EXT4_NOATIME_FL)
3828 new_fl |= S_NOATIME;
3829 if (flags & EXT4_DIRSYNC_FL)
3830 new_fl |= S_DIRSYNC;
3831 if (test_opt(inode->i_sb, DAX))
3833 inode_set_flags(inode, new_fl,
3834 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
3837 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3838 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3840 unsigned int vfs_fl;
3841 unsigned long old_fl, new_fl;
3844 vfs_fl = ei->vfs_inode.i_flags;
3845 old_fl = ei->i_flags;
3846 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3847 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3849 if (vfs_fl & S_SYNC)
3850 new_fl |= EXT4_SYNC_FL;
3851 if (vfs_fl & S_APPEND)
3852 new_fl |= EXT4_APPEND_FL;
3853 if (vfs_fl & S_IMMUTABLE)
3854 new_fl |= EXT4_IMMUTABLE_FL;
3855 if (vfs_fl & S_NOATIME)
3856 new_fl |= EXT4_NOATIME_FL;
3857 if (vfs_fl & S_DIRSYNC)
3858 new_fl |= EXT4_DIRSYNC_FL;
3859 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3862 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3863 struct ext4_inode_info *ei)
3866 struct inode *inode = &(ei->vfs_inode);
3867 struct super_block *sb = inode->i_sb;
3869 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3870 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3871 /* we are using combined 48 bit field */
3872 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3873 le32_to_cpu(raw_inode->i_blocks_lo);
3874 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3875 /* i_blocks represent file system block size */
3876 return i_blocks << (inode->i_blkbits - 9);
3881 return le32_to_cpu(raw_inode->i_blocks_lo);
3885 static inline void ext4_iget_extra_inode(struct inode *inode,
3886 struct ext4_inode *raw_inode,
3887 struct ext4_inode_info *ei)
3889 __le32 *magic = (void *)raw_inode +
3890 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3891 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3892 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3893 ext4_find_inline_data_nolock(inode);
3895 EXT4_I(inode)->i_inline_off = 0;
3898 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3900 struct ext4_iloc iloc;
3901 struct ext4_inode *raw_inode;
3902 struct ext4_inode_info *ei;
3903 struct inode *inode;
3904 journal_t *journal = EXT4_SB(sb)->s_journal;
3910 inode = iget_locked(sb, ino);
3912 return ERR_PTR(-ENOMEM);
3913 if (!(inode->i_state & I_NEW))
3919 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3922 raw_inode = ext4_raw_inode(&iloc);
3924 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3925 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3926 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3927 EXT4_INODE_SIZE(inode->i_sb)) {
3928 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3929 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3930 EXT4_INODE_SIZE(inode->i_sb));
3935 ei->i_extra_isize = 0;
3937 /* Precompute checksum seed for inode metadata */
3938 if (ext4_has_metadata_csum(sb)) {
3939 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3941 __le32 inum = cpu_to_le32(inode->i_ino);
3942 __le32 gen = raw_inode->i_generation;
3943 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3945 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3949 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3950 EXT4_ERROR_INODE(inode, "checksum invalid");
3955 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3956 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3957 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3958 if (!(test_opt(inode->i_sb, NO_UID32))) {
3959 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3960 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3962 i_uid_write(inode, i_uid);
3963 i_gid_write(inode, i_gid);
3964 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3966 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3967 ei->i_inline_off = 0;
3968 ei->i_dir_start_lookup = 0;
3969 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3970 /* We now have enough fields to check if the inode was active or not.
3971 * This is needed because nfsd might try to access dead inodes
3972 * the test is that same one that e2fsck uses
3973 * NeilBrown 1999oct15
3975 if (inode->i_nlink == 0) {
3976 if ((inode->i_mode == 0 ||
3977 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
3978 ino != EXT4_BOOT_LOADER_INO) {
3979 /* this inode is deleted */
3983 /* The only unlinked inodes we let through here have
3984 * valid i_mode and are being read by the orphan
3985 * recovery code: that's fine, we're about to complete
3986 * the process of deleting those.
