2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51 struct ext4_inode_info *ei)
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 csum_lo = raw->i_checksum_lo;
59 raw->i_checksum_lo = 0;
60 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
61 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
62 csum_hi = raw->i_checksum_hi;
63 raw->i_checksum_hi = 0;
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
67 EXT4_INODE_SIZE(inode->i_sb));
69 raw->i_checksum_lo = csum_lo;
70 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
71 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
72 raw->i_checksum_hi = csum_hi;
77 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
78 struct ext4_inode_info *ei)
80 __u32 provided, calculated;
82 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
83 cpu_to_le32(EXT4_OS_LINUX) ||
84 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
85 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
88 provided = le16_to_cpu(raw->i_checksum_lo);
89 calculated = ext4_inode_csum(inode, raw, ei);
90 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
91 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
92 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
96 return provided == calculated;
99 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
100 struct ext4_inode_info *ei)
104 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
105 cpu_to_le32(EXT4_OS_LINUX) ||
106 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
107 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
110 csum = ext4_inode_csum(inode, raw, ei);
111 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
112 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
113 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
114 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
117 static inline int ext4_begin_ordered_truncate(struct inode *inode,
120 trace_ext4_begin_ordered_truncate(inode, new_size);
122 * If jinode is zero, then we never opened the file for
123 * writing, so there's no need to call
124 * jbd2_journal_begin_ordered_truncate() since there's no
125 * outstanding writes we need to flush.
127 if (!EXT4_I(inode)->jinode)
129 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
130 EXT4_I(inode)->jinode,
134 static void ext4_invalidatepage(struct page *page, unsigned long offset);
135 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
136 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
137 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
138 struct inode *inode, struct page *page, loff_t from,
139 loff_t length, int flags);
142 * Test whether an inode is a fast symlink.
144 static int ext4_inode_is_fast_symlink(struct inode *inode)
146 int ea_blocks = EXT4_I(inode)->i_file_acl ?
147 (inode->i_sb->s_blocksize >> 9) : 0;
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 ext4_ioend_wait(inode);
190 if (inode->i_nlink) {
192 * When journalling data dirty buffers are tracked only in the
193 * journal. So although mm thinks everything is clean and
194 * ready for reaping the inode might still have some pages to
195 * write in the running transaction or waiting to be
196 * checkpointed. Thus calling jbd2_journal_invalidatepage()
197 * (via truncate_inode_pages()) to discard these buffers can
198 * cause data loss. Also even if we did not discard these
199 * buffers, we would have no way to find them after the inode
200 * is reaped and thus user could see stale data if he tries to
201 * read them before the transaction is checkpointed. So be
202 * careful and force everything to disk here... We use
203 * ei->i_datasync_tid to store the newest transaction
204 * containing inode's data.
206 * Note that directories do not have this problem because they
207 * don't use page cache.
209 if (ext4_should_journal_data(inode) &&
210 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
211 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
212 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
214 jbd2_log_start_commit(journal, commit_tid);
215 jbd2_log_wait_commit(journal, commit_tid);
216 filemap_write_and_wait(&inode->i_data);
218 truncate_inode_pages(&inode->i_data, 0);
222 if (!is_bad_inode(inode))
223 dquot_initialize(inode);
225 if (ext4_should_order_data(inode))
226 ext4_begin_ordered_truncate(inode, 0);
227 truncate_inode_pages(&inode->i_data, 0);
229 if (is_bad_inode(inode))
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 * Calculate the number of metadata blocks need to reserve
323 * to allocate a block located at @lblock
325 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
327 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
328 return ext4_ext_calc_metadata_amount(inode, lblock);
330 return ext4_ind_calc_metadata_amount(inode, lblock);
334 * Called with i_data_sem down, which is important since we can call
335 * ext4_discard_preallocations() from here.
337 void ext4_da_update_reserve_space(struct inode *inode,
338 int used, int quota_claim)
340 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
341 struct ext4_inode_info *ei = EXT4_I(inode);
343 spin_lock(&ei->i_block_reservation_lock);
344 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
345 if (unlikely(used > ei->i_reserved_data_blocks)) {
346 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
347 "with only %d reserved data blocks",
348 __func__, inode->i_ino, used,
349 ei->i_reserved_data_blocks);
351 used = ei->i_reserved_data_blocks;
354 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
355 ext4_warning(inode->i_sb, "ino %lu, allocated %d "
356 "with only %d reserved metadata blocks "
357 "(releasing %d blocks with reserved %d data blocks)",
358 inode->i_ino, ei->i_allocated_meta_blocks,
359 ei->i_reserved_meta_blocks, used,
360 ei->i_reserved_data_blocks);
362 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
365 /* Update per-inode reservations */
366 ei->i_reserved_data_blocks -= used;
367 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
368 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
369 used + ei->i_allocated_meta_blocks);
370 ei->i_allocated_meta_blocks = 0;
372 if (ei->i_reserved_data_blocks == 0) {
374 * We can release all of the reserved metadata blocks
375 * only when we have written all of the delayed
378 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
379 ei->i_reserved_meta_blocks);
380 ei->i_reserved_meta_blocks = 0;
381 ei->i_da_metadata_calc_len = 0;
383 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
385 /* Update quota subsystem for data blocks */
387 dquot_claim_block(inode, EXT4_C2B(sbi, used));
390 * We did fallocate with an offset that is already delayed
391 * allocated. So on delayed allocated writeback we should
392 * not re-claim the quota for fallocated blocks.
394 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
398 * If we have done all the pending block allocations and if
399 * there aren't any writers on the inode, we can discard the
400 * inode's preallocations.
402 if ((ei->i_reserved_data_blocks == 0) &&
403 (atomic_read(&inode->i_writecount) == 0))
404 ext4_discard_preallocations(inode);
407 static int __check_block_validity(struct inode *inode, const char *func,
409 struct ext4_map_blocks *map)
411 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
413 ext4_error_inode(inode, func, line, map->m_pblk,
414 "lblock %lu mapped to illegal pblock "
415 "(length %d)", (unsigned long) map->m_lblk,
422 #define check_block_validity(inode, map) \
423 __check_block_validity((inode), __func__, __LINE__, (map))
426 * Return the number of contiguous dirty pages in a given inode
427 * starting at page frame idx.
429 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
430 unsigned int max_pages)
432 struct address_space *mapping = inode->i_mapping;
436 int i, nr_pages, done = 0;
440 pagevec_init(&pvec, 0);
443 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
445 (pgoff_t)PAGEVEC_SIZE);
448 for (i = 0; i < nr_pages; i++) {
449 struct page *page = pvec.pages[i];
450 struct buffer_head *bh, *head;
453 if (unlikely(page->mapping != mapping) ||
455 PageWriteback(page) ||
456 page->index != idx) {
461 if (page_has_buffers(page)) {
462 bh = head = page_buffers(page);
464 if (!buffer_delay(bh) &&
465 !buffer_unwritten(bh))
467 bh = bh->b_this_page;
468 } while (!done && (bh != head));
475 if (num >= max_pages) {
480 pagevec_release(&pvec);
486 * The ext4_map_blocks() function tries to look up the requested blocks,
487 * and returns if the blocks are already mapped.
489 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
490 * and store the allocated blocks in the result buffer head and mark it
493 * If file type is extents based, it will call ext4_ext_map_blocks(),
494 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
497 * On success, it returns the number of blocks being mapped or allocate.
498 * if create==0 and the blocks are pre-allocated and uninitialized block,
499 * the result buffer head is unmapped. If the create ==1, it will make sure
500 * the buffer head is mapped.
502 * It returns 0 if plain look up failed (blocks have not been allocated), in
503 * that case, buffer head is unmapped
505 * It returns the error in case of allocation failure.
507 int ext4_map_blocks(handle_t *handle, struct inode *inode,
508 struct ext4_map_blocks *map, int flags)
510 struct extent_status es;
514 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
515 "logical block %lu\n", inode->i_ino, flags, map->m_len,
516 (unsigned long) map->m_lblk);
518 /* Lookup extent status tree firstly */
519 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
520 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
521 map->m_pblk = ext4_es_pblock(&es) +
522 map->m_lblk - es.es_lblk;
523 map->m_flags |= ext4_es_is_written(&es) ?
524 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
525 retval = es.es_len - (map->m_lblk - es.es_lblk);
526 if (retval > map->m_len)
529 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
538 * Try to see if we can get the block without requesting a new
541 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
542 down_read((&EXT4_I(inode)->i_data_sem));
543 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
544 retval = ext4_ext_map_blocks(handle, inode, map, flags &
545 EXT4_GET_BLOCKS_KEEP_SIZE);
547 retval = ext4_ind_map_blocks(handle, inode, map, flags &
548 EXT4_GET_BLOCKS_KEEP_SIZE);
552 unsigned long long status;
554 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
555 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
556 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
557 ext4_find_delalloc_range(inode, map->m_lblk,
558 map->m_lblk + map->m_len - 1))
559 status |= EXTENT_STATUS_DELAYED;
560 ret = ext4_es_insert_extent(inode, map->m_lblk,
561 map->m_len, map->m_pblk, status);
565 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
566 up_read((&EXT4_I(inode)->i_data_sem));
569 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
570 int ret = check_block_validity(inode, map);
575 /* If it is only a block(s) look up */
576 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
580 * Returns if the blocks have already allocated
582 * Note that if blocks have been preallocated
583 * ext4_ext_get_block() returns the create = 0
584 * with buffer head unmapped.
586 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
590 * Here we clear m_flags because after allocating an new extent,
591 * it will be set again.
593 map->m_flags &= ~EXT4_MAP_FLAGS;
596 * New blocks allocate and/or writing to uninitialized extent
597 * will possibly result in updating i_data, so we take
598 * the write lock of i_data_sem, and call get_blocks()
599 * with create == 1 flag.
601 down_write((&EXT4_I(inode)->i_data_sem));
604 * if the caller is from delayed allocation writeout path
605 * we have already reserved fs blocks for allocation
606 * let the underlying get_block() function know to
607 * avoid double accounting
609 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
610 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
612 * We need to check for EXT4 here because migrate
613 * could have changed the inode type in between
615 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
616 retval = ext4_ext_map_blocks(handle, inode, map, flags);
618 retval = ext4_ind_map_blocks(handle, inode, map, flags);
620 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
622 * We allocated new blocks which will result in
623 * i_data's format changing. Force the migrate
624 * to fail by clearing migrate flags
626 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
630 * Update reserved blocks/metadata blocks after successful
631 * block allocation which had been deferred till now. We don't
632 * support fallocate for non extent files. So we can update
633 * reserve space here.
636 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
637 ext4_da_update_reserve_space(inode, retval, 1);
639 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
640 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
644 unsigned long long status;
646 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
647 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
648 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
649 ext4_find_delalloc_range(inode, map->m_lblk,
650 map->m_lblk + map->m_len - 1))
651 status |= EXTENT_STATUS_DELAYED;
652 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
653 map->m_pblk, status);
658 up_write((&EXT4_I(inode)->i_data_sem));
659 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
660 int ret = check_block_validity(inode, map);
667 /* Maximum number of blocks we map for direct IO at once. */
668 #define DIO_MAX_BLOCKS 4096
670 static int _ext4_get_block(struct inode *inode, sector_t iblock,
671 struct buffer_head *bh, int flags)
673 handle_t *handle = ext4_journal_current_handle();
674 struct ext4_map_blocks map;
675 int ret = 0, started = 0;
678 if (ext4_has_inline_data(inode))
682 map.m_len = bh->b_size >> inode->i_blkbits;
684 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
685 /* Direct IO write... */
686 if (map.m_len > DIO_MAX_BLOCKS)
687 map.m_len = DIO_MAX_BLOCKS;
688 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
689 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
691 if (IS_ERR(handle)) {
692 ret = PTR_ERR(handle);
698 ret = ext4_map_blocks(handle, inode, &map, flags);
700 map_bh(bh, inode->i_sb, map.m_pblk);
701 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
702 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
706 ext4_journal_stop(handle);
710 int ext4_get_block(struct inode *inode, sector_t iblock,
711 struct buffer_head *bh, int create)
713 return _ext4_get_block(inode, iblock, bh,
714 create ? EXT4_GET_BLOCKS_CREATE : 0);
718 * `handle' can be NULL if create is zero
720 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
721 ext4_lblk_t block, int create, int *errp)
723 struct ext4_map_blocks map;
724 struct buffer_head *bh;
727 J_ASSERT(handle != NULL || create == 0);
731 err = ext4_map_blocks(handle, inode, &map,
732 create ? EXT4_GET_BLOCKS_CREATE : 0);
734 /* ensure we send some value back into *errp */
737 if (create && err == 0)
738 err = -ENOSPC; /* should never happen */
744 bh = sb_getblk(inode->i_sb, map.m_pblk);
749 if (map.m_flags & EXT4_MAP_NEW) {
750 J_ASSERT(create != 0);
751 J_ASSERT(handle != NULL);
754 * Now that we do not always journal data, we should
755 * keep in mind whether this should always journal the
756 * new buffer as metadata. For now, regular file
757 * writes use ext4_get_block instead, so it's not a
761 BUFFER_TRACE(bh, "call get_create_access");
762 fatal = ext4_journal_get_create_access(handle, bh);
763 if (!fatal && !buffer_uptodate(bh)) {
764 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
765 set_buffer_uptodate(bh);
768 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
769 err = ext4_handle_dirty_metadata(handle, inode, bh);
773 BUFFER_TRACE(bh, "not a new buffer");
783 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
784 ext4_lblk_t block, int create, int *err)
786 struct buffer_head *bh;
788 bh = ext4_getblk(handle, inode, block, create, err);
791 if (buffer_uptodate(bh))
793 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
795 if (buffer_uptodate(bh))
802 int ext4_walk_page_buffers(handle_t *handle,
803 struct buffer_head *head,
807 int (*fn)(handle_t *handle,
808 struct buffer_head *bh))
810 struct buffer_head *bh;
811 unsigned block_start, block_end;
812 unsigned blocksize = head->b_size;
814 struct buffer_head *next;
816 for (bh = head, block_start = 0;
817 ret == 0 && (bh != head || !block_start);
818 block_start = block_end, bh = next) {
819 next = bh->b_this_page;
820 block_end = block_start + blocksize;
821 if (block_end <= from || block_start >= to) {
822 if (partial && !buffer_uptodate(bh))
826 err = (*fn)(handle, bh);
834 * To preserve ordering, it is essential that the hole instantiation and
835 * the data write be encapsulated in a single transaction. We cannot
836 * close off a transaction and start a new one between the ext4_get_block()
837 * and the commit_write(). So doing the jbd2_journal_start at the start of
838 * prepare_write() is the right place.
