2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/quotaops.h>
26 #include <linux/string.h>
27 #include <linux/buffer_head.h>
28 #include <linux/writeback.h>
29 #include <linux/pagevec.h>
30 #include <linux/mpage.h>
31 #include <linux/namei.h>
32 #include <linux/uio.h>
33 #include <linux/bio.h>
34 #include <linux/workqueue.h>
35 #include <linux/kernel.h>
36 #include <linux/printk.h>
37 #include <linux/slab.h>
38 #include <linux/bitops.h>
40 #include "ext4_jbd2.h"
45 #include <trace/events/ext4.h>
47 #define MPAGE_DA_EXTENT_TAIL 0x01
49 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
50 struct ext4_inode_info *ei)
52 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
57 csum_lo = le16_to_cpu(raw->i_checksum_lo);
58 raw->i_checksum_lo = 0;
59 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
60 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
61 csum_hi = le16_to_cpu(raw->i_checksum_hi);
62 raw->i_checksum_hi = 0;
65 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
66 EXT4_INODE_SIZE(inode->i_sb));
68 raw->i_checksum_lo = cpu_to_le16(csum_lo);
69 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
70 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
71 raw->i_checksum_hi = cpu_to_le16(csum_hi);
76 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
77 struct ext4_inode_info *ei)
79 __u32 provided, calculated;
81 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
82 cpu_to_le32(EXT4_OS_LINUX) ||
83 !ext4_has_metadata_csum(inode->i_sb))
86 provided = le16_to_cpu(raw->i_checksum_lo);
87 calculated = ext4_inode_csum(inode, raw, ei);
88 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
89 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
90 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
94 return provided == calculated;
97 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
98 struct ext4_inode_info *ei)
102 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
103 cpu_to_le32(EXT4_OS_LINUX) ||
104 !ext4_has_metadata_csum(inode->i_sb))
107 csum = ext4_inode_csum(inode, raw, ei);
108 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
109 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
110 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
111 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
114 static inline int ext4_begin_ordered_truncate(struct inode *inode,
117 trace_ext4_begin_ordered_truncate(inode, new_size);
119 * If jinode is zero, then we never opened the file for
120 * writing, so there's no need to call
121 * jbd2_journal_begin_ordered_truncate() since there's no
122 * outstanding writes we need to flush.
124 if (!EXT4_I(inode)->jinode)
126 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
127 EXT4_I(inode)->jinode,
131 static void ext4_invalidatepage(struct page *page, unsigned int offset,
132 unsigned int length);
133 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
134 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
135 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
139 * Test whether an inode is a fast symlink.
141 int ext4_inode_is_fast_symlink(struct inode *inode)
143 int ea_blocks = EXT4_I(inode)->i_file_acl ?
144 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
146 if (ext4_has_inline_data(inode))
149 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
153 * Restart the transaction associated with *handle. This does a commit,
154 * so before we call here everything must be consistently dirtied against
157 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
163 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
164 * moment, get_block can be called only for blocks inside i_size since
165 * page cache has been already dropped and writes are blocked by
166 * i_mutex. So we can safely drop the i_data_sem here.
168 BUG_ON(EXT4_JOURNAL(inode) == NULL);
169 jbd_debug(2, "restarting handle %p\n", handle);
170 up_write(&EXT4_I(inode)->i_data_sem);
171 ret = ext4_journal_restart(handle, nblocks);
172 down_write(&EXT4_I(inode)->i_data_sem);
173 ext4_discard_preallocations(inode);
179 * Called at the last iput() if i_nlink is zero.
181 void ext4_evict_inode(struct inode *inode)
186 trace_ext4_evict_inode(inode);
188 if (inode->i_nlink) {
190 * When journalling data dirty buffers are tracked only in the
191 * journal. So although mm thinks everything is clean and
192 * ready for reaping the inode might still have some pages to
193 * write in the running transaction or waiting to be
194 * checkpointed. Thus calling jbd2_journal_invalidatepage()
195 * (via truncate_inode_pages()) to discard these buffers can
196 * cause data loss. Also even if we did not discard these
197 * buffers, we would have no way to find them after the inode
198 * is reaped and thus user could see stale data if he tries to
199 * read them before the transaction is checkpointed. So be
200 * careful and force everything to disk here... We use
201 * ei->i_datasync_tid to store the newest transaction
202 * containing inode's data.
204 * Note that directories do not have this problem because they
205 * don't use page cache.
207 if (ext4_should_journal_data(inode) &&
208 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
209 inode->i_ino != EXT4_JOURNAL_INO) {
210 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
211 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
213 jbd2_complete_transaction(journal, commit_tid);
214 filemap_write_and_wait(&inode->i_data);
216 truncate_inode_pages_final(&inode->i_data);
218 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
222 if (is_bad_inode(inode))
224 dquot_initialize(inode);
226 if (ext4_should_order_data(inode))
227 ext4_begin_ordered_truncate(inode, 0);
228 truncate_inode_pages_final(&inode->i_data);
230 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
233 * Protect us against freezing - iput() caller didn't have to have any
234 * protection against it
236 sb_start_intwrite(inode->i_sb);
237 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
238 ext4_blocks_for_truncate(inode)+3);
239 if (IS_ERR(handle)) {
240 ext4_std_error(inode->i_sb, PTR_ERR(handle));
242 * If we're going to skip the normal cleanup, we still need to
243 * make sure that the in-core orphan linked list is properly
246 ext4_orphan_del(NULL, inode);
247 sb_end_intwrite(inode->i_sb);
252 ext4_handle_sync(handle);
254 err = ext4_mark_inode_dirty(handle, inode);
256 ext4_warning(inode->i_sb,
257 "couldn't mark inode dirty (err %d)", err);
261 ext4_truncate(inode);
264 * ext4_ext_truncate() doesn't reserve any slop when it
265 * restarts journal transactions; therefore there may not be
266 * enough credits left in the handle to remove the inode from
267 * the orphan list and set the dtime field.
269 if (!ext4_handle_has_enough_credits(handle, 3)) {
270 err = ext4_journal_extend(handle, 3);
272 err = ext4_journal_restart(handle, 3);
274 ext4_warning(inode->i_sb,
275 "couldn't extend journal (err %d)", err);
277 ext4_journal_stop(handle);
278 ext4_orphan_del(NULL, inode);
279 sb_end_intwrite(inode->i_sb);
285 * Kill off the orphan record which ext4_truncate created.
286 * AKPM: I think this can be inside the above `if'.
287 * Note that ext4_orphan_del() has to be able to cope with the
288 * deletion of a non-existent orphan - this is because we don't
289 * know if ext4_truncate() actually created an orphan record.
290 * (Well, we could do this if we need to, but heck - it works)
292 ext4_orphan_del(handle, inode);
293 EXT4_I(inode)->i_dtime = get_seconds();
296 * One subtle ordering requirement: if anything has gone wrong
297 * (transaction abort, IO errors, whatever), then we can still
298 * do these next steps (the fs will already have been marked as
299 * having errors), but we can't free the inode if the mark_dirty
302 if (ext4_mark_inode_dirty(handle, inode))
303 /* If that failed, just do the required in-core inode clear. */
304 ext4_clear_inode(inode);
306 ext4_free_inode(handle, inode);
307 ext4_journal_stop(handle);
308 sb_end_intwrite(inode->i_sb);
311 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
315 qsize_t *ext4_get_reserved_space(struct inode *inode)
317 return &EXT4_I(inode)->i_reserved_quota;
322 * Called with i_data_sem down, which is important since we can call
323 * ext4_discard_preallocations() from here.
325 void ext4_da_update_reserve_space(struct inode *inode,
326 int used, int quota_claim)
328 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
329 struct ext4_inode_info *ei = EXT4_I(inode);
331 spin_lock(&ei->i_block_reservation_lock);
332 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
333 if (unlikely(used > ei->i_reserved_data_blocks)) {
334 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
335 "with only %d reserved data blocks",
336 __func__, inode->i_ino, used,
337 ei->i_reserved_data_blocks);
339 used = ei->i_reserved_data_blocks;
342 /* Update per-inode reservations */
343 ei->i_reserved_data_blocks -= used;
344 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
346 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
348 /* Update quota subsystem for data blocks */
350 dquot_claim_block(inode, EXT4_C2B(sbi, used));
353 * We did fallocate with an offset that is already delayed
354 * allocated. So on delayed allocated writeback we should
355 * not re-claim the quota for fallocated blocks.
357 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
361 * If we have done all the pending block allocations and if
362 * there aren't any writers on the inode, we can discard the
363 * inode's preallocations.
365 if ((ei->i_reserved_data_blocks == 0) &&
366 (atomic_read(&inode->i_writecount) == 0))
367 ext4_discard_preallocations(inode);
370 static int __check_block_validity(struct inode *inode, const char *func,
372 struct ext4_map_blocks *map)
374 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
376 ext4_error_inode(inode, func, line, map->m_pblk,
377 "lblock %lu mapped to illegal pblock "
378 "(length %d)", (unsigned long) map->m_lblk,
385 #define check_block_validity(inode, map) \
386 __check_block_validity((inode), __func__, __LINE__, (map))
388 #ifdef ES_AGGRESSIVE_TEST
389 static void ext4_map_blocks_es_recheck(handle_t *handle,
391 struct ext4_map_blocks *es_map,
392 struct ext4_map_blocks *map,
399 * There is a race window that the result is not the same.
400 * e.g. xfstests #223 when dioread_nolock enables. The reason
401 * is that we lookup a block mapping in extent status tree with
402 * out taking i_data_sem. So at the time the unwritten extent
403 * could be converted.
405 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
406 down_read(&EXT4_I(inode)->i_data_sem);
407 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
408 retval = ext4_ext_map_blocks(handle, inode, map, flags &
409 EXT4_GET_BLOCKS_KEEP_SIZE);
411 retval = ext4_ind_map_blocks(handle, inode, map, flags &
412 EXT4_GET_BLOCKS_KEEP_SIZE);
414 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
415 up_read((&EXT4_I(inode)->i_data_sem));
418 * We don't check m_len because extent will be collpased in status
419 * tree. So the m_len might not equal.
421 if (es_map->m_lblk != map->m_lblk ||
422 es_map->m_flags != map->m_flags ||
423 es_map->m_pblk != map->m_pblk) {
424 printk("ES cache assertion failed for inode: %lu "
425 "es_cached ex [%d/%d/%llu/%x] != "
426 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
427 inode->i_ino, es_map->m_lblk, es_map->m_len,
428 es_map->m_pblk, es_map->m_flags, map->m_lblk,
429 map->m_len, map->m_pblk, map->m_flags,
433 #endif /* ES_AGGRESSIVE_TEST */
436 * The ext4_map_blocks() function tries to look up the requested blocks,
437 * and returns if the blocks are already mapped.
439 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
440 * and store the allocated blocks in the result buffer head and mark it
443 * If file type is extents based, it will call ext4_ext_map_blocks(),
444 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
447 * On success, it returns the number of blocks being mapped or allocated.
448 * if create==0 and the blocks are pre-allocated and unwritten block,
449 * the result buffer head is unmapped. If the create ==1, it will make sure
450 * the buffer head is mapped.
452 * It returns 0 if plain look up failed (blocks have not been allocated), in
453 * that case, buffer head is unmapped
455 * It returns the error in case of allocation failure.
457 int ext4_map_blocks(handle_t *handle, struct inode *inode,
458 struct ext4_map_blocks *map, int flags)
460 struct extent_status es;
463 #ifdef ES_AGGRESSIVE_TEST
464 struct ext4_map_blocks orig_map;
466 memcpy(&orig_map, map, sizeof(*map));
470 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
471 "logical block %lu\n", inode->i_ino, flags, map->m_len,
472 (unsigned long) map->m_lblk);
475 * ext4_map_blocks returns an int, and m_len is an unsigned int
477 if (unlikely(map->m_len > INT_MAX))
478 map->m_len = INT_MAX;
480 /* We can handle the block number less than EXT_MAX_BLOCKS */
481 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
484 /* Lookup extent status tree firstly */
485 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
486 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
487 map->m_pblk = ext4_es_pblock(&es) +
488 map->m_lblk - es.es_lblk;
489 map->m_flags |= ext4_es_is_written(&es) ?
490 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
491 retval = es.es_len - (map->m_lblk - es.es_lblk);
492 if (retval > map->m_len)
495 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
500 #ifdef ES_AGGRESSIVE_TEST
501 ext4_map_blocks_es_recheck(handle, inode, map,
508 * Try to see if we can get the block without requesting a new
511 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
512 down_read(&EXT4_I(inode)->i_data_sem);
513 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
514 retval = ext4_ext_map_blocks(handle, inode, map, flags &
515 EXT4_GET_BLOCKS_KEEP_SIZE);
517 retval = ext4_ind_map_blocks(handle, inode, map, flags &
518 EXT4_GET_BLOCKS_KEEP_SIZE);
523 if (unlikely(retval != map->m_len)) {
524 ext4_warning(inode->i_sb,
525 "ES len assertion failed for inode "
526 "%lu: retval %d != map->m_len %d",
527 inode->i_ino, retval, map->m_len);
531 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
532 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
533 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
534 !(status & EXTENT_STATUS_WRITTEN) &&
535 ext4_find_delalloc_range(inode, map->m_lblk,
536 map->m_lblk + map->m_len - 1))
537 status |= EXTENT_STATUS_DELAYED;
538 ret = ext4_es_insert_extent(inode, map->m_lblk,
539 map->m_len, map->m_pblk, status);
543 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
544 up_read((&EXT4_I(inode)->i_data_sem));
547 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
548 ret = check_block_validity(inode, map);
553 /* If it is only a block(s) look up */
554 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
558 * Returns if the blocks have already allocated
560 * Note that if blocks have been preallocated
561 * ext4_ext_get_block() returns the create = 0
562 * with buffer head unmapped.
564 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
566 * If we need to convert extent to unwritten
567 * we continue and do the actual work in
568 * ext4_ext_map_blocks()
570 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
574 * Here we clear m_flags because after allocating an new extent,
575 * it will be set again.
577 map->m_flags &= ~EXT4_MAP_FLAGS;
580 * New blocks allocate and/or writing to unwritten extent
581 * will possibly result in updating i_data, so we take
582 * the write lock of i_data_sem, and call get_block()
583 * with create == 1 flag.
585 down_write(&EXT4_I(inode)->i_data_sem);
588 * We need to check for EXT4 here because migrate
589 * could have changed the inode type in between
591 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
592 retval = ext4_ext_map_blocks(handle, inode, map, flags);
594 retval = ext4_ind_map_blocks(handle, inode, map, flags);
596 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
598 * We allocated new blocks which will result in
599 * i_data's format changing. Force the migrate
600 * to fail by clearing migrate flags
602 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
606 * Update reserved blocks/metadata blocks after successful
607 * block allocation which had been deferred till now. We don't
608 * support fallocate for non extent files. So we can update
609 * reserve space here.
612 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
613 ext4_da_update_reserve_space(inode, retval, 1);
619 if (unlikely(retval != map->m_len)) {
620 ext4_warning(inode->i_sb,
621 "ES len assertion failed for inode "
622 "%lu: retval %d != map->m_len %d",
623 inode->i_ino, retval, map->m_len);
628 * If the extent has been zeroed out, we don't need to update
629 * extent status tree.
