2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
21 #include <linux/module.h>
23 #include <linux/time.h>
24 #include <linux/jbd2.h>
25 #include <linux/highuid.h>
26 #include <linux/pagemap.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/bitops.h>
43 #include "ext4_jbd2.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
52 static inline int ext4_begin_ordered_truncate(struct inode *inode,
55 trace_ext4_begin_ordered_truncate(inode, new_size);
57 * If jinode is zero, then we never opened the file for
58 * writing, so there's no need to call
59 * jbd2_journal_begin_ordered_truncate() since there's no
60 * outstanding writes we need to flush.
62 if (!EXT4_I(inode)->jinode)
64 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
65 EXT4_I(inode)->jinode,
69 static void ext4_invalidatepage(struct page *page, unsigned long offset);
70 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
71 struct buffer_head *bh_result, int create);
72 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
73 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
74 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
75 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
78 * Test whether an inode is a fast symlink.
80 static int ext4_inode_is_fast_symlink(struct inode *inode)
82 int ea_blocks = EXT4_I(inode)->i_file_acl ?
83 (inode->i_sb->s_blocksize >> 9) : 0;
85 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
89 * Restart the transaction associated with *handle. This does a commit,
90 * so before we call here everything must be consistently dirtied against
93 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
99 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
100 * moment, get_block can be called only for blocks inside i_size since
101 * page cache has been already dropped and writes are blocked by
102 * i_mutex. So we can safely drop the i_data_sem here.
104 BUG_ON(EXT4_JOURNAL(inode) == NULL);
105 jbd_debug(2, "restarting handle %p\n", handle);
106 up_write(&EXT4_I(inode)->i_data_sem);
107 ret = ext4_journal_restart(handle, nblocks);
108 down_write(&EXT4_I(inode)->i_data_sem);
109 ext4_discard_preallocations(inode);
115 * Called at the last iput() if i_nlink is zero.
117 void ext4_evict_inode(struct inode *inode)
122 trace_ext4_evict_inode(inode);
124 ext4_ioend_wait(inode);
126 if (inode->i_nlink) {
128 * When journalling data dirty buffers are tracked only in the
129 * journal. So although mm thinks everything is clean and
130 * ready for reaping the inode might still have some pages to
131 * write in the running transaction or waiting to be
132 * checkpointed. Thus calling jbd2_journal_invalidatepage()
133 * (via truncate_inode_pages()) to discard these buffers can
134 * cause data loss. Also even if we did not discard these
135 * buffers, we would have no way to find them after the inode
136 * is reaped and thus user could see stale data if he tries to
137 * read them before the transaction is checkpointed. So be
138 * careful and force everything to disk here... We use
139 * ei->i_datasync_tid to store the newest transaction
140 * containing inode's data.
142 * Note that directories do not have this problem because they
143 * don't use page cache.
145 if (inode->i_ino != EXT4_JOURNAL_INO &&
146 ext4_should_journal_data(inode) &&
147 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
148 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
149 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
151 jbd2_complete_transaction(journal, commit_tid);
152 filemap_write_and_wait(&inode->i_data);
154 truncate_inode_pages(&inode->i_data, 0);
158 if (is_bad_inode(inode))
160 dquot_initialize(inode);
162 if (ext4_should_order_data(inode))
163 ext4_begin_ordered_truncate(inode, 0);
164 truncate_inode_pages(&inode->i_data, 0);
167 handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
168 if (IS_ERR(handle)) {
169 ext4_std_error(inode->i_sb, PTR_ERR(handle));
171 * If we're going to skip the normal cleanup, we still need to
172 * make sure that the in-core orphan linked list is properly
175 ext4_orphan_del(NULL, inode);
180 ext4_handle_sync(handle);
182 err = ext4_mark_inode_dirty(handle, inode);
184 ext4_warning(inode->i_sb,
185 "couldn't mark inode dirty (err %d)", err);
189 ext4_truncate(inode);
192 * ext4_ext_truncate() doesn't reserve any slop when it
193 * restarts journal transactions; therefore there may not be
194 * enough credits left in the handle to remove the inode from
195 * the orphan list and set the dtime field.
197 if (!ext4_handle_has_enough_credits(handle, 3)) {
198 err = ext4_journal_extend(handle, 3);
200 err = ext4_journal_restart(handle, 3);
202 ext4_warning(inode->i_sb,
203 "couldn't extend journal (err %d)", err);
205 ext4_journal_stop(handle);
206 ext4_orphan_del(NULL, inode);
212 * Kill off the orphan record which ext4_truncate created.
213 * AKPM: I think this can be inside the above `if'.
214 * Note that ext4_orphan_del() has to be able to cope with the
215 * deletion of a non-existent orphan - this is because we don't
216 * know if ext4_truncate() actually created an orphan record.
217 * (Well, we could do this if we need to, but heck - it works)
219 ext4_orphan_del(handle, inode);
220 EXT4_I(inode)->i_dtime = get_seconds();
223 * One subtle ordering requirement: if anything has gone wrong
224 * (transaction abort, IO errors, whatever), then we can still
225 * do these next steps (the fs will already have been marked as
226 * having errors), but we can't free the inode if the mark_dirty
229 if (ext4_mark_inode_dirty(handle, inode))
230 /* If that failed, just do the required in-core inode clear. */
231 ext4_clear_inode(inode);
233 ext4_free_inode(handle, inode);
234 ext4_journal_stop(handle);
237 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
241 qsize_t *ext4_get_reserved_space(struct inode *inode)
243 return &EXT4_I(inode)->i_reserved_quota;
248 * Calculate the number of metadata blocks need to reserve
249 * to allocate a block located at @lblock
251 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
253 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
254 return ext4_ext_calc_metadata_amount(inode, lblock);
256 return ext4_ind_calc_metadata_amount(inode, lblock);
260 * Called with i_data_sem down, which is important since we can call
261 * ext4_discard_preallocations() from here.
263 void ext4_da_update_reserve_space(struct inode *inode,
264 int used, int quota_claim)
266 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
267 struct ext4_inode_info *ei = EXT4_I(inode);
269 spin_lock(&ei->i_block_reservation_lock);
270 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
271 if (unlikely(used > ei->i_reserved_data_blocks)) {
272 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
273 "with only %d reserved data blocks\n",
274 __func__, inode->i_ino, used,
275 ei->i_reserved_data_blocks);
277 used = ei->i_reserved_data_blocks;
280 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
281 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, allocated %d "
282 "with only %d reserved metadata blocks\n", __func__,
283 inode->i_ino, ei->i_allocated_meta_blocks,
284 ei->i_reserved_meta_blocks);
286 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
289 /* Update per-inode reservations */
290 ei->i_reserved_data_blocks -= used;
291 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
292 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
293 used + ei->i_allocated_meta_blocks);
294 ei->i_allocated_meta_blocks = 0;
296 if (ei->i_reserved_data_blocks == 0) {
298 * We can release all of the reserved metadata blocks
299 * only when we have written all of the delayed
302 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
303 ei->i_reserved_meta_blocks);
304 ei->i_reserved_meta_blocks = 0;
305 ei->i_da_metadata_calc_len = 0;
307 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
309 /* Update quota subsystem for data blocks */
311 dquot_claim_block(inode, EXT4_C2B(sbi, used));
314 * We did fallocate with an offset that is already delayed
315 * allocated. So on delayed allocated writeback we should
316 * not re-claim the quota for fallocated blocks.
318 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
322 * If we have done all the pending block allocations and if
323 * there aren't any writers on the inode, we can discard the
324 * inode's preallocations.
326 if ((ei->i_reserved_data_blocks == 0) &&
327 (atomic_read(&inode->i_writecount) == 0))
328 ext4_discard_preallocations(inode);
331 static int __check_block_validity(struct inode *inode, const char *func,
333 struct ext4_map_blocks *map)
335 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
337 ext4_error_inode(inode, func, line, map->m_pblk,
338 "lblock %lu mapped to illegal pblock "
339 "(length %d)", (unsigned long) map->m_lblk,
346 #define check_block_validity(inode, map) \
347 __check_block_validity((inode), __func__, __LINE__, (map))
350 * Return the number of contiguous dirty pages in a given inode
351 * starting at page frame idx.
353 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
354 unsigned int max_pages)
356 struct address_space *mapping = inode->i_mapping;
360 int i, nr_pages, done = 0;
364 pagevec_init(&pvec, 0);
367 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
369 (pgoff_t)PAGEVEC_SIZE);
372 for (i = 0; i < nr_pages; i++) {
373 struct page *page = pvec.pages[i];
374 struct buffer_head *bh, *head;
377 if (unlikely(page->mapping != mapping) ||
379 PageWriteback(page) ||
380 page->index != idx) {
385 if (page_has_buffers(page)) {
386 bh = head = page_buffers(page);
388 if (!buffer_delay(bh) &&
389 !buffer_unwritten(bh))
391 bh = bh->b_this_page;
392 } while (!done && (bh != head));
399 if (num >= max_pages) {
404 pagevec_release(&pvec);
410 * Sets the BH_Da_Mapped bit on the buffer heads corresponding to the given map.
412 static void set_buffers_da_mapped(struct inode *inode,
413 struct ext4_map_blocks *map)
415 struct address_space *mapping = inode->i_mapping;
420 index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
421 end = (map->m_lblk + map->m_len - 1) >>
422 (PAGE_CACHE_SHIFT - inode->i_blkbits);
424 pagevec_init(&pvec, 0);
425 while (index <= end) {
426 nr_pages = pagevec_lookup(&pvec, mapping, index,
428 (pgoff_t)PAGEVEC_SIZE));
431 for (i = 0; i < nr_pages; i++) {
432 struct page *page = pvec.pages[i];
433 struct buffer_head *bh, *head;
435 if (unlikely(page->mapping != mapping) ||
439 if (page_has_buffers(page)) {
440 bh = head = page_buffers(page);
442 set_buffer_da_mapped(bh);
443 bh = bh->b_this_page;
444 } while (bh != head);
448 pagevec_release(&pvec);
453 * The ext4_map_blocks() function tries to look up the requested blocks,
454 * and returns if the blocks are already mapped.
456 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
457 * and store the allocated blocks in the result buffer head and mark it
460 * If file type is extents based, it will call ext4_ext_map_blocks(),
461 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
464 * On success, it returns the number of blocks being mapped or allocate.
465 * if create==0 and the blocks are pre-allocated and uninitialized block,
466 * the result buffer head is unmapped. If the create ==1, it will make sure
467 * the buffer head is mapped.
469 * It returns 0 if plain look up failed (blocks have not been allocated), in
470 * that case, buffer head is unmapped
472 * It returns the error in case of allocation failure.
474 int ext4_map_blocks(handle_t *handle, struct inode *inode,
475 struct ext4_map_blocks *map, int flags)
480 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
481 "logical block %lu\n", inode->i_ino, flags, map->m_len,
482 (unsigned long) map->m_lblk);
484 /* We can handle the block number less than EXT_MAX_BLOCKS */
485 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
489 * Try to see if we can get the block without requesting a new
492 down_read((&EXT4_I(inode)->i_data_sem));
493 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
494 retval = ext4_ext_map_blocks(handle, inode, map, flags &
495 EXT4_GET_BLOCKS_KEEP_SIZE);
497 retval = ext4_ind_map_blocks(handle, inode, map, flags &
498 EXT4_GET_BLOCKS_KEEP_SIZE);
500 up_read((&EXT4_I(inode)->i_data_sem));
502 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
503 int ret = check_block_validity(inode, map);
508 /* If it is only a block(s) look up */
509 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
513 * Returns if the blocks have already allocated
515 * Note that if blocks have been preallocated
516 * ext4_ext_get_block() returns the create = 0
517 * with buffer head unmapped.
519 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
523 * When we call get_blocks without the create flag, the
524 * BH_Unwritten flag could have gotten set if the blocks
525 * requested were part of a uninitialized extent. We need to
526 * clear this flag now that we are committed to convert all or
527 * part of the uninitialized extent to be an initialized
528 * extent. This is because we need to avoid the combination
529 * of BH_Unwritten and BH_Mapped flags being simultaneously
530 * set on the buffer_head.
532 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
535 * New blocks allocate and/or writing to uninitialized extent
536 * will possibly result in updating i_data, so we take
537 * the write lock of i_data_sem, and call get_blocks()
538 * with create == 1 flag.
540 down_write((&EXT4_I(inode)->i_data_sem));
543 * if the caller is from delayed allocation writeout path
544 * we have already reserved fs blocks for allocation
545 * let the underlying get_block() function know to
546 * avoid double accounting
548 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
549 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
551 * We need to check for EXT4 here because migrate
552 * could have changed the inode type in between
554 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
555 retval = ext4_ext_map_blocks(handle, inode, map, flags);
557 retval = ext4_ind_map_blocks(handle, inode, map, flags);
559 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
561 * We allocated new blocks which will result in
562 * i_data's format changing. Force the migrate
563 * to fail by clearing migrate flags
565 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
569 * Update reserved blocks/metadata blocks after successful
570 * block allocation which had been deferred till now. We don't
571 * support fallocate for non extent files. So we can update
572 * reserve space here.
575 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
576 ext4_da_update_reserve_space(inode, retval, 1);
578 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
579 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
581 /* If we have successfully mapped the delayed allocated blocks,
582 * set the BH_Da_Mapped bit on them. Its important to do this
583 * under the protection of i_data_sem.
585 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
586 set_buffers_da_mapped(inode, map);
589 up_write((&EXT4_I(inode)->i_data_sem));
590 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
591 int ret = check_block_validity(inode, map);
599 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
600 * we have to be careful as someone else may be manipulating b_state as well.
602 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
604 unsigned long old_state;
605 unsigned long new_state;
607 flags &= EXT4_MAP_FLAGS;
609 /* Dummy buffer_head? Set non-atomically. */
611 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
615 * Someone else may be modifying b_state. Be careful! This is ugly but
616 * once we get rid of using bh as a container for mapping information
617 * to pass to / from get_block functions, this can go away.
