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);
124 ext4_ioend_wait(inode);
126 if (inode->i_nlink) {
128 * When journalling data dirty buffers are tracked only in the
129 * journal. So although mm thinks everything is clean and
130 * ready for reaping the inode might still have some pages to
131 * write in the running transaction or waiting to be
132 * checkpointed. Thus calling jbd2_journal_invalidatepage()
133 * (via truncate_inode_pages()) to discard these buffers can
134 * cause data loss. Also even if we did not discard these
135 * buffers, we would have no way to find them after the inode
136 * is reaped and thus user could see stale data if he tries to
137 * read them before the transaction is checkpointed. So be
138 * careful and force everything to disk here... We use
139 * ei->i_datasync_tid to store the newest transaction
140 * containing inode's data.
142 * Note that directories do not have this problem because they
143 * don't use page cache.
145 if (ext4_should_journal_data(inode) &&
146 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
147 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
148 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
150 jbd2_log_start_commit(journal, commit_tid);
151 jbd2_log_wait_commit(journal, commit_tid);
152 filemap_write_and_wait(&inode->i_data);
154 truncate_inode_pages(&inode->i_data, 0);
158 if (!is_bad_inode(inode))
159 dquot_initialize(inode);
161 if (ext4_should_order_data(inode))
162 ext4_begin_ordered_truncate(inode, 0);
163 truncate_inode_pages(&inode->i_data, 0);
165 if (is_bad_inode(inode))
168 handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
169 if (IS_ERR(handle)) {
170 ext4_std_error(inode->i_sb, PTR_ERR(handle));
172 * If we're going to skip the normal cleanup, we still need to
173 * make sure that the in-core orphan linked list is properly
176 ext4_orphan_del(NULL, inode);
181 ext4_handle_sync(handle);
183 err = ext4_mark_inode_dirty(handle, inode);
185 ext4_warning(inode->i_sb,
186 "couldn't mark inode dirty (err %d)", err);
190 ext4_truncate(inode);
193 * ext4_ext_truncate() doesn't reserve any slop when it
194 * restarts journal transactions; therefore there may not be
195 * enough credits left in the handle to remove the inode from
196 * the orphan list and set the dtime field.
198 if (!ext4_handle_has_enough_credits(handle, 3)) {
199 err = ext4_journal_extend(handle, 3);
201 err = ext4_journal_restart(handle, 3);
203 ext4_warning(inode->i_sb,
204 "couldn't extend journal (err %d)", err);
206 ext4_journal_stop(handle);
207 ext4_orphan_del(NULL, inode);
213 * Kill off the orphan record which ext4_truncate created.
214 * AKPM: I think this can be inside the above `if'.
215 * Note that ext4_orphan_del() has to be able to cope with the
216 * deletion of a non-existent orphan - this is because we don't
217 * know if ext4_truncate() actually created an orphan record.
218 * (Well, we could do this if we need to, but heck - it works)
220 ext4_orphan_del(handle, inode);
221 EXT4_I(inode)->i_dtime = get_seconds();
224 * One subtle ordering requirement: if anything has gone wrong
225 * (transaction abort, IO errors, whatever), then we can still
226 * do these next steps (the fs will already have been marked as
227 * having errors), but we can't free the inode if the mark_dirty
230 if (ext4_mark_inode_dirty(handle, inode))
231 /* If that failed, just do the required in-core inode clear. */
232 ext4_clear_inode(inode);
234 ext4_free_inode(handle, inode);
235 ext4_journal_stop(handle);
238 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
242 qsize_t *ext4_get_reserved_space(struct inode *inode)
244 return &EXT4_I(inode)->i_reserved_quota;
249 * Calculate the number of metadata blocks need to reserve
250 * to allocate a block located at @lblock
252 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
254 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
255 return ext4_ext_calc_metadata_amount(inode, lblock);
257 return ext4_ind_calc_metadata_amount(inode, lblock);
261 * Called with i_data_sem down, which is important since we can call
262 * ext4_discard_preallocations() from here.