631 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
632 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
633 if (ext4_es_is_written(&es))
636 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
637 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
638 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
639 !(status & EXTENT_STATUS_WRITTEN) &&
640 ext4_find_delalloc_range(inode, map->m_lblk,
641 map->m_lblk + map->m_len - 1))
642 status |= EXTENT_STATUS_DELAYED;
643 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
644 map->m_pblk, status);
650 up_write((&EXT4_I(inode)->i_data_sem));
651 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
652 ret = check_block_validity(inode, map);
659 static void ext4_end_io_unwritten(struct buffer_head *bh, int uptodate)
661 struct inode *inode = bh->b_assoc_map->host;
662 /* XXX: breaks on 32-bit > 16GB. Is that even supported? */
663 loff_t offset = (loff_t)(uintptr_t)bh->b_private << inode->i_blkbits;
667 WARN_ON(!buffer_unwritten(bh));
668 err = ext4_convert_unwritten_extents(NULL, inode, offset, bh->b_size);
671 /* Maximum number of blocks we map for direct IO at once. */
672 #define DIO_MAX_BLOCKS 4096
674 static int _ext4_get_block(struct inode *inode, sector_t iblock,
675 struct buffer_head *bh, int flags)
677 handle_t *handle = ext4_journal_current_handle();
678 struct ext4_map_blocks map;
679 int ret = 0, started = 0;
682 if (ext4_has_inline_data(inode))
686 map.m_len = bh->b_size >> inode->i_blkbits;
688 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
689 /* Direct IO write... */
690 if (map.m_len > DIO_MAX_BLOCKS)
691 map.m_len = DIO_MAX_BLOCKS;
692 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
693 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
695 if (IS_ERR(handle)) {
696 ret = PTR_ERR(handle);
702 ret = ext4_map_blocks(handle, inode, &map, flags);
704 ext4_io_end_t *io_end = ext4_inode_aio(inode);
706 map_bh(bh, inode->i_sb, map.m_pblk);
707 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
708 if (IS_DAX(inode) && buffer_unwritten(bh) && !io_end) {
709 bh->b_assoc_map = inode->i_mapping;
710 bh->b_private = (void *)(unsigned long)iblock;
711 bh->b_end_io = ext4_end_io_unwritten;
713 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
714 set_buffer_defer_completion(bh);
715 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
719 ext4_journal_stop(handle);
723 int ext4_get_block(struct inode *inode, sector_t iblock,
724 struct buffer_head *bh, int create)
726 return _ext4_get_block(inode, iblock, bh,
727 create ? EXT4_GET_BLOCKS_CREATE : 0);
731 * `handle' can be NULL if create is zero
733 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
734 ext4_lblk_t block, int map_flags)
736 struct ext4_map_blocks map;
737 struct buffer_head *bh;
738 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
741 J_ASSERT(handle != NULL || create == 0);
745 err = ext4_map_blocks(handle, inode, &map, map_flags);
748 return create ? ERR_PTR(-ENOSPC) : NULL;
752 bh = sb_getblk(inode->i_sb, map.m_pblk);
754 return ERR_PTR(-ENOMEM);
755 if (map.m_flags & EXT4_MAP_NEW) {
756 J_ASSERT(create != 0);
757 J_ASSERT(handle != NULL);
760 * Now that we do not always journal data, we should
761 * keep in mind whether this should always journal the
762 * new buffer as metadata. For now, regular file
763 * writes use ext4_get_block instead, so it's not a
767 BUFFER_TRACE(bh, "call get_create_access");
768 err = ext4_journal_get_create_access(handle, bh);
773 if (!buffer_uptodate(bh)) {
774 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
775 set_buffer_uptodate(bh);
778 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
779 err = ext4_handle_dirty_metadata(handle, inode, bh);
783 BUFFER_TRACE(bh, "not a new buffer");
790 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
791 ext4_lblk_t block, int map_flags)
793 struct buffer_head *bh;
795 bh = ext4_getblk(handle, inode, block, map_flags);
798 if (!bh || buffer_uptodate(bh))
800 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
802 if (buffer_uptodate(bh))
805 return ERR_PTR(-EIO);
808 int ext4_walk_page_buffers(handle_t *handle,
809 struct buffer_head *head,
813 int (*fn)(handle_t *handle,
814 struct buffer_head *bh))
816 struct buffer_head *bh;
817 unsigned block_start, block_end;
818 unsigned blocksize = head->b_size;
820 struct buffer_head *next;
822 for (bh = head, block_start = 0;
823 ret == 0 && (bh != head || !block_start);
824 block_start = block_end, bh = next) {
825 next = bh->b_this_page;
826 block_end = block_start + blocksize;
827 if (block_end <= from || block_start >= to) {
828 if (partial && !buffer_uptodate(bh))
832 err = (*fn)(handle, bh);
840 * To preserve ordering, it is essential that the hole instantiation and
841 * the data write be encapsulated in a single transaction. We cannot
842 * close off a transaction and start a new one between the ext4_get_block()
843 * and the commit_write(). So doing the jbd2_journal_start at the start of
844 * prepare_write() is the right place.
846 * Also, this function can nest inside ext4_writepage(). In that case, we
847 * *know* that ext4_writepage() has generated enough buffer credits to do the
848 * whole page. So we won't block on the journal in that case, which is good,
849 * because the caller may be PF_MEMALLOC.
851 * By accident, ext4 can be reentered when a transaction is open via
852 * quota file writes. If we were to commit the transaction while thus
853 * reentered, there can be a deadlock - we would be holding a quota
854 * lock, and the commit would never complete if another thread had a
855 * transaction open and was blocking on the quota lock - a ranking
858 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
859 * will _not_ run commit under these circumstances because handle->h_ref
860 * is elevated. We'll still have enough credits for the tiny quotafile
863 int do_journal_get_write_access(handle_t *handle,
864 struct buffer_head *bh)
866 int dirty = buffer_dirty(bh);
869 if (!buffer_mapped(bh) || buffer_freed(bh))
872 * __block_write_begin() could have dirtied some buffers. Clean
873 * the dirty bit as jbd2_journal_get_write_access() could complain
874 * otherwise about fs integrity issues. Setting of the dirty bit
875 * by __block_write_begin() isn't a real problem here as we clear
876 * the bit before releasing a page lock and thus writeback cannot
877 * ever write the buffer.
880 clear_buffer_dirty(bh);
881 BUFFER_TRACE(bh, "get write access");
882 ret = ext4_journal_get_write_access(handle, bh);
884 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
888 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
889 struct buffer_head *bh_result, int create);
891 #ifdef CONFIG_EXT4_FS_ENCRYPTION
892 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
893 get_block_t *get_block)
895 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
896 unsigned to = from + len;
897 struct inode *inode = page->mapping->host;
898 unsigned block_start, block_end;
901 unsigned blocksize = inode->i_sb->s_blocksize;
903 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
904 bool decrypt = false;
906 BUG_ON(!PageLocked(page));
907 BUG_ON(from > PAGE_CACHE_SIZE);
908 BUG_ON(to > PAGE_CACHE_SIZE);
911 if (!page_has_buffers(page))
912 create_empty_buffers(page, blocksize, 0);
913 head = page_buffers(page);
914 bbits = ilog2(blocksize);
915 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
917 for (bh = head, block_start = 0; bh != head || !block_start;
918 block++, block_start = block_end, bh = bh->b_this_page) {
919 block_end = block_start + blocksize;
920 if (block_end <= from || block_start >= to) {
921 if (PageUptodate(page)) {
922 if (!buffer_uptodate(bh))
923 set_buffer_uptodate(bh);
928 clear_buffer_new(bh);
929 if (!buffer_mapped(bh)) {
930 WARN_ON(bh->b_size != blocksize);
931 err = get_block(inode, block, bh, 1);
934 if (buffer_new(bh)) {
935 unmap_underlying_metadata(bh->b_bdev,
937 if (PageUptodate(page)) {
938 clear_buffer_new(bh);
939 set_buffer_uptodate(bh);
940 mark_buffer_dirty(bh);
943 if (block_end > to || block_start < from)
944 zero_user_segments(page, to, block_end,
949 if (PageUptodate(page)) {
950 if (!buffer_uptodate(bh))
951 set_buffer_uptodate(bh);
954 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
955 !buffer_unwritten(bh) &&
956 (block_start < from || block_end > to)) {
957 ll_rw_block(READ, 1, &bh);
959 decrypt = ext4_encrypted_inode(inode) &&
960 S_ISREG(inode->i_mode);
964 * If we issued read requests, let them complete.
966 while (wait_bh > wait) {
967 wait_on_buffer(*--wait_bh);
968 if (!buffer_uptodate(*wait_bh))
972 page_zero_new_buffers(page, from, to);
974 err = ext4_decrypt_one(inode, page);
979 static int ext4_write_begin(struct file *file, struct address_space *mapping,
980 loff_t pos, unsigned len, unsigned flags,
981 struct page **pagep, void **fsdata)
983 struct inode *inode = mapping->host;
984 int ret, needed_blocks;
991 trace_ext4_write_begin(inode, pos, len, flags);
993 * Reserve one block more for addition to orphan list in case
994 * we allocate blocks but write fails for some reason
996 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
997 index = pos >> PAGE_CACHE_SHIFT;
998 from = pos & (PAGE_CACHE_SIZE - 1);
1001 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1002 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1011 * grab_cache_page_write_begin() can take a long time if the
1012 * system is thrashing due to memory pressure, or if the page
1013 * is being written back. So grab it first before we start
1014 * the transaction handle. This also allows us to allocate
1015 * the page (if needed) without using GFP_NOFS.
1018 page = grab_cache_page_write_begin(mapping, index, flags);
1024 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1025 if (IS_ERR(handle)) {
1026 page_cache_release(page);
1027 return PTR_ERR(handle);
1031 if (page->mapping != mapping) {
1032 /* The page got truncated from under us */
1034 page_cache_release(page);
1035 ext4_journal_stop(handle);
1038 /* In case writeback began while the page was unlocked */
1039 wait_for_stable_page(page);
1041 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1042 if (ext4_should_dioread_nolock(inode))
1043 ret = ext4_block_write_begin(page, pos, len,
1044 ext4_get_block_write);
1046 ret = ext4_block_write_begin(page, pos, len,
1049 if (ext4_should_dioread_nolock(inode))
1050 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1052 ret = __block_write_begin(page, pos, len, ext4_get_block);
1054 if (!ret && ext4_should_journal_data(inode)) {
1055 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1057 do_journal_get_write_access);
1063 * __block_write_begin may have instantiated a few blocks
1064 * outside i_size. Trim these off again. Don't need
1065 * i_size_read because we hold i_mutex.
1067 * Add inode to orphan list in case we crash before
1070 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1071 ext4_orphan_add(handle, inode);
1073 ext4_journal_stop(handle);
1074 if (pos + len > inode->i_size) {
1075 ext4_truncate_failed_write(inode);
1077 * If truncate failed early the inode might
1078 * still be on the orphan list; we need to
1079 * make sure the inode is removed from the
1080 * orphan list in that case.
1083 ext4_orphan_del(NULL, inode);
1086 if (ret == -ENOSPC &&
1087 ext4_should_retry_alloc(inode->i_sb, &retries))
1089 page_cache_release(page);
1096 /* For write_end() in data=journal mode */
1097 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1100 if (!buffer_mapped(bh) || buffer_freed(bh))
1102 set_buffer_uptodate(bh);
1103 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1104 clear_buffer_meta(bh);
1105 clear_buffer_prio(bh);
1110 * We need to pick up the new inode size which generic_commit_write gave us
1111 * `file' can be NULL - eg, when called from page_symlink().
1113 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1114 * buffers are managed internally.
1116 static int ext4_write_end(struct file *file,
1117 struct address_space *mapping,
1118 loff_t pos, unsigned len, unsigned copied,
1119 struct page *page, void *fsdata)
1121 handle_t *handle = ext4_journal_current_handle();
1122 struct inode *inode = mapping->host;
1123 loff_t old_size = inode->i_size;
1125 int i_size_changed = 0;
1127 trace_ext4_write_end(inode, pos, len, copied);
1128 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1129 ret = ext4_jbd2_file_inode(handle, inode);
1132 page_cache_release(page);
1137 if (ext4_has_inline_data(inode)) {
1138 ret = ext4_write_inline_data_end(inode, pos, len,
1144 copied = block_write_end(file, mapping, pos,
1145 len, copied, page, fsdata);
1147 * it's important to update i_size while still holding page lock:
1148 * page writeout could otherwise come in and zero beyond i_size.
1150 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1152 page_cache_release(page);
1155 pagecache_isize_extended(inode, old_size, pos);
1157 * Don't mark the inode dirty under page lock. First, it unnecessarily
1158 * makes the holding time of page lock longer. Second, it forces lock
1159 * ordering of page lock and transaction start for journaling
1163 ext4_mark_inode_dirty(handle, inode);
1165 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1166 /* if we have allocated more blocks and copied
1167 * less. We will have blocks allocated outside
1168 * inode->i_size. So truncate them
1170 ext4_orphan_add(handle, inode);
1172 ret2 = ext4_journal_stop(handle);
1176 if (pos + len > inode->i_size) {
1177 ext4_truncate_failed_write(inode);
1179 * If truncate failed early the inode might still be
1180 * on the orphan list; we need to make sure the inode
1181 * is removed from the orphan list in that case.
1184 ext4_orphan_del(NULL, inode);
1187 return ret ? ret : copied;
1190 static int ext4_journalled_write_end(struct file *file,
1191 struct address_space *mapping,
1192 loff_t pos, unsigned len, unsigned copied,
1193 struct page *page, void *fsdata)
1195 handle_t *handle = ext4_journal_current_handle();
1196 struct inode *inode = mapping->host;
1197 loff_t old_size = inode->i_size;
1201 int size_changed = 0;
1203 trace_ext4_journalled_write_end(inode, pos, len, copied);
1204 from = pos & (PAGE_CACHE_SIZE - 1);
1207 BUG_ON(!ext4_handle_valid(handle));
1209 if (ext4_has_inline_data(inode))
1210 copied = ext4_write_inline_data_end(inode, pos, len,
1214 if (!PageUptodate(page))
1216 page_zero_new_buffers(page, from+copied, to);
1219 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1220 to, &partial, write_end_fn);
1222 SetPageUptodate(page);
1224 size_changed = ext4_update_inode_size(inode, pos + copied);
1225 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1226 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1228 page_cache_release(page);
1231 pagecache_isize_extended(inode, old_size, pos);
1234 ret2 = ext4_mark_inode_dirty(handle, inode);
1239 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1240 /* if we have allocated more blocks and copied
1241 * less. We will have blocks allocated outside
1242 * inode->i_size. So truncate them
1244 ext4_orphan_add(handle, inode);
1246 ret2 = ext4_journal_stop(handle);
1249 if (pos + len > inode->i_size) {
1250 ext4_truncate_failed_write(inode);
1252 * If truncate failed early the inode might still be
1253 * on the orphan list; we need to make sure the inode
1254 * is removed from the orphan list in that case.
1257 ext4_orphan_del(NULL, inode);
1260 return ret ? ret : copied;
1264 * Reserve space for a single cluster
1266 static int ext4_da_reserve_space(struct inode *inode)
1268 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1269 struct ext4_inode_info *ei = EXT4_I(inode);
1273 * We will charge metadata quota at writeout time; this saves
1274 * us from metadata over-estimation, though we may go over by
1275 * a small amount in the end. Here we just reserve for data.
1277 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1281 spin_lock(&ei->i_block_reservation_lock);
1282 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1283 spin_unlock(&ei->i_block_reservation_lock);
1284 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1287 ei->i_reserved_data_blocks++;
1288 trace_ext4_da_reserve_space(inode);
1289 spin_unlock(&ei->i_block_reservation_lock);
1291 return 0; /* success */
1294 static void ext4_da_release_space(struct inode *inode, int to_free)
1296 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1297 struct ext4_inode_info *ei = EXT4_I(inode);
1300 return; /* Nothing to release, exit */
1302 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1304 trace_ext4_da_release_space(inode, to_free);
1305 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1307 * if there aren't enough reserved blocks, then the
1308 * counter is messed up somewhere. Since this
1309 * function is called from invalidate page, it's
1310 * harmless to return without any action.
1312 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1313 "ino %lu, to_free %d with only %d reserved "
1314 "data blocks", inode->i_ino, to_free,
1315 ei->i_reserved_data_blocks);
1317 to_free = ei->i_reserved_data_blocks;
1319 ei->i_reserved_data_blocks -= to_free;
1321 /* update fs dirty data blocks counter */
1322 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1324 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1326 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1329 static void ext4_da_page_release_reservation(struct page *page,
1330 unsigned int offset,
1331 unsigned int length)
1334 struct buffer_head *head, *bh;
1335 unsigned int curr_off = 0;
1336 struct inode *inode = page->mapping->host;
1337 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1338 unsigned int stop = offset + length;
1342 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1344 head = page_buffers(page);
1347 unsigned int next_off = curr_off + bh->b_size;
1349 if (next_off > stop)
1352 if ((offset <= curr_off) && (buffer_delay(bh))) {
1354 clear_buffer_delay(bh);
1356 curr_off = next_off;
1357 } while ((bh = bh->b_this_page) != head);
1360 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1361 ext4_es_remove_extent(inode, lblk, to_release);
1364 /* If we have released all the blocks belonging to a cluster, then we
1365 * need to release the reserved space for that cluster. */
1366 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1367 while (num_clusters > 0) {
1368 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1369 ((num_clusters - 1) << sbi->s_cluster_bits);
1370 if (sbi->s_cluster_ratio == 1 ||
1371 !ext4_find_delalloc_cluster(inode, lblk))
1372 ext4_da_release_space(inode, 1);
1379 * Delayed allocation stuff
1382 struct mpage_da_data {
1383 struct inode *inode;
1384 struct writeback_control *wbc;
1386 pgoff_t first_page; /* The first page to write */
1387 pgoff_t next_page; /* Current page to examine */
1388 pgoff_t last_page; /* Last page to examine */
1390 * Extent to map - this can be after first_page because that can be
1391 * fully mapped. We somewhat abuse m_flags to store whether the extent
1392 * is delalloc or unwritten.