620 old_state = ACCESS_ONCE(bh->b_state);
621 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
623 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
626 /* Maximum number of blocks we map for direct IO at once. */
627 #define DIO_MAX_BLOCKS 4096
629 static int _ext4_get_block(struct inode *inode, sector_t iblock,
630 struct buffer_head *bh, int flags)
632 handle_t *handle = ext4_journal_current_handle();
633 struct ext4_map_blocks map;
634 int ret = 0, started = 0;
638 map.m_len = bh->b_size >> inode->i_blkbits;
640 if (flags && !handle) {
641 /* Direct IO write... */
642 if (map.m_len > DIO_MAX_BLOCKS)
643 map.m_len = DIO_MAX_BLOCKS;
644 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
645 handle = ext4_journal_start(inode, dio_credits);
646 if (IS_ERR(handle)) {
647 ret = PTR_ERR(handle);
653 ret = ext4_map_blocks(handle, inode, &map, flags);
655 map_bh(bh, inode->i_sb, map.m_pblk);
656 ext4_update_bh_state(bh, map.m_flags);
657 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
661 ext4_journal_stop(handle);
665 int ext4_get_block(struct inode *inode, sector_t iblock,
666 struct buffer_head *bh, int create)
668 return _ext4_get_block(inode, iblock, bh,
669 create ? EXT4_GET_BLOCKS_CREATE : 0);
673 * `handle' can be NULL if create is zero
675 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
676 ext4_lblk_t block, int create, int *errp)
678 struct ext4_map_blocks map;
679 struct buffer_head *bh;
682 J_ASSERT(handle != NULL || create == 0);
686 err = ext4_map_blocks(handle, inode, &map,
687 create ? EXT4_GET_BLOCKS_CREATE : 0);
695 bh = sb_getblk(inode->i_sb, map.m_pblk);
700 if (map.m_flags & EXT4_MAP_NEW) {
701 J_ASSERT(create != 0);
702 J_ASSERT(handle != NULL);
705 * Now that we do not always journal data, we should
706 * keep in mind whether this should always journal the
707 * new buffer as metadata. For now, regular file
708 * writes use ext4_get_block instead, so it's not a
712 BUFFER_TRACE(bh, "call get_create_access");
713 fatal = ext4_journal_get_create_access(handle, bh);
714 if (!fatal && !buffer_uptodate(bh)) {
715 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
716 set_buffer_uptodate(bh);
719 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
720 err = ext4_handle_dirty_metadata(handle, inode, bh);
724 BUFFER_TRACE(bh, "not a new buffer");
734 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
735 ext4_lblk_t block, int create, int *err)
737 struct buffer_head *bh;
739 bh = ext4_getblk(handle, inode, block, create, err);
742 if (buffer_uptodate(bh))
744 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
746 if (buffer_uptodate(bh))
753 static int walk_page_buffers(handle_t *handle,
754 struct buffer_head *head,
758 int (*fn)(handle_t *handle,
759 struct buffer_head *bh))
761 struct buffer_head *bh;
762 unsigned block_start, block_end;
763 unsigned blocksize = head->b_size;
765 struct buffer_head *next;
767 for (bh = head, block_start = 0;
768 ret == 0 && (bh != head || !block_start);
769 block_start = block_end, bh = next) {
770 next = bh->b_this_page;
771 block_end = block_start + blocksize;
772 if (block_end <= from || block_start >= to) {
773 if (partial && !buffer_uptodate(bh))
777 err = (*fn)(handle, bh);
785 * To preserve ordering, it is essential that the hole instantiation and
786 * the data write be encapsulated in a single transaction. We cannot
787 * close off a transaction and start a new one between the ext4_get_block()
788 * and the commit_write(). So doing the jbd2_journal_start at the start of
789 * prepare_write() is the right place.
791 * Also, this function can nest inside ext4_writepage() ->
792 * block_write_full_page(). In that case, we *know* that ext4_writepage()
793 * has generated enough buffer credits to do the whole page. So we won't
794 * block on the journal in that case, which is good, because the caller may
797 * By accident, ext4 can be reentered when a transaction is open via
798 * quota file writes. If we were to commit the transaction while thus
799 * reentered, there can be a deadlock - we would be holding a quota
800 * lock, and the commit would never complete if another thread had a
801 * transaction open and was blocking on the quota lock - a ranking
804 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
805 * will _not_ run commit under these circumstances because handle->h_ref
806 * is elevated. We'll still have enough credits for the tiny quotafile
809 static int do_journal_get_write_access(handle_t *handle,
810 struct buffer_head *bh)
812 int dirty = buffer_dirty(bh);
815 if (!buffer_mapped(bh) || buffer_freed(bh))
818 * __block_write_begin() could have dirtied some buffers. Clean
819 * the dirty bit as jbd2_journal_get_write_access() could complain
820 * otherwise about fs integrity issues. Setting of the dirty bit
821 * by __block_write_begin() isn't a real problem here as we clear
822 * the bit before releasing a page lock and thus writeback cannot
823 * ever write the buffer.
826 clear_buffer_dirty(bh);
827 ret = ext4_journal_get_write_access(handle, bh);
829 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
833 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
834 struct buffer_head *bh_result, int create);
835 static int ext4_write_begin(struct file *file, struct address_space *mapping,
836 loff_t pos, unsigned len, unsigned flags,
837 struct page **pagep, void **fsdata)
839 struct inode *inode = mapping->host;
840 int ret, needed_blocks;
847 trace_ext4_write_begin(inode, pos, len, flags);
849 * Reserve one block more for addition to orphan list in case
850 * we allocate blocks but write fails for some reason
852 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
853 index = pos >> PAGE_CACHE_SHIFT;
854 from = pos & (PAGE_CACHE_SIZE - 1);
858 handle = ext4_journal_start(inode, needed_blocks);
859 if (IS_ERR(handle)) {
860 ret = PTR_ERR(handle);
864 /* We cannot recurse into the filesystem as the transaction is already
866 flags |= AOP_FLAG_NOFS;
868 page = grab_cache_page_write_begin(mapping, index, flags);
870 ext4_journal_stop(handle);
876 if (ext4_should_dioread_nolock(inode))
877 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
879 ret = __block_write_begin(page, pos, len, ext4_get_block);
881 if (!ret && ext4_should_journal_data(inode)) {
882 ret = walk_page_buffers(handle, page_buffers(page),
883 from, to, NULL, do_journal_get_write_access);
888 page_cache_release(page);
890 * __block_write_begin may have instantiated a few blocks
891 * outside i_size. Trim these off again. Don't need
892 * i_size_read because we hold i_mutex.
894 * Add inode to orphan list in case we crash before
897 if (pos + len > inode->i_size && ext4_can_truncate(inode))
898 ext4_orphan_add(handle, inode);
900 ext4_journal_stop(handle);
901 if (pos + len > inode->i_size) {
902 ext4_truncate_failed_write(inode);
904 * If truncate failed early the inode might
905 * still be on the orphan list; we need to
906 * make sure the inode is removed from the
907 * orphan list in that case.
910 ext4_orphan_del(NULL, inode);
914 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
920 /* For write_end() in data=journal mode */
921 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
923 if (!buffer_mapped(bh) || buffer_freed(bh))
925 set_buffer_uptodate(bh);
926 return ext4_handle_dirty_metadata(handle, NULL, bh);
929 static int ext4_generic_write_end(struct file *file,
930 struct address_space *mapping,
931 loff_t pos, unsigned len, unsigned copied,
932 struct page *page, void *fsdata)
934 int i_size_changed = 0;
935 struct inode *inode = mapping->host;
936 handle_t *handle = ext4_journal_current_handle();
938 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
941 * No need to use i_size_read() here, the i_size
942 * cannot change under us because we hold i_mutex.
944 * But it's important to update i_size while still holding page lock:
945 * page writeout could otherwise come in and zero beyond i_size.
947 if (pos + copied > inode->i_size) {
948 i_size_write(inode, pos + copied);
952 if (pos + copied > EXT4_I(inode)->i_disksize) {
953 /* We need to mark inode dirty even if
954 * new_i_size is less that inode->i_size
955 * bu greater than i_disksize.(hint delalloc)
957 ext4_update_i_disksize(inode, (pos + copied));
961 page_cache_release(page);
964 * Don't mark the inode dirty under page lock. First, it unnecessarily
965 * makes the holding time of page lock longer. Second, it forces lock
966 * ordering of page lock and transaction start for journaling
970 ext4_mark_inode_dirty(handle, inode);
976 * We need to pick up the new inode size which generic_commit_write gave us
977 * `file' can be NULL - eg, when called from page_symlink().
979 * ext4 never places buffers on inode->i_mapping->private_list. metadata
980 * buffers are managed internally.
982 static int ext4_ordered_write_end(struct file *file,
983 struct address_space *mapping,
984 loff_t pos, unsigned len, unsigned copied,
985 struct page *page, void *fsdata)
987 handle_t *handle = ext4_journal_current_handle();
988 struct inode *inode = mapping->host;
991 trace_ext4_ordered_write_end(inode, pos, len, copied);
992 ret = ext4_jbd2_file_inode(handle, inode);
995 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
998 if (pos + len > inode->i_size && ext4_can_truncate(inode))
999 /* if we have allocated more blocks and copied
1000 * less. We will have blocks allocated outside
1001 * inode->i_size. So truncate them
1003 ext4_orphan_add(handle, inode);
1008 page_cache_release(page);
1011 ret2 = ext4_journal_stop(handle);
1015 if (pos + len > inode->i_size) {
1016 ext4_truncate_failed_write(inode);
1018 * If truncate failed early the inode might still be
1019 * on the orphan list; we need to make sure the inode
1020 * is removed from the orphan list in that case.
1023 ext4_orphan_del(NULL, inode);
1027 return ret ? ret : copied;
1030 static int ext4_writeback_write_end(struct file *file,
1031 struct address_space *mapping,
1032 loff_t pos, unsigned len, unsigned copied,
1033 struct page *page, void *fsdata)
1035 handle_t *handle = ext4_journal_current_handle();
1036 struct inode *inode = mapping->host;
1039 trace_ext4_writeback_write_end(inode, pos, len, copied);
1040 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1043 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1044 /* if we have allocated more blocks and copied
1045 * less. We will have blocks allocated outside
1046 * inode->i_size. So truncate them
1048 ext4_orphan_add(handle, inode);
1053 ret2 = ext4_journal_stop(handle);
1057 if (pos + len > inode->i_size) {
1058 ext4_truncate_failed_write(inode);
1060 * If truncate failed early the inode might still be
1061 * on the orphan list; we need to make sure the inode
1062 * is removed from the orphan list in that case.
1065 ext4_orphan_del(NULL, inode);
1068 return ret ? ret : copied;
1072 * This is a private version of page_zero_new_buffers() which doesn't
1073 * set the buffer to be dirty, since in data=journalled mode we need
1074 * to call ext4_handle_dirty_metadata() instead.
1076 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1078 unsigned from, unsigned to)
1080 unsigned int block_start = 0, block_end;
1081 struct buffer_head *head, *bh;
1083 bh = head = page_buffers(page);
1085 block_end = block_start + bh->b_size;
1086 if (buffer_new(bh)) {
1087 if (block_end > from && block_start < to) {
1088 if (!PageUptodate(page)) {
1089 unsigned start, size;
1091 start = max(from, block_start);
1092 size = min(to, block_end) - start;
1094 zero_user(page, start, size);
1095 write_end_fn(handle, bh);
1097 clear_buffer_new(bh);
1100 block_start = block_end;
1101 bh = bh->b_this_page;
1102 } while (bh != head);
1105 static int ext4_journalled_write_end(struct file *file,
1106 struct address_space *mapping,
1107 loff_t pos, unsigned len, unsigned copied,
1108 struct page *page, void *fsdata)
1110 handle_t *handle = ext4_journal_current_handle();
1111 struct inode *inode = mapping->host;
1117 trace_ext4_journalled_write_end(inode, pos, len, copied);
1118 from = pos & (PAGE_CACHE_SIZE - 1);
1121 BUG_ON(!ext4_handle_valid(handle));
1123 if (unlikely(copied < len) && !PageUptodate(page)) {
1125 ext4_journalled_zero_new_buffers(handle, page, from, to);
1127 if (unlikely(copied < len))
1128 ext4_journalled_zero_new_buffers(handle, page,
1130 ret = walk_page_buffers(handle, page_buffers(page), from,
1131 from + copied, &partial,
1134 SetPageUptodate(page);
1136 new_i_size = pos + copied;
1137 if (new_i_size > inode->i_size)
1138 i_size_write(inode, pos+copied);
1139 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1140 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1141 if (new_i_size > EXT4_I(inode)->i_disksize) {
1142 ext4_update_i_disksize(inode, new_i_size);
1143 ret2 = ext4_mark_inode_dirty(handle, inode);
1149 page_cache_release(page);
1150 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1151 /* if we have allocated more blocks and copied
1152 * less. We will have blocks allocated outside
1153 * inode->i_size. So truncate them
1155 ext4_orphan_add(handle, inode);
1157 ret2 = ext4_journal_stop(handle);
1160 if (pos + len > inode->i_size) {
1161 ext4_truncate_failed_write(inode);
1163 * If truncate failed early the inode might still be
1164 * on the orphan list; we need to make sure the inode
1165 * is removed from the orphan list in that case.
1168 ext4_orphan_del(NULL, inode);
1171 return ret ? ret : copied;
1175 * Reserve a single cluster located at lblock
1177 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1180 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1181 struct ext4_inode_info *ei = EXT4_I(inode);
1182 unsigned int md_needed;
1184 ext4_lblk_t save_last_lblock;
1188 * We will charge metadata quota at writeout time; this saves
1189 * us from metadata over-estimation, though we may go over by
1190 * a small amount in the end. Here we just reserve for data.
1192 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1197 * recalculate the amount of metadata blocks to reserve
1198 * in order to allocate nrblocks
1199 * worse case is one extent per block
1202 spin_lock(&ei->i_block_reservation_lock);
1204 * ext4_calc_metadata_amount() has side effects, which we have
1205 * to be prepared undo if we fail to claim space.
1207 save_len = ei->i_da_metadata_calc_len;
1208 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1209 md_needed = EXT4_NUM_B2C(sbi,
1210 ext4_calc_metadata_amount(inode, lblock));
1211 trace_ext4_da_reserve_space(inode, md_needed);
1214 * We do still charge estimated metadata to the sb though;
1215 * we cannot afford to run out of free blocks.
1217 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1218 ei->i_da_metadata_calc_len = save_len;
1219 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1220 spin_unlock(&ei->i_block_reservation_lock);
1221 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1225 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1228 ei->i_reserved_data_blocks++;
1229 ei->i_reserved_meta_blocks += md_needed;
1230 spin_unlock(&ei->i_block_reservation_lock);
1232 return 0; /* success */
1235 static void ext4_da_release_space(struct inode *inode, int to_free)
1237 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1238 struct ext4_inode_info *ei = EXT4_I(inode);
1241 return; /* Nothing to release, exit */
1243 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1245 trace_ext4_da_release_space(inode, to_free);
1246 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1248 * if there aren't enough reserved blocks, then the
1249 * counter is messed up somewhere. Since this
1250 * function is called from invalidate page, it's
1251 * harmless to return without any action.
1253 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1254 "ino %lu, to_free %d with only %d reserved "
1255 "data blocks\n", inode->i_ino, to_free,
1256 ei->i_reserved_data_blocks);
1258 to_free = ei->i_reserved_data_blocks;
1260 ei->i_reserved_data_blocks -= to_free;
1262 if (ei->i_reserved_data_blocks == 0) {
1264 * We can release all of the reserved metadata blocks
1265 * only when we have written all of the delayed
1266 * allocation blocks.
1267 * Note that in case of bigalloc, i_reserved_meta_blocks,
1268 * i_reserved_data_blocks, etc. refer to number of clusters.
1270 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1271 ei->i_reserved_meta_blocks);
1272 ei->i_reserved_meta_blocks = 0;
1273 ei->i_da_metadata_calc_len = 0;
1276 /* update fs dirty data blocks counter */
1277 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1279 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1281 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1284 static void ext4_da_page_release_reservation(struct page *page,
1285 unsigned long offset)
1288 struct buffer_head *head, *bh;
1289 unsigned int curr_off = 0;
1290 struct inode *inode = page->mapping->host;
1291 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1294 head = page_buffers(page);
1297 unsigned int next_off = curr_off + bh->b_size;
1299 if ((offset <= curr_off) && (buffer_delay(bh))) {
1301 clear_buffer_delay(bh);
1302 clear_buffer_da_mapped(bh);
1304 curr_off = next_off;
1305 } while ((bh = bh->b_this_page) != head);
1307 /* If we have released all the blocks belonging to a cluster, then we
1308 * need to release the reserved space for that cluster. */
1309 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1310 while (num_clusters > 0) {
1312 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1313 ((num_clusters - 1) << sbi->s_cluster_bits);
1314 if (sbi->s_cluster_ratio == 1 ||
1315 !ext4_find_delalloc_cluster(inode, lblk, 1))
1316 ext4_da_release_space(inode, 1);
1323 * Delayed allocation stuff
1327 * mpage_da_submit_io - walks through extent of pages and try to write
1328 * them with writepage() call back
1330 * @mpd->inode: inode
1331 * @mpd->first_page: first page of the extent
1332 * @mpd->next_page: page after the last page of the extent
1334 * By the time mpage_da_submit_io() is called we expect all blocks
1335 * to be allocated. this may be wrong if allocation failed.
