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>
42 #include "ext4_jbd2.h"
45 #include "ext4_extents.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);
123 if (inode->i_nlink) {
125 * When journalling data dirty buffers are tracked only in the
126 * journal. So although mm thinks everything is clean and
127 * ready for reaping the inode might still have some pages to
128 * write in the running transaction or waiting to be
129 * checkpointed. Thus calling jbd2_journal_invalidatepage()
130 * (via truncate_inode_pages()) to discard these buffers can
131 * cause data loss. Also even if we did not discard these
132 * buffers, we would have no way to find them after the inode
133 * is reaped and thus user could see stale data if he tries to
134 * read them before the transaction is checkpointed. So be
135 * careful and force everything to disk here... We use
136 * ei->i_datasync_tid to store the newest transaction
137 * containing inode's data.
139 * Note that directories do not have this problem because they
140 * don't use page cache.
142 if (ext4_should_journal_data(inode) &&
143 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
144 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
145 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
147 jbd2_log_start_commit(journal, commit_tid);
148 jbd2_log_wait_commit(journal, commit_tid);
149 filemap_write_and_wait(&inode->i_data);
151 truncate_inode_pages(&inode->i_data, 0);
155 if (!is_bad_inode(inode))
156 dquot_initialize(inode);
158 if (ext4_should_order_data(inode))
159 ext4_begin_ordered_truncate(inode, 0);
160 truncate_inode_pages(&inode->i_data, 0);
162 if (is_bad_inode(inode))
165 handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
166 if (IS_ERR(handle)) {
167 ext4_std_error(inode->i_sb, PTR_ERR(handle));
169 * If we're going to skip the normal cleanup, we still need to
170 * make sure that the in-core orphan linked list is properly
173 ext4_orphan_del(NULL, inode);
178 ext4_handle_sync(handle);
180 err = ext4_mark_inode_dirty(handle, inode);
182 ext4_warning(inode->i_sb,
183 "couldn't mark inode dirty (err %d)", err);
187 ext4_truncate(inode);
190 * ext4_ext_truncate() doesn't reserve any slop when it
191 * restarts journal transactions; therefore there may not be
192 * enough credits left in the handle to remove the inode from
193 * the orphan list and set the dtime field.
195 if (!ext4_handle_has_enough_credits(handle, 3)) {
196 err = ext4_journal_extend(handle, 3);
198 err = ext4_journal_restart(handle, 3);
200 ext4_warning(inode->i_sb,
201 "couldn't extend journal (err %d)", err);
203 ext4_journal_stop(handle);
204 ext4_orphan_del(NULL, inode);
210 * Kill off the orphan record which ext4_truncate created.
211 * AKPM: I think this can be inside the above `if'.
212 * Note that ext4_orphan_del() has to be able to cope with the
213 * deletion of a non-existent orphan - this is because we don't
214 * know if ext4_truncate() actually created an orphan record.
215 * (Well, we could do this if we need to, but heck - it works)
217 ext4_orphan_del(handle, inode);
218 EXT4_I(inode)->i_dtime = get_seconds();
221 * One subtle ordering requirement: if anything has gone wrong
222 * (transaction abort, IO errors, whatever), then we can still
223 * do these next steps (the fs will already have been marked as
224 * having errors), but we can't free the inode if the mark_dirty
227 if (ext4_mark_inode_dirty(handle, inode))
228 /* If that failed, just do the required in-core inode clear. */
229 ext4_clear_inode(inode);
231 ext4_free_inode(handle, inode);
232 ext4_journal_stop(handle);
235 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
239 qsize_t *ext4_get_reserved_space(struct inode *inode)
241 return &EXT4_I(inode)->i_reserved_quota;
246 * Calculate the number of metadata blocks need to reserve
247 * to allocate a block located at @lblock
249 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
251 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
252 return ext4_ext_calc_metadata_amount(inode, lblock);
254 return ext4_ind_calc_metadata_amount(inode, lblock);
258 * Called with i_data_sem down, which is important since we can call
259 * ext4_discard_preallocations() from here.
261 void ext4_da_update_reserve_space(struct inode *inode,
262 int used, int quota_claim)
264 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
265 struct ext4_inode_info *ei = EXT4_I(inode);
267 spin_lock(&ei->i_block_reservation_lock);
268 trace_ext4_da_update_reserve_space(inode, used);
269 if (unlikely(used > ei->i_reserved_data_blocks)) {
270 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
271 "with only %d reserved data blocks\n",
272 __func__, inode->i_ino, used,
273 ei->i_reserved_data_blocks);
275 used = ei->i_reserved_data_blocks;
278 /* Update per-inode reservations */
279 ei->i_reserved_data_blocks -= used;
280 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
281 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
282 used + ei->i_allocated_meta_blocks);
283 ei->i_allocated_meta_blocks = 0;
285 if (ei->i_reserved_data_blocks == 0) {
287 * We can release all of the reserved metadata blocks
288 * only when we have written all of the delayed
291 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
292 ei->i_reserved_meta_blocks);
293 ei->i_reserved_meta_blocks = 0;
294 ei->i_da_metadata_calc_len = 0;
296 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
298 /* Update quota subsystem for data blocks */
300 dquot_claim_block(inode, used);
303 * We did fallocate with an offset that is already delayed
304 * allocated. So on delayed allocated writeback we should
305 * not re-claim the quota for fallocated blocks.
307 dquot_release_reservation_block(inode, used);
311 * If we have done all the pending block allocations and if
312 * there aren't any writers on the inode, we can discard the
313 * inode's preallocations.
315 if ((ei->i_reserved_data_blocks == 0) &&
316 (atomic_read(&inode->i_writecount) == 0))
317 ext4_discard_preallocations(inode);
320 static int __check_block_validity(struct inode *inode, const char *func,
322 struct ext4_map_blocks *map)
324 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
326 ext4_error_inode(inode, func, line, map->m_pblk,
327 "lblock %lu mapped to illegal pblock "
328 "(length %d)", (unsigned long) map->m_lblk,
335 #define check_block_validity(inode, map) \
336 __check_block_validity((inode), __func__, __LINE__, (map))
339 * Return the number of contiguous dirty pages in a given inode
340 * starting at page frame idx.
342 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
343 unsigned int max_pages)
345 struct address_space *mapping = inode->i_mapping;
349 int i, nr_pages, done = 0;
353 pagevec_init(&pvec, 0);
356 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
358 (pgoff_t)PAGEVEC_SIZE);
361 for (i = 0; i < nr_pages; i++) {
362 struct page *page = pvec.pages[i];
363 struct buffer_head *bh, *head;
366 if (unlikely(page->mapping != mapping) ||
368 PageWriteback(page) ||
369 page->index != idx) {
374 if (page_has_buffers(page)) {
375 bh = head = page_buffers(page);
377 if (!buffer_delay(bh) &&
378 !buffer_unwritten(bh))
380 bh = bh->b_this_page;
381 } while (!done && (bh != head));
388 if (num >= max_pages) {
393 pagevec_release(&pvec);
399 * The ext4_map_blocks() function tries to look up the requested blocks,
400 * and returns if the blocks are already mapped.
402 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
403 * and store the allocated blocks in the result buffer head and mark it
406 * If file type is extents based, it will call ext4_ext_map_blocks(),
407 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
410 * On success, it returns the number of blocks being mapped or allocate.
411 * if create==0 and the blocks are pre-allocated and uninitialized block,
412 * the result buffer head is unmapped. If the create ==1, it will make sure
413 * the buffer head is mapped.
415 * It returns 0 if plain look up failed (blocks have not been allocated), in
416 * that casem, buffer head is unmapped
418 * It returns the error in case of allocation failure.
420 int ext4_map_blocks(handle_t *handle, struct inode *inode,
421 struct ext4_map_blocks *map, int flags)
426 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
427 "logical block %lu\n", inode->i_ino, flags, map->m_len,
428 (unsigned long) map->m_lblk);
430 * Try to see if we can get the block without requesting a new
433 down_read((&EXT4_I(inode)->i_data_sem));
434 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
435 retval = ext4_ext_map_blocks(handle, inode, map, 0);
437 retval = ext4_ind_map_blocks(handle, inode, map, 0);
439 up_read((&EXT4_I(inode)->i_data_sem));
441 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
442 int ret = check_block_validity(inode, map);
447 /* If it is only a block(s) look up */
448 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
452 * Returns if the blocks have already allocated
454 * Note that if blocks have been preallocated
455 * ext4_ext_get_block() returns th create = 0
456 * with buffer head unmapped.
458 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
462 * When we call get_blocks without the create flag, the
463 * BH_Unwritten flag could have gotten set if the blocks
464 * requested were part of a uninitialized extent. We need to
465 * clear this flag now that we are committed to convert all or
466 * part of the uninitialized extent to be an initialized
467 * extent. This is because we need to avoid the combination
468 * of BH_Unwritten and BH_Mapped flags being simultaneously
469 * set on the buffer_head.
471 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
474 * New blocks allocate and/or writing to uninitialized extent
475 * will possibly result in updating i_data, so we take
476 * the write lock of i_data_sem, and call get_blocks()
477 * with create == 1 flag.
479 down_write((&EXT4_I(inode)->i_data_sem));
482 * if the caller is from delayed allocation writeout path
483 * we have already reserved fs blocks for allocation
484 * let the underlying get_block() function know to
485 * avoid double accounting
487 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
488 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
490 * We need to check for EXT4 here because migrate
491 * could have changed the inode type in between
493 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
494 retval = ext4_ext_map_blocks(handle, inode, map, flags);
496 retval = ext4_ind_map_blocks(handle, inode, map, flags);
498 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
500 * We allocated new blocks which will result in
501 * i_data's format changing. Force the migrate
502 * to fail by clearing migrate flags
504 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
508 * Update reserved blocks/metadata blocks after successful
509 * block allocation which had been deferred till now. We don't
510 * support fallocate for non extent files. So we can update
511 * reserve space here.
514 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
515 ext4_da_update_reserve_space(inode, retval, 1);
517 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
518 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
520 up_write((&EXT4_I(inode)->i_data_sem));
521 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
522 int ret = check_block_validity(inode, map);
529 /* Maximum number of blocks we map for direct IO at once. */
530 #define DIO_MAX_BLOCKS 4096
532 static int _ext4_get_block(struct inode *inode, sector_t iblock,
533 struct buffer_head *bh, int flags)
535 handle_t *handle = ext4_journal_current_handle();
536 struct ext4_map_blocks map;
537 int ret = 0, started = 0;
541 map.m_len = bh->b_size >> inode->i_blkbits;
543 if (flags && !handle) {
544 /* Direct IO write... */
545 if (map.m_len > DIO_MAX_BLOCKS)
546 map.m_len = DIO_MAX_BLOCKS;
547 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
548 handle = ext4_journal_start(inode, dio_credits);
549 if (IS_ERR(handle)) {
550 ret = PTR_ERR(handle);
556 ret = ext4_map_blocks(handle, inode, &map, flags);
558 map_bh(bh, inode->i_sb, map.m_pblk);
559 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
560 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
564 ext4_journal_stop(handle);
568 int ext4_get_block(struct inode *inode, sector_t iblock,
569 struct buffer_head *bh, int create)
571 return _ext4_get_block(inode, iblock, bh,
572 create ? EXT4_GET_BLOCKS_CREATE : 0);
576 * `handle' can be NULL if create is zero
578 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
579 ext4_lblk_t block, int create, int *errp)
581 struct ext4_map_blocks map;
582 struct buffer_head *bh;
585 J_ASSERT(handle != NULL || create == 0);
589 err = ext4_map_blocks(handle, inode, &map,
590 create ? EXT4_GET_BLOCKS_CREATE : 0);
598 bh = sb_getblk(inode->i_sb, map.m_pblk);
603 if (map.m_flags & EXT4_MAP_NEW) {
604 J_ASSERT(create != 0);
605 J_ASSERT(handle != NULL);
608 * Now that we do not always journal data, we should
609 * keep in mind whether this should always journal the
610 * new buffer as metadata. For now, regular file
611 * writes use ext4_get_block instead, so it's not a
615 BUFFER_TRACE(bh, "call get_create_access");
616 fatal = ext4_journal_get_create_access(handle, bh);
617 if (!fatal && !buffer_uptodate(bh)) {
618 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
619 set_buffer_uptodate(bh);
622 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
623 err = ext4_handle_dirty_metadata(handle, inode, bh);
627 BUFFER_TRACE(bh, "not a new buffer");
637 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
638 ext4_lblk_t block, int create, int *err)
640 struct buffer_head *bh;
642 bh = ext4_getblk(handle, inode, block, create, err);
645 if (buffer_uptodate(bh))
647 ll_rw_block(READ_META, 1, &bh);
649 if (buffer_uptodate(bh))
656 static int walk_page_buffers(handle_t *handle,
657 struct buffer_head *head,
661 int (*fn)(handle_t *handle,
662 struct buffer_head *bh))
664 struct buffer_head *bh;
665 unsigned block_start, block_end;
666 unsigned blocksize = head->b_size;
668 struct buffer_head *next;
670 for (bh = head, block_start = 0;
671 ret == 0 && (bh != head || !block_start);
672 block_start = block_end, bh = next) {
673 next = bh->b_this_page;
674 block_end = block_start + blocksize;
675 if (block_end <= from || block_start >= to) {
676 if (partial && !buffer_uptodate(bh))
680 err = (*fn)(handle, bh);
688 * To preserve ordering, it is essential that the hole instantiation and
689 * the data write be encapsulated in a single transaction. We cannot
690 * close off a transaction and start a new one between the ext4_get_block()
691 * and the commit_write(). So doing the jbd2_journal_start at the start of
692 * prepare_write() is the right place.