840 * Also, this function can nest inside ext4_writepage(). In that case, we
841 * *know* that ext4_writepage() has generated enough buffer credits to do the
842 * whole page. So we won't block on the journal in that case, which is good,
843 * because the caller may be PF_MEMALLOC.
845 * By accident, ext4 can be reentered when a transaction is open via
846 * quota file writes. If we were to commit the transaction while thus
847 * reentered, there can be a deadlock - we would be holding a quota
848 * lock, and the commit would never complete if another thread had a
849 * transaction open and was blocking on the quota lock - a ranking
852 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
853 * will _not_ run commit under these circumstances because handle->h_ref
854 * is elevated. We'll still have enough credits for the tiny quotafile
857 int do_journal_get_write_access(handle_t *handle,
858 struct buffer_head *bh)
860 int dirty = buffer_dirty(bh);
863 if (!buffer_mapped(bh) || buffer_freed(bh))
866 * __block_write_begin() could have dirtied some buffers. Clean
867 * the dirty bit as jbd2_journal_get_write_access() could complain
868 * otherwise about fs integrity issues. Setting of the dirty bit
869 * by __block_write_begin() isn't a real problem here as we clear
870 * the bit before releasing a page lock and thus writeback cannot
871 * ever write the buffer.
874 clear_buffer_dirty(bh);
875 ret = ext4_journal_get_write_access(handle, bh);
877 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
881 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
882 struct buffer_head *bh_result, int create);
883 static int ext4_write_begin(struct file *file, struct address_space *mapping,
884 loff_t pos, unsigned len, unsigned flags,
885 struct page **pagep, void **fsdata)
887 struct inode *inode = mapping->host;
888 int ret, needed_blocks;
895 trace_ext4_write_begin(inode, pos, len, flags);
897 * Reserve one block more for addition to orphan list in case
898 * we allocate blocks but write fails for some reason
900 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
901 index = pos >> PAGE_CACHE_SHIFT;
902 from = pos & (PAGE_CACHE_SIZE - 1);
905 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
906 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
915 * grab_cache_page_write_begin() can take a long time if the
916 * system is thrashing due to memory pressure, or if the page
917 * is being written back. So grab it first before we start
918 * the transaction handle. This also allows us to allocate
919 * the page (if needed) without using GFP_NOFS.
922 page = grab_cache_page_write_begin(mapping, index, flags);
928 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
929 if (IS_ERR(handle)) {
930 page_cache_release(page);
931 return PTR_ERR(handle);
935 if (page->mapping != mapping) {
936 /* The page got truncated from under us */
938 page_cache_release(page);
939 ext4_journal_stop(handle);
942 wait_on_page_writeback(page);
944 if (ext4_should_dioread_nolock(inode))
945 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
947 ret = __block_write_begin(page, pos, len, ext4_get_block);
949 if (!ret && ext4_should_journal_data(inode)) {
950 ret = ext4_walk_page_buffers(handle, page_buffers(page),
952 do_journal_get_write_access);
958 * __block_write_begin may have instantiated a few blocks
959 * outside i_size. Trim these off again. Don't need
960 * i_size_read because we hold i_mutex.
962 * Add inode to orphan list in case we crash before
965 if (pos + len > inode->i_size && ext4_can_truncate(inode))
966 ext4_orphan_add(handle, inode);
968 ext4_journal_stop(handle);
969 if (pos + len > inode->i_size) {
970 ext4_truncate_failed_write(inode);
972 * If truncate failed early the inode might
973 * still be on the orphan list; we need to
974 * make sure the inode is removed from the
975 * orphan list in that case.
978 ext4_orphan_del(NULL, inode);
981 if (ret == -ENOSPC &&
982 ext4_should_retry_alloc(inode->i_sb, &retries))
984 page_cache_release(page);
991 /* For write_end() in data=journal mode */
992 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
994 if (!buffer_mapped(bh) || buffer_freed(bh))
996 set_buffer_uptodate(bh);
997 return ext4_handle_dirty_metadata(handle, NULL, bh);
1000 static int ext4_generic_write_end(struct file *file,
1001 struct address_space *mapping,
1002 loff_t pos, unsigned len, unsigned copied,
1003 struct page *page, void *fsdata)
1005 int i_size_changed = 0;
1006 struct inode *inode = mapping->host;
1007 handle_t *handle = ext4_journal_current_handle();
1009 if (ext4_has_inline_data(inode))
1010 copied = ext4_write_inline_data_end(inode, pos, len,
1013 copied = block_write_end(file, mapping, pos,
1014 len, copied, page, fsdata);
1017 * No need to use i_size_read() here, the i_size
1018 * cannot change under us because we hold i_mutex.
1020 * But it's important to update i_size while still holding page lock:
1021 * page writeout could otherwise come in and zero beyond i_size.
1023 if (pos + copied > inode->i_size) {
1024 i_size_write(inode, pos + copied);
1028 if (pos + copied > EXT4_I(inode)->i_disksize) {
1029 /* We need to mark inode dirty even if
1030 * new_i_size is less that inode->i_size
1031 * bu greater than i_disksize.(hint delalloc)
1033 ext4_update_i_disksize(inode, (pos + copied));
1037 page_cache_release(page);
1040 * Don't mark the inode dirty under page lock. First, it unnecessarily
1041 * makes the holding time of page lock longer. Second, it forces lock
1042 * ordering of page lock and transaction start for journaling
1046 ext4_mark_inode_dirty(handle, inode);
1052 * We need to pick up the new inode size which generic_commit_write gave us
1053 * `file' can be NULL - eg, when called from page_symlink().
1055 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1056 * buffers are managed internally.
1058 static int ext4_ordered_write_end(struct file *file,
1059 struct address_space *mapping,
1060 loff_t pos, unsigned len, unsigned copied,
1061 struct page *page, void *fsdata)
1063 handle_t *handle = ext4_journal_current_handle();
1064 struct inode *inode = mapping->host;
1067 trace_ext4_ordered_write_end(inode, pos, len, copied);
1068 ret = ext4_jbd2_file_inode(handle, inode);
1071 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1074 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1075 /* if we have allocated more blocks and copied
1076 * less. We will have blocks allocated outside
1077 * inode->i_size. So truncate them
1079 ext4_orphan_add(handle, inode);
1084 page_cache_release(page);
1087 ret2 = ext4_journal_stop(handle);
1091 if (pos + len > inode->i_size) {
1092 ext4_truncate_failed_write(inode);
1094 * If truncate failed early the inode might still be
1095 * on the orphan list; we need to make sure the inode
1096 * is removed from the orphan list in that case.
1099 ext4_orphan_del(NULL, inode);
1103 return ret ? ret : copied;
1106 static int ext4_writeback_write_end(struct file *file,
1107 struct address_space *mapping,
1108 loff_t pos, unsigned len, unsigned copied,
1109 struct page *page, void *fsdata)
1111 handle_t *handle = ext4_journal_current_handle();
1112 struct inode *inode = mapping->host;
1115 trace_ext4_writeback_write_end(inode, pos, len, copied);
1116 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1119 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1120 /* if we have allocated more blocks and copied
1121 * less. We will have blocks allocated outside
1122 * inode->i_size. So truncate them
1124 ext4_orphan_add(handle, inode);
1129 ret2 = ext4_journal_stop(handle);
1133 if (pos + len > inode->i_size) {
1134 ext4_truncate_failed_write(inode);
1136 * If truncate failed early the inode might still be
1137 * on the orphan list; we need to make sure the inode
1138 * is removed from the orphan list in that case.
1141 ext4_orphan_del(NULL, inode);
1144 return ret ? ret : copied;
1147 static int ext4_journalled_write_end(struct file *file,
1148 struct address_space *mapping,
1149 loff_t pos, unsigned len, unsigned copied,
1150 struct page *page, void *fsdata)
1152 handle_t *handle = ext4_journal_current_handle();
1153 struct inode *inode = mapping->host;
1159 trace_ext4_journalled_write_end(inode, pos, len, copied);
1160 from = pos & (PAGE_CACHE_SIZE - 1);
1163 BUG_ON(!ext4_handle_valid(handle));
1165 if (ext4_has_inline_data(inode))
1166 copied = ext4_write_inline_data_end(inode, pos, len,
1170 if (!PageUptodate(page))
1172 page_zero_new_buffers(page, from+copied, to);
1175 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1176 to, &partial, write_end_fn);
1178 SetPageUptodate(page);
1180 new_i_size = pos + copied;
1181 if (new_i_size > inode->i_size)
1182 i_size_write(inode, pos+copied);
1183 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1184 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1185 if (new_i_size > EXT4_I(inode)->i_disksize) {
1186 ext4_update_i_disksize(inode, new_i_size);
1187 ret2 = ext4_mark_inode_dirty(handle, inode);
1193 page_cache_release(page);
1194 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1195 /* if we have allocated more blocks and copied
1196 * less. We will have blocks allocated outside
1197 * inode->i_size. So truncate them
1199 ext4_orphan_add(handle, inode);
1201 ret2 = ext4_journal_stop(handle);
1204 if (pos + len > inode->i_size) {
1205 ext4_truncate_failed_write(inode);
1207 * If truncate failed early the inode might still be
1208 * on the orphan list; we need to make sure the inode
1209 * is removed from the orphan list in that case.
1212 ext4_orphan_del(NULL, inode);
1215 return ret ? ret : copied;
1219 * Reserve a single cluster located at lblock
1221 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1224 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1225 struct ext4_inode_info *ei = EXT4_I(inode);
1226 unsigned int md_needed;
1228 ext4_lblk_t save_last_lblock;
1232 * We will charge metadata quota at writeout time; this saves
1233 * us from metadata over-estimation, though we may go over by
1234 * a small amount in the end. Here we just reserve for data.
1236 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1241 * recalculate the amount of metadata blocks to reserve
1242 * in order to allocate nrblocks
1243 * worse case is one extent per block
1246 spin_lock(&ei->i_block_reservation_lock);
1248 * ext4_calc_metadata_amount() has side effects, which we have
1249 * to be prepared undo if we fail to claim space.
1251 save_len = ei->i_da_metadata_calc_len;
1252 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1253 md_needed = EXT4_NUM_B2C(sbi,
1254 ext4_calc_metadata_amount(inode, lblock));
1255 trace_ext4_da_reserve_space(inode, md_needed);
1258 * We do still charge estimated metadata to the sb though;
1259 * we cannot afford to run out of free blocks.
1261 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1262 ei->i_da_metadata_calc_len = save_len;
1263 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1264 spin_unlock(&ei->i_block_reservation_lock);
1265 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1269 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1272 ei->i_reserved_data_blocks++;
1273 ei->i_reserved_meta_blocks += md_needed;
1274 spin_unlock(&ei->i_block_reservation_lock);
1276 return 0; /* success */
1279 static void ext4_da_release_space(struct inode *inode, int to_free)
1281 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1282 struct ext4_inode_info *ei = EXT4_I(inode);
1285 return; /* Nothing to release, exit */
1287 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1289 trace_ext4_da_release_space(inode, to_free);
1290 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1292 * if there aren't enough reserved blocks, then the
1293 * counter is messed up somewhere. Since this
1294 * function is called from invalidate page, it's
1295 * harmless to return without any action.
1297 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1298 "ino %lu, to_free %d with only %d reserved "
1299 "data blocks", inode->i_ino, to_free,
1300 ei->i_reserved_data_blocks);
1302 to_free = ei->i_reserved_data_blocks;
1304 ei->i_reserved_data_blocks -= to_free;
1306 if (ei->i_reserved_data_blocks == 0) {
1308 * We can release all of the reserved metadata blocks
1309 * only when we have written all of the delayed
1310 * allocation blocks.
1311 * Note that in case of bigalloc, i_reserved_meta_blocks,
1312 * i_reserved_data_blocks, etc. refer to number of clusters.
1314 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1315 ei->i_reserved_meta_blocks);
1316 ei->i_reserved_meta_blocks = 0;
1317 ei->i_da_metadata_calc_len = 0;
1320 /* update fs dirty data blocks counter */
1321 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1323 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1325 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1328 static void ext4_da_page_release_reservation(struct page *page,
1329 unsigned long offset)
1332 struct buffer_head *head, *bh;
1333 unsigned int curr_off = 0;
1334 struct inode *inode = page->mapping->host;
1335 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1339 head = page_buffers(page);
1342 unsigned int next_off = curr_off + bh->b_size;
1344 if ((offset <= curr_off) && (buffer_delay(bh))) {
1346 clear_buffer_delay(bh);
1348 curr_off = next_off;
1349 } while ((bh = bh->b_this_page) != head);
1352 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1353 ext4_es_remove_extent(inode, lblk, to_release);
1356 /* If we have released all the blocks belonging to a cluster, then we
1357 * need to release the reserved space for that cluster. */
1358 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1359 while (num_clusters > 0) {
1360 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1361 ((num_clusters - 1) << sbi->s_cluster_bits);
1362 if (sbi->s_cluster_ratio == 1 ||
1363 !ext4_find_delalloc_cluster(inode, lblk))
1364 ext4_da_release_space(inode, 1);
1371 * Delayed allocation stuff
1375 * mpage_da_submit_io - walks through extent of pages and try to write
1376 * them with writepage() call back
1378 * @mpd->inode: inode
1379 * @mpd->first_page: first page of the extent
1380 * @mpd->next_page: page after the last page of the extent
1382 * By the time mpage_da_submit_io() is called we expect all blocks
1383 * to be allocated. this may be wrong if allocation failed.
1385 * As pages are already locked by write_cache_pages(), we can't use it
1387 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1388 struct ext4_map_blocks *map)
1390 struct pagevec pvec;
1391 unsigned long index, end;
1392 int ret = 0, err, nr_pages, i;
1393 struct inode *inode = mpd->inode;
1394 struct address_space *mapping = inode->i_mapping;
1395 loff_t size = i_size_read(inode);
1396 unsigned int len, block_start;
1397 struct buffer_head *bh, *page_bufs = NULL;
1398 sector_t pblock = 0, cur_logical = 0;
1399 struct ext4_io_submit io_submit;
1401 BUG_ON(mpd->next_page <= mpd->first_page);
1402 memset(&io_submit, 0, sizeof(io_submit));
1404 * We need to start from the first_page to the next_page - 1
1405 * to make sure we also write the mapped dirty buffer_heads.
1406 * If we look at mpd->b_blocknr we would only be looking
1407 * at the currently mapped buffer_heads.