1337 * As pages are already locked by write_cache_pages(), we can't use it
1339 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1340 struct ext4_map_blocks *map)
1342 struct pagevec pvec;
1343 unsigned long index, end;
1344 int ret = 0, err, nr_pages, i;
1345 struct inode *inode = mpd->inode;
1346 struct address_space *mapping = inode->i_mapping;
1347 unsigned int len, block_start;
1348 struct buffer_head *bh, *page_bufs = NULL;
1349 int journal_data = ext4_should_journal_data(inode);
1350 sector_t pblock = 0, cur_logical = 0;
1351 struct ext4_io_submit io_submit;
1353 BUG_ON(mpd->next_page <= mpd->first_page);
1354 memset(&io_submit, 0, sizeof(io_submit));
1356 * We need to start from the first_page to the next_page - 1
1357 * to make sure we also write the mapped dirty buffer_heads.
1358 * If we look at mpd->b_blocknr we would only be looking
1359 * at the currently mapped buffer_heads.
1361 index = mpd->first_page;
1362 end = mpd->next_page - 1;
1364 pagevec_init(&pvec, 0);
1365 while (index <= end) {
1366 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1369 for (i = 0; i < nr_pages; i++) {
1370 int commit_write = 0, skip_page = 0;
1371 struct page *page = pvec.pages[i];
1372 loff_t size = i_size_read(inode);
1374 index = page->index;
1378 if (index == size >> PAGE_CACHE_SHIFT)
1379 len = size & ~PAGE_CACHE_MASK;
1381 len = PAGE_CACHE_SIZE;
1383 cur_logical = index << (PAGE_CACHE_SHIFT -
1385 pblock = map->m_pblk + (cur_logical -
1390 BUG_ON(!PageLocked(page));
1391 BUG_ON(PageWriteback(page));
1394 * If the page does not have buffers (for
1395 * whatever reason), try to create them using
1396 * __block_write_begin. If this fails,
1397 * skip the page and move on.
1399 if (!page_has_buffers(page)) {
1400 if (__block_write_begin(page, 0, len,
1401 noalloc_get_block_write)) {
1409 bh = page_bufs = page_buffers(page);
1414 if (map && (cur_logical >= map->m_lblk) &&
1415 (cur_logical <= (map->m_lblk +
1416 (map->m_len - 1)))) {
1417 if (buffer_delay(bh)) {
1418 clear_buffer_delay(bh);
1419 bh->b_blocknr = pblock;
1421 if (buffer_da_mapped(bh))
1422 clear_buffer_da_mapped(bh);
1423 if (buffer_unwritten(bh) ||
1425 BUG_ON(bh->b_blocknr != pblock);
1426 if (map->m_flags & EXT4_MAP_UNINIT)
1427 set_buffer_uninit(bh);
1428 clear_buffer_unwritten(bh);
1432 * skip page if block allocation undone and
1435 if (ext4_bh_delay_or_unwritten(NULL, bh))
1437 bh = bh->b_this_page;
1438 block_start += bh->b_size;
1441 } while (bh != page_bufs);
1446 clear_page_dirty_for_io(page);
1448 * We have to be very careful here! Nothing protects
1449 * writeback path against i_size changes and the page
1450 * can be writeably mapped into page tables. So an
1451 * application can be growing i_size and writing data
1452 * through mmap while writeback runs.
1453 * clear_page_dirty_for_io() write-protects our page in
1454 * page tables and the page cannot get written to again
1455 * until we release page lock. So only after
1456 * clear_page_dirty_for_io() we are safe to sample
1457 * i_size for ext4_bio_write_page() to zero-out tail of
1458 * the written page. We rely on the barrier provided by
1459 * TestClearPageDirty in clear_page_dirty_for_io() to
1460 * make sure i_size is really sampled only after page
1461 * tables are updated.
1463 if (size != i_size_read(inode)) {
1464 set_page_dirty(page);
1469 /* mark the buffer_heads as dirty & uptodate */
1470 block_commit_write(page, 0, len);
1472 * Delalloc doesn't support data journalling,
1473 * but eventually maybe we'll lift this
1476 if (unlikely(journal_data && PageChecked(page)))
1477 err = __ext4_journalled_writepage(page, len);
1478 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1479 err = ext4_bio_write_page(&io_submit, page,
1481 else if (buffer_uninit(page_bufs)) {
1482 ext4_set_bh_endio(page_bufs, inode);
1483 err = block_write_full_page_endio(page,
1484 noalloc_get_block_write,
1485 mpd->wbc, ext4_end_io_buffer_write);
1487 err = block_write_full_page(page,
1488 noalloc_get_block_write, mpd->wbc);
1491 mpd->pages_written++;
1493 * In error case, we have to continue because
1494 * remaining pages are still locked
1499 pagevec_release(&pvec);
1501 ext4_io_submit(&io_submit);
1505 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1509 struct pagevec pvec;
1510 struct inode *inode = mpd->inode;
1511 struct address_space *mapping = inode->i_mapping;
1513 index = mpd->first_page;
1514 end = mpd->next_page - 1;
1515 pagevec_init(&pvec, 0);
1516 while (index <= end) {
1517 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1520 for (i = 0; i < nr_pages; i++) {
1521 struct page *page = pvec.pages[i];
1522 if (page->index > end)
1524 BUG_ON(!PageLocked(page));
1525 BUG_ON(PageWriteback(page));
1526 block_invalidatepage(page, 0);
1527 ClearPageUptodate(page);
1530 index = pvec.pages[nr_pages - 1]->index + 1;
1531 pagevec_release(&pvec);
1536 static void ext4_print_free_blocks(struct inode *inode)
1538 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1539 printk(KERN_CRIT "Total free blocks count %lld\n",
1540 EXT4_C2B(EXT4_SB(inode->i_sb),
1541 ext4_count_free_clusters(inode->i_sb)));
1542 printk(KERN_CRIT "Free/Dirty block details\n");
1543 printk(KERN_CRIT "free_blocks=%lld\n",
1544 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1545 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1546 printk(KERN_CRIT "dirty_blocks=%lld\n",
1547 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1548 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1549 printk(KERN_CRIT "Block reservation details\n");
1550 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
1551 EXT4_I(inode)->i_reserved_data_blocks);
1552 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
1553 EXT4_I(inode)->i_reserved_meta_blocks);
1558 * mpage_da_map_and_submit - go through given space, map them
1559 * if necessary, and then submit them for I/O
1561 * @mpd - bh describing space
1563 * The function skips space we know is already mapped to disk blocks.
1566 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1568 int err, blks, get_blocks_flags;
1569 struct ext4_map_blocks map, *mapp = NULL;
1570 sector_t next = mpd->b_blocknr;
1571 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1572 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1573 handle_t *handle = NULL;
1576 * If the blocks are mapped already, or we couldn't accumulate
1577 * any blocks, then proceed immediately to the submission stage.
1579 if ((mpd->b_size == 0) ||
1580 ((mpd->b_state & (1 << BH_Mapped)) &&
1581 !(mpd->b_state & (1 << BH_Delay)) &&
1582 !(mpd->b_state & (1 << BH_Unwritten))))
1585 handle = ext4_journal_current_handle();
1589 * Call ext4_map_blocks() to allocate any delayed allocation
1590 * blocks, or to convert an uninitialized extent to be
1591 * initialized (in the case where we have written into
1592 * one or more preallocated blocks).
1594 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1595 * indicate that we are on the delayed allocation path. This
1596 * affects functions in many different parts of the allocation
1597 * call path. This flag exists primarily because we don't
1598 * want to change *many* call functions, so ext4_map_blocks()
1599 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1600 * inode's allocation semaphore is taken.
1602 * If the blocks in questions were delalloc blocks, set
1603 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1604 * variables are updated after the blocks have been allocated.
1607 map.m_len = max_blocks;
1608 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1609 if (ext4_should_dioread_nolock(mpd->inode))
1610 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1611 if (mpd->b_state & (1 << BH_Delay))
1612 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1614 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1616 struct super_block *sb = mpd->inode->i_sb;
1620 * If get block returns EAGAIN or ENOSPC and there
1621 * appears to be free blocks we will just let
1622 * mpage_da_submit_io() unlock all of the pages.
1627 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1633 * get block failure will cause us to loop in
1634 * writepages, because a_ops->writepage won't be able
1635 * to make progress. The page will be redirtied by
1636 * writepage and writepages will again try to write
1639 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1640 ext4_msg(sb, KERN_CRIT,
1641 "delayed block allocation failed for inode %lu "
1642 "at logical offset %llu with max blocks %zd "
1643 "with error %d", mpd->inode->i_ino,
1644 (unsigned long long) next,
1645 mpd->b_size >> mpd->inode->i_blkbits, err);
1646 ext4_msg(sb, KERN_CRIT,
1647 "This should not happen!! Data will be lost\n");
1649 ext4_print_free_blocks(mpd->inode);
1651 /* invalidate all the pages */
1652 ext4_da_block_invalidatepages(mpd);
1654 /* Mark this page range as having been completed */
1661 if (map.m_flags & EXT4_MAP_NEW) {
1662 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1665 for (i = 0; i < map.m_len; i++)
1666 unmap_underlying_metadata(bdev, map.m_pblk + i);
1668 if (ext4_should_order_data(mpd->inode)) {
1669 err = ext4_jbd2_file_inode(handle, mpd->inode);
1671 /* Only if the journal is aborted */
1679 * Update on-disk size along with block allocation.
1681 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1682 if (disksize > i_size_read(mpd->inode))
1683 disksize = i_size_read(mpd->inode);
1684 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1685 ext4_update_i_disksize(mpd->inode, disksize);
1686 err = ext4_mark_inode_dirty(handle, mpd->inode);
1688 ext4_error(mpd->inode->i_sb,
1689 "Failed to mark inode %lu dirty",
1694 mpage_da_submit_io(mpd, mapp);
1698 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1699 (1 << BH_Delay) | (1 << BH_Unwritten))
1702 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1704 * @mpd->lbh - extent of blocks
1705 * @logical - logical number of the block in the file
1706 * @bh - bh of the block (used to access block's state)
1708 * the function is used to collect contig. blocks in same state
1710 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1711 sector_t logical, size_t b_size,
1712 unsigned long b_state)
1715 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1718 * XXX Don't go larger than mballoc is willing to allocate
1719 * This is a stopgap solution. We eventually need to fold
1720 * mpage_da_submit_io() into this function and then call
1721 * ext4_map_blocks() multiple times in a loop
1723 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1726 /* check if thereserved journal credits might overflow */
1727 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1728 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1730 * With non-extent format we are limited by the journal
1731 * credit available. Total credit needed to insert
1732 * nrblocks contiguous blocks is dependent on the
1733 * nrblocks. So limit nrblocks.
1736 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1737 EXT4_MAX_TRANS_DATA) {
1739 * Adding the new buffer_head would make it cross the
1740 * allowed limit for which we have journal credit
1741 * reserved. So limit the new bh->b_size
1743 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1744 mpd->inode->i_blkbits;
1745 /* we will do mpage_da_submit_io in the next loop */
1749 * First block in the extent
1751 if (mpd->b_size == 0) {
1752 mpd->b_blocknr = logical;
1753 mpd->b_size = b_size;
1754 mpd->b_state = b_state & BH_FLAGS;
1758 next = mpd->b_blocknr + nrblocks;
1760 * Can we merge the block to our big extent?
1762 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1763 mpd->b_size += b_size;
1769 * We couldn't merge the block to our extent, so we
1770 * need to flush current extent and start new one
1772 mpage_da_map_and_submit(mpd);
1776 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1778 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1782 * This function is grabs code from the very beginning of
1783 * ext4_map_blocks, but assumes that the caller is from delayed write
1784 * time. This function looks up the requested blocks and sets the
1785 * buffer delay bit under the protection of i_data_sem.
1787 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1788 struct ext4_map_blocks *map,
1789 struct buffer_head *bh)
1792 sector_t invalid_block = ~((sector_t) 0xffff);
1794 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1798 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1799 "logical block %lu\n", inode->i_ino, map->m_len,
1800 (unsigned long) map->m_lblk);
1802 * Try to see if we can get the block without requesting a new
1803 * file system block.
1805 down_read((&EXT4_I(inode)->i_data_sem));
1806 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1807 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1809 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1813 * XXX: __block_prepare_write() unmaps passed block,
1816 /* If the block was allocated from previously allocated cluster,
1817 * then we dont need to reserve it again. */
1818 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1819 retval = ext4_da_reserve_space(inode, iblock);
1821 /* not enough space to reserve */
1825 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1826 * and it should not appear on the bh->b_state.
1828 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1830 map_bh(bh, inode->i_sb, invalid_block);
1832 set_buffer_delay(bh);
1836 up_read((&EXT4_I(inode)->i_data_sem));
1842 * This is a special get_blocks_t callback which is used by
1843 * ext4_da_write_begin(). It will either return mapped block or
1844 * reserve space for a single block.
1846 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1847 * We also have b_blocknr = -1 and b_bdev initialized properly
1849 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1850 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1851 * initialized properly.
1853 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1854 struct buffer_head *bh, int create)
1856 struct ext4_map_blocks map;
1859 BUG_ON(create == 0);
1860 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1862 map.m_lblk = iblock;
1866 * first, we need to know whether the block is allocated already
1867 * preallocated blocks are unmapped but should treated
1868 * the same as allocated blocks.
1870 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1874 map_bh(bh, inode->i_sb, map.m_pblk);
1875 ext4_update_bh_state(bh, map.m_flags);
1877 if (buffer_unwritten(bh)) {
1878 /* A delayed write to unwritten bh should be marked
1879 * new and mapped. Mapped ensures that we don't do
1880 * get_block multiple times when we write to the same
1881 * offset and new ensures that we do proper zero out
1882 * for partial write.
1885 set_buffer_mapped(bh);
1891 * This function is used as a standard get_block_t calback function
1892 * when there is no desire to allocate any blocks. It is used as a
1893 * callback function for block_write_begin() and block_write_full_page().
1894 * These functions should only try to map a single block at a time.
1896 * Since this function doesn't do block allocations even if the caller
1897 * requests it by passing in create=1, it is critically important that
1898 * any caller checks to make sure that any buffer heads are returned
1899 * by this function are either all already mapped or marked for
1900 * delayed allocation before calling block_write_full_page(). Otherwise,
1901 * b_blocknr could be left unitialized, and the page write functions will
1902 * be taken by surprise.
1904 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1905 struct buffer_head *bh_result, int create)
1907 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1908 return _ext4_get_block(inode, iblock, bh_result, 0);
1911 static int bget_one(handle_t *handle, struct buffer_head *bh)
1917 static int bput_one(handle_t *handle, struct buffer_head *bh)
1923 static int __ext4_journalled_writepage(struct page *page,
1926 struct address_space *mapping = page->mapping;
1927 struct inode *inode = mapping->host;
1928 struct buffer_head *page_bufs;
1929 handle_t *handle = NULL;
1933 ClearPageChecked(page);
1934 page_bufs = page_buffers(page);
1936 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1938 * We need to release the page lock before we start the
1939 * journal, so grab a reference so the page won't disappear
1940 * out from under us.