694 * Also, this function can nest inside ext4_writepage() ->
695 * block_write_full_page(). In that case, we *know* that ext4_writepage()
696 * has generated enough buffer credits to do the whole page. So we won't
697 * block on the journal in that case, which is good, because the caller may
700 * By accident, ext4 can be reentered when a transaction is open via
701 * quota file writes. If we were to commit the transaction while thus
702 * reentered, there can be a deadlock - we would be holding a quota
703 * lock, and the commit would never complete if another thread had a
704 * transaction open and was blocking on the quota lock - a ranking
707 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
708 * will _not_ run commit under these circumstances because handle->h_ref
709 * is elevated. We'll still have enough credits for the tiny quotafile
712 static int do_journal_get_write_access(handle_t *handle,
713 struct buffer_head *bh)
715 int dirty = buffer_dirty(bh);
718 if (!buffer_mapped(bh) || buffer_freed(bh))
721 * __block_write_begin() could have dirtied some buffers. Clean
722 * the dirty bit as jbd2_journal_get_write_access() could complain
723 * otherwise about fs integrity issues. Setting of the dirty bit
724 * by __block_write_begin() isn't a real problem here as we clear
725 * the bit before releasing a page lock and thus writeback cannot
726 * ever write the buffer.
729 clear_buffer_dirty(bh);
730 ret = ext4_journal_get_write_access(handle, bh);
732 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
736 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
737 struct buffer_head *bh_result, int create);
738 static int ext4_write_begin(struct file *file, struct address_space *mapping,
739 loff_t pos, unsigned len, unsigned flags,
740 struct page **pagep, void **fsdata)
742 struct inode *inode = mapping->host;
743 int ret, needed_blocks;
750 trace_ext4_write_begin(inode, pos, len, flags);
752 * Reserve one block more for addition to orphan list in case
753 * we allocate blocks but write fails for some reason
755 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
756 index = pos >> PAGE_CACHE_SHIFT;
757 from = pos & (PAGE_CACHE_SIZE - 1);
761 handle = ext4_journal_start(inode, needed_blocks);
762 if (IS_ERR(handle)) {
763 ret = PTR_ERR(handle);
767 /* We cannot recurse into the filesystem as the transaction is already
769 flags |= AOP_FLAG_NOFS;
771 page = grab_cache_page_write_begin(mapping, index, flags);
773 ext4_journal_stop(handle);
779 if (ext4_should_dioread_nolock(inode))
780 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
782 ret = __block_write_begin(page, pos, len, ext4_get_block);
784 if (!ret && ext4_should_journal_data(inode)) {
785 ret = walk_page_buffers(handle, page_buffers(page),
786 from, to, NULL, do_journal_get_write_access);
791 page_cache_release(page);
793 * __block_write_begin may have instantiated a few blocks
794 * outside i_size. Trim these off again. Don't need
795 * i_size_read because we hold i_mutex.
797 * Add inode to orphan list in case we crash before
800 if (pos + len > inode->i_size && ext4_can_truncate(inode))
801 ext4_orphan_add(handle, inode);
803 ext4_journal_stop(handle);
804 if (pos + len > inode->i_size) {
805 ext4_truncate_failed_write(inode);
807 * If truncate failed early the inode might
808 * still be on the orphan list; we need to
809 * make sure the inode is removed from the
810 * orphan list in that case.
813 ext4_orphan_del(NULL, inode);
817 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
823 /* For write_end() in data=journal mode */
824 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
826 if (!buffer_mapped(bh) || buffer_freed(bh))
828 set_buffer_uptodate(bh);
829 return ext4_handle_dirty_metadata(handle, NULL, bh);
832 static int ext4_generic_write_end(struct file *file,
833 struct address_space *mapping,
834 loff_t pos, unsigned len, unsigned copied,
835 struct page *page, void *fsdata)
837 int i_size_changed = 0;
838 struct inode *inode = mapping->host;
839 handle_t *handle = ext4_journal_current_handle();
841 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
844 * No need to use i_size_read() here, the i_size
845 * cannot change under us because we hold i_mutex.
847 * But it's important to update i_size while still holding page lock:
848 * page writeout could otherwise come in and zero beyond i_size.
850 if (pos + copied > inode->i_size) {
851 i_size_write(inode, pos + copied);
855 if (pos + copied > EXT4_I(inode)->i_disksize) {
856 /* We need to mark inode dirty even if
857 * new_i_size is less that inode->i_size
858 * bu greater than i_disksize.(hint delalloc)
860 ext4_update_i_disksize(inode, (pos + copied));
864 page_cache_release(page);
867 * Don't mark the inode dirty under page lock. First, it unnecessarily
868 * makes the holding time of page lock longer. Second, it forces lock
869 * ordering of page lock and transaction start for journaling
873 ext4_mark_inode_dirty(handle, inode);
879 * We need to pick up the new inode size which generic_commit_write gave us
880 * `file' can be NULL - eg, when called from page_symlink().
882 * ext4 never places buffers on inode->i_mapping->private_list. metadata
883 * buffers are managed internally.
885 static int ext4_ordered_write_end(struct file *file,
886 struct address_space *mapping,
887 loff_t pos, unsigned len, unsigned copied,
888 struct page *page, void *fsdata)
890 handle_t *handle = ext4_journal_current_handle();
891 struct inode *inode = mapping->host;
894 trace_ext4_ordered_write_end(inode, pos, len, copied);
895 ret = ext4_jbd2_file_inode(handle, inode);
898 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
901 if (pos + len > inode->i_size && ext4_can_truncate(inode))
902 /* if we have allocated more blocks and copied
903 * less. We will have blocks allocated outside
904 * inode->i_size. So truncate them
906 ext4_orphan_add(handle, inode);
910 ret2 = ext4_journal_stop(handle);
914 if (pos + len > inode->i_size) {
915 ext4_truncate_failed_write(inode);
917 * If truncate failed early the inode might still be
918 * on the orphan list; we need to make sure the inode
919 * is removed from the orphan list in that case.
922 ext4_orphan_del(NULL, inode);
926 return ret ? ret : copied;
929 static int ext4_writeback_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 handle_t *handle = ext4_journal_current_handle();
935 struct inode *inode = mapping->host;
938 trace_ext4_writeback_write_end(inode, pos, len, copied);
939 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
942 if (pos + len > inode->i_size && ext4_can_truncate(inode))
943 /* if we have allocated more blocks and copied
944 * less. We will have blocks allocated outside
945 * inode->i_size. So truncate them
947 ext4_orphan_add(handle, inode);
952 ret2 = ext4_journal_stop(handle);
956 if (pos + len > inode->i_size) {
957 ext4_truncate_failed_write(inode);
959 * If truncate failed early the inode might still be
960 * on the orphan list; we need to make sure the inode
961 * is removed from the orphan list in that case.
964 ext4_orphan_del(NULL, inode);
967 return ret ? ret : copied;
970 static int ext4_journalled_write_end(struct file *file,
971 struct address_space *mapping,
972 loff_t pos, unsigned len, unsigned copied,
973 struct page *page, void *fsdata)
975 handle_t *handle = ext4_journal_current_handle();
976 struct inode *inode = mapping->host;
982 trace_ext4_journalled_write_end(inode, pos, len, copied);
983 from = pos & (PAGE_CACHE_SIZE - 1);
986 BUG_ON(!ext4_handle_valid(handle));
989 if (!PageUptodate(page))
991 page_zero_new_buffers(page, from+copied, to);
994 ret = walk_page_buffers(handle, page_buffers(page), from,
995 to, &partial, write_end_fn);
997 SetPageUptodate(page);
998 new_i_size = pos + copied;
999 if (new_i_size > inode->i_size)
1000 i_size_write(inode, pos+copied);
1001 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1002 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1003 if (new_i_size > EXT4_I(inode)->i_disksize) {
1004 ext4_update_i_disksize(inode, new_i_size);
1005 ret2 = ext4_mark_inode_dirty(handle, inode);
1011 page_cache_release(page);
1012 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1013 /* if we have allocated more blocks and copied
1014 * less. We will have blocks allocated outside
1015 * inode->i_size. So truncate them
1017 ext4_orphan_add(handle, inode);
1019 ret2 = ext4_journal_stop(handle);
1022 if (pos + len > inode->i_size) {
1023 ext4_truncate_failed_write(inode);
1025 * If truncate failed early the inode might still be
1026 * on the orphan list; we need to make sure the inode
1027 * is removed from the orphan list in that case.
1030 ext4_orphan_del(NULL, inode);
1033 return ret ? ret : copied;
1037 * Reserve a single block located at lblock
1039 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1042 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1043 struct ext4_inode_info *ei = EXT4_I(inode);
1044 unsigned long md_needed;
1048 * recalculate the amount of metadata blocks to reserve
1049 * in order to allocate nrblocks
1050 * worse case is one extent per block
1053 spin_lock(&ei->i_block_reservation_lock);
1054 md_needed = ext4_calc_metadata_amount(inode, lblock);
1055 trace_ext4_da_reserve_space(inode, md_needed);
1056 spin_unlock(&ei->i_block_reservation_lock);
1059 * We will charge metadata quota at writeout time; this saves
1060 * us from metadata over-estimation, though we may go over by
1061 * a small amount in the end. Here we just reserve for data.
1063 ret = dquot_reserve_block(inode, 1);
1067 * We do still charge estimated metadata to the sb though;
1068 * we cannot afford to run out of free blocks.
1070 if (ext4_claim_free_blocks(sbi, md_needed + 1, 0)) {
1071 dquot_release_reservation_block(inode, 1);
1072 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1078 spin_lock(&ei->i_block_reservation_lock);
1079 ei->i_reserved_data_blocks++;
1080 ei->i_reserved_meta_blocks += md_needed;
1081 spin_unlock(&ei->i_block_reservation_lock);
1083 return 0; /* success */
1086 static void ext4_da_release_space(struct inode *inode, int to_free)
1088 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1089 struct ext4_inode_info *ei = EXT4_I(inode);
1092 return; /* Nothing to release, exit */
1094 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1096 trace_ext4_da_release_space(inode, to_free);
1097 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1099 * if there aren't enough reserved blocks, then the
1100 * counter is messed up somewhere. Since this
1101 * function is called from invalidate page, it's
1102 * harmless to return without any action.
1104 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1105 "ino %lu, to_free %d with only %d reserved "
1106 "data blocks\n", inode->i_ino, to_free,
1107 ei->i_reserved_data_blocks);
1109 to_free = ei->i_reserved_data_blocks;
1111 ei->i_reserved_data_blocks -= to_free;
1113 if (ei->i_reserved_data_blocks == 0) {
1115 * We can release all of the reserved metadata blocks
1116 * only when we have written all of the delayed
1117 * allocation blocks.
1119 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1120 ei->i_reserved_meta_blocks);
1121 ei->i_reserved_meta_blocks = 0;
1122 ei->i_da_metadata_calc_len = 0;
1125 /* update fs dirty data blocks counter */
1126 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1128 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1130 dquot_release_reservation_block(inode, to_free);
1133 static void ext4_da_page_release_reservation(struct page *page,
1134 unsigned long offset)
1137 struct buffer_head *head, *bh;
1138 unsigned int curr_off = 0;
1140 head = page_buffers(page);
1143 unsigned int next_off = curr_off + bh->b_size;
1145 if ((offset <= curr_off) && (buffer_delay(bh))) {
1147 clear_buffer_delay(bh);
1149 curr_off = next_off;
1150 } while ((bh = bh->b_this_page) != head);
1151 ext4_da_release_space(page->mapping->host, to_release);
1155 * Delayed allocation stuff
1159 * mpage_da_submit_io - walks through extent of pages and try to write
1160 * them with writepage() call back
1162 * @mpd->inode: inode
1163 * @mpd->first_page: first page of the extent
1164 * @mpd->next_page: page after the last page of the extent
1166 * By the time mpage_da_submit_io() is called we expect all blocks
1167 * to be allocated. this may be wrong if allocation failed.
1169 * As pages are already locked by write_cache_pages(), we can't use it
1171 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1172 struct ext4_map_blocks *map)
1174 struct pagevec pvec;
1175 unsigned long index, end;
1176 int ret = 0, err, nr_pages, i;
1177 struct inode *inode = mpd->inode;
1178 struct address_space *mapping = inode->i_mapping;
1179 loff_t size = i_size_read(inode);
1180 unsigned int len, block_start;
1181 struct buffer_head *bh, *page_bufs = NULL;
1182 int journal_data = ext4_should_journal_data(inode);
1183 sector_t pblock = 0, cur_logical = 0;
1184 struct ext4_io_submit io_submit;
1186 BUG_ON(mpd->next_page <= mpd->first_page);
1187 memset(&io_submit, 0, sizeof(io_submit));
1189 * We need to start from the first_page to the next_page - 1
1190 * to make sure we also write the mapped dirty buffer_heads.
1191 * If we look at mpd->b_blocknr we would only be looking
1192 * at the currently mapped buffer_heads.
1194 index = mpd->first_page;
1195 end = mpd->next_page - 1;
1197 pagevec_init(&pvec, 0);
1198 while (index <= end) {
1199 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1202 for (i = 0; i < nr_pages; i++) {
1203 int commit_write = 0, skip_page = 0;
1204 struct page *page = pvec.pages[i];
1206 index = page->index;
1210 if (index == size >> PAGE_CACHE_SHIFT)
1211 len = size & ~PAGE_CACHE_MASK;
1213 len = PAGE_CACHE_SIZE;
1215 cur_logical = index << (PAGE_CACHE_SHIFT -
1217 pblock = map->m_pblk + (cur_logical -
1222 BUG_ON(!PageLocked(page));
1223 BUG_ON(PageWriteback(page));
1226 * If the page does not have buffers (for
1227 * whatever reason), try to create them using
1228 * __block_write_begin. If this fails,
1229 * skip the page and move on.