1409 index = mpd->first_page;
1410 end = mpd->next_page - 1;
1412 pagevec_init(&pvec, 0);
1413 while (index <= end) {
1414 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1417 for (i = 0; i < nr_pages; i++) {
1419 struct page *page = pvec.pages[i];
1421 index = page->index;
1425 if (index == size >> PAGE_CACHE_SHIFT)
1426 len = size & ~PAGE_CACHE_MASK;
1428 len = PAGE_CACHE_SIZE;
1430 cur_logical = index << (PAGE_CACHE_SHIFT -
1432 pblock = map->m_pblk + (cur_logical -
1437 BUG_ON(!PageLocked(page));
1438 BUG_ON(PageWriteback(page));
1440 bh = page_bufs = page_buffers(page);
1443 if (map && (cur_logical >= map->m_lblk) &&
1444 (cur_logical <= (map->m_lblk +
1445 (map->m_len - 1)))) {
1446 if (buffer_delay(bh)) {
1447 clear_buffer_delay(bh);
1448 bh->b_blocknr = pblock;
1450 if (buffer_unwritten(bh) ||
1452 BUG_ON(bh->b_blocknr != pblock);
1453 if (map->m_flags & EXT4_MAP_UNINIT)
1454 set_buffer_uninit(bh);
1455 clear_buffer_unwritten(bh);
1459 * skip page if block allocation undone and
1462 if (ext4_bh_delay_or_unwritten(NULL, bh))
1464 bh = bh->b_this_page;
1465 block_start += bh->b_size;
1468 } while (bh != page_bufs);
1475 clear_page_dirty_for_io(page);
1476 err = ext4_bio_write_page(&io_submit, page, len,
1479 mpd->pages_written++;
1481 * In error case, we have to continue because
1482 * remaining pages are still locked
1487 pagevec_release(&pvec);
1489 ext4_io_submit(&io_submit);
1493 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1497 struct pagevec pvec;
1498 struct inode *inode = mpd->inode;
1499 struct address_space *mapping = inode->i_mapping;
1500 ext4_lblk_t start, last;
1502 index = mpd->first_page;
1503 end = mpd->next_page - 1;
1505 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1506 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1507 ext4_es_remove_extent(inode, start, last - start + 1);
1509 pagevec_init(&pvec, 0);
1510 while (index <= end) {
1511 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1514 for (i = 0; i < nr_pages; i++) {
1515 struct page *page = pvec.pages[i];
1516 if (page->index > end)
1518 BUG_ON(!PageLocked(page));
1519 BUG_ON(PageWriteback(page));
1520 block_invalidatepage(page, 0);
1521 ClearPageUptodate(page);
1524 index = pvec.pages[nr_pages - 1]->index + 1;
1525 pagevec_release(&pvec);
1530 static void ext4_print_free_blocks(struct inode *inode)
1532 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1533 struct super_block *sb = inode->i_sb;
1535 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1536 EXT4_C2B(EXT4_SB(inode->i_sb),
1537 ext4_count_free_clusters(inode->i_sb)));
1538 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1539 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1540 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1541 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1542 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1543 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1544 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1545 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1546 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1547 EXT4_I(inode)->i_reserved_data_blocks);
1548 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1549 EXT4_I(inode)->i_reserved_meta_blocks);
1554 * mpage_da_map_and_submit - go through given space, map them
1555 * if necessary, and then submit them for I/O
1557 * @mpd - bh describing space
1559 * The function skips space we know is already mapped to disk blocks.
1562 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1564 int err, blks, get_blocks_flags;
1565 struct ext4_map_blocks map, *mapp = NULL;
1566 sector_t next = mpd->b_blocknr;
1567 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1568 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1569 handle_t *handle = NULL;
1572 * If the blocks are mapped already, or we couldn't accumulate
1573 * any blocks, then proceed immediately to the submission stage.
1575 if ((mpd->b_size == 0) ||
1576 ((mpd->b_state & (1 << BH_Mapped)) &&
1577 !(mpd->b_state & (1 << BH_Delay)) &&
1578 !(mpd->b_state & (1 << BH_Unwritten))))
1581 handle = ext4_journal_current_handle();
1585 * Call ext4_map_blocks() to allocate any delayed allocation
1586 * blocks, or to convert an uninitialized extent to be
1587 * initialized (in the case where we have written into
1588 * one or more preallocated blocks).
1590 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1591 * indicate that we are on the delayed allocation path. This
1592 * affects functions in many different parts of the allocation
1593 * call path. This flag exists primarily because we don't
1594 * want to change *many* call functions, so ext4_map_blocks()
1595 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1596 * inode's allocation semaphore is taken.
1598 * If the blocks in questions were delalloc blocks, set
1599 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1600 * variables are updated after the blocks have been allocated.
1603 map.m_len = max_blocks;
1604 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1605 if (ext4_should_dioread_nolock(mpd->inode))
1606 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1607 if (mpd->b_state & (1 << BH_Delay))
1608 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1610 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1612 struct super_block *sb = mpd->inode->i_sb;
1616 * If get block returns EAGAIN or ENOSPC and there
1617 * appears to be free blocks we will just let
1618 * mpage_da_submit_io() unlock all of the pages.
1623 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1629 * get block failure will cause us to loop in
1630 * writepages, because a_ops->writepage won't be able
1631 * to make progress. The page will be redirtied by
1632 * writepage and writepages will again try to write
1635 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1636 ext4_msg(sb, KERN_CRIT,
1637 "delayed block allocation failed for inode %lu "
1638 "at logical offset %llu with max blocks %zd "
1639 "with error %d", mpd->inode->i_ino,
1640 (unsigned long long) next,
1641 mpd->b_size >> mpd->inode->i_blkbits, err);
1642 ext4_msg(sb, KERN_CRIT,
1643 "This should not happen!! Data will be lost");
1645 ext4_print_free_blocks(mpd->inode);
1647 /* invalidate all the pages */
1648 ext4_da_block_invalidatepages(mpd);
1650 /* Mark this page range as having been completed */
1657 if (map.m_flags & EXT4_MAP_NEW) {
1658 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1661 for (i = 0; i < map.m_len; i++)
1662 unmap_underlying_metadata(bdev, map.m_pblk + i);
1666 * Update on-disk size along with block allocation.
1668 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1669 if (disksize > i_size_read(mpd->inode))
1670 disksize = i_size_read(mpd->inode);
1671 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1672 ext4_update_i_disksize(mpd->inode, disksize);
1673 err = ext4_mark_inode_dirty(handle, mpd->inode);
1675 ext4_error(mpd->inode->i_sb,
1676 "Failed to mark inode %lu dirty",
1681 mpage_da_submit_io(mpd, mapp);
1685 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1686 (1 << BH_Delay) | (1 << BH_Unwritten))
1689 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1691 * @mpd->lbh - extent of blocks
1692 * @logical - logical number of the block in the file
1693 * @b_state - b_state of the buffer head added
1695 * the function is used to collect contig. blocks in same state
1697 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical,
1698 unsigned long b_state)
1701 int blkbits = mpd->inode->i_blkbits;
1702 int nrblocks = mpd->b_size >> blkbits;
1705 * XXX Don't go larger than mballoc is willing to allocate
1706 * This is a stopgap solution. We eventually need to fold
1707 * mpage_da_submit_io() into this function and then call
1708 * ext4_map_blocks() multiple times in a loop
1710 if (nrblocks >= (8*1024*1024 >> blkbits))
1713 /* check if the reserved journal credits might overflow */
1714 if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) {
1715 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1717 * With non-extent format we are limited by the journal
1718 * credit available. Total credit needed to insert
1719 * nrblocks contiguous blocks is dependent on the
1720 * nrblocks. So limit nrblocks.
1726 * First block in the extent
1728 if (mpd->b_size == 0) {
1729 mpd->b_blocknr = logical;
1730 mpd->b_size = 1 << blkbits;
1731 mpd->b_state = b_state & BH_FLAGS;
1735 next = mpd->b_blocknr + nrblocks;
1737 * Can we merge the block to our big extent?
1739 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1740 mpd->b_size += 1 << blkbits;
1746 * We couldn't merge the block to our extent, so we
1747 * need to flush current extent and start new one
1749 mpage_da_map_and_submit(mpd);
1753 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1755 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1759 * This function is grabs code from the very beginning of
1760 * ext4_map_blocks, but assumes that the caller is from delayed write
1761 * time. This function looks up the requested blocks and sets the
1762 * buffer delay bit under the protection of i_data_sem.
1764 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1765 struct ext4_map_blocks *map,
1766 struct buffer_head *bh)
1768 struct extent_status es;
1770 sector_t invalid_block = ~((sector_t) 0xffff);
1772 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1776 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1777 "logical block %lu\n", inode->i_ino, map->m_len,
1778 (unsigned long) map->m_lblk);
1780 /* Lookup extent status tree firstly */
1781 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1783 if (ext4_es_is_hole(&es)) {
1785 down_read((&EXT4_I(inode)->i_data_sem));
1790 * Delayed extent could be allocated by fallocate.
1791 * So we need to check it.
1793 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1794 map_bh(bh, inode->i_sb, invalid_block);
1796 set_buffer_delay(bh);
1800 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1801 retval = es.es_len - (iblock - es.es_lblk);
1802 if (retval > map->m_len)
1803 retval = map->m_len;
1804 map->m_len = retval;
1805 if (ext4_es_is_written(&es))
1806 map->m_flags |= EXT4_MAP_MAPPED;
1807 else if (ext4_es_is_unwritten(&es))
1808 map->m_flags |= EXT4_MAP_UNWRITTEN;
1816 * Try to see if we can get the block without requesting a new
1817 * file system block.
1819 down_read((&EXT4_I(inode)->i_data_sem));
1820 if (ext4_has_inline_data(inode)) {
1822 * We will soon create blocks for this page, and let
1823 * us pretend as if the blocks aren't allocated yet.
1824 * In case of clusters, we have to handle the work
1825 * of mapping from cluster so that the reserved space
1826 * is calculated properly.
1828 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1829 ext4_find_delalloc_cluster(inode, map->m_lblk))
1830 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1832 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1833 retval = ext4_ext_map_blocks(NULL, inode, map,
1834 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1836 retval = ext4_ind_map_blocks(NULL, inode, map,
1837 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1843 * XXX: __block_prepare_write() unmaps passed block,
1846 /* If the block was allocated from previously allocated cluster,
1847 * then we dont need to reserve it again. */
1848 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1849 ret = ext4_da_reserve_space(inode, iblock);
1851 /* not enough space to reserve */
1857 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1858 ~0, EXTENT_STATUS_DELAYED);
1864 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1865 * and it should not appear on the bh->b_state.
1867 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1869 map_bh(bh, inode->i_sb, invalid_block);
1871 set_buffer_delay(bh);
1872 } else if (retval > 0) {
1874 unsigned long long status;
1876 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1877 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1878 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1879 map->m_pblk, status);
1885 up_read((&EXT4_I(inode)->i_data_sem));
1891 * This is a special get_blocks_t callback which is used by
1892 * ext4_da_write_begin(). It will either return mapped block or
1893 * reserve space for a single block.
1895 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1896 * We also have b_blocknr = -1 and b_bdev initialized properly
1898 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1899 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1900 * initialized properly.
1902 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1903 struct buffer_head *bh, int create)
1905 struct ext4_map_blocks map;
1908 BUG_ON(create == 0);
1909 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1911 map.m_lblk = iblock;
1915 * first, we need to know whether the block is allocated already
1916 * preallocated blocks are unmapped but should treated
1917 * the same as allocated blocks.
1919 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1923 map_bh(bh, inode->i_sb, map.m_pblk);
1924 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1926 if (buffer_unwritten(bh)) {
1927 /* A delayed write to unwritten bh should be marked
1928 * new and mapped. Mapped ensures that we don't do
1929 * get_block multiple times when we write to the same
1930 * offset and new ensures that we do proper zero out
1931 * for partial write.
1934 set_buffer_mapped(bh);
1939 static int bget_one(handle_t *handle, struct buffer_head *bh)
1945 static int bput_one(handle_t *handle, struct buffer_head *bh)
1951 static int __ext4_journalled_writepage(struct page *page,
1954 struct address_space *mapping = page->mapping;
1955 struct inode *inode = mapping->host;
1956 struct buffer_head *page_bufs = NULL;
1957 handle_t *handle = NULL;
1958 int ret = 0, err = 0;
1959 int inline_data = ext4_has_inline_data(inode);
1960 struct buffer_head *inode_bh = NULL;
1962 ClearPageChecked(page);
1965 BUG_ON(page->index != 0);
1966 BUG_ON(len > ext4_get_max_inline_size(inode));
1967 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1968 if (inode_bh == NULL)
1971 page_bufs = page_buffers(page);
1976 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1979 /* As soon as we unlock the page, it can go away, but we have
1980 * references to buffers so we are safe */
1983 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1984 ext4_writepage_trans_blocks(inode));
1985 if (IS_ERR(handle)) {
1986 ret = PTR_ERR(handle);
1990 BUG_ON(!ext4_handle_valid(handle));
1993 ret = ext4_journal_get_write_access(handle, inode_bh);
1995 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1998 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1999 do_journal_get_write_access);
2001 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2006 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2007 err = ext4_journal_stop(handle);
2011 if (!ext4_has_inline_data(inode))
2012 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2014 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2021 * Note that we don't need to start a transaction unless we're journaling data
2022 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2023 * need to file the inode to the transaction's list in ordered mode because if
2024 * we are writing back data added by write(), the inode is already there and if
2025 * we are writing back data modified via mmap(), no one guarantees in which
2026 * transaction the data will hit the disk. In case we are journaling data, we
2027 * cannot start transaction directly because transaction start ranks above page
2028 * lock so we have to do some magic.
2030 * This function can get called via...
2031 * - ext4_da_writepages after taking page lock (have journal handle)
2032 * - journal_submit_inode_data_buffers (no journal handle)
2033 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2034 * - grab_page_cache when doing write_begin (have journal handle)
2036 * We don't do any block allocation in this function. If we have page with
2037 * multiple blocks we need to write those buffer_heads that are mapped. This
2038 * is important for mmaped based write. So if we do with blocksize 1K
2039 * truncate(f, 1024);
2040 * a = mmap(f, 0, 4096);
2042 * truncate(f, 4096);
2043 * we have in the page first buffer_head mapped via page_mkwrite call back
2044 * but other buffer_heads would be unmapped but dirty (dirty done via the
2045 * do_wp_page). So writepage should write the first block. If we modify
2046 * the mmap area beyond 1024 we will again get a page_fault and the
2047 * page_mkwrite callback will do the block allocation and mark the
2048 * buffer_heads mapped.
2050 * We redirty the page if we have any buffer_heads that is either delay or
2051 * unwritten in the page.
2053 * We can get recursively called as show below.
2055 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2058 * But since we don't do any block allocation we should not deadlock.