3987 * OR it is the EXT4_BOOT_LOADER_INO which is
3988 * not initialized on a new filesystem. */
3990 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3991 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3992 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3993 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3995 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3996 inode->i_size = ext4_isize(raw_inode);
3997 ei->i_disksize = inode->i_size;
3999 ei->i_reserved_quota = 0;
4001 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4002 ei->i_block_group = iloc.block_group;
4003 ei->i_last_alloc_group = ~0;
4005 * NOTE! The in-memory inode i_data array is in little-endian order
4006 * even on big-endian machines: we do NOT byteswap the block numbers!
4008 for (block = 0; block < EXT4_N_BLOCKS; block++)
4009 ei->i_data[block] = raw_inode->i_block[block];
4010 INIT_LIST_HEAD(&ei->i_orphan);
4013 * Set transaction id's of transactions that have to be committed
4014 * to finish f[data]sync. We set them to currently running transaction
4015 * as we cannot be sure that the inode or some of its metadata isn't
4016 * part of the transaction - the inode could have been reclaimed and
4017 * now it is reread from disk.
4020 transaction_t *transaction;
4023 read_lock(&journal->j_state_lock);
4024 if (journal->j_running_transaction)
4025 transaction = journal->j_running_transaction;
4027 transaction = journal->j_committing_transaction;
4029 tid = transaction->t_tid;
4031 tid = journal->j_commit_sequence;
4032 read_unlock(&journal->j_state_lock);
4033 ei->i_sync_tid = tid;
4034 ei->i_datasync_tid = tid;
4037 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4038 if (ei->i_extra_isize == 0) {
4039 /* The extra space is currently unused. Use it. */
4040 ei->i_extra_isize = sizeof(struct ext4_inode) -
4041 EXT4_GOOD_OLD_INODE_SIZE;
4043 ext4_iget_extra_inode(inode, raw_inode, ei);
4047 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4048 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4049 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4050 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4052 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4053 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4054 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4055 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4057 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4062 if (ei->i_file_acl &&
4063 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4064 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4068 } else if (!ext4_has_inline_data(inode)) {
4069 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4070 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4071 (S_ISLNK(inode->i_mode) &&
4072 !ext4_inode_is_fast_symlink(inode))))
4073 /* Validate extent which is part of inode */
4074 ret = ext4_ext_check_inode(inode);
4075 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4076 (S_ISLNK(inode->i_mode) &&
4077 !ext4_inode_is_fast_symlink(inode))) {
4078 /* Validate block references which are part of inode */
4079 ret = ext4_ind_check_inode(inode);
4085 if (S_ISREG(inode->i_mode)) {
4086 inode->i_op = &ext4_file_inode_operations;
4087 if (test_opt(inode->i_sb, DAX))
4088 inode->i_fop = &ext4_dax_file_operations;
4090 inode->i_fop = &ext4_file_operations;
4091 ext4_set_aops(inode);
4092 } else if (S_ISDIR(inode->i_mode)) {
4093 inode->i_op = &ext4_dir_inode_operations;
4094 inode->i_fop = &ext4_dir_operations;
4095 } else if (S_ISLNK(inode->i_mode)) {
4096 if (ext4_inode_is_fast_symlink(inode)) {
4097 inode->i_op = &ext4_fast_symlink_inode_operations;
4098 nd_terminate_link(ei->i_data, inode->i_size,
4099 sizeof(ei->i_data) - 1);
4101 inode->i_op = &ext4_symlink_inode_operations;
4102 ext4_set_aops(inode);
4104 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4105 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4106 inode->i_op = &ext4_special_inode_operations;
4107 if (raw_inode->i_block[0])
4108 init_special_inode(inode, inode->i_mode,
4109 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4111 init_special_inode(inode, inode->i_mode,
4112 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4113 } else if (ino == EXT4_BOOT_LOADER_INO) {
4114 make_bad_inode(inode);
4117 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4121 ext4_set_inode_flags(inode);
4122 unlock_new_inode(inode);
4128 return ERR_PTR(ret);
4131 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4133 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4134 return ERR_PTR(-EIO);
4135 return ext4_iget(sb, ino);
4138 static int ext4_inode_blocks_set(handle_t *handle,
4139 struct ext4_inode *raw_inode,
4140 struct ext4_inode_info *ei)
4142 struct inode *inode = &(ei->vfs_inode);
4143 u64 i_blocks = inode->i_blocks;
4144 struct super_block *sb = inode->i_sb;
4146 if (i_blocks <= ~0U) {
4148 * i_blocks can be represented in a 32 bit variable
4149 * as multiple of 512 bytes
