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
21 #include <linux/module.h>
23 #include <linux/time.h>
24 #include <linux/jbd2.h>
25 #include <linux/highuid.h>
26 #include <linux/pagemap.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/bitops.h>
43 #include "ext4_jbd2.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
52 static inline int ext4_begin_ordered_truncate(struct inode *inode,
55 trace_ext4_begin_ordered_truncate(inode, new_size);
57 * If jinode is zero, then we never opened the file for
58 * writing, so there's no need to call
59 * jbd2_journal_begin_ordered_truncate() since there's no
60 * outstanding writes we need to flush.
62 if (!EXT4_I(inode)->jinode)
64 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
65 EXT4_I(inode)->jinode,
69 static void ext4_invalidatepage(struct page *page, unsigned long offset);
70 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
71 struct buffer_head *bh_result, int create);
72 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
73 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
74 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
75 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
78 * Test whether an inode is a fast symlink.
80 static int ext4_inode_is_fast_symlink(struct inode *inode)
82 int ea_blocks = EXT4_I(inode)->i_file_acl ?
83 (inode->i_sb->s_blocksize >> 9) : 0;
85 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
89 * Restart the transaction associated with *handle. This does a commit,
90 * so before we call here everything must be consistently dirtied against
93 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
99 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
100 * moment, get_block can be called only for blocks inside i_size since
101 * page cache has been already dropped and writes are blocked by
102 * i_mutex. So we can safely drop the i_data_sem here.
104 BUG_ON(EXT4_JOURNAL(inode) == NULL);
105 jbd_debug(2, "restarting handle %p\n", handle);
106 up_write(&EXT4_I(inode)->i_data_sem);
107 ret = ext4_journal_restart(handle, nblocks);
108 down_write(&EXT4_I(inode)->i_data_sem);
109 ext4_discard_preallocations(inode);
115 * Called at the last iput() if i_nlink is zero.
117 void ext4_evict_inode(struct inode *inode)
122 trace_ext4_evict_inode(inode);
124 ext4_ioend_wait(inode);
126 if (inode->i_nlink) {
128 * When journalling data dirty buffers are tracked only in the
129 * journal. So although mm thinks everything is clean and
130 * ready for reaping the inode might still have some pages to
131 * write in the running transaction or waiting to be
132 * checkpointed. Thus calling jbd2_journal_invalidatepage()
133 * (via truncate_inode_pages()) to discard these buffers can
134 * cause data loss. Also even if we did not discard these
135 * buffers, we would have no way to find them after the inode
136 * is reaped and thus user could see stale data if he tries to
137 * read them before the transaction is checkpointed. So be
138 * careful and force everything to disk here... We use
139 * ei->i_datasync_tid to store the newest transaction
140 * containing inode's data.
142 * Note that directories do not have this problem because they
143 * don't use page cache.
145 if (inode->i_ino != EXT4_JOURNAL_INO &&
146 ext4_should_journal_data(inode) &&
147 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
148 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
149 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
151 jbd2_complete_transaction(journal, commit_tid);
152 filemap_write_and_wait(&inode->i_data);
154 truncate_inode_pages(&inode->i_data, 0);
158 if (is_bad_inode(inode))
160 dquot_initialize(inode);
162 if (ext4_should_order_data(inode))
163 ext4_begin_ordered_truncate(inode, 0);
164 truncate_inode_pages(&inode->i_data, 0);
167 handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
168 if (IS_ERR(handle)) {
169 ext4_std_error(inode->i_sb, PTR_ERR(handle));
171 * If we're going to skip the normal cleanup, we still need to
172 * make sure that the in-core orphan linked list is properly
175 ext4_orphan_del(NULL, inode);
180 ext4_handle_sync(handle);
182 err = ext4_mark_inode_dirty(handle, inode);
184 ext4_warning(inode->i_sb,
185 "couldn't mark inode dirty (err %d)", err);
189 ext4_truncate(inode);
192 * ext4_ext_truncate() doesn't reserve any slop when it
193 * restarts journal transactions; therefore there may not be
194 * enough credits left in the handle to remove the inode from
195 * the orphan list and set the dtime field.
197 if (!ext4_handle_has_enough_credits(handle, 3)) {
198 err = ext4_journal_extend(handle, 3);
200 err = ext4_journal_restart(handle, 3);
202 ext4_warning(inode->i_sb,
203 "couldn't extend journal (err %d)", err);
205 ext4_journal_stop(handle);
206 ext4_orphan_del(NULL, inode);
212 * Kill off the orphan record which ext4_truncate created.
213 * AKPM: I think this can be inside the above `if'.
214 * Note that ext4_orphan_del() has to be able to cope with the
215 * deletion of a non-existent orphan - this is because we don't
216 * know if ext4_truncate() actually created an orphan record.
217 * (Well, we could do this if we need to, but heck - it works)
219 ext4_orphan_del(handle, inode);
220 EXT4_I(inode)->i_dtime = get_seconds();
223 * One subtle ordering requirement: if anything has gone wrong
224 * (transaction abort, IO errors, whatever), then we can still
225 * do these next steps (the fs will already have been marked as
226 * having errors), but we can't free the inode if the mark_dirty
229 if (ext4_mark_inode_dirty(handle, inode))
230 /* If that failed, just do the required in-core inode clear. */
231 ext4_clear_inode(inode);
233 ext4_free_inode(handle, inode);
234 ext4_journal_stop(handle);
237 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
241 qsize_t *ext4_get_reserved_space(struct inode *inode)
243 return &EXT4_I(inode)->i_reserved_quota;
248 * Calculate the number of metadata blocks need to reserve
249 * to allocate a block located at @lblock
251 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
253 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
254 return ext4_ext_calc_metadata_amount(inode, lblock);
256 return ext4_ind_calc_metadata_amount(inode, lblock);
260 * Called with i_data_sem down, which is important since we can call
261 * ext4_discard_preallocations() from here.
263 void ext4_da_update_reserve_space(struct inode *inode,
264 int used, int quota_claim)
266 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
267 struct ext4_inode_info *ei = EXT4_I(inode);
269 spin_lock(&ei->i_block_reservation_lock);
270 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
271 if (unlikely(used > ei->i_reserved_data_blocks)) {
272 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
273 "with only %d reserved data blocks\n",
274 __func__, inode->i_ino, used,
275 ei->i_reserved_data_blocks);
277 used = ei->i_reserved_data_blocks;
280 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
281 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, allocated %d "
282 "with only %d reserved metadata blocks\n", __func__,
283 inode->i_ino, ei->i_allocated_meta_blocks,
284 ei->i_reserved_meta_blocks);
286 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
289 /* Update per-inode reservations */
290 ei->i_reserved_data_blocks -= used;
291 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
292 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
293 used + ei->i_allocated_meta_blocks);
294 ei->i_allocated_meta_blocks = 0;
296 if (ei->i_reserved_data_blocks == 0) {
298 * We can release all of the reserved metadata blocks
299 * only when we have written all of the delayed
302 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
303 ei->i_reserved_meta_blocks);
304 ei->i_reserved_meta_blocks = 0;
305 ei->i_da_metadata_calc_len = 0;
307 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
309 /* Update quota subsystem for data blocks */
311 dquot_claim_block(inode, EXT4_C2B(sbi, used));
314 * We did fallocate with an offset that is already delayed
315 * allocated. So on delayed allocated writeback we should
316 * not re-claim the quota for fallocated blocks.
318 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
322 * If we have done all the pending block allocations and if
323 * there aren't any writers on the inode, we can discard the
324 * inode's preallocations.
326 if ((ei->i_reserved_data_blocks == 0) &&
327 (atomic_read(&inode->i_writecount) == 0))
328 ext4_discard_preallocations(inode);
331 static int __check_block_validity(struct inode *inode, const char *func,
333 struct ext4_map_blocks *map)
335 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
337 ext4_error_inode(inode, func, line, map->m_pblk,
338 "lblock %lu mapped to illegal pblock "
339 "(length %d)", (unsigned long) map->m_lblk,
346 #define check_block_validity(inode, map) \
347 __check_block_validity((inode), __func__, __LINE__, (map))
350 * Return the number of contiguous dirty pages in a given inode
351 * starting at page frame idx.
353 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
354 unsigned int max_pages)
356 struct address_space *mapping = inode->i_mapping;
360 int i, nr_pages, done = 0;
364 pagevec_init(&pvec, 0);
367 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
369 (pgoff_t)PAGEVEC_SIZE);
372 for (i = 0; i < nr_pages; i++) {
373 struct page *page = pvec.pages[i];
374 struct buffer_head *bh, *head;
377 if (unlikely(page->mapping != mapping) ||
379 PageWriteback(page) ||
380 page->index != idx) {
385 if (page_has_buffers(page)) {
386 bh = head = page_buffers(page);
388 if (!buffer_delay(bh) &&
389 !buffer_unwritten(bh))
391 bh = bh->b_this_page;
392 } while (!done && (bh != head));
399 if (num >= max_pages) {
404 pagevec_release(&pvec);
410 * Sets the BH_Da_Mapped bit on the buffer heads corresponding to the given map.
412 static void set_buffers_da_mapped(struct inode *inode,
413 struct ext4_map_blocks *map)
415 struct address_space *mapping = inode->i_mapping;
420 index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
421 end = (map->m_lblk + map->m_len - 1) >>
422 (PAGE_CACHE_SHIFT - inode->i_blkbits);
424 pagevec_init(&pvec, 0);
425 while (index <= end) {
426 nr_pages = pagevec_lookup(&pvec, mapping, index,
428 (pgoff_t)PAGEVEC_SIZE));
431 for (i = 0; i < nr_pages; i++) {
432 struct page *page = pvec.pages[i];
433 struct buffer_head *bh, *head;
435 if (unlikely(page->mapping != mapping) ||
439 if (page_has_buffers(page)) {
440 bh = head = page_buffers(page);
442 set_buffer_da_mapped(bh);
443 bh = bh->b_this_page;
444 } while (bh != head);
448 pagevec_release(&pvec);
453 * The ext4_map_blocks() function tries to look up the requested blocks,
454 * and returns if the blocks are already mapped.
456 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
457 * and store the allocated blocks in the result buffer head and mark it
460 * If file type is extents based, it will call ext4_ext_map_blocks(),
461 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
464 * On success, it returns the number of blocks being mapped or allocate.
465 * if create==0 and the blocks are pre-allocated and uninitialized block,
466 * the result buffer head is unmapped. If the create ==1, it will make sure
467 * the buffer head is mapped.
469 * It returns 0 if plain look up failed (blocks have not been allocated), in
470 * that case, buffer head is unmapped
472 * It returns the error in case of allocation failure.
474 int ext4_map_blocks(handle_t *handle, struct inode *inode,
475 struct ext4_map_blocks *map, int flags)
480 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
481 "logical block %lu\n", inode->i_ino, flags, map->m_len,
482 (unsigned long) map->m_lblk);
484 /* We can handle the block number less than EXT_MAX_BLOCKS */
485 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
489 * Try to see if we can get the block without requesting a new
492 down_read((&EXT4_I(inode)->i_data_sem));
493 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
494 retval = ext4_ext_map_blocks(handle, inode, map, flags &
495 EXT4_GET_BLOCKS_KEEP_SIZE);
497 retval = ext4_ind_map_blocks(handle, inode, map, flags &
498 EXT4_GET_BLOCKS_KEEP_SIZE);
500 up_read((&EXT4_I(inode)->i_data_sem));
502 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
503 int ret = check_block_validity(inode, map);
508 /* If it is only a block(s) look up */
509 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
513 * Returns if the blocks have already allocated
515 * Note that if blocks have been preallocated
516 * ext4_ext_get_block() returns the create = 0
517 * with buffer head unmapped.
519 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
523 * When we call get_blocks without the create flag, the
524 * BH_Unwritten flag could have gotten set if the blocks
525 * requested were part of a uninitialized extent. We need to
526 * clear this flag now that we are committed to convert all or
527 * part of the uninitialized extent to be an initialized
528 * extent. This is because we need to avoid the combination
529 * of BH_Unwritten and BH_Mapped flags being simultaneously
530 * set on the buffer_head.
532 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
535 * New blocks allocate and/or writing to uninitialized extent
536 * will possibly result in updating i_data, so we take
537 * the write lock of i_data_sem, and call get_blocks()
538 * with create == 1 flag.
540 down_write((&EXT4_I(inode)->i_data_sem));
543 * if the caller is from delayed allocation writeout path
544 * we have already reserved fs blocks for allocation
545 * let the underlying get_block() function know to
546 * avoid double accounting
548 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
549 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
551 * We need to check for EXT4 here because migrate
552 * could have changed the inode type in between
554 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
555 retval = ext4_ext_map_blocks(handle, inode, map, flags);
557 retval = ext4_ind_map_blocks(handle, inode, map, flags);
559 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
561 * We allocated new blocks which will result in
562 * i_data's format changing. Force the migrate
563 * to fail by clearing migrate flags
565 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
569 * Update reserved blocks/metadata blocks after successful
570 * block allocation which had been deferred till now. We don't
571 * support fallocate for non extent files. So we can update
572 * reserve space here.
575 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
576 ext4_da_update_reserve_space(inode, retval, 1);
578 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
579 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
581 /* If we have successfully mapped the delayed allocated blocks,
582 * set the BH_Da_Mapped bit on them. Its important to do this
583 * under the protection of i_data_sem.
585 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
586 set_buffers_da_mapped(inode, map);
589 up_write((&EXT4_I(inode)->i_data_sem));
590 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
591 int ret = check_block_validity(inode, map);
599 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
600 * we have to be careful as someone else may be manipulating b_state as well.
602 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
604 unsigned long old_state;
605 unsigned long new_state;
607 flags &= EXT4_MAP_FLAGS;
609 /* Dummy buffer_head? Set non-atomically. */
611 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
615 * Someone else may be modifying b_state. Be careful! This is ugly but
616 * once we get rid of using bh as a container for mapping information
617 * to pass to / from get_block functions, this can go away.
