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 (ext4_should_journal_data(inode) &&
146 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
147 inode->i_ino != EXT4_JOURNAL_INO) {
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);
598 /* Maximum number of blocks we map for direct IO at once. */
599 #define DIO_MAX_BLOCKS 4096
601 static int _ext4_get_block(struct inode *inode, sector_t iblock,
602 struct buffer_head *bh, int flags)
604 handle_t *handle = ext4_journal_current_handle();
605 struct ext4_map_blocks map;
606 int ret = 0, started = 0;
610 map.m_len = bh->b_size >> inode->i_blkbits;
612 if (flags && !handle) {
613 /* Direct IO write... */
614 if (map.m_len > DIO_MAX_BLOCKS)
615 map.m_len = DIO_MAX_BLOCKS;
616 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
617 handle = ext4_journal_start(inode, dio_credits);
618 if (IS_ERR(handle)) {
619 ret = PTR_ERR(handle);
625 ret = ext4_map_blocks(handle, inode, &map, flags);
627 map_bh(bh, inode->i_sb, map.m_pblk);
628 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
629 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
633 ext4_journal_stop(handle);
637 int ext4_get_block(struct inode *inode, sector_t iblock,
638 struct buffer_head *bh, int create)
640 return _ext4_get_block(inode, iblock, bh,
641 create ? EXT4_GET_BLOCKS_CREATE : 0);
645 * `handle' can be NULL if create is zero
647 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
648 ext4_lblk_t block, int create, int *errp)
650 struct ext4_map_blocks map;
651 struct buffer_head *bh;
654 J_ASSERT(handle != NULL || create == 0);
658 err = ext4_map_blocks(handle, inode, &map,
659 create ? EXT4_GET_BLOCKS_CREATE : 0);
667 bh = sb_getblk(inode->i_sb, map.m_pblk);
672 if (map.m_flags & EXT4_MAP_NEW) {
673 J_ASSERT(create != 0);
674 J_ASSERT(handle != NULL);
677 * Now that we do not always journal data, we should
678 * keep in mind whether this should always journal the
679 * new buffer as metadata. For now, regular file
680 * writes use ext4_get_block instead, so it's not a
684 BUFFER_TRACE(bh, "call get_create_access");
685 fatal = ext4_journal_get_create_access(handle, bh);
686 if (!fatal && !buffer_uptodate(bh)) {
687 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
688 set_buffer_uptodate(bh);
691 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
692 err = ext4_handle_dirty_metadata(handle, inode, bh);
696 BUFFER_TRACE(bh, "not a new buffer");
706 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
707 ext4_lblk_t block, int create, int *err)
709 struct buffer_head *bh;
711 bh = ext4_getblk(handle, inode, block, create, err);
714 if (buffer_uptodate(bh))
716 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
718 if (buffer_uptodate(bh))
725 static int walk_page_buffers(handle_t *handle,
726 struct buffer_head *head,
730 int (*fn)(handle_t *handle,
731 struct buffer_head *bh))
733 struct buffer_head *bh;
734 unsigned block_start, block_end;
735 unsigned blocksize = head->b_size;
737 struct buffer_head *next;
739 for (bh = head, block_start = 0;
740 ret == 0 && (bh != head || !block_start);
741 block_start = block_end, bh = next) {
742 next = bh->b_this_page;
743 block_end = block_start + blocksize;
744 if (block_end <= from || block_start >= to) {
745 if (partial && !buffer_uptodate(bh))
749 err = (*fn)(handle, bh);
757 * To preserve ordering, it is essential that the hole instantiation and
758 * the data write be encapsulated in a single transaction. We cannot
759 * close off a transaction and start a new one between the ext4_get_block()
760 * and the commit_write(). So doing the jbd2_journal_start at the start of
761 * prepare_write() is the right place.
763 * Also, this function can nest inside ext4_writepage() ->
764 * block_write_full_page(). In that case, we *know* that ext4_writepage()
765 * has generated enough buffer credits to do the whole page. So we won't
766 * block on the journal in that case, which is good, because the caller may
769 * By accident, ext4 can be reentered when a transaction is open via
770 * quota file writes. If we were to commit the transaction while thus
771 * reentered, there can be a deadlock - we would be holding a quota
772 * lock, and the commit would never complete if another thread had a
773 * transaction open and was blocking on the quota lock - a ranking
776 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
777 * will _not_ run commit under these circumstances because handle->h_ref
778 * is elevated. We'll still have enough credits for the tiny quotafile
781 static int do_journal_get_write_access(handle_t *handle,
782 struct buffer_head *bh)
784 int dirty = buffer_dirty(bh);
787 if (!buffer_mapped(bh) || buffer_freed(bh))
790 * __block_write_begin() could have dirtied some buffers. Clean
791 * the dirty bit as jbd2_journal_get_write_access() could complain
792 * otherwise about fs integrity issues. Setting of the dirty bit
793 * by __block_write_begin() isn't a real problem here as we clear
794 * the bit before releasing a page lock and thus writeback cannot
795 * ever write the buffer.
798 clear_buffer_dirty(bh);
799 ret = ext4_journal_get_write_access(handle, bh);
801 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
805 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
806 struct buffer_head *bh_result, int create);
807 static int ext4_write_begin(struct file *file, struct address_space *mapping,
808 loff_t pos, unsigned len, unsigned flags,
809 struct page **pagep, void **fsdata)
811 struct inode *inode = mapping->host;
812 int ret, needed_blocks;
819 trace_ext4_write_begin(inode, pos, len, flags);
821 * Reserve one block more for addition to orphan list in case
822 * we allocate blocks but write fails for some reason
824 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
825 index = pos >> PAGE_CACHE_SHIFT;
826 from = pos & (PAGE_CACHE_SIZE - 1);
830 handle = ext4_journal_start(inode, needed_blocks);
831 if (IS_ERR(handle)) {
832 ret = PTR_ERR(handle);
836 /* We cannot recurse into the filesystem as the transaction is already
838 flags |= AOP_FLAG_NOFS;
840 page = grab_cache_page_write_begin(mapping, index, flags);
842 ext4_journal_stop(handle);
848 if (ext4_should_dioread_nolock(inode))
849 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
851 ret = __block_write_begin(page, pos, len, ext4_get_block);
853 if (!ret && ext4_should_journal_data(inode)) {
854 ret = walk_page_buffers(handle, page_buffers(page),
855 from, to, NULL, do_journal_get_write_access);
860 page_cache_release(page);
862 * __block_write_begin may have instantiated a few blocks
863 * outside i_size. Trim these off again. Don't need
864 * i_size_read because we hold i_mutex.
866 * Add inode to orphan list in case we crash before
869 if (pos + len > inode->i_size && ext4_can_truncate(inode))
870 ext4_orphan_add(handle, inode);
872 ext4_journal_stop(handle);
873 if (pos + len > inode->i_size) {
874 ext4_truncate_failed_write(inode);
876 * If truncate failed early the inode might
877 * still be on the orphan list; we need to
878 * make sure the inode is removed from the
879 * orphan list in that case.
882 ext4_orphan_del(NULL, inode);
886 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
892 /* For write_end() in data=journal mode */
893 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
895 if (!buffer_mapped(bh) || buffer_freed(bh))
897 set_buffer_uptodate(bh);
898 return ext4_handle_dirty_metadata(handle, NULL, bh);
901 static int ext4_generic_write_end(struct file *file,
902 struct address_space *mapping,
903 loff_t pos, unsigned len, unsigned copied,
904 struct page *page, void *fsdata)
906 int i_size_changed = 0;
907 struct inode *inode = mapping->host;
908 handle_t *handle = ext4_journal_current_handle();
910 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
913 * No need to use i_size_read() here, the i_size
914 * cannot change under us because we hold i_mutex.
916 * But it's important to update i_size while still holding page lock:
917 * page writeout could otherwise come in and zero beyond i_size.
919 if (pos + copied > inode->i_size) {
920 i_size_write(inode, pos + copied);
924 if (pos + copied > EXT4_I(inode)->i_disksize) {
925 /* We need to mark inode dirty even if
926 * new_i_size is less that inode->i_size
927 * bu greater than i_disksize.(hint delalloc)
929 ext4_update_i_disksize(inode, (pos + copied));
933 page_cache_release(page);
936 * Don't mark the inode dirty under page lock. First, it unnecessarily
937 * makes the holding time of page lock longer. Second, it forces lock
938 * ordering of page lock and transaction start for journaling
942 ext4_mark_inode_dirty(handle, inode);
948 * We need to pick up the new inode size which generic_commit_write gave us
949 * `file' can be NULL - eg, when called from page_symlink().
951 * ext4 never places buffers on inode->i_mapping->private_list. metadata
952 * buffers are managed internally.
954 static int ext4_ordered_write_end(struct file *file,
955 struct address_space *mapping,
956 loff_t pos, unsigned len, unsigned copied,
957 struct page *page, void *fsdata)
959 handle_t *handle = ext4_journal_current_handle();
960 struct inode *inode = mapping->host;
963 trace_ext4_ordered_write_end(inode, pos, len, copied);
964 ret = ext4_jbd2_file_inode(handle, inode);
967 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
970 if (pos + len > inode->i_size && ext4_can_truncate(inode))
971 /* if we have allocated more blocks and copied
972 * less. We will have blocks allocated outside
973 * inode->i_size. So truncate them
975 ext4_orphan_add(handle, inode);
980 page_cache_release(page);
983 ret2 = ext4_journal_stop(handle);
987 if (pos + len > inode->i_size) {
988 ext4_truncate_failed_write(inode);
990 * If truncate failed early the inode might still be
991 * on the orphan list; we need to make sure the inode
992 * is removed from the orphan list in that case.
995 ext4_orphan_del(NULL, inode);
999 return ret ? ret : copied;
1002 static int ext4_writeback_write_end(struct file *file,
1003 struct address_space *mapping,
1004 loff_t pos, unsigned len, unsigned copied,
1005 struct page *page, void *fsdata)
1007 handle_t *handle = ext4_journal_current_handle();
1008 struct inode *inode = mapping->host;
1011 trace_ext4_writeback_write_end(inode, pos, len, copied);
1012 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1015 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1016 /* if we have allocated more blocks and copied
1017 * less. We will have blocks allocated outside
1018 * inode->i_size. So truncate them
1020 ext4_orphan_add(handle, inode);
1025 ret2 = ext4_journal_stop(handle);
1029 if (pos + len > inode->i_size) {
1030 ext4_truncate_failed_write(inode);
1032 * If truncate failed early the inode might still be
1033 * on the orphan list; we need to make sure the inode
1034 * is removed from the orphan list in that case.
1037 ext4_orphan_del(NULL, inode);
1040 return ret ? ret : copied;
1043 static int ext4_journalled_write_end(struct file *file,
1044 struct address_space *mapping,
1045 loff_t pos, unsigned len, unsigned copied,
1046 struct page *page, void *fsdata)
1048 handle_t *handle = ext4_journal_current_handle();
1049 struct inode *inode = mapping->host;
1055 trace_ext4_journalled_write_end(inode, pos, len, copied);
1056 from = pos & (PAGE_CACHE_SIZE - 1);
1059 BUG_ON(!ext4_handle_valid(handle));
1062 if (!PageUptodate(page))
1064 page_zero_new_buffers(page, from+copied, to);
1067 ret = walk_page_buffers(handle, page_buffers(page), from,
1068 to, &partial, write_end_fn);
1070 SetPageUptodate(page);
1071 new_i_size = pos + copied;
1072 if (new_i_size > inode->i_size)
1073 i_size_write(inode, pos+copied);
1074 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1075 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1076 if (new_i_size > EXT4_I(inode)->i_disksize) {
1077 ext4_update_i_disksize(inode, new_i_size);
1078 ret2 = ext4_mark_inode_dirty(handle, inode);
1084 page_cache_release(page);
1085 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1086 /* if we have allocated more blocks and copied
1087 * less. We will have blocks allocated outside
1088 * inode->i_size. So truncate them
1090 ext4_orphan_add(handle, inode);
1092 ret2 = ext4_journal_stop(handle);
1095 if (pos + len > inode->i_size) {
1096 ext4_truncate_failed_write(inode);
1098 * If truncate failed early the inode might still be
1099 * on the orphan list; we need to make sure the inode
1100 * is removed from the orphan list in that case.
1103 ext4_orphan_del(NULL, inode);
1106 return ret ? ret : copied;
1110 * Reserve a single cluster located at lblock
1112 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1115 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1116 struct ext4_inode_info *ei = EXT4_I(inode);
1117 unsigned int md_needed;
1119 ext4_lblk_t save_last_lblock;
1123 * We will charge metadata quota at writeout time; this saves
1124 * us from metadata over-estimation, though we may go over by
1125 * a small amount in the end. Here we just reserve for data.
1127 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1132 * recalculate the amount of metadata blocks to reserve
1133 * in order to allocate nrblocks
1134 * worse case is one extent per block
1137 spin_lock(&ei->i_block_reservation_lock);
1139 * ext4_calc_metadata_amount() has side effects, which we have
1140 * to be prepared undo if we fail to claim space.
1142 save_len = ei->i_da_metadata_calc_len;
1143 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1144 md_needed = EXT4_NUM_B2C(sbi,
1145 ext4_calc_metadata_amount(inode, lblock));
1146 trace_ext4_da_reserve_space(inode, md_needed);
1149 * We do still charge estimated metadata to the sb though;
1150 * we cannot afford to run out of free blocks.