264 void ext4_da_update_reserve_space(struct inode *inode,
265 int used, int quota_claim)
267 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
268 struct ext4_inode_info *ei = EXT4_I(inode);
270 spin_lock(&ei->i_block_reservation_lock);
271 trace_ext4_da_update_reserve_space(inode, used);
272 if (unlikely(used > ei->i_reserved_data_blocks)) {
273 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
274 "with only %d reserved data blocks\n",
275 __func__, inode->i_ino, used,
276 ei->i_reserved_data_blocks);
278 used = ei->i_reserved_data_blocks;
281 /* Update per-inode reservations */
282 ei->i_reserved_data_blocks -= used;
283 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
284 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
285 used + ei->i_allocated_meta_blocks);
286 ei->i_allocated_meta_blocks = 0;
288 if (ei->i_reserved_data_blocks == 0) {
290 * We can release all of the reserved metadata blocks
291 * only when we have written all of the delayed
294 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
295 ei->i_reserved_meta_blocks);
296 ei->i_reserved_meta_blocks = 0;
297 ei->i_da_metadata_calc_len = 0;
299 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
301 /* Update quota subsystem for data blocks */
303 dquot_claim_block(inode, used);
306 * We did fallocate with an offset that is already delayed
307 * allocated. So on delayed allocated writeback we should
308 * not re-claim the quota for fallocated blocks.
310 dquot_release_reservation_block(inode, used);
314 * If we have done all the pending block allocations and if
315 * there aren't any writers on the inode, we can discard the
316 * inode's preallocations.
318 if ((ei->i_reserved_data_blocks == 0) &&
319 (atomic_read(&inode->i_writecount) == 0))
320 ext4_discard_preallocations(inode);
323 static int __check_block_validity(struct inode *inode, const char *func,
325 struct ext4_map_blocks *map)
327 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
329 ext4_error_inode(inode, func, line, map->m_pblk,
330 "lblock %lu mapped to illegal pblock "
331 "(length %d)", (unsigned long) map->m_lblk,
338 #define check_block_validity(inode, map) \
339 __check_block_validity((inode), __func__, __LINE__, (map))
342 * Return the number of contiguous dirty pages in a given inode
343 * starting at page frame idx.
345 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
346 unsigned int max_pages)
348 struct address_space *mapping = inode->i_mapping;
352 int i, nr_pages, done = 0;
356 pagevec_init(&pvec, 0);
359 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
361 (pgoff_t)PAGEVEC_SIZE);
364 for (i = 0; i < nr_pages; i++) {
365 struct page *page = pvec.pages[i];
366 struct buffer_head *bh, *head;
369 if (unlikely(page->mapping != mapping) ||
371 PageWriteback(page) ||
372 page->index != idx) {
377 if (page_has_buffers(page)) {
378 bh = head = page_buffers(page);
380 if (!buffer_delay(bh) &&
381 !buffer_unwritten(bh))
383 bh = bh->b_this_page;
384 } while (!done && (bh != head));
391 if (num >= max_pages) {
396 pagevec_release(&pvec);
402 * The ext4_map_blocks() function tries to look up the requested blocks,
403 * and returns if the blocks are already mapped.
405 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
406 * and store the allocated blocks in the result buffer head and mark it
409 * If file type is extents based, it will call ext4_ext_map_blocks(),
410 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
413 * On success, it returns the number of blocks being mapped or allocate.
414 * if create==0 and the blocks are pre-allocated and uninitialized block,
415 * the result buffer head is unmapped. If the create ==1, it will make sure
416 * the buffer head is mapped.
418 * It returns 0 if plain look up failed (blocks have not been allocated), in
419 * that casem, buffer head is unmapped
421 * It returns the error in case of allocation failure.
423 int ext4_map_blocks(handle_t *handle, struct inode *inode,
424 struct ext4_map_blocks *map, int flags)
429 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
430 "logical block %lu\n", inode->i_ino, flags, map->m_len,
431 (unsigned long) map->m_lblk);
433 * Try to see if we can get the block without requesting a new
436 down_read((&EXT4_I(inode)->i_data_sem));
437 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
438 retval = ext4_ext_map_blocks(handle, inode, map, 0);
440 retval = ext4_ind_map_blocks(handle, inode, map, 0);
442 up_read((&EXT4_I(inode)->i_data_sem));
444 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
445 int ret = check_block_validity(inode, map);
450 /* If it is only a block(s) look up */
451 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
455 * Returns if the blocks have already allocated
457 * Note that if blocks have been preallocated
458 * ext4_ext_get_block() returns th create = 0
459 * with buffer head unmapped.