4151 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4152 raw_inode->i_blocks_high = 0;
4153 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4156 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4159 if (i_blocks <= 0xffffffffffffULL) {
4161 * i_blocks can be represented in a 48 bit variable
4162 * as multiple of 512 bytes
4164 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4165 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4166 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4168 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4169 /* i_block is stored in file system block size */
4170 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4171 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4172 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4177 struct other_inode {
4178 unsigned long orig_ino;
4179 struct ext4_inode *raw_inode;
4182 static int other_inode_match(struct inode * inode, unsigned long ino,
4185 struct other_inode *oi = (struct other_inode *) data;
4187 if ((inode->i_ino != ino) ||
4188 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4189 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4190 ((inode->i_state & I_DIRTY_TIME) == 0))
4192 spin_lock(&inode->i_lock);
4193 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4194 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4195 (inode->i_state & I_DIRTY_TIME)) {
4196 struct ext4_inode_info *ei = EXT4_I(inode);
4198 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4199 spin_unlock(&inode->i_lock);
4201 spin_lock(&ei->i_raw_lock);
4202 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4203 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4204 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4205 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4206 spin_unlock(&ei->i_raw_lock);
4207 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4210 spin_unlock(&inode->i_lock);
4215 * Opportunistically update the other time fields for other inodes in
4216 * the same inode table block.
4218 static void ext4_update_other_inodes_time(struct super_block *sb,
4219 unsigned long orig_ino, char *buf)
4221 struct other_inode oi;
4223 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4224 int inode_size = EXT4_INODE_SIZE(sb);
4226 oi.orig_ino = orig_ino;
4227 ino = orig_ino & ~(inodes_per_block - 1);
4228 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4229 if (ino == orig_ino)
4231 oi.raw_inode = (struct ext4_inode *) buf;
4232 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4237 * Post the struct inode info into an on-disk inode location in the
4238 * buffer-cache. This gobbles the caller's reference to the
4239 * buffer_head in the inode location struct.
4241 * The caller must have write access to iloc->bh.
4243 static int ext4_do_update_inode(handle_t *handle,
4244 struct inode *inode,
4245 struct ext4_iloc *iloc)
4247 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4248 struct ext4_inode_info *ei = EXT4_I(inode);
4249 struct buffer_head *bh = iloc->bh;
4250 struct super_block *sb = inode->i_sb;
4251 int err = 0, rc, block;
4252 int need_datasync = 0, set_large_file = 0;
4256 spin_lock(&ei->i_raw_lock);
4258 /* For fields not tracked in the in-memory inode,
4259 * initialise them to zero for new inodes. */
4260 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4261 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4263 ext4_get_inode_flags(ei);
4264 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4265 i_uid = i_uid_read(inode);
4266 i_gid = i_gid_read(inode);
4267 if (!(test_opt(inode->i_sb, NO_UID32))) {
4268 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4269 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4271 * Fix up interoperability with old kernels. Otherwise, old inodes get
4272 * re-used with the upper 16 bits of the uid/gid intact
4275 raw_inode->i_uid_high =
4276 cpu_to_le16(high_16_bits(i_uid));
4277 raw_inode->i_gid_high =
4278 cpu_to_le16(high_16_bits(i_gid));
4280 raw_inode->i_uid_high = 0;
4281 raw_inode->i_gid_high = 0;
4284 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4285 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4286 raw_inode->i_uid_high = 0;
4287 raw_inode->i_gid_high = 0;
4289 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4291 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4292 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4293 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4294 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4296 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4298 spin_unlock(&ei->i_raw_lock);
4301 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4302 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4303 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4304 raw_inode->i_file_acl_high =
4305 cpu_to_le16(ei->i_file_acl >> 32);
4306 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4307 if (ei->i_disksize != ext4_isize(raw_inode)) {
4308 ext4_isize_set(raw_inode, ei->i_disksize);
4311 if (ei->i_disksize > 0x7fffffffULL) {