1231 if (!page_has_buffers(page)) {
1232 if (__block_write_begin(page, 0, len,
1233 noalloc_get_block_write)) {
1241 bh = page_bufs = page_buffers(page);
1246 if (map && (cur_logical >= map->m_lblk) &&
1247 (cur_logical <= (map->m_lblk +
1248 (map->m_len - 1)))) {
1249 if (buffer_delay(bh)) {
1250 clear_buffer_delay(bh);
1251 bh->b_blocknr = pblock;
1253 if (buffer_unwritten(bh) ||
1255 BUG_ON(bh->b_blocknr != pblock);
1256 if (map->m_flags & EXT4_MAP_UNINIT)
1257 set_buffer_uninit(bh);
1258 clear_buffer_unwritten(bh);
1261 /* skip page if block allocation undone */
1262 if (buffer_delay(bh) || buffer_unwritten(bh))
1264 bh = bh->b_this_page;
1265 block_start += bh->b_size;
1268 } while (bh != page_bufs);
1274 /* mark the buffer_heads as dirty & uptodate */
1275 block_commit_write(page, 0, len);
1277 clear_page_dirty_for_io(page);
1279 * Delalloc doesn't support data journalling,
1280 * but eventually maybe we'll lift this
1283 if (unlikely(journal_data && PageChecked(page)))
1284 err = __ext4_journalled_writepage(page, len);
1285 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1286 err = ext4_bio_write_page(&io_submit, page,
1289 err = block_write_full_page(page,
1290 noalloc_get_block_write, mpd->wbc);
1293 mpd->pages_written++;
1295 * In error case, we have to continue because
1296 * remaining pages are still locked
1301 pagevec_release(&pvec);
1303 ext4_io_submit(&io_submit);
1307 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1311 struct pagevec pvec;
1312 struct inode *inode = mpd->inode;
1313 struct address_space *mapping = inode->i_mapping;
1315 index = mpd->first_page;
1316 end = mpd->next_page - 1;
1317 while (index <= end) {
1318 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1321 for (i = 0; i < nr_pages; i++) {
1322 struct page *page = pvec.pages[i];
1323 if (page->index > end)
1325 BUG_ON(!PageLocked(page));
1326 BUG_ON(PageWriteback(page));
1327 block_invalidatepage(page, 0);
1328 ClearPageUptodate(page);
1331 index = pvec.pages[nr_pages - 1]->index + 1;
1332 pagevec_release(&pvec);
1337 static void ext4_print_free_blocks(struct inode *inode)
1339 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1340 printk(KERN_CRIT "Total free blocks count %lld\n",
1341 ext4_count_free_blocks(inode->i_sb));
1342 printk(KERN_CRIT "Free/Dirty block details\n");
1343 printk(KERN_CRIT "free_blocks=%lld\n",
1344 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
1345 printk(KERN_CRIT "dirty_blocks=%lld\n",
1346 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1347 printk(KERN_CRIT "Block reservation details\n");
1348 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
1349 EXT4_I(inode)->i_reserved_data_blocks);
1350 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
1351 EXT4_I(inode)->i_reserved_meta_blocks);
1356 * mpage_da_map_and_submit - go through given space, map them
1357 * if necessary, and then submit them for I/O
1359 * @mpd - bh describing space
1361 * The function skips space we know is already mapped to disk blocks.
1364 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1366 int err, blks, get_blocks_flags;
1367 struct ext4_map_blocks map, *mapp = NULL;
1368 sector_t next = mpd->b_blocknr;
1369 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1370 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1371 handle_t *handle = NULL;
1374 * If the blocks are mapped already, or we couldn't accumulate
1375 * any blocks, then proceed immediately to the submission stage.
1377 if ((mpd->b_size == 0) ||
1378 ((mpd->b_state & (1 << BH_Mapped)) &&
1379 !(mpd->b_state & (1 << BH_Delay)) &&
1380 !(mpd->b_state & (1 << BH_Unwritten))))
1383 handle = ext4_journal_current_handle();
1387 * Call ext4_map_blocks() to allocate any delayed allocation
1388 * blocks, or to convert an uninitialized extent to be
1389 * initialized (in the case where we have written into
1390 * one or more preallocated blocks).
1392 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1393 * indicate that we are on the delayed allocation path. This
1394 * affects functions in many different parts of the allocation
1395 * call path. This flag exists primarily because we don't
1396 * want to change *many* call functions, so ext4_map_blocks()
1397 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1398 * inode's allocation semaphore is taken.
1400 * If the blocks in questions were delalloc blocks, set
1401 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1402 * variables are updated after the blocks have been allocated.
1405 map.m_len = max_blocks;
1406 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1407 if (ext4_should_dioread_nolock(mpd->inode))
1408 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1409 if (mpd->b_state & (1 << BH_Delay))
1410 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1412 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1414 struct super_block *sb = mpd->inode->i_sb;
1418 * If get block returns EAGAIN or ENOSPC and there
1419 * appears to be free blocks we will just let
1420 * mpage_da_submit_io() unlock all of the pages.
1425 if (err == -ENOSPC &&
1426 ext4_count_free_blocks(sb)) {
1432 * get block failure will cause us to loop in
1433 * writepages, because a_ops->writepage won't be able
1434 * to make progress. The page will be redirtied by
1435 * writepage and writepages will again try to write
1438 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1439 ext4_msg(sb, KERN_CRIT,
1440 "delayed block allocation failed for inode %lu "
1441 "at logical offset %llu with max blocks %zd "
1442 "with error %d", mpd->inode->i_ino,
1443 (unsigned long long) next,
1444 mpd->b_size >> mpd->inode->i_blkbits, err);
1445 ext4_msg(sb, KERN_CRIT,
1446 "This should not happen!! Data will be lost\n");
1448 ext4_print_free_blocks(mpd->inode);
1450 /* invalidate all the pages */
1451 ext4_da_block_invalidatepages(mpd);
1453 /* Mark this page range as having been completed */
1460 if (map.m_flags & EXT4_MAP_NEW) {
1461 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1464 for (i = 0; i < map.m_len; i++)
1465 unmap_underlying_metadata(bdev, map.m_pblk + i);
1468 if (ext4_should_order_data(mpd->inode)) {
1469 err = ext4_jbd2_file_inode(handle, mpd->inode);
1471 /* This only happens if the journal is aborted */
1476 * Update on-disk size along with block allocation.
1478 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1479 if (disksize > i_size_read(mpd->inode))
1480 disksize = i_size_read(mpd->inode);
1481 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1482 ext4_update_i_disksize(mpd->inode, disksize);
1483 err = ext4_mark_inode_dirty(handle, mpd->inode);
1485 ext4_error(mpd->inode->i_sb,
1486 "Failed to mark inode %lu dirty",
1491 mpage_da_submit_io(mpd, mapp);
1495 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1496 (1 << BH_Delay) | (1 << BH_Unwritten))
1499 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1501 * @mpd->lbh - extent of blocks
1502 * @logical - logical number of the block in the file
1503 * @bh - bh of the block (used to access block's state)
1505 * the function is used to collect contig. blocks in same state
1507 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1508 sector_t logical, size_t b_size,
1509 unsigned long b_state)
1512 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1515 * XXX Don't go larger than mballoc is willing to allocate
1516 * This is a stopgap solution. We eventually need to fold
1517 * mpage_da_submit_io() into this function and then call
1518 * ext4_map_blocks() multiple times in a loop
1520 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1523 /* check if thereserved journal credits might overflow */
1524 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1525 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1527 * With non-extent format we are limited by the journal
1528 * credit available. Total credit needed to insert
1529 * nrblocks contiguous blocks is dependent on the
1530 * nrblocks. So limit nrblocks.
1533 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1534 EXT4_MAX_TRANS_DATA) {
1536 * Adding the new buffer_head would make it cross the
1537 * allowed limit for which we have journal credit
1538 * reserved. So limit the new bh->b_size
1540 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1541 mpd->inode->i_blkbits;
1542 /* we will do mpage_da_submit_io in the next loop */
1546 * First block in the extent
1548 if (mpd->b_size == 0) {
1549 mpd->b_blocknr = logical;
1550 mpd->b_size = b_size;
1551 mpd->b_state = b_state & BH_FLAGS;
1555 next = mpd->b_blocknr + nrblocks;
1557 * Can we merge the block to our big extent?
1559 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1560 mpd->b_size += b_size;
1566 * We couldn't merge the block to our extent, so we
1567 * need to flush current extent and start new one
1569 mpage_da_map_and_submit(mpd);
1573 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1575 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1579 * This is a special get_blocks_t callback which is used by
1580 * ext4_da_write_begin(). It will either return mapped block or
1581 * reserve space for a single block.
1583 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1584 * We also have b_blocknr = -1 and b_bdev initialized properly
1586 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1587 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1588 * initialized properly.
1590 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1591 struct buffer_head *bh, int create)
1593 struct ext4_map_blocks map;
1595 sector_t invalid_block = ~((sector_t) 0xffff);
1597 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1600 BUG_ON(create == 0);
1601 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1603 map.m_lblk = iblock;
1607 * first, we need to know whether the block is allocated already
1608 * preallocated blocks are unmapped but should treated
1609 * the same as allocated blocks.
1611 ret = ext4_map_blocks(NULL, inode, &map, 0);
1615 if (buffer_delay(bh))
1616 return 0; /* Not sure this could or should happen */
1618 * XXX: __block_write_begin() unmaps passed block, is it OK?
1620 ret = ext4_da_reserve_space(inode, iblock);
1622 /* not enough space to reserve */
1625 map_bh(bh, inode->i_sb, invalid_block);
1627 set_buffer_delay(bh);
1631 map_bh(bh, inode->i_sb, map.m_pblk);
1632 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1634 if (buffer_unwritten(bh)) {
1635 /* A delayed write to unwritten bh should be marked
1636 * new and mapped. Mapped ensures that we don't do
1637 * get_block multiple times when we write to the same
1638 * offset and new ensures that we do proper zero out
1639 * for partial write.
1642 set_buffer_mapped(bh);
1648 * This function is used as a standard get_block_t calback function
1649 * when there is no desire to allocate any blocks. It is used as a
1650 * callback function for block_write_begin() and block_write_full_page().
1651 * These functions should only try to map a single block at a time.
1653 * Since this function doesn't do block allocations even if the caller
1654 * requests it by passing in create=1, it is critically important that
1655 * any caller checks to make sure that any buffer heads are returned
1656 * by this function are either all already mapped or marked for
1657 * delayed allocation before calling block_write_full_page(). Otherwise,
1658 * b_blocknr could be left unitialized, and the page write functions will
1659 * be taken by surprise.
1661 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1662 struct buffer_head *bh_result, int create)
1664 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1665 return _ext4_get_block(inode, iblock, bh_result, 0);
1668 static int bget_one(handle_t *handle, struct buffer_head *bh)
1674 static int bput_one(handle_t *handle, struct buffer_head *bh)
1680 static int __ext4_journalled_writepage(struct page *page,
1683 struct address_space *mapping = page->mapping;
1684 struct inode *inode = mapping->host;
1685 struct buffer_head *page_bufs;
1686 handle_t *handle = NULL;
1690 ClearPageChecked(page);
1691 page_bufs = page_buffers(page);
1693 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1694 /* As soon as we unlock the page, it can go away, but we have
1695 * references to buffers so we are safe */
1698 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1699 if (IS_ERR(handle)) {
1700 ret = PTR_ERR(handle);
1704 BUG_ON(!ext4_handle_valid(handle));
1706 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1707 do_journal_get_write_access);
1709 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1713 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1714 err = ext4_journal_stop(handle);
1718 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
1719 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1724 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
1725 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
1728 * Note that we don't need to start a transaction unless we're journaling data
1729 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1730 * need to file the inode to the transaction's list in ordered mode because if
1731 * we are writing back data added by write(), the inode is already there and if
1732 * we are writing back data modified via mmap(), no one guarantees in which
1733 * transaction the data will hit the disk. In case we are journaling data, we
1734 * cannot start transaction directly because transaction start ranks above page
1735 * lock so we have to do some magic.
1737 * This function can get called via...
1738 * - ext4_da_writepages after taking page lock (have journal handle)
1739 * - journal_submit_inode_data_buffers (no journal handle)
1740 * - shrink_page_list via pdflush (no journal handle)
1741 * - grab_page_cache when doing write_begin (have journal handle)
1743 * We don't do any block allocation in this function. If we have page with
1744 * multiple blocks we need to write those buffer_heads that are mapped. This
1745 * is important for mmaped based write. So if we do with blocksize 1K
1746 * truncate(f, 1024);
1747 * a = mmap(f, 0, 4096);
1749 * truncate(f, 4096);
1750 * we have in the page first buffer_head mapped via page_mkwrite call back
1751 * but other bufer_heads would be unmapped but dirty(dirty done via the
1752 * do_wp_page). So writepage should write the first block. If we modify
1753 * the mmap area beyond 1024 we will again get a page_fault and the
1754 * page_mkwrite callback will do the block allocation and mark the
1755 * buffer_heads mapped.
1757 * We redirty the page if we have any buffer_heads that is either delay or
1758 * unwritten in the page.
1760 * We can get recursively called as show below.
1762 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1765 * But since we don't do any block allocation we should not deadlock.
1766 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1768 static int ext4_writepage(struct page *page,
1769 struct writeback_control *wbc)
1771 int ret = 0, commit_write = 0;
1774 struct buffer_head *page_bufs = NULL;
1775 struct inode *inode = page->mapping->host;
1777 trace_ext4_writepage(page);
1778 size = i_size_read(inode);
1779 if (page->index == size >> PAGE_CACHE_SHIFT)
1780 len = size & ~PAGE_CACHE_MASK;
1782 len = PAGE_CACHE_SIZE;
1785 * If the page does not have buffers (for whatever reason),
1786 * try to create them using __block_write_begin. If this
1787 * fails, redirty the page and move on.