461 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
465 * When we call get_blocks without the create flag, the
466 * BH_Unwritten flag could have gotten set if the blocks
467 * requested were part of a uninitialized extent. We need to
468 * clear this flag now that we are committed to convert all or
469 * part of the uninitialized extent to be an initialized
470 * extent. This is because we need to avoid the combination
471 * of BH_Unwritten and BH_Mapped flags being simultaneously
472 * set on the buffer_head.
474 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
477 * New blocks allocate and/or writing to uninitialized extent
478 * will possibly result in updating i_data, so we take
479 * the write lock of i_data_sem, and call get_blocks()
480 * with create == 1 flag.
482 down_write((&EXT4_I(inode)->i_data_sem));
485 * if the caller is from delayed allocation writeout path
486 * we have already reserved fs blocks for allocation
487 * let the underlying get_block() function know to
488 * avoid double accounting
490 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
491 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
493 * We need to check for EXT4 here because migrate
494 * could have changed the inode type in between
496 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
497 retval = ext4_ext_map_blocks(handle, inode, map, flags);
499 retval = ext4_ind_map_blocks(handle, inode, map, flags);
501 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
503 * We allocated new blocks which will result in
504 * i_data's format changing. Force the migrate
505 * to fail by clearing migrate flags
507 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
511 * Update reserved blocks/metadata blocks after successful
512 * block allocation which had been deferred till now. We don't
513 * support fallocate for non extent files. So we can update
514 * reserve space here.
517 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
518 ext4_da_update_reserve_space(inode, retval, 1);
520 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
521 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
523 up_write((&EXT4_I(inode)->i_data_sem));
524 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
525 int ret = check_block_validity(inode, map);
532 /* Maximum number of blocks we map for direct IO at once. */
533 #define DIO_MAX_BLOCKS 4096
535 static int _ext4_get_block(struct inode *inode, sector_t iblock,
536 struct buffer_head *bh, int flags)
538 handle_t *handle = ext4_journal_current_handle();
539 struct ext4_map_blocks map;
540 int ret = 0, started = 0;
544 map.m_len = bh->b_size >> inode->i_blkbits;
546 if (flags && !handle) {
547 /* Direct IO write... */
548 if (map.m_len > DIO_MAX_BLOCKS)
549 map.m_len = DIO_MAX_BLOCKS;
550 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
551 handle = ext4_journal_start(inode, dio_credits);
552 if (IS_ERR(handle)) {
553 ret = PTR_ERR(handle);
559 ret = ext4_map_blocks(handle, inode, &map, flags);
561 map_bh(bh, inode->i_sb, map.m_pblk);
562 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
563 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
567 ext4_journal_stop(handle);
571 int ext4_get_block(struct inode *inode, sector_t iblock,
572 struct buffer_head *bh, int create)
574 return _ext4_get_block(inode, iblock, bh,
575 create ? EXT4_GET_BLOCKS_CREATE : 0);
579 * `handle' can be NULL if create is zero
581 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
582 ext4_lblk_t block, int create, int *errp)
584 struct ext4_map_blocks map;
585 struct buffer_head *bh;
588 J_ASSERT(handle != NULL || create == 0);
592 err = ext4_map_blocks(handle, inode, &map,
593 create ? EXT4_GET_BLOCKS_CREATE : 0);
601 bh = sb_getblk(inode->i_sb, map.m_pblk);
606 if (map.m_flags & EXT4_MAP_NEW) {
607 J_ASSERT(create != 0);
608 J_ASSERT(handle != NULL);
611 * Now that we do not always journal data, we should
612 * keep in mind whether this should always journal the
613 * new buffer as metadata. For now, regular file
614 * writes use ext4_get_block instead, so it's not a
618 BUFFER_TRACE(bh, "call get_create_access");
619 fatal = ext4_journal_get_create_access(handle, bh);
620 if (!fatal && !buffer_uptodate(bh)) {
621 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
622 set_buffer_uptodate(bh);
625 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
626 err = ext4_handle_dirty_metadata(handle, inode, bh);
630 BUFFER_TRACE(bh, "not a new buffer");
640 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
641 ext4_lblk_t block, int create, int *err)
643 struct buffer_head *bh;
645 bh = ext4_getblk(handle, inode, block, create, err);
648 if (buffer_uptodate(bh))
650 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
652 if (buffer_uptodate(bh))
659 static int walk_page_buffers(handle_t *handle,
660 struct buffer_head *head,
664 int (*fn)(handle_t *handle,
665 struct buffer_head *bh))
667 struct buffer_head *bh;
668 unsigned block_start, block_end;
669 unsigned blocksize = head->b_size;
671 struct buffer_head *next;
673 for (bh = head, block_start = 0;
674 ret == 0 && (bh != head || !block_start);
675 block_start = block_end, bh = next) {
676 next = bh->b_this_page;
677 block_end = block_start + blocksize;
678 if (block_end <= from || block_start >= to) {
679 if (partial && !buffer_uptodate(bh))
683 err = (*fn)(handle, bh);
691 * To preserve ordering, it is essential that the hole instantiation and
692 * the data write be encapsulated in a single transaction. We cannot
693 * close off a transaction and start a new one between the ext4_get_block()
694 * and the commit_write(). So doing the jbd2_journal_start at the start of
695 * prepare_write() is the right place.