2059 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2061 static int ext4_writepage(struct page *page,
2062 struct writeback_control *wbc)
2067 struct buffer_head *page_bufs = NULL;
2068 struct inode *inode = page->mapping->host;
2069 struct ext4_io_submit io_submit;
2071 trace_ext4_writepage(page);
2072 size = i_size_read(inode);
2073 if (page->index == size >> PAGE_CACHE_SHIFT)
2074 len = size & ~PAGE_CACHE_MASK;
2076 len = PAGE_CACHE_SIZE;
2078 page_bufs = page_buffers(page);
2080 * We cannot do block allocation or other extent handling in this
2081 * function. If there are buffers needing that, we have to redirty
2082 * the page. But we may reach here when we do a journal commit via
2083 * journal_submit_inode_data_buffers() and in that case we must write
2084 * allocated buffers to achieve data=ordered mode guarantees.
2086 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2087 ext4_bh_delay_or_unwritten)) {
2088 redirty_page_for_writepage(wbc, page);
2089 if (current->flags & PF_MEMALLOC) {
2091 * For memory cleaning there's no point in writing only
2092 * some buffers. So just bail out. Warn if we came here
2093 * from direct reclaim.
2095 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2102 if (PageChecked(page) && ext4_should_journal_data(inode))
2104 * It's mmapped pagecache. Add buffers and journal it. There
2105 * doesn't seem much point in redirtying the page here.
2107 return __ext4_journalled_writepage(page, len);
2109 memset(&io_submit, 0, sizeof(io_submit));
2110 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2111 ext4_io_submit(&io_submit);
2116 * This is called via ext4_da_writepages() to
2117 * calculate the total number of credits to reserve to fit
2118 * a single extent allocation into a single transaction,
2119 * ext4_da_writpeages() will loop calling this before
2120 * the block allocation.
2123 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2125 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2128 * With non-extent format the journal credit needed to
2129 * insert nrblocks contiguous block is dependent on
2130 * number of contiguous block. So we will limit
2131 * number of contiguous block to a sane value
2133 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2134 (max_blocks > EXT4_MAX_TRANS_DATA))
2135 max_blocks = EXT4_MAX_TRANS_DATA;
2137 return ext4_chunk_trans_blocks(inode, max_blocks);
2141 * write_cache_pages_da - walk the list of dirty pages of the given
2142 * address space and accumulate pages that need writing, and call
2143 * mpage_da_map_and_submit to map a single contiguous memory region
2144 * and then write them.
2146 static int write_cache_pages_da(handle_t *handle,
2147 struct address_space *mapping,
2148 struct writeback_control *wbc,
2149 struct mpage_da_data *mpd,
2150 pgoff_t *done_index)
2152 struct buffer_head *bh, *head;
2153 struct inode *inode = mapping->host;
2154 struct pagevec pvec;
2155 unsigned int nr_pages;
2158 long nr_to_write = wbc->nr_to_write;
2159 int i, tag, ret = 0;
2161 memset(mpd, 0, sizeof(struct mpage_da_data));
2164 pagevec_init(&pvec, 0);
2165 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2166 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2168 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2169 tag = PAGECACHE_TAG_TOWRITE;
2171 tag = PAGECACHE_TAG_DIRTY;
2173 *done_index = index;
2174 while (index <= end) {
2175 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2176 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2180 for (i = 0; i < nr_pages; i++) {
2181 struct page *page = pvec.pages[i];
2184 * At this point, the page may be truncated or
2185 * invalidated (changing page->mapping to NULL), or
2186 * even swizzled back from swapper_space to tmpfs file
2187 * mapping. However, page->index will not change
2188 * because we have a reference on the page.
2190 if (page->index > end)
2193 *done_index = page->index + 1;
2196 * If we can't merge this page, and we have
2197 * accumulated an contiguous region, write it
2199 if ((mpd->next_page != page->index) &&
2200 (mpd->next_page != mpd->first_page)) {
2201 mpage_da_map_and_submit(mpd);
2202 goto ret_extent_tail;
2208 * If the page is no longer dirty, or its
2209 * mapping no longer corresponds to inode we
2210 * are writing (which means it has been
2211 * truncated or invalidated), or the page is
2212 * already under writeback and we are not
2213 * doing a data integrity writeback, skip the page
2215 if (!PageDirty(page) ||
2216 (PageWriteback(page) &&
2217 (wbc->sync_mode == WB_SYNC_NONE)) ||
2218 unlikely(page->mapping != mapping)) {
2223 wait_on_page_writeback(page);
2224 BUG_ON(PageWriteback(page));
2227 * If we have inline data and arrive here, it means that
2228 * we will soon create the block for the 1st page, so
2229 * we'd better clear the inline data here.
2231 if (ext4_has_inline_data(inode)) {
2232 BUG_ON(ext4_test_inode_state(inode,
2233 EXT4_STATE_MAY_INLINE_DATA));
2234 ext4_destroy_inline_data(handle, inode);
2237 if (mpd->next_page != page->index)
2238 mpd->first_page = page->index;
2239 mpd->next_page = page->index + 1;
2240 logical = (sector_t) page->index <<
2241 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2243 /* Add all dirty buffers to mpd */
2244 head = page_buffers(page);
2247 BUG_ON(buffer_locked(bh));
2249 * We need to try to allocate unmapped blocks
2250 * in the same page. Otherwise we won't make
2251 * progress with the page in ext4_writepage
2253 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2254 mpage_add_bh_to_extent(mpd, logical,
2257 goto ret_extent_tail;
2258 } else if (buffer_dirty(bh) &&
2259 buffer_mapped(bh)) {
2261 * mapped dirty buffer. We need to
2262 * update the b_state because we look
2263 * at b_state in mpage_da_map_blocks.
2264 * We don't update b_size because if we
2265 * find an unmapped buffer_head later
2266 * we need to use the b_state flag of
2269 if (mpd->b_size == 0)
2271 bh->b_state & BH_FLAGS;
2274 } while ((bh = bh->b_this_page) != head);
2276 if (nr_to_write > 0) {
2278 if (nr_to_write == 0 &&
2279 wbc->sync_mode == WB_SYNC_NONE)
2281 * We stop writing back only if we are
2282 * not doing integrity sync. In case of
2283 * integrity sync we have to keep going
2284 * because someone may be concurrently
2285 * dirtying pages, and we might have
2286 * synced a lot of newly appeared dirty
2287 * pages, but have not synced all of the
2293 pagevec_release(&pvec);
2298 ret = MPAGE_DA_EXTENT_TAIL;
2300 pagevec_release(&pvec);
2306 static int ext4_da_writepages(struct address_space *mapping,
2307 struct writeback_control *wbc)
2310 int range_whole = 0;
2311 handle_t *handle = NULL;
2312 struct mpage_da_data mpd;
2313 struct inode *inode = mapping->host;
2314 int pages_written = 0;
2315 unsigned int max_pages;
2316 int range_cyclic, cycled = 1, io_done = 0;
2317 int needed_blocks, ret = 0;
2318 long desired_nr_to_write, nr_to_writebump = 0;
2319 loff_t range_start = wbc->range_start;
2320 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2321 pgoff_t done_index = 0;
2323 struct blk_plug plug;
2325 trace_ext4_da_writepages(inode, wbc);
2328 * No pages to write? This is mainly a kludge to avoid starting
2329 * a transaction for special inodes like journal inode on last iput()
2330 * because that could violate lock ordering on umount
2332 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2336 * If the filesystem has aborted, it is read-only, so return
2337 * right away instead of dumping stack traces later on that
2338 * will obscure the real source of the problem. We test
2339 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2340 * the latter could be true if the filesystem is mounted
2341 * read-only, and in that case, ext4_da_writepages should
2342 * *never* be called, so if that ever happens, we would want
2345 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2348 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2351 range_cyclic = wbc->range_cyclic;
2352 if (wbc->range_cyclic) {
2353 index = mapping->writeback_index;
2356 wbc->range_start = index << PAGE_CACHE_SHIFT;
2357 wbc->range_end = LLONG_MAX;
2358 wbc->range_cyclic = 0;
2361 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2362 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2366 * This works around two forms of stupidity. The first is in
2367 * the writeback code, which caps the maximum number of pages
2368 * written to be 1024 pages. This is wrong on multiple
2369 * levels; different architectues have a different page size,
2370 * which changes the maximum amount of data which gets
2371 * written. Secondly, 4 megabytes is way too small. XFS
2372 * forces this value to be 16 megabytes by multiplying
2373 * nr_to_write parameter by four, and then relies on its
2374 * allocator to allocate larger extents to make them
2375 * contiguous. Unfortunately this brings us to the second
2376 * stupidity, which is that ext4's mballoc code only allocates
2377 * at most 2048 blocks. So we force contiguous writes up to
2378 * the number of dirty blocks in the inode, or
2379 * sbi->max_writeback_mb_bump whichever is smaller.
2381 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2382 if (!range_cyclic && range_whole) {
2383 if (wbc->nr_to_write == LONG_MAX)
2384 desired_nr_to_write = wbc->nr_to_write;
2386 desired_nr_to_write = wbc->nr_to_write * 8;
2388 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2390 if (desired_nr_to_write > max_pages)
2391 desired_nr_to_write = max_pages;
2393 if (wbc->nr_to_write < desired_nr_to_write) {
2394 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2395 wbc->nr_to_write = desired_nr_to_write;
2399 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2400 tag_pages_for_writeback(mapping, index, end);
2402 blk_start_plug(&plug);
2403 while (!ret && wbc->nr_to_write > 0) {
2406 * we insert one extent at a time. So we need
2407 * credit needed for single extent allocation.
2408 * journalled mode is currently not supported
2411 BUG_ON(ext4_should_journal_data(inode));
2412 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2414 /* start a new transaction*/
2415 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2417 if (IS_ERR(handle)) {
2418 ret = PTR_ERR(handle);
2419 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2420 "%ld pages, ino %lu; err %d", __func__,
2421 wbc->nr_to_write, inode->i_ino, ret);
2422 blk_finish_plug(&plug);
2423 goto out_writepages;
2427 * Now call write_cache_pages_da() to find the next
2428 * contiguous region of logical blocks that need
2429 * blocks to be allocated by ext4 and submit them.
2431 ret = write_cache_pages_da(handle, mapping,
2432 wbc, &mpd, &done_index);
2434 * If we have a contiguous extent of pages and we
2435 * haven't done the I/O yet, map the blocks and submit
2438 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2439 mpage_da_map_and_submit(&mpd);
2440 ret = MPAGE_DA_EXTENT_TAIL;
2442 trace_ext4_da_write_pages(inode, &mpd);
2443 wbc->nr_to_write -= mpd.pages_written;
2445 ext4_journal_stop(handle);
2447 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2448 /* commit the transaction which would
2449 * free blocks released in the transaction
2452 jbd2_journal_force_commit_nested(sbi->s_journal);
2454 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2456 * Got one extent now try with rest of the pages.
2457 * If mpd.retval is set -EIO, journal is aborted.
2458 * So we don't need to write any more.
2460 pages_written += mpd.pages_written;
2463 } else if (wbc->nr_to_write)
2465 * There is no more writeout needed
2466 * or we requested for a noblocking writeout
2467 * and we found the device congested
2471 blk_finish_plug(&plug);
2472 if (!io_done && !cycled) {
2475 wbc->range_start = index << PAGE_CACHE_SHIFT;
2476 wbc->range_end = mapping->writeback_index - 1;
2481 wbc->range_cyclic = range_cyclic;
2482 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2484 * set the writeback_index so that range_cyclic
2485 * mode will write it back later
2487 mapping->writeback_index = done_index;
2490 wbc->nr_to_write -= nr_to_writebump;
2491 wbc->range_start = range_start;
2492 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2496 static int ext4_nonda_switch(struct super_block *sb)
2498 s64 free_blocks, dirty_blocks;
2499 struct ext4_sb_info *sbi = EXT4_SB(sb);
2502 * switch to non delalloc mode if we are running low
2503 * on free block. The free block accounting via percpu
2504 * counters can get slightly wrong with percpu_counter_batch getting
2505 * accumulated on each CPU without updating global counters
2506 * Delalloc need an accurate free block accounting. So switch
2507 * to non delalloc when we are near to error range.
2509 free_blocks = EXT4_C2B(sbi,
2510 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2511 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2513 * Start pushing delalloc when 1/2 of free blocks are dirty.
2515 if (dirty_blocks && (free_blocks < 2 * dirty_blocks) &&
2516 !writeback_in_progress(sb->s_bdi) &&
2517 down_read_trylock(&sb->s_umount)) {
2518 writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2519 up_read(&sb->s_umount);
2522 if (2 * free_blocks < 3 * dirty_blocks ||
2523 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2525 * free block count is less than 150% of dirty blocks
2526 * or free blocks is less than watermark
2533 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2534 loff_t pos, unsigned len, unsigned flags,
2535 struct page **pagep, void **fsdata)
2537 int ret, retries = 0;
2540 struct inode *inode = mapping->host;
2543 index = pos >> PAGE_CACHE_SHIFT;
2545 if (ext4_nonda_switch(inode->i_sb)) {
2546 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2547 return ext4_write_begin(file, mapping, pos,
2548 len, flags, pagep, fsdata);
2550 *fsdata = (void *)0;
2551 trace_ext4_da_write_begin(inode, pos, len, flags);
2553 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2554 ret = ext4_da_write_inline_data_begin(mapping, inode,
2564 * grab_cache_page_write_begin() can take a long time if the
2565 * system is thrashing due to memory pressure, or if the page
2566 * is being written back. So grab it first before we start
2567 * the transaction handle. This also allows us to allocate
2568 * the page (if needed) without using GFP_NOFS.
2571 page = grab_cache_page_write_begin(mapping, index, flags);
2577 * With delayed allocation, we don't log the i_disksize update
2578 * if there is delayed block allocation. But we still need
2579 * to journalling the i_disksize update if writes to the end
2580 * of file which has an already mapped buffer.
2583 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2584 if (IS_ERR(handle)) {
2585 page_cache_release(page);
2586 return PTR_ERR(handle);
2590 if (page->mapping != mapping) {
2591 /* The page got truncated from under us */
2593 page_cache_release(page);
2594 ext4_journal_stop(handle);
2597 /* In case writeback began while the page was unlocked */
2598 wait_on_page_writeback(page);
2600 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2603 ext4_journal_stop(handle);
2605 * block_write_begin may have instantiated a few blocks
2606 * outside i_size. Trim these off again. Don't need
2607 * i_size_read because we hold i_mutex.