1394 struct ext4_map_blocks map;
1395 struct ext4_io_submit io_submit; /* IO submission data */
1398 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1403 struct pagevec pvec;
1404 struct inode *inode = mpd->inode;
1405 struct address_space *mapping = inode->i_mapping;
1407 /* This is necessary when next_page == 0. */
1408 if (mpd->first_page >= mpd->next_page)
1411 index = mpd->first_page;
1412 end = mpd->next_page - 1;
1414 ext4_lblk_t start, last;
1415 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1416 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1417 ext4_es_remove_extent(inode, start, last - start + 1);
1420 pagevec_init(&pvec, 0);
1421 while (index <= end) {
1422 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1425 for (i = 0; i < nr_pages; i++) {
1426 struct page *page = pvec.pages[i];
1427 if (page->index > end)
1429 BUG_ON(!PageLocked(page));
1430 BUG_ON(PageWriteback(page));
1432 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1433 ClearPageUptodate(page);
1437 index = pvec.pages[nr_pages - 1]->index + 1;
1438 pagevec_release(&pvec);
1442 static void ext4_print_free_blocks(struct inode *inode)
1444 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1445 struct super_block *sb = inode->i_sb;
1446 struct ext4_inode_info *ei = EXT4_I(inode);
1448 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1449 EXT4_C2B(EXT4_SB(inode->i_sb),
1450 ext4_count_free_clusters(sb)));
1451 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1452 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1453 (long long) EXT4_C2B(EXT4_SB(sb),
1454 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1455 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1456 (long long) EXT4_C2B(EXT4_SB(sb),
1457 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1458 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1459 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1460 ei->i_reserved_data_blocks);
1464 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1466 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1470 * This function is grabs code from the very beginning of
1471 * ext4_map_blocks, but assumes that the caller is from delayed write
1472 * time. This function looks up the requested blocks and sets the
1473 * buffer delay bit under the protection of i_data_sem.
1475 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1476 struct ext4_map_blocks *map,
1477 struct buffer_head *bh)
1479 struct extent_status es;
1481 sector_t invalid_block = ~((sector_t) 0xffff);
1482 #ifdef ES_AGGRESSIVE_TEST
1483 struct ext4_map_blocks orig_map;
1485 memcpy(&orig_map, map, sizeof(*map));
1488 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1492 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1493 "logical block %lu\n", inode->i_ino, map->m_len,
1494 (unsigned long) map->m_lblk);
1496 /* Lookup extent status tree firstly */
1497 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1498 if (ext4_es_is_hole(&es)) {
1500 down_read(&EXT4_I(inode)->i_data_sem);
1505 * Delayed extent could be allocated by fallocate.
1506 * So we need to check it.
1508 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1509 map_bh(bh, inode->i_sb, invalid_block);
1511 set_buffer_delay(bh);
1515 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1516 retval = es.es_len - (iblock - es.es_lblk);
1517 if (retval > map->m_len)
1518 retval = map->m_len;
1519 map->m_len = retval;
1520 if (ext4_es_is_written(&es))
1521 map->m_flags |= EXT4_MAP_MAPPED;
1522 else if (ext4_es_is_unwritten(&es))
1523 map->m_flags |= EXT4_MAP_UNWRITTEN;
1527 #ifdef ES_AGGRESSIVE_TEST
1528 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1534 * Try to see if we can get the block without requesting a new
1535 * file system block.
1537 down_read(&EXT4_I(inode)->i_data_sem);
1538 if (ext4_has_inline_data(inode))
1540 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1541 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1543 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1549 * XXX: __block_prepare_write() unmaps passed block,
1553 * If the block was allocated from previously allocated cluster,
1554 * then we don't need to reserve it again. However we still need
1555 * to reserve metadata for every block we're going to write.
1557 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1558 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1559 ret = ext4_da_reserve_space(inode);
1561 /* not enough space to reserve */
1567 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1568 ~0, EXTENT_STATUS_DELAYED);
1574 map_bh(bh, inode->i_sb, invalid_block);
1576 set_buffer_delay(bh);
1577 } else if (retval > 0) {
1579 unsigned int status;
1581 if (unlikely(retval != map->m_len)) {
1582 ext4_warning(inode->i_sb,
1583 "ES len assertion failed for inode "
1584 "%lu: retval %d != map->m_len %d",
1585 inode->i_ino, retval, map->m_len);
1589 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1590 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1591 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1592 map->m_pblk, status);
1598 up_read((&EXT4_I(inode)->i_data_sem));
1604 * This is a special get_block_t callback which is used by
1605 * ext4_da_write_begin(). It will either return mapped block or
1606 * reserve space for a single block.
1608 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1609 * We also have b_blocknr = -1 and b_bdev initialized properly
1611 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1612 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1613 * initialized properly.
1615 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1616 struct buffer_head *bh, int create)
1618 struct ext4_map_blocks map;
1621 BUG_ON(create == 0);
1622 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1624 map.m_lblk = iblock;
1628 * first, we need to know whether the block is allocated already
1629 * preallocated blocks are unmapped but should treated
1630 * the same as allocated blocks.
1632 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1636 map_bh(bh, inode->i_sb, map.m_pblk);
1637 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1639 if (buffer_unwritten(bh)) {
1640 /* A delayed write to unwritten bh should be marked
1641 * new and mapped. Mapped ensures that we don't do
1642 * get_block multiple times when we write to the same
1643 * offset and new ensures that we do proper zero out
1644 * for partial write.
1647 set_buffer_mapped(bh);
1652 static int bget_one(handle_t *handle, struct buffer_head *bh)
1658 static int bput_one(handle_t *handle, struct buffer_head *bh)
1664 static int __ext4_journalled_writepage(struct page *page,
1667 struct address_space *mapping = page->mapping;
1668 struct inode *inode = mapping->host;
1669 struct buffer_head *page_bufs = NULL;
1670 handle_t *handle = NULL;
1671 int ret = 0, err = 0;
1672 int inline_data = ext4_has_inline_data(inode);
1673 struct buffer_head *inode_bh = NULL;
1675 ClearPageChecked(page);
1678 BUG_ON(page->index != 0);
1679 BUG_ON(len > ext4_get_max_inline_size(inode));
1680 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1681 if (inode_bh == NULL)
1684 page_bufs = page_buffers(page);
1689 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1693 * We need to release the page lock before we start the
1694 * journal, so grab a reference so the page won't disappear
1695 * out from under us.
1700 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1701 ext4_writepage_trans_blocks(inode));
1702 if (IS_ERR(handle)) {
1703 ret = PTR_ERR(handle);
1705 goto out_no_pagelock;
1707 BUG_ON(!ext4_handle_valid(handle));
1711 if (page->mapping != mapping) {
1712 /* The page got truncated from under us */
1713 ext4_journal_stop(handle);
1719 BUFFER_TRACE(inode_bh, "get write access");
1720 ret = ext4_journal_get_write_access(handle, inode_bh);
1722 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1725 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1726 do_journal_get_write_access);
1728 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1733 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1734 err = ext4_journal_stop(handle);
1738 if (!ext4_has_inline_data(inode))
1739 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1741 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1750 * Note that we don't need to start a transaction unless we're journaling data
1751 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1752 * need to file the inode to the transaction's list in ordered mode because if
1753 * we are writing back data added by write(), the inode is already there and if
1754 * we are writing back data modified via mmap(), no one guarantees in which
1755 * transaction the data will hit the disk. In case we are journaling data, we
1756 * cannot start transaction directly because transaction start ranks above page
1757 * lock so we have to do some magic.
1759 * This function can get called via...
1760 * - ext4_writepages after taking page lock (have journal handle)
1761 * - journal_submit_inode_data_buffers (no journal handle)
1762 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1763 * - grab_page_cache when doing write_begin (have journal handle)
1765 * We don't do any block allocation in this function. If we have page with
1766 * multiple blocks we need to write those buffer_heads that are mapped. This
1767 * is important for mmaped based write. So if we do with blocksize 1K
1768 * truncate(f, 1024);
1769 * a = mmap(f, 0, 4096);
1771 * truncate(f, 4096);
1772 * we have in the page first buffer_head mapped via page_mkwrite call back
1773 * but other buffer_heads would be unmapped but dirty (dirty done via the
1774 * do_wp_page). So writepage should write the first block. If we modify
1775 * the mmap area beyond 1024 we will again get a page_fault and the
1776 * page_mkwrite callback will do the block allocation and mark the
1777 * buffer_heads mapped.
1779 * We redirty the page if we have any buffer_heads that is either delay or
1780 * unwritten in the page.
1782 * We can get recursively called as show below.
1784 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1787 * But since we don't do any block allocation we should not deadlock.
1788 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1790 static int ext4_writepage(struct page *page,
1791 struct writeback_control *wbc)
1796 struct buffer_head *page_bufs = NULL;
1797 struct inode *inode = page->mapping->host;
1798 struct ext4_io_submit io_submit;
1799 bool keep_towrite = false;
1801 trace_ext4_writepage(page);
1802 size = i_size_read(inode);
1803 if (page->index == size >> PAGE_CACHE_SHIFT)
1804 len = size & ~PAGE_CACHE_MASK;
1806 len = PAGE_CACHE_SIZE;
1808 page_bufs = page_buffers(page);
1810 * We cannot do block allocation or other extent handling in this
1811 * function. If there are buffers needing that, we have to redirty
1812 * the page. But we may reach here when we do a journal commit via
1813 * journal_submit_inode_data_buffers() and in that case we must write
1814 * allocated buffers to achieve data=ordered mode guarantees.
1816 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1817 ext4_bh_delay_or_unwritten)) {
1818 redirty_page_for_writepage(wbc, page);
1819 if (current->flags & PF_MEMALLOC) {
1821 * For memory cleaning there's no point in writing only
1822 * some buffers. So just bail out. Warn if we came here
1823 * from direct reclaim.
1825 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1830 keep_towrite = true;
1833 if (PageChecked(page) && ext4_should_journal_data(inode))
1835 * It's mmapped pagecache. Add buffers and journal it. There
1836 * doesn't seem much point in redirtying the page here.
1838 return __ext4_journalled_writepage(page, len);
1840 ext4_io_submit_init(&io_submit, wbc);
1841 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1842 if (!io_submit.io_end) {
1843 redirty_page_for_writepage(wbc, page);
1847 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1848 ext4_io_submit(&io_submit);
1849 /* Drop io_end reference we got from init */
1850 ext4_put_io_end_defer(io_submit.io_end);
1854 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1857 loff_t size = i_size_read(mpd->inode);
1860 BUG_ON(page->index != mpd->first_page);
1861 if (page->index == size >> PAGE_CACHE_SHIFT)
1862 len = size & ~PAGE_CACHE_MASK;
1864 len = PAGE_CACHE_SIZE;
1865 clear_page_dirty_for_io(page);
1866 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1868 mpd->wbc->nr_to_write--;
1874 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1877 * mballoc gives us at most this number of blocks...
1878 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1879 * The rest of mballoc seems to handle chunks up to full group size.
1881 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1884 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1886 * @mpd - extent of blocks
1887 * @lblk - logical number of the block in the file
1888 * @bh - buffer head we want to add to the extent
1890 * The function is used to collect contig. blocks in the same state. If the
1891 * buffer doesn't require mapping for writeback and we haven't started the
1892 * extent of buffers to map yet, the function returns 'true' immediately - the
1893 * caller can write the buffer right away. Otherwise the function returns true
1894 * if the block has been added to the extent, false if the block couldn't be
1897 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1898 struct buffer_head *bh)
1900 struct ext4_map_blocks *map = &mpd->map;
1902 /* Buffer that doesn't need mapping for writeback? */
1903 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1904 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1905 /* So far no extent to map => we write the buffer right away */
1906 if (map->m_len == 0)
1911 /* First block in the extent? */
1912 if (map->m_len == 0) {
1915 map->m_flags = bh->b_state & BH_FLAGS;
1919 /* Don't go larger than mballoc is willing to allocate */
1920 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1923 /* Can we merge the block to our big extent? */
1924 if (lblk == map->m_lblk + map->m_len &&
1925 (bh->b_state & BH_FLAGS) == map->m_flags) {
1933 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1935 * @mpd - extent of blocks for mapping
1936 * @head - the first buffer in the page
1937 * @bh - buffer we should start processing from
1938 * @lblk - logical number of the block in the file corresponding to @bh
1940 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1941 * the page for IO if all buffers in this page were mapped and there's no
1942 * accumulated extent of buffers to map or add buffers in the page to the
1943 * extent of buffers to map. The function returns 1 if the caller can continue
1944 * by processing the next page, 0 if it should stop adding buffers to the
1945 * extent to map because we cannot extend it anymore. It can also return value
1946 * < 0 in case of error during IO submission.
1948 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1949 struct buffer_head *head,
1950 struct buffer_head *bh,
1953 struct inode *inode = mpd->inode;
1955 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
1956 >> inode->i_blkbits;
1959 BUG_ON(buffer_locked(bh));
1961 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
1962 /* Found extent to map? */
1965 /* Everything mapped so far and we hit EOF */
1968 } while (lblk++, (bh = bh->b_this_page) != head);
1969 /* So far everything mapped? Submit the page for IO. */
1970 if (mpd->map.m_len == 0) {
1971 err = mpage_submit_page(mpd, head->b_page);
1975 return lblk < blocks;
1979 * mpage_map_buffers - update buffers corresponding to changed extent and
1980 * submit fully mapped pages for IO
1982 * @mpd - description of extent to map, on return next extent to map
1984 * Scan buffers corresponding to changed extent (we expect corresponding pages
1985 * to be already locked) and update buffer state according to new extent state.
1986 * We map delalloc buffers to their physical location, clear unwritten bits,
1987 * and mark buffers as uninit when we perform writes to unwritten extents
1988 * and do extent conversion after IO is finished. If the last page is not fully
1989 * mapped, we update @map to the next extent in the last page that needs
1990 * mapping. Otherwise we submit the page for IO.
1992 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
1994 struct pagevec pvec;
1996 struct inode *inode = mpd->inode;
1997 struct buffer_head *head, *bh;
1998 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
2004 start = mpd->map.m_lblk >> bpp_bits;
2005 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2006 lblk = start << bpp_bits;
2007 pblock = mpd->map.m_pblk;
2009 pagevec_init(&pvec, 0);
2010 while (start <= end) {
2011 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2015 for (i = 0; i < nr_pages; i++) {
2016 struct page *page = pvec.pages[i];
2018 if (page->index > end)
2020 /* Up to 'end' pages must be contiguous */
2021 BUG_ON(page->index != start);
2022 bh = head = page_buffers(page);
2024 if (lblk < mpd->map.m_lblk)
2026 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2028 * Buffer after end of mapped extent.
2029 * Find next buffer in the page to map.
2032 mpd->map.m_flags = 0;
2034 * FIXME: If dioread_nolock supports
2035 * blocksize < pagesize, we need to make
2036 * sure we add size mapped so far to
2037 * io_end->size as the following call
2038 * can submit the page for IO.
2040 err = mpage_process_page_bufs(mpd, head,
2042 pagevec_release(&pvec);
2047 if (buffer_delay(bh)) {
2048 clear_buffer_delay(bh);
2049 bh->b_blocknr = pblock++;
2051 clear_buffer_unwritten(bh);
2052 } while (lblk++, (bh = bh->b_this_page) != head);
2055 * FIXME: This is going to break if dioread_nolock
2056 * supports blocksize < pagesize as we will try to
2057 * convert potentially unmapped parts of inode.