1152 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1153 ei->i_da_metadata_calc_len = save_len;
1154 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1155 spin_unlock(&ei->i_block_reservation_lock);
1156 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1160 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1163 ei->i_reserved_data_blocks++;
1164 ei->i_reserved_meta_blocks += md_needed;
1165 spin_unlock(&ei->i_block_reservation_lock);
1167 return 0; /* success */
1170 static void ext4_da_release_space(struct inode *inode, int to_free)
1172 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1173 struct ext4_inode_info *ei = EXT4_I(inode);
1176 return; /* Nothing to release, exit */
1178 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1180 trace_ext4_da_release_space(inode, to_free);
1181 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1183 * if there aren't enough reserved blocks, then the
1184 * counter is messed up somewhere. Since this
1185 * function is called from invalidate page, it's
1186 * harmless to return without any action.
1188 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1189 "ino %lu, to_free %d with only %d reserved "
1190 "data blocks\n", inode->i_ino, to_free,
1191 ei->i_reserved_data_blocks);
1193 to_free = ei->i_reserved_data_blocks;
1195 ei->i_reserved_data_blocks -= to_free;
1197 if (ei->i_reserved_data_blocks == 0) {
1199 * We can release all of the reserved metadata blocks
1200 * only when we have written all of the delayed
1201 * allocation blocks.
1202 * Note that in case of bigalloc, i_reserved_meta_blocks,
1203 * i_reserved_data_blocks, etc. refer to number of clusters.
1205 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1206 ei->i_reserved_meta_blocks);
1207 ei->i_reserved_meta_blocks = 0;
1208 ei->i_da_metadata_calc_len = 0;
1211 /* update fs dirty data blocks counter */
1212 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1214 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1216 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1219 static void ext4_da_page_release_reservation(struct page *page,
1220 unsigned long offset)
1223 struct buffer_head *head, *bh;
1224 unsigned int curr_off = 0;
1225 struct inode *inode = page->mapping->host;
1226 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1229 head = page_buffers(page);
1232 unsigned int next_off = curr_off + bh->b_size;
1234 if ((offset <= curr_off) && (buffer_delay(bh))) {
1236 clear_buffer_delay(bh);
1237 clear_buffer_da_mapped(bh);
1239 curr_off = next_off;
1240 } while ((bh = bh->b_this_page) != head);
1242 /* If we have released all the blocks belonging to a cluster, then we
1243 * need to release the reserved space for that cluster. */
1244 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1245 while (num_clusters > 0) {
1247 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1248 ((num_clusters - 1) << sbi->s_cluster_bits);
1249 if (sbi->s_cluster_ratio == 1 ||
1250 !ext4_find_delalloc_cluster(inode, lblk, 1))
1251 ext4_da_release_space(inode, 1);
1258 * Delayed allocation stuff
1262 * mpage_da_submit_io - walks through extent of pages and try to write
1263 * them with writepage() call back
1265 * @mpd->inode: inode
1266 * @mpd->first_page: first page of the extent
1267 * @mpd->next_page: page after the last page of the extent
1269 * By the time mpage_da_submit_io() is called we expect all blocks
1270 * to be allocated. this may be wrong if allocation failed.
1272 * As pages are already locked by write_cache_pages(), we can't use it
1274 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1275 struct ext4_map_blocks *map)
1277 struct pagevec pvec;
1278 unsigned long index, end;
1279 int ret = 0, err, nr_pages, i;
1280 struct inode *inode = mpd->inode;
1281 struct address_space *mapping = inode->i_mapping;
1282 loff_t size = i_size_read(inode);
1283 unsigned int len, block_start;
1284 struct buffer_head *bh, *page_bufs = NULL;
1285 int journal_data = ext4_should_journal_data(inode);
1286 sector_t pblock = 0, cur_logical = 0;
1287 struct ext4_io_submit io_submit;
1289 BUG_ON(mpd->next_page <= mpd->first_page);
1290 memset(&io_submit, 0, sizeof(io_submit));
1292 * We need to start from the first_page to the next_page - 1
1293 * to make sure we also write the mapped dirty buffer_heads.
1294 * If we look at mpd->b_blocknr we would only be looking
1295 * at the currently mapped buffer_heads.
1297 index = mpd->first_page;
1298 end = mpd->next_page - 1;
1300 pagevec_init(&pvec, 0);
1301 while (index <= end) {
1302 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1305 for (i = 0; i < nr_pages; i++) {
1306 int commit_write = 0, skip_page = 0;
1307 struct page *page = pvec.pages[i];
1309 index = page->index;
1313 if (index == size >> PAGE_CACHE_SHIFT)
1314 len = size & ~PAGE_CACHE_MASK;
1316 len = PAGE_CACHE_SIZE;
1318 cur_logical = index << (PAGE_CACHE_SHIFT -
1320 pblock = map->m_pblk + (cur_logical -
1325 BUG_ON(!PageLocked(page));
1326 BUG_ON(PageWriteback(page));
1329 * If the page does not have buffers (for
1330 * whatever reason), try to create them using
1331 * __block_write_begin. If this fails,
1332 * skip the page and move on.
1334 if (!page_has_buffers(page)) {
1335 if (__block_write_begin(page, 0, len,
1336 noalloc_get_block_write)) {
1344 bh = page_bufs = page_buffers(page);
1349 if (map && (cur_logical >= map->m_lblk) &&
1350 (cur_logical <= (map->m_lblk +
1351 (map->m_len - 1)))) {
1352 if (buffer_delay(bh)) {
1353 clear_buffer_delay(bh);
1354 bh->b_blocknr = pblock;
1356 if (buffer_da_mapped(bh))
1357 clear_buffer_da_mapped(bh);
1358 if (buffer_unwritten(bh) ||
1360 BUG_ON(bh->b_blocknr != pblock);
1361 if (map->m_flags & EXT4_MAP_UNINIT)
1362 set_buffer_uninit(bh);
1363 clear_buffer_unwritten(bh);
1367 * skip page if block allocation undone and
1370 if (ext4_bh_delay_or_unwritten(NULL, bh))
1372 bh = bh->b_this_page;
1373 block_start += bh->b_size;
1376 } while (bh != page_bufs);
1382 /* mark the buffer_heads as dirty & uptodate */
1383 block_commit_write(page, 0, len);
1385 clear_page_dirty_for_io(page);
1387 * Delalloc doesn't support data journalling,
1388 * but eventually maybe we'll lift this
1391 if (unlikely(journal_data && PageChecked(page)))
1392 err = __ext4_journalled_writepage(page, len);
1393 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1394 err = ext4_bio_write_page(&io_submit, page,
1396 else if (buffer_uninit(page_bufs)) {
1397 ext4_set_bh_endio(page_bufs, inode);
1398 err = block_write_full_page_endio(page,
1399 noalloc_get_block_write,
1400 mpd->wbc, ext4_end_io_buffer_write);
1402 err = block_write_full_page(page,
1403 noalloc_get_block_write, mpd->wbc);
1406 mpd->pages_written++;
1408 * In error case, we have to continue because
1409 * remaining pages are still locked
1414 pagevec_release(&pvec);
1416 ext4_io_submit(&io_submit);
1420 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1424 struct pagevec pvec;
1425 struct inode *inode = mpd->inode;
1426 struct address_space *mapping = inode->i_mapping;
1428 index = mpd->first_page;
1429 end = mpd->next_page - 1;
1430 pagevec_init(&pvec, 0);
1431 while (index <= end) {
1432 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1435 for (i = 0; i < nr_pages; i++) {
1436 struct page *page = pvec.pages[i];
1437 if (page->index > end)
1439 BUG_ON(!PageLocked(page));
1440 BUG_ON(PageWriteback(page));
1441 block_invalidatepage(page, 0);
1442 ClearPageUptodate(page);
1445 index = pvec.pages[nr_pages - 1]->index + 1;
1446 pagevec_release(&pvec);
1451 static void ext4_print_free_blocks(struct inode *inode)
1453 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1454 printk(KERN_CRIT "Total free blocks count %lld\n",
1455 EXT4_C2B(EXT4_SB(inode->i_sb),
1456 ext4_count_free_clusters(inode->i_sb)));
1457 printk(KERN_CRIT "Free/Dirty block details\n");
1458 printk(KERN_CRIT "free_blocks=%lld\n",
1459 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1460 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1461 printk(KERN_CRIT "dirty_blocks=%lld\n",
1462 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1463 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1464 printk(KERN_CRIT "Block reservation details\n");
1465 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
1466 EXT4_I(inode)->i_reserved_data_blocks);
1467 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
1468 EXT4_I(inode)->i_reserved_meta_blocks);
1473 * mpage_da_map_and_submit - go through given space, map them
1474 * if necessary, and then submit them for I/O
1476 * @mpd - bh describing space
1478 * The function skips space we know is already mapped to disk blocks.
1481 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1483 int err, blks, get_blocks_flags;
1484 struct ext4_map_blocks map, *mapp = NULL;
1485 sector_t next = mpd->b_blocknr;
1486 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1487 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1488 handle_t *handle = NULL;
1491 * If the blocks are mapped already, or we couldn't accumulate
1492 * any blocks, then proceed immediately to the submission stage.
1494 if ((mpd->b_size == 0) ||
1495 ((mpd->b_state & (1 << BH_Mapped)) &&
1496 !(mpd->b_state & (1 << BH_Delay)) &&
1497 !(mpd->b_state & (1 << BH_Unwritten))))
1500 handle = ext4_journal_current_handle();
1504 * Call ext4_map_blocks() to allocate any delayed allocation
1505 * blocks, or to convert an uninitialized extent to be
1506 * initialized (in the case where we have written into
1507 * one or more preallocated blocks).
1509 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1510 * indicate that we are on the delayed allocation path. This
1511 * affects functions in many different parts of the allocation
1512 * call path. This flag exists primarily because we don't
1513 * want to change *many* call functions, so ext4_map_blocks()
1514 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1515 * inode's allocation semaphore is taken.
1517 * If the blocks in questions were delalloc blocks, set
1518 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1519 * variables are updated after the blocks have been allocated.
1522 map.m_len = max_blocks;
1523 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1524 if (ext4_should_dioread_nolock(mpd->inode))
1525 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1526 if (mpd->b_state & (1 << BH_Delay))
1527 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1529 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1531 struct super_block *sb = mpd->inode->i_sb;
1535 * If get block returns EAGAIN or ENOSPC and there
1536 * appears to be free blocks we will just let
1537 * mpage_da_submit_io() unlock all of the pages.
1542 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1548 * get block failure will cause us to loop in
1549 * writepages, because a_ops->writepage won't be able
1550 * to make progress. The page will be redirtied by
1551 * writepage and writepages will again try to write
1554 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1555 ext4_msg(sb, KERN_CRIT,
1556 "delayed block allocation failed for inode %lu "
1557 "at logical offset %llu with max blocks %zd "
1558 "with error %d", mpd->inode->i_ino,
1559 (unsigned long long) next,
1560 mpd->b_size >> mpd->inode->i_blkbits, err);
1561 ext4_msg(sb, KERN_CRIT,
1562 "This should not happen!! Data will be lost\n");
1564 ext4_print_free_blocks(mpd->inode);
1566 /* invalidate all the pages */
1567 ext4_da_block_invalidatepages(mpd);
1569 /* Mark this page range as having been completed */
1576 if (map.m_flags & EXT4_MAP_NEW) {
1577 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1580 for (i = 0; i < map.m_len; i++)
1581 unmap_underlying_metadata(bdev, map.m_pblk + i);
1583 if (ext4_should_order_data(mpd->inode)) {
1584 err = ext4_jbd2_file_inode(handle, mpd->inode);
1586 /* Only if the journal is aborted */
1594 * Update on-disk size along with block allocation.
1596 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1597 if (disksize > i_size_read(mpd->inode))
1598 disksize = i_size_read(mpd->inode);
1599 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1600 ext4_update_i_disksize(mpd->inode, disksize);
1601 err = ext4_mark_inode_dirty(handle, mpd->inode);
1603 ext4_error(mpd->inode->i_sb,
1604 "Failed to mark inode %lu dirty",
1609 mpage_da_submit_io(mpd, mapp);
1613 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1614 (1 << BH_Delay) | (1 << BH_Unwritten))
1617 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1619 * @mpd->lbh - extent of blocks
1620 * @logical - logical number of the block in the file
1621 * @bh - bh of the block (used to access block's state)
1623 * the function is used to collect contig. blocks in same state
1625 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1626 sector_t logical, size_t b_size,
1627 unsigned long b_state)
1630 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1633 * XXX Don't go larger than mballoc is willing to allocate
1634 * This is a stopgap solution. We eventually need to fold
1635 * mpage_da_submit_io() into this function and then call
1636 * ext4_map_blocks() multiple times in a loop
1638 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1641 /* check if thereserved journal credits might overflow */
1642 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1643 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1645 * With non-extent format we are limited by the journal
1646 * credit available. Total credit needed to insert
1647 * nrblocks contiguous blocks is dependent on the
1648 * nrblocks. So limit nrblocks.
1651 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1652 EXT4_MAX_TRANS_DATA) {
1654 * Adding the new buffer_head would make it cross the
1655 * allowed limit for which we have journal credit
1656 * reserved. So limit the new bh->b_size
1658 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1659 mpd->inode->i_blkbits;
1660 /* we will do mpage_da_submit_io in the next loop */
1664 * First block in the extent
1666 if (mpd->b_size == 0) {
1667 mpd->b_blocknr = logical;
1668 mpd->b_size = b_size;
1669 mpd->b_state = b_state & BH_FLAGS;
1673 next = mpd->b_blocknr + nrblocks;
1675 * Can we merge the block to our big extent?