4312 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4313 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4314 EXT4_SB(sb)->s_es->s_rev_level ==
4315 cpu_to_le32(EXT4_GOOD_OLD_REV))
4318 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4319 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4320 if (old_valid_dev(inode->i_rdev)) {
4321 raw_inode->i_block[0] =
4322 cpu_to_le32(old_encode_dev(inode->i_rdev));
4323 raw_inode->i_block[1] = 0;
4325 raw_inode->i_block[0] = 0;
4326 raw_inode->i_block[1] =
4327 cpu_to_le32(new_encode_dev(inode->i_rdev));
4328 raw_inode->i_block[2] = 0;
4330 } else if (!ext4_has_inline_data(inode)) {
4331 for (block = 0; block < EXT4_N_BLOCKS; block++)
4332 raw_inode->i_block[block] = ei->i_data[block];
4335 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4336 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4337 if (ei->i_extra_isize) {
4338 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4339 raw_inode->i_version_hi =
4340 cpu_to_le32(inode->i_version >> 32);
4341 raw_inode->i_extra_isize =
4342 cpu_to_le16(ei->i_extra_isize);
4345 ext4_inode_csum_set(inode, raw_inode, ei);
4346 spin_unlock(&ei->i_raw_lock);
4347 if (inode->i_sb->s_flags & MS_LAZYTIME)
4348 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4351 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4352 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4355 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4356 if (set_large_file) {
4357 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4358 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4361 ext4_update_dynamic_rev(sb);
4362 EXT4_SET_RO_COMPAT_FEATURE(sb,
4363 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4364 ext4_handle_sync(handle);
4365 err = ext4_handle_dirty_super(handle, sb);
4367 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4370 ext4_std_error(inode->i_sb, err);
4375 * ext4_write_inode()
4377 * We are called from a few places:
4379 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4380 * Here, there will be no transaction running. We wait for any running
4381 * transaction to commit.
4383 * - Within flush work (sys_sync(), kupdate and such).
4384 * We wait on commit, if told to.
4386 * - Within iput_final() -> write_inode_now()
4387 * We wait on commit, if told to.
4389 * In all cases it is actually safe for us to return without doing anything,
4390 * because the inode has been copied into a raw inode buffer in
4391 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4394 * Note that we are absolutely dependent upon all inode dirtiers doing the
4395 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4396 * which we are interested.
4398 * It would be a bug for them to not do this. The code:
4400 * mark_inode_dirty(inode)
4402 * inode->i_size = expr;
4404 * is in error because write_inode() could occur while `stuff()' is running,
4405 * and the new i_size will be lost. Plus the inode will no longer be on the
4406 * superblock's dirty inode list.
4408 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4412 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4415 if (EXT4_SB(inode->i_sb)->s_journal) {
4416 if (ext4_journal_current_handle()) {
4417 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4423 * No need to force transaction in WB_SYNC_NONE mode. Also
4424 * ext4_sync_fs() will force the commit after everything is
4427 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4430 err = ext4_force_commit(inode->i_sb);
4432 struct ext4_iloc iloc;
4434 err = __ext4_get_inode_loc(inode, &iloc, 0);
4438 * sync(2) will flush the whole buffer cache. No need to do
4439 * it here separately for each inode.
4441 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4442 sync_dirty_buffer(iloc.bh);
4443 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4444 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4445 "IO error syncing inode");
4454 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4455 * buffers that are attached to a page stradding i_size and are undergoing
4456 * commit. In that case we have to wait for commit to finish and try again.
4458 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4462 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4463 tid_t commit_tid = 0;
4466 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4468 * All buffers in the last page remain valid? Then there's nothing to
4469 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4472 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4475 page = find_lock_page(inode->i_mapping,
4476 inode->i_size >> PAGE_CACHE_SHIFT);
4479 ret = __ext4_journalled_invalidatepage(page, offset,
4480 PAGE_CACHE_SIZE - offset);
4482 page_cache_release(page);
4486 read_lock(&journal->j_state_lock);
4487 if (journal->j_committing_transaction)
4488 commit_tid = journal->j_committing_transaction->t_tid;
4489 read_unlock(&journal->j_state_lock);
4491 jbd2_log_wait_commit(journal, commit_tid);
4498 * Called from notify_change.