2059 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2060 /* Page fully mapped - let IO run! */
2061 err = mpage_submit_page(mpd, page);
2063 pagevec_release(&pvec);
2068 pagevec_release(&pvec);
2070 /* Extent fully mapped and matches with page boundary. We are done. */
2072 mpd->map.m_flags = 0;
2076 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2078 struct inode *inode = mpd->inode;
2079 struct ext4_map_blocks *map = &mpd->map;
2080 int get_blocks_flags;
2081 int err, dioread_nolock;
2083 trace_ext4_da_write_pages_extent(inode, map);
2085 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2086 * to convert an unwritten extent to be initialized (in the case
2087 * where we have written into one or more preallocated blocks). It is
2088 * possible that we're going to need more metadata blocks than
2089 * previously reserved. However we must not fail because we're in
2090 * writeback and there is nothing we can do about it so it might result
2091 * in data loss. So use reserved blocks to allocate metadata if
2094 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2095 * the blocks in question are delalloc blocks. This indicates
2096 * that the blocks and quotas has already been checked when
2097 * the data was copied into the page cache.
2099 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2100 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2101 dioread_nolock = ext4_should_dioread_nolock(inode);
2103 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2104 if (map->m_flags & (1 << BH_Delay))
2105 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2107 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2110 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2111 if (!mpd->io_submit.io_end->handle &&
2112 ext4_handle_valid(handle)) {
2113 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2114 handle->h_rsv_handle = NULL;
2116 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2119 BUG_ON(map->m_len == 0);
2120 if (map->m_flags & EXT4_MAP_NEW) {
2121 struct block_device *bdev = inode->i_sb->s_bdev;
2124 for (i = 0; i < map->m_len; i++)
2125 unmap_underlying_metadata(bdev, map->m_pblk + i);
2131 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2132 * mpd->len and submit pages underlying it for IO
2134 * @handle - handle for journal operations
2135 * @mpd - extent to map
2136 * @give_up_on_write - we set this to true iff there is a fatal error and there
2137 * is no hope of writing the data. The caller should discard
2138 * dirty pages to avoid infinite loops.
2140 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2141 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2142 * them to initialized or split the described range from larger unwritten
2143 * extent. Note that we need not map all the described range since allocation
2144 * can return less blocks or the range is covered by more unwritten extents. We
2145 * cannot map more because we are limited by reserved transaction credits. On
2146 * the other hand we always make sure that the last touched page is fully
2147 * mapped so that it can be written out (and thus forward progress is
2148 * guaranteed). After mapping we submit all mapped pages for IO.
2150 static int mpage_map_and_submit_extent(handle_t *handle,
2151 struct mpage_da_data *mpd,
2152 bool *give_up_on_write)
2154 struct inode *inode = mpd->inode;
2155 struct ext4_map_blocks *map = &mpd->map;
2160 mpd->io_submit.io_end->offset =
2161 ((loff_t)map->m_lblk) << inode->i_blkbits;
2163 err = mpage_map_one_extent(handle, mpd);
2165 struct super_block *sb = inode->i_sb;
2167 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2168 goto invalidate_dirty_pages;
2170 * Let the uper layers retry transient errors.
2171 * In the case of ENOSPC, if ext4_count_free_blocks()
2172 * is non-zero, a commit should free up blocks.
2174 if ((err == -ENOMEM) ||
2175 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2177 goto update_disksize;
2180 ext4_msg(sb, KERN_CRIT,
2181 "Delayed block allocation failed for "
2182 "inode %lu at logical offset %llu with"
2183 " max blocks %u with error %d",
2185 (unsigned long long)map->m_lblk,
2186 (unsigned)map->m_len, -err);
2187 ext4_msg(sb, KERN_CRIT,
2188 "This should not happen!! Data will "
2191 ext4_print_free_blocks(inode);
2192 invalidate_dirty_pages:
2193 *give_up_on_write = true;
2198 * Update buffer state, submit mapped pages, and get us new
2201 err = mpage_map_and_submit_buffers(mpd);
2203 goto update_disksize;
2204 } while (map->m_len);
2208 * Update on-disk size after IO is submitted. Races with
2209 * truncate are avoided by checking i_size under i_data_sem.
2211 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2212 if (disksize > EXT4_I(inode)->i_disksize) {
2216 down_write(&EXT4_I(inode)->i_data_sem);
2217 i_size = i_size_read(inode);
2218 if (disksize > i_size)
2220 if (disksize > EXT4_I(inode)->i_disksize)
2221 EXT4_I(inode)->i_disksize = disksize;
2222 err2 = ext4_mark_inode_dirty(handle, inode);
2223 up_write(&EXT4_I(inode)->i_data_sem);
2225 ext4_error(inode->i_sb,
2226 "Failed to mark inode %lu dirty",
2235 * Calculate the total number of credits to reserve for one writepages
2236 * iteration. This is called from ext4_writepages(). We map an extent of
2237 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2238 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2239 * bpp - 1 blocks in bpp different extents.
2241 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2243 int bpp = ext4_journal_blocks_per_page(inode);
2245 return ext4_meta_trans_blocks(inode,
2246 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2250 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2251 * and underlying extent to map
2253 * @mpd - where to look for pages
2255 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2256 * IO immediately. When we find a page which isn't mapped we start accumulating
2257 * extent of buffers underlying these pages that needs mapping (formed by
2258 * either delayed or unwritten buffers). We also lock the pages containing
2259 * these buffers. The extent found is returned in @mpd structure (starting at
2260 * mpd->lblk with length mpd->len blocks).
2262 * Note that this function can attach bios to one io_end structure which are
2263 * neither logically nor physically contiguous. Although it may seem as an
2264 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2265 * case as we need to track IO to all buffers underlying a page in one io_end.
2267 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2269 struct address_space *mapping = mpd->inode->i_mapping;
2270 struct pagevec pvec;
2271 unsigned int nr_pages;
2272 long left = mpd->wbc->nr_to_write;
2273 pgoff_t index = mpd->first_page;
2274 pgoff_t end = mpd->last_page;
2277 int blkbits = mpd->inode->i_blkbits;
2279 struct buffer_head *head;
2281 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2282 tag = PAGECACHE_TAG_TOWRITE;
2284 tag = PAGECACHE_TAG_DIRTY;
2286 pagevec_init(&pvec, 0);
2288 mpd->next_page = index;
2289 while (index <= end) {
2290 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2291 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2295 for (i = 0; i < nr_pages; i++) {
2296 struct page *page = pvec.pages[i];
2299 * At this point, the page may be truncated or
2300 * invalidated (changing page->mapping to NULL), or
2301 * even swizzled back from swapper_space to tmpfs file
2302 * mapping. However, page->index will not change
2303 * because we have a reference on the page.
2305 if (page->index > end)
2309 * Accumulated enough dirty pages? This doesn't apply
2310 * to WB_SYNC_ALL mode. For integrity sync we have to
2311 * keep going because someone may be concurrently
2312 * dirtying pages, and we might have synced a lot of
2313 * newly appeared dirty pages, but have not synced all
2314 * of the old dirty pages.
2316 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2319 /* If we can't merge this page, we are done. */
2320 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2325 * If the page is no longer dirty, or its mapping no
2326 * longer corresponds to inode we are writing (which
2327 * means it has been truncated or invalidated), or the
2328 * page is already under writeback and we are not doing
2329 * a data integrity writeback, skip the page
2331 if (!PageDirty(page) ||
2332 (PageWriteback(page) &&
2333 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2334 unlikely(page->mapping != mapping)) {
2339 wait_on_page_writeback(page);
2340 BUG_ON(PageWriteback(page));
2342 if (mpd->map.m_len == 0)
2343 mpd->first_page = page->index;
2344 mpd->next_page = page->index + 1;
2345 /* Add all dirty buffers to mpd */
2346 lblk = ((ext4_lblk_t)page->index) <<
2347 (PAGE_CACHE_SHIFT - blkbits);
2348 head = page_buffers(page);
2349 err = mpage_process_page_bufs(mpd, head, head, lblk);
2355 pagevec_release(&pvec);
2360 pagevec_release(&pvec);
2364 static int __writepage(struct page *page, struct writeback_control *wbc,
2367 struct address_space *mapping = data;
2368 int ret = ext4_writepage(page, wbc);
2369 mapping_set_error(mapping, ret);
2373 static int ext4_writepages(struct address_space *mapping,
2374 struct writeback_control *wbc)
2376 pgoff_t writeback_index = 0;
2377 long nr_to_write = wbc->nr_to_write;
2378 int range_whole = 0;
2380 handle_t *handle = NULL;
2381 struct mpage_da_data mpd;
2382 struct inode *inode = mapping->host;
2383 int needed_blocks, rsv_blocks = 0, ret = 0;
2384 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2386 struct blk_plug plug;
2387 bool give_up_on_write = false;
2389 trace_ext4_writepages(inode, wbc);
2392 * No pages to write? This is mainly a kludge to avoid starting
2393 * a transaction for special inodes like journal inode on last iput()
2394 * because that could violate lock ordering on umount
2396 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2397 goto out_writepages;
2399 if (ext4_should_journal_data(inode)) {
2400 struct blk_plug plug;
2402 blk_start_plug(&plug);
2403 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2404 blk_finish_plug(&plug);
2405 goto out_writepages;
2409 * If the filesystem has aborted, it is read-only, so return
2410 * right away instead of dumping stack traces later on that
2411 * will obscure the real source of the problem. We test
2412 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2413 * the latter could be true if the filesystem is mounted
2414 * read-only, and in that case, ext4_writepages should
2415 * *never* be called, so if that ever happens, we would want
2418 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2420 goto out_writepages;
2423 if (ext4_should_dioread_nolock(inode)) {
2425 * We may need to convert up to one extent per block in
2426 * the page and we may dirty the inode.
2428 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2432 * If we have inline data and arrive here, it means that
2433 * we will soon create the block for the 1st page, so
2434 * we'd better clear the inline data here.
2436 if (ext4_has_inline_data(inode)) {
2437 /* Just inode will be modified... */
2438 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2439 if (IS_ERR(handle)) {
2440 ret = PTR_ERR(handle);
2441 goto out_writepages;
2443 BUG_ON(ext4_test_inode_state(inode,
2444 EXT4_STATE_MAY_INLINE_DATA));
2445 ext4_destroy_inline_data(handle, inode);
2446 ext4_journal_stop(handle);
2449 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2452 if (wbc->range_cyclic) {
2453 writeback_index = mapping->writeback_index;
2454 if (writeback_index)
2456 mpd.first_page = writeback_index;
2459 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2460 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2465 ext4_io_submit_init(&mpd.io_submit, wbc);
2467 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2468 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2470 blk_start_plug(&plug);
2471 while (!done && mpd.first_page <= mpd.last_page) {
2472 /* For each extent of pages we use new io_end */
2473 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2474 if (!mpd.io_submit.io_end) {
2480 * We have two constraints: We find one extent to map and we
2481 * must always write out whole page (makes a difference when
2482 * blocksize < pagesize) so that we don't block on IO when we
2483 * try to write out the rest of the page. Journalled mode is
2484 * not supported by delalloc.
2486 BUG_ON(ext4_should_journal_data(inode));
2487 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2489 /* start a new transaction */
2490 handle = ext4_journal_start_with_reserve(inode,
2491 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2492 if (IS_ERR(handle)) {
2493 ret = PTR_ERR(handle);
2494 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2495 "%ld pages, ino %lu; err %d", __func__,
2496 wbc->nr_to_write, inode->i_ino, ret);
2497 /* Release allocated io_end */
2498 ext4_put_io_end(mpd.io_submit.io_end);
2502 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2503 ret = mpage_prepare_extent_to_map(&mpd);
2506 ret = mpage_map_and_submit_extent(handle, &mpd,
2510 * We scanned the whole range (or exhausted
2511 * nr_to_write), submitted what was mapped and
2512 * didn't find anything needing mapping. We are
2518 ext4_journal_stop(handle);
2519 /* Submit prepared bio */
2520 ext4_io_submit(&mpd.io_submit);
2521 /* Unlock pages we didn't use */
2522 mpage_release_unused_pages(&mpd, give_up_on_write);
2523 /* Drop our io_end reference we got from init */
2524 ext4_put_io_end(mpd.io_submit.io_end);
2526 if (ret == -ENOSPC && sbi->s_journal) {
2528 * Commit the transaction which would
2529 * free blocks released in the transaction
2532 jbd2_journal_force_commit_nested(sbi->s_journal);
2536 /* Fatal error - ENOMEM, EIO... */
2540 blk_finish_plug(&plug);
2541 if (!ret && !cycled && wbc->nr_to_write > 0) {
2543 mpd.last_page = writeback_index - 1;
2549 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2551 * Set the writeback_index so that range_cyclic
2552 * mode will write it back later
2554 mapping->writeback_index = mpd.first_page;
2557 trace_ext4_writepages_result(inode, wbc, ret,
2558 nr_to_write - wbc->nr_to_write);
2562 static int ext4_nonda_switch(struct super_block *sb)
2564 s64 free_clusters, dirty_clusters;
2565 struct ext4_sb_info *sbi = EXT4_SB(sb);
2568 * switch to non delalloc mode if we are running low
2569 * on free block. The free block accounting via percpu
2570 * counters can get slightly wrong with percpu_counter_batch getting
2571 * accumulated on each CPU without updating global counters
2572 * Delalloc need an accurate free block accounting. So switch
2573 * to non delalloc when we are near to error range.
2576 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2578 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2580 * Start pushing delalloc when 1/2 of free blocks are dirty.
2582 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2583 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2585 if (2 * free_clusters < 3 * dirty_clusters ||
2586 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2588 * free block count is less than 150% of dirty blocks
2589 * or free blocks is less than watermark
2596 /* We always reserve for an inode update; the superblock could be there too */
2597 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2599 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
2600 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2603 if (pos + len <= 0x7fffffffULL)
2606 /* We might need to update the superblock to set LARGE_FILE */
2610 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2611 loff_t pos, unsigned len, unsigned flags,
2612 struct page **pagep, void **fsdata)
2614 int ret, retries = 0;
2617 struct inode *inode = mapping->host;
2620 index = pos >> PAGE_CACHE_SHIFT;
2622 if (ext4_nonda_switch(inode->i_sb)) {
2623 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2624 return ext4_write_begin(file, mapping, pos,
2625 len, flags, pagep, fsdata);
2627 *fsdata = (void *)0;
2628 trace_ext4_da_write_begin(inode, pos, len, flags);
2630 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2631 ret = ext4_da_write_inline_data_begin(mapping, inode,
2641 * grab_cache_page_write_begin() can take a long time if the
2642 * system is thrashing due to memory pressure, or if the page
2643 * is being written back. So grab it first before we start
2644 * the transaction handle. This also allows us to allocate
2645 * the page (if needed) without using GFP_NOFS.
2648 page = grab_cache_page_write_begin(mapping, index, flags);
2654 * With delayed allocation, we don't log the i_disksize update
2655 * if there is delayed block allocation. But we still need
2656 * to journalling the i_disksize update if writes to the end
2657 * of file which has an already mapped buffer.
2660 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2661 ext4_da_write_credits(inode, pos, len));
2662 if (IS_ERR(handle)) {
2663 page_cache_release(page);
2664 return PTR_ERR(handle);
2668 if (page->mapping != mapping) {
2669 /* The page got truncated from under us */
2671 page_cache_release(page);
2672 ext4_journal_stop(handle);
2675 /* In case writeback began while the page was unlocked */
2676 wait_for_stable_page(page);
2678 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2679 ret = ext4_block_write_begin(page, pos, len,
2680 ext4_da_get_block_prep);
2682 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2686 ext4_journal_stop(handle);
2688 * block_write_begin may have instantiated a few blocks
2689 * outside i_size. Trim these off again. Don't need
2690 * i_size_read because we hold i_mutex.