620 old_state = ACCESS_ONCE(bh->b_state);
621 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
623 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
626 /* Maximum number of blocks we map for direct IO at once. */
627 #define DIO_MAX_BLOCKS 4096
629 static int _ext4_get_block(struct inode *inode, sector_t iblock,
630 struct buffer_head *bh, int flags)
632 handle_t *handle = ext4_journal_current_handle();
633 struct ext4_map_blocks map;
634 int ret = 0, started = 0;
638 map.m_len = bh->b_size >> inode->i_blkbits;
640 if (flags && !handle) {
641 /* Direct IO write... */
642 if (map.m_len > DIO_MAX_BLOCKS)
643 map.m_len = DIO_MAX_BLOCKS;
644 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
645 handle = ext4_journal_start(inode, dio_credits);
646 if (IS_ERR(handle)) {
647 ret = PTR_ERR(handle);
653 ret = ext4_map_blocks(handle, inode, &map, flags);
655 map_bh(bh, inode->i_sb, map.m_pblk);
656 ext4_update_bh_state(bh, map.m_flags);
657 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
661 ext4_journal_stop(handle);
665 int ext4_get_block(struct inode *inode, sector_t iblock,
666 struct buffer_head *bh, int create)
668 return _ext4_get_block(inode, iblock, bh,
669 create ? EXT4_GET_BLOCKS_CREATE : 0);
673 * `handle' can be NULL if create is zero
675 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
676 ext4_lblk_t block, int create, int *errp)
678 struct ext4_map_blocks map;
679 struct buffer_head *bh;
682 J_ASSERT(handle != NULL || create == 0);
686 err = ext4_map_blocks(handle, inode, &map,
687 create ? EXT4_GET_BLOCKS_CREATE : 0);
695 bh = sb_getblk(inode->i_sb, map.m_pblk);
700 if (map.m_flags & EXT4_MAP_NEW) {
701 J_ASSERT(create != 0);
702 J_ASSERT(handle != NULL);
705 * Now that we do not always journal data, we should
706 * keep in mind whether this should always journal the
707 * new buffer as metadata. For now, regular file
708 * writes use ext4_get_block instead, so it's not a
712 BUFFER_TRACE(bh, "call get_create_access");
713 fatal = ext4_journal_get_create_access(handle, bh);
714 if (!fatal && !buffer_uptodate(bh)) {
715 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
716 set_buffer_uptodate(bh);
719 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
720 err = ext4_handle_dirty_metadata(handle, inode, bh);
724 BUFFER_TRACE(bh, "not a new buffer");
734 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
735 ext4_lblk_t block, int create, int *err)
737 struct buffer_head *bh;
739 bh = ext4_getblk(handle, inode, block, create, err);
742 if (buffer_uptodate(bh))
744 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
746 if (buffer_uptodate(bh))
753 static int walk_page_buffers(handle_t *handle,
754 struct buffer_head *head,
758 int (*fn)(handle_t *handle,
759 struct buffer_head *bh))
761 struct buffer_head *bh;
762 unsigned block_start, block_end;
763 unsigned blocksize = head->b_size;
765 struct buffer_head *next;
767 for (bh = head, block_start = 0;
768 ret == 0 && (bh != head || !block_start);
769 block_start = block_end, bh = next) {
770 next = bh->b_this_page;
771 block_end = block_start + blocksize;
772 if (block_end <= from || block_start >= to) {
773 if (partial && !buffer_uptodate(bh))
777 err = (*fn)(handle, bh);
785 * To preserve ordering, it is essential that the hole instantiation and
786 * the data write be encapsulated in a single transaction. We cannot
787 * close off a transaction and start a new one between the ext4_get_block()
788 * and the commit_write(). So doing the jbd2_journal_start at the start of
789 * prepare_write() is the right place.
791 * Also, this function can nest inside ext4_writepage() ->
792 * block_write_full_page(). In that case, we *know* that ext4_writepage()
793 * has generated enough buffer credits to do the whole page. So we won't
794 * block on the journal in that case, which is good, because the caller may
797 * By accident, ext4 can be reentered when a transaction is open via
798 * quota file writes. If we were to commit the transaction while thus
799 * reentered, there can be a deadlock - we would be holding a quota
800 * lock, and the commit would never complete if another thread had a
801 * transaction open and was blocking on the quota lock - a ranking
804 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
805 * will _not_ run commit under these circumstances because handle->h_ref
806 * is elevated. We'll still have enough credits for the tiny quotafile
809 static int do_journal_get_write_access(handle_t *handle,
810 struct buffer_head *bh)
812 int dirty = buffer_dirty(bh);
815 if (!buffer_mapped(bh) || buffer_freed(bh))
818 * __block_write_begin() could have dirtied some buffers. Clean
819 * the dirty bit as jbd2_journal_get_write_access() could complain
820 * otherwise about fs integrity issues. Setting of the dirty bit
821 * by __block_write_begin() isn't a real problem here as we clear
822 * the bit before releasing a page lock and thus writeback cannot
823 * ever write the buffer.
826 clear_buffer_dirty(bh);
827 ret = ext4_journal_get_write_access(handle, bh);
829 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
833 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
834 struct buffer_head *bh_result, int create);
835 static int ext4_write_begin(struct file *file, struct address_space *mapping,
836 loff_t pos, unsigned len, unsigned flags,
837 struct page **pagep, void **fsdata)
839 struct inode *inode = mapping->host;
840 int ret, needed_blocks;
847 trace_ext4_write_begin(inode, pos, len, flags);
849 * Reserve one block more for addition to orphan list in case
850 * we allocate blocks but write fails for some reason
852 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
853 index = pos >> PAGE_CACHE_SHIFT;
854 from = pos & (PAGE_CACHE_SIZE - 1);
858 handle = ext4_journal_start(inode, needed_blocks);
859 if (IS_ERR(handle)) {
860 ret = PTR_ERR(handle);
864 /* We cannot recurse into the filesystem as the transaction is already
866 flags |= AOP_FLAG_NOFS;
868 page = grab_cache_page_write_begin(mapping, index, flags);
870 ext4_journal_stop(handle);
876 if (ext4_should_dioread_nolock(inode))
877 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
879 ret = __block_write_begin(page, pos, len, ext4_get_block);
881 if (!ret && ext4_should_journal_data(inode)) {
882 ret = walk_page_buffers(handle, page_buffers(page),
883 from, to, NULL, do_journal_get_write_access);
888 page_cache_release(page);
890 * __block_write_begin may have instantiated a few blocks
891 * outside i_size. Trim these off again. Don't need
892 * i_size_read because we hold i_mutex.
894 * Add inode to orphan list in case we crash before
897 if (pos + len > inode->i_size && ext4_can_truncate(inode))
898 ext4_orphan_add(handle, inode);
900 ext4_journal_stop(handle);
901 if (pos + len > inode->i_size) {
902 ext4_truncate_failed_write(inode);
904 * If truncate failed early the inode might
905 * still be on the orphan list; we need to
906 * make sure the inode is removed from the
907 * orphan list in that case.
910 ext4_orphan_del(NULL, inode);
914 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
920 /* For write_end() in data=journal mode */
921 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
923 if (!buffer_mapped(bh) || buffer_freed(bh))
925 set_buffer_uptodate(bh);
926 return ext4_handle_dirty_metadata(handle, NULL, bh);
929 static int ext4_generic_write_end(struct file *file,
930 struct address_space *mapping,
931 loff_t pos, unsigned len, unsigned copied,
932 struct page *page, void *fsdata)
934 int i_size_changed = 0;
935 struct inode *inode = mapping->host;
936 handle_t *handle = ext4_journal_current_handle();
938 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
941 * No need to use i_size_read() here, the i_size
942 * cannot change under us because we hold i_mutex.
944 * But it's important to update i_size while still holding page lock:
945 * page writeout could otherwise come in and zero beyond i_size.
947 if (pos + copied > inode->i_size) {
948 i_size_write(inode, pos + copied);
952 if (pos + copied > EXT4_I(inode)->i_disksize) {
953 /* We need to mark inode dirty even if
954 * new_i_size is less that inode->i_size
955 * bu greater than i_disksize.(hint delalloc)
957 ext4_update_i_disksize(inode, (pos + copied));
961 page_cache_release(page);
964 * Don't mark the inode dirty under page lock. First, it unnecessarily
965 * makes the holding time of page lock longer. Second, it forces lock
966 * ordering of page lock and transaction start for journaling
970 ext4_mark_inode_dirty(handle, inode);
976 * We need to pick up the new inode size which generic_commit_write gave us
977 * `file' can be NULL - eg, when called from page_symlink().
979 * ext4 never places buffers on inode->i_mapping->private_list. metadata
980 * buffers are managed internally.
982 static int ext4_ordered_write_end(struct file *file,
983 struct address_space *mapping,
984 loff_t pos, unsigned len, unsigned copied,
985 struct page *page, void *fsdata)
987 handle_t *handle = ext4_journal_current_handle();
988 struct inode *inode = mapping->host;
991 trace_ext4_ordered_write_end(inode, pos, len, copied);
992 ret = ext4_jbd2_file_inode(handle, inode);
995 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
998 if (pos + len > inode->i_size && ext4_can_truncate(inode))
999 /* if we have allocated more blocks and copied
1000 * less. We will have blocks allocated outside
1001 * inode->i_size. So truncate them
1003 ext4_orphan_add(handle, inode);
1008 page_cache_release(page);
1011 ret2 = ext4_journal_stop(handle);
1015 if (pos + len > inode->i_size) {
1016 ext4_truncate_failed_write(inode);
1018 * If truncate failed early the inode might still be
1019 * on the orphan list; we need to make sure the inode
1020 * is removed from the orphan list in that case.
1023 ext4_orphan_del(NULL, inode);
1027 return ret ? ret : copied;
1030 static int ext4_writeback_write_end(struct file *file,
1031 struct address_space *mapping,
1032 loff_t pos, unsigned len, unsigned copied,
1033 struct page *page, void *fsdata)
1035 handle_t *handle = ext4_journal_current_handle();
1036 struct inode *inode = mapping->host;
1039 trace_ext4_writeback_write_end(inode, pos, len, copied);
1040 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1043 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1044 /* if we have allocated more blocks and copied
1045 * less. We will have blocks allocated outside
1046 * inode->i_size. So truncate them
1048 ext4_orphan_add(handle, inode);
1053 ret2 = ext4_journal_stop(handle);
1057 if (pos + len > inode->i_size) {
1058 ext4_truncate_failed_write(inode);
1060 * If truncate failed early the inode might still be
1061 * on the orphan list; we need to make sure the inode
1062 * is removed from the orphan list in that case.
1065 ext4_orphan_del(NULL, inode);
1068 return ret ? ret : copied;
1072 * This is a private version of page_zero_new_buffers() which doesn't
1073 * set the buffer to be dirty, since in data=journalled mode we need
1074 * to call ext4_handle_dirty_metadata() instead.
1076 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1078 unsigned from, unsigned to)
1080 unsigned int block_start = 0, block_end;
1081 struct buffer_head *head, *bh;
1083 bh = head = page_buffers(page);
1085 block_end = block_start + bh->b_size;
1086 if (buffer_new(bh)) {
1087 if (block_end > from && block_start < to) {
1088 if (!PageUptodate(page)) {
1089 unsigned start, size;
1091 start = max(from, block_start);
1092 size = min(to, block_end) - start;
1094 zero_user(page, start, size);
1095 write_end_fn(handle, bh);
1097 clear_buffer_new(bh);
1100 block_start = block_end;
1101 bh = bh->b_this_page;
1102 } while (bh != head);
1105 static int ext4_journalled_write_end(struct file *file,
1106 struct address_space *mapping,
1107 loff_t pos, unsigned len, unsigned copied,
1108 struct page *page, void *fsdata)
1110 handle_t *handle = ext4_journal_current_handle();
1111 struct inode *inode = mapping->host;
1117 trace_ext4_journalled_write_end(inode, pos, len, copied);
1118 from = pos & (PAGE_CACHE_SIZE - 1);
1121 BUG_ON(!ext4_handle_valid(handle));
1123 if (unlikely(copied < len) && !PageUptodate(page)) {
1125 ext4_journalled_zero_new_buffers(handle, page, from, to);
1127 if (unlikely(copied < len))
1128 ext4_journalled_zero_new_buffers(handle, page,
1130 ret = walk_page_buffers(handle, page_buffers(page), from,
1131 from + copied, &partial,
1134 SetPageUptodate(page);
1136 new_i_size = pos + copied;
1137 if (new_i_size > inode->i_size)
1138 i_size_write(inode, pos+copied);
1139 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1140 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1141 if (new_i_size > EXT4_I(inode)->i_disksize) {
1142 ext4_update_i_disksize(inode, new_i_size);
1143 ret2 = ext4_mark_inode_dirty(handle, inode);
1149 page_cache_release(page);
1150 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1151 /* if we have allocated more blocks and copied
1152 * less. We will have blocks allocated outside
1153 * inode->i_size. So truncate them
1155 ext4_orphan_add(handle, inode);
1157 ret2 = ext4_journal_stop(handle);
1160 if (pos + len > inode->i_size) {
1161 ext4_truncate_failed_write(inode);
1163 * If truncate failed early the inode might still be
1164 * on the orphan list; we need to make sure the inode
1165 * is removed from the orphan list in that case.
1168 ext4_orphan_del(NULL, inode);
1171 return ret ? ret : copied;
1175 * Reserve a single cluster located at lblock
1177 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1180 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1181 struct ext4_inode_info *ei = EXT4_I(inode);
1182 unsigned int md_needed;
1184 ext4_lblk_t save_last_lblock;
1188 * We will charge metadata quota at writeout time; this saves
1189 * us from metadata over-estimation, though we may go over by
1190 * a small amount in the end. Here we just reserve for data.
1192 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1197 * recalculate the amount of metadata blocks to reserve
1198 * in order to allocate nrblocks
1199 * worse case is one extent per block
1202 spin_lock(&ei->i_block_reservation_lock);
1204 * ext4_calc_metadata_amount() has side effects, which we have
1205 * to be prepared undo if we fail to claim space.
1207 save_len = ei->i_da_metadata_calc_len;
1208 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1209 md_needed = EXT4_NUM_B2C(sbi,
1210 ext4_calc_metadata_amount(inode, lblock));
1211 trace_ext4_da_reserve_space(inode, md_needed);
1214 * We do still charge estimated metadata to the sb though;
1215 * we cannot afford to run out of free blocks.
1217 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1218 ei->i_da_metadata_calc_len = save_len;
1219 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1220 spin_unlock(&ei->i_block_reservation_lock);
1221 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1225 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1228 ei->i_reserved_data_blocks++;
1229 ei->i_reserved_meta_blocks += md_needed;
1230 spin_unlock(&ei->i_block_reservation_lock);
1232 return 0; /* success */
1235 static void ext4_da_release_space(struct inode *inode, int to_free)
1237 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1238 struct ext4_inode_info *ei = EXT4_I(inode);
1241 return; /* Nothing to release, exit */
1243 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1245 trace_ext4_da_release_space(inode, to_free);
1246 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1248 * if there aren't enough reserved blocks, then the
1249 * counter is messed up somewhere. Since this
1250 * function is called from invalidate page, it's
1251 * harmless to return without any action.
1253 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1254 "ino %lu, to_free %d with only %d reserved "
1255 "data blocks\n", inode->i_ino, to_free,
1256 ei->i_reserved_data_blocks);
1258 to_free = ei->i_reserved_data_blocks;
1260 ei->i_reserved_data_blocks -= to_free;
1262 if (ei->i_reserved_data_blocks == 0) {
1264 * We can release all of the reserved metadata blocks
1265 * only when we have written all of the delayed
1266 * allocation blocks.
1267 * Note that in case of bigalloc, i_reserved_meta_blocks,
1268 * i_reserved_data_blocks, etc. refer to number of clusters.