1677 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1678 mpd->b_size += b_size;
1684 * We couldn't merge the block to our extent, so we
1685 * need to flush current extent and start new one
1687 mpage_da_map_and_submit(mpd);
1691 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1693 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1697 * This function is grabs code from the very beginning of
1698 * ext4_map_blocks, but assumes that the caller is from delayed write
1699 * time. This function looks up the requested blocks and sets the
1700 * buffer delay bit under the protection of i_data_sem.
1702 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1703 struct ext4_map_blocks *map,
1704 struct buffer_head *bh)
1707 sector_t invalid_block = ~((sector_t) 0xffff);
1709 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1713 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1714 "logical block %lu\n", inode->i_ino, map->m_len,
1715 (unsigned long) map->m_lblk);
1717 * Try to see if we can get the block without requesting a new
1718 * file system block.
1720 down_read((&EXT4_I(inode)->i_data_sem));
1721 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1722 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1724 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1728 * XXX: __block_prepare_write() unmaps passed block,
1731 /* If the block was allocated from previously allocated cluster,
1732 * then we dont need to reserve it again. */
1733 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1734 retval = ext4_da_reserve_space(inode, iblock);
1736 /* not enough space to reserve */
1740 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1741 * and it should not appear on the bh->b_state.
1743 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1745 map_bh(bh, inode->i_sb, invalid_block);
1747 set_buffer_delay(bh);
1751 up_read((&EXT4_I(inode)->i_data_sem));
1757 * This is a special get_blocks_t callback which is used by
1758 * ext4_da_write_begin(). It will either return mapped block or
1759 * reserve space for a single block.
1761 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1762 * We also have b_blocknr = -1 and b_bdev initialized properly
1764 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1765 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1766 * initialized properly.
1768 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1769 struct buffer_head *bh, int create)
1771 struct ext4_map_blocks map;
1774 BUG_ON(create == 0);
1775 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1777 map.m_lblk = iblock;
1781 * first, we need to know whether the block is allocated already
1782 * preallocated blocks are unmapped but should treated
1783 * the same as allocated blocks.
1785 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1789 map_bh(bh, inode->i_sb, map.m_pblk);
1790 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1792 if (buffer_unwritten(bh)) {
1793 /* A delayed write to unwritten bh should be marked
1794 * new and mapped. Mapped ensures that we don't do
1795 * get_block multiple times when we write to the same
1796 * offset and new ensures that we do proper zero out
1797 * for partial write.
1800 set_buffer_mapped(bh);
1806 * This function is used as a standard get_block_t calback function
1807 * when there is no desire to allocate any blocks. It is used as a
1808 * callback function for block_write_begin() and block_write_full_page().
1809 * These functions should only try to map a single block at a time.
1811 * Since this function doesn't do block allocations even if the caller
1812 * requests it by passing in create=1, it is critically important that
1813 * any caller checks to make sure that any buffer heads are returned
1814 * by this function are either all already mapped or marked for
1815 * delayed allocation before calling block_write_full_page(). Otherwise,
1816 * b_blocknr could be left unitialized, and the page write functions will
1817 * be taken by surprise.
1819 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1820 struct buffer_head *bh_result, int create)
1822 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1823 return _ext4_get_block(inode, iblock, bh_result, 0);
1826 static int bget_one(handle_t *handle, struct buffer_head *bh)
1832 static int bput_one(handle_t *handle, struct buffer_head *bh)
1838 static int __ext4_journalled_writepage(struct page *page,
1841 struct address_space *mapping = page->mapping;
1842 struct inode *inode = mapping->host;
1843 struct buffer_head *page_bufs;
1844 handle_t *handle = NULL;
1848 ClearPageChecked(page);
1849 page_bufs = page_buffers(page);
1851 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1853 * We need to release the page lock before we start the
1854 * journal, so grab a reference so the page won't disappear
1855 * out from under us.
1860 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1861 if (IS_ERR(handle)) {
1862 ret = PTR_ERR(handle);
1864 goto out_no_pagelock;
1866 BUG_ON(!ext4_handle_valid(handle));
1870 if (page->mapping != mapping) {
1871 /* The page got truncated from under us */
1872 ext4_journal_stop(handle);
1877 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1878 do_journal_get_write_access);
1880 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1884 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1885 err = ext4_journal_stop(handle);
1889 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
1890 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1897 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
1898 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
1901 * Note that we don't need to start a transaction unless we're journaling data
1902 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1903 * need to file the inode to the transaction's list in ordered mode because if
1904 * we are writing back data added by write(), the inode is already there and if
1905 * we are writing back data modified via mmap(), no one guarantees in which
1906 * transaction the data will hit the disk. In case we are journaling data, we
1907 * cannot start transaction directly because transaction start ranks above page
1908 * lock so we have to do some magic.
1910 * This function can get called via...
1911 * - ext4_da_writepages after taking page lock (have journal handle)
1912 * - journal_submit_inode_data_buffers (no journal handle)
1913 * - shrink_page_list via pdflush (no journal handle)
1914 * - grab_page_cache when doing write_begin (have journal handle)
1916 * We don't do any block allocation in this function. If we have page with
1917 * multiple blocks we need to write those buffer_heads that are mapped. This
1918 * is important for mmaped based write. So if we do with blocksize 1K
1919 * truncate(f, 1024);
1920 * a = mmap(f, 0, 4096);
1922 * truncate(f, 4096);
1923 * we have in the page first buffer_head mapped via page_mkwrite call back
1924 * but other bufer_heads would be unmapped but dirty(dirty done via the
1925 * do_wp_page). So writepage should write the first block. If we modify
1926 * the mmap area beyond 1024 we will again get a page_fault and the
1927 * page_mkwrite callback will do the block allocation and mark the
1928 * buffer_heads mapped.
1930 * We redirty the page if we have any buffer_heads that is either delay or
1931 * unwritten in the page.
1933 * We can get recursively called as show below.
1935 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1938 * But since we don't do any block allocation we should not deadlock.
1939 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1941 static int ext4_writepage(struct page *page,
1942 struct writeback_control *wbc)
1944 int ret = 0, commit_write = 0;
1947 struct buffer_head *page_bufs = NULL;
1948 struct inode *inode = page->mapping->host;
1950 trace_ext4_writepage(page);
1951 size = i_size_read(inode);
1952 if (page->index == size >> PAGE_CACHE_SHIFT)
1953 len = size & ~PAGE_CACHE_MASK;
1955 len = PAGE_CACHE_SIZE;
1958 * If the page does not have buffers (for whatever reason),
1959 * try to create them using __block_write_begin. If this
1960 * fails, redirty the page and move on.
1962 if (!page_has_buffers(page)) {
1963 if (__block_write_begin(page, 0, len,
1964 noalloc_get_block_write)) {
1966 redirty_page_for_writepage(wbc, page);
1972 page_bufs = page_buffers(page);
1973 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1974 ext4_bh_delay_or_unwritten)) {
1976 * We don't want to do block allocation, so redirty
1977 * the page and return. We may reach here when we do
1978 * a journal commit via journal_submit_inode_data_buffers.
1979 * We can also reach here via shrink_page_list but it
1980 * should never be for direct reclaim so warn if that
1983 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
1988 /* now mark the buffer_heads as dirty and uptodate */
1989 block_commit_write(page, 0, len);
1991 if (PageChecked(page) && ext4_should_journal_data(inode))
1993 * It's mmapped pagecache. Add buffers and journal it. There
1994 * doesn't seem much point in redirtying the page here.
1996 return __ext4_journalled_writepage(page, len);
1998 if (buffer_uninit(page_bufs)) {
1999 ext4_set_bh_endio(page_bufs, inode);
2000 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2001 wbc, ext4_end_io_buffer_write);
2003 ret = block_write_full_page(page, noalloc_get_block_write,
2010 * This is called via ext4_da_writepages() to
2011 * calculate the total number of credits to reserve to fit
2012 * a single extent allocation into a single transaction,
2013 * ext4_da_writpeages() will loop calling this before
2014 * the block allocation.
2017 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2019 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2022 * With non-extent format the journal credit needed to
2023 * insert nrblocks contiguous block is dependent on
2024 * number of contiguous block. So we will limit
2025 * number of contiguous block to a sane value
2027 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2028 (max_blocks > EXT4_MAX_TRANS_DATA))
2029 max_blocks = EXT4_MAX_TRANS_DATA;
2031 return ext4_chunk_trans_blocks(inode, max_blocks);
2035 * write_cache_pages_da - walk the list of dirty pages of the given
2036 * address space and accumulate pages that need writing, and call
2037 * mpage_da_map_and_submit to map a single contiguous memory region
2038 * and then write them.
2040 static int write_cache_pages_da(struct address_space *mapping,
2041 struct writeback_control *wbc,
2042 struct mpage_da_data *mpd,
2043 pgoff_t *done_index)
2045 struct buffer_head *bh, *head;
2046 struct inode *inode = mapping->host;
2047 struct pagevec pvec;
2048 unsigned int nr_pages;
2051 long nr_to_write = wbc->nr_to_write;
2052 int i, tag, ret = 0;
2054 memset(mpd, 0, sizeof(struct mpage_da_data));
2057 pagevec_init(&pvec, 0);
2058 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2059 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2061 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2062 tag = PAGECACHE_TAG_TOWRITE;
2064 tag = PAGECACHE_TAG_DIRTY;
2066 *done_index = index;
2067 while (index <= end) {
2068 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2069 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2073 for (i = 0; i < nr_pages; i++) {
2074 struct page *page = pvec.pages[i];
2077 * At this point, the page may be truncated or
2078 * invalidated (changing page->mapping to NULL), or
2079 * even swizzled back from swapper_space to tmpfs file
2080 * mapping. However, page->index will not change
2081 * because we have a reference on the page.
2083 if (page->index > end)
2086 *done_index = page->index + 1;
2089 * If we can't merge this page, and we have
2090 * accumulated an contiguous region, write it
2092 if ((mpd->next_page != page->index) &&
2093 (mpd->next_page != mpd->first_page)) {
2094 mpage_da_map_and_submit(mpd);
2095 goto ret_extent_tail;
2101 * If the page is no longer dirty, or its
2102 * mapping no longer corresponds to inode we
2103 * are writing (which means it has been
2104 * truncated or invalidated), or the page is
2105 * already under writeback and we are not
2106 * doing a data integrity writeback, skip the page
2108 if (!PageDirty(page) ||
2109 (PageWriteback(page) &&
2110 (wbc->sync_mode == WB_SYNC_NONE)) ||
2111 unlikely(page->mapping != mapping)) {
2116 wait_on_page_writeback(page);
2117 BUG_ON(PageWriteback(page));
2119 if (mpd->next_page != page->index)
2120 mpd->first_page = page->index;
2121 mpd->next_page = page->index + 1;
2122 logical = (sector_t) page->index <<
2123 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2125 if (!page_has_buffers(page)) {
2126 mpage_add_bh_to_extent(mpd, logical,
2128 (1 << BH_Dirty) | (1 << BH_Uptodate));
2130 goto ret_extent_tail;
2133 * Page with regular buffer heads,
2134 * just add all dirty ones
2136 head = page_buffers(page);
2139 BUG_ON(buffer_locked(bh));
2141 * We need to try to allocate
2142 * unmapped blocks in the same page.
2143 * Otherwise we won't make progress
2144 * with the page in ext4_writepage
2146 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2147 mpage_add_bh_to_extent(mpd, logical,
2151 goto ret_extent_tail;
2152 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2154 * mapped dirty buffer. We need
2155 * to update the b_state
2156 * because we look at b_state
2157 * in mpage_da_map_blocks. We
2158 * don't update b_size because
2159 * if we find an unmapped
2160 * buffer_head later we need to
2161 * use the b_state flag of that
2164 if (mpd->b_size == 0)
2165 mpd->b_state = bh->b_state & BH_FLAGS;
2168 } while ((bh = bh->b_this_page) != head);
2171 if (nr_to_write > 0) {
2173 if (nr_to_write == 0 &&
2174 wbc->sync_mode == WB_SYNC_NONE)
2176 * We stop writing back only if we are
2177 * not doing integrity sync. In case of
2178 * integrity sync we have to keep going
2179 * because someone may be concurrently
2180 * dirtying pages, and we might have
2181 * synced a lot of newly appeared dirty
2182 * pages, but have not synced all of the
2188 pagevec_release(&pvec);
2193 ret = MPAGE_DA_EXTENT_TAIL;
2195 pagevec_release(&pvec);
2201 static int ext4_da_writepages(struct address_space *mapping,
2202 struct writeback_control *wbc)
2205 int range_whole = 0;
2206 handle_t *handle = NULL;
2207 struct mpage_da_data mpd;
2208 struct inode *inode = mapping->host;
2209 int pages_written = 0;
2210 unsigned int max_pages;
2211 int range_cyclic, cycled = 1, io_done = 0;
2212 int needed_blocks, ret = 0;
2213 long desired_nr_to_write, nr_to_writebump = 0;
2214 loff_t range_start = wbc->range_start;
2215 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2216 pgoff_t done_index = 0;
2218 struct blk_plug plug;
2220 trace_ext4_da_writepages(inode, wbc);
2223 * No pages to write? This is mainly a kludge to avoid starting
2224 * a transaction for special inodes like journal inode on last iput()
2225 * because that could violate lock ordering on umount
2227 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2231 * If the filesystem has aborted, it is read-only, so return
2232 * right away instead of dumping stack traces later on that
2233 * will obscure the real source of the problem. We test
2234 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2235 * the latter could be true if the filesystem is mounted
2236 * read-only, and in that case, ext4_da_writepages should
2237 * *never* be called, so if that ever happens, we would want
2240 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2243 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2246 range_cyclic = wbc->range_cyclic;
2247 if (wbc->range_cyclic) {
2248 index = mapping->writeback_index;
2251 wbc->range_start = index << PAGE_CACHE_SHIFT;
2252 wbc->range_end = LLONG_MAX;
2253 wbc->range_cyclic = 0;
2256 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2257 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2261 * This works around two forms of stupidity. The first is in
2262 * the writeback code, which caps the maximum number of pages
2263 * written to be 1024 pages. This is wrong on multiple
2264 * levels; different architectues have a different page size,
2265 * which changes the maximum amount of data which gets
2266 * written. Secondly, 4 megabytes is way too small. XFS
2267 * forces this value to be 16 megabytes by multiplying
2268 * nr_to_write parameter by four, and then relies on its
2269 * allocator to allocate larger extents to make them
2270 * contiguous. Unfortunately this brings us to the second
2271 * stupidity, which is that ext4's mballoc code only allocates
2272 * at most 2048 blocks. So we force contiguous writes up to
2273 * the number of dirty blocks in the inode, or
2274 * sbi->max_writeback_mb_bump whichever is smaller.