4500 * We want to trap VFS attempts to truncate the file as soon as
4501 * possible. In particular, we want to make sure that when the VFS
4502 * shrinks i_size, we put the inode on the orphan list and modify
4503 * i_disksize immediately, so that during the subsequent flushing of
4504 * dirty pages and freeing of disk blocks, we can guarantee that any
4505 * commit will leave the blocks being flushed in an unused state on
4506 * disk. (On recovery, the inode will get truncated and the blocks will
4507 * be freed, so we have a strong guarantee that no future commit will
4508 * leave these blocks visible to the user.)
4510 * Another thing we have to assure is that if we are in ordered mode
4511 * and inode is still attached to the committing transaction, we must
4512 * we start writeout of all the dirty pages which are being truncated.
4513 * This way we are sure that all the data written in the previous
4514 * transaction are already on disk (truncate waits for pages under
4517 * Called with inode->i_mutex down.
4519 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4521 struct inode *inode = dentry->d_inode;
4524 const unsigned int ia_valid = attr->ia_valid;
4526 error = inode_change_ok(inode, attr);
4530 if (is_quota_modification(inode, attr))
4531 dquot_initialize(inode);
4532 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4533 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4536 /* (user+group)*(old+new) structure, inode write (sb,
4537 * inode block, ? - but truncate inode update has it) */
4538 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4539 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4540 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4541 if (IS_ERR(handle)) {
4542 error = PTR_ERR(handle);
4545 error = dquot_transfer(inode, attr);
4547 ext4_journal_stop(handle);
4550 /* Update corresponding info in inode so that everything is in
4551 * one transaction */
4552 if (attr->ia_valid & ATTR_UID)
4553 inode->i_uid = attr->ia_uid;
4554 if (attr->ia_valid & ATTR_GID)
4555 inode->i_gid = attr->ia_gid;
4556 error = ext4_mark_inode_dirty(handle, inode);
4557 ext4_journal_stop(handle);
4560 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4563 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4564 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4566 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4570 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4571 inode_inc_iversion(inode);
4573 if (S_ISREG(inode->i_mode) &&
4574 (attr->ia_size < inode->i_size)) {
4575 if (ext4_should_order_data(inode)) {
4576 error = ext4_begin_ordered_truncate(inode,
4581 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4582 if (IS_ERR(handle)) {
4583 error = PTR_ERR(handle);
4586 if (ext4_handle_valid(handle)) {
4587 error = ext4_orphan_add(handle, inode);
4590 down_write(&EXT4_I(inode)->i_data_sem);
4591 EXT4_I(inode)->i_disksize = attr->ia_size;
4592 rc = ext4_mark_inode_dirty(handle, inode);
4596 * We have to update i_size under i_data_sem together
4597 * with i_disksize to avoid races with writeback code
4598 * running ext4_wb_update_i_disksize().
4601 i_size_write(inode, attr->ia_size);
4602 up_write(&EXT4_I(inode)->i_data_sem);
4603 ext4_journal_stop(handle);
4605 ext4_orphan_del(NULL, inode);
4609 loff_t oldsize = inode->i_size;
4611 i_size_write(inode, attr->ia_size);
4612 pagecache_isize_extended(inode, oldsize, inode->i_size);
4616 * Blocks are going to be removed from the inode. Wait
4617 * for dio in flight. Temporarily disable
4618 * dioread_nolock to prevent livelock.
4621 if (!ext4_should_journal_data(inode)) {
4622 ext4_inode_block_unlocked_dio(inode);
4623 inode_dio_wait(inode);
4624 ext4_inode_resume_unlocked_dio(inode);
4626 ext4_wait_for_tail_page_commit(inode);
4629 * Truncate pagecache after we've waited for commit
4630 * in data=journal mode to make pages freeable.
4632 truncate_pagecache(inode, inode->i_size);
4635 * We want to call ext4_truncate() even if attr->ia_size ==
4636 * inode->i_size for cases like truncation of fallocated space
4638 if (attr->ia_valid & ATTR_SIZE)
4639 ext4_truncate(inode);
4642 setattr_copy(inode, attr);
4643 mark_inode_dirty(inode);
4647 * If the call to ext4_truncate failed to get a transaction handle at
4648 * all, we need to clean up the in-core orphan list manually.