2609 if (pos + len > inode->i_size)
2610 ext4_truncate_failed_write(inode);
2612 if (ret == -ENOSPC &&
2613 ext4_should_retry_alloc(inode->i_sb, &retries))
2616 page_cache_release(page);
2625 * Check if we should update i_disksize
2626 * when write to the end of file but not require block allocation
2628 static int ext4_da_should_update_i_disksize(struct page *page,
2629 unsigned long offset)
2631 struct buffer_head *bh;
2632 struct inode *inode = page->mapping->host;
2636 bh = page_buffers(page);
2637 idx = offset >> inode->i_blkbits;
2639 for (i = 0; i < idx; i++)
2640 bh = bh->b_this_page;
2642 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2647 static int ext4_da_write_end(struct file *file,
2648 struct address_space *mapping,
2649 loff_t pos, unsigned len, unsigned copied,
2650 struct page *page, void *fsdata)
2652 struct inode *inode = mapping->host;
2654 handle_t *handle = ext4_journal_current_handle();
2656 unsigned long start, end;
2657 int write_mode = (int)(unsigned long)fsdata;
2659 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2660 switch (ext4_inode_journal_mode(inode)) {
2661 case EXT4_INODE_ORDERED_DATA_MODE:
2662 return ext4_ordered_write_end(file, mapping, pos,
2663 len, copied, page, fsdata);
2664 case EXT4_INODE_WRITEBACK_DATA_MODE:
2665 return ext4_writeback_write_end(file, mapping, pos,
2666 len, copied, page, fsdata);
2672 trace_ext4_da_write_end(inode, pos, len, copied);
2673 start = pos & (PAGE_CACHE_SIZE - 1);
2674 end = start + copied - 1;
2677 * generic_write_end() will run mark_inode_dirty() if i_size
2678 * changes. So let's piggyback the i_disksize mark_inode_dirty
2681 new_i_size = pos + copied;
2682 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2683 if (ext4_has_inline_data(inode) ||
2684 ext4_da_should_update_i_disksize(page, end)) {
2685 down_write(&EXT4_I(inode)->i_data_sem);
2686 if (new_i_size > EXT4_I(inode)->i_disksize)
2687 EXT4_I(inode)->i_disksize = new_i_size;
2688 up_write(&EXT4_I(inode)->i_data_sem);
2689 /* We need to mark inode dirty even if
2690 * new_i_size is less that inode->i_size
2691 * bu greater than i_disksize.(hint delalloc)
2693 ext4_mark_inode_dirty(handle, inode);
2697 if (write_mode != CONVERT_INLINE_DATA &&
2698 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2699 ext4_has_inline_data(inode))
2700 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2703 ret2 = generic_write_end(file, mapping, pos, len, copied,
2709 ret2 = ext4_journal_stop(handle);
2713 return ret ? ret : copied;
2716 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2719 * Drop reserved blocks
2721 BUG_ON(!PageLocked(page));
2722 if (!page_has_buffers(page))
2725 ext4_da_page_release_reservation(page, offset);
2728 ext4_invalidatepage(page, offset);
2734 * Force all delayed allocation blocks to be allocated for a given inode.
2736 int ext4_alloc_da_blocks(struct inode *inode)
2738 trace_ext4_alloc_da_blocks(inode);
2740 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2741 !EXT4_I(inode)->i_reserved_meta_blocks)
2745 * We do something simple for now. The filemap_flush() will
2746 * also start triggering a write of the data blocks, which is
2747 * not strictly speaking necessary (and for users of
2748 * laptop_mode, not even desirable). However, to do otherwise
2749 * would require replicating code paths in:
2751 * ext4_da_writepages() ->
2752 * write_cache_pages() ---> (via passed in callback function)
2753 * __mpage_da_writepage() -->
2754 * mpage_add_bh_to_extent()
2755 * mpage_da_map_blocks()
2757 * The problem is that write_cache_pages(), located in
2758 * mm/page-writeback.c, marks pages clean in preparation for
2759 * doing I/O, which is not desirable if we're not planning on
2762 * We could call write_cache_pages(), and then redirty all of
2763 * the pages by calling redirty_page_for_writepage() but that
2764 * would be ugly in the extreme. So instead we would need to
2765 * replicate parts of the code in the above functions,
2766 * simplifying them because we wouldn't actually intend to
2767 * write out the pages, but rather only collect contiguous
2768 * logical block extents, call the multi-block allocator, and
2769 * then update the buffer heads with the block allocations.
2771 * For now, though, we'll cheat by calling filemap_flush(),
2772 * which will map the blocks, and start the I/O, but not
2773 * actually wait for the I/O to complete.
2775 return filemap_flush(inode->i_mapping);
2779 * bmap() is special. It gets used by applications such as lilo and by
2780 * the swapper to find the on-disk block of a specific piece of data.
2782 * Naturally, this is dangerous if the block concerned is still in the
2783 * journal. If somebody makes a swapfile on an ext4 data-journaling
2784 * filesystem and enables swap, then they may get a nasty shock when the
2785 * data getting swapped to that swapfile suddenly gets overwritten by
2786 * the original zero's written out previously to the journal and
2787 * awaiting writeback in the kernel's buffer cache.
2789 * So, if we see any bmap calls here on a modified, data-journaled file,
2790 * take extra steps to flush any blocks which might be in the cache.
2792 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2794 struct inode *inode = mapping->host;
2799 * We can get here for an inline file via the FIBMAP ioctl
2801 if (ext4_has_inline_data(inode))
2804 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2805 test_opt(inode->i_sb, DELALLOC)) {
2807 * With delalloc we want to sync the file
2808 * so that we can make sure we allocate
2811 filemap_write_and_wait(mapping);
2814 if (EXT4_JOURNAL(inode) &&
2815 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2817 * This is a REALLY heavyweight approach, but the use of
2818 * bmap on dirty files is expected to be extremely rare:
2819 * only if we run lilo or swapon on a freshly made file
2820 * do we expect this to happen.
2822 * (bmap requires CAP_SYS_RAWIO so this does not
2823 * represent an unprivileged user DOS attack --- we'd be
2824 * in trouble if mortal users could trigger this path at
2827 * NB. EXT4_STATE_JDATA is not set on files other than
2828 * regular files. If somebody wants to bmap a directory
2829 * or symlink and gets confused because the buffer
2830 * hasn't yet been flushed to disk, they deserve
2831 * everything they get.
2834 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2835 journal = EXT4_JOURNAL(inode);
2836 jbd2_journal_lock_updates(journal);
2837 err = jbd2_journal_flush(journal);
2838 jbd2_journal_unlock_updates(journal);
2844 return generic_block_bmap(mapping, block, ext4_get_block);
2847 static int ext4_readpage(struct file *file, struct page *page)
2850 struct inode *inode = page->mapping->host;
2852 trace_ext4_readpage(page);
2854 if (ext4_has_inline_data(inode))
2855 ret = ext4_readpage_inline(inode, page);
2858 return mpage_readpage(page, ext4_get_block);
2864 ext4_readpages(struct file *file, struct address_space *mapping,
2865 struct list_head *pages, unsigned nr_pages)
2867 struct inode *inode = mapping->host;
2869 /* If the file has inline data, no need to do readpages. */
2870 if (ext4_has_inline_data(inode))
2873 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2876 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2878 trace_ext4_invalidatepage(page, offset);
2880 /* No journalling happens on data buffers when this function is used */
2881 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2883 block_invalidatepage(page, offset);
2886 static int __ext4_journalled_invalidatepage(struct page *page,
2887 unsigned long offset)
2889 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2891 trace_ext4_journalled_invalidatepage(page, offset);
2894 * If it's a full truncate we just forget about the pending dirtying
2897 ClearPageChecked(page);
2899 return jbd2_journal_invalidatepage(journal, page, offset);
2902 /* Wrapper for aops... */
2903 static void ext4_journalled_invalidatepage(struct page *page,
2904 unsigned long offset)
2906 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
2909 static int ext4_releasepage(struct page *page, gfp_t wait)
2911 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2913 trace_ext4_releasepage(page);
2915 WARN_ON(PageChecked(page));
2916 if (!page_has_buffers(page))
2919 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2921 return try_to_free_buffers(page);
2925 * ext4_get_block used when preparing for a DIO write or buffer write.
2926 * We allocate an uinitialized extent if blocks haven't been allocated.
2927 * The extent will be converted to initialized after the IO is complete.
2929 int ext4_get_block_write(struct inode *inode, sector_t iblock,
2930 struct buffer_head *bh_result, int create)
2932 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2933 inode->i_ino, create);
2934 return _ext4_get_block(inode, iblock, bh_result,
2935 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2938 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2939 struct buffer_head *bh_result, int create)
2941 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2942 inode->i_ino, create);
2943 return _ext4_get_block(inode, iblock, bh_result,
2944 EXT4_GET_BLOCKS_NO_LOCK);
2947 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2948 ssize_t size, void *private, int ret,
2951 struct inode *inode = file_inode(iocb->ki_filp);
2952 ext4_io_end_t *io_end = iocb->private;
2954 /* if not async direct IO or dio with 0 bytes write, just return */
2955 if (!io_end || !size)
2958 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2959 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2960 iocb->private, io_end->inode->i_ino, iocb, offset,
2963 iocb->private = NULL;
2965 /* if not aio dio with unwritten extents, just free io and return */
2966 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2967 ext4_free_io_end(io_end);
2969 inode_dio_done(inode);
2971 aio_complete(iocb, ret, 0);
2975 io_end->offset = offset;
2976 io_end->size = size;
2978 io_end->iocb = iocb;
2979 io_end->result = ret;
2982 ext4_add_complete_io(io_end);
2986 * For ext4 extent files, ext4 will do direct-io write to holes,
2987 * preallocated extents, and those write extend the file, no need to
2988 * fall back to buffered IO.
2990 * For holes, we fallocate those blocks, mark them as uninitialized
2991 * If those blocks were preallocated, we mark sure they are split, but
2992 * still keep the range to write as uninitialized.
2994 * The unwritten extents will be converted to written when DIO is completed.
2995 * For async direct IO, since the IO may still pending when return, we
2996 * set up an end_io call back function, which will do the conversion
2997 * when async direct IO completed.
2999 * If the O_DIRECT write will extend the file then add this inode to the
3000 * orphan list. So recovery will truncate it back to the original size
3001 * if the machine crashes during the write.
3004 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3005 const struct iovec *iov, loff_t offset,
3006 unsigned long nr_segs)
3008 struct file *file = iocb->ki_filp;
3009 struct inode *inode = file->f_mapping->host;
3011 size_t count = iov_length(iov, nr_segs);
3013 get_block_t *get_block_func = NULL;
3015 loff_t final_size = offset + count;
3017 /* Use the old path for reads and writes beyond i_size. */
3018 if (rw != WRITE || final_size > inode->i_size)
3019 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3021 BUG_ON(iocb->private == NULL);
3023 /* If we do a overwrite dio, i_mutex locking can be released */
3024 overwrite = *((int *)iocb->private);
3027 atomic_inc(&inode->i_dio_count);
3028 down_read(&EXT4_I(inode)->i_data_sem);
3029 mutex_unlock(&inode->i_mutex);
3033 * We could direct write to holes and fallocate.
3035 * Allocated blocks to fill the hole are marked as
3036 * uninitialized to prevent parallel buffered read to expose
3037 * the stale data before DIO complete the data IO.
3039 * As to previously fallocated extents, ext4 get_block will
3040 * just simply mark the buffer mapped but still keep the
3041 * extents uninitialized.
3043 * For non AIO case, we will convert those unwritten extents
3044 * to written after return back from blockdev_direct_IO.
3046 * For async DIO, the conversion needs to be deferred when the
3047 * IO is completed. The ext4 end_io callback function will be
3048 * called to take care of the conversion work. Here for async
3049 * case, we allocate an io_end structure to hook to the iocb.
3051 iocb->private = NULL;
3052 ext4_inode_aio_set(inode, NULL);
3053 if (!is_sync_kiocb(iocb)) {
3054 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
3059 io_end->flag |= EXT4_IO_END_DIRECT;
3060 iocb->private = io_end;
3062 * we save the io structure for current async direct
3063 * IO, so that later ext4_map_blocks() could flag the
3064 * io structure whether there is a unwritten extents
3065 * needs to be converted when IO is completed.
3067 ext4_inode_aio_set(inode, io_end);
3071 get_block_func = ext4_get_block_write_nolock;
3073 get_block_func = ext4_get_block_write;
3074 dio_flags = DIO_LOCKING;
3076 ret = __blockdev_direct_IO(rw, iocb, inode,
3077 inode->i_sb->s_bdev, iov,
3085 ext4_inode_aio_set(inode, NULL);
3087 * The io_end structure takes a reference to the inode, that
3088 * structure needs to be destroyed and the reference to the
3089 * inode need to be dropped, when IO is complete, even with 0
3090 * byte write, or failed.
3092 * In the successful AIO DIO case, the io_end structure will
3093 * be destroyed and the reference to the inode will be dropped
3094 * after the end_io call back function is called.
3096 * In the case there is 0 byte write, or error case, since VFS
3097 * direct IO won't invoke the end_io call back function, we
3098 * need to free the end_io structure here.
3100 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3101 ext4_free_io_end(iocb->private);
3102 iocb->private = NULL;
3103 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3104 EXT4_STATE_DIO_UNWRITTEN)) {
3107 * for non AIO case, since the IO is already
3108 * completed, we could do the conversion right here
3110 err = ext4_convert_unwritten_extents(inode,
3114 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3118 /* take i_mutex locking again if we do a ovewrite dio */
3120 inode_dio_done(inode);
3121 up_read(&EXT4_I(inode)->i_data_sem);
3122 mutex_lock(&inode->i_mutex);
3128 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3129 const struct iovec *iov, loff_t offset,
3130 unsigned long nr_segs)
3132 struct file *file = iocb->ki_filp;
3133 struct inode *inode = file->f_mapping->host;
3137 * If we are doing data journalling we don't support O_DIRECT
3139 if (ext4_should_journal_data(inode))
3142 /* Let buffer I/O handle the inline data case. */
3143 if (ext4_has_inline_data(inode))
3146 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3147 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3148 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3150 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3151 trace_ext4_direct_IO_exit(inode, offset,
3152 iov_length(iov, nr_segs), rw, ret);
3157 * Pages can be marked dirty completely asynchronously from ext4's journalling
3158 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3159 * much here because ->set_page_dirty is called under VFS locks. The page is
3160 * not necessarily locked.
3162 * We cannot just dirty the page and leave attached buffers clean, because the
3163 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3164 * or jbddirty because all the journalling code will explode.
3166 * So what we do is to mark the page "pending dirty" and next time writepage
3167 * is called, propagate that into the buffers appropriately.