2276 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2277 if (!range_cyclic && range_whole) {
2278 if (wbc->nr_to_write == LONG_MAX)
2279 desired_nr_to_write = wbc->nr_to_write;
2281 desired_nr_to_write = wbc->nr_to_write * 8;
2283 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2285 if (desired_nr_to_write > max_pages)
2286 desired_nr_to_write = max_pages;
2288 if (wbc->nr_to_write < desired_nr_to_write) {
2289 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2290 wbc->nr_to_write = desired_nr_to_write;
2294 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2295 tag_pages_for_writeback(mapping, index, end);
2297 blk_start_plug(&plug);
2298 while (!ret && wbc->nr_to_write > 0) {
2301 * we insert one extent at a time. So we need
2302 * credit needed for single extent allocation.
2303 * journalled mode is currently not supported
2306 BUG_ON(ext4_should_journal_data(inode));
2307 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2309 /* start a new transaction*/
2310 handle = ext4_journal_start(inode, needed_blocks);
2311 if (IS_ERR(handle)) {
2312 ret = PTR_ERR(handle);
2313 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2314 "%ld pages, ino %lu; err %d", __func__,
2315 wbc->nr_to_write, inode->i_ino, ret);
2316 blk_finish_plug(&plug);
2317 goto out_writepages;
2321 * Now call write_cache_pages_da() to find the next
2322 * contiguous region of logical blocks that need
2323 * blocks to be allocated by ext4 and submit them.
2325 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
2327 * If we have a contiguous extent of pages and we
2328 * haven't done the I/O yet, map the blocks and submit
2331 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2332 mpage_da_map_and_submit(&mpd);
2333 ret = MPAGE_DA_EXTENT_TAIL;
2335 trace_ext4_da_write_pages(inode, &mpd);
2336 wbc->nr_to_write -= mpd.pages_written;
2338 ext4_journal_stop(handle);
2340 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2341 /* commit the transaction which would
2342 * free blocks released in the transaction
2345 jbd2_journal_force_commit_nested(sbi->s_journal);
2347 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2349 * Got one extent now try with rest of the pages.
2350 * If mpd.retval is set -EIO, journal is aborted.
2351 * So we don't need to write any more.
2353 pages_written += mpd.pages_written;
2356 } else if (wbc->nr_to_write)
2358 * There is no more writeout needed
2359 * or we requested for a noblocking writeout
2360 * and we found the device congested
2364 blk_finish_plug(&plug);
2365 if (!io_done && !cycled) {
2368 wbc->range_start = index << PAGE_CACHE_SHIFT;
2369 wbc->range_end = mapping->writeback_index - 1;
2374 wbc->range_cyclic = range_cyclic;
2375 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2377 * set the writeback_index so that range_cyclic
2378 * mode will write it back later
2380 mapping->writeback_index = done_index;
2383 wbc->nr_to_write -= nr_to_writebump;
2384 wbc->range_start = range_start;
2385 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2389 #define FALL_BACK_TO_NONDELALLOC 1
2390 static int ext4_nonda_switch(struct super_block *sb)
2392 s64 free_blocks, dirty_blocks;
2393 struct ext4_sb_info *sbi = EXT4_SB(sb);
2396 * switch to non delalloc mode if we are running low
2397 * on free block. The free block accounting via percpu
2398 * counters can get slightly wrong with percpu_counter_batch getting
2399 * accumulated on each CPU without updating global counters
2400 * Delalloc need an accurate free block accounting. So switch
2401 * to non delalloc when we are near to error range.
2403 free_blocks = EXT4_C2B(sbi,
2404 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2405 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2407 * Start pushing delalloc when 1/2 of free blocks are dirty.
2409 if (dirty_blocks && (free_blocks < 2 * dirty_blocks) &&
2410 !writeback_in_progress(sb->s_bdi) &&
2411 down_read_trylock(&sb->s_umount)) {
2412 writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2413 up_read(&sb->s_umount);
2416 if (2 * free_blocks < 3 * dirty_blocks ||
2417 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2419 * free block count is less than 150% of dirty blocks
2420 * or free blocks is less than watermark
2427 /* We always reserve for an inode update; the superblock could be there too */
2428 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2430 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
2431 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2434 if (pos + len <= 0x7fffffffULL)
2437 /* We might need to update the superblock to set LARGE_FILE */
2441 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2442 loff_t pos, unsigned len, unsigned flags,
2443 struct page **pagep, void **fsdata)
2445 int ret, retries = 0;
2448 struct inode *inode = mapping->host;
2451 index = pos >> PAGE_CACHE_SHIFT;
2453 if (ext4_nonda_switch(inode->i_sb)) {
2454 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2455 return ext4_write_begin(file, mapping, pos,
2456 len, flags, pagep, fsdata);
2458 *fsdata = (void *)0;
2459 trace_ext4_da_write_begin(inode, pos, len, flags);
2462 * With delayed allocation, we don't log the i_disksize update
2463 * if there is delayed block allocation. But we still need
2464 * to journalling the i_disksize update if writes to the end
2465 * of file which has an already mapped buffer.
2467 handle = ext4_journal_start(inode,
2468 ext4_da_write_credits(inode, pos, len));
2469 if (IS_ERR(handle)) {
2470 ret = PTR_ERR(handle);
2473 /* We cannot recurse into the filesystem as the transaction is already
2475 flags |= AOP_FLAG_NOFS;
2477 page = grab_cache_page_write_begin(mapping, index, flags);
2479 ext4_journal_stop(handle);
2485 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2488 ext4_journal_stop(handle);
2489 page_cache_release(page);
2491 * block_write_begin may have instantiated a few blocks
2492 * outside i_size. Trim these off again. Don't need
2493 * i_size_read because we hold i_mutex.
2495 if (pos + len > inode->i_size)
2496 ext4_truncate_failed_write(inode);
2499 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2506 * Check if we should update i_disksize
2507 * when write to the end of file but not require block allocation
2509 static int ext4_da_should_update_i_disksize(struct page *page,
2510 unsigned long offset)
2512 struct buffer_head *bh;
2513 struct inode *inode = page->mapping->host;
2517 bh = page_buffers(page);
2518 idx = offset >> inode->i_blkbits;
2520 for (i = 0; i < idx; i++)
2521 bh = bh->b_this_page;
2523 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2528 static int ext4_da_write_end(struct file *file,
2529 struct address_space *mapping,
2530 loff_t pos, unsigned len, unsigned copied,
2531 struct page *page, void *fsdata)
2533 struct inode *inode = mapping->host;
2535 handle_t *handle = ext4_journal_current_handle();
2537 unsigned long start, end;
2538 int write_mode = (int)(unsigned long)fsdata;
2540 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2541 switch (ext4_inode_journal_mode(inode)) {
2542 case EXT4_INODE_ORDERED_DATA_MODE:
2543 return ext4_ordered_write_end(file, mapping, pos,
2544 len, copied, page, fsdata);
2545 case EXT4_INODE_WRITEBACK_DATA_MODE:
2546 return ext4_writeback_write_end(file, mapping, pos,
2547 len, copied, page, fsdata);
2553 trace_ext4_da_write_end(inode, pos, len, copied);
2554 start = pos & (PAGE_CACHE_SIZE - 1);
2555 end = start + copied - 1;
2558 * generic_write_end() will run mark_inode_dirty() if i_size
2559 * changes. So let's piggyback the i_disksize mark_inode_dirty
2563 new_i_size = pos + copied;
2564 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2565 if (ext4_da_should_update_i_disksize(page, end)) {
2566 down_write(&EXT4_I(inode)->i_data_sem);
2567 if (new_i_size > EXT4_I(inode)->i_disksize) {
2569 * Updating i_disksize when extending file
2570 * without needing block allocation
2572 if (ext4_should_order_data(inode))
2573 ret = ext4_jbd2_file_inode(handle,
2576 EXT4_I(inode)->i_disksize = new_i_size;
2578 up_write(&EXT4_I(inode)->i_data_sem);
2579 /* We need to mark inode dirty even if
2580 * new_i_size is less that inode->i_size
2581 * bu greater than i_disksize.(hint delalloc)
2583 ext4_mark_inode_dirty(handle, inode);
2586 ret2 = generic_write_end(file, mapping, pos, len, copied,
2591 ret2 = ext4_journal_stop(handle);
2595 return ret ? ret : copied;
2598 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2601 * Drop reserved blocks
2603 BUG_ON(!PageLocked(page));
2604 if (!page_has_buffers(page))
2607 ext4_da_page_release_reservation(page, offset);
2610 ext4_invalidatepage(page, offset);
2616 * Force all delayed allocation blocks to be allocated for a given inode.
2618 int ext4_alloc_da_blocks(struct inode *inode)
2620 trace_ext4_alloc_da_blocks(inode);
2622 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2623 !EXT4_I(inode)->i_reserved_meta_blocks)
2627 * We do something simple for now. The filemap_flush() will
2628 * also start triggering a write of the data blocks, which is
2629 * not strictly speaking necessary (and for users of
2630 * laptop_mode, not even desirable). However, to do otherwise
2631 * would require replicating code paths in:
2633 * ext4_da_writepages() ->
2634 * write_cache_pages() ---> (via passed in callback function)
2635 * __mpage_da_writepage() -->
2636 * mpage_add_bh_to_extent()
2637 * mpage_da_map_blocks()
2639 * The problem is that write_cache_pages(), located in
2640 * mm/page-writeback.c, marks pages clean in preparation for
2641 * doing I/O, which is not desirable if we're not planning on
2644 * We could call write_cache_pages(), and then redirty all of
2645 * the pages by calling redirty_page_for_writepage() but that
2646 * would be ugly in the extreme. So instead we would need to
2647 * replicate parts of the code in the above functions,
2648 * simplifying them because we wouldn't actually intend to
2649 * write out the pages, but rather only collect contiguous
2650 * logical block extents, call the multi-block allocator, and
2651 * then update the buffer heads with the block allocations.
2653 * For now, though, we'll cheat by calling filemap_flush(),
2654 * which will map the blocks, and start the I/O, but not
2655 * actually wait for the I/O to complete.
2657 return filemap_flush(inode->i_mapping);
2661 * bmap() is special. It gets used by applications such as lilo and by
2662 * the swapper to find the on-disk block of a specific piece of data.
2664 * Naturally, this is dangerous if the block concerned is still in the
2665 * journal. If somebody makes a swapfile on an ext4 data-journaling
2666 * filesystem and enables swap, then they may get a nasty shock when the
2667 * data getting swapped to that swapfile suddenly gets overwritten by
2668 * the original zero's written out previously to the journal and
2669 * awaiting writeback in the kernel's buffer cache.
2671 * So, if we see any bmap calls here on a modified, data-journaled file,
2672 * take extra steps to flush any blocks which might be in the cache.
2674 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2676 struct inode *inode = mapping->host;
2680 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2681 test_opt(inode->i_sb, DELALLOC)) {
2683 * With delalloc we want to sync the file
2684 * so that we can make sure we allocate
2687 filemap_write_and_wait(mapping);
2690 if (EXT4_JOURNAL(inode) &&
2691 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2693 * This is a REALLY heavyweight approach, but the use of
2694 * bmap on dirty files is expected to be extremely rare:
2695 * only if we run lilo or swapon on a freshly made file
2696 * do we expect this to happen.
2698 * (bmap requires CAP_SYS_RAWIO so this does not
2699 * represent an unprivileged user DOS attack --- we'd be
2700 * in trouble if mortal users could trigger this path at
2703 * NB. EXT4_STATE_JDATA is not set on files other than
2704 * regular files. If somebody wants to bmap a directory
2705 * or symlink and gets confused because the buffer
2706 * hasn't yet been flushed to disk, they deserve
2707 * everything they get.