4650 if (orphan && inode->i_nlink)
4651 ext4_orphan_del(NULL, inode);
4653 if (!rc && (ia_valid & ATTR_MODE))
4654 rc = posix_acl_chmod(inode, inode->i_mode);
4657 ext4_std_error(inode->i_sb, error);
4663 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4666 struct inode *inode;
4667 unsigned long long delalloc_blocks;
4669 inode = dentry->d_inode;
4670 generic_fillattr(inode, stat);
4673 * If there is inline data in the inode, the inode will normally not
4674 * have data blocks allocated (it may have an external xattr block).
4675 * Report at least one sector for such files, so tools like tar, rsync,
4676 * others doen't incorrectly think the file is completely sparse.
4678 if (unlikely(ext4_has_inline_data(inode)))
4679 stat->blocks += (stat->size + 511) >> 9;
4682 * We can't update i_blocks if the block allocation is delayed
4683 * otherwise in the case of system crash before the real block
4684 * allocation is done, we will have i_blocks inconsistent with
4685 * on-disk file blocks.
4686 * We always keep i_blocks updated together with real
4687 * allocation. But to not confuse with user, stat
4688 * will return the blocks that include the delayed allocation
4689 * blocks for this file.
4691 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4692 EXT4_I(inode)->i_reserved_data_blocks);
4693 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
4697 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4700 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4701 return ext4_ind_trans_blocks(inode, lblocks);
4702 return ext4_ext_index_trans_blocks(inode, pextents);
4706 * Account for index blocks, block groups bitmaps and block group
4707 * descriptor blocks if modify datablocks and index blocks
4708 * worse case, the indexs blocks spread over different block groups
4710 * If datablocks are discontiguous, they are possible to spread over
4711 * different block groups too. If they are contiguous, with flexbg,
4712 * they could still across block group boundary.
4714 * Also account for superblock, inode, quota and xattr blocks
4716 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4719 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4725 * How many index blocks need to touch to map @lblocks logical blocks
4726 * to @pextents physical extents?
4728 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
4733 * Now let's see how many group bitmaps and group descriptors need
4736 groups = idxblocks + pextents;
4738 if (groups > ngroups)
4740 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4741 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4743 /* bitmaps and block group descriptor blocks */
4744 ret += groups + gdpblocks;
4746 /* Blocks for super block, inode, quota and xattr blocks */
4747 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4753 * Calculate the total number of credits to reserve to fit
4754 * the modification of a single pages into a single transaction,
4755 * which may include multiple chunks of block allocations.
4757 * This could be called via ext4_write_begin()
4759 * We need to consider the worse case, when
4760 * one new block per extent.
4762 int ext4_writepage_trans_blocks(struct inode *inode)
4764 int bpp = ext4_journal_blocks_per_page(inode);
4767 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
4769 /* Account for data blocks for journalled mode */
4770 if (ext4_should_journal_data(inode))
4776 * Calculate the journal credits for a chunk of data modification.
4778 * This is called from DIO, fallocate or whoever calling
4779 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4781 * journal buffers for data blocks are not included here, as DIO
4782 * and fallocate do no need to journal data buffers.
4784 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4786 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4790 * The caller must have previously called ext4_reserve_inode_write().
4791 * Give this, we know that the caller already has write access to iloc->bh.
4793 int ext4_mark_iloc_dirty(handle_t *handle,
4794 struct inode *inode, struct ext4_iloc *iloc)
4798 if (IS_I_VERSION(inode))
4799 inode_inc_iversion(inode);
4801 /* the do_update_inode consumes one bh->b_count */
4804 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4805 err = ext4_do_update_inode(handle, inode, iloc);
4811 * On success, We end up with an outstanding reference count against
4812 * iloc->bh. This _must_ be cleaned up later.