1945 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1946 if (IS_ERR(handle)) {
1947 ret = PTR_ERR(handle);
1949 goto out_no_pagelock;
1951 BUG_ON(!ext4_handle_valid(handle));
1955 if (page->mapping != mapping) {
1956 /* The page got truncated from under us */
1957 ext4_journal_stop(handle);
1962 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1963 do_journal_get_write_access);
1965 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1969 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1970 err = ext4_journal_stop(handle);
1974 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
1975 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1982 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
1983 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
1986 * Note that we don't need to start a transaction unless we're journaling data
1987 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1988 * need to file the inode to the transaction's list in ordered mode because if
1989 * we are writing back data added by write(), the inode is already there and if
1990 * we are writing back data modified via mmap(), no one guarantees in which
1991 * transaction the data will hit the disk. In case we are journaling data, we
1992 * cannot start transaction directly because transaction start ranks above page
1993 * lock so we have to do some magic.
1995 * This function can get called via...
1996 * - ext4_da_writepages after taking page lock (have journal handle)
1997 * - journal_submit_inode_data_buffers (no journal handle)
1998 * - shrink_page_list via pdflush (no journal handle)
1999 * - grab_page_cache when doing write_begin (have journal handle)
2001 * We don't do any block allocation in this function. If we have page with
2002 * multiple blocks we need to write those buffer_heads that are mapped. This
2003 * is important for mmaped based write. So if we do with blocksize 1K
2004 * truncate(f, 1024);
2005 * a = mmap(f, 0, 4096);
2007 * truncate(f, 4096);
2008 * we have in the page first buffer_head mapped via page_mkwrite call back
2009 * but other bufer_heads would be unmapped but dirty(dirty done via the
2010 * do_wp_page). So writepage should write the first block. If we modify
2011 * the mmap area beyond 1024 we will again get a page_fault and the
2012 * page_mkwrite callback will do the block allocation and mark the
2013 * buffer_heads mapped.
2015 * We redirty the page if we have any buffer_heads that is either delay or
2016 * unwritten in the page.
2018 * We can get recursively called as show below.
2020 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2023 * But since we don't do any block allocation we should not deadlock.
2024 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2026 static int ext4_writepage(struct page *page,
2027 struct writeback_control *wbc)
2029 int ret = 0, commit_write = 0;
2032 struct buffer_head *page_bufs = NULL;
2033 struct inode *inode = page->mapping->host;
2035 trace_ext4_writepage(page);
2036 size = i_size_read(inode);
2037 if (page->index == size >> PAGE_CACHE_SHIFT)
2038 len = size & ~PAGE_CACHE_MASK;
2040 len = PAGE_CACHE_SIZE;
2043 * If the page does not have buffers (for whatever reason),
2044 * try to create them using __block_write_begin. If this
2045 * fails, redirty the page and move on.
2047 if (!page_has_buffers(page)) {
2048 if (__block_write_begin(page, 0, len,
2049 noalloc_get_block_write)) {
2051 redirty_page_for_writepage(wbc, page);
2057 page_bufs = page_buffers(page);
2058 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2059 ext4_bh_delay_or_unwritten)) {
2061 * We don't want to do block allocation, so redirty
2062 * the page and return. We may reach here when we do
2063 * a journal commit via journal_submit_inode_data_buffers.
2064 * We can also reach here via shrink_page_list but it
2065 * should never be for direct reclaim so warn if that
2068 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
2073 /* now mark the buffer_heads as dirty and uptodate */
2074 block_commit_write(page, 0, len);
2076 if (PageChecked(page) && ext4_should_journal_data(inode))
2078 * It's mmapped pagecache. Add buffers and journal it. There
2079 * doesn't seem much point in redirtying the page here.
2081 return __ext4_journalled_writepage(page, len);
2083 if (buffer_uninit(page_bufs)) {
2084 ext4_set_bh_endio(page_bufs, inode);
2085 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2086 wbc, ext4_end_io_buffer_write);
2088 ret = block_write_full_page(page, noalloc_get_block_write,
2095 * This is called via ext4_da_writepages() to
2096 * calculate the total number of credits to reserve to fit
2097 * a single extent allocation into a single transaction,
2098 * ext4_da_writpeages() will loop calling this before
2099 * the block allocation.
2102 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2104 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2107 * With non-extent format the journal credit needed to
2108 * insert nrblocks contiguous block is dependent on
2109 * number of contiguous block. So we will limit
2110 * number of contiguous block to a sane value
2112 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2113 (max_blocks > EXT4_MAX_TRANS_DATA))
2114 max_blocks = EXT4_MAX_TRANS_DATA;
2116 return ext4_chunk_trans_blocks(inode, max_blocks);
2120 * write_cache_pages_da - walk the list of dirty pages of the given
2121 * address space and accumulate pages that need writing, and call
2122 * mpage_da_map_and_submit to map a single contiguous memory region
2123 * and then write them.
2125 static int write_cache_pages_da(struct address_space *mapping,
2126 struct writeback_control *wbc,
2127 struct mpage_da_data *mpd,
2128 pgoff_t *done_index)
2130 struct buffer_head *bh, *head;
2131 struct inode *inode = mapping->host;
2132 struct pagevec pvec;
2133 unsigned int nr_pages;
2136 long nr_to_write = wbc->nr_to_write;
2137 int i, tag, ret = 0;
2139 memset(mpd, 0, sizeof(struct mpage_da_data));
2142 pagevec_init(&pvec, 0);
2143 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2144 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2146 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2147 tag = PAGECACHE_TAG_TOWRITE;
2149 tag = PAGECACHE_TAG_DIRTY;
2151 *done_index = index;
2152 while (index <= end) {
2153 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2154 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2158 for (i = 0; i < nr_pages; i++) {
2159 struct page *page = pvec.pages[i];
2162 * At this point, the page may be truncated or
2163 * invalidated (changing page->mapping to NULL), or
2164 * even swizzled back from swapper_space to tmpfs file
2165 * mapping. However, page->index will not change
2166 * because we have a reference on the page.
2168 if (page->index > end)
2171 *done_index = page->index + 1;
2174 * If we can't merge this page, and we have
2175 * accumulated an contiguous region, write it
2177 if ((mpd->next_page != page->index) &&
2178 (mpd->next_page != mpd->first_page)) {
2179 mpage_da_map_and_submit(mpd);
2180 goto ret_extent_tail;
2186 * If the page is no longer dirty, or its
2187 * mapping no longer corresponds to inode we
2188 * are writing (which means it has been
2189 * truncated or invalidated), or the page is
2190 * already under writeback and we are not
2191 * doing a data integrity writeback, skip the page
2193 if (!PageDirty(page) ||
2194 (PageWriteback(page) &&
2195 (wbc->sync_mode == WB_SYNC_NONE)) ||
2196 unlikely(page->mapping != mapping)) {
2201 wait_on_page_writeback(page);
2202 BUG_ON(PageWriteback(page));
2204 if (mpd->next_page != page->index)
2205 mpd->first_page = page->index;
2206 mpd->next_page = page->index + 1;
2207 logical = (sector_t) page->index <<
2208 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2210 if (!page_has_buffers(page)) {
2211 mpage_add_bh_to_extent(mpd, logical,
2213 (1 << BH_Dirty) | (1 << BH_Uptodate));
2215 goto ret_extent_tail;
2218 * Page with regular buffer heads,
2219 * just add all dirty ones
2221 head = page_buffers(page);
2224 BUG_ON(buffer_locked(bh));
2226 * We need to try to allocate
2227 * unmapped blocks in the same page.
2228 * Otherwise we won't make progress
2229 * with the page in ext4_writepage
2231 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2232 mpage_add_bh_to_extent(mpd, logical,
2236 goto ret_extent_tail;
2237 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2239 * mapped dirty buffer. We need
2240 * to update the b_state
2241 * because we look at b_state
2242 * in mpage_da_map_blocks. We
2243 * don't update b_size because
2244 * if we find an unmapped
2245 * buffer_head later we need to
2246 * use the b_state flag of that
2249 if (mpd->b_size == 0)
2250 mpd->b_state = bh->b_state & BH_FLAGS;
2253 } while ((bh = bh->b_this_page) != head);
2256 if (nr_to_write > 0) {
2258 if (nr_to_write == 0 &&
2259 wbc->sync_mode == WB_SYNC_NONE)
2261 * We stop writing back only if we are
2262 * not doing integrity sync. In case of
2263 * integrity sync we have to keep going
2264 * because someone may be concurrently
2265 * dirtying pages, and we might have
2266 * synced a lot of newly appeared dirty
2267 * pages, but have not synced all of the
2273 pagevec_release(&pvec);
2278 ret = MPAGE_DA_EXTENT_TAIL;
2280 pagevec_release(&pvec);
2286 static int ext4_da_writepages(struct address_space *mapping,
2287 struct writeback_control *wbc)
2290 int range_whole = 0;
2291 handle_t *handle = NULL;
2292 struct mpage_da_data mpd;
2293 struct inode *inode = mapping->host;
2294 int pages_written = 0;
2295 unsigned int max_pages;
2296 int range_cyclic, cycled = 1, io_done = 0;
2297 int needed_blocks, ret = 0;
2298 long desired_nr_to_write, nr_to_writebump = 0;
2299 loff_t range_start = wbc->range_start;
2300 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2301 pgoff_t done_index = 0;
2303 struct blk_plug plug;
2305 trace_ext4_da_writepages(inode, wbc);
2308 * No pages to write? This is mainly a kludge to avoid starting
2309 * a transaction for special inodes like journal inode on last iput()
2310 * because that could violate lock ordering on umount
2312 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2316 * If the filesystem has aborted, it is read-only, so return
2317 * right away instead of dumping stack traces later on that
2318 * will obscure the real source of the problem. We test
2319 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2320 * the latter could be true if the filesystem is mounted
2321 * read-only, and in that case, ext4_da_writepages should
2322 * *never* be called, so if that ever happens, we would want
2325 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2328 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2331 range_cyclic = wbc->range_cyclic;
2332 if (wbc->range_cyclic) {
2333 index = mapping->writeback_index;
2336 wbc->range_start = index << PAGE_CACHE_SHIFT;
2337 wbc->range_end = LLONG_MAX;
2338 wbc->range_cyclic = 0;
2341 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2342 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2346 * This works around two forms of stupidity. The first is in
2347 * the writeback code, which caps the maximum number of pages
2348 * written to be 1024 pages. This is wrong on multiple
2349 * levels; different architectues have a different page size,
2350 * which changes the maximum amount of data which gets
2351 * written. Secondly, 4 megabytes is way too small. XFS
2352 * forces this value to be 16 megabytes by multiplying
2353 * nr_to_write parameter by four, and then relies on its
2354 * allocator to allocate larger extents to make them
2355 * contiguous. Unfortunately this brings us to the second
2356 * stupidity, which is that ext4's mballoc code only allocates
2357 * at most 2048 blocks. So we force contiguous writes up to
2358 * the number of dirty blocks in the inode, or
2359 * sbi->max_writeback_mb_bump whichever is smaller.
2361 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2362 if (!range_cyclic && range_whole) {
2363 if (wbc->nr_to_write == LONG_MAX)
2364 desired_nr_to_write = wbc->nr_to_write;
2366 desired_nr_to_write = wbc->nr_to_write * 8;
2368 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2370 if (desired_nr_to_write > max_pages)
2371 desired_nr_to_write = max_pages;
2373 if (wbc->nr_to_write < desired_nr_to_write) {
2374 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2375 wbc->nr_to_write = desired_nr_to_write;
2379 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2380 tag_pages_for_writeback(mapping, index, end);
2382 blk_start_plug(&plug);
2383 while (!ret && wbc->nr_to_write > 0) {
2386 * we insert one extent at a time. So we need
2387 * credit needed for single extent allocation.
2388 * journalled mode is currently not supported
2391 BUG_ON(ext4_should_journal_data(inode));
2392 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2394 /* start a new transaction*/
2395 handle = ext4_journal_start(inode, needed_blocks);
2396 if (IS_ERR(handle)) {
2397 ret = PTR_ERR(handle);
2398 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2399 "%ld pages, ino %lu; err %d", __func__,
2400 wbc->nr_to_write, inode->i_ino, ret);
2401 blk_finish_plug(&plug);
2402 goto out_writepages;
2406 * Now call write_cache_pages_da() to find the next
2407 * contiguous region of logical blocks that need
2408 * blocks to be allocated by ext4 and submit them.
2410 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
2412 * If we have a contiguous extent of pages and we
2413 * haven't done the I/O yet, map the blocks and submit
2416 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2417 mpage_da_map_and_submit(&mpd);
2418 ret = MPAGE_DA_EXTENT_TAIL;
2420 trace_ext4_da_write_pages(inode, &mpd);
2421 wbc->nr_to_write -= mpd.pages_written;
2423 ext4_journal_stop(handle);
2425 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2426 /* commit the transaction which would
2427 * free blocks released in the transaction
2430 jbd2_journal_force_commit_nested(sbi->s_journal);
2432 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2434 * Got one extent now try with rest of the pages.
2435 * If mpd.retval is set -EIO, journal is aborted.
2436 * So we don't need to write any more.
2438 pages_written += mpd.pages_written;
2441 } else if (wbc->nr_to_write)
2443 * There is no more writeout needed
2444 * or we requested for a noblocking writeout
2445 * and we found the device congested
2449 blk_finish_plug(&plug);
2450 if (!io_done && !cycled) {
2453 wbc->range_start = index << PAGE_CACHE_SHIFT;
2454 wbc->range_end = mapping->writeback_index - 1;
2459 wbc->range_cyclic = range_cyclic;
2460 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2462 * set the writeback_index so that range_cyclic
2463 * mode will write it back later
2465 mapping->writeback_index = done_index;
2468 wbc->nr_to_write -= nr_to_writebump;
2469 wbc->range_start = range_start;
2470 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2474 #define FALL_BACK_TO_NONDELALLOC 1
2475 static int ext4_nonda_switch(struct super_block *sb)
2477 s64 free_blocks, dirty_blocks;
2478 struct ext4_sb_info *sbi = EXT4_SB(sb);
2481 * switch to non delalloc mode if we are running low
2482 * on free block. The free block accounting via percpu
2483 * counters can get slightly wrong with percpu_counter_batch getting
2484 * accumulated on each CPU without updating global counters
2485 * Delalloc need an accurate free block accounting. So switch
2486 * to non delalloc when we are near to error range.
2488 free_blocks = EXT4_C2B(sbi,
2489 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2490 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2492 * Start pushing delalloc when 1/2 of free blocks are dirty.