1270 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1271 ei->i_reserved_meta_blocks);
1272 ei->i_reserved_meta_blocks = 0;
1273 ei->i_da_metadata_calc_len = 0;
1276 /* update fs dirty data blocks counter */
1277 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1279 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1281 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1284 static void ext4_da_page_release_reservation(struct page *page,
1285 unsigned long offset)
1288 struct buffer_head *head, *bh;
1289 unsigned int curr_off = 0;
1290 struct inode *inode = page->mapping->host;
1291 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1294 head = page_buffers(page);
1297 unsigned int next_off = curr_off + bh->b_size;
1299 if ((offset <= curr_off) && (buffer_delay(bh))) {
1301 clear_buffer_delay(bh);
1302 clear_buffer_da_mapped(bh);
1304 curr_off = next_off;
1305 } while ((bh = bh->b_this_page) != head);
1307 /* If we have released all the blocks belonging to a cluster, then we
1308 * need to release the reserved space for that cluster. */
1309 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1310 while (num_clusters > 0) {
1312 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1313 ((num_clusters - 1) << sbi->s_cluster_bits);
1314 if (sbi->s_cluster_ratio == 1 ||
1315 !ext4_find_delalloc_cluster(inode, lblk, 1))
1316 ext4_da_release_space(inode, 1);
1323 * Delayed allocation stuff
1327 * mpage_da_submit_io - walks through extent of pages and try to write
1328 * them with writepage() call back
1330 * @mpd->inode: inode
1331 * @mpd->first_page: first page of the extent
1332 * @mpd->next_page: page after the last page of the extent
1334 * By the time mpage_da_submit_io() is called we expect all blocks
1335 * to be allocated. this may be wrong if allocation failed.
1337 * As pages are already locked by write_cache_pages(), we can't use it
1339 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1340 struct ext4_map_blocks *map)
1342 struct pagevec pvec;
1343 unsigned long index, end;
1344 int ret = 0, err, nr_pages, i;
1345 struct inode *inode = mpd->inode;
1346 struct address_space *mapping = inode->i_mapping;
1347 loff_t size = i_size_read(inode);
1348 unsigned int len, block_start;
1349 struct buffer_head *bh, *page_bufs = NULL;
1350 int journal_data = ext4_should_journal_data(inode);
1351 sector_t pblock = 0, cur_logical = 0;
1352 struct ext4_io_submit io_submit;
1354 BUG_ON(mpd->next_page <= mpd->first_page);
1355 memset(&io_submit, 0, sizeof(io_submit));
1357 * We need to start from the first_page to the next_page - 1
1358 * to make sure we also write the mapped dirty buffer_heads.
1359 * If we look at mpd->b_blocknr we would only be looking
1360 * at the currently mapped buffer_heads.
1362 index = mpd->first_page;
1363 end = mpd->next_page - 1;
1365 pagevec_init(&pvec, 0);
1366 while (index <= end) {
1367 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1370 for (i = 0; i < nr_pages; i++) {
1371 int commit_write = 0, skip_page = 0;
1372 struct page *page = pvec.pages[i];
1374 index = page->index;
1378 if (index == size >> PAGE_CACHE_SHIFT)
1379 len = size & ~PAGE_CACHE_MASK;
1381 len = PAGE_CACHE_SIZE;
1383 cur_logical = index << (PAGE_CACHE_SHIFT -
1385 pblock = map->m_pblk + (cur_logical -
1390 BUG_ON(!PageLocked(page));
1391 BUG_ON(PageWriteback(page));
1394 * If the page does not have buffers (for
1395 * whatever reason), try to create them using
1396 * __block_write_begin. If this fails,
1397 * skip the page and move on.
1399 if (!page_has_buffers(page)) {
1400 if (__block_write_begin(page, 0, len,
1401 noalloc_get_block_write)) {
1409 bh = page_bufs = page_buffers(page);
1414 if (map && (cur_logical >= map->m_lblk) &&
1415 (cur_logical <= (map->m_lblk +
1416 (map->m_len - 1)))) {
1417 if (buffer_delay(bh)) {
1418 clear_buffer_delay(bh);
1419 bh->b_blocknr = pblock;
1421 if (buffer_da_mapped(bh))
1422 clear_buffer_da_mapped(bh);
1423 if (buffer_unwritten(bh) ||
1425 BUG_ON(bh->b_blocknr != pblock);
1426 if (map->m_flags & EXT4_MAP_UNINIT)
1427 set_buffer_uninit(bh);
1428 clear_buffer_unwritten(bh);
1432 * skip page if block allocation undone and
1435 if (ext4_bh_delay_or_unwritten(NULL, bh))
1437 bh = bh->b_this_page;
1438 block_start += bh->b_size;
1441 } while (bh != page_bufs);
1447 /* mark the buffer_heads as dirty & uptodate */
1448 block_commit_write(page, 0, len);
1450 clear_page_dirty_for_io(page);
1452 * Delalloc doesn't support data journalling,
1453 * but eventually maybe we'll lift this
1456 if (unlikely(journal_data && PageChecked(page)))
1457 err = __ext4_journalled_writepage(page, len);
1458 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1459 err = ext4_bio_write_page(&io_submit, page,
1461 else if (buffer_uninit(page_bufs)) {
1462 ext4_set_bh_endio(page_bufs, inode);
1463 err = block_write_full_page_endio(page,
1464 noalloc_get_block_write,
1465 mpd->wbc, ext4_end_io_buffer_write);
1467 err = block_write_full_page(page,
1468 noalloc_get_block_write, mpd->wbc);
1471 mpd->pages_written++;
1473 * In error case, we have to continue because
1474 * remaining pages are still locked
1479 pagevec_release(&pvec);
1481 ext4_io_submit(&io_submit);
1485 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1489 struct pagevec pvec;
1490 struct inode *inode = mpd->inode;
1491 struct address_space *mapping = inode->i_mapping;
1493 index = mpd->first_page;
1494 end = mpd->next_page - 1;
1495 pagevec_init(&pvec, 0);
1496 while (index <= end) {
1497 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1500 for (i = 0; i < nr_pages; i++) {
1501 struct page *page = pvec.pages[i];
1502 if (page->index > end)
1504 BUG_ON(!PageLocked(page));
1505 BUG_ON(PageWriteback(page));
1506 block_invalidatepage(page, 0);
1507 ClearPageUptodate(page);
1510 index = pvec.pages[nr_pages - 1]->index + 1;
1511 pagevec_release(&pvec);
1516 static void ext4_print_free_blocks(struct inode *inode)
1518 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1519 printk(KERN_CRIT "Total free blocks count %lld\n",
1520 EXT4_C2B(EXT4_SB(inode->i_sb),
1521 ext4_count_free_clusters(inode->i_sb)));
1522 printk(KERN_CRIT "Free/Dirty block details\n");
1523 printk(KERN_CRIT "free_blocks=%lld\n",
1524 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1525 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1526 printk(KERN_CRIT "dirty_blocks=%lld\n",
1527 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1528 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1529 printk(KERN_CRIT "Block reservation details\n");
1530 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
1531 EXT4_I(inode)->i_reserved_data_blocks);
1532 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
1533 EXT4_I(inode)->i_reserved_meta_blocks);
1538 * mpage_da_map_and_submit - go through given space, map them
1539 * if necessary, and then submit them for I/O
1541 * @mpd - bh describing space
1543 * The function skips space we know is already mapped to disk blocks.
1546 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1548 int err, blks, get_blocks_flags;
1549 struct ext4_map_blocks map, *mapp = NULL;
1550 sector_t next = mpd->b_blocknr;
1551 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1552 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1553 handle_t *handle = NULL;
1556 * If the blocks are mapped already, or we couldn't accumulate
1557 * any blocks, then proceed immediately to the submission stage.
1559 if ((mpd->b_size == 0) ||
1560 ((mpd->b_state & (1 << BH_Mapped)) &&
1561 !(mpd->b_state & (1 << BH_Delay)) &&
1562 !(mpd->b_state & (1 << BH_Unwritten))))
1565 handle = ext4_journal_current_handle();
1569 * Call ext4_map_blocks() to allocate any delayed allocation
1570 * blocks, or to convert an uninitialized extent to be
1571 * initialized (in the case where we have written into
1572 * one or more preallocated blocks).
1574 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1575 * indicate that we are on the delayed allocation path. This
1576 * affects functions in many different parts of the allocation
1577 * call path. This flag exists primarily because we don't
1578 * want to change *many* call functions, so ext4_map_blocks()
1579 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1580 * inode's allocation semaphore is taken.
1582 * If the blocks in questions were delalloc blocks, set
1583 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1584 * variables are updated after the blocks have been allocated.
1587 map.m_len = max_blocks;
1588 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1589 if (ext4_should_dioread_nolock(mpd->inode))
1590 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1591 if (mpd->b_state & (1 << BH_Delay))
1592 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1594 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1596 struct super_block *sb = mpd->inode->i_sb;
1600 * If get block returns EAGAIN or ENOSPC and there
1601 * appears to be free blocks we will just let
1602 * mpage_da_submit_io() unlock all of the pages.
1607 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1613 * get block failure will cause us to loop in
1614 * writepages, because a_ops->writepage won't be able
1615 * to make progress. The page will be redirtied by
1616 * writepage and writepages will again try to write
1619 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1620 ext4_msg(sb, KERN_CRIT,
1621 "delayed block allocation failed for inode %lu "
1622 "at logical offset %llu with max blocks %zd "
1623 "with error %d", mpd->inode->i_ino,
1624 (unsigned long long) next,
1625 mpd->b_size >> mpd->inode->i_blkbits, err);
1626 ext4_msg(sb, KERN_CRIT,
1627 "This should not happen!! Data will be lost\n");
1629 ext4_print_free_blocks(mpd->inode);
1631 /* invalidate all the pages */
1632 ext4_da_block_invalidatepages(mpd);
1634 /* Mark this page range as having been completed */
1641 if (map.m_flags & EXT4_MAP_NEW) {
1642 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1645 for (i = 0; i < map.m_len; i++)
1646 unmap_underlying_metadata(bdev, map.m_pblk + i);
1648 if (ext4_should_order_data(mpd->inode)) {
1649 err = ext4_jbd2_file_inode(handle, mpd->inode);
1651 /* Only if the journal is aborted */
1659 * Update on-disk size along with block allocation.
1661 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1662 if (disksize > i_size_read(mpd->inode))
1663 disksize = i_size_read(mpd->inode);
1664 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1665 ext4_update_i_disksize(mpd->inode, disksize);
1666 err = ext4_mark_inode_dirty(handle, mpd->inode);
1668 ext4_error(mpd->inode->i_sb,
1669 "Failed to mark inode %lu dirty",
1674 mpage_da_submit_io(mpd, mapp);
1678 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1679 (1 << BH_Delay) | (1 << BH_Unwritten))
1682 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1684 * @mpd->lbh - extent of blocks
1685 * @logical - logical number of the block in the file
1686 * @bh - bh of the block (used to access block's state)
1688 * the function is used to collect contig. blocks in same state
1690 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1691 sector_t logical, size_t b_size,
1692 unsigned long b_state)
1695 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1698 * XXX Don't go larger than mballoc is willing to allocate
1699 * This is a stopgap solution. We eventually need to fold
1700 * mpage_da_submit_io() into this function and then call
1701 * ext4_map_blocks() multiple times in a loop
1703 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1706 /* check if thereserved journal credits might overflow */
1707 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1708 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1710 * With non-extent format we are limited by the journal
1711 * credit available. Total credit needed to insert
1712 * nrblocks contiguous blocks is dependent on the
1713 * nrblocks. So limit nrblocks.
1716 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1717 EXT4_MAX_TRANS_DATA) {
1719 * Adding the new buffer_head would make it cross the
1720 * allowed limit for which we have journal credit
1721 * reserved. So limit the new bh->b_size
1723 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1724 mpd->inode->i_blkbits;
1725 /* we will do mpage_da_submit_io in the next loop */
1729 * First block in the extent
1731 if (mpd->b_size == 0) {
1732 mpd->b_blocknr = logical;
1733 mpd->b_size = b_size;
1734 mpd->b_state = b_state & BH_FLAGS;
1738 next = mpd->b_blocknr + nrblocks;
1740 * Can we merge the block to our big extent?
1742 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1743 mpd->b_size += b_size;
1749 * We couldn't merge the block to our extent, so we
1750 * need to flush current extent and start new one
1752 mpage_da_map_and_submit(mpd);
1756 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1758 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1762 * This function is grabs code from the very beginning of
1763 * ext4_map_blocks, but assumes that the caller is from delayed write
1764 * time. This function looks up the requested blocks and sets the
1765 * buffer delay bit under the protection of i_data_sem.
1767 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1768 struct ext4_map_blocks *map,
1769 struct buffer_head *bh)
1772 sector_t invalid_block = ~((sector_t) 0xffff);
1774 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1778 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1779 "logical block %lu\n", inode->i_ino, map->m_len,
1780 (unsigned long) map->m_lblk);
1782 * Try to see if we can get the block without requesting a new
1783 * file system block.
1785 down_read((&EXT4_I(inode)->i_data_sem));
1786 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1787 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1789 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1793 * XXX: __block_prepare_write() unmaps passed block,
1796 /* If the block was allocated from previously allocated cluster,
1797 * then we dont need to reserve it again. */
1798 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1799 retval = ext4_da_reserve_space(inode, iblock);
1801 /* not enough space to reserve */
1805 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1806 * and it should not appear on the bh->b_state.
1808 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1810 map_bh(bh, inode->i_sb, invalid_block);
1812 set_buffer_delay(bh);
1816 up_read((&EXT4_I(inode)->i_data_sem));
1822 * This is a special get_blocks_t callback which is used by
1823 * ext4_da_write_begin(). It will either return mapped block or
1824 * reserve space for a single block.
1826 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1827 * We also have b_blocknr = -1 and b_bdev initialized properly
1829 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1830 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1831 * initialized properly.
1833 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1834 struct buffer_head *bh, int create)
1836 struct ext4_map_blocks map;
1839 BUG_ON(create == 0);
1840 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1842 map.m_lblk = iblock;
1846 * first, we need to know whether the block is allocated already
1847 * preallocated blocks are unmapped but should treated
1848 * the same as allocated blocks.
1850 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1854 map_bh(bh, inode->i_sb, map.m_pblk);
1855 ext4_update_bh_state(bh, map.m_flags);
1857 if (buffer_unwritten(bh)) {
1858 /* A delayed write to unwritten bh should be marked
1859 * new and mapped. Mapped ensures that we don't do
1860 * get_block multiple times when we write to the same
1861 * offset and new ensures that we do proper zero out
1862 * for partial write.
1865 set_buffer_mapped(bh);
1871 * This function is used as a standard get_block_t calback function
1872 * when there is no desire to allocate any blocks. It is used as a
1873 * callback function for block_write_begin() and block_write_full_page().
1874 * These functions should only try to map a single block at a time.
1876 * Since this function doesn't do block allocations even if the caller
1877 * requests it by passing in create=1, it is critically important that
1878 * any caller checks to make sure that any buffer heads are returned
1879 * by this function are either all already mapped or marked for
1880 * delayed allocation before calling block_write_full_page(). Otherwise,
1881 * b_blocknr could be left unitialized, and the page write functions will
1882 * be taken by surprise.
1884 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1885 struct buffer_head *bh_result, int create)
1887 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1888 return _ext4_get_block(inode, iblock, bh_result, 0);
1891 static int bget_one(handle_t *handle, struct buffer_head *bh)
1897 static int bput_one(handle_t *handle, struct buffer_head *bh)
1903 static int __ext4_journalled_writepage(struct page *page,
1906 struct address_space *mapping = page->mapping;
1907 struct inode *inode = mapping->host;
1908 struct buffer_head *page_bufs;
1909 handle_t *handle = NULL;
1913 ClearPageChecked(page);
1914 page_bufs = page_buffers(page);
1916 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1918 * We need to release the page lock before we start the
1919 * journal, so grab a reference so the page won't disappear
1920 * out from under us.