2692 if (pos + len > inode->i_size)
2693 ext4_truncate_failed_write(inode);
2695 if (ret == -ENOSPC &&
2696 ext4_should_retry_alloc(inode->i_sb, &retries))
2699 page_cache_release(page);
2708 * Check if we should update i_disksize
2709 * when write to the end of file but not require block allocation
2711 static int ext4_da_should_update_i_disksize(struct page *page,
2712 unsigned long offset)
2714 struct buffer_head *bh;
2715 struct inode *inode = page->mapping->host;
2719 bh = page_buffers(page);
2720 idx = offset >> inode->i_blkbits;
2722 for (i = 0; i < idx; i++)
2723 bh = bh->b_this_page;
2725 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2730 static int ext4_da_write_end(struct file *file,
2731 struct address_space *mapping,
2732 loff_t pos, unsigned len, unsigned copied,
2733 struct page *page, void *fsdata)
2735 struct inode *inode = mapping->host;
2737 handle_t *handle = ext4_journal_current_handle();
2739 unsigned long start, end;
2740 int write_mode = (int)(unsigned long)fsdata;
2742 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2743 return ext4_write_end(file, mapping, pos,
2744 len, copied, page, fsdata);
2746 trace_ext4_da_write_end(inode, pos, len, copied);
2747 start = pos & (PAGE_CACHE_SIZE - 1);
2748 end = start + copied - 1;
2751 * generic_write_end() will run mark_inode_dirty() if i_size
2752 * changes. So let's piggyback the i_disksize mark_inode_dirty
2755 new_i_size = pos + copied;
2756 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2757 if (ext4_has_inline_data(inode) ||
2758 ext4_da_should_update_i_disksize(page, end)) {
2759 ext4_update_i_disksize(inode, new_i_size);
2760 /* We need to mark inode dirty even if
2761 * new_i_size is less that inode->i_size
2762 * bu greater than i_disksize.(hint delalloc)
2764 ext4_mark_inode_dirty(handle, inode);
2768 if (write_mode != CONVERT_INLINE_DATA &&
2769 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2770 ext4_has_inline_data(inode))
2771 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2774 ret2 = generic_write_end(file, mapping, pos, len, copied,
2780 ret2 = ext4_journal_stop(handle);
2784 return ret ? ret : copied;
2787 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2788 unsigned int length)
2791 * Drop reserved blocks
2793 BUG_ON(!PageLocked(page));
2794 if (!page_has_buffers(page))
2797 ext4_da_page_release_reservation(page, offset, length);
2800 ext4_invalidatepage(page, offset, length);
2806 * Force all delayed allocation blocks to be allocated for a given inode.
2808 int ext4_alloc_da_blocks(struct inode *inode)
2810 trace_ext4_alloc_da_blocks(inode);
2812 if (!EXT4_I(inode)->i_reserved_data_blocks)
2816 * We do something simple for now. The filemap_flush() will
2817 * also start triggering a write of the data blocks, which is
2818 * not strictly speaking necessary (and for users of
2819 * laptop_mode, not even desirable). However, to do otherwise
2820 * would require replicating code paths in:
2822 * ext4_writepages() ->
2823 * write_cache_pages() ---> (via passed in callback function)
2824 * __mpage_da_writepage() -->
2825 * mpage_add_bh_to_extent()
2826 * mpage_da_map_blocks()
2828 * The problem is that write_cache_pages(), located in
2829 * mm/page-writeback.c, marks pages clean in preparation for
2830 * doing I/O, which is not desirable if we're not planning on
2833 * We could call write_cache_pages(), and then redirty all of
2834 * the pages by calling redirty_page_for_writepage() but that
2835 * would be ugly in the extreme. So instead we would need to
2836 * replicate parts of the code in the above functions,
2837 * simplifying them because we wouldn't actually intend to
2838 * write out the pages, but rather only collect contiguous
2839 * logical block extents, call the multi-block allocator, and
2840 * then update the buffer heads with the block allocations.
2842 * For now, though, we'll cheat by calling filemap_flush(),
2843 * which will map the blocks, and start the I/O, but not
2844 * actually wait for the I/O to complete.
2846 return filemap_flush(inode->i_mapping);
2850 * bmap() is special. It gets used by applications such as lilo and by
2851 * the swapper to find the on-disk block of a specific piece of data.
2853 * Naturally, this is dangerous if the block concerned is still in the
2854 * journal. If somebody makes a swapfile on an ext4 data-journaling
2855 * filesystem and enables swap, then they may get a nasty shock when the
2856 * data getting swapped to that swapfile suddenly gets overwritten by
2857 * the original zero's written out previously to the journal and
2858 * awaiting writeback in the kernel's buffer cache.
2860 * So, if we see any bmap calls here on a modified, data-journaled file,
2861 * take extra steps to flush any blocks which might be in the cache.
2863 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2865 struct inode *inode = mapping->host;
2870 * We can get here for an inline file via the FIBMAP ioctl
2872 if (ext4_has_inline_data(inode))
2875 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2876 test_opt(inode->i_sb, DELALLOC)) {
2878 * With delalloc we want to sync the file
2879 * so that we can make sure we allocate
2882 filemap_write_and_wait(mapping);
2885 if (EXT4_JOURNAL(inode) &&
2886 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2888 * This is a REALLY heavyweight approach, but the use of
2889 * bmap on dirty files is expected to be extremely rare:
2890 * only if we run lilo or swapon on a freshly made file
2891 * do we expect this to happen.
2893 * (bmap requires CAP_SYS_RAWIO so this does not
2894 * represent an unprivileged user DOS attack --- we'd be
2895 * in trouble if mortal users could trigger this path at
2898 * NB. EXT4_STATE_JDATA is not set on files other than
2899 * regular files. If somebody wants to bmap a directory
2900 * or symlink and gets confused because the buffer
2901 * hasn't yet been flushed to disk, they deserve
2902 * everything they get.
2905 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2906 journal = EXT4_JOURNAL(inode);
2907 jbd2_journal_lock_updates(journal);
2908 err = jbd2_journal_flush(journal);
2909 jbd2_journal_unlock_updates(journal);
2915 return generic_block_bmap(mapping, block, ext4_get_block);
2918 static int ext4_readpage(struct file *file, struct page *page)
2921 struct inode *inode = page->mapping->host;
2923 trace_ext4_readpage(page);
2925 if (ext4_has_inline_data(inode))
2926 ret = ext4_readpage_inline(inode, page);
2929 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
2935 ext4_readpages(struct file *file, struct address_space *mapping,
2936 struct list_head *pages, unsigned nr_pages)
2938 struct inode *inode = mapping->host;
2940 /* If the file has inline data, no need to do readpages. */
2941 if (ext4_has_inline_data(inode))
2944 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
2947 static void ext4_invalidatepage(struct page *page, unsigned int offset,
2948 unsigned int length)
2950 trace_ext4_invalidatepage(page, offset, length);
2952 /* No journalling happens on data buffers when this function is used */
2953 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2955 block_invalidatepage(page, offset, length);
2958 static int __ext4_journalled_invalidatepage(struct page *page,
2959 unsigned int offset,
2960 unsigned int length)
2962 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2964 trace_ext4_journalled_invalidatepage(page, offset, length);
2967 * If it's a full truncate we just forget about the pending dirtying
2969 if (offset == 0 && length == PAGE_CACHE_SIZE)
2970 ClearPageChecked(page);
2972 return jbd2_journal_invalidatepage(journal, page, offset, length);
2975 /* Wrapper for aops... */
2976 static void ext4_journalled_invalidatepage(struct page *page,
2977 unsigned int offset,
2978 unsigned int length)
2980 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
2983 static int ext4_releasepage(struct page *page, gfp_t wait)
2985 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2987 trace_ext4_releasepage(page);
2989 /* Page has dirty journalled data -> cannot release */
2990 if (PageChecked(page))
2993 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2995 return try_to_free_buffers(page);
2999 * ext4_get_block used when preparing for a DIO write or buffer write.
3000 * We allocate an uinitialized extent if blocks haven't been allocated.
3001 * The extent will be converted to initialized after the IO is complete.
3003 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3004 struct buffer_head *bh_result, int create)
3006 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3007 inode->i_ino, create);
3008 return _ext4_get_block(inode, iblock, bh_result,
3009 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3012 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3013 struct buffer_head *bh_result, int create)
3015 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3016 inode->i_ino, create);
3017 return _ext4_get_block(inode, iblock, bh_result,
3018 EXT4_GET_BLOCKS_NO_LOCK);
3021 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3022 ssize_t size, void *private)
3024 ext4_io_end_t *io_end = iocb->private;
3026 /* if not async direct IO just return */
3030 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3031 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3032 iocb->private, io_end->inode->i_ino, iocb, offset,
3035 iocb->private = NULL;
3036 io_end->offset = offset;
3037 io_end->size = size;
3038 ext4_put_io_end(io_end);
3042 * For ext4 extent files, ext4 will do direct-io write to holes,
3043 * preallocated extents, and those write extend the file, no need to
3044 * fall back to buffered IO.
3046 * For holes, we fallocate those blocks, mark them as unwritten
3047 * If those blocks were preallocated, we mark sure they are split, but
3048 * still keep the range to write as unwritten.
3050 * The unwritten extents will be converted to written when DIO is completed.
3051 * For async direct IO, since the IO may still pending when return, we
3052 * set up an end_io call back function, which will do the conversion
3053 * when async direct IO completed.
3055 * If the O_DIRECT write will extend the file then add this inode to the
3056 * orphan list. So recovery will truncate it back to the original size
3057 * if the machine crashes during the write.
3060 static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3063 struct file *file = iocb->ki_filp;
3064 struct inode *inode = file->f_mapping->host;
3066 size_t count = iov_iter_count(iter);
3068 get_block_t *get_block_func = NULL;
3070 loff_t final_size = offset + count;
3071 ext4_io_end_t *io_end = NULL;
3073 /* Use the old path for reads and writes beyond i_size. */
3074 if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size)
3075 return ext4_ind_direct_IO(iocb, iter, offset);
3077 BUG_ON(iocb->private == NULL);
3080 * Make all waiters for direct IO properly wait also for extent
3081 * conversion. This also disallows race between truncate() and
3082 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3084 if (iov_iter_rw(iter) == WRITE)
3085 inode_dio_begin(inode);
3087 /* If we do a overwrite dio, i_mutex locking can be released */
3088 overwrite = *((int *)iocb->private);
3091 down_read(&EXT4_I(inode)->i_data_sem);
3092 mutex_unlock(&inode->i_mutex);
3096 * We could direct write to holes and fallocate.
3098 * Allocated blocks to fill the hole are marked as
3099 * unwritten to prevent parallel buffered read to expose
3100 * the stale data before DIO complete the data IO.
3102 * As to previously fallocated extents, ext4 get_block will
3103 * just simply mark the buffer mapped but still keep the
3104 * extents unwritten.
3106 * For non AIO case, we will convert those unwritten extents
3107 * to written after return back from blockdev_direct_IO.
3109 * For async DIO, the conversion needs to be deferred when the
3110 * IO is completed. The ext4 end_io callback function will be
3111 * called to take care of the conversion work. Here for async
3112 * case, we allocate an io_end structure to hook to the iocb.
3114 iocb->private = NULL;
3115 ext4_inode_aio_set(inode, NULL);
3116 if (!is_sync_kiocb(iocb)) {
3117 io_end = ext4_init_io_end(inode, GFP_NOFS);
3123 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3125 iocb->private = ext4_get_io_end(io_end);
3127 * we save the io structure for current async direct
3128 * IO, so that later ext4_map_blocks() could flag the
3129 * io structure whether there is a unwritten extents
3130 * needs to be converted when IO is completed.
3132 ext4_inode_aio_set(inode, io_end);
3136 get_block_func = ext4_get_block_write_nolock;
3138 get_block_func = ext4_get_block_write;
3139 dio_flags = DIO_LOCKING;
3141 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3142 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3145 ret = dax_do_io(iocb, inode, iter, offset, get_block_func,
3146 ext4_end_io_dio, dio_flags);
3148 ret = __blockdev_direct_IO(iocb, inode,
3149 inode->i_sb->s_bdev, iter, offset,
3151 ext4_end_io_dio, NULL, dio_flags);
3154 * Put our reference to io_end. This can free the io_end structure e.g.
3155 * in sync IO case or in case of error. It can even perform extent
3156 * conversion if all bios we submitted finished before we got here.
3157 * Note that in that case iocb->private can be already set to NULL
3161 ext4_inode_aio_set(inode, NULL);
3162 ext4_put_io_end(io_end);
3164 * When no IO was submitted ext4_end_io_dio() was not
3165 * called so we have to put iocb's reference.
3167 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3168 WARN_ON(iocb->private != io_end);
3169 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3170 ext4_put_io_end(io_end);
3171 iocb->private = NULL;
3174 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3175 EXT4_STATE_DIO_UNWRITTEN)) {
3178 * for non AIO case, since the IO is already
3179 * completed, we could do the conversion right here
3181 err = ext4_convert_unwritten_extents(NULL, inode,
3185 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3189 if (iov_iter_rw(iter) == WRITE)
3190 inode_dio_end(inode);
3191 /* take i_mutex locking again if we do a ovewrite dio */
3193 up_read(&EXT4_I(inode)->i_data_sem);
3194 mutex_lock(&inode->i_mutex);
3200 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3203 struct file *file = iocb->ki_filp;
3204 struct inode *inode = file->f_mapping->host;
3205 size_t count = iov_iter_count(iter);
3208 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3209 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3214 * If we are doing data journalling we don't support O_DIRECT
3216 if (ext4_should_journal_data(inode))
3219 /* Let buffer I/O handle the inline data case. */
3220 if (ext4_has_inline_data(inode))
3223 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3224 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3225 ret = ext4_ext_direct_IO(iocb, iter, offset);
3227 ret = ext4_ind_direct_IO(iocb, iter, offset);
3228 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3233 * Pages can be marked dirty completely asynchronously from ext4's journalling
3234 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3235 * much here because ->set_page_dirty is called under VFS locks. The page is
3236 * not necessarily locked.
3238 * We cannot just dirty the page and leave attached buffers clean, because the
3239 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3240 * or jbddirty because all the journalling code will explode.
3242 * So what we do is to mark the page "pending dirty" and next time writepage
3243 * is called, propagate that into the buffers appropriately.