1789 if (!page_has_buffers(page)) {
1790 if (__block_write_begin(page, 0, len,
1791 noalloc_get_block_write)) {
1793 redirty_page_for_writepage(wbc, page);
1799 page_bufs = page_buffers(page);
1800 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1801 ext4_bh_delay_or_unwritten)) {
1803 * We don't want to do block allocation, so redirty
1804 * the page and return. We may reach here when we do
1805 * a journal commit via journal_submit_inode_data_buffers.
1806 * We can also reach here via shrink_page_list
1811 /* now mark the buffer_heads as dirty and uptodate */
1812 block_commit_write(page, 0, len);
1814 if (PageChecked(page) && ext4_should_journal_data(inode))
1816 * It's mmapped pagecache. Add buffers and journal it. There
1817 * doesn't seem much point in redirtying the page here.
1819 return __ext4_journalled_writepage(page, len);
1821 if (buffer_uninit(page_bufs)) {
1822 ext4_set_bh_endio(page_bufs, inode);
1823 ret = block_write_full_page_endio(page, noalloc_get_block_write,
1824 wbc, ext4_end_io_buffer_write);
1826 ret = block_write_full_page(page, noalloc_get_block_write,
1833 * This is called via ext4_da_writepages() to
1834 * calculate the total number of credits to reserve to fit
1835 * a single extent allocation into a single transaction,
1836 * ext4_da_writpeages() will loop calling this before
1837 * the block allocation.
1840 static int ext4_da_writepages_trans_blocks(struct inode *inode)
1842 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
1845 * With non-extent format the journal credit needed to
1846 * insert nrblocks contiguous block is dependent on
1847 * number of contiguous block. So we will limit
1848 * number of contiguous block to a sane value
1850 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
1851 (max_blocks > EXT4_MAX_TRANS_DATA))
1852 max_blocks = EXT4_MAX_TRANS_DATA;
1854 return ext4_chunk_trans_blocks(inode, max_blocks);
1858 * write_cache_pages_da - walk the list of dirty pages of the given
1859 * address space and accumulate pages that need writing, and call
1860 * mpage_da_map_and_submit to map a single contiguous memory region
1861 * and then write them.
1863 static int write_cache_pages_da(struct address_space *mapping,
1864 struct writeback_control *wbc,
1865 struct mpage_da_data *mpd,
1866 pgoff_t *done_index)
1868 struct buffer_head *bh, *head;
1869 struct inode *inode = mapping->host;
1870 struct pagevec pvec;
1871 unsigned int nr_pages;
1874 long nr_to_write = wbc->nr_to_write;
1875 int i, tag, ret = 0;
1877 memset(mpd, 0, sizeof(struct mpage_da_data));
1880 pagevec_init(&pvec, 0);
1881 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1882 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1884 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1885 tag = PAGECACHE_TAG_TOWRITE;
1887 tag = PAGECACHE_TAG_DIRTY;
1889 *done_index = index;
1890 while (index <= end) {
1891 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1892 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1896 for (i = 0; i < nr_pages; i++) {
1897 struct page *page = pvec.pages[i];
1900 * At this point, the page may be truncated or
1901 * invalidated (changing page->mapping to NULL), or
1902 * even swizzled back from swapper_space to tmpfs file
1903 * mapping. However, page->index will not change
1904 * because we have a reference on the page.
1906 if (page->index > end)
1909 *done_index = page->index + 1;
1912 * If we can't merge this page, and we have
1913 * accumulated an contiguous region, write it
1915 if ((mpd->next_page != page->index) &&
1916 (mpd->next_page != mpd->first_page)) {
1917 mpage_da_map_and_submit(mpd);
1918 goto ret_extent_tail;
1924 * If the page is no longer dirty, or its
1925 * mapping no longer corresponds to inode we
1926 * are writing (which means it has been
1927 * truncated or invalidated), or the page is
1928 * already under writeback and we are not
1929 * doing a data integrity writeback, skip the page
1931 if (!PageDirty(page) ||
1932 (PageWriteback(page) &&
1933 (wbc->sync_mode == WB_SYNC_NONE)) ||
1934 unlikely(page->mapping != mapping)) {
1939 wait_on_page_writeback(page);
1940 BUG_ON(PageWriteback(page));
1942 if (mpd->next_page != page->index)
1943 mpd->first_page = page->index;
1944 mpd->next_page = page->index + 1;
1945 logical = (sector_t) page->index <<
1946 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1948 if (!page_has_buffers(page)) {
1949 mpage_add_bh_to_extent(mpd, logical,
1951 (1 << BH_Dirty) | (1 << BH_Uptodate));
1953 goto ret_extent_tail;
1956 * Page with regular buffer heads,
1957 * just add all dirty ones
1959 head = page_buffers(page);
1962 BUG_ON(buffer_locked(bh));
1964 * We need to try to allocate
1965 * unmapped blocks in the same page.
1966 * Otherwise we won't make progress
1967 * with the page in ext4_writepage
1969 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
1970 mpage_add_bh_to_extent(mpd, logical,
1974 goto ret_extent_tail;
1975 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
1977 * mapped dirty buffer. We need
1978 * to update the b_state
1979 * because we look at b_state
1980 * in mpage_da_map_blocks. We
1981 * don't update b_size because
1982 * if we find an unmapped
1983 * buffer_head later we need to
1984 * use the b_state flag of that
1987 if (mpd->b_size == 0)
1988 mpd->b_state = bh->b_state & BH_FLAGS;
1991 } while ((bh = bh->b_this_page) != head);
1994 if (nr_to_write > 0) {
1996 if (nr_to_write == 0 &&
1997 wbc->sync_mode == WB_SYNC_NONE)
1999 * We stop writing back only if we are
2000 * not doing integrity sync. In case of
2001 * integrity sync we have to keep going
2002 * because someone may be concurrently
2003 * dirtying pages, and we might have
2004 * synced a lot of newly appeared dirty
2005 * pages, but have not synced all of the
2011 pagevec_release(&pvec);
2016 ret = MPAGE_DA_EXTENT_TAIL;
2018 pagevec_release(&pvec);
2024 static int ext4_da_writepages(struct address_space *mapping,
2025 struct writeback_control *wbc)
2028 int range_whole = 0;
2029 handle_t *handle = NULL;
2030 struct mpage_da_data mpd;
2031 struct inode *inode = mapping->host;
2032 int pages_written = 0;
2033 unsigned int max_pages;
2034 int range_cyclic, cycled = 1, io_done = 0;
2035 int needed_blocks, ret = 0;
2036 long desired_nr_to_write, nr_to_writebump = 0;
2037 loff_t range_start = wbc->range_start;
2038 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2039 pgoff_t done_index = 0;
2042 trace_ext4_da_writepages(inode, wbc);
2045 * No pages to write? This is mainly a kludge to avoid starting
2046 * a transaction for special inodes like journal inode on last iput()
2047 * because that could violate lock ordering on umount
2049 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2053 * If the filesystem has aborted, it is read-only, so return
2054 * right away instead of dumping stack traces later on that
2055 * will obscure the real source of the problem. We test
2056 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2057 * the latter could be true if the filesystem is mounted
2058 * read-only, and in that case, ext4_da_writepages should
2059 * *never* be called, so if that ever happens, we would want
2062 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2065 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2068 range_cyclic = wbc->range_cyclic;
2069 if (wbc->range_cyclic) {
2070 index = mapping->writeback_index;
2073 wbc->range_start = index << PAGE_CACHE_SHIFT;
2074 wbc->range_end = LLONG_MAX;
2075 wbc->range_cyclic = 0;
2078 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2079 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2083 * This works around two forms of stupidity. The first is in
2084 * the writeback code, which caps the maximum number of pages
2085 * written to be 1024 pages. This is wrong on multiple
2086 * levels; different architectues have a different page size,
2087 * which changes the maximum amount of data which gets
2088 * written. Secondly, 4 megabytes is way too small. XFS
2089 * forces this value to be 16 megabytes by multiplying
2090 * nr_to_write parameter by four, and then relies on its
2091 * allocator to allocate larger extents to make them
2092 * contiguous. Unfortunately this brings us to the second
2093 * stupidity, which is that ext4's mballoc code only allocates
2094 * at most 2048 blocks. So we force contiguous writes up to
2095 * the number of dirty blocks in the inode, or
2096 * sbi->max_writeback_mb_bump whichever is smaller.
2098 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2099 if (!range_cyclic && range_whole) {
2100 if (wbc->nr_to_write == LONG_MAX)
2101 desired_nr_to_write = wbc->nr_to_write;
2103 desired_nr_to_write = wbc->nr_to_write * 8;
2105 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2107 if (desired_nr_to_write > max_pages)
2108 desired_nr_to_write = max_pages;
2110 if (wbc->nr_to_write < desired_nr_to_write) {
2111 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2112 wbc->nr_to_write = desired_nr_to_write;
2116 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2117 tag_pages_for_writeback(mapping, index, end);
2119 while (!ret && wbc->nr_to_write > 0) {
2122 * we insert one extent at a time. So we need
2123 * credit needed for single extent allocation.
2124 * journalled mode is currently not supported
2127 BUG_ON(ext4_should_journal_data(inode));
2128 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2130 /* start a new transaction*/
2131 handle = ext4_journal_start(inode, needed_blocks);
2132 if (IS_ERR(handle)) {
2133 ret = PTR_ERR(handle);
2134 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2135 "%ld pages, ino %lu; err %d", __func__,
2136 wbc->nr_to_write, inode->i_ino, ret);
2137 goto out_writepages;
2141 * Now call write_cache_pages_da() to find the next
2142 * contiguous region of logical blocks that need
2143 * blocks to be allocated by ext4 and submit them.
2145 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
2147 * If we have a contiguous extent of pages and we
2148 * haven't done the I/O yet, map the blocks and submit
2151 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2152 mpage_da_map_and_submit(&mpd);
2153 ret = MPAGE_DA_EXTENT_TAIL;
2155 trace_ext4_da_write_pages(inode, &mpd);
2156 wbc->nr_to_write -= mpd.pages_written;
2158 ext4_journal_stop(handle);
2160 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2161 /* commit the transaction which would
2162 * free blocks released in the transaction
2165 jbd2_journal_force_commit_nested(sbi->s_journal);
2167 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2169 * got one extent now try with
2172 pages_written += mpd.pages_written;
2175 } else if (wbc->nr_to_write)
2177 * There is no more writeout needed
2178 * or we requested for a noblocking writeout
2179 * and we found the device congested
2183 if (!io_done && !cycled) {
2186 wbc->range_start = index << PAGE_CACHE_SHIFT;
2187 wbc->range_end = mapping->writeback_index - 1;
2192 wbc->range_cyclic = range_cyclic;
2193 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2195 * set the writeback_index so that range_cyclic
2196 * mode will write it back later
2198 mapping->writeback_index = done_index;
2201 wbc->nr_to_write -= nr_to_writebump;
2202 wbc->range_start = range_start;
2203 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2207 #define FALL_BACK_TO_NONDELALLOC 1
2208 static int ext4_nonda_switch(struct super_block *sb)
2210 s64 free_blocks, dirty_blocks;
2211 struct ext4_sb_info *sbi = EXT4_SB(sb);
2214 * switch to non delalloc mode if we are running low
2215 * on free block. The free block accounting via percpu
2216 * counters can get slightly wrong with percpu_counter_batch getting
2217 * accumulated on each CPU without updating global counters
2218 * Delalloc need an accurate free block accounting. So switch
2219 * to non delalloc when we are near to error range.
2221 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2222 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2223 if (2 * free_blocks < 3 * dirty_blocks ||
2224 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2226 * free block count is less than 150% of dirty blocks
2227 * or free blocks is less than watermark
2232 * Even if we don't switch but are nearing capacity,
2233 * start pushing delalloc when 1/2 of free blocks are dirty.
2235 if (free_blocks < 2 * dirty_blocks)
2236 writeback_inodes_sb_if_idle(sb);
2241 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2242 loff_t pos, unsigned len, unsigned flags,
2243 struct page **pagep, void **fsdata)
2245 int ret, retries = 0;
2248 struct inode *inode = mapping->host;
2251 index = pos >> PAGE_CACHE_SHIFT;
2253 if (ext4_nonda_switch(inode->i_sb)) {
2254 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2255 return ext4_write_begin(file, mapping, pos,
2256 len, flags, pagep, fsdata);
2258 *fsdata = (void *)0;
2259 trace_ext4_da_write_begin(inode, pos, len, flags);
2262 * With delayed allocation, we don't log the i_disksize update
2263 * if there is delayed block allocation. But we still need
2264 * to journalling the i_disksize update if writes to the end
2265 * of file which has an already mapped buffer.
2267 handle = ext4_journal_start(inode, 1);
2268 if (IS_ERR(handle)) {
2269 ret = PTR_ERR(handle);
2272 /* We cannot recurse into the filesystem as the transaction is already
2274 flags |= AOP_FLAG_NOFS;
2276 page = grab_cache_page_write_begin(mapping, index, flags);
2278 ext4_journal_stop(handle);
2284 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2287 ext4_journal_stop(handle);
2288 page_cache_release(page);
2290 * block_write_begin may have instantiated a few blocks
2291 * outside i_size. Trim these off again. Don't need
2292 * i_size_read because we hold i_mutex.