1925 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1926 if (IS_ERR(handle)) {
1927 ret = PTR_ERR(handle);
1929 goto out_no_pagelock;
1931 BUG_ON(!ext4_handle_valid(handle));
1935 if (page->mapping != mapping) {
1936 /* The page got truncated from under us */
1937 ext4_journal_stop(handle);
1942 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1943 do_journal_get_write_access);
1945 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1949 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1950 err = ext4_journal_stop(handle);
1954 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
1955 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1962 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
1963 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
1966 * Note that we don't need to start a transaction unless we're journaling data
1967 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1968 * need to file the inode to the transaction's list in ordered mode because if
1969 * we are writing back data added by write(), the inode is already there and if
1970 * we are writing back data modified via mmap(), no one guarantees in which
1971 * transaction the data will hit the disk. In case we are journaling data, we
1972 * cannot start transaction directly because transaction start ranks above page
1973 * lock so we have to do some magic.
1975 * This function can get called via...
1976 * - ext4_da_writepages after taking page lock (have journal handle)
1977 * - journal_submit_inode_data_buffers (no journal handle)
1978 * - shrink_page_list via pdflush (no journal handle)
1979 * - grab_page_cache when doing write_begin (have journal handle)
1981 * We don't do any block allocation in this function. If we have page with
1982 * multiple blocks we need to write those buffer_heads that are mapped. This
1983 * is important for mmaped based write. So if we do with blocksize 1K
1984 * truncate(f, 1024);
1985 * a = mmap(f, 0, 4096);
1987 * truncate(f, 4096);
1988 * we have in the page first buffer_head mapped via page_mkwrite call back
1989 * but other bufer_heads would be unmapped but dirty(dirty done via the
1990 * do_wp_page). So writepage should write the first block. If we modify
1991 * the mmap area beyond 1024 we will again get a page_fault and the
1992 * page_mkwrite callback will do the block allocation and mark the
1993 * buffer_heads mapped.
1995 * We redirty the page if we have any buffer_heads that is either delay or
1996 * unwritten in the page.
1998 * We can get recursively called as show below.
2000 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2003 * But since we don't do any block allocation we should not deadlock.
2004 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2006 static int ext4_writepage(struct page *page,
2007 struct writeback_control *wbc)
2009 int ret = 0, commit_write = 0;
2012 struct buffer_head *page_bufs = NULL;
2013 struct inode *inode = page->mapping->host;
2015 trace_ext4_writepage(page);
2016 size = i_size_read(inode);
2017 if (page->index == size >> PAGE_CACHE_SHIFT)
2018 len = size & ~PAGE_CACHE_MASK;
2020 len = PAGE_CACHE_SIZE;
2023 * If the page does not have buffers (for whatever reason),
2024 * try to create them using __block_write_begin. If this
2025 * fails, redirty the page and move on.
2027 if (!page_has_buffers(page)) {
2028 if (__block_write_begin(page, 0, len,
2029 noalloc_get_block_write)) {
2031 redirty_page_for_writepage(wbc, page);
2037 page_bufs = page_buffers(page);
2038 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2039 ext4_bh_delay_or_unwritten)) {
2041 * We don't want to do block allocation, so redirty
2042 * the page and return. We may reach here when we do
2043 * a journal commit via journal_submit_inode_data_buffers.
2044 * We can also reach here via shrink_page_list but it
2045 * should never be for direct reclaim so warn if that
2048 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
2053 /* now mark the buffer_heads as dirty and uptodate */
2054 block_commit_write(page, 0, len);
2056 if (PageChecked(page) && ext4_should_journal_data(inode))
2058 * It's mmapped pagecache. Add buffers and journal it. There
2059 * doesn't seem much point in redirtying the page here.
2061 return __ext4_journalled_writepage(page, len);
2063 if (buffer_uninit(page_bufs)) {
2064 ext4_set_bh_endio(page_bufs, inode);
2065 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2066 wbc, ext4_end_io_buffer_write);
2068 ret = block_write_full_page(page, noalloc_get_block_write,
2075 * This is called via ext4_da_writepages() to
2076 * calculate the total number of credits to reserve to fit
2077 * a single extent allocation into a single transaction,
2078 * ext4_da_writpeages() will loop calling this before
2079 * the block allocation.
2082 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2084 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2087 * With non-extent format the journal credit needed to
2088 * insert nrblocks contiguous block is dependent on
2089 * number of contiguous block. So we will limit
2090 * number of contiguous block to a sane value
2092 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2093 (max_blocks > EXT4_MAX_TRANS_DATA))
2094 max_blocks = EXT4_MAX_TRANS_DATA;
2096 return ext4_chunk_trans_blocks(inode, max_blocks);
2100 * write_cache_pages_da - walk the list of dirty pages of the given
2101 * address space and accumulate pages that need writing, and call
2102 * mpage_da_map_and_submit to map a single contiguous memory region
2103 * and then write them.
2105 static int write_cache_pages_da(struct address_space *mapping,
2106 struct writeback_control *wbc,
2107 struct mpage_da_data *mpd,
2108 pgoff_t *done_index)
2110 struct buffer_head *bh, *head;
2111 struct inode *inode = mapping->host;
2112 struct pagevec pvec;
2113 unsigned int nr_pages;
2116 long nr_to_write = wbc->nr_to_write;
2117 int i, tag, ret = 0;
2119 memset(mpd, 0, sizeof(struct mpage_da_data));
2122 pagevec_init(&pvec, 0);
2123 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2124 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2126 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2127 tag = PAGECACHE_TAG_TOWRITE;
2129 tag = PAGECACHE_TAG_DIRTY;
2131 *done_index = index;
2132 while (index <= end) {
2133 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2134 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2138 for (i = 0; i < nr_pages; i++) {
2139 struct page *page = pvec.pages[i];
2142 * At this point, the page may be truncated or
2143 * invalidated (changing page->mapping to NULL), or
2144 * even swizzled back from swapper_space to tmpfs file
2145 * mapping. However, page->index will not change
2146 * because we have a reference on the page.
2148 if (page->index > end)
2151 *done_index = page->index + 1;
2154 * If we can't merge this page, and we have
2155 * accumulated an contiguous region, write it
2157 if ((mpd->next_page != page->index) &&
2158 (mpd->next_page != mpd->first_page)) {
2159 mpage_da_map_and_submit(mpd);
2160 goto ret_extent_tail;
2166 * If the page is no longer dirty, or its
2167 * mapping no longer corresponds to inode we
2168 * are writing (which means it has been
2169 * truncated or invalidated), or the page is
2170 * already under writeback and we are not
2171 * doing a data integrity writeback, skip the page
2173 if (!PageDirty(page) ||
2174 (PageWriteback(page) &&
2175 (wbc->sync_mode == WB_SYNC_NONE)) ||
2176 unlikely(page->mapping != mapping)) {
2181 wait_on_page_writeback(page);
2182 BUG_ON(PageWriteback(page));
2184 if (mpd->next_page != page->index)
2185 mpd->first_page = page->index;
2186 mpd->next_page = page->index + 1;
2187 logical = (sector_t) page->index <<
2188 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2190 if (!page_has_buffers(page)) {
2191 mpage_add_bh_to_extent(mpd, logical,
2193 (1 << BH_Dirty) | (1 << BH_Uptodate));
2195 goto ret_extent_tail;
2198 * Page with regular buffer heads,
2199 * just add all dirty ones
2201 head = page_buffers(page);
2204 BUG_ON(buffer_locked(bh));
2206 * We need to try to allocate
2207 * unmapped blocks in the same page.
2208 * Otherwise we won't make progress
2209 * with the page in ext4_writepage
2211 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2212 mpage_add_bh_to_extent(mpd, logical,
2216 goto ret_extent_tail;
2217 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2219 * mapped dirty buffer. We need
2220 * to update the b_state
2221 * because we look at b_state
2222 * in mpage_da_map_blocks. We
2223 * don't update b_size because
2224 * if we find an unmapped
2225 * buffer_head later we need to
2226 * use the b_state flag of that
2229 if (mpd->b_size == 0)
2230 mpd->b_state = bh->b_state & BH_FLAGS;
2233 } while ((bh = bh->b_this_page) != head);
2236 if (nr_to_write > 0) {
2238 if (nr_to_write == 0 &&
2239 wbc->sync_mode == WB_SYNC_NONE)
2241 * We stop writing back only if we are
2242 * not doing integrity sync. In case of
2243 * integrity sync we have to keep going
2244 * because someone may be concurrently
2245 * dirtying pages, and we might have
2246 * synced a lot of newly appeared dirty
2247 * pages, but have not synced all of the
2253 pagevec_release(&pvec);
2258 ret = MPAGE_DA_EXTENT_TAIL;
2260 pagevec_release(&pvec);
2266 static int ext4_da_writepages(struct address_space *mapping,
2267 struct writeback_control *wbc)
2270 int range_whole = 0;
2271 handle_t *handle = NULL;
2272 struct mpage_da_data mpd;
2273 struct inode *inode = mapping->host;
2274 int pages_written = 0;
2275 unsigned int max_pages;
2276 int range_cyclic, cycled = 1, io_done = 0;
2277 int needed_blocks, ret = 0;
2278 long desired_nr_to_write, nr_to_writebump = 0;
2279 loff_t range_start = wbc->range_start;
2280 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2281 pgoff_t done_index = 0;
2283 struct blk_plug plug;
2285 trace_ext4_da_writepages(inode, wbc);
2288 * No pages to write? This is mainly a kludge to avoid starting
2289 * a transaction for special inodes like journal inode on last iput()
2290 * because that could violate lock ordering on umount
2292 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2296 * If the filesystem has aborted, it is read-only, so return
2297 * right away instead of dumping stack traces later on that
2298 * will obscure the real source of the problem. We test
2299 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2300 * the latter could be true if the filesystem is mounted
2301 * read-only, and in that case, ext4_da_writepages should
2302 * *never* be called, so if that ever happens, we would want
2305 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2308 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2311 range_cyclic = wbc->range_cyclic;
2312 if (wbc->range_cyclic) {
2313 index = mapping->writeback_index;
2316 wbc->range_start = index << PAGE_CACHE_SHIFT;
2317 wbc->range_end = LLONG_MAX;
2318 wbc->range_cyclic = 0;
2321 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2322 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2326 * This works around two forms of stupidity. The first is in
2327 * the writeback code, which caps the maximum number of pages
2328 * written to be 1024 pages. This is wrong on multiple
2329 * levels; different architectues have a different page size,
2330 * which changes the maximum amount of data which gets
2331 * written. Secondly, 4 megabytes is way too small. XFS
2332 * forces this value to be 16 megabytes by multiplying
2333 * nr_to_write parameter by four, and then relies on its
2334 * allocator to allocate larger extents to make them
2335 * contiguous. Unfortunately this brings us to the second
2336 * stupidity, which is that ext4's mballoc code only allocates
2337 * at most 2048 blocks. So we force contiguous writes up to
2338 * the number of dirty blocks in the inode, or
2339 * sbi->max_writeback_mb_bump whichever is smaller.
2341 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2342 if (!range_cyclic && range_whole) {
2343 if (wbc->nr_to_write == LONG_MAX)
2344 desired_nr_to_write = wbc->nr_to_write;
2346 desired_nr_to_write = wbc->nr_to_write * 8;
2348 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2350 if (desired_nr_to_write > max_pages)
2351 desired_nr_to_write = max_pages;
2353 if (wbc->nr_to_write < desired_nr_to_write) {
2354 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2355 wbc->nr_to_write = desired_nr_to_write;
2359 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2360 tag_pages_for_writeback(mapping, index, end);
2362 blk_start_plug(&plug);
2363 while (!ret && wbc->nr_to_write > 0) {
2366 * we insert one extent at a time. So we need
2367 * credit needed for single extent allocation.
2368 * journalled mode is currently not supported
2371 BUG_ON(ext4_should_journal_data(inode));
2372 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2374 /* start a new transaction*/
2375 handle = ext4_journal_start(inode, needed_blocks);
2376 if (IS_ERR(handle)) {
2377 ret = PTR_ERR(handle);
2378 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2379 "%ld pages, ino %lu; err %d", __func__,
2380 wbc->nr_to_write, inode->i_ino, ret);
2381 blk_finish_plug(&plug);
2382 goto out_writepages;
2386 * Now call write_cache_pages_da() to find the next
2387 * contiguous region of logical blocks that need
2388 * blocks to be allocated by ext4 and submit them.
2390 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
2392 * If we have a contiguous extent of pages and we
2393 * haven't done the I/O yet, map the blocks and submit
2396 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2397 mpage_da_map_and_submit(&mpd);
2398 ret = MPAGE_DA_EXTENT_TAIL;
2400 trace_ext4_da_write_pages(inode, &mpd);
2401 wbc->nr_to_write -= mpd.pages_written;
2403 ext4_journal_stop(handle);
2405 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2406 /* commit the transaction which would
2407 * free blocks released in the transaction
2410 jbd2_journal_force_commit_nested(sbi->s_journal);
2412 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2414 * Got one extent now try with rest of the pages.
2415 * If mpd.retval is set -EIO, journal is aborted.
2416 * So we don't need to write any more.
2418 pages_written += mpd.pages_written;
2421 } else if (wbc->nr_to_write)
2423 * There is no more writeout needed
2424 * or we requested for a noblocking writeout
2425 * and we found the device congested
2429 blk_finish_plug(&plug);
2430 if (!io_done && !cycled) {
2433 wbc->range_start = index << PAGE_CACHE_SHIFT;
2434 wbc->range_end = mapping->writeback_index - 1;
2439 wbc->range_cyclic = range_cyclic;
2440 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2442 * set the writeback_index so that range_cyclic
2443 * mode will write it back later
2445 mapping->writeback_index = done_index;
2448 wbc->nr_to_write -= nr_to_writebump;
2449 wbc->range_start = range_start;
2450 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2454 #define FALL_BACK_TO_NONDELALLOC 1
2455 static int ext4_nonda_switch(struct super_block *sb)
2457 s64 free_blocks, dirty_blocks;
2458 struct ext4_sb_info *sbi = EXT4_SB(sb);
2461 * switch to non delalloc mode if we are running low
2462 * on free block. The free block accounting via percpu
2463 * counters can get slightly wrong with percpu_counter_batch getting
2464 * accumulated on each CPU without updating global counters
2465 * Delalloc need an accurate free block accounting. So switch
2466 * to non delalloc when we are near to error range.
2468 free_blocks = EXT4_C2B(sbi,
2469 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2470 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2472 * Start pushing delalloc when 1/2 of free blocks are dirty.
2474 if (dirty_blocks && (free_blocks < 2 * dirty_blocks) &&
2475 !writeback_in_progress(sb->s_bdi) &&
2476 down_read_trylock(&sb->s_umount)) {
2477 writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2478 up_read(&sb->s_umount);
2481 if (2 * free_blocks < 3 * dirty_blocks ||
2482 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2484 * free block count is less than 150% of dirty blocks
2485 * or free blocks is less than watermark
2492 /* We always reserve for an inode update; the superblock could be there too */
2493 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2495 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
2496 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2499 if (pos + len <= 0x7fffffffULL)
2502 /* We might need to update the superblock to set LARGE_FILE */
2506 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2507 loff_t pos, unsigned len, unsigned flags,
2508 struct page **pagep, void **fsdata)
2510 int ret, retries = 0;
2513 struct inode *inode = mapping->host;
2516 index = pos >> PAGE_CACHE_SHIFT;
2518 if (ext4_nonda_switch(inode->i_sb)) {
2519 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2520 return ext4_write_begin(file, mapping, pos,
2521 len, flags, pagep, fsdata);
2523 *fsdata = (void *)0;
2524 trace_ext4_da_write_begin(inode, pos, len, flags);
2527 * With delayed allocation, we don't log the i_disksize update
2528 * if there is delayed block allocation. But we still need
2529 * to journalling the i_disksize update if writes to the end
2530 * of file which has an already mapped buffer.