3169 static int ext4_journalled_set_page_dirty(struct page *page)
3171 SetPageChecked(page);
3172 return __set_page_dirty_nobuffers(page);
3175 static const struct address_space_operations ext4_ordered_aops = {
3176 .readpage = ext4_readpage,
3177 .readpages = ext4_readpages,
3178 .writepage = ext4_writepage,
3179 .write_begin = ext4_write_begin,
3180 .write_end = ext4_ordered_write_end,
3182 .invalidatepage = ext4_invalidatepage,
3183 .releasepage = ext4_releasepage,
3184 .direct_IO = ext4_direct_IO,
3185 .migratepage = buffer_migrate_page,
3186 .is_partially_uptodate = block_is_partially_uptodate,
3187 .error_remove_page = generic_error_remove_page,
3190 static const struct address_space_operations ext4_writeback_aops = {
3191 .readpage = ext4_readpage,
3192 .readpages = ext4_readpages,
3193 .writepage = ext4_writepage,
3194 .write_begin = ext4_write_begin,
3195 .write_end = ext4_writeback_write_end,
3197 .invalidatepage = ext4_invalidatepage,
3198 .releasepage = ext4_releasepage,
3199 .direct_IO = ext4_direct_IO,
3200 .migratepage = buffer_migrate_page,
3201 .is_partially_uptodate = block_is_partially_uptodate,
3202 .error_remove_page = generic_error_remove_page,
3205 static const struct address_space_operations ext4_journalled_aops = {
3206 .readpage = ext4_readpage,
3207 .readpages = ext4_readpages,
3208 .writepage = ext4_writepage,
3209 .write_begin = ext4_write_begin,
3210 .write_end = ext4_journalled_write_end,
3211 .set_page_dirty = ext4_journalled_set_page_dirty,
3213 .invalidatepage = ext4_journalled_invalidatepage,
3214 .releasepage = ext4_releasepage,
3215 .direct_IO = ext4_direct_IO,
3216 .is_partially_uptodate = block_is_partially_uptodate,
3217 .error_remove_page = generic_error_remove_page,
3220 static const struct address_space_operations ext4_da_aops = {
3221 .readpage = ext4_readpage,
3222 .readpages = ext4_readpages,
3223 .writepage = ext4_writepage,
3224 .writepages = ext4_da_writepages,
3225 .write_begin = ext4_da_write_begin,
3226 .write_end = ext4_da_write_end,
3228 .invalidatepage = ext4_da_invalidatepage,
3229 .releasepage = ext4_releasepage,
3230 .direct_IO = ext4_direct_IO,
3231 .migratepage = buffer_migrate_page,
3232 .is_partially_uptodate = block_is_partially_uptodate,
3233 .error_remove_page = generic_error_remove_page,
3236 void ext4_set_aops(struct inode *inode)
3238 switch (ext4_inode_journal_mode(inode)) {
3239 case EXT4_INODE_ORDERED_DATA_MODE:
3240 if (test_opt(inode->i_sb, DELALLOC))
3241 inode->i_mapping->a_ops = &ext4_da_aops;
3243 inode->i_mapping->a_ops = &ext4_ordered_aops;
3245 case EXT4_INODE_WRITEBACK_DATA_MODE:
3246 if (test_opt(inode->i_sb, DELALLOC))
3247 inode->i_mapping->a_ops = &ext4_da_aops;
3249 inode->i_mapping->a_ops = &ext4_writeback_aops;
3251 case EXT4_INODE_JOURNAL_DATA_MODE:
3252 inode->i_mapping->a_ops = &ext4_journalled_aops;
3261 * ext4_discard_partial_page_buffers()
3262 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3263 * This function finds and locks the page containing the offset
3264 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3265 * Calling functions that already have the page locked should call
3266 * ext4_discard_partial_page_buffers_no_lock directly.
3268 int ext4_discard_partial_page_buffers(handle_t *handle,
3269 struct address_space *mapping, loff_t from,
3270 loff_t length, int flags)
3272 struct inode *inode = mapping->host;
3276 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3277 mapping_gfp_mask(mapping) & ~__GFP_FS);
3281 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3282 from, length, flags);
3285 page_cache_release(page);
3290 * ext4_discard_partial_page_buffers_no_lock()
3291 * Zeros a page range of length 'length' starting from offset 'from'.
3292 * Buffer heads that correspond to the block aligned regions of the
3293 * zeroed range will be unmapped. Unblock aligned regions
3294 * will have the corresponding buffer head mapped if needed so that
3295 * that region of the page can be updated with the partial zero out.
3297 * This function assumes that the page has already been locked. The
3298 * The range to be discarded must be contained with in the given page.
3299 * If the specified range exceeds the end of the page it will be shortened
3300 * to the end of the page that corresponds to 'from'. This function is
3301 * appropriate for updating a page and it buffer heads to be unmapped and
3302 * zeroed for blocks that have been either released, or are going to be
3305 * handle: The journal handle
3306 * inode: The files inode
3307 * page: A locked page that contains the offset "from"
3308 * from: The starting byte offset (from the beginning of the file)
3309 * to begin discarding
3310 * len: The length of bytes to discard
3311 * flags: Optional flags that may be used:
3313 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3314 * Only zero the regions of the page whose buffer heads
3315 * have already been unmapped. This flag is appropriate
3316 * for updating the contents of a page whose blocks may
3317 * have already been released, and we only want to zero
3318 * out the regions that correspond to those released blocks.
3320 * Returns zero on success or negative on failure.
3322 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3323 struct inode *inode, struct page *page, loff_t from,
3324 loff_t length, int flags)
3326 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3327 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3328 unsigned int blocksize, max, pos;
3330 struct buffer_head *bh;
3333 blocksize = inode->i_sb->s_blocksize;
3334 max = PAGE_CACHE_SIZE - offset;
3336 if (index != page->index)
3340 * correct length if it does not fall between
3341 * 'from' and the end of the page
3343 if (length > max || length < 0)
3346 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3348 if (!page_has_buffers(page))
3349 create_empty_buffers(page, blocksize, 0);
3351 /* Find the buffer that contains "offset" */
3352 bh = page_buffers(page);
3354 while (offset >= pos) {
3355 bh = bh->b_this_page;
3361 while (pos < offset + length) {
3362 unsigned int end_of_block, range_to_discard;
3366 /* The length of space left to zero and unmap */
3367 range_to_discard = offset + length - pos;
3369 /* The length of space until the end of the block */
3370 end_of_block = blocksize - (pos & (blocksize-1));
3373 * Do not unmap or zero past end of block
3374 * for this buffer head
3376 if (range_to_discard > end_of_block)
3377 range_to_discard = end_of_block;
3381 * Skip this buffer head if we are only zeroing unampped
3382 * regions of the page
3384 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3388 /* If the range is block aligned, unmap */
3389 if (range_to_discard == blocksize) {
3390 clear_buffer_dirty(bh);
3392 clear_buffer_mapped(bh);
3393 clear_buffer_req(bh);
3394 clear_buffer_new(bh);
3395 clear_buffer_delay(bh);
3396 clear_buffer_unwritten(bh);
3397 clear_buffer_uptodate(bh);
3398 zero_user(page, pos, range_to_discard);
3399 BUFFER_TRACE(bh, "Buffer discarded");
3404 * If this block is not completely contained in the range
3405 * to be discarded, then it is not going to be released. Because
3406 * we need to keep this block, we need to make sure this part
3407 * of the page is uptodate before we modify it by writeing
3408 * partial zeros on it.
3410 if (!buffer_mapped(bh)) {
3412 * Buffer head must be mapped before we can read
3415 BUFFER_TRACE(bh, "unmapped");
3416 ext4_get_block(inode, iblock, bh, 0);
3417 /* unmapped? It's a hole - nothing to do */
3418 if (!buffer_mapped(bh)) {
3419 BUFFER_TRACE(bh, "still unmapped");
3424 /* Ok, it's mapped. Make sure it's up-to-date */
3425 if (PageUptodate(page))
3426 set_buffer_uptodate(bh);
3428 if (!buffer_uptodate(bh)) {
3430 ll_rw_block(READ, 1, &bh);
3432 /* Uhhuh. Read error. Complain and punt.*/
3433 if (!buffer_uptodate(bh))
3437 if (ext4_should_journal_data(inode)) {
3438 BUFFER_TRACE(bh, "get write access");
3439 err = ext4_journal_get_write_access(handle, bh);
3444 zero_user(page, pos, range_to_discard);
3447 if (ext4_should_journal_data(inode)) {
3448 err = ext4_handle_dirty_metadata(handle, inode, bh);
3450 mark_buffer_dirty(bh);
3452 BUFFER_TRACE(bh, "Partial buffer zeroed");
3454 bh = bh->b_this_page;
3456 pos += range_to_discard;
3462 int ext4_can_truncate(struct inode *inode)
3464 if (S_ISREG(inode->i_mode))
3466 if (S_ISDIR(inode->i_mode))
3468 if (S_ISLNK(inode->i_mode))
3469 return !ext4_inode_is_fast_symlink(inode);
3474 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3475 * associated with the given offset and length
3477 * @inode: File inode
3478 * @offset: The offset where the hole will begin
3479 * @len: The length of the hole
3481 * Returns: 0 on success or negative on failure
3484 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3486 struct inode *inode = file_inode(file);
3487 if (!S_ISREG(inode->i_mode))
3490 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3491 return ext4_ind_punch_hole(file, offset, length);
3493 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3494 /* TODO: Add support for bigalloc file systems */
3498 trace_ext4_punch_hole(inode, offset, length);
3500 return ext4_ext_punch_hole(file, offset, length);
3506 * We block out ext4_get_block() block instantiations across the entire
3507 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3508 * simultaneously on behalf of the same inode.
3510 * As we work through the truncate and commit bits of it to the journal there
3511 * is one core, guiding principle: the file's tree must always be consistent on
3512 * disk. We must be able to restart the truncate after a crash.
3514 * The file's tree may be transiently inconsistent in memory (although it
3515 * probably isn't), but whenever we close off and commit a journal transaction,
3516 * the contents of (the filesystem + the journal) must be consistent and
3517 * restartable. It's pretty simple, really: bottom up, right to left (although
3518 * left-to-right works OK too).
3520 * Note that at recovery time, journal replay occurs *before* the restart of
3521 * truncate against the orphan inode list.
3523 * The committed inode has the new, desired i_size (which is the same as
3524 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3525 * that this inode's truncate did not complete and it will again call
3526 * ext4_truncate() to have another go. So there will be instantiated blocks
3527 * to the right of the truncation point in a crashed ext4 filesystem. But
3528 * that's fine - as long as they are linked from the inode, the post-crash
3529 * ext4_truncate() run will find them and release them.
3531 void ext4_truncate(struct inode *inode)
3533 trace_ext4_truncate_enter(inode);
3535 if (!ext4_can_truncate(inode))
3538 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3540 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3541 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3543 if (ext4_has_inline_data(inode)) {
3546 ext4_inline_data_truncate(inode, &has_inline);
3551 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3552 ext4_ext_truncate(inode);
3554 ext4_ind_truncate(inode);
3556 trace_ext4_truncate_exit(inode);
3560 * ext4_get_inode_loc returns with an extra refcount against the inode's
3561 * underlying buffer_head on success. If 'in_mem' is true, we have all
3562 * data in memory that is needed to recreate the on-disk version of this
3565 static int __ext4_get_inode_loc(struct inode *inode,
3566 struct ext4_iloc *iloc, int in_mem)
3568 struct ext4_group_desc *gdp;
3569 struct buffer_head *bh;
3570 struct super_block *sb = inode->i_sb;
3572 int inodes_per_block, inode_offset;
3575 if (!ext4_valid_inum(sb, inode->i_ino))
3578 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3579 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3584 * Figure out the offset within the block group inode table
3586 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3587 inode_offset = ((inode->i_ino - 1) %
3588 EXT4_INODES_PER_GROUP(sb));
3589 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3590 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3592 bh = sb_getblk(sb, block);
3595 if (!buffer_uptodate(bh)) {
3599 * If the buffer has the write error flag, we have failed
3600 * to write out another inode in the same block. In this
3601 * case, we don't have to read the block because we may
3602 * read the old inode data successfully.
3604 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3605 set_buffer_uptodate(bh);
3607 if (buffer_uptodate(bh)) {
3608 /* someone brought it uptodate while we waited */
3614 * If we have all information of the inode in memory and this
3615 * is the only valid inode in the block, we need not read the
3619 struct buffer_head *bitmap_bh;
3622 start = inode_offset & ~(inodes_per_block - 1);
3624 /* Is the inode bitmap in cache? */
3625 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3626 if (unlikely(!bitmap_bh))
3630 * If the inode bitmap isn't in cache then the
3631 * optimisation may end up performing two reads instead
3632 * of one, so skip it.
3634 if (!buffer_uptodate(bitmap_bh)) {
3638 for (i = start; i < start + inodes_per_block; i++) {
3639 if (i == inode_offset)
3641 if (ext4_test_bit(i, bitmap_bh->b_data))
3645 if (i == start + inodes_per_block) {
3646 /* all other inodes are free, so skip I/O */
3647 memset(bh->b_data, 0, bh->b_size);
3648 set_buffer_uptodate(bh);
3656 * If we need to do any I/O, try to pre-readahead extra
3657 * blocks from the inode table.
3659 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3660 ext4_fsblk_t b, end, table;
3663 table = ext4_inode_table(sb, gdp);
3664 /* s_inode_readahead_blks is always a power of 2 */
3665 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3668 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3669 num = EXT4_INODES_PER_GROUP(sb);
3670 if (ext4_has_group_desc_csum(sb))
3671 num -= ext4_itable_unused_count(sb, gdp);
3672 table += num / inodes_per_block;
3676 sb_breadahead(sb, b++);
3680 * There are other valid inodes in the buffer, this inode
3681 * has in-inode xattrs, or we don't have this inode in memory.
3682 * Read the block from disk.