2294 if (pos + len > inode->i_size)
2295 ext4_truncate_failed_write(inode);
2298 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2305 * Check if we should update i_disksize
2306 * when write to the end of file but not require block allocation
2308 static int ext4_da_should_update_i_disksize(struct page *page,
2309 unsigned long offset)
2311 struct buffer_head *bh;
2312 struct inode *inode = page->mapping->host;
2316 bh = page_buffers(page);
2317 idx = offset >> inode->i_blkbits;
2319 for (i = 0; i < idx; i++)
2320 bh = bh->b_this_page;
2322 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2327 static int ext4_da_write_end(struct file *file,
2328 struct address_space *mapping,
2329 loff_t pos, unsigned len, unsigned copied,
2330 struct page *page, void *fsdata)
2332 struct inode *inode = mapping->host;
2334 handle_t *handle = ext4_journal_current_handle();
2336 unsigned long start, end;
2337 int write_mode = (int)(unsigned long)fsdata;
2339 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2340 if (ext4_should_order_data(inode)) {
2341 return ext4_ordered_write_end(file, mapping, pos,
2342 len, copied, page, fsdata);
2343 } else if (ext4_should_writeback_data(inode)) {
2344 return ext4_writeback_write_end(file, mapping, pos,
2345 len, copied, page, fsdata);
2351 trace_ext4_da_write_end(inode, pos, len, copied);
2352 start = pos & (PAGE_CACHE_SIZE - 1);
2353 end = start + copied - 1;
2356 * generic_write_end() will run mark_inode_dirty() if i_size
2357 * changes. So let's piggyback the i_disksize mark_inode_dirty
2361 new_i_size = pos + copied;
2362 if (new_i_size > EXT4_I(inode)->i_disksize) {
2363 if (ext4_da_should_update_i_disksize(page, end)) {
2364 down_write(&EXT4_I(inode)->i_data_sem);
2365 if (new_i_size > EXT4_I(inode)->i_disksize) {
2367 * Updating i_disksize when extending file
2368 * without needing block allocation
2370 if (ext4_should_order_data(inode))
2371 ret = ext4_jbd2_file_inode(handle,
2374 EXT4_I(inode)->i_disksize = new_i_size;
2376 up_write(&EXT4_I(inode)->i_data_sem);
2377 /* We need to mark inode dirty even if
2378 * new_i_size is less that inode->i_size
2379 * bu greater than i_disksize.(hint delalloc)
2381 ext4_mark_inode_dirty(handle, inode);
2384 ret2 = generic_write_end(file, mapping, pos, len, copied,
2389 ret2 = ext4_journal_stop(handle);
2393 return ret ? ret : copied;
2396 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2399 * Drop reserved blocks
2401 BUG_ON(!PageLocked(page));
2402 if (!page_has_buffers(page))
2405 ext4_da_page_release_reservation(page, offset);
2408 ext4_invalidatepage(page, offset);
2414 * Force all delayed allocation blocks to be allocated for a given inode.
2416 int ext4_alloc_da_blocks(struct inode *inode)
2418 trace_ext4_alloc_da_blocks(inode);
2420 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2421 !EXT4_I(inode)->i_reserved_meta_blocks)
2425 * We do something simple for now. The filemap_flush() will
2426 * also start triggering a write of the data blocks, which is
2427 * not strictly speaking necessary (and for users of
2428 * laptop_mode, not even desirable). However, to do otherwise
2429 * would require replicating code paths in:
2431 * ext4_da_writepages() ->
2432 * write_cache_pages() ---> (via passed in callback function)
2433 * __mpage_da_writepage() -->
2434 * mpage_add_bh_to_extent()
2435 * mpage_da_map_blocks()
2437 * The problem is that write_cache_pages(), located in
2438 * mm/page-writeback.c, marks pages clean in preparation for
2439 * doing I/O, which is not desirable if we're not planning on
2442 * We could call write_cache_pages(), and then redirty all of
2443 * the pages by calling redirty_page_for_writepage() but that
2444 * would be ugly in the extreme. So instead we would need to
2445 * replicate parts of the code in the above functions,
2446 * simplifying them because we wouldn't actually intend to
2447 * write out the pages, but rather only collect contiguous
2448 * logical block extents, call the multi-block allocator, and
2449 * then update the buffer heads with the block allocations.
2451 * For now, though, we'll cheat by calling filemap_flush(),
2452 * which will map the blocks, and start the I/O, but not
2453 * actually wait for the I/O to complete.
2455 return filemap_flush(inode->i_mapping);
2459 * bmap() is special. It gets used by applications such as lilo and by
2460 * the swapper to find the on-disk block of a specific piece of data.
2462 * Naturally, this is dangerous if the block concerned is still in the
2463 * journal. If somebody makes a swapfile on an ext4 data-journaling
2464 * filesystem and enables swap, then they may get a nasty shock when the
2465 * data getting swapped to that swapfile suddenly gets overwritten by
2466 * the original zero's written out previously to the journal and
2467 * awaiting writeback in the kernel's buffer cache.
2469 * So, if we see any bmap calls here on a modified, data-journaled file,
2470 * take extra steps to flush any blocks which might be in the cache.
2472 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2474 struct inode *inode = mapping->host;
2478 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2479 test_opt(inode->i_sb, DELALLOC)) {
2481 * With delalloc we want to sync the file
2482 * so that we can make sure we allocate
2485 filemap_write_and_wait(mapping);
2488 if (EXT4_JOURNAL(inode) &&
2489 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2491 * This is a REALLY heavyweight approach, but the use of
2492 * bmap on dirty files is expected to be extremely rare:
2493 * only if we run lilo or swapon on a freshly made file
2494 * do we expect this to happen.
2496 * (bmap requires CAP_SYS_RAWIO so this does not
2497 * represent an unprivileged user DOS attack --- we'd be
2498 * in trouble if mortal users could trigger this path at
2501 * NB. EXT4_STATE_JDATA is not set on files other than
2502 * regular files. If somebody wants to bmap a directory
2503 * or symlink and gets confused because the buffer
2504 * hasn't yet been flushed to disk, they deserve
2505 * everything they get.
2508 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2509 journal = EXT4_JOURNAL(inode);
2510 jbd2_journal_lock_updates(journal);
2511 err = jbd2_journal_flush(journal);
2512 jbd2_journal_unlock_updates(journal);
2518 return generic_block_bmap(mapping, block, ext4_get_block);
2521 static int ext4_readpage(struct file *file, struct page *page)
2523 trace_ext4_readpage(page);
2524 return mpage_readpage(page, ext4_get_block);
2528 ext4_readpages(struct file *file, struct address_space *mapping,
2529 struct list_head *pages, unsigned nr_pages)
2531 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2534 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
2536 struct buffer_head *head, *bh;
2537 unsigned int curr_off = 0;
2539 if (!page_has_buffers(page))
2541 head = bh = page_buffers(page);
2543 if (offset <= curr_off && test_clear_buffer_uninit(bh)
2545 ext4_free_io_end(bh->b_private);
2546 bh->b_private = NULL;
2547 bh->b_end_io = NULL;
2549 curr_off = curr_off + bh->b_size;
2550 bh = bh->b_this_page;
2551 } while (bh != head);
2554 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2556 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2558 trace_ext4_invalidatepage(page, offset);
2561 * free any io_end structure allocated for buffers to be discarded
2563 if (ext4_should_dioread_nolock(page->mapping->host))
2564 ext4_invalidatepage_free_endio(page, offset);
2566 * If it's a full truncate we just forget about the pending dirtying
2569 ClearPageChecked(page);
2572 jbd2_journal_invalidatepage(journal, page, offset);
2574 block_invalidatepage(page, offset);
2577 static int ext4_releasepage(struct page *page, gfp_t wait)
2579 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2581 trace_ext4_releasepage(page);
2583 WARN_ON(PageChecked(page));
2584 if (!page_has_buffers(page))
2587 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2589 return try_to_free_buffers(page);
2593 * ext4_get_block used when preparing for a DIO write or buffer write.
2594 * We allocate an uinitialized extent if blocks haven't been allocated.
2595 * The extent will be converted to initialized after the IO is complete.
2597 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
2598 struct buffer_head *bh_result, int create)
2600 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2601 inode->i_ino, create);
2602 return _ext4_get_block(inode, iblock, bh_result,
2603 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2606 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2607 ssize_t size, void *private, int ret,
2610 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2611 ext4_io_end_t *io_end = iocb->private;
2612 struct workqueue_struct *wq;
2613 unsigned long flags;
2614 struct ext4_inode_info *ei;
2616 /* if not async direct IO or dio with 0 bytes write, just return */
2617 if (!io_end || !size)
2620 ext_debug("ext4_end_io_dio(): io_end 0x%p"
2621 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
2622 iocb->private, io_end->inode->i_ino, iocb, offset,
2625 /* if not aio dio with unwritten extents, just free io and return */
2626 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2627 ext4_free_io_end(io_end);
2628 iocb->private = NULL;
2631 aio_complete(iocb, ret, 0);
2632 inode_dio_done(inode);
2636 io_end->offset = offset;
2637 io_end->size = size;
2639 io_end->iocb = iocb;
2640 io_end->result = ret;
2642 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
2644 /* Add the io_end to per-inode completed aio dio list*/
2645 ei = EXT4_I(io_end->inode);
2646 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
2647 list_add_tail(&io_end->list, &ei->i_completed_io_list);
2648 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
2650 /* queue the work to convert unwritten extents to written */
2651 queue_work(wq, &io_end->work);
2652 iocb->private = NULL;
2654 /* XXX: probably should move into the real I/O completion handler */
2655 inode_dio_done(inode);
2658 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
2660 ext4_io_end_t *io_end = bh->b_private;
2661 struct workqueue_struct *wq;
2662 struct inode *inode;
2663 unsigned long flags;
2665 if (!test_clear_buffer_uninit(bh) || !io_end)
2668 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2669 printk("sb umounted, discard end_io request for inode %lu\n",
2670 io_end->inode->i_ino);
2671 ext4_free_io_end(io_end);
2675 io_end->flag = EXT4_IO_END_UNWRITTEN;
2676 inode = io_end->inode;
2678 /* Add the io_end to per-inode completed io list*/
2679 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
2680 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
2681 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
2683 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
2684 /* queue the work to convert unwritten extents to written */
2685 queue_work(wq, &io_end->work);
2687 bh->b_private = NULL;
2688 bh->b_end_io = NULL;
2689 clear_buffer_uninit(bh);
2690 end_buffer_async_write(bh, uptodate);
2693 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2695 ext4_io_end_t *io_end;
2696 struct page *page = bh->b_page;
2697 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2698 size_t size = bh->b_size;
2701 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2703 pr_warn_ratelimited("%s: allocation fail\n", __func__);
2707 io_end->offset = offset;
2708 io_end->size = size;
2710 * We need to hold a reference to the page to make sure it
2711 * doesn't get evicted before ext4_end_io_work() has a chance
2712 * to convert the extent from written to unwritten.
2714 io_end->page = page;
2715 get_page(io_end->page);
2717 bh->b_private = io_end;
2718 bh->b_end_io = ext4_end_io_buffer_write;
2723 * For ext4 extent files, ext4 will do direct-io write to holes,
2724 * preallocated extents, and those write extend the file, no need to
2725 * fall back to buffered IO.
2727 * For holes, we fallocate those blocks, mark them as uninitialized
2728 * If those blocks were preallocated, we mark sure they are splited, but
2729 * still keep the range to write as uninitialized.
2731 * The unwrritten extents will be converted to written when DIO is completed.
2732 * For async direct IO, since the IO may still pending when return, we
2733 * set up an end_io call back function, which will do the conversion
2734 * when async direct IO completed.
2736 * If the O_DIRECT write will extend the file then add this inode to the
2737 * orphan list. So recovery will truncate it back to the original size
2738 * if the machine crashes during the write.
2741 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2742 const struct iovec *iov, loff_t offset,
2743 unsigned long nr_segs)
2745 struct file *file = iocb->ki_filp;
2746 struct inode *inode = file->f_mapping->host;
2748 size_t count = iov_length(iov, nr_segs);
2750 loff_t final_size = offset + count;
2751 if (rw == WRITE && final_size <= inode->i_size) {
2753 * We could direct write to holes and fallocate.
2755 * Allocated blocks to fill the hole are marked as uninitialized
2756 * to prevent parallel buffered read to expose the stale data
2757 * before DIO complete the data IO.
2759 * As to previously fallocated extents, ext4 get_block
2760 * will just simply mark the buffer mapped but still
2761 * keep the extents uninitialized.
2763 * for non AIO case, we will convert those unwritten extents
2764 * to written after return back from blockdev_direct_IO.
2766 * for async DIO, the conversion needs to be defered when
2767 * the IO is completed. The ext4 end_io callback function
2768 * will be called to take care of the conversion work.
2769 * Here for async case, we allocate an io_end structure to
2772 iocb->private = NULL;
2773 EXT4_I(inode)->cur_aio_dio = NULL;
2774 if (!is_sync_kiocb(iocb)) {
2775 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
2779 * we save the io structure for current async
2780 * direct IO, so that later ext4_map_blocks()
2781 * could flag the io structure whether there
2782 * is a unwritten extents needs to be converted
2783 * when IO is completed.
2785 EXT4_I(inode)->cur_aio_dio = iocb->private;
2788 ret = __blockdev_direct_IO(rw, iocb, inode,
2789 inode->i_sb->s_bdev, iov,
2791 ext4_get_block_write,
2794 DIO_LOCKING | DIO_SKIP_HOLES);
2796 EXT4_I(inode)->cur_aio_dio = NULL;
2798 * The io_end structure takes a reference to the inode,
2799 * that structure needs to be destroyed and the
2800 * reference to the inode need to be dropped, when IO is
2801 * complete, even with 0 byte write, or failed.