2532 handle = ext4_journal_start(inode,
2533 ext4_da_write_credits(inode, pos, len));
2534 if (IS_ERR(handle)) {
2535 ret = PTR_ERR(handle);
2538 /* We cannot recurse into the filesystem as the transaction is already
2540 flags |= AOP_FLAG_NOFS;
2542 page = grab_cache_page_write_begin(mapping, index, flags);
2544 ext4_journal_stop(handle);
2550 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2553 ext4_journal_stop(handle);
2554 page_cache_release(page);
2556 * block_write_begin may have instantiated a few blocks
2557 * outside i_size. Trim these off again. Don't need
2558 * i_size_read because we hold i_mutex.
2560 if (pos + len > inode->i_size)
2561 ext4_truncate_failed_write(inode);
2564 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2571 * Check if we should update i_disksize
2572 * when write to the end of file but not require block allocation
2574 static int ext4_da_should_update_i_disksize(struct page *page,
2575 unsigned long offset)
2577 struct buffer_head *bh;
2578 struct inode *inode = page->mapping->host;
2582 bh = page_buffers(page);
2583 idx = offset >> inode->i_blkbits;
2585 for (i = 0; i < idx; i++)
2586 bh = bh->b_this_page;
2588 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2593 static int ext4_da_write_end(struct file *file,
2594 struct address_space *mapping,
2595 loff_t pos, unsigned len, unsigned copied,
2596 struct page *page, void *fsdata)
2598 struct inode *inode = mapping->host;
2600 handle_t *handle = ext4_journal_current_handle();
2602 unsigned long start, end;
2603 int write_mode = (int)(unsigned long)fsdata;
2605 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2606 switch (ext4_inode_journal_mode(inode)) {
2607 case EXT4_INODE_ORDERED_DATA_MODE:
2608 return ext4_ordered_write_end(file, mapping, pos,
2609 len, copied, page, fsdata);
2610 case EXT4_INODE_WRITEBACK_DATA_MODE:
2611 return ext4_writeback_write_end(file, mapping, pos,
2612 len, copied, page, fsdata);
2618 trace_ext4_da_write_end(inode, pos, len, copied);
2619 start = pos & (PAGE_CACHE_SIZE - 1);
2620 end = start + copied - 1;
2623 * generic_write_end() will run mark_inode_dirty() if i_size
2624 * changes. So let's piggyback the i_disksize mark_inode_dirty
2628 new_i_size = pos + copied;
2629 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2630 if (ext4_da_should_update_i_disksize(page, end)) {
2631 down_write(&EXT4_I(inode)->i_data_sem);
2632 if (new_i_size > EXT4_I(inode)->i_disksize) {
2634 * Updating i_disksize when extending file
2635 * without needing block allocation
2637 if (ext4_should_order_data(inode))
2638 ret = ext4_jbd2_file_inode(handle,
2641 EXT4_I(inode)->i_disksize = new_i_size;
2643 up_write(&EXT4_I(inode)->i_data_sem);
2644 /* We need to mark inode dirty even if
2645 * new_i_size is less that inode->i_size
2646 * bu greater than i_disksize.(hint delalloc)
2648 ext4_mark_inode_dirty(handle, inode);
2651 ret2 = generic_write_end(file, mapping, pos, len, copied,
2656 ret2 = ext4_journal_stop(handle);
2660 return ret ? ret : copied;
2663 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2666 * Drop reserved blocks
2668 BUG_ON(!PageLocked(page));
2669 if (!page_has_buffers(page))
2672 ext4_da_page_release_reservation(page, offset);
2675 ext4_invalidatepage(page, offset);
2681 * Force all delayed allocation blocks to be allocated for a given inode.
2683 int ext4_alloc_da_blocks(struct inode *inode)
2685 trace_ext4_alloc_da_blocks(inode);
2687 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2688 !EXT4_I(inode)->i_reserved_meta_blocks)
2692 * We do something simple for now. The filemap_flush() will
2693 * also start triggering a write of the data blocks, which is
2694 * not strictly speaking necessary (and for users of
2695 * laptop_mode, not even desirable). However, to do otherwise
2696 * would require replicating code paths in:
2698 * ext4_da_writepages() ->
2699 * write_cache_pages() ---> (via passed in callback function)
2700 * __mpage_da_writepage() -->
2701 * mpage_add_bh_to_extent()
2702 * mpage_da_map_blocks()
2704 * The problem is that write_cache_pages(), located in
2705 * mm/page-writeback.c, marks pages clean in preparation for
2706 * doing I/O, which is not desirable if we're not planning on
2709 * We could call write_cache_pages(), and then redirty all of
2710 * the pages by calling redirty_page_for_writepage() but that
2711 * would be ugly in the extreme. So instead we would need to
2712 * replicate parts of the code in the above functions,
2713 * simplifying them because we wouldn't actually intend to
2714 * write out the pages, but rather only collect contiguous
2715 * logical block extents, call the multi-block allocator, and
2716 * then update the buffer heads with the block allocations.
2718 * For now, though, we'll cheat by calling filemap_flush(),
2719 * which will map the blocks, and start the I/O, but not
2720 * actually wait for the I/O to complete.
2722 return filemap_flush(inode->i_mapping);
2726 * bmap() is special. It gets used by applications such as lilo and by
2727 * the swapper to find the on-disk block of a specific piece of data.
2729 * Naturally, this is dangerous if the block concerned is still in the
2730 * journal. If somebody makes a swapfile on an ext4 data-journaling
2731 * filesystem and enables swap, then they may get a nasty shock when the
2732 * data getting swapped to that swapfile suddenly gets overwritten by
2733 * the original zero's written out previously to the journal and
2734 * awaiting writeback in the kernel's buffer cache.
2736 * So, if we see any bmap calls here on a modified, data-journaled file,
2737 * take extra steps to flush any blocks which might be in the cache.
2739 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2741 struct inode *inode = mapping->host;
2745 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2746 test_opt(inode->i_sb, DELALLOC)) {
2748 * With delalloc we want to sync the file
2749 * so that we can make sure we allocate
2752 filemap_write_and_wait(mapping);
2755 if (EXT4_JOURNAL(inode) &&
2756 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2758 * This is a REALLY heavyweight approach, but the use of
2759 * bmap on dirty files is expected to be extremely rare:
2760 * only if we run lilo or swapon on a freshly made file
2761 * do we expect this to happen.
2763 * (bmap requires CAP_SYS_RAWIO so this does not
2764 * represent an unprivileged user DOS attack --- we'd be
2765 * in trouble if mortal users could trigger this path at
2768 * NB. EXT4_STATE_JDATA is not set on files other than
2769 * regular files. If somebody wants to bmap a directory
2770 * or symlink and gets confused because the buffer
2771 * hasn't yet been flushed to disk, they deserve
2772 * everything they get.
2775 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2776 journal = EXT4_JOURNAL(inode);
2777 jbd2_journal_lock_updates(journal);
2778 err = jbd2_journal_flush(journal);
2779 jbd2_journal_unlock_updates(journal);
2785 return generic_block_bmap(mapping, block, ext4_get_block);
2788 static int ext4_readpage(struct file *file, struct page *page)
2790 trace_ext4_readpage(page);
2791 return mpage_readpage(page, ext4_get_block);
2795 ext4_readpages(struct file *file, struct address_space *mapping,
2796 struct list_head *pages, unsigned nr_pages)
2798 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2801 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
2803 struct buffer_head *head, *bh;
2804 unsigned int curr_off = 0;
2806 if (!page_has_buffers(page))
2808 head = bh = page_buffers(page);
2810 if (offset <= curr_off && test_clear_buffer_uninit(bh)
2812 ext4_free_io_end(bh->b_private);
2813 bh->b_private = NULL;
2814 bh->b_end_io = NULL;
2816 curr_off = curr_off + bh->b_size;
2817 bh = bh->b_this_page;
2818 } while (bh != head);
2821 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2823 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2825 trace_ext4_invalidatepage(page, offset);
2828 * free any io_end structure allocated for buffers to be discarded
2830 if (ext4_should_dioread_nolock(page->mapping->host))
2831 ext4_invalidatepage_free_endio(page, offset);
2833 * If it's a full truncate we just forget about the pending dirtying
2836 ClearPageChecked(page);
2839 jbd2_journal_invalidatepage(journal, page, offset);
2841 block_invalidatepage(page, offset);
2844 static int ext4_releasepage(struct page *page, gfp_t wait)
2846 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2848 trace_ext4_releasepage(page);
2850 WARN_ON(PageChecked(page));
2851 if (!page_has_buffers(page))
2854 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2856 return try_to_free_buffers(page);
2860 * ext4_get_block used when preparing for a DIO write or buffer write.
2861 * We allocate an uinitialized extent if blocks haven't been allocated.
2862 * The extent will be converted to initialized after the IO is complete.
2864 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
2865 struct buffer_head *bh_result, int create)
2867 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2868 inode->i_ino, create);
2869 return _ext4_get_block(inode, iblock, bh_result,
2870 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2873 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2874 ssize_t size, void *private, int ret,
2877 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2878 ext4_io_end_t *io_end = iocb->private;
2879 struct workqueue_struct *wq;
2880 unsigned long flags;
2881 struct ext4_inode_info *ei;
2883 /* if not async direct IO or dio with 0 bytes write, just return */
2884 if (!io_end || !size)
2887 ext_debug("ext4_end_io_dio(): io_end 0x%p"
2888 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
2889 iocb->private, io_end->inode->i_ino, iocb, offset,
2892 iocb->private = NULL;
2894 /* if not aio dio with unwritten extents, just free io and return */
2895 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2896 ext4_free_io_end(io_end);
2898 inode_dio_done(inode);
2900 aio_complete(iocb, ret, 0);
2904 io_end->offset = offset;
2905 io_end->size = size;
2907 io_end->iocb = iocb;
2908 io_end->result = ret;
2910 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
2912 /* Add the io_end to per-inode completed aio dio list*/
2913 ei = EXT4_I(io_end->inode);
2914 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
2915 list_add_tail(&io_end->list, &ei->i_completed_io_list);
2916 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
2918 /* queue the work to convert unwritten extents to written */
2919 queue_work(wq, &io_end->work);
2922 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
2924 ext4_io_end_t *io_end = bh->b_private;
2925 struct workqueue_struct *wq;
2926 struct inode *inode;
2927 unsigned long flags;
2929 if (!test_clear_buffer_uninit(bh) || !io_end)
2932 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2933 printk("sb umounted, discard end_io request for inode %lu\n",
2934 io_end->inode->i_ino);
2935 ext4_free_io_end(io_end);
2940 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2941 * but being more careful is always safe for the future change.
2943 inode = io_end->inode;
2944 ext4_set_io_unwritten_flag(inode, io_end);
2946 /* Add the io_end to per-inode completed io list*/
2947 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
2948 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
2949 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
2951 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
2952 /* queue the work to convert unwritten extents to written */
2953 queue_work(wq, &io_end->work);
2955 bh->b_private = NULL;
2956 bh->b_end_io = NULL;
2957 clear_buffer_uninit(bh);
2958 end_buffer_async_write(bh, uptodate);
2961 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2963 ext4_io_end_t *io_end;
2964 struct page *page = bh->b_page;
2965 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2966 size_t size = bh->b_size;
2969 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2971 pr_warn_ratelimited("%s: allocation fail\n", __func__);
2975 io_end->offset = offset;
2976 io_end->size = size;
2978 * We need to hold a reference to the page to make sure it
2979 * doesn't get evicted before ext4_end_io_work() has a chance
2980 * to convert the extent from written to unwritten.
2982 io_end->page = page;
2983 get_page(io_end->page);
2985 bh->b_private = io_end;
2986 bh->b_end_io = ext4_end_io_buffer_write;
2991 * For ext4 extent files, ext4 will do direct-io write to holes,
2992 * preallocated extents, and those write extend the file, no need to
2993 * fall back to buffered IO.
2995 * For holes, we fallocate those blocks, mark them as uninitialized
2996 * If those blocks were preallocated, we mark sure they are splited, but
2997 * still keep the range to write as uninitialized.
2999 * The unwrritten extents will be converted to written when DIO is completed.
3000 * For async direct IO, since the IO may still pending when return, we
3001 * set up an end_io call back function, which will do the conversion
3002 * when async direct IO completed.
3004 * If the O_DIRECT write will extend the file then add this inode to the
3005 * orphan list. So recovery will truncate it back to the original size
3006 * if the machine crashes during the write.
3009 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3010 const struct iovec *iov, loff_t offset,
3011 unsigned long nr_segs)
3013 struct file *file = iocb->ki_filp;
3014 struct inode *inode = file->f_mapping->host;
3016 size_t count = iov_length(iov, nr_segs);
3018 loff_t final_size = offset + count;
3019 if (rw == WRITE && final_size <= inode->i_size) {
3021 * We could direct write to holes and fallocate.
3023 * Allocated blocks to fill the hole are marked as uninitialized
3024 * to prevent parallel buffered read to expose the stale data
3025 * before DIO complete the data IO.
3027 * As to previously fallocated extents, ext4 get_block
3028 * will just simply mark the buffer mapped but still
3029 * keep the extents uninitialized.
3031 * for non AIO case, we will convert those unwritten extents
3032 * to written after return back from blockdev_direct_IO.
3034 * for async DIO, the conversion needs to be defered when
3035 * the IO is completed. The ext4 end_io callback function
3036 * will be called to take care of the conversion work.
3037 * Here for async case, we allocate an io_end structure to
3040 iocb->private = NULL;
3041 EXT4_I(inode)->cur_aio_dio = NULL;
3042 if (!is_sync_kiocb(iocb)) {
3043 ext4_io_end_t *io_end =
3044 ext4_init_io_end(inode, GFP_NOFS);
3047 io_end->flag |= EXT4_IO_END_DIRECT;
3048 iocb->private = io_end;
3050 * we save the io structure for current async
3051 * direct IO, so that later ext4_map_blocks()
3052 * could flag the io structure whether there
3053 * is a unwritten extents needs to be converted
3054 * when IO is completed.
3056 EXT4_I(inode)->cur_aio_dio = iocb->private;
3059 ret = __blockdev_direct_IO(rw, iocb, inode,
3060 inode->i_sb->s_bdev, iov,
3062 ext4_get_block_write,
3065 DIO_LOCKING | DIO_SKIP_HOLES);
3067 EXT4_I(inode)->cur_aio_dio = NULL;
3069 * The io_end structure takes a reference to the inode,
3070 * that structure needs to be destroyed and the
3071 * reference to the inode need to be dropped, when IO is
3072 * complete, even with 0 byte write, or failed.
3074 * In the successful AIO DIO case, the io_end structure will be
3075 * desctroyed and the reference to the inode will be dropped
3076 * after the end_io call back function is called.
3078 * In the case there is 0 byte write, or error case, since
3079 * VFS direct IO won't invoke the end_io call back function,
3080 * we need to free the end_io structure here.