697 * Also, this function can nest inside ext4_writepage() ->
698 * block_write_full_page(). In that case, we *know* that ext4_writepage()
699 * has generated enough buffer credits to do the whole page. So we won't
700 * block on the journal in that case, which is good, because the caller may
703 * By accident, ext4 can be reentered when a transaction is open via
704 * quota file writes. If we were to commit the transaction while thus
705 * reentered, there can be a deadlock - we would be holding a quota
706 * lock, and the commit would never complete if another thread had a
707 * transaction open and was blocking on the quota lock - a ranking
710 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
711 * will _not_ run commit under these circumstances because handle->h_ref
712 * is elevated. We'll still have enough credits for the tiny quotafile
715 static int do_journal_get_write_access(handle_t *handle,
716 struct buffer_head *bh)
718 int dirty = buffer_dirty(bh);
721 if (!buffer_mapped(bh) || buffer_freed(bh))
724 * __block_write_begin() could have dirtied some buffers. Clean
725 * the dirty bit as jbd2_journal_get_write_access() could complain
726 * otherwise about fs integrity issues. Setting of the dirty bit
727 * by __block_write_begin() isn't a real problem here as we clear
728 * the bit before releasing a page lock and thus writeback cannot
729 * ever write the buffer.
732 clear_buffer_dirty(bh);
733 ret = ext4_journal_get_write_access(handle, bh);
735 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
739 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
740 struct buffer_head *bh_result, int create);
741 static int ext4_write_begin(struct file *file, struct address_space *mapping,
742 loff_t pos, unsigned len, unsigned flags,
743 struct page **pagep, void **fsdata)
745 struct inode *inode = mapping->host;
746 int ret, needed_blocks;
753 trace_ext4_write_begin(inode, pos, len, flags);
755 * Reserve one block more for addition to orphan list in case
756 * we allocate blocks but write fails for some reason
758 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
759 index = pos >> PAGE_CACHE_SHIFT;
760 from = pos & (PAGE_CACHE_SIZE - 1);
764 handle = ext4_journal_start(inode, needed_blocks);
765 if (IS_ERR(handle)) {
766 ret = PTR_ERR(handle);
770 /* We cannot recurse into the filesystem as the transaction is already
772 flags |= AOP_FLAG_NOFS;
774 page = grab_cache_page_write_begin(mapping, index, flags);
776 ext4_journal_stop(handle);
782 if (ext4_should_dioread_nolock(inode))
783 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
785 ret = __block_write_begin(page, pos, len, ext4_get_block);
787 if (!ret && ext4_should_journal_data(inode)) {
788 ret = walk_page_buffers(handle, page_buffers(page),
789 from, to, NULL, do_journal_get_write_access);
794 page_cache_release(page);
796 * __block_write_begin may have instantiated a few blocks
797 * outside i_size. Trim these off again. Don't need
798 * i_size_read because we hold i_mutex.
800 * Add inode to orphan list in case we crash before
803 if (pos + len > inode->i_size && ext4_can_truncate(inode))
804 ext4_orphan_add(handle, inode);
806 ext4_journal_stop(handle);
807 if (pos + len > inode->i_size) {
808 ext4_truncate_failed_write(inode);
810 * If truncate failed early the inode might
811 * still be on the orphan list; we need to
812 * make sure the inode is removed from the
813 * orphan list in that case.