2494 if (dirty_blocks && (free_blocks < 2 * dirty_blocks) &&
2495 !writeback_in_progress(sb->s_bdi) &&
2496 down_read_trylock(&sb->s_umount)) {
2497 writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2498 up_read(&sb->s_umount);
2501 if (2 * free_blocks < 3 * dirty_blocks ||
2502 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2504 * free block count is less than 150% of dirty blocks
2505 * or free blocks is less than watermark
2512 /* We always reserve for an inode update; the superblock could be there too */
2513 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2515 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
2516 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2519 if (pos + len <= 0x7fffffffULL)
2522 /* We might need to update the superblock to set LARGE_FILE */
2526 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2527 loff_t pos, unsigned len, unsigned flags,
2528 struct page **pagep, void **fsdata)
2530 int ret, retries = 0;
2533 struct inode *inode = mapping->host;
2536 index = pos >> PAGE_CACHE_SHIFT;
2538 if (ext4_nonda_switch(inode->i_sb)) {
2539 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2540 return ext4_write_begin(file, mapping, pos,
2541 len, flags, pagep, fsdata);
2543 *fsdata = (void *)0;
2544 trace_ext4_da_write_begin(inode, pos, len, flags);
2547 * With delayed allocation, we don't log the i_disksize update
2548 * if there is delayed block allocation. But we still need
2549 * to journalling the i_disksize update if writes to the end
2550 * of file which has an already mapped buffer.
2552 handle = ext4_journal_start(inode,
2553 ext4_da_write_credits(inode, pos, len));
2554 if (IS_ERR(handle)) {
2555 ret = PTR_ERR(handle);
2558 /* We cannot recurse into the filesystem as the transaction is already
2560 flags |= AOP_FLAG_NOFS;
2562 page = grab_cache_page_write_begin(mapping, index, flags);
2564 ext4_journal_stop(handle);
2570 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2573 ext4_journal_stop(handle);
2574 page_cache_release(page);
2576 * block_write_begin may have instantiated a few blocks
2577 * outside i_size. Trim these off again. Don't need
2578 * i_size_read because we hold i_mutex.
2580 if (pos + len > inode->i_size)
2581 ext4_truncate_failed_write(inode);
2584 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2591 * Check if we should update i_disksize
2592 * when write to the end of file but not require block allocation
2594 static int ext4_da_should_update_i_disksize(struct page *page,
2595 unsigned long offset)
2597 struct buffer_head *bh;
2598 struct inode *inode = page->mapping->host;
2602 bh = page_buffers(page);
2603 idx = offset >> inode->i_blkbits;
2605 for (i = 0; i < idx; i++)
2606 bh = bh->b_this_page;
2608 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2613 static int ext4_da_write_end(struct file *file,
2614 struct address_space *mapping,
2615 loff_t pos, unsigned len, unsigned copied,
2616 struct page *page, void *fsdata)
2618 struct inode *inode = mapping->host;
2620 handle_t *handle = ext4_journal_current_handle();
2622 unsigned long start, end;
2623 int write_mode = (int)(unsigned long)fsdata;
2625 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2626 switch (ext4_inode_journal_mode(inode)) {
2627 case EXT4_INODE_ORDERED_DATA_MODE:
2628 return ext4_ordered_write_end(file, mapping, pos,
2629 len, copied, page, fsdata);
2630 case EXT4_INODE_WRITEBACK_DATA_MODE:
2631 return ext4_writeback_write_end(file, mapping, pos,
2632 len, copied, page, fsdata);
2638 trace_ext4_da_write_end(inode, pos, len, copied);
2639 start = pos & (PAGE_CACHE_SIZE - 1);
2640 end = start + copied - 1;
2643 * generic_write_end() will run mark_inode_dirty() if i_size
2644 * changes. So let's piggyback the i_disksize mark_inode_dirty
2648 new_i_size = pos + copied;
2649 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2650 if (ext4_da_should_update_i_disksize(page, end)) {
2651 down_write(&EXT4_I(inode)->i_data_sem);
2652 if (new_i_size > EXT4_I(inode)->i_disksize) {
2654 * Updating i_disksize when extending file
2655 * without needing block allocation
2657 if (ext4_should_order_data(inode))
2658 ret = ext4_jbd2_file_inode(handle,
2661 EXT4_I(inode)->i_disksize = new_i_size;
2663 up_write(&EXT4_I(inode)->i_data_sem);
2664 /* We need to mark inode dirty even if
2665 * new_i_size is less that inode->i_size
2666 * bu greater than i_disksize.(hint delalloc)
2668 ext4_mark_inode_dirty(handle, inode);
2671 ret2 = generic_write_end(file, mapping, pos, len, copied,
2676 ret2 = ext4_journal_stop(handle);
2680 return ret ? ret : copied;
2683 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2686 * Drop reserved blocks
2688 BUG_ON(!PageLocked(page));
2689 if (!page_has_buffers(page))
2692 ext4_da_page_release_reservation(page, offset);
2695 ext4_invalidatepage(page, offset);
2701 * Force all delayed allocation blocks to be allocated for a given inode.
2703 int ext4_alloc_da_blocks(struct inode *inode)
2705 trace_ext4_alloc_da_blocks(inode);
2707 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2708 !EXT4_I(inode)->i_reserved_meta_blocks)
2712 * We do something simple for now. The filemap_flush() will
2713 * also start triggering a write of the data blocks, which is
2714 * not strictly speaking necessary (and for users of
2715 * laptop_mode, not even desirable). However, to do otherwise
2716 * would require replicating code paths in:
2718 * ext4_da_writepages() ->
2719 * write_cache_pages() ---> (via passed in callback function)
2720 * __mpage_da_writepage() -->
2721 * mpage_add_bh_to_extent()
2722 * mpage_da_map_blocks()
2724 * The problem is that write_cache_pages(), located in
2725 * mm/page-writeback.c, marks pages clean in preparation for
2726 * doing I/O, which is not desirable if we're not planning on
2729 * We could call write_cache_pages(), and then redirty all of
2730 * the pages by calling redirty_page_for_writepage() but that
2731 * would be ugly in the extreme. So instead we would need to
2732 * replicate parts of the code in the above functions,
2733 * simplifying them because we wouldn't actually intend to
2734 * write out the pages, but rather only collect contiguous
2735 * logical block extents, call the multi-block allocator, and
2736 * then update the buffer heads with the block allocations.
2738 * For now, though, we'll cheat by calling filemap_flush(),
2739 * which will map the blocks, and start the I/O, but not
2740 * actually wait for the I/O to complete.
2742 return filemap_flush(inode->i_mapping);
2746 * bmap() is special. It gets used by applications such as lilo and by
2747 * the swapper to find the on-disk block of a specific piece of data.
2749 * Naturally, this is dangerous if the block concerned is still in the
2750 * journal. If somebody makes a swapfile on an ext4 data-journaling
2751 * filesystem and enables swap, then they may get a nasty shock when the
2752 * data getting swapped to that swapfile suddenly gets overwritten by
2753 * the original zero's written out previously to the journal and
2754 * awaiting writeback in the kernel's buffer cache.
2756 * So, if we see any bmap calls here on a modified, data-journaled file,
2757 * take extra steps to flush any blocks which might be in the cache.
2759 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2761 struct inode *inode = mapping->host;
2765 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2766 test_opt(inode->i_sb, DELALLOC)) {
2768 * With delalloc we want to sync the file
2769 * so that we can make sure we allocate
2772 filemap_write_and_wait(mapping);
2775 if (EXT4_JOURNAL(inode) &&
2776 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2778 * This is a REALLY heavyweight approach, but the use of
2779 * bmap on dirty files is expected to be extremely rare:
2780 * only if we run lilo or swapon on a freshly made file
2781 * do we expect this to happen.
2783 * (bmap requires CAP_SYS_RAWIO so this does not
2784 * represent an unprivileged user DOS attack --- we'd be
2785 * in trouble if mortal users could trigger this path at
2788 * NB. EXT4_STATE_JDATA is not set on files other than
2789 * regular files. If somebody wants to bmap a directory
2790 * or symlink and gets confused because the buffer
2791 * hasn't yet been flushed to disk, they deserve
2792 * everything they get.
2795 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2796 journal = EXT4_JOURNAL(inode);
2797 jbd2_journal_lock_updates(journal);
2798 err = jbd2_journal_flush(journal);
2799 jbd2_journal_unlock_updates(journal);
2805 return generic_block_bmap(mapping, block, ext4_get_block);
2808 static int ext4_readpage(struct file *file, struct page *page)
2810 trace_ext4_readpage(page);
2811 return mpage_readpage(page, ext4_get_block);
2815 ext4_readpages(struct file *file, struct address_space *mapping,
2816 struct list_head *pages, unsigned nr_pages)
2818 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2821 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
2823 struct buffer_head *head, *bh;
2824 unsigned int curr_off = 0;
2826 if (!page_has_buffers(page))
2828 head = bh = page_buffers(page);
2830 if (offset <= curr_off && test_clear_buffer_uninit(bh)
2832 ext4_free_io_end(bh->b_private);
2833 bh->b_private = NULL;
2834 bh->b_end_io = NULL;
2836 curr_off = curr_off + bh->b_size;
2837 bh = bh->b_this_page;
2838 } while (bh != head);
2841 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2843 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2845 trace_ext4_invalidatepage(page, offset);
2848 * free any io_end structure allocated for buffers to be discarded
2850 if (ext4_should_dioread_nolock(page->mapping->host))
2851 ext4_invalidatepage_free_endio(page, offset);
2853 * If it's a full truncate we just forget about the pending dirtying
2856 ClearPageChecked(page);
2859 jbd2_journal_invalidatepage(journal, page, offset);
2861 block_invalidatepage(page, offset);
2864 static int ext4_releasepage(struct page *page, gfp_t wait)
2866 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2868 trace_ext4_releasepage(page);
2870 WARN_ON(PageChecked(page));
2871 if (!page_has_buffers(page))
2874 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2876 return try_to_free_buffers(page);
2880 * ext4_get_block used when preparing for a DIO write or buffer write.
2881 * We allocate an uinitialized extent if blocks haven't been allocated.
2882 * The extent will be converted to initialized after the IO is complete.
2884 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
2885 struct buffer_head *bh_result, int create)
2887 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2888 inode->i_ino, create);
2889 return _ext4_get_block(inode, iblock, bh_result,
2890 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2893 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2894 ssize_t size, void *private, int ret,
2897 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2898 ext4_io_end_t *io_end = iocb->private;
2899 struct workqueue_struct *wq;
2900 unsigned long flags;
2901 struct ext4_inode_info *ei;
2903 /* if not async direct IO or dio with 0 bytes write, just return */
2904 if (!io_end || !size)
2907 ext_debug("ext4_end_io_dio(): io_end 0x%p"
2908 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
2909 iocb->private, io_end->inode->i_ino, iocb, offset,
2912 iocb->private = NULL;
2914 /* if not aio dio with unwritten extents, just free io and return */
2915 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2916 ext4_free_io_end(io_end);
2918 inode_dio_done(inode);
2920 aio_complete(iocb, ret, 0);
2924 io_end->offset = offset;
2925 io_end->size = size;
2927 io_end->iocb = iocb;
2928 io_end->result = ret;
2930 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
2932 /* Add the io_end to per-inode completed aio dio list*/
2933 ei = EXT4_I(io_end->inode);
2934 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
2935 list_add_tail(&io_end->list, &ei->i_completed_io_list);
2936 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
2938 /* queue the work to convert unwritten extents to written */
2939 queue_work(wq, &io_end->work);
2942 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
2944 ext4_io_end_t *io_end = bh->b_private;
2945 struct workqueue_struct *wq;
2946 struct inode *inode;
2947 unsigned long flags;
2949 if (!test_clear_buffer_uninit(bh) || !io_end)
2952 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2953 printk("sb umounted, discard end_io request for inode %lu\n",
2954 io_end->inode->i_ino);
2955 ext4_free_io_end(io_end);
2960 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2961 * but being more careful is always safe for the future change.
2963 inode = io_end->inode;
2964 ext4_set_io_unwritten_flag(inode, io_end);
2966 /* Add the io_end to per-inode completed io list*/
2967 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
2968 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
2969 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
2971 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
2972 /* queue the work to convert unwritten extents to written */
2973 queue_work(wq, &io_end->work);
2975 bh->b_private = NULL;
2976 bh->b_end_io = NULL;
2977 clear_buffer_uninit(bh);
2978 end_buffer_async_write(bh, uptodate);
2981 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2983 ext4_io_end_t *io_end;
2984 struct page *page = bh->b_page;
2985 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2986 size_t size = bh->b_size;
2989 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2991 pr_warn_ratelimited("%s: allocation fail\n", __func__);
2995 io_end->offset = offset;
2996 io_end->size = size;
2998 * We need to hold a reference to the page to make sure it
2999 * doesn't get evicted before ext4_end_io_work() has a chance
3000 * to convert the extent from written to unwritten.
3002 io_end->page = page;
3003 get_page(io_end->page);
3005 bh->b_private = io_end;
3006 bh->b_end_io = ext4_end_io_buffer_write;
3011 * For ext4 extent files, ext4 will do direct-io write to holes,
3012 * preallocated extents, and those write extend the file, no need to
3013 * fall back to buffered IO.
3015 * For holes, we fallocate those blocks, mark them as uninitialized
3016 * If those blocks were preallocated, we mark sure they are splited, but
3017 * still keep the range to write as uninitialized.
3019 * The unwrritten extents will be converted to written when DIO is completed.
3020 * For async direct IO, since the IO may still pending when return, we
3021 * set up an end_io call back function, which will do the conversion
3022 * when async direct IO completed.
3024 * If the O_DIRECT write will extend the file then add this inode to the
3025 * orphan list. So recovery will truncate it back to the original size
3026 * if the machine crashes during the write.
3029 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3030 const struct iovec *iov, loff_t offset,
3031 unsigned long nr_segs)
3033 struct file *file = iocb->ki_filp;
3034 struct inode *inode = file->f_mapping->host;
3036 size_t count = iov_length(iov, nr_segs);
3038 loff_t final_size = offset + count;
3039 if (rw == WRITE && final_size <= inode->i_size) {
3041 * We could direct write to holes and fallocate.
3043 * Allocated blocks to fill the hole are marked as uninitialized
3044 * to prevent parallel buffered read to expose the stale data
3045 * before DIO complete the data IO.
3047 * As to previously fallocated extents, ext4 get_block
3048 * will just simply mark the buffer mapped but still
3049 * keep the extents uninitialized.
3051 * for non AIO case, we will convert those unwritten extents
3052 * to written after return back from blockdev_direct_IO.
3054 * for async DIO, the conversion needs to be defered when
3055 * the IO is completed. The ext4 end_io callback function
3056 * will be called to take care of the conversion work.
3057 * Here for async case, we allocate an io_end structure to
3060 iocb->private = NULL;
3061 EXT4_I(inode)->cur_aio_dio = NULL;
3062 if (!is_sync_kiocb(iocb)) {
3063 ext4_io_end_t *io_end =
3064 ext4_init_io_end(inode, GFP_NOFS);
3067 io_end->flag |= EXT4_IO_END_DIRECT;
3068 iocb->private = io_end;
3070 * we save the io structure for current async
3071 * direct IO, so that later ext4_map_blocks()
3072 * could flag the io structure whether there
3073 * is a unwritten extents needs to be converted
3074 * when IO is completed.
3076 EXT4_I(inode)->cur_aio_dio = iocb->private;
3079 ret = __blockdev_direct_IO(rw, iocb, inode,
3080 inode->i_sb->s_bdev, iov,
3082 ext4_get_block_write,
3085 DIO_LOCKING | DIO_SKIP_HOLES);
3087 EXT4_I(inode)->cur_aio_dio = NULL;
3089 * The io_end structure takes a reference to the inode,
3090 * that structure needs to be destroyed and the
3091 * reference to the inode need to be dropped, when IO is
3092 * complete, even with 0 byte write, or failed.