2803 * In the successful AIO DIO case, the io_end structure will be
2804 * desctroyed and the reference to the inode will be dropped
2805 * after the end_io call back function is called.
2807 * In the case there is 0 byte write, or error case, since
2808 * VFS direct IO won't invoke the end_io call back function,
2809 * we need to free the end_io structure here.
2811 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
2812 ext4_free_io_end(iocb->private);
2813 iocb->private = NULL;
2814 } else if (ret > 0 && ext4_test_inode_state(inode,
2815 EXT4_STATE_DIO_UNWRITTEN)) {
2818 * for non AIO case, since the IO is already
2819 * completed, we could do the conversion right here
2821 err = ext4_convert_unwritten_extents(inode,
2825 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
2830 /* for write the the end of file case, we fall back to old way */
2831 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
2834 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
2835 const struct iovec *iov, loff_t offset,
2836 unsigned long nr_segs)
2838 struct file *file = iocb->ki_filp;
2839 struct inode *inode = file->f_mapping->host;
2842 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
2843 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
2844 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
2846 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
2847 trace_ext4_direct_IO_exit(inode, offset,
2848 iov_length(iov, nr_segs), rw, ret);
2853 * Pages can be marked dirty completely asynchronously from ext4's journalling
2854 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
2855 * much here because ->set_page_dirty is called under VFS locks. The page is
2856 * not necessarily locked.
2858 * We cannot just dirty the page and leave attached buffers clean, because the
2859 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
2860 * or jbddirty because all the journalling code will explode.
2862 * So what we do is to mark the page "pending dirty" and next time writepage
2863 * is called, propagate that into the buffers appropriately.
2865 static int ext4_journalled_set_page_dirty(struct page *page)
2867 SetPageChecked(page);
2868 return __set_page_dirty_nobuffers(page);
2871 static const struct address_space_operations ext4_ordered_aops = {
2872 .readpage = ext4_readpage,
2873 .readpages = ext4_readpages,
2874 .writepage = ext4_writepage,
2875 .write_begin = ext4_write_begin,
2876 .write_end = ext4_ordered_write_end,
2878 .invalidatepage = ext4_invalidatepage,
2879 .releasepage = ext4_releasepage,
2880 .direct_IO = ext4_direct_IO,
2881 .migratepage = buffer_migrate_page,
2882 .is_partially_uptodate = block_is_partially_uptodate,
2883 .error_remove_page = generic_error_remove_page,
2886 static const struct address_space_operations ext4_writeback_aops = {
2887 .readpage = ext4_readpage,
2888 .readpages = ext4_readpages,
2889 .writepage = ext4_writepage,
2890 .write_begin = ext4_write_begin,
2891 .write_end = ext4_writeback_write_end,
2893 .invalidatepage = ext4_invalidatepage,
2894 .releasepage = ext4_releasepage,
2895 .direct_IO = ext4_direct_IO,
2896 .migratepage = buffer_migrate_page,
2897 .is_partially_uptodate = block_is_partially_uptodate,
2898 .error_remove_page = generic_error_remove_page,
2901 static const struct address_space_operations ext4_journalled_aops = {
2902 .readpage = ext4_readpage,
2903 .readpages = ext4_readpages,
2904 .writepage = ext4_writepage,
2905 .write_begin = ext4_write_begin,
2906 .write_end = ext4_journalled_write_end,
2907 .set_page_dirty = ext4_journalled_set_page_dirty,
2909 .invalidatepage = ext4_invalidatepage,
2910 .releasepage = ext4_releasepage,
2911 .is_partially_uptodate = block_is_partially_uptodate,
2912 .error_remove_page = generic_error_remove_page,
2915 static const struct address_space_operations ext4_da_aops = {
2916 .readpage = ext4_readpage,
2917 .readpages = ext4_readpages,
2918 .writepage = ext4_writepage,
2919 .writepages = ext4_da_writepages,
2920 .write_begin = ext4_da_write_begin,
2921 .write_end = ext4_da_write_end,
2923 .invalidatepage = ext4_da_invalidatepage,
2924 .releasepage = ext4_releasepage,
2925 .direct_IO = ext4_direct_IO,
2926 .migratepage = buffer_migrate_page,
2927 .is_partially_uptodate = block_is_partially_uptodate,
2928 .error_remove_page = generic_error_remove_page,
2931 void ext4_set_aops(struct inode *inode)
2933 if (ext4_should_order_data(inode) &&
2934 test_opt(inode->i_sb, DELALLOC))
2935 inode->i_mapping->a_ops = &ext4_da_aops;
2936 else if (ext4_should_order_data(inode))
2937 inode->i_mapping->a_ops = &ext4_ordered_aops;
2938 else if (ext4_should_writeback_data(inode) &&
2939 test_opt(inode->i_sb, DELALLOC))
2940 inode->i_mapping->a_ops = &ext4_da_aops;
2941 else if (ext4_should_writeback_data(inode))
2942 inode->i_mapping->a_ops = &ext4_writeback_aops;
2944 inode->i_mapping->a_ops = &ext4_journalled_aops;
2948 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
2949 * up to the end of the block which corresponds to `from'.
2950 * This required during truncate. We need to physically zero the tail end
2951 * of that block so it doesn't yield old data if the file is later grown.
2953 int ext4_block_truncate_page(handle_t *handle,
2954 struct address_space *mapping, loff_t from)
2956 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2959 struct inode *inode = mapping->host;
2961 blocksize = inode->i_sb->s_blocksize;
2962 length = blocksize - (offset & (blocksize - 1));
2964 return ext4_block_zero_page_range(handle, mapping, from, length);
2968 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
2969 * starting from file offset 'from'. The range to be zero'd must
2970 * be contained with in one block. If the specified range exceeds
2971 * the end of the block it will be shortened to end of the block
2972 * that cooresponds to 'from'
2974 int ext4_block_zero_page_range(handle_t *handle,
2975 struct address_space *mapping, loff_t from, loff_t length)
2977 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
2978 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2979 unsigned blocksize, max, pos;
2981 struct inode *inode = mapping->host;
2982 struct buffer_head *bh;
2986 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
2987 mapping_gfp_mask(mapping) & ~__GFP_FS);
2991 blocksize = inode->i_sb->s_blocksize;
2992 max = blocksize - (offset & (blocksize - 1));
2995 * correct length if it does not fall between
2996 * 'from' and the end of the block
2998 if (length > max || length < 0)
3001 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3003 if (!page_has_buffers(page))
3004 create_empty_buffers(page, blocksize, 0);
3006 /* Find the buffer that contains "offset" */
3007 bh = page_buffers(page);
3009 while (offset >= pos) {
3010 bh = bh->b_this_page;
3016 if (buffer_freed(bh)) {
3017 BUFFER_TRACE(bh, "freed: skip");
3021 if (!buffer_mapped(bh)) {
3022 BUFFER_TRACE(bh, "unmapped");
3023 ext4_get_block(inode, iblock, bh, 0);
3024 /* unmapped? It's a hole - nothing to do */
3025 if (!buffer_mapped(bh)) {
3026 BUFFER_TRACE(bh, "still unmapped");
3031 /* Ok, it's mapped. Make sure it's up-to-date */
3032 if (PageUptodate(page))
3033 set_buffer_uptodate(bh);
3035 if (!buffer_uptodate(bh)) {
3037 ll_rw_block(READ, 1, &bh);
3039 /* Uhhuh. Read error. Complain and punt. */
3040 if (!buffer_uptodate(bh))
3044 if (ext4_should_journal_data(inode)) {
3045 BUFFER_TRACE(bh, "get write access");
3046 err = ext4_journal_get_write_access(handle, bh);
3051 zero_user(page, offset, length);
3053 BUFFER_TRACE(bh, "zeroed end of block");
3056 if (ext4_should_journal_data(inode)) {
3057 err = ext4_handle_dirty_metadata(handle, inode, bh);
3059 if (ext4_should_order_data(inode) && EXT4_I(inode)->jinode)
3060 err = ext4_jbd2_file_inode(handle, inode);
3061 mark_buffer_dirty(bh);
3066 page_cache_release(page);
3070 int ext4_can_truncate(struct inode *inode)
3072 if (S_ISREG(inode->i_mode))
3074 if (S_ISDIR(inode->i_mode))
3076 if (S_ISLNK(inode->i_mode))
3077 return !ext4_inode_is_fast_symlink(inode);
3082 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3083 * associated with the given offset and length
3085 * @inode: File inode
3086 * @offset: The offset where the hole will begin
3087 * @len: The length of the hole
3089 * Returns: 0 on sucess or negative on failure
3092 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3094 struct inode *inode = file->f_path.dentry->d_inode;
3095 if (!S_ISREG(inode->i_mode))
3098 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3099 /* TODO: Add support for non extent hole punching */
3103 return ext4_ext_punch_hole(file, offset, length);
3109 * We block out ext4_get_block() block instantiations across the entire
3110 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3111 * simultaneously on behalf of the same inode.
3113 * As we work through the truncate and commmit bits of it to the journal there
3114 * is one core, guiding principle: the file's tree must always be consistent on
3115 * disk. We must be able to restart the truncate after a crash.
3117 * The file's tree may be transiently inconsistent in memory (although it
3118 * probably isn't), but whenever we close off and commit a journal transaction,
3119 * the contents of (the filesystem + the journal) must be consistent and
3120 * restartable. It's pretty simple, really: bottom up, right to left (although
3121 * left-to-right works OK too).
3123 * Note that at recovery time, journal replay occurs *before* the restart of
3124 * truncate against the orphan inode list.
3126 * The committed inode has the new, desired i_size (which is the same as
3127 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3128 * that this inode's truncate did not complete and it will again call
3129 * ext4_truncate() to have another go. So there will be instantiated blocks
3130 * to the right of the truncation point in a crashed ext4 filesystem. But
3131 * that's fine - as long as they are linked from the inode, the post-crash
3132 * ext4_truncate() run will find them and release them.
3134 void ext4_truncate(struct inode *inode)
3136 trace_ext4_truncate_enter(inode);
3138 if (!ext4_can_truncate(inode))
3141 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3143 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3144 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3146 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3147 ext4_ext_truncate(inode);
3149 ext4_ind_truncate(inode);
3151 trace_ext4_truncate_exit(inode);
3155 * ext4_get_inode_loc returns with an extra refcount against the inode's
3156 * underlying buffer_head on success. If 'in_mem' is true, we have all
3157 * data in memory that is needed to recreate the on-disk version of this
3160 static int __ext4_get_inode_loc(struct inode *inode,
3161 struct ext4_iloc *iloc, int in_mem)
3163 struct ext4_group_desc *gdp;
3164 struct buffer_head *bh;
3165 struct super_block *sb = inode->i_sb;
3167 int inodes_per_block, inode_offset;
3170 if (!ext4_valid_inum(sb, inode->i_ino))
3173 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3174 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3179 * Figure out the offset within the block group inode table
3181 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3182 inode_offset = ((inode->i_ino - 1) %
3183 EXT4_INODES_PER_GROUP(sb));
3184 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3185 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3187 bh = sb_getblk(sb, block);
3189 EXT4_ERROR_INODE_BLOCK(inode, block,
3190 "unable to read itable block");
3193 if (!buffer_uptodate(bh)) {
3197 * If the buffer has the write error flag, we have failed
3198 * to write out another inode in the same block. In this
3199 * case, we don't have to read the block because we may
3200 * read the old inode data successfully.
3202 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3203 set_buffer_uptodate(bh);
3205 if (buffer_uptodate(bh)) {
3206 /* someone brought it uptodate while we waited */
3212 * If we have all information of the inode in memory and this
3213 * is the only valid inode in the block, we need not read the
3217 struct buffer_head *bitmap_bh;
3220 start = inode_offset & ~(inodes_per_block - 1);
3222 /* Is the inode bitmap in cache? */
3223 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3228 * If the inode bitmap isn't in cache then the
3229 * optimisation may end up performing two reads instead
3230 * of one, so skip it.
3232 if (!buffer_uptodate(bitmap_bh)) {
3236 for (i = start; i < start + inodes_per_block; i++) {
3237 if (i == inode_offset)
3239 if (ext4_test_bit(i, bitmap_bh->b_data))
3243 if (i == start + inodes_per_block) {
3244 /* all other inodes are free, so skip I/O */
3245 memset(bh->b_data, 0, bh->b_size);
3246 set_buffer_uptodate(bh);
3254 * If we need to do any I/O, try to pre-readahead extra
3255 * blocks from the inode table.
3257 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3258 ext4_fsblk_t b, end, table;
3261 table = ext4_inode_table(sb, gdp);
3262 /* s_inode_readahead_blks is always a power of 2 */
3263 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3266 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3267 num = EXT4_INODES_PER_GROUP(sb);
3268 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3269 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3270 num -= ext4_itable_unused_count(sb, gdp);
3271 table += num / inodes_per_block;
3275 sb_breadahead(sb, b++);
3279 * There are other valid inodes in the buffer, this inode
3280 * has in-inode xattrs, or we don't have this inode in memory.
3281 * Read the block from disk.