3245 static int ext4_journalled_set_page_dirty(struct page *page)
3247 SetPageChecked(page);
3248 return __set_page_dirty_nobuffers(page);
3251 static const struct address_space_operations ext4_aops = {
3252 .readpage = ext4_readpage,
3253 .readpages = ext4_readpages,
3254 .writepage = ext4_writepage,
3255 .writepages = ext4_writepages,
3256 .write_begin = ext4_write_begin,
3257 .write_end = ext4_write_end,
3259 .invalidatepage = ext4_invalidatepage,
3260 .releasepage = ext4_releasepage,
3261 .direct_IO = ext4_direct_IO,
3262 .migratepage = buffer_migrate_page,
3263 .is_partially_uptodate = block_is_partially_uptodate,
3264 .error_remove_page = generic_error_remove_page,
3267 static const struct address_space_operations ext4_journalled_aops = {
3268 .readpage = ext4_readpage,
3269 .readpages = ext4_readpages,
3270 .writepage = ext4_writepage,
3271 .writepages = ext4_writepages,
3272 .write_begin = ext4_write_begin,
3273 .write_end = ext4_journalled_write_end,
3274 .set_page_dirty = ext4_journalled_set_page_dirty,
3276 .invalidatepage = ext4_journalled_invalidatepage,
3277 .releasepage = ext4_releasepage,
3278 .direct_IO = ext4_direct_IO,
3279 .is_partially_uptodate = block_is_partially_uptodate,
3280 .error_remove_page = generic_error_remove_page,
3283 static const struct address_space_operations ext4_da_aops = {
3284 .readpage = ext4_readpage,
3285 .readpages = ext4_readpages,
3286 .writepage = ext4_writepage,
3287 .writepages = ext4_writepages,
3288 .write_begin = ext4_da_write_begin,
3289 .write_end = ext4_da_write_end,
3291 .invalidatepage = ext4_da_invalidatepage,
3292 .releasepage = ext4_releasepage,
3293 .direct_IO = ext4_direct_IO,
3294 .migratepage = buffer_migrate_page,
3295 .is_partially_uptodate = block_is_partially_uptodate,
3296 .error_remove_page = generic_error_remove_page,
3299 void ext4_set_aops(struct inode *inode)
3301 switch (ext4_inode_journal_mode(inode)) {
3302 case EXT4_INODE_ORDERED_DATA_MODE:
3303 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3305 case EXT4_INODE_WRITEBACK_DATA_MODE:
3306 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3308 case EXT4_INODE_JOURNAL_DATA_MODE:
3309 inode->i_mapping->a_ops = &ext4_journalled_aops;
3314 if (test_opt(inode->i_sb, DELALLOC))
3315 inode->i_mapping->a_ops = &ext4_da_aops;
3317 inode->i_mapping->a_ops = &ext4_aops;
3320 static int __ext4_block_zero_page_range(handle_t *handle,
3321 struct address_space *mapping, loff_t from, loff_t length)
3323 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3324 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3325 unsigned blocksize, pos;
3327 struct inode *inode = mapping->host;
3328 struct buffer_head *bh;
3332 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3333 mapping_gfp_mask(mapping) & ~__GFP_FS);
3337 blocksize = inode->i_sb->s_blocksize;
3339 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3341 if (!page_has_buffers(page))
3342 create_empty_buffers(page, blocksize, 0);
3344 /* Find the buffer that contains "offset" */
3345 bh = page_buffers(page);
3347 while (offset >= pos) {
3348 bh = bh->b_this_page;
3352 if (buffer_freed(bh)) {
3353 BUFFER_TRACE(bh, "freed: skip");
3356 if (!buffer_mapped(bh)) {
3357 BUFFER_TRACE(bh, "unmapped");
3358 ext4_get_block(inode, iblock, bh, 0);
3359 /* unmapped? It's a hole - nothing to do */
3360 if (!buffer_mapped(bh)) {
3361 BUFFER_TRACE(bh, "still unmapped");
3366 /* Ok, it's mapped. Make sure it's up-to-date */
3367 if (PageUptodate(page))
3368 set_buffer_uptodate(bh);
3370 if (!buffer_uptodate(bh)) {
3372 ll_rw_block(READ, 1, &bh);
3374 /* Uhhuh. Read error. Complain and punt. */
3375 if (!buffer_uptodate(bh))
3377 if (S_ISREG(inode->i_mode) &&
3378 ext4_encrypted_inode(inode)) {
3379 /* We expect the key to be set. */
3380 BUG_ON(!ext4_has_encryption_key(inode));
3381 BUG_ON(blocksize != PAGE_CACHE_SIZE);
3382 WARN_ON_ONCE(ext4_decrypt_one(inode, page));
3385 if (ext4_should_journal_data(inode)) {
3386 BUFFER_TRACE(bh, "get write access");
3387 err = ext4_journal_get_write_access(handle, bh);
3391 zero_user(page, offset, length);
3392 BUFFER_TRACE(bh, "zeroed end of block");
3394 if (ext4_should_journal_data(inode)) {
3395 err = ext4_handle_dirty_metadata(handle, inode, bh);
3398 mark_buffer_dirty(bh);
3399 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3400 err = ext4_jbd2_file_inode(handle, inode);
3405 page_cache_release(page);
3410 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3411 * starting from file offset 'from'. The range to be zero'd must
3412 * be contained with in one block. If the specified range exceeds
3413 * the end of the block it will be shortened to end of the block
3414 * that cooresponds to 'from'
3416 static int ext4_block_zero_page_range(handle_t *handle,
3417 struct address_space *mapping, loff_t from, loff_t length)
3419 struct inode *inode = mapping->host;
3420 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3421 unsigned blocksize = inode->i_sb->s_blocksize;
3422 unsigned max = blocksize - (offset & (blocksize - 1));
3425 * correct length if it does not fall between
3426 * 'from' and the end of the block
3428 if (length > max || length < 0)
3432 return dax_zero_page_range(inode, from, length, ext4_get_block);
3433 return __ext4_block_zero_page_range(handle, mapping, from, length);
3437 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3438 * up to the end of the block which corresponds to `from'.
3439 * This required during truncate. We need to physically zero the tail end
3440 * of that block so it doesn't yield old data if the file is later grown.
3442 static int ext4_block_truncate_page(handle_t *handle,
3443 struct address_space *mapping, loff_t from)
3445 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3448 struct inode *inode = mapping->host;
3450 blocksize = inode->i_sb->s_blocksize;
3451 length = blocksize - (offset & (blocksize - 1));
3453 return ext4_block_zero_page_range(handle, mapping, from, length);
3456 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3457 loff_t lstart, loff_t length)
3459 struct super_block *sb = inode->i_sb;
3460 struct address_space *mapping = inode->i_mapping;
3461 unsigned partial_start, partial_end;
3462 ext4_fsblk_t start, end;
3463 loff_t byte_end = (lstart + length - 1);
3466 partial_start = lstart & (sb->s_blocksize - 1);
3467 partial_end = byte_end & (sb->s_blocksize - 1);
3469 start = lstart >> sb->s_blocksize_bits;
3470 end = byte_end >> sb->s_blocksize_bits;
3472 /* Handle partial zero within the single block */
3474 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3475 err = ext4_block_zero_page_range(handle, mapping,
3479 /* Handle partial zero out on the start of the range */
3480 if (partial_start) {
3481 err = ext4_block_zero_page_range(handle, mapping,
3482 lstart, sb->s_blocksize);
3486 /* Handle partial zero out on the end of the range */
3487 if (partial_end != sb->s_blocksize - 1)
3488 err = ext4_block_zero_page_range(handle, mapping,
3489 byte_end - partial_end,
3494 int ext4_can_truncate(struct inode *inode)
3496 if (S_ISREG(inode->i_mode))
3498 if (S_ISDIR(inode->i_mode))
3500 if (S_ISLNK(inode->i_mode))
3501 return !ext4_inode_is_fast_symlink(inode);
3506 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3507 * associated with the given offset and length
3509 * @inode: File inode
3510 * @offset: The offset where the hole will begin
3511 * @len: The length of the hole
3513 * Returns: 0 on success or negative on failure
3516 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3518 struct super_block *sb = inode->i_sb;
3519 ext4_lblk_t first_block, stop_block;
3520 struct address_space *mapping = inode->i_mapping;
3521 loff_t first_block_offset, last_block_offset;
3523 unsigned int credits;
3526 if (!S_ISREG(inode->i_mode))
3529 trace_ext4_punch_hole(inode, offset, length, 0);
3532 * Write out all dirty pages to avoid race conditions
3533 * Then release them.
3535 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3536 ret = filemap_write_and_wait_range(mapping, offset,
3537 offset + length - 1);
3542 mutex_lock(&inode->i_mutex);
3544 /* No need to punch hole beyond i_size */
3545 if (offset >= inode->i_size)
3549 * If the hole extends beyond i_size, set the hole
3550 * to end after the page that contains i_size
3552 if (offset + length > inode->i_size) {
3553 length = inode->i_size +
3554 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3558 if (offset & (sb->s_blocksize - 1) ||
3559 (offset + length) & (sb->s_blocksize - 1)) {
3561 * Attach jinode to inode for jbd2 if we do any zeroing of
3564 ret = ext4_inode_attach_jinode(inode);
3570 first_block_offset = round_up(offset, sb->s_blocksize);
3571 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3573 /* Now release the pages and zero block aligned part of pages*/
3574 if (last_block_offset > first_block_offset)
3575 truncate_pagecache_range(inode, first_block_offset,
3578 /* Wait all existing dio workers, newcomers will block on i_mutex */
3579 ext4_inode_block_unlocked_dio(inode);
3580 inode_dio_wait(inode);
3582 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3583 credits = ext4_writepage_trans_blocks(inode);
3585 credits = ext4_blocks_for_truncate(inode);
3586 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3587 if (IS_ERR(handle)) {
3588 ret = PTR_ERR(handle);
3589 ext4_std_error(sb, ret);
3593 ret = ext4_zero_partial_blocks(handle, inode, offset,
3598 first_block = (offset + sb->s_blocksize - 1) >>
3599 EXT4_BLOCK_SIZE_BITS(sb);
3600 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3602 /* If there are no blocks to remove, return now */
3603 if (first_block >= stop_block)
3606 down_write(&EXT4_I(inode)->i_data_sem);
3607 ext4_discard_preallocations(inode);
3609 ret = ext4_es_remove_extent(inode, first_block,
3610 stop_block - first_block);
3612 up_write(&EXT4_I(inode)->i_data_sem);
3616 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3617 ret = ext4_ext_remove_space(inode, first_block,
3620 ret = ext4_ind_remove_space(handle, inode, first_block,
3623 up_write(&EXT4_I(inode)->i_data_sem);
3625 ext4_handle_sync(handle);
3627 /* Now release the pages again to reduce race window */
3628 if (last_block_offset > first_block_offset)
3629 truncate_pagecache_range(inode, first_block_offset,
3632 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3633 ext4_mark_inode_dirty(handle, inode);
3635 ext4_journal_stop(handle);
3637 ext4_inode_resume_unlocked_dio(inode);
3639 mutex_unlock(&inode->i_mutex);
3643 int ext4_inode_attach_jinode(struct inode *inode)
3645 struct ext4_inode_info *ei = EXT4_I(inode);
3646 struct jbd2_inode *jinode;
3648 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3651 jinode = jbd2_alloc_inode(GFP_KERNEL);
3652 spin_lock(&inode->i_lock);
3655 spin_unlock(&inode->i_lock);
3658 ei->jinode = jinode;
3659 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3662 spin_unlock(&inode->i_lock);
3663 if (unlikely(jinode != NULL))
3664 jbd2_free_inode(jinode);
3671 * We block out ext4_get_block() block instantiations across the entire
3672 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3673 * simultaneously on behalf of the same inode.
3675 * As we work through the truncate and commit bits of it to the journal there
3676 * is one core, guiding principle: the file's tree must always be consistent on
3677 * disk. We must be able to restart the truncate after a crash.
3679 * The file's tree may be transiently inconsistent in memory (although it
3680 * probably isn't), but whenever we close off and commit a journal transaction,
3681 * the contents of (the filesystem + the journal) must be consistent and
3682 * restartable. It's pretty simple, really: bottom up, right to left (although
3683 * left-to-right works OK too).
3685 * Note that at recovery time, journal replay occurs *before* the restart of
3686 * truncate against the orphan inode list.
3688 * The committed inode has the new, desired i_size (which is the same as
3689 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3690 * that this inode's truncate did not complete and it will again call
3691 * ext4_truncate() to have another go. So there will be instantiated blocks
3692 * to the right of the truncation point in a crashed ext4 filesystem. But
3693 * that's fine - as long as they are linked from the inode, the post-crash
3694 * ext4_truncate() run will find them and release them.
3696 void ext4_truncate(struct inode *inode)
3698 struct ext4_inode_info *ei = EXT4_I(inode);
3699 unsigned int credits;
3701 struct address_space *mapping = inode->i_mapping;
3704 * There is a possibility that we're either freeing the inode
3705 * or it's a completely new inode. In those cases we might not
3706 * have i_mutex locked because it's not necessary.
3708 if (!(inode->i_state & (I_NEW|I_FREEING)))
3709 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3710 trace_ext4_truncate_enter(inode);
3712 if (!ext4_can_truncate(inode))
3715 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3717 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3718 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3720 if (ext4_has_inline_data(inode)) {
3723 ext4_inline_data_truncate(inode, &has_inline);
3728 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3729 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3730 if (ext4_inode_attach_jinode(inode) < 0)
3734 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3735 credits = ext4_writepage_trans_blocks(inode);
3737 credits = ext4_blocks_for_truncate(inode);
3739 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3740 if (IS_ERR(handle)) {
3741 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3745 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3746 ext4_block_truncate_page(handle, mapping, inode->i_size);
3749 * We add the inode to the orphan list, so that if this
3750 * truncate spans multiple transactions, and we crash, we will
3751 * resume the truncate when the filesystem recovers. It also
3752 * marks the inode dirty, to catch the new size.
3754 * Implication: the file must always be in a sane, consistent
3755 * truncatable state while each transaction commits.
3757 if (ext4_orphan_add(handle, inode))
3760 down_write(&EXT4_I(inode)->i_data_sem);
3762 ext4_discard_preallocations(inode);
3764 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3765 ext4_ext_truncate(handle, inode);
3767 ext4_ind_truncate(handle, inode);
3769 up_write(&ei->i_data_sem);
3772 ext4_handle_sync(handle);
3776 * If this was a simple ftruncate() and the file will remain alive,
3777 * then we need to clear up the orphan record which we created above.
3778 * However, if this was a real unlink then we were called by
3779 * ext4_evict_inode(), and we allow that function to clean up the
3780 * orphan info for us.
3783 ext4_orphan_del(handle, inode);
3785 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3786 ext4_mark_inode_dirty(handle, inode);
3787 ext4_journal_stop(handle);
3789 trace_ext4_truncate_exit(inode);
3793 * ext4_get_inode_loc returns with an extra refcount against the inode's
3794 * underlying buffer_head on success. If 'in_mem' is true, we have all
3795 * data in memory that is needed to recreate the on-disk version of this
3798 static int __ext4_get_inode_loc(struct inode *inode,
3799 struct ext4_iloc *iloc, int in_mem)
3801 struct ext4_group_desc *gdp;
3802 struct buffer_head *bh;
3803 struct super_block *sb = inode->i_sb;
3805 int inodes_per_block, inode_offset;
3808 if (!ext4_valid_inum(sb, inode->i_ino))
3811 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3812 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3817 * Figure out the offset within the block group inode table
3819 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3820 inode_offset = ((inode->i_ino - 1) %
3821 EXT4_INODES_PER_GROUP(sb));
3822 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3823 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3825 bh = sb_getblk(sb, block);
3828 if (!buffer_uptodate(bh)) {
3832 * If the buffer has the write error flag, we have failed
3833 * to write out another inode in the same block. In this
3834 * case, we don't have to read the block because we may
3835 * read the old inode data successfully.
3837 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3838 set_buffer_uptodate(bh);
3840 if (buffer_uptodate(bh)) {
3841 /* someone brought it uptodate while we waited */
3847 * If we have all information of the inode in memory and this
3848 * is the only valid inode in the block, we need not read the
3852 struct buffer_head *bitmap_bh;
3855 start = inode_offset & ~(inodes_per_block - 1);
3857 /* Is the inode bitmap in cache? */
3858 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3859 if (unlikely(!bitmap_bh))
3863 * If the inode bitmap isn't in cache then the
3864 * optimisation may end up performing two reads instead
3865 * of one, so skip it.
3867 if (!buffer_uptodate(bitmap_bh)) {
3871 for (i = start; i < start + inodes_per_block; i++) {
3872 if (i == inode_offset)
3874 if (ext4_test_bit(i, bitmap_bh->b_data))
3878 if (i == start + inodes_per_block) {
3879 /* all other inodes are free, so skip I/O */
3880 memset(bh->b_data, 0, bh->b_size);
3881 set_buffer_uptodate(bh);
3889 * If we need to do any I/O, try to pre-readahead extra
3890 * blocks from the inode table.
3892 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3893 ext4_fsblk_t b, end, table;
3895 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
3897 table = ext4_inode_table(sb, gdp);
3898 /* s_inode_readahead_blks is always a power of 2 */
3899 b = block & ~((ext4_fsblk_t) ra_blks - 1);
3903 num = EXT4_INODES_PER_GROUP(sb);
3904 if (ext4_has_group_desc_csum(sb))
3905 num -= ext4_itable_unused_count(sb, gdp);
3906 table += num / inodes_per_block;
3910 sb_breadahead(sb, b++);
3914 * There are other valid inodes in the buffer, this inode
3915 * has in-inode xattrs, or we don't have this inode in memory.
3916 * Read the block from disk.