3094 * In the successful AIO DIO case, the io_end structure will be
3095 * desctroyed and the reference to the inode will be dropped
3096 * after the end_io call back function is called.
3098 * In the case there is 0 byte write, or error case, since
3099 * VFS direct IO won't invoke the end_io call back function,
3100 * we need to free the end_io structure here.
3102 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3103 ext4_free_io_end(iocb->private);
3104 iocb->private = NULL;
3105 } else if (ret > 0 && ext4_test_inode_state(inode,
3106 EXT4_STATE_DIO_UNWRITTEN)) {
3109 * for non AIO case, since the IO is already
3110 * completed, we could do the conversion right here
3112 err = ext4_convert_unwritten_extents(inode,
3116 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3121 /* for write the the end of file case, we fall back to old way */
3122 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3125 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3126 const struct iovec *iov, loff_t offset,
3127 unsigned long nr_segs)
3129 struct file *file = iocb->ki_filp;
3130 struct inode *inode = file->f_mapping->host;
3134 * If we are doing data journalling we don't support O_DIRECT
3136 if (ext4_should_journal_data(inode))
3139 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3140 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3141 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3143 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3144 trace_ext4_direct_IO_exit(inode, offset,
3145 iov_length(iov, nr_segs), rw, ret);
3150 * Pages can be marked dirty completely asynchronously from ext4's journalling
3151 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3152 * much here because ->set_page_dirty is called under VFS locks. The page is
3153 * not necessarily locked.
3155 * We cannot just dirty the page and leave attached buffers clean, because the
3156 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3157 * or jbddirty because all the journalling code will explode.
3159 * So what we do is to mark the page "pending dirty" and next time writepage
3160 * is called, propagate that into the buffers appropriately.
3162 static int ext4_journalled_set_page_dirty(struct page *page)
3164 SetPageChecked(page);
3165 return __set_page_dirty_nobuffers(page);
3168 static const struct address_space_operations ext4_ordered_aops = {
3169 .readpage = ext4_readpage,
3170 .readpages = ext4_readpages,
3171 .writepage = ext4_writepage,
3172 .write_begin = ext4_write_begin,
3173 .write_end = ext4_ordered_write_end,
3175 .invalidatepage = ext4_invalidatepage,
3176 .releasepage = ext4_releasepage,
3177 .direct_IO = ext4_direct_IO,
3178 .migratepage = buffer_migrate_page,
3179 .is_partially_uptodate = block_is_partially_uptodate,
3180 .error_remove_page = generic_error_remove_page,
3183 static const struct address_space_operations ext4_writeback_aops = {
3184 .readpage = ext4_readpage,
3185 .readpages = ext4_readpages,
3186 .writepage = ext4_writepage,
3187 .write_begin = ext4_write_begin,
3188 .write_end = ext4_writeback_write_end,
3190 .invalidatepage = ext4_invalidatepage,
3191 .releasepage = ext4_releasepage,
3192 .direct_IO = ext4_direct_IO,
3193 .migratepage = buffer_migrate_page,
3194 .is_partially_uptodate = block_is_partially_uptodate,
3195 .error_remove_page = generic_error_remove_page,
3198 static const struct address_space_operations ext4_journalled_aops = {
3199 .readpage = ext4_readpage,
3200 .readpages = ext4_readpages,
3201 .writepage = ext4_writepage,
3202 .write_begin = ext4_write_begin,
3203 .write_end = ext4_journalled_write_end,
3204 .set_page_dirty = ext4_journalled_set_page_dirty,
3206 .invalidatepage = ext4_invalidatepage,
3207 .releasepage = ext4_releasepage,
3208 .direct_IO = ext4_direct_IO,
3209 .is_partially_uptodate = block_is_partially_uptodate,
3210 .error_remove_page = generic_error_remove_page,
3213 static const struct address_space_operations ext4_da_aops = {
3214 .readpage = ext4_readpage,
3215 .readpages = ext4_readpages,
3216 .writepage = ext4_writepage,
3217 .writepages = ext4_da_writepages,
3218 .write_begin = ext4_da_write_begin,
3219 .write_end = ext4_da_write_end,
3221 .invalidatepage = ext4_da_invalidatepage,
3222 .releasepage = ext4_releasepage,
3223 .direct_IO = ext4_direct_IO,
3224 .migratepage = buffer_migrate_page,
3225 .is_partially_uptodate = block_is_partially_uptodate,
3226 .error_remove_page = generic_error_remove_page,
3229 void ext4_set_aops(struct inode *inode)
3231 switch (ext4_inode_journal_mode(inode)) {
3232 case EXT4_INODE_ORDERED_DATA_MODE:
3233 if (test_opt(inode->i_sb, DELALLOC))
3234 inode->i_mapping->a_ops = &ext4_da_aops;
3236 inode->i_mapping->a_ops = &ext4_ordered_aops;
3238 case EXT4_INODE_WRITEBACK_DATA_MODE:
3239 if (test_opt(inode->i_sb, DELALLOC))
3240 inode->i_mapping->a_ops = &ext4_da_aops;
3242 inode->i_mapping->a_ops = &ext4_writeback_aops;
3244 case EXT4_INODE_JOURNAL_DATA_MODE:
3245 inode->i_mapping->a_ops = &ext4_journalled_aops;
3254 * ext4_discard_partial_page_buffers()
3255 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3256 * This function finds and locks the page containing the offset
3257 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3258 * Calling functions that already have the page locked should call
3259 * ext4_discard_partial_page_buffers_no_lock directly.
3261 int ext4_discard_partial_page_buffers(handle_t *handle,
3262 struct address_space *mapping, loff_t from,
3263 loff_t length, int flags)
3265 struct inode *inode = mapping->host;
3269 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3270 mapping_gfp_mask(mapping) & ~__GFP_FS);
3274 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3275 from, length, flags);
3278 page_cache_release(page);
3283 * ext4_discard_partial_page_buffers_no_lock()
3284 * Zeros a page range of length 'length' starting from offset 'from'.
3285 * Buffer heads that correspond to the block aligned regions of the
3286 * zeroed range will be unmapped. Unblock aligned regions
3287 * will have the corresponding buffer head mapped if needed so that
3288 * that region of the page can be updated with the partial zero out.
3290 * This function assumes that the page has already been locked. The
3291 * The range to be discarded must be contained with in the given page.
3292 * If the specified range exceeds the end of the page it will be shortened
3293 * to the end of the page that corresponds to 'from'. This function is
3294 * appropriate for updating a page and it buffer heads to be unmapped and
3295 * zeroed for blocks that have been either released, or are going to be
3298 * handle: The journal handle
3299 * inode: The files inode
3300 * page: A locked page that contains the offset "from"
3301 * from: The starting byte offset (from the begining of the file)
3302 * to begin discarding
3303 * len: The length of bytes to discard
3304 * flags: Optional flags that may be used:
3306 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3307 * Only zero the regions of the page whose buffer heads
3308 * have already been unmapped. This flag is appropriate
3309 * for updateing the contents of a page whose blocks may
3310 * have already been released, and we only want to zero
3311 * out the regions that correspond to those released blocks.
3313 * Returns zero on sucess or negative on failure.
3315 int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3316 struct inode *inode, struct page *page, loff_t from,
3317 loff_t length, int flags)
3319 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3320 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3321 unsigned int blocksize, max, pos;
3323 struct buffer_head *bh;
3326 blocksize = inode->i_sb->s_blocksize;
3327 max = PAGE_CACHE_SIZE - offset;
3329 if (index != page->index)
3333 * correct length if it does not fall between
3334 * 'from' and the end of the page
3336 if (length > max || length < 0)
3339 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3341 if (!page_has_buffers(page))
3342 create_empty_buffers(page, blocksize, 0);
3344 /* Find the buffer that contains "offset" */
3345 bh = page_buffers(page);
3347 while (offset >= pos) {
3348 bh = bh->b_this_page;
3354 while (pos < offset + length) {
3355 unsigned int end_of_block, range_to_discard;
3359 /* The length of space left to zero and unmap */
3360 range_to_discard = offset + length - pos;
3362 /* The length of space until the end of the block */
3363 end_of_block = blocksize - (pos & (blocksize-1));
3366 * Do not unmap or zero past end of block
3367 * for this buffer head
3369 if (range_to_discard > end_of_block)
3370 range_to_discard = end_of_block;
3374 * Skip this buffer head if we are only zeroing unampped
3375 * regions of the page
3377 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3381 /* If the range is block aligned, unmap */
3382 if (range_to_discard == blocksize) {
3383 clear_buffer_dirty(bh);
3385 clear_buffer_mapped(bh);
3386 clear_buffer_req(bh);
3387 clear_buffer_new(bh);
3388 clear_buffer_delay(bh);
3389 clear_buffer_unwritten(bh);
3390 clear_buffer_uptodate(bh);
3391 zero_user(page, pos, range_to_discard);
3392 BUFFER_TRACE(bh, "Buffer discarded");
3397 * If this block is not completely contained in the range
3398 * to be discarded, then it is not going to be released. Because
3399 * we need to keep this block, we need to make sure this part
3400 * of the page is uptodate before we modify it by writeing
3401 * partial zeros on it.
3403 if (!buffer_mapped(bh)) {
3405 * Buffer head must be mapped before we can read
3408 BUFFER_TRACE(bh, "unmapped");
3409 ext4_get_block(inode, iblock, bh, 0);
3410 /* unmapped? It's a hole - nothing to do */
3411 if (!buffer_mapped(bh)) {
3412 BUFFER_TRACE(bh, "still unmapped");
3417 /* Ok, it's mapped. Make sure it's up-to-date */
3418 if (PageUptodate(page))
3419 set_buffer_uptodate(bh);
3421 if (!buffer_uptodate(bh)) {
3423 ll_rw_block(READ, 1, &bh);
3425 /* Uhhuh. Read error. Complain and punt.*/
3426 if (!buffer_uptodate(bh))
3430 if (ext4_should_journal_data(inode)) {
3431 BUFFER_TRACE(bh, "get write access");
3432 err = ext4_journal_get_write_access(handle, bh);
3437 zero_user(page, pos, range_to_discard);
3440 if (ext4_should_journal_data(inode)) {
3441 err = ext4_handle_dirty_metadata(handle, inode, bh);
3443 mark_buffer_dirty(bh);
3445 BUFFER_TRACE(bh, "Partial buffer zeroed");
3447 bh = bh->b_this_page;
3449 pos += range_to_discard;
3456 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3457 * up to the end of the block which corresponds to `from'.
3458 * This required during truncate. We need to physically zero the tail end
3459 * of that block so it doesn't yield old data if the file is later grown.
3461 int ext4_block_truncate_page(handle_t *handle,
3462 struct address_space *mapping, loff_t from)
3464 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3467 struct inode *inode = mapping->host;
3469 blocksize = inode->i_sb->s_blocksize;
3470 length = blocksize - (offset & (blocksize - 1));
3472 return ext4_block_zero_page_range(handle, mapping, from, length);
3476 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3477 * starting from file offset 'from'. The range to be zero'd must
3478 * be contained with in one block. If the specified range exceeds
3479 * the end of the block it will be shortened to end of the block
3480 * that cooresponds to 'from'
3482 int ext4_block_zero_page_range(handle_t *handle,
3483 struct address_space *mapping, loff_t from, loff_t length)
3485 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3486 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3487 unsigned blocksize, max, pos;
3489 struct inode *inode = mapping->host;
3490 struct buffer_head *bh;
3494 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3495 mapping_gfp_mask(mapping) & ~__GFP_FS);
3499 blocksize = inode->i_sb->s_blocksize;
3500 max = blocksize - (offset & (blocksize - 1));
3503 * correct length if it does not fall between
3504 * 'from' and the end of the block
3506 if (length > max || length < 0)
3509 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3511 if (!page_has_buffers(page))
3512 create_empty_buffers(page, blocksize, 0);
3514 /* Find the buffer that contains "offset" */
3515 bh = page_buffers(page);
3517 while (offset >= pos) {
3518 bh = bh->b_this_page;
3524 if (buffer_freed(bh)) {
3525 BUFFER_TRACE(bh, "freed: skip");
3529 if (!buffer_mapped(bh)) {
3530 BUFFER_TRACE(bh, "unmapped");
3531 ext4_get_block(inode, iblock, bh, 0);
3532 /* unmapped? It's a hole - nothing to do */
3533 if (!buffer_mapped(bh)) {
3534 BUFFER_TRACE(bh, "still unmapped");
3539 /* Ok, it's mapped. Make sure it's up-to-date */
3540 if (PageUptodate(page))
3541 set_buffer_uptodate(bh);
3543 if (!buffer_uptodate(bh)) {
3545 ll_rw_block(READ, 1, &bh);
3547 /* Uhhuh. Read error. Complain and punt. */
3548 if (!buffer_uptodate(bh))
3552 if (ext4_should_journal_data(inode)) {
3553 BUFFER_TRACE(bh, "get write access");
3554 err = ext4_journal_get_write_access(handle, bh);
3559 zero_user(page, offset, length);
3561 BUFFER_TRACE(bh, "zeroed end of block");
3564 if (ext4_should_journal_data(inode)) {
3565 err = ext4_handle_dirty_metadata(handle, inode, bh);
3567 mark_buffer_dirty(bh);
3571 page_cache_release(page);
3575 int ext4_can_truncate(struct inode *inode)
3577 if (S_ISREG(inode->i_mode))
3579 if (S_ISDIR(inode->i_mode))
3581 if (S_ISLNK(inode->i_mode))
3582 return !ext4_inode_is_fast_symlink(inode);
3587 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3588 * associated with the given offset and length
3590 * @inode: File inode
3591 * @offset: The offset where the hole will begin
3592 * @len: The length of the hole
3594 * Returns: 0 on sucess or negative on failure
3597 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3599 struct inode *inode = file->f_path.dentry->d_inode;
3600 if (!S_ISREG(inode->i_mode))
3603 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3604 /* TODO: Add support for non extent hole punching */
3608 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3609 /* TODO: Add support for bigalloc file systems */
3613 return ext4_ext_punch_hole(file, offset, length);
3619 * We block out ext4_get_block() block instantiations across the entire
3620 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3621 * simultaneously on behalf of the same inode.
3623 * As we work through the truncate and commmit bits of it to the journal there
3624 * is one core, guiding principle: the file's tree must always be consistent on
3625 * disk. We must be able to restart the truncate after a crash.
3627 * The file's tree may be transiently inconsistent in memory (although it
3628 * probably isn't), but whenever we close off and commit a journal transaction,
3629 * the contents of (the filesystem + the journal) must be consistent and
3630 * restartable. It's pretty simple, really: bottom up, right to left (although
3631 * left-to-right works OK too).
3633 * Note that at recovery time, journal replay occurs *before* the restart of
3634 * truncate against the orphan inode list.
3636 * The committed inode has the new, desired i_size (which is the same as
3637 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3638 * that this inode's truncate did not complete and it will again call
3639 * ext4_truncate() to have another go. So there will be instantiated blocks
3640 * to the right of the truncation point in a crashed ext4 filesystem. But
3641 * that's fine - as long as they are linked from the inode, the post-crash
3642 * ext4_truncate() run will find them and release them.