3283 trace_ext4_load_inode(inode);
3285 bh->b_end_io = end_buffer_read_sync;
3286 submit_bh(READ_META, bh);
3288 if (!buffer_uptodate(bh)) {
3289 EXT4_ERROR_INODE_BLOCK(inode, block,
3290 "unable to read itable block");
3300 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3302 /* We have all inode data except xattrs in memory here. */
3303 return __ext4_get_inode_loc(inode, iloc,
3304 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3307 void ext4_set_inode_flags(struct inode *inode)
3309 unsigned int flags = EXT4_I(inode)->i_flags;
3311 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3312 if (flags & EXT4_SYNC_FL)
3313 inode->i_flags |= S_SYNC;
3314 if (flags & EXT4_APPEND_FL)
3315 inode->i_flags |= S_APPEND;
3316 if (flags & EXT4_IMMUTABLE_FL)
3317 inode->i_flags |= S_IMMUTABLE;
3318 if (flags & EXT4_NOATIME_FL)
3319 inode->i_flags |= S_NOATIME;
3320 if (flags & EXT4_DIRSYNC_FL)
3321 inode->i_flags |= S_DIRSYNC;
3324 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3325 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3327 unsigned int vfs_fl;
3328 unsigned long old_fl, new_fl;
3331 vfs_fl = ei->vfs_inode.i_flags;
3332 old_fl = ei->i_flags;
3333 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3334 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3336 if (vfs_fl & S_SYNC)
3337 new_fl |= EXT4_SYNC_FL;
3338 if (vfs_fl & S_APPEND)
3339 new_fl |= EXT4_APPEND_FL;
3340 if (vfs_fl & S_IMMUTABLE)
3341 new_fl |= EXT4_IMMUTABLE_FL;
3342 if (vfs_fl & S_NOATIME)
3343 new_fl |= EXT4_NOATIME_FL;
3344 if (vfs_fl & S_DIRSYNC)
3345 new_fl |= EXT4_DIRSYNC_FL;
3346 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3349 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3350 struct ext4_inode_info *ei)
3353 struct inode *inode = &(ei->vfs_inode);
3354 struct super_block *sb = inode->i_sb;
3356 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3357 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3358 /* we are using combined 48 bit field */
3359 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3360 le32_to_cpu(raw_inode->i_blocks_lo);
3361 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3362 /* i_blocks represent file system block size */
3363 return i_blocks << (inode->i_blkbits - 9);
3368 return le32_to_cpu(raw_inode->i_blocks_lo);
3372 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3374 struct ext4_iloc iloc;
3375 struct ext4_inode *raw_inode;
3376 struct ext4_inode_info *ei;
3377 struct inode *inode;
3378 journal_t *journal = EXT4_SB(sb)->s_journal;
3382 inode = iget_locked(sb, ino);
3384 return ERR_PTR(-ENOMEM);
3385 if (!(inode->i_state & I_NEW))
3391 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3394 raw_inode = ext4_raw_inode(&iloc);
3395 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3396 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3397 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3398 if (!(test_opt(inode->i_sb, NO_UID32))) {
3399 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3400 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3402 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
3404 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3405 ei->i_dir_start_lookup = 0;
3406 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3407 /* We now have enough fields to check if the inode was active or not.
3408 * This is needed because nfsd might try to access dead inodes
3409 * the test is that same one that e2fsck uses
3410 * NeilBrown 1999oct15
3412 if (inode->i_nlink == 0) {
3413 if (inode->i_mode == 0 ||
3414 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3415 /* this inode is deleted */
3419 /* The only unlinked inodes we let through here have
3420 * valid i_mode and are being read by the orphan
3421 * recovery code: that's fine, we're about to complete
3422 * the process of deleting those. */
3424 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3425 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3426 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3427 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3429 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3430 inode->i_size = ext4_isize(raw_inode);
3431 ei->i_disksize = inode->i_size;
3433 ei->i_reserved_quota = 0;
3435 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3436 ei->i_block_group = iloc.block_group;
3437 ei->i_last_alloc_group = ~0;
3439 * NOTE! The in-memory inode i_data array is in little-endian order
3440 * even on big-endian machines: we do NOT byteswap the block numbers!
3442 for (block = 0; block < EXT4_N_BLOCKS; block++)
3443 ei->i_data[block] = raw_inode->i_block[block];
3444 INIT_LIST_HEAD(&ei->i_orphan);
3447 * Set transaction id's of transactions that have to be committed
3448 * to finish f[data]sync. We set them to currently running transaction
3449 * as we cannot be sure that the inode or some of its metadata isn't
3450 * part of the transaction - the inode could have been reclaimed and
3451 * now it is reread from disk.
3454 transaction_t *transaction;
3457 read_lock(&journal->j_state_lock);
3458 if (journal->j_running_transaction)
3459 transaction = journal->j_running_transaction;
3461 transaction = journal->j_committing_transaction;
3463 tid = transaction->t_tid;
3465 tid = journal->j_commit_sequence;
3466 read_unlock(&journal->j_state_lock);
3467 ei->i_sync_tid = tid;
3468 ei->i_datasync_tid = tid;
3471 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3472 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3473 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3474 EXT4_INODE_SIZE(inode->i_sb)) {
3478 if (ei->i_extra_isize == 0) {
3479 /* The extra space is currently unused. Use it. */
3480 ei->i_extra_isize = sizeof(struct ext4_inode) -
3481 EXT4_GOOD_OLD_INODE_SIZE;
3483 __le32 *magic = (void *)raw_inode +
3484 EXT4_GOOD_OLD_INODE_SIZE +
3486 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3487 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3490 ei->i_extra_isize = 0;
3492 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3493 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3494 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3495 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3497 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3498 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3499 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3501 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3505 if (ei->i_file_acl &&
3506 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3507 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
3511 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3512 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3513 (S_ISLNK(inode->i_mode) &&
3514 !ext4_inode_is_fast_symlink(inode)))
3515 /* Validate extent which is part of inode */
3516 ret = ext4_ext_check_inode(inode);
3517 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3518 (S_ISLNK(inode->i_mode) &&
3519 !ext4_inode_is_fast_symlink(inode))) {
3520 /* Validate block references which are part of inode */
3521 ret = ext4_ind_check_inode(inode);
3526 if (S_ISREG(inode->i_mode)) {
3527 inode->i_op = &ext4_file_inode_operations;
3528 inode->i_fop = &ext4_file_operations;
3529 ext4_set_aops(inode);
3530 } else if (S_ISDIR(inode->i_mode)) {
3531 inode->i_op = &ext4_dir_inode_operations;
3532 inode->i_fop = &ext4_dir_operations;
3533 } else if (S_ISLNK(inode->i_mode)) {
3534 if (ext4_inode_is_fast_symlink(inode)) {
3535 inode->i_op = &ext4_fast_symlink_inode_operations;
3536 nd_terminate_link(ei->i_data, inode->i_size,
3537 sizeof(ei->i_data) - 1);
3539 inode->i_op = &ext4_symlink_inode_operations;
3540 ext4_set_aops(inode);
3542 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3543 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3544 inode->i_op = &ext4_special_inode_operations;
3545 if (raw_inode->i_block[0])
3546 init_special_inode(inode, inode->i_mode,
3547 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3549 init_special_inode(inode, inode->i_mode,
3550 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3553 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
3557 ext4_set_inode_flags(inode);
3558 unlock_new_inode(inode);
3564 return ERR_PTR(ret);
3567 static int ext4_inode_blocks_set(handle_t *handle,
3568 struct ext4_inode *raw_inode,
3569 struct ext4_inode_info *ei)
3571 struct inode *inode = &(ei->vfs_inode);
3572 u64 i_blocks = inode->i_blocks;
3573 struct super_block *sb = inode->i_sb;
3575 if (i_blocks <= ~0U) {
3577 * i_blocks can be represnted in a 32 bit variable
3578 * as multiple of 512 bytes
3580 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3581 raw_inode->i_blocks_high = 0;
3582 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3585 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
3588 if (i_blocks <= 0xffffffffffffULL) {
3590 * i_blocks can be represented in a 48 bit variable
3591 * as multiple of 512 bytes
3593 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3594 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3595 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3597 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3598 /* i_block is stored in file system block size */
3599 i_blocks = i_blocks >> (inode->i_blkbits - 9);
3600 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3601 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3607 * Post the struct inode info into an on-disk inode location in the
3608 * buffer-cache. This gobbles the caller's reference to the
3609 * buffer_head in the inode location struct.
3611 * The caller must have write access to iloc->bh.
3613 static int ext4_do_update_inode(handle_t *handle,
3614 struct inode *inode,
3615 struct ext4_iloc *iloc)
3617 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
3618 struct ext4_inode_info *ei = EXT4_I(inode);
3619 struct buffer_head *bh = iloc->bh;
3620 int err = 0, rc, block;
3622 /* For fields not not tracking in the in-memory inode,
3623 * initialise them to zero for new inodes. */
3624 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
3625 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
3627 ext4_get_inode_flags(ei);
3628 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
3629 if (!(test_opt(inode->i_sb, NO_UID32))) {
3630 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
3631 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
3633 * Fix up interoperability with old kernels. Otherwise, old inodes get
3634 * re-used with the upper 16 bits of the uid/gid intact
3637 raw_inode->i_uid_high =
3638 cpu_to_le16(high_16_bits(inode->i_uid));
3639 raw_inode->i_gid_high =
3640 cpu_to_le16(high_16_bits(inode->i_gid));
3642 raw_inode->i_uid_high = 0;
3643 raw_inode->i_gid_high = 0;
3646 raw_inode->i_uid_low =
3647 cpu_to_le16(fs_high2lowuid(inode->i_uid));
3648 raw_inode->i_gid_low =
3649 cpu_to_le16(fs_high2lowgid(inode->i_gid));
3650 raw_inode->i_uid_high = 0;
3651 raw_inode->i_gid_high = 0;
3653 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
3655 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
3656 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
3657 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
3658 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
3660 if (ext4_inode_blocks_set(handle, raw_inode, ei))
3662 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3663 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
3664 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
3665 cpu_to_le32(EXT4_OS_HURD))
3666 raw_inode->i_file_acl_high =
3667 cpu_to_le16(ei->i_file_acl >> 32);
3668 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
3669 ext4_isize_set(raw_inode, ei->i_disksize);
3670 if (ei->i_disksize > 0x7fffffffULL) {
3671 struct super_block *sb = inode->i_sb;
3672 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
3673 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
3674 EXT4_SB(sb)->s_es->s_rev_level ==
3675 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
3676 /* If this is the first large file
3677 * created, add a flag to the superblock.
3679 err = ext4_journal_get_write_access(handle,
3680 EXT4_SB(sb)->s_sbh);
3683 ext4_update_dynamic_rev(sb);
3684 EXT4_SET_RO_COMPAT_FEATURE(sb,
3685 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
3687 ext4_handle_sync(handle);
3688 err = ext4_handle_dirty_metadata(handle, NULL,
3689 EXT4_SB(sb)->s_sbh);
3692 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3693 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3694 if (old_valid_dev(inode->i_rdev)) {
3695 raw_inode->i_block[0] =
3696 cpu_to_le32(old_encode_dev(inode->i_rdev));
3697 raw_inode->i_block[1] = 0;
3699 raw_inode->i_block[0] = 0;
3700 raw_inode->i_block[1] =
3701 cpu_to_le32(new_encode_dev(inode->i_rdev));
3702 raw_inode->i_block[2] = 0;
3705 for (block = 0; block < EXT4_N_BLOCKS; block++)
3706 raw_inode->i_block[block] = ei->i_data[block];
3708 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
3709 if (ei->i_extra_isize) {
3710 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3711 raw_inode->i_version_hi =
3712 cpu_to_le32(inode->i_version >> 32);
3713 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
3716 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
3717 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
3720 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
3722 ext4_update_inode_fsync_trans(handle, inode, 0);
3725 ext4_std_error(inode->i_sb, err);
3730 * ext4_write_inode()
3732 * We are called from a few places:
3734 * - Within generic_file_write() for O_SYNC files.
3735 * Here, there will be no transaction running. We wait for any running
3736 * trasnaction to commit.
3738 * - Within sys_sync(), kupdate and such.
3739 * We wait on commit, if tol to.
3741 * - Within prune_icache() (PF_MEMALLOC == true)
3742 * Here we simply return. We can't afford to block kswapd on the
3745 * In all cases it is actually safe for us to return without doing anything,
3746 * because the inode has been copied into a raw inode buffer in
3747 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3750 * Note that we are absolutely dependent upon all inode dirtiers doing the
3751 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3752 * which we are interested.
3754 * It would be a bug for them to not do this. The code:
3756 * mark_inode_dirty(inode)
3758 * inode->i_size = expr;
3760 * is in error because a kswapd-driven write_inode() could occur while
3761 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3762 * will no longer be on the superblock's dirty inode list.
3764 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
3768 if (current->flags & PF_MEMALLOC)
3771 if (EXT4_SB(inode->i_sb)->s_journal) {
3772 if (ext4_journal_current_handle()) {
3773 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3778 if (wbc->sync_mode != WB_SYNC_ALL)
3781 err = ext4_force_commit(inode->i_sb);
3783 struct ext4_iloc iloc;
3785 err = __ext4_get_inode_loc(inode, &iloc, 0);
3788 if (wbc->sync_mode == WB_SYNC_ALL)
3789 sync_dirty_buffer(iloc.bh);
3790 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
3791 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
3792 "IO error syncing inode");
3803 * Called from notify_change.
3805 * We want to trap VFS attempts to truncate the file as soon as
3806 * possible. In particular, we want to make sure that when the VFS
3807 * shrinks i_size, we put the inode on the orphan list and modify
3808 * i_disksize immediately, so that during the subsequent flushing of
3809 * dirty pages and freeing of disk blocks, we can guarantee that any
3810 * commit will leave the blocks being flushed in an unused state on
3811 * disk. (On recovery, the inode will get truncated and the blocks will
3812 * be freed, so we have a strong guarantee that no future commit will
3813 * leave these blocks visible to the user.)
3815 * Another thing we have to assure is that if we are in ordered mode
3816 * and inode is still attached to the committing transaction, we must
3817 * we start writeout of all the dirty pages which are being truncated.
3818 * This way we are sure that all the data written in the previous
3819 * transaction are already on disk (truncate waits for pages under
3822 * Called with inode->i_mutex down.