3082 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3083 ext4_free_io_end(iocb->private);
3084 iocb->private = NULL;
3085 } else if (ret > 0 && ext4_test_inode_state(inode,
3086 EXT4_STATE_DIO_UNWRITTEN)) {
3089 * for non AIO case, since the IO is already
3090 * completed, we could do the conversion right here
3092 err = ext4_convert_unwritten_extents(inode,
3096 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3101 /* for write the the end of file case, we fall back to old way */
3102 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3105 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3106 const struct iovec *iov, loff_t offset,
3107 unsigned long nr_segs)
3109 struct file *file = iocb->ki_filp;
3110 struct inode *inode = file->f_mapping->host;
3114 * If we are doing data journalling we don't support O_DIRECT
3116 if (ext4_should_journal_data(inode))
3119 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3120 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3121 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3123 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3124 trace_ext4_direct_IO_exit(inode, offset,
3125 iov_length(iov, nr_segs), rw, ret);
3130 * Pages can be marked dirty completely asynchronously from ext4's journalling
3131 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3132 * much here because ->set_page_dirty is called under VFS locks. The page is
3133 * not necessarily locked.
3135 * We cannot just dirty the page and leave attached buffers clean, because the
3136 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3137 * or jbddirty because all the journalling code will explode.
3139 * So what we do is to mark the page "pending dirty" and next time writepage
3140 * is called, propagate that into the buffers appropriately.
3142 static int ext4_journalled_set_page_dirty(struct page *page)
3144 SetPageChecked(page);
3145 return __set_page_dirty_nobuffers(page);
3148 static const struct address_space_operations ext4_ordered_aops = {
3149 .readpage = ext4_readpage,
3150 .readpages = ext4_readpages,
3151 .writepage = ext4_writepage,
3152 .write_begin = ext4_write_begin,
3153 .write_end = ext4_ordered_write_end,
3155 .invalidatepage = ext4_invalidatepage,
3156 .releasepage = ext4_releasepage,
3157 .direct_IO = ext4_direct_IO,
3158 .migratepage = buffer_migrate_page,
3159 .is_partially_uptodate = block_is_partially_uptodate,
3160 .error_remove_page = generic_error_remove_page,
3163 static const struct address_space_operations ext4_writeback_aops = {
3164 .readpage = ext4_readpage,
3165 .readpages = ext4_readpages,
3166 .writepage = ext4_writepage,
3167 .write_begin = ext4_write_begin,
3168 .write_end = ext4_writeback_write_end,
3170 .invalidatepage = ext4_invalidatepage,
3171 .releasepage = ext4_releasepage,
3172 .direct_IO = ext4_direct_IO,
3173 .migratepage = buffer_migrate_page,
3174 .is_partially_uptodate = block_is_partially_uptodate,
3175 .error_remove_page = generic_error_remove_page,
3178 static const struct address_space_operations ext4_journalled_aops = {
3179 .readpage = ext4_readpage,
3180 .readpages = ext4_readpages,
3181 .writepage = ext4_writepage,
3182 .write_begin = ext4_write_begin,
3183 .write_end = ext4_journalled_write_end,
3184 .set_page_dirty = ext4_journalled_set_page_dirty,
3186 .invalidatepage = ext4_invalidatepage,
3187 .releasepage = ext4_releasepage,
3188 .direct_IO = ext4_direct_IO,
3189 .is_partially_uptodate = block_is_partially_uptodate,
3190 .error_remove_page = generic_error_remove_page,
3193 static const struct address_space_operations ext4_da_aops = {
3194 .readpage = ext4_readpage,
3195 .readpages = ext4_readpages,
3196 .writepage = ext4_writepage,
3197 .writepages = ext4_da_writepages,
3198 .write_begin = ext4_da_write_begin,
3199 .write_end = ext4_da_write_end,
3201 .invalidatepage = ext4_da_invalidatepage,
3202 .releasepage = ext4_releasepage,
3203 .direct_IO = ext4_direct_IO,
3204 .migratepage = buffer_migrate_page,
3205 .is_partially_uptodate = block_is_partially_uptodate,
3206 .error_remove_page = generic_error_remove_page,
3209 void ext4_set_aops(struct inode *inode)
3211 switch (ext4_inode_journal_mode(inode)) {
3212 case EXT4_INODE_ORDERED_DATA_MODE:
3213 if (test_opt(inode->i_sb, DELALLOC))
3214 inode->i_mapping->a_ops = &ext4_da_aops;
3216 inode->i_mapping->a_ops = &ext4_ordered_aops;
3218 case EXT4_INODE_WRITEBACK_DATA_MODE:
3219 if (test_opt(inode->i_sb, DELALLOC))
3220 inode->i_mapping->a_ops = &ext4_da_aops;
3222 inode->i_mapping->a_ops = &ext4_writeback_aops;
3224 case EXT4_INODE_JOURNAL_DATA_MODE:
3225 inode->i_mapping->a_ops = &ext4_journalled_aops;
3234 * ext4_discard_partial_page_buffers()
3235 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3236 * This function finds and locks the page containing the offset
3237 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3238 * Calling functions that already have the page locked should call
3239 * ext4_discard_partial_page_buffers_no_lock directly.
3241 int ext4_discard_partial_page_buffers(handle_t *handle,
3242 struct address_space *mapping, loff_t from,
3243 loff_t length, int flags)
3245 struct inode *inode = mapping->host;
3249 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3250 mapping_gfp_mask(mapping) & ~__GFP_FS);
3254 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3255 from, length, flags);
3258 page_cache_release(page);
3263 * ext4_discard_partial_page_buffers_no_lock()
3264 * Zeros a page range of length 'length' starting from offset 'from'.
3265 * Buffer heads that correspond to the block aligned regions of the
3266 * zeroed range will be unmapped. Unblock aligned regions
3267 * will have the corresponding buffer head mapped if needed so that
3268 * that region of the page can be updated with the partial zero out.
3270 * This function assumes that the page has already been locked. The
3271 * The range to be discarded must be contained with in the given page.
3272 * If the specified range exceeds the end of the page it will be shortened
3273 * to the end of the page that corresponds to 'from'. This function is
3274 * appropriate for updating a page and it buffer heads to be unmapped and
3275 * zeroed for blocks that have been either released, or are going to be
3278 * handle: The journal handle
3279 * inode: The files inode
3280 * page: A locked page that contains the offset "from"
3281 * from: The starting byte offset (from the begining of the file)
3282 * to begin discarding
3283 * len: The length of bytes to discard
3284 * flags: Optional flags that may be used:
3286 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3287 * Only zero the regions of the page whose buffer heads
3288 * have already been unmapped. This flag is appropriate
3289 * for updateing the contents of a page whose blocks may
3290 * have already been released, and we only want to zero
3291 * out the regions that correspond to those released blocks.
3293 * Returns zero on sucess or negative on failure.
3295 int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3296 struct inode *inode, struct page *page, loff_t from,
3297 loff_t length, int flags)
3299 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3300 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3301 unsigned int blocksize, max, pos;
3303 struct buffer_head *bh;
3306 blocksize = inode->i_sb->s_blocksize;
3307 max = PAGE_CACHE_SIZE - offset;
3309 if (index != page->index)
3313 * correct length if it does not fall between
3314 * 'from' and the end of the page
3316 if (length > max || length < 0)
3319 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3321 if (!page_has_buffers(page))
3322 create_empty_buffers(page, blocksize, 0);
3324 /* Find the buffer that contains "offset" */
3325 bh = page_buffers(page);
3327 while (offset >= pos) {
3328 bh = bh->b_this_page;
3334 while (pos < offset + length) {
3335 unsigned int end_of_block, range_to_discard;
3339 /* The length of space left to zero and unmap */
3340 range_to_discard = offset + length - pos;
3342 /* The length of space until the end of the block */
3343 end_of_block = blocksize - (pos & (blocksize-1));
3346 * Do not unmap or zero past end of block
3347 * for this buffer head
3349 if (range_to_discard > end_of_block)
3350 range_to_discard = end_of_block;
3354 * Skip this buffer head if we are only zeroing unampped
3355 * regions of the page
3357 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3361 /* If the range is block aligned, unmap */
3362 if (range_to_discard == blocksize) {
3363 clear_buffer_dirty(bh);
3365 clear_buffer_mapped(bh);
3366 clear_buffer_req(bh);
3367 clear_buffer_new(bh);
3368 clear_buffer_delay(bh);
3369 clear_buffer_unwritten(bh);
3370 clear_buffer_uptodate(bh);
3371 zero_user(page, pos, range_to_discard);
3372 BUFFER_TRACE(bh, "Buffer discarded");
3377 * If this block is not completely contained in the range
3378 * to be discarded, then it is not going to be released. Because
3379 * we need to keep this block, we need to make sure this part
3380 * of the page is uptodate before we modify it by writeing
3381 * partial zeros on it.
3383 if (!buffer_mapped(bh)) {
3385 * Buffer head must be mapped before we can read
3388 BUFFER_TRACE(bh, "unmapped");
3389 ext4_get_block(inode, iblock, bh, 0);
3390 /* unmapped? It's a hole - nothing to do */
3391 if (!buffer_mapped(bh)) {
3392 BUFFER_TRACE(bh, "still unmapped");
3397 /* Ok, it's mapped. Make sure it's up-to-date */
3398 if (PageUptodate(page))
3399 set_buffer_uptodate(bh);
3401 if (!buffer_uptodate(bh)) {
3403 ll_rw_block(READ, 1, &bh);
3405 /* Uhhuh. Read error. Complain and punt.*/
3406 if (!buffer_uptodate(bh))
3410 if (ext4_should_journal_data(inode)) {
3411 BUFFER_TRACE(bh, "get write access");
3412 err = ext4_journal_get_write_access(handle, bh);
3417 zero_user(page, pos, range_to_discard);
3420 if (ext4_should_journal_data(inode)) {
3421 err = ext4_handle_dirty_metadata(handle, inode, bh);
3423 mark_buffer_dirty(bh);
3425 BUFFER_TRACE(bh, "Partial buffer zeroed");
3427 bh = bh->b_this_page;
3429 pos += range_to_discard;
3436 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3437 * up to the end of the block which corresponds to `from'.
3438 * This required during truncate. We need to physically zero the tail end
3439 * of that block so it doesn't yield old data if the file is later grown.
3441 int ext4_block_truncate_page(handle_t *handle,
3442 struct address_space *mapping, loff_t from)
3444 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3447 struct inode *inode = mapping->host;
3449 blocksize = inode->i_sb->s_blocksize;
3450 length = blocksize - (offset & (blocksize - 1));
3452 return ext4_block_zero_page_range(handle, mapping, from, length);
3456 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3457 * starting from file offset 'from'. The range to be zero'd must
3458 * be contained with in one block. If the specified range exceeds
3459 * the end of the block it will be shortened to end of the block
3460 * that cooresponds to 'from'
3462 int ext4_block_zero_page_range(handle_t *handle,
3463 struct address_space *mapping, loff_t from, loff_t length)
3465 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3466 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3467 unsigned blocksize, max, pos;
3469 struct inode *inode = mapping->host;
3470 struct buffer_head *bh;
3474 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3475 mapping_gfp_mask(mapping) & ~__GFP_FS);
3479 blocksize = inode->i_sb->s_blocksize;
3480 max = blocksize - (offset & (blocksize - 1));
3483 * correct length if it does not fall between
3484 * 'from' and the end of the block
3486 if (length > max || length < 0)
3489 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3491 if (!page_has_buffers(page))
3492 create_empty_buffers(page, blocksize, 0);
3494 /* Find the buffer that contains "offset" */
3495 bh = page_buffers(page);
3497 while (offset >= pos) {
3498 bh = bh->b_this_page;
3504 if (buffer_freed(bh)) {
3505 BUFFER_TRACE(bh, "freed: skip");
3509 if (!buffer_mapped(bh)) {
3510 BUFFER_TRACE(bh, "unmapped");
3511 ext4_get_block(inode, iblock, bh, 0);
3512 /* unmapped? It's a hole - nothing to do */
3513 if (!buffer_mapped(bh)) {
3514 BUFFER_TRACE(bh, "still unmapped");
3519 /* Ok, it's mapped. Make sure it's up-to-date */
3520 if (PageUptodate(page))
3521 set_buffer_uptodate(bh);
3523 if (!buffer_uptodate(bh)) {
3525 ll_rw_block(READ, 1, &bh);
3527 /* Uhhuh. Read error. Complain and punt. */
3528 if (!buffer_uptodate(bh))
3532 if (ext4_should_journal_data(inode)) {
3533 BUFFER_TRACE(bh, "get write access");
3534 err = ext4_journal_get_write_access(handle, bh);
3539 zero_user(page, offset, length);
3541 BUFFER_TRACE(bh, "zeroed end of block");
3544 if (ext4_should_journal_data(inode)) {
3545 err = ext4_handle_dirty_metadata(handle, inode, bh);
3547 mark_buffer_dirty(bh);
3551 page_cache_release(page);
3555 int ext4_can_truncate(struct inode *inode)
3557 if (S_ISREG(inode->i_mode))
3559 if (S_ISDIR(inode->i_mode))
3561 if (S_ISLNK(inode->i_mode))
3562 return !ext4_inode_is_fast_symlink(inode);
3567 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3568 * associated with the given offset and length
3570 * @inode: File inode
3571 * @offset: The offset where the hole will begin
3572 * @len: The length of the hole
3574 * Returns: 0 on sucess or negative on failure
3577 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3579 struct inode *inode = file->f_path.dentry->d_inode;
3580 if (!S_ISREG(inode->i_mode))
3583 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3584 /* TODO: Add support for non extent hole punching */
3588 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3589 /* TODO: Add support for bigalloc file systems */
3593 return ext4_ext_punch_hole(file, offset, length);
3599 * We block out ext4_get_block() block instantiations across the entire
3600 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3601 * simultaneously on behalf of the same inode.
3603 * As we work through the truncate and commmit bits of it to the journal there
3604 * is one core, guiding principle: the file's tree must always be consistent on
3605 * disk. We must be able to restart the truncate after a crash.
3607 * The file's tree may be transiently inconsistent in memory (although it
3608 * probably isn't), but whenever we close off and commit a journal transaction,
3609 * the contents of (the filesystem + the journal) must be consistent and
3610 * restartable. It's pretty simple, really: bottom up, right to left (although
3611 * left-to-right works OK too).
3613 * Note that at recovery time, journal replay occurs *before* the restart of
3614 * truncate against the orphan inode list.
3616 * The committed inode has the new, desired i_size (which is the same as
3617 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3618 * that this inode's truncate did not complete and it will again call
3619 * ext4_truncate() to have another go. So there will be instantiated blocks
3620 * to the right of the truncation point in a crashed ext4 filesystem. But
3621 * that's fine - as long as they are linked from the inode, the post-crash
3622 * ext4_truncate() run will find them and release them.