3644 void ext4_truncate(struct inode *inode)
3646 trace_ext4_truncate_enter(inode);
3648 if (!ext4_can_truncate(inode))
3651 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3653 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3654 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3656 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3657 ext4_ext_truncate(inode);
3659 ext4_ind_truncate(inode);
3661 trace_ext4_truncate_exit(inode);
3665 * ext4_get_inode_loc returns with an extra refcount against the inode's
3666 * underlying buffer_head on success. If 'in_mem' is true, we have all
3667 * data in memory that is needed to recreate the on-disk version of this
3670 static int __ext4_get_inode_loc(struct inode *inode,
3671 struct ext4_iloc *iloc, int in_mem)
3673 struct ext4_group_desc *gdp;
3674 struct buffer_head *bh;
3675 struct super_block *sb = inode->i_sb;
3677 int inodes_per_block, inode_offset;
3680 if (!ext4_valid_inum(sb, inode->i_ino))
3683 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3684 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3689 * Figure out the offset within the block group inode table
3691 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3692 inode_offset = ((inode->i_ino - 1) %
3693 EXT4_INODES_PER_GROUP(sb));
3694 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3695 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3697 bh = sb_getblk(sb, block);
3700 if (!buffer_uptodate(bh)) {
3704 * If the buffer has the write error flag, we have failed
3705 * to write out another inode in the same block. In this
3706 * case, we don't have to read the block because we may
3707 * read the old inode data successfully.
3709 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3710 set_buffer_uptodate(bh);
3712 if (buffer_uptodate(bh)) {
3713 /* someone brought it uptodate while we waited */
3719 * If we have all information of the inode in memory and this
3720 * is the only valid inode in the block, we need not read the
3724 struct buffer_head *bitmap_bh;
3727 start = inode_offset & ~(inodes_per_block - 1);
3729 /* Is the inode bitmap in cache? */
3730 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3735 * If the inode bitmap isn't in cache then the
3736 * optimisation may end up performing two reads instead
3737 * of one, so skip it.
3739 if (!buffer_uptodate(bitmap_bh)) {
3743 for (i = start; i < start + inodes_per_block; i++) {
3744 if (i == inode_offset)
3746 if (ext4_test_bit(i, bitmap_bh->b_data))
3750 if (i == start + inodes_per_block) {
3751 /* all other inodes are free, so skip I/O */
3752 memset(bh->b_data, 0, bh->b_size);
3753 set_buffer_uptodate(bh);
3761 * If we need to do any I/O, try to pre-readahead extra
3762 * blocks from the inode table.
3764 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3765 ext4_fsblk_t b, end, table;
3768 table = ext4_inode_table(sb, gdp);
3769 /* s_inode_readahead_blks is always a power of 2 */
3770 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3773 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3774 num = EXT4_INODES_PER_GROUP(sb);
3775 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3776 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3777 num -= ext4_itable_unused_count(sb, gdp);
3778 table += num / inodes_per_block;
3782 sb_breadahead(sb, b++);
3786 * There are other valid inodes in the buffer, this inode
3787 * has in-inode xattrs, or we don't have this inode in memory.
3788 * Read the block from disk.
3790 trace_ext4_load_inode(inode);
3792 bh->b_end_io = end_buffer_read_sync;
3793 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3795 if (!buffer_uptodate(bh)) {
3796 EXT4_ERROR_INODE_BLOCK(inode, block,
3797 "unable to read itable block");
3807 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3809 /* We have all inode data except xattrs in memory here. */
3810 return __ext4_get_inode_loc(inode, iloc,
3811 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3814 void ext4_set_inode_flags(struct inode *inode)
3816 unsigned int flags = EXT4_I(inode)->i_flags;
3817 unsigned int new_fl = 0;
3819 if (flags & EXT4_SYNC_FL)
3821 if (flags & EXT4_APPEND_FL)
3823 if (flags & EXT4_IMMUTABLE_FL)
3824 new_fl |= S_IMMUTABLE;
3825 if (flags & EXT4_NOATIME_FL)
3826 new_fl |= S_NOATIME;
3827 if (flags & EXT4_DIRSYNC_FL)
3828 new_fl |= S_DIRSYNC;
3829 set_mask_bits(&inode->i_flags,
3830 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC, new_fl);
3833 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3834 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3836 unsigned int vfs_fl;
3837 unsigned long old_fl, new_fl;
3840 vfs_fl = ei->vfs_inode.i_flags;
3841 old_fl = ei->i_flags;
3842 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3843 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3845 if (vfs_fl & S_SYNC)
3846 new_fl |= EXT4_SYNC_FL;
3847 if (vfs_fl & S_APPEND)
3848 new_fl |= EXT4_APPEND_FL;
3849 if (vfs_fl & S_IMMUTABLE)
3850 new_fl |= EXT4_IMMUTABLE_FL;
3851 if (vfs_fl & S_NOATIME)
3852 new_fl |= EXT4_NOATIME_FL;
3853 if (vfs_fl & S_DIRSYNC)
3854 new_fl |= EXT4_DIRSYNC_FL;
3855 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3858 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3859 struct ext4_inode_info *ei)
3862 struct inode *inode = &(ei->vfs_inode);
3863 struct super_block *sb = inode->i_sb;
3865 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3866 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3867 /* we are using combined 48 bit field */
3868 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3869 le32_to_cpu(raw_inode->i_blocks_lo);
3870 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3871 /* i_blocks represent file system block size */
3872 return i_blocks << (inode->i_blkbits - 9);
3877 return le32_to_cpu(raw_inode->i_blocks_lo);
3881 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3883 struct ext4_iloc iloc;
3884 struct ext4_inode *raw_inode;
3885 struct ext4_inode_info *ei;
3886 struct inode *inode;
3887 journal_t *journal = EXT4_SB(sb)->s_journal;
3892 inode = iget_locked(sb, ino);
3894 return ERR_PTR(-ENOMEM);
3895 if (!(inode->i_state & I_NEW))
3901 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3904 raw_inode = ext4_raw_inode(&iloc);
3905 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3906 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3907 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3908 if (!(test_opt(inode->i_sb, NO_UID32))) {
3909 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3910 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3912 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3914 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3915 ei->i_dir_start_lookup = 0;
3916 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3917 /* We now have enough fields to check if the inode was active or not.
3918 * This is needed because nfsd might try to access dead inodes
3919 * the test is that same one that e2fsck uses
3920 * NeilBrown 1999oct15
3922 if (inode->i_nlink == 0) {
3923 if (inode->i_mode == 0 ||
3924 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3925 /* this inode is deleted */
3929 /* The only unlinked inodes we let through here have
3930 * valid i_mode and are being read by the orphan
3931 * recovery code: that's fine, we're about to complete
3932 * the process of deleting those. */
3934 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3935 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3936 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3937 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3939 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3940 inode->i_size = ext4_isize(raw_inode);
3941 if ((size = i_size_read(inode)) < 0) {
3942 EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size);
3946 ei->i_disksize = inode->i_size;
3948 ei->i_reserved_quota = 0;
3950 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3951 ei->i_block_group = iloc.block_group;
3952 ei->i_last_alloc_group = ~0;
3954 * NOTE! The in-memory inode i_data array is in little-endian order
3955 * even on big-endian machines: we do NOT byteswap the block numbers!
3957 for (block = 0; block < EXT4_N_BLOCKS; block++)
3958 ei->i_data[block] = raw_inode->i_block[block];
3959 INIT_LIST_HEAD(&ei->i_orphan);
3962 * Set transaction id's of transactions that have to be committed
3963 * to finish f[data]sync. We set them to currently running transaction
3964 * as we cannot be sure that the inode or some of its metadata isn't
3965 * part of the transaction - the inode could have been reclaimed and
3966 * now it is reread from disk.
3969 transaction_t *transaction;
3972 read_lock(&journal->j_state_lock);
3973 if (journal->j_running_transaction)
3974 transaction = journal->j_running_transaction;
3976 transaction = journal->j_committing_transaction;
3978 tid = transaction->t_tid;
3980 tid = journal->j_commit_sequence;
3981 read_unlock(&journal->j_state_lock);
3982 ei->i_sync_tid = tid;
3983 ei->i_datasync_tid = tid;
3986 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3987 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3988 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3989 EXT4_INODE_SIZE(inode->i_sb)) {
3993 if (ei->i_extra_isize == 0) {
3994 /* The extra space is currently unused. Use it. */
3995 ei->i_extra_isize = sizeof(struct ext4_inode) -
3996 EXT4_GOOD_OLD_INODE_SIZE;
3998 __le32 *magic = (void *)raw_inode +
3999 EXT4_GOOD_OLD_INODE_SIZE +
4001 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4002 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4005 ei->i_extra_isize = 0;
4007 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4008 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4009 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4010 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4012 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4013 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4014 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4016 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4020 if (ei->i_file_acl &&
4021 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4022 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4026 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4027 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4028 (S_ISLNK(inode->i_mode) &&
4029 !ext4_inode_is_fast_symlink(inode)))
4030 /* Validate extent which is part of inode */
4031 ret = ext4_ext_check_inode(inode);
4032 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4033 (S_ISLNK(inode->i_mode) &&
4034 !ext4_inode_is_fast_symlink(inode))) {
4035 /* Validate block references which are part of inode */
4036 ret = ext4_ind_check_inode(inode);
4041 if (S_ISREG(inode->i_mode)) {
4042 inode->i_op = &ext4_file_inode_operations;
4043 inode->i_fop = &ext4_file_operations;
4044 ext4_set_aops(inode);
4045 } else if (S_ISDIR(inode->i_mode)) {
4046 inode->i_op = &ext4_dir_inode_operations;
4047 inode->i_fop = &ext4_dir_operations;
4048 } else if (S_ISLNK(inode->i_mode)) {
4049 if (ext4_inode_is_fast_symlink(inode)) {
4050 inode->i_op = &ext4_fast_symlink_inode_operations;
4051 nd_terminate_link(ei->i_data, inode->i_size,
4052 sizeof(ei->i_data) - 1);
4054 inode->i_op = &ext4_symlink_inode_operations;
4055 ext4_set_aops(inode);
4057 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4058 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4059 inode->i_op = &ext4_special_inode_operations;
4060 if (raw_inode->i_block[0])
4061 init_special_inode(inode, inode->i_mode,
4062 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4064 init_special_inode(inode, inode->i_mode,
4065 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4068 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4072 ext4_set_inode_flags(inode);
4073 unlock_new_inode(inode);
4079 return ERR_PTR(ret);
4082 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4084 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4085 return ERR_PTR(-EIO);
4086 return ext4_iget(sb, ino);
4089 static int ext4_inode_blocks_set(handle_t *handle,
4090 struct ext4_inode *raw_inode,
4091 struct ext4_inode_info *ei)
4093 struct inode *inode = &(ei->vfs_inode);
4094 u64 i_blocks = inode->i_blocks;
4095 struct super_block *sb = inode->i_sb;
4097 if (i_blocks <= ~0U) {
4099 * i_blocks can be represnted in a 32 bit variable
4100 * as multiple of 512 bytes
4102 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4103 raw_inode->i_blocks_high = 0;
4104 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4107 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4110 if (i_blocks <= 0xffffffffffffULL) {
4112 * i_blocks can be represented in a 48 bit variable
4113 * as multiple of 512 bytes
4115 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4116 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4117 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4119 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4120 /* i_block is stored in file system block size */
4121 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4122 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4123 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4129 * Post the struct inode info into an on-disk inode location in the
4130 * buffer-cache. This gobbles the caller's reference to the
4131 * buffer_head in the inode location struct.
4133 * The caller must have write access to iloc->bh.
4135 static int ext4_do_update_inode(handle_t *handle,
4136 struct inode *inode,
4137 struct ext4_iloc *iloc)
4139 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4140 struct ext4_inode_info *ei = EXT4_I(inode);
4141 struct buffer_head *bh = iloc->bh;
4142 int err = 0, rc, block;
4143 int need_datasync = 0;
4145 /* For fields not not tracking in the in-memory inode,
4146 * initialise them to zero for new inodes. */
4147 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4148 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4150 ext4_get_inode_flags(ei);
4151 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4152 if (!(test_opt(inode->i_sb, NO_UID32))) {
4153 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4154 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4156 * Fix up interoperability with old kernels. Otherwise, old inodes get
4157 * re-used with the upper 16 bits of the uid/gid intact
4159 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4160 raw_inode->i_uid_high = 0;
4161 raw_inode->i_gid_high = 0;
4163 raw_inode->i_uid_high =
4164 cpu_to_le16(high_16_bits(inode->i_uid));
4165 raw_inode->i_gid_high =
4166 cpu_to_le16(high_16_bits(inode->i_gid));
4169 raw_inode->i_uid_low =
4170 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4171 raw_inode->i_gid_low =
4172 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4173 raw_inode->i_uid_high = 0;
4174 raw_inode->i_gid_high = 0;
4176 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4178 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4179 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4180 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4181 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4183 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4185 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4186 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4187 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4188 cpu_to_le32(EXT4_OS_HURD))
4189 raw_inode->i_file_acl_high =
4190 cpu_to_le16(ei->i_file_acl >> 32);
4191 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4192 if (ei->i_disksize != ext4_isize(raw_inode)) {
4193 ext4_isize_set(raw_inode, ei->i_disksize);
4196 if (ei->i_disksize > 0x7fffffffULL) {
4197 struct super_block *sb = inode->i_sb;
4198 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4199 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4200 EXT4_SB(sb)->s_es->s_rev_level ==
4201 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4202 /* If this is the first large file
4203 * created, add a flag to the superblock.
4205 err = ext4_journal_get_write_access(handle,
4206 EXT4_SB(sb)->s_sbh);
4209 ext4_update_dynamic_rev(sb);
4210 EXT4_SET_RO_COMPAT_FEATURE(sb,
4211 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4213 ext4_handle_sync(handle);
4214 err = ext4_handle_dirty_metadata(handle, NULL,
4215 EXT4_SB(sb)->s_sbh);
4218 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4219 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4220 if (old_valid_dev(inode->i_rdev)) {
4221 raw_inode->i_block[0] =
4222 cpu_to_le32(old_encode_dev(inode->i_rdev));
4223 raw_inode->i_block[1] = 0;
4225 raw_inode->i_block[0] = 0;
4226 raw_inode->i_block[1] =
4227 cpu_to_le32(new_encode_dev(inode->i_rdev));
4228 raw_inode->i_block[2] = 0;
4231 for (block = 0; block < EXT4_N_BLOCKS; block++)
4232 raw_inode->i_block[block] = ei->i_data[block];
4234 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4235 if (ei->i_extra_isize) {
4236 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4237 raw_inode->i_version_hi =
4238 cpu_to_le32(inode->i_version >> 32);
4239 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4242 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4243 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4246 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4248 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4251 ext4_std_error(inode->i_sb, err);
4256 * ext4_write_inode()
4258 * We are called from a few places:
4260 * - Within generic_file_write() for O_SYNC files.
4261 * Here, there will be no transaction running. We wait for any running
4262 * trasnaction to commit.
4264 * - Within sys_sync(), kupdate and such.
4265 * We wait on commit, if tol to.
4267 * - Within prune_icache() (PF_MEMALLOC == true)
4268 * Here we simply return. We can't afford to block kswapd on the
4271 * In all cases it is actually safe for us to return without doing anything,
4272 * because the inode has been copied into a raw inode buffer in
4273 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4276 * Note that we are absolutely dependent upon all inode dirtiers doing the
4277 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4278 * which we are interested.
4280 * It would be a bug for them to not do this. The code:
4282 * mark_inode_dirty(inode)
4284 * inode->i_size = expr;
4286 * is in error because a kswapd-driven write_inode() could occur while
4287 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4288 * will no longer be on the superblock's dirty inode list.