3824 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
3826 struct inode *inode = dentry->d_inode;
3829 const unsigned int ia_valid = attr->ia_valid;
3831 error = inode_change_ok(inode, attr);
3835 if (is_quota_modification(inode, attr))
3836 dquot_initialize(inode);
3837 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3838 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3841 /* (user+group)*(old+new) structure, inode write (sb,
3842 * inode block, ? - but truncate inode update has it) */
3843 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
3844 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
3845 if (IS_ERR(handle)) {
3846 error = PTR_ERR(handle);
3849 error = dquot_transfer(inode, attr);
3851 ext4_journal_stop(handle);
3854 /* Update corresponding info in inode so that everything is in
3855 * one transaction */
3856 if (attr->ia_valid & ATTR_UID)
3857 inode->i_uid = attr->ia_uid;
3858 if (attr->ia_valid & ATTR_GID)
3859 inode->i_gid = attr->ia_gid;
3860 error = ext4_mark_inode_dirty(handle, inode);
3861 ext4_journal_stop(handle);
3864 if (attr->ia_valid & ATTR_SIZE) {
3865 inode_dio_wait(inode);
3867 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3868 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3870 if (attr->ia_size > sbi->s_bitmap_maxbytes)
3875 if (S_ISREG(inode->i_mode) &&
3876 attr->ia_valid & ATTR_SIZE &&
3877 (attr->ia_size < inode->i_size)) {
3880 handle = ext4_journal_start(inode, 3);
3881 if (IS_ERR(handle)) {
3882 error = PTR_ERR(handle);
3885 if (ext4_handle_valid(handle)) {
3886 error = ext4_orphan_add(handle, inode);
3889 EXT4_I(inode)->i_disksize = attr->ia_size;
3890 rc = ext4_mark_inode_dirty(handle, inode);
3893 ext4_journal_stop(handle);
3895 if (ext4_should_order_data(inode)) {
3896 error = ext4_begin_ordered_truncate(inode,
3899 /* Do as much error cleanup as possible */
3900 handle = ext4_journal_start(inode, 3);
3901 if (IS_ERR(handle)) {
3902 ext4_orphan_del(NULL, inode);
3905 ext4_orphan_del(handle, inode);
3907 ext4_journal_stop(handle);
3913 if (attr->ia_valid & ATTR_SIZE) {
3914 if (attr->ia_size != i_size_read(inode)) {
3915 truncate_setsize(inode, attr->ia_size);
3916 ext4_truncate(inode);
3917 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3918 ext4_truncate(inode);
3922 setattr_copy(inode, attr);
3923 mark_inode_dirty(inode);
3927 * If the call to ext4_truncate failed to get a transaction handle at
3928 * all, we need to clean up the in-core orphan list manually.
3930 if (orphan && inode->i_nlink)
3931 ext4_orphan_del(NULL, inode);
3933 if (!rc && (ia_valid & ATTR_MODE))
3934 rc = ext4_acl_chmod(inode);
3937 ext4_std_error(inode->i_sb, error);
3943 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
3946 struct inode *inode;
3947 unsigned long delalloc_blocks;
3949 inode = dentry->d_inode;
3950 generic_fillattr(inode, stat);
3953 * We can't update i_blocks if the block allocation is delayed
3954 * otherwise in the case of system crash before the real block
3955 * allocation is done, we will have i_blocks inconsistent with
3956 * on-disk file blocks.
3957 * We always keep i_blocks updated together with real
3958 * allocation. But to not confuse with user, stat
3959 * will return the blocks that include the delayed allocation
3960 * blocks for this file.
3962 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
3964 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
3968 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
3970 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3971 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
3972 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
3976 * Account for index blocks, block groups bitmaps and block group
3977 * descriptor blocks if modify datablocks and index blocks
3978 * worse case, the indexs blocks spread over different block groups
3980 * If datablocks are discontiguous, they are possible to spread over
3981 * different block groups too. If they are contiuguous, with flexbg,
3982 * they could still across block group boundary.
3984 * Also account for superblock, inode, quota and xattr blocks
3986 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
3988 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
3994 * How many index blocks need to touch to modify nrblocks?
3995 * The "Chunk" flag indicating whether the nrblocks is
3996 * physically contiguous on disk
3998 * For Direct IO and fallocate, they calls get_block to allocate
3999 * one single extent at a time, so they could set the "Chunk" flag
4001 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4006 * Now let's see how many group bitmaps and group descriptors need
4016 if (groups > ngroups)
4018 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4019 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4021 /* bitmaps and block group descriptor blocks */
4022 ret += groups + gdpblocks;
4024 /* Blocks for super block, inode, quota and xattr blocks */
4025 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4031 * Calculate the total number of credits to reserve to fit
4032 * the modification of a single pages into a single transaction,
4033 * which may include multiple chunks of block allocations.
4035 * This could be called via ext4_write_begin()
4037 * We need to consider the worse case, when
4038 * one new block per extent.
4040 int ext4_writepage_trans_blocks(struct inode *inode)
4042 int bpp = ext4_journal_blocks_per_page(inode);
4045 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4047 /* Account for data blocks for journalled mode */
4048 if (ext4_should_journal_data(inode))
4054 * Calculate the journal credits for a chunk of data modification.
4056 * This is called from DIO, fallocate or whoever calling
4057 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4059 * journal buffers for data blocks are not included here, as DIO
4060 * and fallocate do no need to journal data buffers.
4062 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4064 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4068 * The caller must have previously called ext4_reserve_inode_write().
4069 * Give this, we know that the caller already has write access to iloc->bh.
4071 int ext4_mark_iloc_dirty(handle_t *handle,
4072 struct inode *inode, struct ext4_iloc *iloc)
4076 if (test_opt(inode->i_sb, I_VERSION))
4077 inode_inc_iversion(inode);
4079 /* the do_update_inode consumes one bh->b_count */
4082 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4083 err = ext4_do_update_inode(handle, inode, iloc);
4089 * On success, We end up with an outstanding reference count against
4090 * iloc->bh. This _must_ be cleaned up later.
4094 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4095 struct ext4_iloc *iloc)
4099 err = ext4_get_inode_loc(inode, iloc);
4101 BUFFER_TRACE(iloc->bh, "get_write_access");
4102 err = ext4_journal_get_write_access(handle, iloc->bh);
4108 ext4_std_error(inode->i_sb, err);
4113 * Expand an inode by new_extra_isize bytes.
4114 * Returns 0 on success or negative error number on failure.
4116 static int ext4_expand_extra_isize(struct inode *inode,
4117 unsigned int new_extra_isize,
4118 struct ext4_iloc iloc,
4121 struct ext4_inode *raw_inode;
4122 struct ext4_xattr_ibody_header *header;
4124 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4127 raw_inode = ext4_raw_inode(&iloc);
4129 header = IHDR(inode, raw_inode);
4131 /* No extended attributes present */
4132 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4133 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4134 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4136 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4140 /* try to expand with EAs present */
4141 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4146 * What we do here is to mark the in-core inode as clean with respect to inode
4147 * dirtiness (it may still be data-dirty).
4148 * This means that the in-core inode may be reaped by prune_icache
4149 * without having to perform any I/O. This is a very good thing,
4150 * because *any* task may call prune_icache - even ones which
4151 * have a transaction open against a different journal.
4153 * Is this cheating? Not really. Sure, we haven't written the
4154 * inode out, but prune_icache isn't a user-visible syncing function.
4155 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4156 * we start and wait on commits.
4158 * Is this efficient/effective? Well, we're being nice to the system
4159 * by cleaning up our inodes proactively so they can be reaped
4160 * without I/O. But we are potentially leaving up to five seconds'
4161 * worth of inodes floating about which prune_icache wants us to
4162 * write out. One way to fix that would be to get prune_icache()
4163 * to do a write_super() to free up some memory. It has the desired
4166 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4168 struct ext4_iloc iloc;
4169 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4170 static unsigned int mnt_count;
4174 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4175 err = ext4_reserve_inode_write(handle, inode, &iloc);
4176 if (ext4_handle_valid(handle) &&
4177 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4178 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4180 * We need extra buffer credits since we may write into EA block
4181 * with this same handle. If journal_extend fails, then it will
4182 * only result in a minor loss of functionality for that inode.
4183 * If this is felt to be critical, then e2fsck should be run to
4184 * force a large enough s_min_extra_isize.
4186 if ((jbd2_journal_extend(handle,
4187 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4188 ret = ext4_expand_extra_isize(inode,
4189 sbi->s_want_extra_isize,
4192 ext4_set_inode_state(inode,
4193 EXT4_STATE_NO_EXPAND);
4195 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4196 ext4_warning(inode->i_sb,
4197 "Unable to expand inode %lu. Delete"
4198 " some EAs or run e2fsck.",
4201 le16_to_cpu(sbi->s_es->s_mnt_count);
4207 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4212 * ext4_dirty_inode() is called from __mark_inode_dirty()
4214 * We're really interested in the case where a file is being extended.
4215 * i_size has been changed by generic_commit_write() and we thus need
4216 * to include the updated inode in the current transaction.
4218 * Also, dquot_alloc_block() will always dirty the inode when blocks
4219 * are allocated to the file.
4221 * If the inode is marked synchronous, we don't honour that here - doing
4222 * so would cause a commit on atime updates, which we don't bother doing.
4223 * We handle synchronous inodes at the highest possible level.
4225 void ext4_dirty_inode(struct inode *inode, int flags)
4229 handle = ext4_journal_start(inode, 2);
4233 ext4_mark_inode_dirty(handle, inode);
4235 ext4_journal_stop(handle);
4242 * Bind an inode's backing buffer_head into this transaction, to prevent
4243 * it from being flushed to disk early. Unlike
4244 * ext4_reserve_inode_write, this leaves behind no bh reference and
4245 * returns no iloc structure, so the caller needs to repeat the iloc
4246 * lookup to mark the inode dirty later.
4248 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4250 struct ext4_iloc iloc;
4254 err = ext4_get_inode_loc(inode, &iloc);
4256 BUFFER_TRACE(iloc.bh, "get_write_access");
4257 err = jbd2_journal_get_write_access(handle, iloc.bh);
4259 err = ext4_handle_dirty_metadata(handle,
4265 ext4_std_error(inode->i_sb, err);
4270 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4277 * We have to be very careful here: changing a data block's
4278 * journaling status dynamically is dangerous. If we write a
4279 * data block to the journal, change the status and then delete
4280 * that block, we risk forgetting to revoke the old log record
4281 * from the journal and so a subsequent replay can corrupt data.
4282 * So, first we make sure that the journal is empty and that
4283 * nobody is changing anything.
4286 journal = EXT4_JOURNAL(inode);
4289 if (is_journal_aborted(journal))
4292 jbd2_journal_lock_updates(journal);
4293 jbd2_journal_flush(journal);
4296 * OK, there are no updates running now, and all cached data is
4297 * synced to disk. We are now in a completely consistent state
4298 * which doesn't have anything in the journal, and we know that
4299 * no filesystem updates are running, so it is safe to modify
4300 * the inode's in-core data-journaling state flag now.
4304 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4306 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4307 ext4_set_aops(inode);
4309 jbd2_journal_unlock_updates(journal);
4311 /* Finally we can mark the inode as dirty. */
4313 handle = ext4_journal_start(inode, 1);
4315 return PTR_ERR(handle);
4317 err = ext4_mark_inode_dirty(handle, inode);
4318 ext4_handle_sync(handle);
4319 ext4_journal_stop(handle);
4320 ext4_std_error(inode->i_sb, err);
4325 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4327 return !buffer_mapped(bh);
4330 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4332 struct page *page = vmf->page;
4336 struct file *file = vma->vm_file;
4337 struct inode *inode = file->f_path.dentry->d_inode;
4338 struct address_space *mapping = inode->i_mapping;
4340 get_block_t *get_block;
4344 * This check is racy but catches the common case. We rely on
4345 * __block_page_mkwrite() to do a reliable check.
4347 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
4348 /* Delalloc case is easy... */
4349 if (test_opt(inode->i_sb, DELALLOC) &&
4350 !ext4_should_journal_data(inode) &&
4351 !ext4_nonda_switch(inode->i_sb)) {
4353 ret = __block_page_mkwrite(vma, vmf,
4354 ext4_da_get_block_prep);
4355 } while (ret == -ENOSPC &&
4356 ext4_should_retry_alloc(inode->i_sb, &retries));
4361 size = i_size_read(inode);
4362 /* Page got truncated from under us? */
4363 if (page->mapping != mapping || page_offset(page) > size) {
4365 ret = VM_FAULT_NOPAGE;
4369 if (page->index == size >> PAGE_CACHE_SHIFT)
4370 len = size & ~PAGE_CACHE_MASK;
4372 len = PAGE_CACHE_SIZE;
4374 * Return if we have all the buffers mapped. This avoids the need to do
4375 * journal_start/journal_stop which can block and take a long time
4377 if (page_has_buffers(page)) {
4378 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4379 ext4_bh_unmapped)) {
4380 /* Wait so that we don't change page under IO */
4381 wait_on_page_writeback(page);
4382 ret = VM_FAULT_LOCKED;
4387 /* OK, we need to fill the hole... */
4388 if (ext4_should_dioread_nolock(inode))
4389 get_block = ext4_get_block_write;
4391 get_block = ext4_get_block;
4393 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4394 if (IS_ERR(handle)) {
4395 ret = VM_FAULT_SIGBUS;
4398 ret = __block_page_mkwrite(vma, vmf, get_block);
4399 if (!ret && ext4_should_journal_data(inode)) {
4400 if (walk_page_buffers(handle, page_buffers(page), 0,
4401 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4403 ret = VM_FAULT_SIGBUS;
4406 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4408 ext4_journal_stop(handle);
4409 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4412 ret = block_page_mkwrite_return(ret);