3624 void ext4_truncate(struct inode *inode)
3626 trace_ext4_truncate_enter(inode);
3628 if (!ext4_can_truncate(inode))
3631 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3633 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3634 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3636 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3637 ext4_ext_truncate(inode);
3639 ext4_ind_truncate(inode);
3641 trace_ext4_truncate_exit(inode);
3645 * ext4_get_inode_loc returns with an extra refcount against the inode's
3646 * underlying buffer_head on success. If 'in_mem' is true, we have all
3647 * data in memory that is needed to recreate the on-disk version of this
3650 static int __ext4_get_inode_loc(struct inode *inode,
3651 struct ext4_iloc *iloc, int in_mem)
3653 struct ext4_group_desc *gdp;
3654 struct buffer_head *bh;
3655 struct super_block *sb = inode->i_sb;
3657 int inodes_per_block, inode_offset;
3660 if (!ext4_valid_inum(sb, inode->i_ino))
3663 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3664 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3669 * Figure out the offset within the block group inode table
3671 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3672 inode_offset = ((inode->i_ino - 1) %
3673 EXT4_INODES_PER_GROUP(sb));
3674 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3675 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3677 bh = sb_getblk(sb, block);
3680 if (!buffer_uptodate(bh)) {
3684 * If the buffer has the write error flag, we have failed
3685 * to write out another inode in the same block. In this
3686 * case, we don't have to read the block because we may
3687 * read the old inode data successfully.
3689 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3690 set_buffer_uptodate(bh);
3692 if (buffer_uptodate(bh)) {
3693 /* someone brought it uptodate while we waited */
3699 * If we have all information of the inode in memory and this
3700 * is the only valid inode in the block, we need not read the
3704 struct buffer_head *bitmap_bh;
3707 start = inode_offset & ~(inodes_per_block - 1);
3709 /* Is the inode bitmap in cache? */
3710 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3715 * If the inode bitmap isn't in cache then the
3716 * optimisation may end up performing two reads instead
3717 * of one, so skip it.
3719 if (!buffer_uptodate(bitmap_bh)) {
3723 for (i = start; i < start + inodes_per_block; i++) {
3724 if (i == inode_offset)
3726 if (ext4_test_bit(i, bitmap_bh->b_data))
3730 if (i == start + inodes_per_block) {
3731 /* all other inodes are free, so skip I/O */
3732 memset(bh->b_data, 0, bh->b_size);
3733 set_buffer_uptodate(bh);
3741 * If we need to do any I/O, try to pre-readahead extra
3742 * blocks from the inode table.
3744 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3745 ext4_fsblk_t b, end, table;
3748 table = ext4_inode_table(sb, gdp);
3749 /* s_inode_readahead_blks is always a power of 2 */
3750 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3753 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3754 num = EXT4_INODES_PER_GROUP(sb);
3755 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3756 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3757 num -= ext4_itable_unused_count(sb, gdp);
3758 table += num / inodes_per_block;
3762 sb_breadahead(sb, b++);
3766 * There are other valid inodes in the buffer, this inode
3767 * has in-inode xattrs, or we don't have this inode in memory.
3768 * Read the block from disk.
3770 trace_ext4_load_inode(inode);
3772 bh->b_end_io = end_buffer_read_sync;
3773 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3775 if (!buffer_uptodate(bh)) {
3776 EXT4_ERROR_INODE_BLOCK(inode, block,
3777 "unable to read itable block");
3787 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3789 /* We have all inode data except xattrs in memory here. */
3790 return __ext4_get_inode_loc(inode, iloc,
3791 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3794 void ext4_set_inode_flags(struct inode *inode)
3796 unsigned int flags = EXT4_I(inode)->i_flags;
3797 unsigned int new_fl = 0;
3799 if (flags & EXT4_SYNC_FL)
3801 if (flags & EXT4_APPEND_FL)
3803 if (flags & EXT4_IMMUTABLE_FL)
3804 new_fl |= S_IMMUTABLE;
3805 if (flags & EXT4_NOATIME_FL)
3806 new_fl |= S_NOATIME;
3807 if (flags & EXT4_DIRSYNC_FL)
3808 new_fl |= S_DIRSYNC;
3809 set_mask_bits(&inode->i_flags,
3810 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC, new_fl);
3813 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3814 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3816 unsigned int vfs_fl;
3817 unsigned long old_fl, new_fl;
3820 vfs_fl = ei->vfs_inode.i_flags;
3821 old_fl = ei->i_flags;
3822 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3823 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3825 if (vfs_fl & S_SYNC)
3826 new_fl |= EXT4_SYNC_FL;
3827 if (vfs_fl & S_APPEND)
3828 new_fl |= EXT4_APPEND_FL;
3829 if (vfs_fl & S_IMMUTABLE)
3830 new_fl |= EXT4_IMMUTABLE_FL;
3831 if (vfs_fl & S_NOATIME)
3832 new_fl |= EXT4_NOATIME_FL;
3833 if (vfs_fl & S_DIRSYNC)
3834 new_fl |= EXT4_DIRSYNC_FL;
3835 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3838 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3839 struct ext4_inode_info *ei)
3842 struct inode *inode = &(ei->vfs_inode);
3843 struct super_block *sb = inode->i_sb;
3845 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3846 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3847 /* we are using combined 48 bit field */
3848 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3849 le32_to_cpu(raw_inode->i_blocks_lo);
3850 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3851 /* i_blocks represent file system block size */
3852 return i_blocks << (inode->i_blkbits - 9);
3857 return le32_to_cpu(raw_inode->i_blocks_lo);
3861 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3863 struct ext4_iloc iloc;
3864 struct ext4_inode *raw_inode;
3865 struct ext4_inode_info *ei;
3866 struct inode *inode;
3867 journal_t *journal = EXT4_SB(sb)->s_journal;
3872 inode = iget_locked(sb, ino);
3874 return ERR_PTR(-ENOMEM);
3875 if (!(inode->i_state & I_NEW))
3881 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3884 raw_inode = ext4_raw_inode(&iloc);
3885 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3886 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3887 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3888 if (!(test_opt(inode->i_sb, NO_UID32))) {
3889 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3890 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3892 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3894 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3895 ei->i_dir_start_lookup = 0;
3896 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3897 /* We now have enough fields to check if the inode was active or not.
3898 * This is needed because nfsd might try to access dead inodes
3899 * the test is that same one that e2fsck uses
3900 * NeilBrown 1999oct15
3902 if (inode->i_nlink == 0) {
3903 if (inode->i_mode == 0 ||
3904 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3905 /* this inode is deleted */
3909 /* The only unlinked inodes we let through here have
3910 * valid i_mode and are being read by the orphan
3911 * recovery code: that's fine, we're about to complete
3912 * the process of deleting those. */
3914 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3915 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3916 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3917 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3919 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3920 inode->i_size = ext4_isize(raw_inode);
3921 if ((size = i_size_read(inode)) < 0) {
3922 EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size);
3926 ei->i_disksize = inode->i_size;
3928 ei->i_reserved_quota = 0;
3930 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3931 ei->i_block_group = iloc.block_group;
3932 ei->i_last_alloc_group = ~0;
3934 * NOTE! The in-memory inode i_data array is in little-endian order
3935 * even on big-endian machines: we do NOT byteswap the block numbers!
3937 for (block = 0; block < EXT4_N_BLOCKS; block++)
3938 ei->i_data[block] = raw_inode->i_block[block];
3939 INIT_LIST_HEAD(&ei->i_orphan);
3942 * Set transaction id's of transactions that have to be committed
3943 * to finish f[data]sync. We set them to currently running transaction
3944 * as we cannot be sure that the inode or some of its metadata isn't
3945 * part of the transaction - the inode could have been reclaimed and
3946 * now it is reread from disk.
3949 transaction_t *transaction;
3952 read_lock(&journal->j_state_lock);
3953 if (journal->j_running_transaction)
3954 transaction = journal->j_running_transaction;
3956 transaction = journal->j_committing_transaction;
3958 tid = transaction->t_tid;
3960 tid = journal->j_commit_sequence;
3961 read_unlock(&journal->j_state_lock);
3962 ei->i_sync_tid = tid;
3963 ei->i_datasync_tid = tid;
3966 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3967 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3968 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3969 EXT4_INODE_SIZE(inode->i_sb)) {
3973 if (ei->i_extra_isize == 0) {
3974 /* The extra space is currently unused. Use it. */
3975 ei->i_extra_isize = sizeof(struct ext4_inode) -
3976 EXT4_GOOD_OLD_INODE_SIZE;
3978 __le32 *magic = (void *)raw_inode +
3979 EXT4_GOOD_OLD_INODE_SIZE +
3981 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3982 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3985 ei->i_extra_isize = 0;
3987 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3988 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3989 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3990 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3992 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3993 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3994 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3996 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4000 if (ei->i_file_acl &&
4001 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4002 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4006 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4007 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4008 (S_ISLNK(inode->i_mode) &&
4009 !ext4_inode_is_fast_symlink(inode)))
4010 /* Validate extent which is part of inode */
4011 ret = ext4_ext_check_inode(inode);
4012 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4013 (S_ISLNK(inode->i_mode) &&
4014 !ext4_inode_is_fast_symlink(inode))) {
4015 /* Validate block references which are part of inode */
4016 ret = ext4_ind_check_inode(inode);
4021 if (S_ISREG(inode->i_mode)) {
4022 inode->i_op = &ext4_file_inode_operations;
4023 inode->i_fop = &ext4_file_operations;
4024 ext4_set_aops(inode);
4025 } else if (S_ISDIR(inode->i_mode)) {
4026 inode->i_op = &ext4_dir_inode_operations;
4027 inode->i_fop = &ext4_dir_operations;
4028 } else if (S_ISLNK(inode->i_mode)) {
4029 if (ext4_inode_is_fast_symlink(inode)) {
4030 inode->i_op = &ext4_fast_symlink_inode_operations;
4031 nd_terminate_link(ei->i_data, inode->i_size,
4032 sizeof(ei->i_data) - 1);
4034 inode->i_op = &ext4_symlink_inode_operations;
4035 ext4_set_aops(inode);
4037 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4038 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4039 inode->i_op = &ext4_special_inode_operations;
4040 if (raw_inode->i_block[0])
4041 init_special_inode(inode, inode->i_mode,
4042 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4044 init_special_inode(inode, inode->i_mode,
4045 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4048 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4052 ext4_set_inode_flags(inode);
4053 unlock_new_inode(inode);
4059 return ERR_PTR(ret);
4062 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4064 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4065 return ERR_PTR(-EIO);
4066 return ext4_iget(sb, ino);
4069 static int ext4_inode_blocks_set(handle_t *handle,
4070 struct ext4_inode *raw_inode,
4071 struct ext4_inode_info *ei)
4073 struct inode *inode = &(ei->vfs_inode);
4074 u64 i_blocks = inode->i_blocks;
4075 struct super_block *sb = inode->i_sb;
4077 if (i_blocks <= ~0U) {
4079 * i_blocks can be represnted in a 32 bit variable
4080 * as multiple of 512 bytes
4082 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4083 raw_inode->i_blocks_high = 0;
4084 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4087 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4090 if (i_blocks <= 0xffffffffffffULL) {
4092 * i_blocks can be represented in a 48 bit variable
4093 * as multiple of 512 bytes
4095 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4096 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4097 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4099 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4100 /* i_block is stored in file system block size */
4101 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4102 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4103 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4109 * Post the struct inode info into an on-disk inode location in the
4110 * buffer-cache. This gobbles the caller's reference to the
4111 * buffer_head in the inode location struct.
4113 * The caller must have write access to iloc->bh.
4115 static int ext4_do_update_inode(handle_t *handle,
4116 struct inode *inode,
4117 struct ext4_iloc *iloc)
4119 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4120 struct ext4_inode_info *ei = EXT4_I(inode);
4121 struct buffer_head *bh = iloc->bh;
4122 int err = 0, rc, block;
4123 int need_datasync = 0;
4125 /* For fields not not tracking in the in-memory inode,
4126 * initialise them to zero for new inodes. */
4127 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4128 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4130 ext4_get_inode_flags(ei);
4131 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4132 if (!(test_opt(inode->i_sb, NO_UID32))) {
4133 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4134 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4136 * Fix up interoperability with old kernels. Otherwise, old inodes get
4137 * re-used with the upper 16 bits of the uid/gid intact
4139 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4140 raw_inode->i_uid_high = 0;
4141 raw_inode->i_gid_high = 0;
4143 raw_inode->i_uid_high =
4144 cpu_to_le16(high_16_bits(inode->i_uid));
4145 raw_inode->i_gid_high =
4146 cpu_to_le16(high_16_bits(inode->i_gid));
4149 raw_inode->i_uid_low =
4150 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4151 raw_inode->i_gid_low =
4152 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4153 raw_inode->i_uid_high = 0;
4154 raw_inode->i_gid_high = 0;
4156 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4158 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4159 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4160 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4161 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4163 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4165 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4166 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4167 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4168 cpu_to_le32(EXT4_OS_HURD))
4169 raw_inode->i_file_acl_high =
4170 cpu_to_le16(ei->i_file_acl >> 32);
4171 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4172 if (ei->i_disksize != ext4_isize(raw_inode)) {
4173 ext4_isize_set(raw_inode, ei->i_disksize);
4176 if (ei->i_disksize > 0x7fffffffULL) {
4177 struct super_block *sb = inode->i_sb;
4178 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4179 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4180 EXT4_SB(sb)->s_es->s_rev_level ==
4181 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4182 /* If this is the first large file
4183 * created, add a flag to the superblock.
4185 err = ext4_journal_get_write_access(handle,
4186 EXT4_SB(sb)->s_sbh);
4189 ext4_update_dynamic_rev(sb);
4190 EXT4_SET_RO_COMPAT_FEATURE(sb,
4191 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4193 ext4_handle_sync(handle);
4194 err = ext4_handle_dirty_metadata(handle, NULL,
4195 EXT4_SB(sb)->s_sbh);
4198 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4199 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4200 if (old_valid_dev(inode->i_rdev)) {
4201 raw_inode->i_block[0] =
4202 cpu_to_le32(old_encode_dev(inode->i_rdev));
4203 raw_inode->i_block[1] = 0;
4205 raw_inode->i_block[0] = 0;
4206 raw_inode->i_block[1] =
4207 cpu_to_le32(new_encode_dev(inode->i_rdev));
4208 raw_inode->i_block[2] = 0;
4211 for (block = 0; block < EXT4_N_BLOCKS; block++)
4212 raw_inode->i_block[block] = ei->i_data[block];
4214 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4215 if (ei->i_extra_isize) {
4216 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4217 raw_inode->i_version_hi =
4218 cpu_to_le32(inode->i_version >> 32);
4219 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4222 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4223 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4226 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4228 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4231 ext4_std_error(inode->i_sb, err);
4236 * ext4_write_inode()
4238 * We are called from a few places:
4240 * - Within generic_file_write() for O_SYNC files.
4241 * Here, there will be no transaction running. We wait for any running
4242 * trasnaction to commit.
4244 * - Within sys_sync(), kupdate and such.
4245 * We wait on commit, if tol to.
4247 * - Within prune_icache() (PF_MEMALLOC == true)
4248 * Here we simply return. We can't afford to block kswapd on the
4251 * In all cases it is actually safe for us to return without doing anything,
4252 * because the inode has been copied into a raw inode buffer in
4253 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4256 * Note that we are absolutely dependent upon all inode dirtiers doing the
4257 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4258 * which we are interested.
4260 * It would be a bug for them to not do this. The code:
4262 * mark_inode_dirty(inode)
4264 * inode->i_size = expr;
4266 * is in error because a kswapd-driven write_inode() could occur while
4267 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4268 * will no longer be on the superblock's dirty inode list.