4290 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4294 if (current->flags & PF_MEMALLOC)
4297 if (EXT4_SB(inode->i_sb)->s_journal) {
4298 if (ext4_journal_current_handle()) {
4299 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4304 if (wbc->sync_mode != WB_SYNC_ALL)
4307 err = ext4_force_commit(inode->i_sb);
4309 struct ext4_iloc iloc;
4311 err = __ext4_get_inode_loc(inode, &iloc, 0);
4314 if (wbc->sync_mode == WB_SYNC_ALL)
4315 sync_dirty_buffer(iloc.bh);
4316 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4317 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4318 "IO error syncing inode");
4329 * Called from notify_change.
4331 * We want to trap VFS attempts to truncate the file as soon as
4332 * possible. In particular, we want to make sure that when the VFS
4333 * shrinks i_size, we put the inode on the orphan list and modify
4334 * i_disksize immediately, so that during the subsequent flushing of
4335 * dirty pages and freeing of disk blocks, we can guarantee that any
4336 * commit will leave the blocks being flushed in an unused state on
4337 * disk. (On recovery, the inode will get truncated and the blocks will
4338 * be freed, so we have a strong guarantee that no future commit will
4339 * leave these blocks visible to the user.)
4341 * Another thing we have to assure is that if we are in ordered mode
4342 * and inode is still attached to the committing transaction, we must
4343 * we start writeout of all the dirty pages which are being truncated.
4344 * This way we are sure that all the data written in the previous
4345 * transaction are already on disk (truncate waits for pages under
4348 * Called with inode->i_mutex down.
4350 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4352 struct inode *inode = dentry->d_inode;
4355 const unsigned int ia_valid = attr->ia_valid;
4357 error = setattr_prepare(dentry, attr);
4361 if (is_quota_modification(inode, attr))
4362 dquot_initialize(inode);
4363 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4364 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4367 /* (user+group)*(old+new) structure, inode write (sb,
4368 * inode block, ? - but truncate inode update has it) */
4369 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
4370 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
4371 if (IS_ERR(handle)) {
4372 error = PTR_ERR(handle);
4375 error = dquot_transfer(inode, attr);
4377 ext4_journal_stop(handle);
4380 /* Update corresponding info in inode so that everything is in
4381 * one transaction */
4382 if (attr->ia_valid & ATTR_UID)
4383 inode->i_uid = attr->ia_uid;
4384 if (attr->ia_valid & ATTR_GID)
4385 inode->i_gid = attr->ia_gid;
4386 error = ext4_mark_inode_dirty(handle, inode);
4387 ext4_journal_stop(handle);
4390 if (attr->ia_valid & ATTR_SIZE) {
4391 inode_dio_wait(inode);
4393 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4394 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4396 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4401 if (S_ISREG(inode->i_mode) &&
4402 attr->ia_valid & ATTR_SIZE &&
4403 (attr->ia_size < inode->i_size)) {
4406 handle = ext4_journal_start(inode, 3);
4407 if (IS_ERR(handle)) {
4408 error = PTR_ERR(handle);
4411 if (ext4_handle_valid(handle)) {
4412 error = ext4_orphan_add(handle, inode);
4415 EXT4_I(inode)->i_disksize = attr->ia_size;
4416 rc = ext4_mark_inode_dirty(handle, inode);
4419 ext4_journal_stop(handle);
4421 if (ext4_should_order_data(inode)) {
4422 error = ext4_begin_ordered_truncate(inode,
4425 /* Do as much error cleanup as possible */
4426 handle = ext4_journal_start(inode, 3);
4427 if (IS_ERR(handle)) {
4428 ext4_orphan_del(NULL, inode);
4431 ext4_orphan_del(handle, inode);
4433 ext4_journal_stop(handle);
4439 if (attr->ia_valid & ATTR_SIZE) {
4440 if (attr->ia_size != i_size_read(inode)) {
4441 truncate_setsize(inode, attr->ia_size);
4442 ext4_truncate(inode);
4443 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
4444 ext4_truncate(inode);
4448 setattr_copy(inode, attr);
4449 mark_inode_dirty(inode);
4453 * If the call to ext4_truncate failed to get a transaction handle at
4454 * all, we need to clean up the in-core orphan list manually.
4456 if (orphan && inode->i_nlink)
4457 ext4_orphan_del(NULL, inode);
4459 if (!rc && (ia_valid & ATTR_MODE))
4460 rc = ext4_acl_chmod(inode);
4463 ext4_std_error(inode->i_sb, error);
4469 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4472 struct inode *inode;
4473 unsigned long long delalloc_blocks;
4475 inode = dentry->d_inode;
4476 generic_fillattr(inode, stat);
4479 * We can't update i_blocks if the block allocation is delayed
4480 * otherwise in the case of system crash before the real block
4481 * allocation is done, we will have i_blocks inconsistent with
4482 * on-disk file blocks.
4483 * We always keep i_blocks updated together with real
4484 * allocation. But to not confuse with user, stat
4485 * will return the blocks that include the delayed allocation
4486 * blocks for this file.
4488 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4490 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits-9);
4494 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4496 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4497 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4498 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4502 * Account for index blocks, block groups bitmaps and block group
4503 * descriptor blocks if modify datablocks and index blocks
4504 * worse case, the indexs blocks spread over different block groups
4506 * If datablocks are discontiguous, they are possible to spread over
4507 * different block groups too. If they are contiuguous, with flexbg,
4508 * they could still across block group boundary.
4510 * Also account for superblock, inode, quota and xattr blocks
4512 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4514 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4520 * How many index blocks need to touch to modify nrblocks?
4521 * The "Chunk" flag indicating whether the nrblocks is
4522 * physically contiguous on disk
4524 * For Direct IO and fallocate, they calls get_block to allocate
4525 * one single extent at a time, so they could set the "Chunk" flag
4527 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4532 * Now let's see how many group bitmaps and group descriptors need
4542 if (groups > ngroups)
4544 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4545 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4547 /* bitmaps and block group descriptor blocks */
4548 ret += groups + gdpblocks;
4550 /* Blocks for super block, inode, quota and xattr blocks */
4551 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4557 * Calculate the total number of credits to reserve to fit
4558 * the modification of a single pages into a single transaction,
4559 * which may include multiple chunks of block allocations.
4561 * This could be called via ext4_write_begin()
4563 * We need to consider the worse case, when
4564 * one new block per extent.
4566 int ext4_writepage_trans_blocks(struct inode *inode)
4568 int bpp = ext4_journal_blocks_per_page(inode);
4571 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4573 /* Account for data blocks for journalled mode */
4574 if (ext4_should_journal_data(inode))
4580 * Calculate the journal credits for a chunk of data modification.
4582 * This is called from DIO, fallocate or whoever calling
4583 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4585 * journal buffers for data blocks are not included here, as DIO
4586 * and fallocate do no need to journal data buffers.
4588 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4590 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4594 * The caller must have previously called ext4_reserve_inode_write().
4595 * Give this, we know that the caller already has write access to iloc->bh.
4597 int ext4_mark_iloc_dirty(handle_t *handle,
4598 struct inode *inode, struct ext4_iloc *iloc)
4602 if (test_opt(inode->i_sb, I_VERSION))
4603 inode_inc_iversion(inode);
4605 /* the do_update_inode consumes one bh->b_count */
4608 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4609 err = ext4_do_update_inode(handle, inode, iloc);
4615 * On success, We end up with an outstanding reference count against
4616 * iloc->bh. This _must_ be cleaned up later.
4620 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4621 struct ext4_iloc *iloc)
4625 err = ext4_get_inode_loc(inode, iloc);
4627 BUFFER_TRACE(iloc->bh, "get_write_access");
4628 err = ext4_journal_get_write_access(handle, iloc->bh);
4634 ext4_std_error(inode->i_sb, err);
4639 * Expand an inode by new_extra_isize bytes.
4640 * Returns 0 on success or negative error number on failure.
4642 static int ext4_expand_extra_isize(struct inode *inode,
4643 unsigned int new_extra_isize,
4644 struct ext4_iloc iloc,
4647 struct ext4_inode *raw_inode;
4648 struct ext4_xattr_ibody_header *header;
4650 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4653 raw_inode = ext4_raw_inode(&iloc);
4655 header = IHDR(inode, raw_inode);
4657 /* No extended attributes present */
4658 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4659 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4660 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
4661 EXT4_I(inode)->i_extra_isize, 0,
4662 new_extra_isize - EXT4_I(inode)->i_extra_isize);
4663 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4667 /* try to expand with EAs present */
4668 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4673 * What we do here is to mark the in-core inode as clean with respect to inode
4674 * dirtiness (it may still be data-dirty).
4675 * This means that the in-core inode may be reaped by prune_icache
4676 * without having to perform any I/O. This is a very good thing,
4677 * because *any* task may call prune_icache - even ones which
4678 * have a transaction open against a different journal.
4680 * Is this cheating? Not really. Sure, we haven't written the
4681 * inode out, but prune_icache isn't a user-visible syncing function.
4682 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4683 * we start and wait on commits.
4685 * Is this efficient/effective? Well, we're being nice to the system
4686 * by cleaning up our inodes proactively so they can be reaped
4687 * without I/O. But we are potentially leaving up to five seconds'
4688 * worth of inodes floating about which prune_icache wants us to
4689 * write out. One way to fix that would be to get prune_icache()
4690 * to do a write_super() to free up some memory. It has the desired
4693 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4695 struct ext4_iloc iloc;
4696 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4697 static unsigned int mnt_count;
4701 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4702 err = ext4_reserve_inode_write(handle, inode, &iloc);
4705 if (ext4_handle_valid(handle) &&
4706 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4707 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4709 * We need extra buffer credits since we may write into EA block
4710 * with this same handle. If journal_extend fails, then it will
4711 * only result in a minor loss of functionality for that inode.
4712 * If this is felt to be critical, then e2fsck should be run to
4713 * force a large enough s_min_extra_isize.
4715 if ((jbd2_journal_extend(handle,
4716 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4717 ret = ext4_expand_extra_isize(inode,
4718 sbi->s_want_extra_isize,
4721 ext4_set_inode_state(inode,
4722 EXT4_STATE_NO_EXPAND);
4724 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4725 ext4_warning(inode->i_sb,
4726 "Unable to expand inode %lu. Delete"
4727 " some EAs or run e2fsck.",
4730 le16_to_cpu(sbi->s_es->s_mnt_count);
4735 return ext4_mark_iloc_dirty(handle, inode, &iloc);
4739 * ext4_dirty_inode() is called from __mark_inode_dirty()
4741 * We're really interested in the case where a file is being extended.
4742 * i_size has been changed by generic_commit_write() and we thus need
4743 * to include the updated inode in the current transaction.
4745 * Also, dquot_alloc_block() will always dirty the inode when blocks
4746 * are allocated to the file.
4748 * If the inode is marked synchronous, we don't honour that here - doing
4749 * so would cause a commit on atime updates, which we don't bother doing.
4750 * We handle synchronous inodes at the highest possible level.
4752 void ext4_dirty_inode(struct inode *inode, int flags)
4756 handle = ext4_journal_start(inode, 2);
4760 ext4_mark_inode_dirty(handle, inode);
4762 ext4_journal_stop(handle);
4769 * Bind an inode's backing buffer_head into this transaction, to prevent
4770 * it from being flushed to disk early. Unlike
4771 * ext4_reserve_inode_write, this leaves behind no bh reference and
4772 * returns no iloc structure, so the caller needs to repeat the iloc
4773 * lookup to mark the inode dirty later.
4775 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4777 struct ext4_iloc iloc;
4781 err = ext4_get_inode_loc(inode, &iloc);
4783 BUFFER_TRACE(iloc.bh, "get_write_access");
4784 err = jbd2_journal_get_write_access(handle, iloc.bh);
4786 err = ext4_handle_dirty_metadata(handle,
4792 ext4_std_error(inode->i_sb, err);
4797 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4804 * We have to be very careful here: changing a data block's
4805 * journaling status dynamically is dangerous. If we write a
4806 * data block to the journal, change the status and then delete
4807 * that block, we risk forgetting to revoke the old log record
4808 * from the journal and so a subsequent replay can corrupt data.
4809 * So, first we make sure that the journal is empty and that
4810 * nobody is changing anything.
4813 journal = EXT4_JOURNAL(inode);
4816 if (is_journal_aborted(journal))
4819 jbd2_journal_lock_updates(journal);
4820 jbd2_journal_flush(journal);
4823 * OK, there are no updates running now, and all cached data is
4824 * synced to disk. We are now in a completely consistent state
4825 * which doesn't have anything in the journal, and we know that
4826 * no filesystem updates are running, so it is safe to modify
4827 * the inode's in-core data-journaling state flag now.
4831 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4833 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4834 ext4_set_aops(inode);
4836 jbd2_journal_unlock_updates(journal);
4838 /* Finally we can mark the inode as dirty. */
4840 handle = ext4_journal_start(inode, 1);
4842 return PTR_ERR(handle);
4844 err = ext4_mark_inode_dirty(handle, inode);
4845 ext4_handle_sync(handle);
4846 ext4_journal_stop(handle);
4847 ext4_std_error(inode->i_sb, err);
4852 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4854 return !buffer_mapped(bh);
4857 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4859 struct page *page = vmf->page;
4863 struct file *file = vma->vm_file;
4864 struct inode *inode = file->f_path.dentry->d_inode;
4865 struct address_space *mapping = inode->i_mapping;
4867 get_block_t *get_block;
4871 * This check is racy but catches the common case. We rely on
4872 * __block_page_mkwrite() to do a reliable check.
4874 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
4875 /* Delalloc case is easy... */
4876 if (test_opt(inode->i_sb, DELALLOC) &&
4877 !ext4_should_journal_data(inode) &&
4878 !ext4_nonda_switch(inode->i_sb)) {
4880 ret = __block_page_mkwrite(vma, vmf,
4881 ext4_da_get_block_prep);
4882 } while (ret == -ENOSPC &&
4883 ext4_should_retry_alloc(inode->i_sb, &retries));
4888 size = i_size_read(inode);
4889 /* Page got truncated from under us? */
4890 if (page->mapping != mapping || page_offset(page) > size) {
4892 ret = VM_FAULT_NOPAGE;
4896 if (page->index == size >> PAGE_CACHE_SHIFT)
4897 len = size & ~PAGE_CACHE_MASK;
4899 len = PAGE_CACHE_SIZE;
4901 * Return if we have all the buffers mapped. This avoids the need to do
4902 * journal_start/journal_stop which can block and take a long time
4904 if (page_has_buffers(page)) {
4905 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4906 ext4_bh_unmapped)) {
4907 /* Wait so that we don't change page under IO */
4908 wait_on_page_writeback(page);
4909 ret = VM_FAULT_LOCKED;
4914 /* OK, we need to fill the hole... */
4915 if (ext4_should_dioread_nolock(inode))
4916 get_block = ext4_get_block_write;
4918 get_block = ext4_get_block;
4920 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4921 if (IS_ERR(handle)) {
4922 ret = VM_FAULT_SIGBUS;
4925 ret = __block_page_mkwrite(vma, vmf, get_block);
4926 if (!ret && ext4_should_journal_data(inode)) {
4927 if (walk_page_buffers(handle, page_buffers(page), 0,
4928 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4930 ret = VM_FAULT_SIGBUS;
4931 ext4_journal_stop(handle);
4934 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4936 ext4_journal_stop(handle);
4937 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4940 ret = block_page_mkwrite_return(ret);