2710 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2711 journal = EXT4_JOURNAL(inode);
2712 jbd2_journal_lock_updates(journal);
2713 err = jbd2_journal_flush(journal);
2714 jbd2_journal_unlock_updates(journal);
2720 return generic_block_bmap(mapping, block, ext4_get_block);
2723 static int ext4_readpage(struct file *file, struct page *page)
2725 trace_ext4_readpage(page);
2726 return mpage_readpage(page, ext4_get_block);
2730 ext4_readpages(struct file *file, struct address_space *mapping,
2731 struct list_head *pages, unsigned nr_pages)
2733 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2736 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
2738 struct buffer_head *head, *bh;
2739 unsigned int curr_off = 0;
2741 if (!page_has_buffers(page))
2743 head = bh = page_buffers(page);
2745 if (offset <= curr_off && test_clear_buffer_uninit(bh)
2747 ext4_free_io_end(bh->b_private);
2748 bh->b_private = NULL;
2749 bh->b_end_io = NULL;
2751 curr_off = curr_off + bh->b_size;
2752 bh = bh->b_this_page;
2753 } while (bh != head);
2756 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2758 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2760 trace_ext4_invalidatepage(page, offset);
2763 * free any io_end structure allocated for buffers to be discarded
2765 if (ext4_should_dioread_nolock(page->mapping->host))
2766 ext4_invalidatepage_free_endio(page, offset);
2768 * If it's a full truncate we just forget about the pending dirtying
2771 ClearPageChecked(page);
2774 jbd2_journal_invalidatepage(journal, page, offset);
2776 block_invalidatepage(page, offset);
2779 static int ext4_releasepage(struct page *page, gfp_t wait)
2781 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2783 trace_ext4_releasepage(page);
2785 WARN_ON(PageChecked(page));
2786 if (!page_has_buffers(page))
2789 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2791 return try_to_free_buffers(page);
2795 * ext4_get_block used when preparing for a DIO write or buffer write.
2796 * We allocate an uinitialized extent if blocks haven't been allocated.
2797 * The extent will be converted to initialized after the IO is complete.
2799 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
2800 struct buffer_head *bh_result, int create)
2802 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2803 inode->i_ino, create);
2804 return _ext4_get_block(inode, iblock, bh_result,
2805 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2808 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2809 ssize_t size, void *private, int ret,
2812 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2813 ext4_io_end_t *io_end = iocb->private;
2814 struct workqueue_struct *wq;
2815 unsigned long flags;
2816 struct ext4_inode_info *ei;
2818 /* if not async direct IO or dio with 0 bytes write, just return */
2819 if (!io_end || !size)
2822 ext_debug("ext4_end_io_dio(): io_end 0x%p"
2823 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
2824 iocb->private, io_end->inode->i_ino, iocb, offset,
2827 iocb->private = NULL;
2829 /* if not aio dio with unwritten extents, just free io and return */
2830 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2831 ext4_free_io_end(io_end);
2833 inode_dio_done(inode);
2835 aio_complete(iocb, ret, 0);
2839 io_end->offset = offset;
2840 io_end->size = size;
2842 io_end->iocb = iocb;
2843 io_end->result = ret;
2845 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
2847 /* Add the io_end to per-inode completed aio dio list*/
2848 ei = EXT4_I(io_end->inode);
2849 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
2850 list_add_tail(&io_end->list, &ei->i_completed_io_list);
2851 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
2853 /* queue the work to convert unwritten extents to written */
2854 queue_work(wq, &io_end->work);
2857 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
2859 ext4_io_end_t *io_end = bh->b_private;
2860 struct workqueue_struct *wq;
2861 struct inode *inode;
2862 unsigned long flags;
2864 if (!test_clear_buffer_uninit(bh) || !io_end)
2867 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2868 printk("sb umounted, discard end_io request for inode %lu\n",
2869 io_end->inode->i_ino);
2870 ext4_free_io_end(io_end);
2875 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2876 * but being more careful is always safe for the future change.
2878 inode = io_end->inode;
2879 ext4_set_io_unwritten_flag(inode, io_end);
2881 /* Add the io_end to per-inode completed io list*/
2882 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
2883 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
2884 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
2886 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
2887 /* queue the work to convert unwritten extents to written */
2888 queue_work(wq, &io_end->work);
2890 bh->b_private = NULL;
2891 bh->b_end_io = NULL;
2892 clear_buffer_uninit(bh);
2893 end_buffer_async_write(bh, uptodate);
2896 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2898 ext4_io_end_t *io_end;
2899 struct page *page = bh->b_page;
2900 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2901 size_t size = bh->b_size;
2904 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2906 pr_warn_ratelimited("%s: allocation fail\n", __func__);
2910 io_end->offset = offset;
2911 io_end->size = size;
2913 * We need to hold a reference to the page to make sure it
2914 * doesn't get evicted before ext4_end_io_work() has a chance
2915 * to convert the extent from written to unwritten.
2917 io_end->page = page;
2918 get_page(io_end->page);
2920 bh->b_private = io_end;
2921 bh->b_end_io = ext4_end_io_buffer_write;
2926 * For ext4 extent files, ext4 will do direct-io write to holes,
2927 * preallocated extents, and those write extend the file, no need to
2928 * fall back to buffered IO.
2930 * For holes, we fallocate those blocks, mark them as uninitialized
2931 * If those blocks were preallocated, we mark sure they are splited, but
2932 * still keep the range to write as uninitialized.
2934 * The unwrritten extents will be converted to written when DIO is completed.
2935 * For async direct IO, since the IO may still pending when return, we
2936 * set up an end_io call back function, which will do the conversion
2937 * when async direct IO completed.
2939 * If the O_DIRECT write will extend the file then add this inode to the
2940 * orphan list. So recovery will truncate it back to the original size
2941 * if the machine crashes during the write.
2944 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2945 const struct iovec *iov, loff_t offset,
2946 unsigned long nr_segs)
2948 struct file *file = iocb->ki_filp;
2949 struct inode *inode = file->f_mapping->host;
2951 size_t count = iov_length(iov, nr_segs);
2953 loff_t final_size = offset + count;
2954 if (rw == WRITE && final_size <= inode->i_size) {
2956 * We could direct write to holes and fallocate.
2958 * Allocated blocks to fill the hole are marked as uninitialized
2959 * to prevent parallel buffered read to expose the stale data
2960 * before DIO complete the data IO.
2962 * As to previously fallocated extents, ext4 get_block
2963 * will just simply mark the buffer mapped but still
2964 * keep the extents uninitialized.
2966 * for non AIO case, we will convert those unwritten extents
2967 * to written after return back from blockdev_direct_IO.
2969 * for async DIO, the conversion needs to be defered when
2970 * the IO is completed. The ext4 end_io callback function
2971 * will be called to take care of the conversion work.
2972 * Here for async case, we allocate an io_end structure to
2975 iocb->private = NULL;
2976 EXT4_I(inode)->cur_aio_dio = NULL;
2977 if (!is_sync_kiocb(iocb)) {
2978 ext4_io_end_t *io_end =
2979 ext4_init_io_end(inode, GFP_NOFS);
2982 io_end->flag |= EXT4_IO_END_DIRECT;
2983 iocb->private = io_end;
2985 * we save the io structure for current async
2986 * direct IO, so that later ext4_map_blocks()
2987 * could flag the io structure whether there
2988 * is a unwritten extents needs to be converted
2989 * when IO is completed.
2991 EXT4_I(inode)->cur_aio_dio = iocb->private;
2994 ret = __blockdev_direct_IO(rw, iocb, inode,
2995 inode->i_sb->s_bdev, iov,
2997 ext4_get_block_write,
3000 DIO_LOCKING | DIO_SKIP_HOLES);
3002 EXT4_I(inode)->cur_aio_dio = NULL;
3004 * The io_end structure takes a reference to the inode,
3005 * that structure needs to be destroyed and the
3006 * reference to the inode need to be dropped, when IO is
3007 * complete, even with 0 byte write, or failed.
3009 * In the successful AIO DIO case, the io_end structure will be
3010 * desctroyed and the reference to the inode will be dropped
3011 * after the end_io call back function is called.
3013 * In the case there is 0 byte write, or error case, since
3014 * VFS direct IO won't invoke the end_io call back function,
3015 * we need to free the end_io structure here.
3017 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3018 ext4_free_io_end(iocb->private);
3019 iocb->private = NULL;
3020 } else if (ret > 0 && ext4_test_inode_state(inode,
3021 EXT4_STATE_DIO_UNWRITTEN)) {
3024 * for non AIO case, since the IO is already
3025 * completed, we could do the conversion right here
3027 err = ext4_convert_unwritten_extents(inode,
3031 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3036 /* for write the the end of file case, we fall back to old way */
3037 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3040 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3041 const struct iovec *iov, loff_t offset,
3042 unsigned long nr_segs)
3044 struct file *file = iocb->ki_filp;
3045 struct inode *inode = file->f_mapping->host;
3049 * If we are doing data journalling we don't support O_DIRECT
3051 if (ext4_should_journal_data(inode))
3054 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3055 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3056 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3058 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3059 trace_ext4_direct_IO_exit(inode, offset,
3060 iov_length(iov, nr_segs), rw, ret);
3065 * Pages can be marked dirty completely asynchronously from ext4's journalling
3066 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3067 * much here because ->set_page_dirty is called under VFS locks. The page is
3068 * not necessarily locked.
3070 * We cannot just dirty the page and leave attached buffers clean, because the
3071 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3072 * or jbddirty because all the journalling code will explode.
3074 * So what we do is to mark the page "pending dirty" and next time writepage
3075 * is called, propagate that into the buffers appropriately.
3077 static int ext4_journalled_set_page_dirty(struct page *page)
3079 SetPageChecked(page);
3080 return __set_page_dirty_nobuffers(page);
3083 static const struct address_space_operations ext4_ordered_aops = {
3084 .readpage = ext4_readpage,
3085 .readpages = ext4_readpages,
3086 .writepage = ext4_writepage,
3087 .write_begin = ext4_write_begin,
3088 .write_end = ext4_ordered_write_end,
3090 .invalidatepage = ext4_invalidatepage,
3091 .releasepage = ext4_releasepage,
3092 .direct_IO = ext4_direct_IO,
3093 .migratepage = buffer_migrate_page,
3094 .is_partially_uptodate = block_is_partially_uptodate,
3095 .error_remove_page = generic_error_remove_page,
3098 static const struct address_space_operations ext4_writeback_aops = {
3099 .readpage = ext4_readpage,
3100 .readpages = ext4_readpages,
3101 .writepage = ext4_writepage,
3102 .write_begin = ext4_write_begin,
3103 .write_end = ext4_writeback_write_end,
3105 .invalidatepage = ext4_invalidatepage,
3106 .releasepage = ext4_releasepage,
3107 .direct_IO = ext4_direct_IO,
3108 .migratepage = buffer_migrate_page,
3109 .is_partially_uptodate = block_is_partially_uptodate,
3110 .error_remove_page = generic_error_remove_page,
3113 static const struct address_space_operations ext4_journalled_aops = {
3114 .readpage = ext4_readpage,
3115 .readpages = ext4_readpages,
3116 .writepage = ext4_writepage,
3117 .write_begin = ext4_write_begin,
3118 .write_end = ext4_journalled_write_end,
3119 .set_page_dirty = ext4_journalled_set_page_dirty,
3121 .invalidatepage = ext4_invalidatepage,
3122 .releasepage = ext4_releasepage,
3123 .direct_IO = ext4_direct_IO,
3124 .is_partially_uptodate = block_is_partially_uptodate,
3125 .error_remove_page = generic_error_remove_page,
3128 static const struct address_space_operations ext4_da_aops = {
3129 .readpage = ext4_readpage,
3130 .readpages = ext4_readpages,
3131 .writepage = ext4_writepage,
3132 .writepages = ext4_da_writepages,
3133 .write_begin = ext4_da_write_begin,
3134 .write_end = ext4_da_write_end,
3136 .invalidatepage = ext4_da_invalidatepage,
3137 .releasepage = ext4_releasepage,
3138 .direct_IO = ext4_direct_IO,
3139 .migratepage = buffer_migrate_page,
3140 .is_partially_uptodate = block_is_partially_uptodate,
3141 .error_remove_page = generic_error_remove_page,
3144 void ext4_set_aops(struct inode *inode)
3146 switch (ext4_inode_journal_mode(inode)) {
3147 case EXT4_INODE_ORDERED_DATA_MODE:
3148 if (test_opt(inode->i_sb, DELALLOC))
3149 inode->i_mapping->a_ops = &ext4_da_aops;
3151 inode->i_mapping->a_ops = &ext4_ordered_aops;
3153 case EXT4_INODE_WRITEBACK_DATA_MODE:
3154 if (test_opt(inode->i_sb, DELALLOC))
3155 inode->i_mapping->a_ops = &ext4_da_aops;
3157 inode->i_mapping->a_ops = &ext4_writeback_aops;
3159 case EXT4_INODE_JOURNAL_DATA_MODE:
3160 inode->i_mapping->a_ops = &ext4_journalled_aops;
3169 * ext4_discard_partial_page_buffers()
3170 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3171 * This function finds and locks the page containing the offset
3172 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3173 * Calling functions that already have the page locked should call
3174 * ext4_discard_partial_page_buffers_no_lock directly.
3176 int ext4_discard_partial_page_buffers(handle_t *handle,
3177 struct address_space *mapping, loff_t from,
3178 loff_t length, int flags)
3180 struct inode *inode = mapping->host;
3184 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3185 mapping_gfp_mask(mapping) & ~__GFP_FS);
3189 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3190 from, length, flags);
3193 page_cache_release(page);
3198 * ext4_discard_partial_page_buffers_no_lock()
3199 * Zeros a page range of length 'length' starting from offset 'from'.