4270 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4274 if (current->flags & PF_MEMALLOC)
4277 if (EXT4_SB(inode->i_sb)->s_journal) {
4278 if (ext4_journal_current_handle()) {
4279 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4284 if (wbc->sync_mode != WB_SYNC_ALL)
4287 err = ext4_force_commit(inode->i_sb);
4289 struct ext4_iloc iloc;
4291 err = __ext4_get_inode_loc(inode, &iloc, 0);
4294 if (wbc->sync_mode == WB_SYNC_ALL)
4295 sync_dirty_buffer(iloc.bh);
4296 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4297 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4298 "IO error syncing inode");
4309 * Called from notify_change.
4311 * We want to trap VFS attempts to truncate the file as soon as
4312 * possible. In particular, we want to make sure that when the VFS
4313 * shrinks i_size, we put the inode on the orphan list and modify
4314 * i_disksize immediately, so that during the subsequent flushing of
4315 * dirty pages and freeing of disk blocks, we can guarantee that any
4316 * commit will leave the blocks being flushed in an unused state on
4317 * disk. (On recovery, the inode will get truncated and the blocks will
4318 * be freed, so we have a strong guarantee that no future commit will
4319 * leave these blocks visible to the user.)
4321 * Another thing we have to assure is that if we are in ordered mode
4322 * and inode is still attached to the committing transaction, we must
4323 * we start writeout of all the dirty pages which are being truncated.
4324 * This way we are sure that all the data written in the previous
4325 * transaction are already on disk (truncate waits for pages under
4328 * Called with inode->i_mutex down.
4330 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4332 struct inode *inode = dentry->d_inode;
4335 const unsigned int ia_valid = attr->ia_valid;
4337 error = setattr_prepare(dentry, attr);
4341 if (is_quota_modification(inode, attr))
4342 dquot_initialize(inode);
4343 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4344 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4347 /* (user+group)*(old+new) structure, inode write (sb,
4348 * inode block, ? - but truncate inode update has it) */
4349 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
4350 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
4351 if (IS_ERR(handle)) {
4352 error = PTR_ERR(handle);
4355 error = dquot_transfer(inode, attr);
4357 ext4_journal_stop(handle);
4360 /* Update corresponding info in inode so that everything is in
4361 * one transaction */
4362 if (attr->ia_valid & ATTR_UID)
4363 inode->i_uid = attr->ia_uid;
4364 if (attr->ia_valid & ATTR_GID)
4365 inode->i_gid = attr->ia_gid;
4366 error = ext4_mark_inode_dirty(handle, inode);
4367 ext4_journal_stop(handle);
4370 if (attr->ia_valid & ATTR_SIZE) {
4371 inode_dio_wait(inode);
4373 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4374 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4376 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4381 if (S_ISREG(inode->i_mode) &&
4382 attr->ia_valid & ATTR_SIZE &&
4383 (attr->ia_size < inode->i_size)) {
4386 handle = ext4_journal_start(inode, 3);
4387 if (IS_ERR(handle)) {
4388 error = PTR_ERR(handle);
4391 if (ext4_handle_valid(handle)) {
4392 error = ext4_orphan_add(handle, inode);
4395 EXT4_I(inode)->i_disksize = attr->ia_size;
4396 rc = ext4_mark_inode_dirty(handle, inode);
4399 ext4_journal_stop(handle);
4401 if (ext4_should_order_data(inode)) {
4402 error = ext4_begin_ordered_truncate(inode,
4405 /* Do as much error cleanup as possible */
4406 handle = ext4_journal_start(inode, 3);
4407 if (IS_ERR(handle)) {
4408 ext4_orphan_del(NULL, inode);
4411 ext4_orphan_del(handle, inode);
4413 ext4_journal_stop(handle);
4419 if (attr->ia_valid & ATTR_SIZE) {
4420 if (attr->ia_size != i_size_read(inode)) {
4421 truncate_setsize(inode, attr->ia_size);
4422 ext4_truncate(inode);
4423 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
4424 ext4_truncate(inode);
4428 setattr_copy(inode, attr);
4429 mark_inode_dirty(inode);
4433 * If the call to ext4_truncate failed to get a transaction handle at
4434 * all, we need to clean up the in-core orphan list manually.
4436 if (orphan && inode->i_nlink)
4437 ext4_orphan_del(NULL, inode);
4439 if (!rc && (ia_valid & ATTR_MODE))
4440 rc = ext4_acl_chmod(inode);
4443 ext4_std_error(inode->i_sb, error);
4449 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4452 struct inode *inode;
4453 unsigned long long delalloc_blocks;
4455 inode = dentry->d_inode;
4456 generic_fillattr(inode, stat);
4459 * We can't update i_blocks if the block allocation is delayed
4460 * otherwise in the case of system crash before the real block
4461 * allocation is done, we will have i_blocks inconsistent with
4462 * on-disk file blocks.
4463 * We always keep i_blocks updated together with real
4464 * allocation. But to not confuse with user, stat
4465 * will return the blocks that include the delayed allocation
4466 * blocks for this file.
4468 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4470 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits-9);
4474 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4476 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4477 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4478 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4482 * Account for index blocks, block groups bitmaps and block group
4483 * descriptor blocks if modify datablocks and index blocks
4484 * worse case, the indexs blocks spread over different block groups
4486 * If datablocks are discontiguous, they are possible to spread over
4487 * different block groups too. If they are contiuguous, with flexbg,
4488 * they could still across block group boundary.
4490 * Also account for superblock, inode, quota and xattr blocks
4492 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4494 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4500 * How many index blocks need to touch to modify nrblocks?
4501 * The "Chunk" flag indicating whether the nrblocks is
4502 * physically contiguous on disk
4504 * For Direct IO and fallocate, they calls get_block to allocate
4505 * one single extent at a time, so they could set the "Chunk" flag
4507 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4512 * Now let's see how many group bitmaps and group descriptors need
4522 if (groups > ngroups)
4524 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4525 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4527 /* bitmaps and block group descriptor blocks */
4528 ret += groups + gdpblocks;
4530 /* Blocks for super block, inode, quota and xattr blocks */
4531 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4537 * Calculate the total number of credits to reserve to fit
4538 * the modification of a single pages into a single transaction,
4539 * which may include multiple chunks of block allocations.
4541 * This could be called via ext4_write_begin()
4543 * We need to consider the worse case, when
4544 * one new block per extent.
4546 int ext4_writepage_trans_blocks(struct inode *inode)
4548 int bpp = ext4_journal_blocks_per_page(inode);
4551 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4553 /* Account for data blocks for journalled mode */
4554 if (ext4_should_journal_data(inode))
4560 * Calculate the journal credits for a chunk of data modification.
4562 * This is called from DIO, fallocate or whoever calling
4563 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4565 * journal buffers for data blocks are not included here, as DIO
4566 * and fallocate do no need to journal data buffers.
4568 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4570 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4574 * The caller must have previously called ext4_reserve_inode_write().
4575 * Give this, we know that the caller already has write access to iloc->bh.
4577 int ext4_mark_iloc_dirty(handle_t *handle,
4578 struct inode *inode, struct ext4_iloc *iloc)
4582 if (test_opt(inode->i_sb, I_VERSION))
4583 inode_inc_iversion(inode);
4585 /* the do_update_inode consumes one bh->b_count */
4588 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4589 err = ext4_do_update_inode(handle, inode, iloc);
4595 * On success, We end up with an outstanding reference count against
4596 * iloc->bh. This _must_ be cleaned up later.
4600 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4601 struct ext4_iloc *iloc)
4605 err = ext4_get_inode_loc(inode, iloc);
4607 BUFFER_TRACE(iloc->bh, "get_write_access");
4608 err = ext4_journal_get_write_access(handle, iloc->bh);
4614 ext4_std_error(inode->i_sb, err);
4619 * Expand an inode by new_extra_isize bytes.
4620 * Returns 0 on success or negative error number on failure.
4622 static int ext4_expand_extra_isize(struct inode *inode,
4623 unsigned int new_extra_isize,
4624 struct ext4_iloc iloc,
4627 struct ext4_inode *raw_inode;
4628 struct ext4_xattr_ibody_header *header;
4630 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4633 raw_inode = ext4_raw_inode(&iloc);
4635 header = IHDR(inode, raw_inode);
4637 /* No extended attributes present */
4638 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4639 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4640 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4642 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4646 /* try to expand with EAs present */
4647 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4652 * What we do here is to mark the in-core inode as clean with respect to inode
4653 * dirtiness (it may still be data-dirty).
4654 * This means that the in-core inode may be reaped by prune_icache
4655 * without having to perform any I/O. This is a very good thing,
4656 * because *any* task may call prune_icache - even ones which
4657 * have a transaction open against a different journal.
4659 * Is this cheating? Not really. Sure, we haven't written the
4660 * inode out, but prune_icache isn't a user-visible syncing function.
4661 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4662 * we start and wait on commits.
4664 * Is this efficient/effective? Well, we're being nice to the system
4665 * by cleaning up our inodes proactively so they can be reaped
4666 * without I/O. But we are potentially leaving up to five seconds'
4667 * worth of inodes floating about which prune_icache wants us to
4668 * write out. One way to fix that would be to get prune_icache()
4669 * to do a write_super() to free up some memory. It has the desired
4672 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4674 struct ext4_iloc iloc;
4675 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4676 static unsigned int mnt_count;
4680 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4681 err = ext4_reserve_inode_write(handle, inode, &iloc);
4684 if (ext4_handle_valid(handle) &&
4685 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4686 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4688 * We need extra buffer credits since we may write into EA block
4689 * with this same handle. If journal_extend fails, then it will
4690 * only result in a minor loss of functionality for that inode.
4691 * If this is felt to be critical, then e2fsck should be run to
4692 * force a large enough s_min_extra_isize.
4694 if ((jbd2_journal_extend(handle,
4695 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4696 ret = ext4_expand_extra_isize(inode,
4697 sbi->s_want_extra_isize,
4700 ext4_set_inode_state(inode,
4701 EXT4_STATE_NO_EXPAND);
4703 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4704 ext4_warning(inode->i_sb,
4705 "Unable to expand inode %lu. Delete"
4706 " some EAs or run e2fsck.",
4709 le16_to_cpu(sbi->s_es->s_mnt_count);
4714 return ext4_mark_iloc_dirty(handle, inode, &iloc);
4718 * ext4_dirty_inode() is called from __mark_inode_dirty()
4720 * We're really interested in the case where a file is being extended.
4721 * i_size has been changed by generic_commit_write() and we thus need
4722 * to include the updated inode in the current transaction.
4724 * Also, dquot_alloc_block() will always dirty the inode when blocks
4725 * are allocated to the file.
4727 * If the inode is marked synchronous, we don't honour that here - doing
4728 * so would cause a commit on atime updates, which we don't bother doing.
4729 * We handle synchronous inodes at the highest possible level.
4731 void ext4_dirty_inode(struct inode *inode, int flags)
4735 handle = ext4_journal_start(inode, 2);
4739 ext4_mark_inode_dirty(handle, inode);
4741 ext4_journal_stop(handle);
4748 * Bind an inode's backing buffer_head into this transaction, to prevent
4749 * it from being flushed to disk early. Unlike
4750 * ext4_reserve_inode_write, this leaves behind no bh reference and
4751 * returns no iloc structure, so the caller needs to repeat the iloc
4752 * lookup to mark the inode dirty later.
4754 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4756 struct ext4_iloc iloc;
4760 err = ext4_get_inode_loc(inode, &iloc);
4762 BUFFER_TRACE(iloc.bh, "get_write_access");
4763 err = jbd2_journal_get_write_access(handle, iloc.bh);
4765 err = ext4_handle_dirty_metadata(handle,
4771 ext4_std_error(inode->i_sb, err);
4776 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4783 * We have to be very careful here: changing a data block's
4784 * journaling status dynamically is dangerous. If we write a
4785 * data block to the journal, change the status and then delete
4786 * that block, we risk forgetting to revoke the old log record
4787 * from the journal and so a subsequent replay can corrupt data.
4788 * So, first we make sure that the journal is empty and that
4789 * nobody is changing anything.
4792 journal = EXT4_JOURNAL(inode);
4795 if (is_journal_aborted(journal))
4798 jbd2_journal_lock_updates(journal);
4799 jbd2_journal_flush(journal);
4802 * OK, there are no updates running now, and all cached data is
4803 * synced to disk. We are now in a completely consistent state
4804 * which doesn't have anything in the journal, and we know that
4805 * no filesystem updates are running, so it is safe to modify
4806 * the inode's in-core data-journaling state flag now.
4810 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4812 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4813 ext4_set_aops(inode);
4815 jbd2_journal_unlock_updates(journal);
4817 /* Finally we can mark the inode as dirty. */
4819 handle = ext4_journal_start(inode, 1);
4821 return PTR_ERR(handle);
4823 err = ext4_mark_inode_dirty(handle, inode);
4824 ext4_handle_sync(handle);
4825 ext4_journal_stop(handle);
4826 ext4_std_error(inode->i_sb, err);
4831 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4833 return !buffer_mapped(bh);
4836 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4838 struct page *page = vmf->page;
4842 struct file *file = vma->vm_file;
4843 struct inode *inode = file->f_path.dentry->d_inode;
4844 struct address_space *mapping = inode->i_mapping;
4846 get_block_t *get_block;
4850 * This check is racy but catches the common case. We rely on
4851 * __block_page_mkwrite() to do a reliable check.
4853 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
4854 /* Delalloc case is easy... */
4855 if (test_opt(inode->i_sb, DELALLOC) &&
4856 !ext4_should_journal_data(inode) &&
4857 !ext4_nonda_switch(inode->i_sb)) {
4859 ret = __block_page_mkwrite(vma, vmf,
4860 ext4_da_get_block_prep);
4861 } while (ret == -ENOSPC &&
4862 ext4_should_retry_alloc(inode->i_sb, &retries));
4867 size = i_size_read(inode);
4868 /* Page got truncated from under us? */
4869 if (page->mapping != mapping || page_offset(page) > size) {
4871 ret = VM_FAULT_NOPAGE;
4875 if (page->index == size >> PAGE_CACHE_SHIFT)
4876 len = size & ~PAGE_CACHE_MASK;
4878 len = PAGE_CACHE_SIZE;
4880 * Return if we have all the buffers mapped. This avoids the need to do
4881 * journal_start/journal_stop which can block and take a long time
4883 if (page_has_buffers(page)) {
4884 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4885 ext4_bh_unmapped)) {
4886 /* Wait so that we don't change page under IO */
4887 wait_on_page_writeback(page);
4888 ret = VM_FAULT_LOCKED;
4893 /* OK, we need to fill the hole... */
4894 if (ext4_should_dioread_nolock(inode))
4895 get_block = ext4_get_block_write;
4897 get_block = ext4_get_block;
4899 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4900 if (IS_ERR(handle)) {
4901 ret = VM_FAULT_SIGBUS;
4904 ret = __block_page_mkwrite(vma, vmf, get_block);
4905 if (!ret && ext4_should_journal_data(inode)) {
4906 if (walk_page_buffers(handle, page_buffers(page), 0,
4907 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4909 ret = VM_FAULT_SIGBUS;
4910 ext4_journal_stop(handle);
4913 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4915 ext4_journal_stop(handle);
4916 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4919 ret = block_page_mkwrite_return(ret);