4816 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4817 struct ext4_iloc *iloc)
4821 err = ext4_get_inode_loc(inode, iloc);
4823 BUFFER_TRACE(iloc->bh, "get_write_access");
4824 err = ext4_journal_get_write_access(handle, iloc->bh);
4830 ext4_std_error(inode->i_sb, err);
4835 * Expand an inode by new_extra_isize bytes.
4836 * Returns 0 on success or negative error number on failure.
4838 static int ext4_expand_extra_isize(struct inode *inode,
4839 unsigned int new_extra_isize,
4840 struct ext4_iloc iloc,
4843 struct ext4_inode *raw_inode;
4844 struct ext4_xattr_ibody_header *header;
4846 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4849 raw_inode = ext4_raw_inode(&iloc);
4851 header = IHDR(inode, raw_inode);
4853 /* No extended attributes present */
4854 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4855 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4856 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4858 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4862 /* try to expand with EAs present */
4863 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4868 * What we do here is to mark the in-core inode as clean with respect to inode
4869 * dirtiness (it may still be data-dirty).
4870 * This means that the in-core inode may be reaped by prune_icache
4871 * without having to perform any I/O. This is a very good thing,
4872 * because *any* task may call prune_icache - even ones which
4873 * have a transaction open against a different journal.
4875 * Is this cheating? Not really. Sure, we haven't written the
4876 * inode out, but prune_icache isn't a user-visible syncing function.
4877 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4878 * we start and wait on commits.
4880 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4882 struct ext4_iloc iloc;
4883 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4884 static unsigned int mnt_count;
4888 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4889 err = ext4_reserve_inode_write(handle, inode, &iloc);
4890 if (ext4_handle_valid(handle) &&
4891 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4892 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4894 * We need extra buffer credits since we may write into EA block
4895 * with this same handle. If journal_extend fails, then it will
4896 * only result in a minor loss of functionality for that inode.
4897 * If this is felt to be critical, then e2fsck should be run to
4898 * force a large enough s_min_extra_isize.
4900 if ((jbd2_journal_extend(handle,
4901 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4902 ret = ext4_expand_extra_isize(inode,
4903 sbi->s_want_extra_isize,
4906 ext4_set_inode_state(inode,
4907 EXT4_STATE_NO_EXPAND);
4909 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4910 ext4_warning(inode->i_sb,
4911 "Unable to expand inode %lu. Delete"
4912 " some EAs or run e2fsck.",
4915 le16_to_cpu(sbi->s_es->s_mnt_count);
4921 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4926 * ext4_dirty_inode() is called from __mark_inode_dirty()
4928 * We're really interested in the case where a file is being extended.
4929 * i_size has been changed by generic_commit_write() and we thus need
4930 * to include the updated inode in the current transaction.
4932 * Also, dquot_alloc_block() will always dirty the inode when blocks
4933 * are allocated to the file.
4935 * If the inode is marked synchronous, we don't honour that here - doing
4936 * so would cause a commit on atime updates, which we don't bother doing.
4937 * We handle synchronous inodes at the highest possible level.
4939 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
4940 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
4941 * to copy into the on-disk inode structure are the timestamp files.
4943 void ext4_dirty_inode(struct inode *inode, int flags)
4947 if (flags == I_DIRTY_TIME)
4949 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
4953 ext4_mark_inode_dirty(handle, inode);
4955 ext4_journal_stop(handle);
4962 * Bind an inode's backing buffer_head into this transaction, to prevent
4963 * it from being flushed to disk early. Unlike
4964 * ext4_reserve_inode_write, this leaves behind no bh reference and
4965 * returns no iloc structure, so the caller needs to repeat the iloc
4966 * lookup to mark the inode dirty later.
4968 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4970 struct ext4_iloc iloc;
4974 err = ext4_get_inode_loc(inode, &iloc);
4976 BUFFER_TRACE(iloc.bh, "get_write_access");
4977 err = jbd2_journal_get_write_access(handle, iloc.bh);
4979 err = ext4_handle_dirty_metadata(handle,
4985 ext4_std_error(inode->i_sb, err);
4990 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4997 * We have to be very careful here: changing a data block's
4998 * journaling status dynamically is dangerous. If we write a
4999 * data block to the journal, change the status and then delete
5000 * that block, we risk forgetting to revoke the old log record
5001 * from the journal and so a subsequent replay can corrupt data.