3200 * Buffer heads that correspond to the block aligned regions of the
3201 * zeroed range will be unmapped. Unblock aligned regions
3202 * will have the corresponding buffer head mapped if needed so that
3203 * that region of the page can be updated with the partial zero out.
3205 * This function assumes that the page has already been locked. The
3206 * The range to be discarded must be contained with in the given page.
3207 * If the specified range exceeds the end of the page it will be shortened
3208 * to the end of the page that corresponds to 'from'. This function is
3209 * appropriate for updating a page and it buffer heads to be unmapped and
3210 * zeroed for blocks that have been either released, or are going to be
3213 * handle: The journal handle
3214 * inode: The files inode
3215 * page: A locked page that contains the offset "from"
3216 * from: The starting byte offset (from the begining of the file)
3217 * to begin discarding
3218 * len: The length of bytes to discard
3219 * flags: Optional flags that may be used:
3221 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3222 * Only zero the regions of the page whose buffer heads
3223 * have already been unmapped. This flag is appropriate
3224 * for updateing the contents of a page whose blocks may
3225 * have already been released, and we only want to zero
3226 * out the regions that correspond to those released blocks.
3228 * Returns zero on sucess or negative on failure.
3230 int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3231 struct inode *inode, struct page *page, loff_t from,
3232 loff_t length, int flags)
3234 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3235 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3236 unsigned int blocksize, max, pos;
3238 struct buffer_head *bh;
3241 blocksize = inode->i_sb->s_blocksize;
3242 max = PAGE_CACHE_SIZE - offset;
3244 if (index != page->index)
3248 * correct length if it does not fall between
3249 * 'from' and the end of the page
3251 if (length > max || length < 0)
3254 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3256 if (!page_has_buffers(page))
3257 create_empty_buffers(page, blocksize, 0);
3259 /* Find the buffer that contains "offset" */
3260 bh = page_buffers(page);
3262 while (offset >= pos) {
3263 bh = bh->b_this_page;
3269 while (pos < offset + length) {
3270 unsigned int end_of_block, range_to_discard;
3274 /* The length of space left to zero and unmap */
3275 range_to_discard = offset + length - pos;
3277 /* The length of space until the end of the block */
3278 end_of_block = blocksize - (pos & (blocksize-1));
3281 * Do not unmap or zero past end of block
3282 * for this buffer head
3284 if (range_to_discard > end_of_block)
3285 range_to_discard = end_of_block;
3289 * Skip this buffer head if we are only zeroing unampped
3290 * regions of the page
3292 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3296 /* If the range is block aligned, unmap */
3297 if (range_to_discard == blocksize) {
3298 clear_buffer_dirty(bh);
3300 clear_buffer_mapped(bh);
3301 clear_buffer_req(bh);
3302 clear_buffer_new(bh);
3303 clear_buffer_delay(bh);
3304 clear_buffer_unwritten(bh);
3305 clear_buffer_uptodate(bh);
3306 zero_user(page, pos, range_to_discard);
3307 BUFFER_TRACE(bh, "Buffer discarded");
3312 * If this block is not completely contained in the range
3313 * to be discarded, then it is not going to be released. Because
3314 * we need to keep this block, we need to make sure this part
3315 * of the page is uptodate before we modify it by writeing
3316 * partial zeros on it.
3318 if (!buffer_mapped(bh)) {
3320 * Buffer head must be mapped before we can read
3323 BUFFER_TRACE(bh, "unmapped");
3324 ext4_get_block(inode, iblock, bh, 0);
3325 /* unmapped? It's a hole - nothing to do */
3326 if (!buffer_mapped(bh)) {
3327 BUFFER_TRACE(bh, "still unmapped");
3332 /* Ok, it's mapped. Make sure it's up-to-date */
3333 if (PageUptodate(page))
3334 set_buffer_uptodate(bh);
3336 if (!buffer_uptodate(bh)) {
3338 ll_rw_block(READ, 1, &bh);
3340 /* Uhhuh. Read error. Complain and punt.*/
3341 if (!buffer_uptodate(bh))
3345 if (ext4_should_journal_data(inode)) {
3346 BUFFER_TRACE(bh, "get write access");
3347 err = ext4_journal_get_write_access(handle, bh);
3352 zero_user(page, pos, range_to_discard);
3355 if (ext4_should_journal_data(inode)) {
3356 err = ext4_handle_dirty_metadata(handle, inode, bh);
3358 mark_buffer_dirty(bh);
3360 BUFFER_TRACE(bh, "Partial buffer zeroed");
3362 bh = bh->b_this_page;
3364 pos += range_to_discard;
3371 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3372 * up to the end of the block which corresponds to `from'.
3373 * This required during truncate. We need to physically zero the tail end
3374 * of that block so it doesn't yield old data if the file is later grown.
3376 int ext4_block_truncate_page(handle_t *handle,
3377 struct address_space *mapping, loff_t from)
3379 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3382 struct inode *inode = mapping->host;
3384 blocksize = inode->i_sb->s_blocksize;
3385 length = blocksize - (offset & (blocksize - 1));
3387 return ext4_block_zero_page_range(handle, mapping, from, length);
3391 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3392 * starting from file offset 'from'. The range to be zero'd must
3393 * be contained with in one block. If the specified range exceeds
3394 * the end of the block it will be shortened to end of the block
3395 * that cooresponds to 'from'
3397 int ext4_block_zero_page_range(handle_t *handle,
3398 struct address_space *mapping, loff_t from, loff_t length)
3400 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3401 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3402 unsigned blocksize, max, pos;
3404 struct inode *inode = mapping->host;
3405 struct buffer_head *bh;
3409 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3410 mapping_gfp_mask(mapping) & ~__GFP_FS);
3414 blocksize = inode->i_sb->s_blocksize;
3415 max = blocksize - (offset & (blocksize - 1));
3418 * correct length if it does not fall between
3419 * 'from' and the end of the block
3421 if (length > max || length < 0)
3424 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3426 if (!page_has_buffers(page))
3427 create_empty_buffers(page, blocksize, 0);
3429 /* Find the buffer that contains "offset" */
3430 bh = page_buffers(page);
3432 while (offset >= pos) {
3433 bh = bh->b_this_page;
3439 if (buffer_freed(bh)) {
3440 BUFFER_TRACE(bh, "freed: skip");
3444 if (!buffer_mapped(bh)) {
3445 BUFFER_TRACE(bh, "unmapped");
3446 ext4_get_block(inode, iblock, bh, 0);
3447 /* unmapped? It's a hole - nothing to do */
3448 if (!buffer_mapped(bh)) {
3449 BUFFER_TRACE(bh, "still unmapped");
3454 /* Ok, it's mapped. Make sure it's up-to-date */
3455 if (PageUptodate(page))
3456 set_buffer_uptodate(bh);
3458 if (!buffer_uptodate(bh)) {
3460 ll_rw_block(READ, 1, &bh);
3462 /* Uhhuh. Read error. Complain and punt. */
3463 if (!buffer_uptodate(bh))
3467 if (ext4_should_journal_data(inode)) {
3468 BUFFER_TRACE(bh, "get write access");
3469 err = ext4_journal_get_write_access(handle, bh);
3474 zero_user(page, offset, length);
3476 BUFFER_TRACE(bh, "zeroed end of block");
3479 if (ext4_should_journal_data(inode)) {
3480 err = ext4_handle_dirty_metadata(handle, inode, bh);
3482 mark_buffer_dirty(bh);
3486 page_cache_release(page);
3490 int ext4_can_truncate(struct inode *inode)
3492 if (S_ISREG(inode->i_mode))
3494 if (S_ISDIR(inode->i_mode))
3496 if (S_ISLNK(inode->i_mode))
3497 return !ext4_inode_is_fast_symlink(inode);
3502 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3503 * associated with the given offset and length
3505 * @inode: File inode
3506 * @offset: The offset where the hole will begin
3507 * @len: The length of the hole
3509 * Returns: 0 on sucess or negative on failure
3512 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3514 struct inode *inode = file->f_path.dentry->d_inode;
3515 if (!S_ISREG(inode->i_mode))
3518 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3519 /* TODO: Add support for non extent hole punching */
3523 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3524 /* TODO: Add support for bigalloc file systems */
3528 return ext4_ext_punch_hole(file, offset, length);
3534 * We block out ext4_get_block() block instantiations across the entire
3535 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3536 * simultaneously on behalf of the same inode.
3538 * As we work through the truncate and commmit bits of it to the journal there
3539 * is one core, guiding principle: the file's tree must always be consistent on
3540 * disk. We must be able to restart the truncate after a crash.
3542 * The file's tree may be transiently inconsistent in memory (although it
3543 * probably isn't), but whenever we close off and commit a journal transaction,
3544 * the contents of (the filesystem + the journal) must be consistent and
3545 * restartable. It's pretty simple, really: bottom up, right to left (although
3546 * left-to-right works OK too).
3548 * Note that at recovery time, journal replay occurs *before* the restart of
3549 * truncate against the orphan inode list.
3551 * The committed inode has the new, desired i_size (which is the same as
3552 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3553 * that this inode's truncate did not complete and it will again call
3554 * ext4_truncate() to have another go. So there will be instantiated blocks
3555 * to the right of the truncation point in a crashed ext4 filesystem. But
3556 * that's fine - as long as they are linked from the inode, the post-crash
3557 * ext4_truncate() run will find them and release them.
3559 void ext4_truncate(struct inode *inode)
3561 trace_ext4_truncate_enter(inode);
3563 if (!ext4_can_truncate(inode))
3566 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3568 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3569 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3571 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3572 ext4_ext_truncate(inode);
3574 ext4_ind_truncate(inode);
3576 trace_ext4_truncate_exit(inode);
3580 * ext4_get_inode_loc returns with an extra refcount against the inode's
3581 * underlying buffer_head on success. If 'in_mem' is true, we have all
3582 * data in memory that is needed to recreate the on-disk version of this
3585 static int __ext4_get_inode_loc(struct inode *inode,
3586 struct ext4_iloc *iloc, int in_mem)
3588 struct ext4_group_desc *gdp;
3589 struct buffer_head *bh;
3590 struct super_block *sb = inode->i_sb;
3592 int inodes_per_block, inode_offset;
3595 if (!ext4_valid_inum(sb, inode->i_ino))
3598 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3599 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3604 * Figure out the offset within the block group inode table
3606 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3607 inode_offset = ((inode->i_ino - 1) %
3608 EXT4_INODES_PER_GROUP(sb));
3609 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3610 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3612 bh = sb_getblk(sb, block);
3615 if (!buffer_uptodate(bh)) {
3619 * If the buffer has the write error flag, we have failed
3620 * to write out another inode in the same block. In this
3621 * case, we don't have to read the block because we may
3622 * read the old inode data successfully.
3624 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3625 set_buffer_uptodate(bh);
3627 if (buffer_uptodate(bh)) {
3628 /* someone brought it uptodate while we waited */
3634 * If we have all information of the inode in memory and this
3635 * is the only valid inode in the block, we need not read the
3639 struct buffer_head *bitmap_bh;
3642 start = inode_offset & ~(inodes_per_block - 1);
3644 /* Is the inode bitmap in cache? */
3645 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3650 * If the inode bitmap isn't in cache then the
3651 * optimisation may end up performing two reads instead
3652 * of one, so skip it.
3654 if (!buffer_uptodate(bitmap_bh)) {
3658 for (i = start; i < start + inodes_per_block; i++) {
3659 if (i == inode_offset)
3661 if (ext4_test_bit(i, bitmap_bh->b_data))
3665 if (i == start + inodes_per_block) {
3666 /* all other inodes are free, so skip I/O */
3667 memset(bh->b_data, 0, bh->b_size);
3668 set_buffer_uptodate(bh);
3676 * If we need to do any I/O, try to pre-readahead extra
3677 * blocks from the inode table.
3679 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3680 ext4_fsblk_t b, end, table;
3683 table = ext4_inode_table(sb, gdp);
3684 /* s_inode_readahead_blks is always a power of 2 */
3685 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3688 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3689 num = EXT4_INODES_PER_GROUP(sb);
3690 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3691 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3692 num -= ext4_itable_unused_count(sb, gdp);
3693 table += num / inodes_per_block;
3697 sb_breadahead(sb, b++);
3701 * There are other valid inodes in the buffer, this inode
3702 * has in-inode xattrs, or we don't have this inode in memory.
3703 * Read the block from disk.