3684 trace_ext4_load_inode(inode);
3686 bh->b_end_io = end_buffer_read_sync;
3687 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3689 if (!buffer_uptodate(bh)) {
3690 EXT4_ERROR_INODE_BLOCK(inode, block,
3691 "unable to read itable block");
3701 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3703 /* We have all inode data except xattrs in memory here. */
3704 return __ext4_get_inode_loc(inode, iloc,
3705 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3708 void ext4_set_inode_flags(struct inode *inode)
3710 unsigned int flags = EXT4_I(inode)->i_flags;
3712 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3713 if (flags & EXT4_SYNC_FL)
3714 inode->i_flags |= S_SYNC;
3715 if (flags & EXT4_APPEND_FL)
3716 inode->i_flags |= S_APPEND;
3717 if (flags & EXT4_IMMUTABLE_FL)
3718 inode->i_flags |= S_IMMUTABLE;
3719 if (flags & EXT4_NOATIME_FL)
3720 inode->i_flags |= S_NOATIME;
3721 if (flags & EXT4_DIRSYNC_FL)
3722 inode->i_flags |= S_DIRSYNC;
3725 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3726 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3728 unsigned int vfs_fl;
3729 unsigned long old_fl, new_fl;
3732 vfs_fl = ei->vfs_inode.i_flags;
3733 old_fl = ei->i_flags;
3734 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3735 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3737 if (vfs_fl & S_SYNC)
3738 new_fl |= EXT4_SYNC_FL;
3739 if (vfs_fl & S_APPEND)
3740 new_fl |= EXT4_APPEND_FL;
3741 if (vfs_fl & S_IMMUTABLE)
3742 new_fl |= EXT4_IMMUTABLE_FL;
3743 if (vfs_fl & S_NOATIME)
3744 new_fl |= EXT4_NOATIME_FL;
3745 if (vfs_fl & S_DIRSYNC)
3746 new_fl |= EXT4_DIRSYNC_FL;
3747 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3750 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3751 struct ext4_inode_info *ei)
3754 struct inode *inode = &(ei->vfs_inode);
3755 struct super_block *sb = inode->i_sb;
3757 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3758 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3759 /* we are using combined 48 bit field */
3760 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3761 le32_to_cpu(raw_inode->i_blocks_lo);
3762 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3763 /* i_blocks represent file system block size */
3764 return i_blocks << (inode->i_blkbits - 9);
3769 return le32_to_cpu(raw_inode->i_blocks_lo);
3773 static inline void ext4_iget_extra_inode(struct inode *inode,
3774 struct ext4_inode *raw_inode,
3775 struct ext4_inode_info *ei)
3777 __le32 *magic = (void *)raw_inode +
3778 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3779 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3780 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3781 ext4_find_inline_data_nolock(inode);
3783 EXT4_I(inode)->i_inline_off = 0;
3786 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3788 struct ext4_iloc iloc;
3789 struct ext4_inode *raw_inode;
3790 struct ext4_inode_info *ei;
3791 struct inode *inode;
3792 journal_t *journal = EXT4_SB(sb)->s_journal;
3798 inode = iget_locked(sb, ino);
3800 return ERR_PTR(-ENOMEM);
3801 if (!(inode->i_state & I_NEW))
3807 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3810 raw_inode = ext4_raw_inode(&iloc);
3812 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3813 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3814 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3815 EXT4_INODE_SIZE(inode->i_sb)) {
3816 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3817 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3818 EXT4_INODE_SIZE(inode->i_sb));
3823 ei->i_extra_isize = 0;
3825 /* Precompute checksum seed for inode metadata */
3826 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3827 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3828 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3830 __le32 inum = cpu_to_le32(inode->i_ino);
3831 __le32 gen = raw_inode->i_generation;
3832 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3834 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3838 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3839 EXT4_ERROR_INODE(inode, "checksum invalid");
3844 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3845 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3846 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3847 if (!(test_opt(inode->i_sb, NO_UID32))) {
3848 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3849 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3851 i_uid_write(inode, i_uid);
3852 i_gid_write(inode, i_gid);
3853 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3855 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3856 ei->i_inline_off = 0;
3857 ei->i_dir_start_lookup = 0;
3858 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3859 /* We now have enough fields to check if the inode was active or not.
3860 * This is needed because nfsd might try to access dead inodes
3861 * the test is that same one that e2fsck uses
3862 * NeilBrown 1999oct15
3864 if (inode->i_nlink == 0) {
3865 if (inode->i_mode == 0 ||
3866 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3867 /* this inode is deleted */
3871 /* The only unlinked inodes we let through here have
3872 * valid i_mode and are being read by the orphan
3873 * recovery code: that's fine, we're about to complete
3874 * the process of deleting those. */
3876 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3877 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3878 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3879 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3881 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3882 inode->i_size = ext4_isize(raw_inode);
3883 ei->i_disksize = inode->i_size;
3885 ei->i_reserved_quota = 0;
3887 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3888 ei->i_block_group = iloc.block_group;
3889 ei->i_last_alloc_group = ~0;
3891 * NOTE! The in-memory inode i_data array is in little-endian order
3892 * even on big-endian machines: we do NOT byteswap the block numbers!
3894 for (block = 0; block < EXT4_N_BLOCKS; block++)
3895 ei->i_data[block] = raw_inode->i_block[block];
3896 INIT_LIST_HEAD(&ei->i_orphan);
3899 * Set transaction id's of transactions that have to be committed
3900 * to finish f[data]sync. We set them to currently running transaction
3901 * as we cannot be sure that the inode or some of its metadata isn't
3902 * part of the transaction - the inode could have been reclaimed and
3903 * now it is reread from disk.
3906 transaction_t *transaction;
3909 read_lock(&journal->j_state_lock);
3910 if (journal->j_running_transaction)
3911 transaction = journal->j_running_transaction;
3913 transaction = journal->j_committing_transaction;
3915 tid = transaction->t_tid;
3917 tid = journal->j_commit_sequence;
3918 read_unlock(&journal->j_state_lock);
3919 ei->i_sync_tid = tid;
3920 ei->i_datasync_tid = tid;
3923 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3924 if (ei->i_extra_isize == 0) {
3925 /* The extra space is currently unused. Use it. */
3926 ei->i_extra_isize = sizeof(struct ext4_inode) -
3927 EXT4_GOOD_OLD_INODE_SIZE;
3929 ext4_iget_extra_inode(inode, raw_inode, ei);
3933 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3934 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3935 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3936 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3938 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3939 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3940 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3942 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3946 if (ei->i_file_acl &&
3947 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3948 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
3952 } else if (!ext4_has_inline_data(inode)) {
3953 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3954 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3955 (S_ISLNK(inode->i_mode) &&
3956 !ext4_inode_is_fast_symlink(inode))))
3957 /* Validate extent which is part of inode */
3958 ret = ext4_ext_check_inode(inode);
3959 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3960 (S_ISLNK(inode->i_mode) &&
3961 !ext4_inode_is_fast_symlink(inode))) {
3962 /* Validate block references which are part of inode */
3963 ret = ext4_ind_check_inode(inode);
3969 if (S_ISREG(inode->i_mode)) {
3970 inode->i_op = &ext4_file_inode_operations;
3971 inode->i_fop = &ext4_file_operations;
3972 ext4_set_aops(inode);
3973 } else if (S_ISDIR(inode->i_mode)) {
3974 inode->i_op = &ext4_dir_inode_operations;
3975 inode->i_fop = &ext4_dir_operations;
3976 } else if (S_ISLNK(inode->i_mode)) {
3977 if (ext4_inode_is_fast_symlink(inode)) {
3978 inode->i_op = &ext4_fast_symlink_inode_operations;
3979 nd_terminate_link(ei->i_data, inode->i_size,
3980 sizeof(ei->i_data) - 1);
3982 inode->i_op = &ext4_symlink_inode_operations;
3983 ext4_set_aops(inode);
3985 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3986 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3987 inode->i_op = &ext4_special_inode_operations;
3988 if (raw_inode->i_block[0])
3989 init_special_inode(inode, inode->i_mode,
3990 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3992 init_special_inode(inode, inode->i_mode,
3993 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3996 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4000 ext4_set_inode_flags(inode);
4001 unlock_new_inode(inode);
4007 return ERR_PTR(ret);
4010 static int ext4_inode_blocks_set(handle_t *handle,
4011 struct ext4_inode *raw_inode,
4012 struct ext4_inode_info *ei)
4014 struct inode *inode = &(ei->vfs_inode);
4015 u64 i_blocks = inode->i_blocks;
4016 struct super_block *sb = inode->i_sb;
4018 if (i_blocks <= ~0U) {
4020 * i_blocks can be represented in a 32 bit variable
4021 * as multiple of 512 bytes
4023 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4024 raw_inode->i_blocks_high = 0;
4025 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4028 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4031 if (i_blocks <= 0xffffffffffffULL) {
4033 * i_blocks can be represented in a 48 bit variable
4034 * as multiple of 512 bytes
4036 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4037 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4038 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4040 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4041 /* i_block is stored in file system block size */
4042 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4043 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4044 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4050 * Post the struct inode info into an on-disk inode location in the
4051 * buffer-cache. This gobbles the caller's reference to the
4052 * buffer_head in the inode location struct.
4054 * The caller must have write access to iloc->bh.
4056 static int ext4_do_update_inode(handle_t *handle,
4057 struct inode *inode,
4058 struct ext4_iloc *iloc)
4060 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4061 struct ext4_inode_info *ei = EXT4_I(inode);
4062 struct buffer_head *bh = iloc->bh;
4063 int err = 0, rc, block;
4064 int need_datasync = 0;
4068 /* For fields not not tracking in the in-memory inode,
4069 * initialise them to zero for new inodes. */
4070 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4071 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4073 ext4_get_inode_flags(ei);
4074 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4075 i_uid = i_uid_read(inode);
4076 i_gid = i_gid_read(inode);
4077 if (!(test_opt(inode->i_sb, NO_UID32))) {
4078 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4079 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4081 * Fix up interoperability with old kernels. Otherwise, old inodes get
4082 * re-used with the upper 16 bits of the uid/gid intact
4085 raw_inode->i_uid_high =
4086 cpu_to_le16(high_16_bits(i_uid));
4087 raw_inode->i_gid_high =
4088 cpu_to_le16(high_16_bits(i_gid));
4090 raw_inode->i_uid_high = 0;
4091 raw_inode->i_gid_high = 0;
4094 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4095 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4096 raw_inode->i_uid_high = 0;
4097 raw_inode->i_gid_high = 0;
4099 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4101 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4102 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4103 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4104 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4106 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4108 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4109 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4110 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4111 cpu_to_le32(EXT4_OS_HURD))
4112 raw_inode->i_file_acl_high =
4113 cpu_to_le16(ei->i_file_acl >> 32);
4114 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4115 if (ei->i_disksize != ext4_isize(raw_inode)) {
4116 ext4_isize_set(raw_inode, ei->i_disksize);
4119 if (ei->i_disksize > 0x7fffffffULL) {
4120 struct super_block *sb = inode->i_sb;
4121 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4122 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4123 EXT4_SB(sb)->s_es->s_rev_level ==
4124 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4125 /* If this is the first large file
4126 * created, add a flag to the superblock.
4128 err = ext4_journal_get_write_access(handle,
4129 EXT4_SB(sb)->s_sbh);
4132 ext4_update_dynamic_rev(sb);
4133 EXT4_SET_RO_COMPAT_FEATURE(sb,
4134 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4135 ext4_handle_sync(handle);
4136 err = ext4_handle_dirty_super(handle, sb);
4139 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4140 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4141 if (old_valid_dev(inode->i_rdev)) {
4142 raw_inode->i_block[0] =
4143 cpu_to_le32(old_encode_dev(inode->i_rdev));
4144 raw_inode->i_block[1] = 0;
4146 raw_inode->i_block[0] = 0;
4147 raw_inode->i_block[1] =
4148 cpu_to_le32(new_encode_dev(inode->i_rdev));
4149 raw_inode->i_block[2] = 0;
4151 } else if (!ext4_has_inline_data(inode)) {
4152 for (block = 0; block < EXT4_N_BLOCKS; block++)
4153 raw_inode->i_block[block] = ei->i_data[block];
4156 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4157 if (ei->i_extra_isize) {
4158 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4159 raw_inode->i_version_hi =
4160 cpu_to_le32(inode->i_version >> 32);
4161 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4164 ext4_inode_csum_set(inode, raw_inode, ei);
4166 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4167 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4170 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4172 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4175 ext4_std_error(inode->i_sb, err);
4180 * ext4_write_inode()
4182 * We are called from a few places:
4184 * - Within generic_file_write() for O_SYNC files.
4185 * Here, there will be no transaction running. We wait for any running
4186 * transaction to commit.
4188 * - Within sys_sync(), kupdate and such.
4189 * We wait on commit, if tol to.
4191 * - Within prune_icache() (PF_MEMALLOC == true)
4192 * Here we simply return. We can't afford to block kswapd on the
4195 * In all cases it is actually safe for us to return without doing anything,
4196 * because the inode has been copied into a raw inode buffer in
4197 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4200 * Note that we are absolutely dependent upon all inode dirtiers doing the
4201 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4202 * which we are interested.
4204 * It would be a bug for them to not do this. The code:
4206 * mark_inode_dirty(inode)
4208 * inode->i_size = expr;
4210 * is in error because a kswapd-driven write_inode() could occur while
4211 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4212 * will no longer be on the superblock's dirty inode list.
4214 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4218 if (current->flags & PF_MEMALLOC)
4221 if (EXT4_SB(inode->i_sb)->s_journal) {
4222 if (ext4_journal_current_handle()) {
4223 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4228 if (wbc->sync_mode != WB_SYNC_ALL)
4231 err = ext4_force_commit(inode->i_sb);
4233 struct ext4_iloc iloc;
4235 err = __ext4_get_inode_loc(inode, &iloc, 0);
4238 if (wbc->sync_mode == WB_SYNC_ALL)
4239 sync_dirty_buffer(iloc.bh);
4240 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4241 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4242 "IO error syncing inode");
4251 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4252 * buffers that are attached to a page stradding i_size and are undergoing
4253 * commit. In that case we have to wait for commit to finish and try again.
4255 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4259 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4260 tid_t commit_tid = 0;
4263 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4265 * All buffers in the last page remain valid? Then there's nothing to
4266 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4269 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4272 page = find_lock_page(inode->i_mapping,
4273 inode->i_size >> PAGE_CACHE_SHIFT);
4276 ret = __ext4_journalled_invalidatepage(page, offset);
4278 page_cache_release(page);
4282 read_lock(&journal->j_state_lock);
4283 if (journal->j_committing_transaction)
4284 commit_tid = journal->j_committing_transaction->t_tid;
4285 read_unlock(&journal->j_state_lock);
4287 jbd2_log_wait_commit(journal, commit_tid);
4294 * Called from notify_change.
4296 * We want to trap VFS attempts to truncate the file as soon as
4297 * possible. In particular, we want to make sure that when the VFS
4298 * shrinks i_size, we put the inode on the orphan list and modify
4299 * i_disksize immediately, so that during the subsequent flushing of
4300 * dirty pages and freeing of disk blocks, we can guarantee that any
4301 * commit will leave the blocks being flushed in an unused state on
4302 * disk. (On recovery, the inode will get truncated and the blocks will
4303 * be freed, so we have a strong guarantee that no future commit will
4304 * leave these blocks visible to the user.)
4306 * Another thing we have to assure is that if we are in ordered mode
4307 * and inode is still attached to the committing transaction, we must
4308 * we start writeout of all the dirty pages which are being truncated.