816 ext4_orphan_del(NULL, inode);
820 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
826 /* For write_end() in data=journal mode */
827 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
829 if (!buffer_mapped(bh) || buffer_freed(bh))
831 set_buffer_uptodate(bh);
832 return ext4_handle_dirty_metadata(handle, NULL, bh);
835 static int ext4_generic_write_end(struct file *file,
836 struct address_space *mapping,
837 loff_t pos, unsigned len, unsigned copied,
838 struct page *page, void *fsdata)
840 int i_size_changed = 0;
841 struct inode *inode = mapping->host;
842 handle_t *handle = ext4_journal_current_handle();
844 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
847 * No need to use i_size_read() here, the i_size
848 * cannot change under us because we hold i_mutex.
850 * But it's important to update i_size while still holding page lock:
851 * page writeout could otherwise come in and zero beyond i_size.
853 if (pos + copied > inode->i_size) {
854 i_size_write(inode, pos + copied);
858 if (pos + copied > EXT4_I(inode)->i_disksize) {
859 /* We need to mark inode dirty even if
860 * new_i_size is less that inode->i_size
861 * bu greater than i_disksize.(hint delalloc)
863 ext4_update_i_disksize(inode, (pos + copied));
867 page_cache_release(page);
870 * Don't mark the inode dirty under page lock. First, it unnecessarily
871 * makes the holding time of page lock longer. Second, it forces lock
872 * ordering of page lock and transaction start for journaling
876 ext4_mark_inode_dirty(handle, inode);
882 * We need to pick up the new inode size which generic_commit_write gave us
883 * `file' can be NULL - eg, when called from page_symlink().
885 * ext4 never places buffers on inode->i_mapping->private_list. metadata
886 * buffers are managed internally.
888 static int ext4_ordered_write_end(struct file *file,
889 struct address_space *mapping,
890 loff_t pos, unsigned len, unsigned copied,
891 struct page *page, void *fsdata)
893 handle_t *handle = ext4_journal_current_handle();
894 struct inode *inode = mapping->host;
897 trace_ext4_ordered_write_end(inode, pos, len, copied);
898 ret = ext4_jbd2_file_inode(handle, inode);
901 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
904 if (pos + len > inode->i_size && ext4_can_truncate(inode))
905 /* if we have allocated more blocks and copied
906 * less. We will have blocks allocated outside
907 * inode->i_size. So truncate them
909 ext4_orphan_add(handle, inode);
913 ret2 = ext4_journal_stop(handle);
917 if (pos + len > inode->i_size) {
918 ext4_truncate_failed_write(inode);
920 * If truncate failed early the inode might still be
921 * on the orphan list; we need to make sure the inode
922 * is removed from the orphan list in that case.
925 ext4_orphan_del(NULL, inode);
929 return ret ? ret : copied;
932 static int ext4_writeback_write_end(struct file *file,
933 struct address_space *mapping,
934 loff_t pos, unsigned len, unsigned copied,
935 struct page *page, void *fsdata)
937 handle_t *handle = ext4_journal_current_handle();
938 struct inode *inode = mapping->host;
941 trace_ext4_writeback_write_end(inode, pos, len, copied);
942 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
945 if (pos + len > inode->i_size && ext4_can_truncate(inode))
946 /* if we have allocated more blocks and copied
947 * less. We will have blocks allocated outside
948 * inode->i_size. So truncate them
950 ext4_orphan_add(handle, inode);
955 ret2 = ext4_journal_stop(handle);
959 if (pos + len > inode->i_size) {
960 ext4_truncate_failed_write(inode);
962 * If truncate failed early the inode might still be
963 * on the orphan list; we need to make sure the inode
964 * is removed from the orphan list in that case.
967 ext4_orphan_del(NULL, inode);
970 return ret ? ret : copied;
973 static int ext4_journalled_write_end(struct file *file,
974 struct address_space *mapping,
975 loff_t pos, unsigned len, unsigned copied,
976 struct page *page, void *fsdata)
978 handle_t *handle = ext4_journal_current_handle();
979 struct inode *inode = mapping->host;
985 trace_ext4_journalled_write_end(inode, pos, len, copied);
986 from = pos & (PAGE_CACHE_SIZE - 1);
989 BUG_ON(!ext4_handle_valid(handle));
992 if (!PageUptodate(page))
994 page_zero_new_buffers(page, from+copied, to);
997 ret = walk_page_buffers(handle, page_buffers(page), from,
998 to, &partial, write_end_fn);
1000 SetPageUptodate(page);
1001 new_i_size = pos + copied;
1002 if (new_i_size > inode->i_size)
1003 i_size_write(inode, pos+copied);
1004 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1005 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1006 if (new_i_size > EXT4_I(inode)->i_disksize) {
1007 ext4_update_i_disksize(inode, new_i_size);
1008 ret2 = ext4_mark_inode_dirty(handle, inode);
1014 page_cache_release(page);
1015 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1016 /* if we have allocated more blocks and copied
1017 * less. We will have blocks allocated outside
1018 * inode->i_size. So truncate them
1020 ext4_orphan_add(handle, inode);
1022 ret2 = ext4_journal_stop(handle);
1025 if (pos + len > inode->i_size) {
1026 ext4_truncate_failed_write(inode);
1028 * If truncate failed early the inode might still be
1029 * on the orphan list; we need to make sure the inode
1030 * is removed from the orphan list in that case.