3705 trace_ext4_load_inode(inode);
3707 bh->b_end_io = end_buffer_read_sync;
3708 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3710 if (!buffer_uptodate(bh)) {
3711 EXT4_ERROR_INODE_BLOCK(inode, block,
3712 "unable to read itable block");
3722 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3724 /* We have all inode data except xattrs in memory here. */
3725 return __ext4_get_inode_loc(inode, iloc,
3726 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3729 void ext4_set_inode_flags(struct inode *inode)
3731 unsigned int flags = EXT4_I(inode)->i_flags;
3732 unsigned int new_fl = 0;
3734 if (flags & EXT4_SYNC_FL)
3736 if (flags & EXT4_APPEND_FL)
3738 if (flags & EXT4_IMMUTABLE_FL)
3739 new_fl |= S_IMMUTABLE;
3740 if (flags & EXT4_NOATIME_FL)
3741 new_fl |= S_NOATIME;
3742 if (flags & EXT4_DIRSYNC_FL)
3743 new_fl |= S_DIRSYNC;
3744 set_mask_bits(&inode->i_flags,
3745 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC, new_fl);
3748 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3749 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3751 unsigned int vfs_fl;
3752 unsigned long old_fl, new_fl;
3755 vfs_fl = ei->vfs_inode.i_flags;
3756 old_fl = ei->i_flags;
3757 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3758 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3760 if (vfs_fl & S_SYNC)
3761 new_fl |= EXT4_SYNC_FL;
3762 if (vfs_fl & S_APPEND)
3763 new_fl |= EXT4_APPEND_FL;
3764 if (vfs_fl & S_IMMUTABLE)
3765 new_fl |= EXT4_IMMUTABLE_FL;
3766 if (vfs_fl & S_NOATIME)
3767 new_fl |= EXT4_NOATIME_FL;
3768 if (vfs_fl & S_DIRSYNC)
3769 new_fl |= EXT4_DIRSYNC_FL;
3770 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3773 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3774 struct ext4_inode_info *ei)
3777 struct inode *inode = &(ei->vfs_inode);
3778 struct super_block *sb = inode->i_sb;
3780 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3781 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3782 /* we are using combined 48 bit field */
3783 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3784 le32_to_cpu(raw_inode->i_blocks_lo);
3785 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3786 /* i_blocks represent file system block size */
3787 return i_blocks << (inode->i_blkbits - 9);
3792 return le32_to_cpu(raw_inode->i_blocks_lo);
3796 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3798 struct ext4_iloc iloc;
3799 struct ext4_inode *raw_inode;
3800 struct ext4_inode_info *ei;
3801 struct inode *inode;
3802 journal_t *journal = EXT4_SB(sb)->s_journal;
3806 inode = iget_locked(sb, ino);
3808 return ERR_PTR(-ENOMEM);
3809 if (!(inode->i_state & I_NEW))
3815 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3818 raw_inode = ext4_raw_inode(&iloc);
3819 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3820 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3821 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3822 if (!(test_opt(inode->i_sb, NO_UID32))) {
3823 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3824 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3826 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3828 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3829 ei->i_dir_start_lookup = 0;
3830 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3831 /* We now have enough fields to check if the inode was active or not.
3832 * This is needed because nfsd might try to access dead inodes
3833 * the test is that same one that e2fsck uses
3834 * NeilBrown 1999oct15
3836 if (inode->i_nlink == 0) {
3837 if (inode->i_mode == 0 ||
3838 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3839 /* this inode is deleted */
3843 /* The only unlinked inodes we let through here have
3844 * valid i_mode and are being read by the orphan
3845 * recovery code: that's fine, we're about to complete
3846 * the process of deleting those. */
3848 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3849 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3850 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3851 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3853 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3854 inode->i_size = ext4_isize(raw_inode);
3855 ei->i_disksize = inode->i_size;
3857 ei->i_reserved_quota = 0;
3859 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3860 ei->i_block_group = iloc.block_group;
3861 ei->i_last_alloc_group = ~0;
3863 * NOTE! The in-memory inode i_data array is in little-endian order
3864 * even on big-endian machines: we do NOT byteswap the block numbers!
3866 for (block = 0; block < EXT4_N_BLOCKS; block++)
3867 ei->i_data[block] = raw_inode->i_block[block];
3868 INIT_LIST_HEAD(&ei->i_orphan);
3871 * Set transaction id's of transactions that have to be committed
3872 * to finish f[data]sync. We set them to currently running transaction
3873 * as we cannot be sure that the inode or some of its metadata isn't
3874 * part of the transaction - the inode could have been reclaimed and
3875 * now it is reread from disk.
3878 transaction_t *transaction;
3881 read_lock(&journal->j_state_lock);
3882 if (journal->j_running_transaction)
3883 transaction = journal->j_running_transaction;
3885 transaction = journal->j_committing_transaction;
3887 tid = transaction->t_tid;
3889 tid = journal->j_commit_sequence;
3890 read_unlock(&journal->j_state_lock);
3891 ei->i_sync_tid = tid;
3892 ei->i_datasync_tid = tid;
3895 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3896 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3897 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3898 EXT4_INODE_SIZE(inode->i_sb)) {
3902 if (ei->i_extra_isize == 0) {
3903 /* The extra space is currently unused. Use it. */
3904 ei->i_extra_isize = sizeof(struct ext4_inode) -
3905 EXT4_GOOD_OLD_INODE_SIZE;
3907 __le32 *magic = (void *)raw_inode +
3908 EXT4_GOOD_OLD_INODE_SIZE +
3910 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3911 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3914 ei->i_extra_isize = 0;
3916 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3917 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3918 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3919 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3921 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3922 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3923 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3925 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3929 if (ei->i_file_acl &&
3930 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3931 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
3935 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3936 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3937 (S_ISLNK(inode->i_mode) &&
3938 !ext4_inode_is_fast_symlink(inode)))
3939 /* Validate extent which is part of inode */
3940 ret = ext4_ext_check_inode(inode);
3941 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3942 (S_ISLNK(inode->i_mode) &&
3943 !ext4_inode_is_fast_symlink(inode))) {
3944 /* Validate block references which are part of inode */
3945 ret = ext4_ind_check_inode(inode);
3950 if (S_ISREG(inode->i_mode)) {
3951 inode->i_op = &ext4_file_inode_operations;
3952 inode->i_fop = &ext4_file_operations;
3953 ext4_set_aops(inode);
3954 } else if (S_ISDIR(inode->i_mode)) {
3955 inode->i_op = &ext4_dir_inode_operations;
3956 inode->i_fop = &ext4_dir_operations;
3957 } else if (S_ISLNK(inode->i_mode)) {
3958 if (ext4_inode_is_fast_symlink(inode)) {
3959 inode->i_op = &ext4_fast_symlink_inode_operations;
3960 nd_terminate_link(ei->i_data, inode->i_size,
3961 sizeof(ei->i_data) - 1);
3963 inode->i_op = &ext4_symlink_inode_operations;
3964 ext4_set_aops(inode);
3966 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3967 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3968 inode->i_op = &ext4_special_inode_operations;
3969 if (raw_inode->i_block[0])
3970 init_special_inode(inode, inode->i_mode,
3971 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3973 init_special_inode(inode, inode->i_mode,
3974 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3977 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
3981 ext4_set_inode_flags(inode);
3982 unlock_new_inode(inode);
3988 return ERR_PTR(ret);
3991 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
3993 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
3994 return ERR_PTR(-EIO);
3995 return ext4_iget(sb, ino);
3998 static int ext4_inode_blocks_set(handle_t *handle,
3999 struct ext4_inode *raw_inode,
4000 struct ext4_inode_info *ei)
4002 struct inode *inode = &(ei->vfs_inode);
4003 u64 i_blocks = inode->i_blocks;
4004 struct super_block *sb = inode->i_sb;
4006 if (i_blocks <= ~0U) {
4008 * i_blocks can be represnted in a 32 bit variable
4009 * as multiple of 512 bytes
4011 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4012 raw_inode->i_blocks_high = 0;
4013 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4016 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4019 if (i_blocks <= 0xffffffffffffULL) {
4021 * i_blocks can be represented in a 48 bit variable
4022 * as multiple of 512 bytes
4024 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4025 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4026 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4028 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4029 /* i_block is stored in file system block size */
4030 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4031 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4032 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4038 * Post the struct inode info into an on-disk inode location in the
4039 * buffer-cache. This gobbles the caller's reference to the
4040 * buffer_head in the inode location struct.
4042 * The caller must have write access to iloc->bh.
4044 static int ext4_do_update_inode(handle_t *handle,
4045 struct inode *inode,
4046 struct ext4_iloc *iloc)
4048 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4049 struct ext4_inode_info *ei = EXT4_I(inode);
4050 struct buffer_head *bh = iloc->bh;
4051 int err = 0, rc, block;
4052 int need_datasync = 0;
4054 /* For fields not not tracking in the in-memory inode,
4055 * initialise them to zero for new inodes. */
4056 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4057 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4059 ext4_get_inode_flags(ei);
4060 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4061 if (!(test_opt(inode->i_sb, NO_UID32))) {
4062 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4063 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4065 * Fix up interoperability with old kernels. Otherwise, old inodes get
4066 * re-used with the upper 16 bits of the uid/gid intact
4069 raw_inode->i_uid_high =
4070 cpu_to_le16(high_16_bits(inode->i_uid));
4071 raw_inode->i_gid_high =
4072 cpu_to_le16(high_16_bits(inode->i_gid));
4074 raw_inode->i_uid_high = 0;
4075 raw_inode->i_gid_high = 0;
4078 raw_inode->i_uid_low =
4079 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4080 raw_inode->i_gid_low =
4081 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4082 raw_inode->i_uid_high = 0;
4083 raw_inode->i_gid_high = 0;
4085 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4087 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4088 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4089 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4090 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4092 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4094 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4095 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4096 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4097 cpu_to_le32(EXT4_OS_HURD))
4098 raw_inode->i_file_acl_high =
4099 cpu_to_le16(ei->i_file_acl >> 32);
4100 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4101 if (ei->i_disksize != ext4_isize(raw_inode)) {
4102 ext4_isize_set(raw_inode, ei->i_disksize);
4105 if (ei->i_disksize > 0x7fffffffULL) {
4106 struct super_block *sb = inode->i_sb;
4107 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4108 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4109 EXT4_SB(sb)->s_es->s_rev_level ==
4110 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4111 /* If this is the first large file
4112 * created, add a flag to the superblock.
4114 err = ext4_journal_get_write_access(handle,
4115 EXT4_SB(sb)->s_sbh);
4118 ext4_update_dynamic_rev(sb);
4119 EXT4_SET_RO_COMPAT_FEATURE(sb,
4120 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4122 ext4_handle_sync(handle);
4123 err = ext4_handle_dirty_metadata(handle, NULL,
4124 EXT4_SB(sb)->s_sbh);
4127 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4128 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4129 if (old_valid_dev(inode->i_rdev)) {
4130 raw_inode->i_block[0] =
4131 cpu_to_le32(old_encode_dev(inode->i_rdev));
4132 raw_inode->i_block[1] = 0;
4134 raw_inode->i_block[0] = 0;
4135 raw_inode->i_block[1] =
4136 cpu_to_le32(new_encode_dev(inode->i_rdev));
4137 raw_inode->i_block[2] = 0;
4140 for (block = 0; block < EXT4_N_BLOCKS; block++)
4141 raw_inode->i_block[block] = ei->i_data[block];
4143 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4144 if (ei->i_extra_isize) {
4145 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4146 raw_inode->i_version_hi =
4147 cpu_to_le32(inode->i_version >> 32);
4148 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4151 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4152 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4155 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4157 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4160 ext4_std_error(inode->i_sb, err);
4165 * ext4_write_inode()
4167 * We are called from a few places:
4169 * - Within generic_file_write() for O_SYNC files.
4170 * Here, there will be no transaction running. We wait for any running
4171 * trasnaction to commit.
4173 * - Within sys_sync(), kupdate and such.
4174 * We wait on commit, if tol to.
4176 * - Within prune_icache() (PF_MEMALLOC == true)
4177 * Here we simply return. We can't afford to block kswapd on the
4180 * In all cases it is actually safe for us to return without doing anything,
4181 * because the inode has been copied into a raw inode buffer in
4182 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4185 * Note that we are absolutely dependent upon all inode dirtiers doing the
4186 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4187 * which we are interested.
4189 * It would be a bug for them to not do this. The code:
4191 * mark_inode_dirty(inode)
4193 * inode->i_size = expr;
4195 * is in error because a kswapd-driven write_inode() could occur while
4196 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4197 * will no longer be on the superblock's dirty inode list.
4199 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4203 if (current->flags & PF_MEMALLOC)
4206 if (EXT4_SB(inode->i_sb)->s_journal) {
4207 if (ext4_journal_current_handle()) {
4208 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4213 if (wbc->sync_mode != WB_SYNC_ALL)
4216 err = ext4_force_commit(inode->i_sb);
4218 struct ext4_iloc iloc;
4220 err = __ext4_get_inode_loc(inode, &iloc, 0);
4223 if (wbc->sync_mode == WB_SYNC_ALL)
4224 sync_dirty_buffer(iloc.bh);
4225 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4226 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4227 "IO error syncing inode");
4238 * Called from notify_change.
4240 * We want to trap VFS attempts to truncate the file as soon as
4241 * possible. In particular, we want to make sure that when the VFS
4242 * shrinks i_size, we put the inode on the orphan list and modify
4243 * i_disksize immediately, so that during the subsequent flushing of
4244 * dirty pages and freeing of disk blocks, we can guarantee that any
4245 * commit will leave the blocks being flushed in an unused state on
4246 * disk. (On recovery, the inode will get truncated and the blocks will
4247 * be freed, so we have a strong guarantee that no future commit will
4248 * leave these blocks visible to the user.)
4250 * Another thing we have to assure is that if we are in ordered mode
4251 * and inode is still attached to the committing transaction, we must
4252 * we start writeout of all the dirty pages which are being truncated.
4253 * This way we are sure that all the data written in the previous
4254 * transaction are already on disk (truncate waits for pages under
4257 * Called with inode->i_mutex down.