5002 * So, first we make sure that the journal is empty and that
5003 * nobody is changing anything.
5006 journal = EXT4_JOURNAL(inode);
5009 if (is_journal_aborted(journal))
5011 /* We have to allocate physical blocks for delalloc blocks
5012 * before flushing journal. otherwise delalloc blocks can not
5013 * be allocated any more. even more truncate on delalloc blocks
5014 * could trigger BUG by flushing delalloc blocks in journal.
5015 * There is no delalloc block in non-journal data mode.
5017 if (val && test_opt(inode->i_sb, DELALLOC)) {
5018 err = ext4_alloc_da_blocks(inode);
5023 /* Wait for all existing dio workers */
5024 ext4_inode_block_unlocked_dio(inode);
5025 inode_dio_wait(inode);
5027 jbd2_journal_lock_updates(journal);
5030 * OK, there are no updates running now, and all cached data is
5031 * synced to disk. We are now in a completely consistent state
5032 * which doesn't have anything in the journal, and we know that
5033 * no filesystem updates are running, so it is safe to modify
5034 * the inode's in-core data-journaling state flag now.
5038 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5040 err = jbd2_journal_flush(journal);
5042 jbd2_journal_unlock_updates(journal);
5043 ext4_inode_resume_unlocked_dio(inode);
5046 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5048 ext4_set_aops(inode);
5050 jbd2_journal_unlock_updates(journal);
5051 ext4_inode_resume_unlocked_dio(inode);
5053 /* Finally we can mark the inode as dirty. */
5055 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5057 return PTR_ERR(handle);
5059 err = ext4_mark_inode_dirty(handle, inode);
5060 ext4_handle_sync(handle);
5061 ext4_journal_stop(handle);
5062 ext4_std_error(inode->i_sb, err);
5067 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5069 return !buffer_mapped(bh);
5072 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5074 struct page *page = vmf->page;
5078 struct file *file = vma->vm_file;
5079 struct inode *inode = file_inode(file);
5080 struct address_space *mapping = inode->i_mapping;
5082 get_block_t *get_block;
5085 sb_start_pagefault(inode->i_sb);
5086 file_update_time(vma->vm_file);
5087 /* Delalloc case is easy... */
5088 if (test_opt(inode->i_sb, DELALLOC) &&
5089 !ext4_should_journal_data(inode) &&
5090 !ext4_nonda_switch(inode->i_sb)) {
5092 ret = __block_page_mkwrite(vma, vmf,
5093 ext4_da_get_block_prep);
5094 } while (ret == -ENOSPC &&
5095 ext4_should_retry_alloc(inode->i_sb, &retries));
5100 size = i_size_read(inode);
5101 /* Page got truncated from under us? */
5102 if (page->mapping != mapping || page_offset(page) > size) {
5104 ret = VM_FAULT_NOPAGE;
5108 if (page->index == size >> PAGE_CACHE_SHIFT)
5109 len = size & ~PAGE_CACHE_MASK;
5111 len = PAGE_CACHE_SIZE;
5113 * Return if we have all the buffers mapped. This avoids the need to do
5114 * journal_start/journal_stop which can block and take a long time
5116 if (page_has_buffers(page)) {
5117 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5119 ext4_bh_unmapped)) {
5120 /* Wait so that we don't change page under IO */
5121 wait_for_stable_page(page);
5122 ret = VM_FAULT_LOCKED;
5127 /* OK, we need to fill the hole... */
5128 if (ext4_should_dioread_nolock(inode))
5129 get_block = ext4_get_block_write;
5131 get_block = ext4_get_block;
5133 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5134 ext4_writepage_trans_blocks(inode));
5135 if (IS_ERR(handle)) {
5136 ret = VM_FAULT_SIGBUS;
5139 ret = __block_page_mkwrite(vma, vmf, get_block);
5140 if (!ret && ext4_should_journal_data(inode)) {
5141 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5142 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5144 ret = VM_FAULT_SIGBUS;
5145 ext4_journal_stop(handle);
5148 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5150 ext4_journal_stop(handle);
5151 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5154 ret = block_page_mkwrite_return(ret);
5156 sb_end_pagefault(inode->i_sb);