3918 trace_ext4_load_inode(inode);
3920 bh->b_end_io = end_buffer_read_sync;
3921 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3923 if (!buffer_uptodate(bh)) {
3924 EXT4_ERROR_INODE_BLOCK(inode, block,
3925 "unable to read itable block");
3935 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3937 /* We have all inode data except xattrs in memory here. */
3938 return __ext4_get_inode_loc(inode, iloc,
3939 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3942 void ext4_set_inode_flags(struct inode *inode)
3944 unsigned int flags = EXT4_I(inode)->i_flags;
3945 unsigned int new_fl = 0;
3947 if (flags & EXT4_SYNC_FL)
3949 if (flags & EXT4_APPEND_FL)
3951 if (flags & EXT4_IMMUTABLE_FL)
3952 new_fl |= S_IMMUTABLE;
3953 if (flags & EXT4_NOATIME_FL)
3954 new_fl |= S_NOATIME;
3955 if (flags & EXT4_DIRSYNC_FL)
3956 new_fl |= S_DIRSYNC;
3957 if (test_opt(inode->i_sb, DAX))
3959 inode_set_flags(inode, new_fl,
3960 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
3963 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3964 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3966 unsigned int vfs_fl;
3967 unsigned long old_fl, new_fl;
3970 vfs_fl = ei->vfs_inode.i_flags;
3971 old_fl = ei->i_flags;
3972 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3973 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3975 if (vfs_fl & S_SYNC)
3976 new_fl |= EXT4_SYNC_FL;
3977 if (vfs_fl & S_APPEND)
3978 new_fl |= EXT4_APPEND_FL;
3979 if (vfs_fl & S_IMMUTABLE)
3980 new_fl |= EXT4_IMMUTABLE_FL;
3981 if (vfs_fl & S_NOATIME)
3982 new_fl |= EXT4_NOATIME_FL;
3983 if (vfs_fl & S_DIRSYNC)
3984 new_fl |= EXT4_DIRSYNC_FL;
3985 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3988 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3989 struct ext4_inode_info *ei)
3992 struct inode *inode = &(ei->vfs_inode);
3993 struct super_block *sb = inode->i_sb;
3995 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3996 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3997 /* we are using combined 48 bit field */
3998 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3999 le32_to_cpu(raw_inode->i_blocks_lo);
4000 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4001 /* i_blocks represent file system block size */
4002 return i_blocks << (inode->i_blkbits - 9);
4007 return le32_to_cpu(raw_inode->i_blocks_lo);
4011 static inline void ext4_iget_extra_inode(struct inode *inode,
4012 struct ext4_inode *raw_inode,
4013 struct ext4_inode_info *ei)
4015 __le32 *magic = (void *)raw_inode +
4016 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4017 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4018 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4019 ext4_find_inline_data_nolock(inode);
4021 EXT4_I(inode)->i_inline_off = 0;
4024 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4026 struct ext4_iloc iloc;
4027 struct ext4_inode *raw_inode;
4028 struct ext4_inode_info *ei;
4029 struct inode *inode;
4030 journal_t *journal = EXT4_SB(sb)->s_journal;
4036 inode = iget_locked(sb, ino);
4038 return ERR_PTR(-ENOMEM);
4039 if (!(inode->i_state & I_NEW))
4045 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4048 raw_inode = ext4_raw_inode(&iloc);
4050 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4051 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4052 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4053 EXT4_INODE_SIZE(inode->i_sb)) {
4054 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4055 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4056 EXT4_INODE_SIZE(inode->i_sb));
4061 ei->i_extra_isize = 0;
4063 /* Precompute checksum seed for inode metadata */
4064 if (ext4_has_metadata_csum(sb)) {
4065 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4067 __le32 inum = cpu_to_le32(inode->i_ino);
4068 __le32 gen = raw_inode->i_generation;
4069 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4071 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4075 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4076 EXT4_ERROR_INODE(inode, "checksum invalid");
4081 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4082 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4083 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4084 if (!(test_opt(inode->i_sb, NO_UID32))) {
4085 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4086 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4088 i_uid_write(inode, i_uid);
4089 i_gid_write(inode, i_gid);
4090 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4092 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4093 ei->i_inline_off = 0;
4094 ei->i_dir_start_lookup = 0;
4095 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4096 /* We now have enough fields to check if the inode was active or not.
4097 * This is needed because nfsd might try to access dead inodes
4098 * the test is that same one that e2fsck uses
4099 * NeilBrown 1999oct15
4101 if (inode->i_nlink == 0) {
4102 if ((inode->i_mode == 0 ||
4103 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4104 ino != EXT4_BOOT_LOADER_INO) {
4105 /* this inode is deleted */
4109 /* The only unlinked inodes we let through here have
4110 * valid i_mode and are being read by the orphan
4111 * recovery code: that's fine, we're about to complete
4112 * the process of deleting those.
4113 * OR it is the EXT4_BOOT_LOADER_INO which is
4114 * not initialized on a new filesystem. */
4116 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4117 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4118 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4119 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4121 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4122 inode->i_size = ext4_isize(raw_inode);
4123 ei->i_disksize = inode->i_size;
4125 ei->i_reserved_quota = 0;
4127 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4128 ei->i_block_group = iloc.block_group;
4129 ei->i_last_alloc_group = ~0;
4131 * NOTE! The in-memory inode i_data array is in little-endian order
4132 * even on big-endian machines: we do NOT byteswap the block numbers!
4134 for (block = 0; block < EXT4_N_BLOCKS; block++)
4135 ei->i_data[block] = raw_inode->i_block[block];
4136 INIT_LIST_HEAD(&ei->i_orphan);
4139 * Set transaction id's of transactions that have to be committed
4140 * to finish f[data]sync. We set them to currently running transaction
4141 * as we cannot be sure that the inode or some of its metadata isn't
4142 * part of the transaction - the inode could have been reclaimed and
4143 * now it is reread from disk.
4146 transaction_t *transaction;
4149 read_lock(&journal->j_state_lock);
4150 if (journal->j_running_transaction)
4151 transaction = journal->j_running_transaction;
4153 transaction = journal->j_committing_transaction;
4155 tid = transaction->t_tid;
4157 tid = journal->j_commit_sequence;
4158 read_unlock(&journal->j_state_lock);
4159 ei->i_sync_tid = tid;
4160 ei->i_datasync_tid = tid;
4163 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4164 if (ei->i_extra_isize == 0) {
4165 /* The extra space is currently unused. Use it. */
4166 ei->i_extra_isize = sizeof(struct ext4_inode) -
4167 EXT4_GOOD_OLD_INODE_SIZE;
4169 ext4_iget_extra_inode(inode, raw_inode, ei);
4173 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4174 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4175 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4176 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4178 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4179 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4180 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4181 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4183 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4188 if (ei->i_file_acl &&
4189 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4190 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4194 } else if (!ext4_has_inline_data(inode)) {
4195 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4196 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4197 (S_ISLNK(inode->i_mode) &&
4198 !ext4_inode_is_fast_symlink(inode))))
4199 /* Validate extent which is part of inode */
4200 ret = ext4_ext_check_inode(inode);
4201 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4202 (S_ISLNK(inode->i_mode) &&
4203 !ext4_inode_is_fast_symlink(inode))) {
4204 /* Validate block references which are part of inode */
4205 ret = ext4_ind_check_inode(inode);
4211 if (S_ISREG(inode->i_mode)) {
4212 inode->i_op = &ext4_file_inode_operations;
4213 inode->i_fop = &ext4_file_operations;
4214 ext4_set_aops(inode);
4215 } else if (S_ISDIR(inode->i_mode)) {
4216 inode->i_op = &ext4_dir_inode_operations;
4217 inode->i_fop = &ext4_dir_operations;
4218 } else if (S_ISLNK(inode->i_mode)) {
4219 if (ext4_encrypted_inode(inode)) {
4220 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4221 ext4_set_aops(inode);
4222 } else if (ext4_inode_is_fast_symlink(inode)) {
4223 inode->i_link = (char *)ei->i_data;
4224 inode->i_op = &ext4_fast_symlink_inode_operations;
4225 nd_terminate_link(ei->i_data, inode->i_size,
4226 sizeof(ei->i_data) - 1);
4228 inode->i_op = &ext4_symlink_inode_operations;
4229 ext4_set_aops(inode);
4231 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4232 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4233 inode->i_op = &ext4_special_inode_operations;
4234 if (raw_inode->i_block[0])
4235 init_special_inode(inode, inode->i_mode,
4236 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4238 init_special_inode(inode, inode->i_mode,
4239 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4240 } else if (ino == EXT4_BOOT_LOADER_INO) {
4241 make_bad_inode(inode);
4244 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4248 ext4_set_inode_flags(inode);
4249 unlock_new_inode(inode);
4255 return ERR_PTR(ret);
4258 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4260 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4261 return ERR_PTR(-EIO);
4262 return ext4_iget(sb, ino);
4265 static int ext4_inode_blocks_set(handle_t *handle,
4266 struct ext4_inode *raw_inode,
4267 struct ext4_inode_info *ei)
4269 struct inode *inode = &(ei->vfs_inode);
4270 u64 i_blocks = inode->i_blocks;
4271 struct super_block *sb = inode->i_sb;
4273 if (i_blocks <= ~0U) {
4275 * i_blocks can be represented in a 32 bit variable
4276 * as multiple of 512 bytes
4278 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4279 raw_inode->i_blocks_high = 0;
4280 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4283 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4286 if (i_blocks <= 0xffffffffffffULL) {
4288 * i_blocks can be represented in a 48 bit variable
4289 * as multiple of 512 bytes
4291 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4292 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4293 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4295 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4296 /* i_block is stored in file system block size */
4297 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4298 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4299 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4304 struct other_inode {
4305 unsigned long orig_ino;
4306 struct ext4_inode *raw_inode;
4309 static int other_inode_match(struct inode * inode, unsigned long ino,
4312 struct other_inode *oi = (struct other_inode *) data;
4314 if ((inode->i_ino != ino) ||
4315 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4316 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4317 ((inode->i_state & I_DIRTY_TIME) == 0))
4319 spin_lock(&inode->i_lock);
4320 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4321 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4322 (inode->i_state & I_DIRTY_TIME)) {
4323 struct ext4_inode_info *ei = EXT4_I(inode);
4325 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4326 spin_unlock(&inode->i_lock);
4328 spin_lock(&ei->i_raw_lock);
4329 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4330 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4331 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4332 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4333 spin_unlock(&ei->i_raw_lock);
4334 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4337 spin_unlock(&inode->i_lock);
4342 * Opportunistically update the other time fields for other inodes in
4343 * the same inode table block.
4345 static void ext4_update_other_inodes_time(struct super_block *sb,
4346 unsigned long orig_ino, char *buf)
4348 struct other_inode oi;
4350 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4351 int inode_size = EXT4_INODE_SIZE(sb);
4353 oi.orig_ino = orig_ino;
4354 ino = (orig_ino & ~(inodes_per_block - 1)) + 1;
4355 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4356 if (ino == orig_ino)
4358 oi.raw_inode = (struct ext4_inode *) buf;
4359 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4364 * Post the struct inode info into an on-disk inode location in the
4365 * buffer-cache. This gobbles the caller's reference to the
4366 * buffer_head in the inode location struct.
4368 * The caller must have write access to iloc->bh.
4370 static int ext4_do_update_inode(handle_t *handle,
4371 struct inode *inode,
4372 struct ext4_iloc *iloc)
4374 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4375 struct ext4_inode_info *ei = EXT4_I(inode);
4376 struct buffer_head *bh = iloc->bh;
4377 struct super_block *sb = inode->i_sb;
4378 int err = 0, rc, block;
4379 int need_datasync = 0, set_large_file = 0;
4383 spin_lock(&ei->i_raw_lock);
4385 /* For fields not tracked in the in-memory inode,
4386 * initialise them to zero for new inodes. */
4387 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4388 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4390 ext4_get_inode_flags(ei);
4391 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4392 i_uid = i_uid_read(inode);
4393 i_gid = i_gid_read(inode);
4394 if (!(test_opt(inode->i_sb, NO_UID32))) {
4395 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4396 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4398 * Fix up interoperability with old kernels. Otherwise, old inodes get
4399 * re-used with the upper 16 bits of the uid/gid intact
4402 raw_inode->i_uid_high =
4403 cpu_to_le16(high_16_bits(i_uid));
4404 raw_inode->i_gid_high =
4405 cpu_to_le16(high_16_bits(i_gid));
4407 raw_inode->i_uid_high = 0;
4408 raw_inode->i_gid_high = 0;
4411 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4412 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4413 raw_inode->i_uid_high = 0;
4414 raw_inode->i_gid_high = 0;
4416 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4418 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4419 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4420 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4421 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4423 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4425 spin_unlock(&ei->i_raw_lock);
4428 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4429 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4430 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4431 raw_inode->i_file_acl_high =
4432 cpu_to_le16(ei->i_file_acl >> 32);
4433 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4434 if (ei->i_disksize != ext4_isize(raw_inode)) {
4435 ext4_isize_set(raw_inode, ei->i_disksize);
4438 if (ei->i_disksize > 0x7fffffffULL) {
4439 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4440 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4441 EXT4_SB(sb)->s_es->s_rev_level ==
4442 cpu_to_le32(EXT4_GOOD_OLD_REV))
4445 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4446 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4447 if (old_valid_dev(inode->i_rdev)) {
4448 raw_inode->i_block[0] =
4449 cpu_to_le32(old_encode_dev(inode->i_rdev));
4450 raw_inode->i_block[1] = 0;
4452 raw_inode->i_block[0] = 0;
4453 raw_inode->i_block[1] =
4454 cpu_to_le32(new_encode_dev(inode->i_rdev));
4455 raw_inode->i_block[2] = 0;
4457 } else if (!ext4_has_inline_data(inode)) {
4458 for (block = 0; block < EXT4_N_BLOCKS; block++)
4459 raw_inode->i_block[block] = ei->i_data[block];
4462 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4463 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4464 if (ei->i_extra_isize) {
4465 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4466 raw_inode->i_version_hi =
4467 cpu_to_le32(inode->i_version >> 32);
4468 raw_inode->i_extra_isize =
4469 cpu_to_le16(ei->i_extra_isize);
4472 ext4_inode_csum_set(inode, raw_inode, ei);
4473 spin_unlock(&ei->i_raw_lock);
4474 if (inode->i_sb->s_flags & MS_LAZYTIME)
4475 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4478 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4479 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4482 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4483 if (set_large_file) {
4484 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4485 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4488 ext4_update_dynamic_rev(sb);
4489 EXT4_SET_RO_COMPAT_FEATURE(sb,
4490 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4491 ext4_handle_sync(handle);
4492 err = ext4_handle_dirty_super(handle, sb);
4494 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4497 ext4_std_error(inode->i_sb, err);
4502 * ext4_write_inode()
4504 * We are called from a few places:
4506 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4507 * Here, there will be no transaction running. We wait for any running
4508 * transaction to commit.
4510 * - Within flush work (sys_sync(), kupdate and such).
4511 * We wait on commit, if told to.
4513 * - Within iput_final() -> write_inode_now()
4514 * We wait on commit, if told to.
4516 * In all cases it is actually safe for us to return without doing anything,
4517 * because the inode has been copied into a raw inode buffer in
4518 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4521 * Note that we are absolutely dependent upon all inode dirtiers doing the
4522 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4523 * which we are interested.
4525 * It would be a bug for them to not do this. The code:
4527 * mark_inode_dirty(inode)
4529 * inode->i_size = expr;
4531 * is in error because write_inode() could occur while `stuff()' is running,
4532 * and the new i_size will be lost. Plus the inode will no longer be on the
4533 * superblock's dirty inode list.
4535 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4539 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4542 if (EXT4_SB(inode->i_sb)->s_journal) {
4543 if (ext4_journal_current_handle()) {
4544 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4550 * No need to force transaction in WB_SYNC_NONE mode. Also
4551 * ext4_sync_fs() will force the commit after everything is
4554 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4557 err = ext4_force_commit(inode->i_sb);
4559 struct ext4_iloc iloc;
4561 err = __ext4_get_inode_loc(inode, &iloc, 0);
4565 * sync(2) will flush the whole buffer cache. No need to do
4566 * it here separately for each inode.
4568 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4569 sync_dirty_buffer(iloc.bh);
4570 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4571 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4572 "IO error syncing inode");
4581 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4582 * buffers that are attached to a page stradding i_size and are undergoing
4583 * commit. In that case we have to wait for commit to finish and try again.
4585 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4589 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4590 tid_t commit_tid = 0;
4593 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4595 * All buffers in the last page remain valid? Then there's nothing to
4596 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4599 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4602 page = find_lock_page(inode->i_mapping,
4603 inode->i_size >> PAGE_CACHE_SHIFT);
4606 ret = __ext4_journalled_invalidatepage(page, offset,
4607 PAGE_CACHE_SIZE - offset);
4609 page_cache_release(page);
4613 read_lock(&journal->j_state_lock);
4614 if (journal->j_committing_transaction)
4615 commit_tid = journal->j_committing_transaction->t_tid;
4616 read_unlock(&journal->j_state_lock);
4618 jbd2_log_wait_commit(journal, commit_tid);
4625 * Called from notify_change.