4259 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4261 struct inode *inode = dentry->d_inode;
4264 const unsigned int ia_valid = attr->ia_valid;
4266 error = inode_change_ok(inode, attr);
4270 if (is_quota_modification(inode, attr))
4271 dquot_initialize(inode);
4272 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4273 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4276 /* (user+group)*(old+new) structure, inode write (sb,
4277 * inode block, ? - but truncate inode update has it) */
4278 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
4279 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
4280 if (IS_ERR(handle)) {
4281 error = PTR_ERR(handle);
4284 error = dquot_transfer(inode, attr);
4286 ext4_journal_stop(handle);
4289 /* Update corresponding info in inode so that everything is in
4290 * one transaction */
4291 if (attr->ia_valid & ATTR_UID)
4292 inode->i_uid = attr->ia_uid;
4293 if (attr->ia_valid & ATTR_GID)
4294 inode->i_gid = attr->ia_gid;
4295 error = ext4_mark_inode_dirty(handle, inode);
4296 ext4_journal_stop(handle);
4299 if (attr->ia_valid & ATTR_SIZE) {
4300 inode_dio_wait(inode);
4302 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4303 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4305 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4310 if (S_ISREG(inode->i_mode) &&
4311 attr->ia_valid & ATTR_SIZE &&
4312 (attr->ia_size < inode->i_size)) {
4315 handle = ext4_journal_start(inode, 3);
4316 if (IS_ERR(handle)) {
4317 error = PTR_ERR(handle);
4320 if (ext4_handle_valid(handle)) {
4321 error = ext4_orphan_add(handle, inode);
4324 EXT4_I(inode)->i_disksize = attr->ia_size;
4325 rc = ext4_mark_inode_dirty(handle, inode);
4328 ext4_journal_stop(handle);
4330 if (ext4_should_order_data(inode)) {
4331 error = ext4_begin_ordered_truncate(inode,
4334 /* Do as much error cleanup as possible */
4335 handle = ext4_journal_start(inode, 3);
4336 if (IS_ERR(handle)) {
4337 ext4_orphan_del(NULL, inode);
4340 ext4_orphan_del(handle, inode);
4342 ext4_journal_stop(handle);
4348 if (attr->ia_valid & ATTR_SIZE) {
4349 if (attr->ia_size != i_size_read(inode)) {
4350 truncate_setsize(inode, attr->ia_size);
4351 ext4_truncate(inode);
4352 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
4353 ext4_truncate(inode);
4357 setattr_copy(inode, attr);
4358 mark_inode_dirty(inode);
4362 * If the call to ext4_truncate failed to get a transaction handle at
4363 * all, we need to clean up the in-core orphan list manually.
4365 if (orphan && inode->i_nlink)
4366 ext4_orphan_del(NULL, inode);
4368 if (!rc && (ia_valid & ATTR_MODE))
4369 rc = ext4_acl_chmod(inode);
4372 ext4_std_error(inode->i_sb, error);
4378 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4381 struct inode *inode;
4382 unsigned long long delalloc_blocks;
4384 inode = dentry->d_inode;
4385 generic_fillattr(inode, stat);
4388 * We can't update i_blocks if the block allocation is delayed
4389 * otherwise in the case of system crash before the real block
4390 * allocation is done, we will have i_blocks inconsistent with
4391 * on-disk file blocks.
4392 * We always keep i_blocks updated together with real
4393 * allocation. But to not confuse with user, stat
4394 * will return the blocks that include the delayed allocation
4395 * blocks for this file.
4397 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4399 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits-9);
4403 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4405 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4406 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4407 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4411 * Account for index blocks, block groups bitmaps and block group
4412 * descriptor blocks if modify datablocks and index blocks
4413 * worse case, the indexs blocks spread over different block groups
4415 * If datablocks are discontiguous, they are possible to spread over
4416 * different block groups too. If they are contiuguous, with flexbg,
4417 * they could still across block group boundary.
4419 * Also account for superblock, inode, quota and xattr blocks
4421 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4423 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4429 * How many index blocks need to touch to modify nrblocks?
4430 * The "Chunk" flag indicating whether the nrblocks is
4431 * physically contiguous on disk
4433 * For Direct IO and fallocate, they calls get_block to allocate
4434 * one single extent at a time, so they could set the "Chunk" flag
4436 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4441 * Now let's see how many group bitmaps and group descriptors need
4451 if (groups > ngroups)
4453 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4454 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4456 /* bitmaps and block group descriptor blocks */
4457 ret += groups + gdpblocks;
4459 /* Blocks for super block, inode, quota and xattr blocks */
4460 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4466 * Calculate the total number of credits to reserve to fit
4467 * the modification of a single pages into a single transaction,
4468 * which may include multiple chunks of block allocations.
4470 * This could be called via ext4_write_begin()
4472 * We need to consider the worse case, when
4473 * one new block per extent.
4475 int ext4_writepage_trans_blocks(struct inode *inode)
4477 int bpp = ext4_journal_blocks_per_page(inode);
4480 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4482 /* Account for data blocks for journalled mode */
4483 if (ext4_should_journal_data(inode))
4489 * Calculate the journal credits for a chunk of data modification.
4491 * This is called from DIO, fallocate or whoever calling
4492 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4494 * journal buffers for data blocks are not included here, as DIO
4495 * and fallocate do no need to journal data buffers.
4497 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4499 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4503 * The caller must have previously called ext4_reserve_inode_write().
4504 * Give this, we know that the caller already has write access to iloc->bh.
4506 int ext4_mark_iloc_dirty(handle_t *handle,
4507 struct inode *inode, struct ext4_iloc *iloc)
4511 if (test_opt(inode->i_sb, I_VERSION))
4512 inode_inc_iversion(inode);
4514 /* the do_update_inode consumes one bh->b_count */
4517 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4518 err = ext4_do_update_inode(handle, inode, iloc);
4524 * On success, We end up with an outstanding reference count against
4525 * iloc->bh. This _must_ be cleaned up later.
4529 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4530 struct ext4_iloc *iloc)
4534 err = ext4_get_inode_loc(inode, iloc);
4536 BUFFER_TRACE(iloc->bh, "get_write_access");
4537 err = ext4_journal_get_write_access(handle, iloc->bh);
4543 ext4_std_error(inode->i_sb, err);
4548 * Expand an inode by new_extra_isize bytes.
4549 * Returns 0 on success or negative error number on failure.
4551 static int ext4_expand_extra_isize(struct inode *inode,
4552 unsigned int new_extra_isize,
4553 struct ext4_iloc iloc,
4556 struct ext4_inode *raw_inode;
4557 struct ext4_xattr_ibody_header *header;
4559 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4562 raw_inode = ext4_raw_inode(&iloc);
4564 header = IHDR(inode, raw_inode);
4566 /* No extended attributes present */
4567 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4568 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4569 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4571 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4575 /* try to expand with EAs present */
4576 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4581 * What we do here is to mark the in-core inode as clean with respect to inode
4582 * dirtiness (it may still be data-dirty).
4583 * This means that the in-core inode may be reaped by prune_icache
4584 * without having to perform any I/O. This is a very good thing,
4585 * because *any* task may call prune_icache - even ones which
4586 * have a transaction open against a different journal.
4588 * Is this cheating? Not really. Sure, we haven't written the
4589 * inode out, but prune_icache isn't a user-visible syncing function.
4590 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4591 * we start and wait on commits.
4593 * Is this efficient/effective? Well, we're being nice to the system
4594 * by cleaning up our inodes proactively so they can be reaped
4595 * without I/O. But we are potentially leaving up to five seconds'
4596 * worth of inodes floating about which prune_icache wants us to
4597 * write out. One way to fix that would be to get prune_icache()
4598 * to do a write_super() to free up some memory. It has the desired
4601 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4603 struct ext4_iloc iloc;
4604 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4605 static unsigned int mnt_count;
4609 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4610 err = ext4_reserve_inode_write(handle, inode, &iloc);
4611 if (ext4_handle_valid(handle) &&
4612 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4613 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4615 * We need extra buffer credits since we may write into EA block
4616 * with this same handle. If journal_extend fails, then it will
4617 * only result in a minor loss of functionality for that inode.
4618 * If this is felt to be critical, then e2fsck should be run to
4619 * force a large enough s_min_extra_isize.
4621 if ((jbd2_journal_extend(handle,
4622 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4623 ret = ext4_expand_extra_isize(inode,
4624 sbi->s_want_extra_isize,
4627 ext4_set_inode_state(inode,
4628 EXT4_STATE_NO_EXPAND);
4630 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4631 ext4_warning(inode->i_sb,
4632 "Unable to expand inode %lu. Delete"
4633 " some EAs or run e2fsck.",
4636 le16_to_cpu(sbi->s_es->s_mnt_count);
4642 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4647 * ext4_dirty_inode() is called from __mark_inode_dirty()
4649 * We're really interested in the case where a file is being extended.
4650 * i_size has been changed by generic_commit_write() and we thus need
4651 * to include the updated inode in the current transaction.
4653 * Also, dquot_alloc_block() will always dirty the inode when blocks
4654 * are allocated to the file.
4656 * If the inode is marked synchronous, we don't honour that here - doing
4657 * so would cause a commit on atime updates, which we don't bother doing.
4658 * We handle synchronous inodes at the highest possible level.
4660 void ext4_dirty_inode(struct inode *inode, int flags)
4664 handle = ext4_journal_start(inode, 2);
4668 ext4_mark_inode_dirty(handle, inode);
4670 ext4_journal_stop(handle);
4677 * Bind an inode's backing buffer_head into this transaction, to prevent
4678 * it from being flushed to disk early. Unlike
4679 * ext4_reserve_inode_write, this leaves behind no bh reference and
4680 * returns no iloc structure, so the caller needs to repeat the iloc
4681 * lookup to mark the inode dirty later.
4683 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4685 struct ext4_iloc iloc;
4689 err = ext4_get_inode_loc(inode, &iloc);
4691 BUFFER_TRACE(iloc.bh, "get_write_access");
4692 err = jbd2_journal_get_write_access(handle, iloc.bh);
4694 err = ext4_handle_dirty_metadata(handle,
4700 ext4_std_error(inode->i_sb, err);
4705 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4712 * We have to be very careful here: changing a data block's
4713 * journaling status dynamically is dangerous. If we write a
4714 * data block to the journal, change the status and then delete
4715 * that block, we risk forgetting to revoke the old log record
4716 * from the journal and so a subsequent replay can corrupt data.
4717 * So, first we make sure that the journal is empty and that
4718 * nobody is changing anything.
4721 journal = EXT4_JOURNAL(inode);
4724 if (is_journal_aborted(journal))
4727 jbd2_journal_lock_updates(journal);
4728 jbd2_journal_flush(journal);
4731 * OK, there are no updates running now, and all cached data is
4732 * synced to disk. We are now in a completely consistent state
4733 * which doesn't have anything in the journal, and we know that
4734 * no filesystem updates are running, so it is safe to modify
4735 * the inode's in-core data-journaling state flag now.
4739 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4741 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4742 ext4_set_aops(inode);
4744 jbd2_journal_unlock_updates(journal);
4746 /* Finally we can mark the inode as dirty. */
4748 handle = ext4_journal_start(inode, 1);
4750 return PTR_ERR(handle);
4752 err = ext4_mark_inode_dirty(handle, inode);
4753 ext4_handle_sync(handle);
4754 ext4_journal_stop(handle);
4755 ext4_std_error(inode->i_sb, err);
4760 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4762 return !buffer_mapped(bh);
4765 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4767 struct page *page = vmf->page;
4771 struct file *file = vma->vm_file;
4772 struct inode *inode = file->f_path.dentry->d_inode;
4773 struct address_space *mapping = inode->i_mapping;
4775 get_block_t *get_block;
4779 * This check is racy but catches the common case. We rely on
4780 * __block_page_mkwrite() to do a reliable check.
4782 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
4783 /* Delalloc case is easy... */
4784 if (test_opt(inode->i_sb, DELALLOC) &&
4785 !ext4_should_journal_data(inode) &&
4786 !ext4_nonda_switch(inode->i_sb)) {
4788 ret = __block_page_mkwrite(vma, vmf,
4789 ext4_da_get_block_prep);
4790 } while (ret == -ENOSPC &&
4791 ext4_should_retry_alloc(inode->i_sb, &retries));
4796 size = i_size_read(inode);
4797 /* Page got truncated from under us? */
4798 if (page->mapping != mapping || page_offset(page) > size) {
4800 ret = VM_FAULT_NOPAGE;
4804 if (page->index == size >> PAGE_CACHE_SHIFT)
4805 len = size & ~PAGE_CACHE_MASK;
4807 len = PAGE_CACHE_SIZE;
4809 * Return if we have all the buffers mapped. This avoids the need to do
4810 * journal_start/journal_stop which can block and take a long time
4812 if (page_has_buffers(page)) {
4813 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4814 ext4_bh_unmapped)) {
4815 /* Wait so that we don't change page under IO */
4816 wait_on_page_writeback(page);
4817 ret = VM_FAULT_LOCKED;
4822 /* OK, we need to fill the hole... */
4823 if (ext4_should_dioread_nolock(inode))
4824 get_block = ext4_get_block_write;
4826 get_block = ext4_get_block;
4828 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4829 if (IS_ERR(handle)) {
4830 ret = VM_FAULT_SIGBUS;
4833 ret = __block_page_mkwrite(vma, vmf, get_block);
4834 if (!ret && ext4_should_journal_data(inode)) {
4835 if (walk_page_buffers(handle, page_buffers(page), 0,
4836 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4838 ret = VM_FAULT_SIGBUS;
4839 ext4_journal_stop(handle);
4842 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4844 ext4_journal_stop(handle);
4845 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4848 ret = block_page_mkwrite_return(ret);