4309 * This way we are sure that all the data written in the previous
4310 * transaction are already on disk (truncate waits for pages under
4313 * Called with inode->i_mutex down.
4315 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4317 struct inode *inode = dentry->d_inode;
4320 const unsigned int ia_valid = attr->ia_valid;
4322 error = inode_change_ok(inode, attr);
4326 if (is_quota_modification(inode, attr))
4327 dquot_initialize(inode);
4328 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4329 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4332 /* (user+group)*(old+new) structure, inode write (sb,
4333 * inode block, ? - but truncate inode update has it) */
4334 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4335 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4336 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4337 if (IS_ERR(handle)) {
4338 error = PTR_ERR(handle);
4341 error = dquot_transfer(inode, attr);
4343 ext4_journal_stop(handle);
4346 /* Update corresponding info in inode so that everything is in
4347 * one transaction */
4348 if (attr->ia_valid & ATTR_UID)
4349 inode->i_uid = attr->ia_uid;
4350 if (attr->ia_valid & ATTR_GID)
4351 inode->i_gid = attr->ia_gid;
4352 error = ext4_mark_inode_dirty(handle, inode);
4353 ext4_journal_stop(handle);
4356 if (attr->ia_valid & ATTR_SIZE) {
4358 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4359 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4361 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4366 if (S_ISREG(inode->i_mode) &&
4367 attr->ia_valid & ATTR_SIZE &&
4368 (attr->ia_size < inode->i_size)) {
4371 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4372 if (IS_ERR(handle)) {
4373 error = PTR_ERR(handle);
4376 if (ext4_handle_valid(handle)) {
4377 error = ext4_orphan_add(handle, inode);
4380 EXT4_I(inode)->i_disksize = attr->ia_size;
4381 rc = ext4_mark_inode_dirty(handle, inode);
4384 ext4_journal_stop(handle);
4386 if (ext4_should_order_data(inode)) {
4387 error = ext4_begin_ordered_truncate(inode,
4390 /* Do as much error cleanup as possible */
4391 handle = ext4_journal_start(inode,
4393 if (IS_ERR(handle)) {
4394 ext4_orphan_del(NULL, inode);
4397 ext4_orphan_del(handle, inode);
4399 ext4_journal_stop(handle);
4405 if (attr->ia_valid & ATTR_SIZE) {
4406 if (attr->ia_size != inode->i_size) {
4407 loff_t oldsize = inode->i_size;
4409 i_size_write(inode, attr->ia_size);
4411 * Blocks are going to be removed from the inode. Wait
4412 * for dio in flight. Temporarily disable
4413 * dioread_nolock to prevent livelock.
4416 if (!ext4_should_journal_data(inode)) {
4417 ext4_inode_block_unlocked_dio(inode);
4418 inode_dio_wait(inode);
4419 ext4_inode_resume_unlocked_dio(inode);
4421 ext4_wait_for_tail_page_commit(inode);
4424 * Truncate pagecache after we've waited for commit
4425 * in data=journal mode to make pages freeable.
4427 truncate_pagecache(inode, oldsize, inode->i_size);
4429 ext4_truncate(inode);
4433 setattr_copy(inode, attr);
4434 mark_inode_dirty(inode);
4438 * If the call to ext4_truncate failed to get a transaction handle at
4439 * all, we need to clean up the in-core orphan list manually.
4441 if (orphan && inode->i_nlink)
4442 ext4_orphan_del(NULL, inode);
4444 if (!rc && (ia_valid & ATTR_MODE))
4445 rc = ext4_acl_chmod(inode);
4448 ext4_std_error(inode->i_sb, error);
4454 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4457 struct inode *inode;
4458 unsigned long delalloc_blocks;
4460 inode = dentry->d_inode;
4461 generic_fillattr(inode, stat);
4464 * We can't update i_blocks if the block allocation is delayed
4465 * otherwise in the case of system crash before the real block
4466 * allocation is done, we will have i_blocks inconsistent with
4467 * on-disk file blocks.
4468 * We always keep i_blocks updated together with real
4469 * allocation. But to not confuse with user, stat
4470 * will return the blocks that include the delayed allocation
4471 * blocks for this file.
4473 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4474 EXT4_I(inode)->i_reserved_data_blocks);
4476 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4480 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4482 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4483 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4484 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4488 * Account for index blocks, block groups bitmaps and block group
4489 * descriptor blocks if modify datablocks and index blocks
4490 * worse case, the indexs blocks spread over different block groups
4492 * If datablocks are discontiguous, they are possible to spread over
4493 * different block groups too. If they are contiguous, with flexbg,
4494 * they could still across block group boundary.
4496 * Also account for superblock, inode, quota and xattr blocks
4498 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4500 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4506 * How many index blocks need to touch to modify nrblocks?
4507 * The "Chunk" flag indicating whether the nrblocks is
4508 * physically contiguous on disk
4510 * For Direct IO and fallocate, they calls get_block to allocate
4511 * one single extent at a time, so they could set the "Chunk" flag
4513 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4518 * Now let's see how many group bitmaps and group descriptors need
4528 if (groups > ngroups)
4530 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4531 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4533 /* bitmaps and block group descriptor blocks */
4534 ret += groups + gdpblocks;
4536 /* Blocks for super block, inode, quota and xattr blocks */
4537 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4543 * Calculate the total number of credits to reserve to fit
4544 * the modification of a single pages into a single transaction,
4545 * which may include multiple chunks of block allocations.
4547 * This could be called via ext4_write_begin()
4549 * We need to consider the worse case, when
4550 * one new block per extent.
4552 int ext4_writepage_trans_blocks(struct inode *inode)
4554 int bpp = ext4_journal_blocks_per_page(inode);
4557 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4559 /* Account for data blocks for journalled mode */
4560 if (ext4_should_journal_data(inode))
4566 * Calculate the journal credits for a chunk of data modification.
4568 * This is called from DIO, fallocate or whoever calling
4569 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4571 * journal buffers for data blocks are not included here, as DIO
4572 * and fallocate do no need to journal data buffers.
4574 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4576 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4580 * The caller must have previously called ext4_reserve_inode_write().
4581 * Give this, we know that the caller already has write access to iloc->bh.
4583 int ext4_mark_iloc_dirty(handle_t *handle,
4584 struct inode *inode, struct ext4_iloc *iloc)
4588 if (IS_I_VERSION(inode))
4589 inode_inc_iversion(inode);
4591 /* the do_update_inode consumes one bh->b_count */
4594 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4595 err = ext4_do_update_inode(handle, inode, iloc);
4601 * On success, We end up with an outstanding reference count against
4602 * iloc->bh. This _must_ be cleaned up later.
4606 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4607 struct ext4_iloc *iloc)
4611 err = ext4_get_inode_loc(inode, iloc);
4613 BUFFER_TRACE(iloc->bh, "get_write_access");
4614 err = ext4_journal_get_write_access(handle, iloc->bh);
4620 ext4_std_error(inode->i_sb, err);
4625 * Expand an inode by new_extra_isize bytes.
4626 * Returns 0 on success or negative error number on failure.
4628 static int ext4_expand_extra_isize(struct inode *inode,
4629 unsigned int new_extra_isize,
4630 struct ext4_iloc iloc,
4633 struct ext4_inode *raw_inode;
4634 struct ext4_xattr_ibody_header *header;
4636 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4639 raw_inode = ext4_raw_inode(&iloc);
4641 header = IHDR(inode, raw_inode);
4643 /* No extended attributes present */
4644 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4645 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4646 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4648 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4652 /* try to expand with EAs present */
4653 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4658 * What we do here is to mark the in-core inode as clean with respect to inode
4659 * dirtiness (it may still be data-dirty).
4660 * This means that the in-core inode may be reaped by prune_icache
4661 * without having to perform any I/O. This is a very good thing,
4662 * because *any* task may call prune_icache - even ones which
4663 * have a transaction open against a different journal.
4665 * Is this cheating? Not really. Sure, we haven't written the
4666 * inode out, but prune_icache isn't a user-visible syncing function.
4667 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4668 * we start and wait on commits.
4670 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4672 struct ext4_iloc iloc;
4673 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4674 static unsigned int mnt_count;
4678 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4679 err = ext4_reserve_inode_write(handle, inode, &iloc);
4680 if (ext4_handle_valid(handle) &&
4681 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4682 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4684 * We need extra buffer credits since we may write into EA block
4685 * with this same handle. If journal_extend fails, then it will
4686 * only result in a minor loss of functionality for that inode.
4687 * If this is felt to be critical, then e2fsck should be run to
4688 * force a large enough s_min_extra_isize.
4690 if ((jbd2_journal_extend(handle,
4691 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4692 ret = ext4_expand_extra_isize(inode,
4693 sbi->s_want_extra_isize,
4696 ext4_set_inode_state(inode,
4697 EXT4_STATE_NO_EXPAND);
4699 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4700 ext4_warning(inode->i_sb,
4701 "Unable to expand inode %lu. Delete"
4702 " some EAs or run e2fsck.",
4705 le16_to_cpu(sbi->s_es->s_mnt_count);
4711 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4716 * ext4_dirty_inode() is called from __mark_inode_dirty()
4718 * We're really interested in the case where a file is being extended.
4719 * i_size has been changed by generic_commit_write() and we thus need
4720 * to include the updated inode in the current transaction.
4722 * Also, dquot_alloc_block() will always dirty the inode when blocks
4723 * are allocated to the file.
4725 * If the inode is marked synchronous, we don't honour that here - doing
4726 * so would cause a commit on atime updates, which we don't bother doing.
4727 * We handle synchronous inodes at the highest possible level.
4729 void ext4_dirty_inode(struct inode *inode, int flags)
4733 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
4737 ext4_mark_inode_dirty(handle, inode);
4739 ext4_journal_stop(handle);
4746 * Bind an inode's backing buffer_head into this transaction, to prevent
4747 * it from being flushed to disk early. Unlike
4748 * ext4_reserve_inode_write, this leaves behind no bh reference and
4749 * returns no iloc structure, so the caller needs to repeat the iloc
4750 * lookup to mark the inode dirty later.
4752 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4754 struct ext4_iloc iloc;
4758 err = ext4_get_inode_loc(inode, &iloc);
4760 BUFFER_TRACE(iloc.bh, "get_write_access");
4761 err = jbd2_journal_get_write_access(handle, iloc.bh);
4763 err = ext4_handle_dirty_metadata(handle,
4769 ext4_std_error(inode->i_sb, err);
4774 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4781 * We have to be very careful here: changing a data block's
4782 * journaling status dynamically is dangerous. If we write a
4783 * data block to the journal, change the status and then delete
4784 * that block, we risk forgetting to revoke the old log record
4785 * from the journal and so a subsequent replay can corrupt data.
4786 * So, first we make sure that the journal is empty and that
4787 * nobody is changing anything.
4790 journal = EXT4_JOURNAL(inode);
4793 if (is_journal_aborted(journal))
4795 /* We have to allocate physical blocks for delalloc blocks
4796 * before flushing journal. otherwise delalloc blocks can not
4797 * be allocated any more. even more truncate on delalloc blocks
4798 * could trigger BUG by flushing delalloc blocks in journal.
4799 * There is no delalloc block in non-journal data mode.
4801 if (val && test_opt(inode->i_sb, DELALLOC)) {
4802 err = ext4_alloc_da_blocks(inode);
4807 /* Wait for all existing dio workers */
4808 ext4_inode_block_unlocked_dio(inode);
4809 inode_dio_wait(inode);
4811 jbd2_journal_lock_updates(journal);
4814 * OK, there are no updates running now, and all cached data is
4815 * synced to disk. We are now in a completely consistent state
4816 * which doesn't have anything in the journal, and we know that
4817 * no filesystem updates are running, so it is safe to modify
4818 * the inode's in-core data-journaling state flag now.
4822 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4824 jbd2_journal_flush(journal);
4825 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4827 ext4_set_aops(inode);
4829 jbd2_journal_unlock_updates(journal);
4830 ext4_inode_resume_unlocked_dio(inode);
4832 /* Finally we can mark the inode as dirty. */
4834 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
4836 return PTR_ERR(handle);
4838 err = ext4_mark_inode_dirty(handle, inode);
4839 ext4_handle_sync(handle);
4840 ext4_journal_stop(handle);
4841 ext4_std_error(inode->i_sb, err);
4846 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4848 return !buffer_mapped(bh);
4851 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4853 struct page *page = vmf->page;
4857 struct file *file = vma->vm_file;
4858 struct inode *inode = file_inode(file);
4859 struct address_space *mapping = inode->i_mapping;
4861 get_block_t *get_block;
4864 sb_start_pagefault(inode->i_sb);
4865 file_update_time(vma->vm_file);
4866 /* Delalloc case is easy... */
4867 if (test_opt(inode->i_sb, DELALLOC) &&
4868 !ext4_should_journal_data(inode) &&
4869 !ext4_nonda_switch(inode->i_sb)) {
4871 ret = __block_page_mkwrite(vma, vmf,
4872 ext4_da_get_block_prep);
4873 } while (ret == -ENOSPC &&
4874 ext4_should_retry_alloc(inode->i_sb, &retries));
4879 size = i_size_read(inode);
4880 /* Page got truncated from under us? */
4881 if (page->mapping != mapping || page_offset(page) > size) {
4883 ret = VM_FAULT_NOPAGE;
4887 if (page->index == size >> PAGE_CACHE_SHIFT)
4888 len = size & ~PAGE_CACHE_MASK;
4890 len = PAGE_CACHE_SIZE;
4892 * Return if we have all the buffers mapped. This avoids the need to do
4893 * journal_start/journal_stop which can block and take a long time
4895 if (page_has_buffers(page)) {
4896 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
4898 ext4_bh_unmapped)) {
4899 /* Wait so that we don't change page under IO */
4900 wait_for_stable_page(page);
4901 ret = VM_FAULT_LOCKED;
4906 /* OK, we need to fill the hole... */
4907 if (ext4_should_dioread_nolock(inode))
4908 get_block = ext4_get_block_write;
4910 get_block = ext4_get_block;
4912 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
4913 ext4_writepage_trans_blocks(inode));
4914 if (IS_ERR(handle)) {
4915 ret = VM_FAULT_SIGBUS;
4918 ret = __block_page_mkwrite(vma, vmf, get_block);
4919 if (!ret && ext4_should_journal_data(inode)) {
4920 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
4921 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4923 ret = VM_FAULT_SIGBUS;
4924 ext4_journal_stop(handle);
4927 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4929 ext4_journal_stop(handle);
4930 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4933 ret = block_page_mkwrite_return(ret);
4935 sb_end_pagefault(inode->i_sb);