4627 * We want to trap VFS attempts to truncate the file as soon as
4628 * possible. In particular, we want to make sure that when the VFS
4629 * shrinks i_size, we put the inode on the orphan list and modify
4630 * i_disksize immediately, so that during the subsequent flushing of
4631 * dirty pages and freeing of disk blocks, we can guarantee that any
4632 * commit will leave the blocks being flushed in an unused state on
4633 * disk. (On recovery, the inode will get truncated and the blocks will
4634 * be freed, so we have a strong guarantee that no future commit will
4635 * leave these blocks visible to the user.)
4637 * Another thing we have to assure is that if we are in ordered mode
4638 * and inode is still attached to the committing transaction, we must
4639 * we start writeout of all the dirty pages which are being truncated.
4640 * This way we are sure that all the data written in the previous
4641 * transaction are already on disk (truncate waits for pages under
4644 * Called with inode->i_mutex down.
4646 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4648 struct inode *inode = d_inode(dentry);
4651 const unsigned int ia_valid = attr->ia_valid;
4653 error = inode_change_ok(inode, attr);
4657 if (is_quota_modification(inode, attr))
4658 dquot_initialize(inode);
4659 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4660 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4663 /* (user+group)*(old+new) structure, inode write (sb,
4664 * inode block, ? - but truncate inode update has it) */
4665 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4666 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4667 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4668 if (IS_ERR(handle)) {
4669 error = PTR_ERR(handle);
4672 error = dquot_transfer(inode, attr);
4674 ext4_journal_stop(handle);
4677 /* Update corresponding info in inode so that everything is in
4678 * one transaction */
4679 if (attr->ia_valid & ATTR_UID)
4680 inode->i_uid = attr->ia_uid;
4681 if (attr->ia_valid & ATTR_GID)
4682 inode->i_gid = attr->ia_gid;
4683 error = ext4_mark_inode_dirty(handle, inode);
4684 ext4_journal_stop(handle);
4687 if (attr->ia_valid & ATTR_SIZE) {
4689 loff_t oldsize = inode->i_size;
4690 int shrink = (attr->ia_size <= inode->i_size);
4692 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4693 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4695 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4698 if (!S_ISREG(inode->i_mode))
4701 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4702 inode_inc_iversion(inode);
4704 if (ext4_should_order_data(inode) &&
4705 (attr->ia_size < inode->i_size)) {
4706 error = ext4_begin_ordered_truncate(inode,
4711 if (attr->ia_size != inode->i_size) {
4712 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4713 if (IS_ERR(handle)) {
4714 error = PTR_ERR(handle);
4717 if (ext4_handle_valid(handle) && shrink) {
4718 error = ext4_orphan_add(handle, inode);
4721 down_write(&EXT4_I(inode)->i_data_sem);
4722 EXT4_I(inode)->i_disksize = attr->ia_size;
4723 rc = ext4_mark_inode_dirty(handle, inode);
4727 * We have to update i_size under i_data_sem together
4728 * with i_disksize to avoid races with writeback code
4729 * running ext4_wb_update_i_disksize().
4732 i_size_write(inode, attr->ia_size);
4733 up_write(&EXT4_I(inode)->i_data_sem);
4734 ext4_journal_stop(handle);
4737 ext4_orphan_del(NULL, inode);
4742 pagecache_isize_extended(inode, oldsize, inode->i_size);
4745 * Blocks are going to be removed from the inode. Wait
4746 * for dio in flight. Temporarily disable
4747 * dioread_nolock to prevent livelock.
4750 if (!ext4_should_journal_data(inode)) {
4751 ext4_inode_block_unlocked_dio(inode);
4752 inode_dio_wait(inode);
4753 ext4_inode_resume_unlocked_dio(inode);
4755 ext4_wait_for_tail_page_commit(inode);
4758 * Truncate pagecache after we've waited for commit
4759 * in data=journal mode to make pages freeable.
4761 truncate_pagecache(inode, inode->i_size);
4763 ext4_truncate(inode);
4767 setattr_copy(inode, attr);
4768 mark_inode_dirty(inode);
4772 * If the call to ext4_truncate failed to get a transaction handle at
4773 * all, we need to clean up the in-core orphan list manually.
4775 if (orphan && inode->i_nlink)
4776 ext4_orphan_del(NULL, inode);
4778 if (!rc && (ia_valid & ATTR_MODE))
4779 rc = posix_acl_chmod(inode, inode->i_mode);
4782 ext4_std_error(inode->i_sb, error);
4788 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4791 struct inode *inode;
4792 unsigned long long delalloc_blocks;
4794 inode = d_inode(dentry);
4795 generic_fillattr(inode, stat);
4798 * If there is inline data in the inode, the inode will normally not
4799 * have data blocks allocated (it may have an external xattr block).
4800 * Report at least one sector for such files, so tools like tar, rsync,
4801 * others doen't incorrectly think the file is completely sparse.
4803 if (unlikely(ext4_has_inline_data(inode)))
4804 stat->blocks += (stat->size + 511) >> 9;
4807 * We can't update i_blocks if the block allocation is delayed
4808 * otherwise in the case of system crash before the real block
4809 * allocation is done, we will have i_blocks inconsistent with
4810 * on-disk file blocks.
4811 * We always keep i_blocks updated together with real
4812 * allocation. But to not confuse with user, stat
4813 * will return the blocks that include the delayed allocation
4814 * blocks for this file.
4816 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4817 EXT4_I(inode)->i_reserved_data_blocks);
4818 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
4822 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4825 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4826 return ext4_ind_trans_blocks(inode, lblocks);
4827 return ext4_ext_index_trans_blocks(inode, pextents);
4831 * Account for index blocks, block groups bitmaps and block group
4832 * descriptor blocks if modify datablocks and index blocks
4833 * worse case, the indexs blocks spread over different block groups
4835 * If datablocks are discontiguous, they are possible to spread over
4836 * different block groups too. If they are contiguous, with flexbg,
4837 * they could still across block group boundary.
4839 * Also account for superblock, inode, quota and xattr blocks
4841 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4844 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4850 * How many index blocks need to touch to map @lblocks logical blocks
4851 * to @pextents physical extents?
4853 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
4858 * Now let's see how many group bitmaps and group descriptors need
4861 groups = idxblocks + pextents;
4863 if (groups > ngroups)
4865 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4866 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4868 /* bitmaps and block group descriptor blocks */
4869 ret += groups + gdpblocks;
4871 /* Blocks for super block, inode, quota and xattr blocks */
4872 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4878 * Calculate the total number of credits to reserve to fit
4879 * the modification of a single pages into a single transaction,
4880 * which may include multiple chunks of block allocations.
4882 * This could be called via ext4_write_begin()
4884 * We need to consider the worse case, when
4885 * one new block per extent.
4887 int ext4_writepage_trans_blocks(struct inode *inode)
4889 int bpp = ext4_journal_blocks_per_page(inode);
4892 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
4894 /* Account for data blocks for journalled mode */
4895 if (ext4_should_journal_data(inode))
4901 * Calculate the journal credits for a chunk of data modification.
4903 * This is called from DIO, fallocate or whoever calling
4904 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4906 * journal buffers for data blocks are not included here, as DIO
4907 * and fallocate do no need to journal data buffers.
4909 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4911 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4915 * The caller must have previously called ext4_reserve_inode_write().
4916 * Give this, we know that the caller already has write access to iloc->bh.
4918 int ext4_mark_iloc_dirty(handle_t *handle,
4919 struct inode *inode, struct ext4_iloc *iloc)
4923 if (IS_I_VERSION(inode))
4924 inode_inc_iversion(inode);
4926 /* the do_update_inode consumes one bh->b_count */
4929 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4930 err = ext4_do_update_inode(handle, inode, iloc);
4936 * On success, We end up with an outstanding reference count against
4937 * iloc->bh. This _must_ be cleaned up later.
4941 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4942 struct ext4_iloc *iloc)
4946 err = ext4_get_inode_loc(inode, iloc);
4948 BUFFER_TRACE(iloc->bh, "get_write_access");
4949 err = ext4_journal_get_write_access(handle, iloc->bh);
4955 ext4_std_error(inode->i_sb, err);
4960 * Expand an inode by new_extra_isize bytes.
4961 * Returns 0 on success or negative error number on failure.
4963 static int ext4_expand_extra_isize(struct inode *inode,
4964 unsigned int new_extra_isize,
4965 struct ext4_iloc iloc,
4968 struct ext4_inode *raw_inode;
4969 struct ext4_xattr_ibody_header *header;
4971 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4974 raw_inode = ext4_raw_inode(&iloc);
4976 header = IHDR(inode, raw_inode);
4978 /* No extended attributes present */
4979 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4980 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4981 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4983 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4987 /* try to expand with EAs present */
4988 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4993 * What we do here is to mark the in-core inode as clean with respect to inode
4994 * dirtiness (it may still be data-dirty).
4995 * This means that the in-core inode may be reaped by prune_icache
4996 * without having to perform any I/O. This is a very good thing,
4997 * because *any* task may call prune_icache - even ones which
4998 * have a transaction open against a different journal.
5000 * Is this cheating? Not really. Sure, we haven't written the
5001 * inode out, but prune_icache isn't a user-visible syncing function.
5002 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5003 * we start and wait on commits.
5005 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5007 struct ext4_iloc iloc;
5008 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5009 static unsigned int mnt_count;
5013 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5014 err = ext4_reserve_inode_write(handle, inode, &iloc);
5015 if (ext4_handle_valid(handle) &&
5016 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5017 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5019 * We need extra buffer credits since we may write into EA block
5020 * with this same handle. If journal_extend fails, then it will
5021 * only result in a minor loss of functionality for that inode.
5022 * If this is felt to be critical, then e2fsck should be run to
5023 * force a large enough s_min_extra_isize.
5025 if ((jbd2_journal_extend(handle,
5026 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5027 ret = ext4_expand_extra_isize(inode,
5028 sbi->s_want_extra_isize,
5031 ext4_set_inode_state(inode,
5032 EXT4_STATE_NO_EXPAND);
5034 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5035 ext4_warning(inode->i_sb,
5036 "Unable to expand inode %lu. Delete"
5037 " some EAs or run e2fsck.",
5040 le16_to_cpu(sbi->s_es->s_mnt_count);
5046 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5051 * ext4_dirty_inode() is called from __mark_inode_dirty()
5053 * We're really interested in the case where a file is being extended.
5054 * i_size has been changed by generic_commit_write() and we thus need
5055 * to include the updated inode in the current transaction.
5057 * Also, dquot_alloc_block() will always dirty the inode when blocks
5058 * are allocated to the file.
5060 * If the inode is marked synchronous, we don't honour that here - doing
5061 * so would cause a commit on atime updates, which we don't bother doing.
5062 * We handle synchronous inodes at the highest possible level.
5064 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5065 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5066 * to copy into the on-disk inode structure are the timestamp files.
5068 void ext4_dirty_inode(struct inode *inode, int flags)
5072 if (flags == I_DIRTY_TIME)
5074 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5078 ext4_mark_inode_dirty(handle, inode);
5080 ext4_journal_stop(handle);
5087 * Bind an inode's backing buffer_head into this transaction, to prevent
5088 * it from being flushed to disk early. Unlike
5089 * ext4_reserve_inode_write, this leaves behind no bh reference and
5090 * returns no iloc structure, so the caller needs to repeat the iloc
5091 * lookup to mark the inode dirty later.
5093 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5095 struct ext4_iloc iloc;
5099 err = ext4_get_inode_loc(inode, &iloc);
5101 BUFFER_TRACE(iloc.bh, "get_write_access");
5102 err = jbd2_journal_get_write_access(handle, iloc.bh);
5104 err = ext4_handle_dirty_metadata(handle,
5110 ext4_std_error(inode->i_sb, err);
5115 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5122 * We have to be very careful here: changing a data block's
5123 * journaling status dynamically is dangerous. If we write a
5124 * data block to the journal, change the status and then delete
5125 * that block, we risk forgetting to revoke the old log record
5126 * from the journal and so a subsequent replay can corrupt data.
5127 * So, first we make sure that the journal is empty and that
5128 * nobody is changing anything.
5131 journal = EXT4_JOURNAL(inode);
5134 if (is_journal_aborted(journal))
5136 /* We have to allocate physical blocks for delalloc blocks
5137 * before flushing journal. otherwise delalloc blocks can not
5138 * be allocated any more. even more truncate on delalloc blocks
5139 * could trigger BUG by flushing delalloc blocks in journal.
5140 * There is no delalloc block in non-journal data mode.
5142 if (val && test_opt(inode->i_sb, DELALLOC)) {
5143 err = ext4_alloc_da_blocks(inode);
5148 /* Wait for all existing dio workers */
5149 ext4_inode_block_unlocked_dio(inode);
5150 inode_dio_wait(inode);
5152 jbd2_journal_lock_updates(journal);
5155 * OK, there are no updates running now, and all cached data is
5156 * synced to disk. We are now in a completely consistent state
5157 * which doesn't have anything in the journal, and we know that
5158 * no filesystem updates are running, so it is safe to modify
5159 * the inode's in-core data-journaling state flag now.
5163 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5165 err = jbd2_journal_flush(journal);
5167 jbd2_journal_unlock_updates(journal);
5168 ext4_inode_resume_unlocked_dio(inode);
5171 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5173 ext4_set_aops(inode);
5175 jbd2_journal_unlock_updates(journal);
5176 ext4_inode_resume_unlocked_dio(inode);
5178 /* Finally we can mark the inode as dirty. */
5180 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5182 return PTR_ERR(handle);
5184 err = ext4_mark_inode_dirty(handle, inode);
5185 ext4_handle_sync(handle);
5186 ext4_journal_stop(handle);
5187 ext4_std_error(inode->i_sb, err);
5192 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5194 return !buffer_mapped(bh);
5197 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5199 struct page *page = vmf->page;
5203 struct file *file = vma->vm_file;
5204 struct inode *inode = file_inode(file);
5205 struct address_space *mapping = inode->i_mapping;
5207 get_block_t *get_block;
5210 sb_start_pagefault(inode->i_sb);
5211 file_update_time(vma->vm_file);
5212 /* Delalloc case is easy... */
5213 if (test_opt(inode->i_sb, DELALLOC) &&
5214 !ext4_should_journal_data(inode) &&
5215 !ext4_nonda_switch(inode->i_sb)) {
5217 ret = __block_page_mkwrite(vma, vmf,
5218 ext4_da_get_block_prep);
5219 } while (ret == -ENOSPC &&
5220 ext4_should_retry_alloc(inode->i_sb, &retries));
5225 size = i_size_read(inode);
5226 /* Page got truncated from under us? */
5227 if (page->mapping != mapping || page_offset(page) > size) {
5229 ret = VM_FAULT_NOPAGE;
5233 if (page->index == size >> PAGE_CACHE_SHIFT)
5234 len = size & ~PAGE_CACHE_MASK;
5236 len = PAGE_CACHE_SIZE;
5238 * Return if we have all the buffers mapped. This avoids the need to do
5239 * journal_start/journal_stop which can block and take a long time
5241 if (page_has_buffers(page)) {
5242 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5244 ext4_bh_unmapped)) {
5245 /* Wait so that we don't change page under IO */
5246 wait_for_stable_page(page);
5247 ret = VM_FAULT_LOCKED;
5252 /* OK, we need to fill the hole... */
5253 if (ext4_should_dioread_nolock(inode))
5254 get_block = ext4_get_block_write;
5256 get_block = ext4_get_block;
5258 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5259 ext4_writepage_trans_blocks(inode));
5260 if (IS_ERR(handle)) {
5261 ret = VM_FAULT_SIGBUS;
5264 ret = __block_page_mkwrite(vma, vmf, get_block);
5265 if (!ret && ext4_should_journal_data(inode)) {
5266 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5267 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5269 ret = VM_FAULT_SIGBUS;
5270 ext4_journal_stop(handle);
5273 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5275 ext4_journal_stop(handle);
5276 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5279 ret = block_page_mkwrite_return(ret);
5281 sb_end_pagefault(inode->i_sb);