1033 ext4_orphan_del(NULL, inode);
1036 return ret ? ret : copied;
1040 * Reserve a single block located at lblock
1042 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1045 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1046 struct ext4_inode_info *ei = EXT4_I(inode);
1047 unsigned long md_needed;
1051 * recalculate the amount of metadata blocks to reserve
1052 * in order to allocate nrblocks
1053 * worse case is one extent per block
1056 spin_lock(&ei->i_block_reservation_lock);
1057 md_needed = ext4_calc_metadata_amount(inode, lblock);
1058 trace_ext4_da_reserve_space(inode, md_needed);
1059 spin_unlock(&ei->i_block_reservation_lock);
1062 * We will charge metadata quota at writeout time; this saves
1063 * us from metadata over-estimation, though we may go over by
1064 * a small amount in the end. Here we just reserve for data.
1066 ret = dquot_reserve_block(inode, 1);
1070 * We do still charge estimated metadata to the sb though;
1071 * we cannot afford to run out of free blocks.
1073 if (ext4_claim_free_blocks(sbi, md_needed + 1, 0)) {
1074 dquot_release_reservation_block(inode, 1);
1075 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1081 spin_lock(&ei->i_block_reservation_lock);
1082 ei->i_reserved_data_blocks++;
1083 ei->i_reserved_meta_blocks += md_needed;
1084 spin_unlock(&ei->i_block_reservation_lock);
1086 return 0; /* success */
1089 static void ext4_da_release_space(struct inode *inode, int to_free)
1091 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1092 struct ext4_inode_info *ei = EXT4_I(inode);
1095 return; /* Nothing to release, exit */
1097 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1099 trace_ext4_da_release_space(inode, to_free);
1100 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1102 * if there aren't enough reserved blocks, then the
1103 * counter is messed up somewhere. Since this
1104 * function is called from invalidate page, it's
1105 * harmless to return without any action.
1107 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1108 "ino %lu, to_free %d with only %d reserved "
1109 "data blocks\n", inode->i_ino, to_free,
1110 ei->i_reserved_data_blocks);
1112 to_free = ei->i_reserved_data_blocks;
1114 ei->i_reserved_data_blocks -= to_free;
1116 if (ei->i_reserved_data_blocks == 0) {
1118 * We can release all of the reserved metadata blocks
1119 * only when we have written all of the delayed
1120 * allocation blocks.
1122 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1123 ei->i_reserved_meta_blocks);
1124 ei->i_reserved_meta_blocks = 0;
1125 ei->i_da_metadata_calc_len = 0;
1128 /* update fs dirty data blocks counter */
1129 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1131 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1133 dquot_release_reservation_block(inode, to_free);
1136 static void ext4_da_page_release_reservation(struct page *page,
1137 unsigned long offset)
1140 struct buffer_head *head, *bh;
1141 unsigned int curr_off = 0;
1143 head = page_buffers(page);
1146 unsigned int next_off = curr_off + bh->b_size;
1148 if ((offset <= curr_off) && (buffer_delay(bh))) {
1150 clear_buffer_delay(bh);
1152 curr_off = next_off;
1153 } while ((bh = bh->b_this_page) != head);
1154 ext4_da_release_space(page->mapping->host, to_release);
1158 * Delayed allocation stuff
1162 * mpage_da_submit_io - walks through extent of pages and try to write
1163 * them with writepage() call back
1165 * @mpd->inode: inode
1166 * @mpd->first_page: first page of the extent
1167 * @mpd->next_page: page after the last page of the extent
1169 * By the time mpage_da_submit_io() is called we expect all blocks
1170 * to be allocated. this may be wrong if allocation failed.
1172 * As pages are already locked by write_cache_pages(), we can't use it
1174 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1175 struct ext4_map_blocks *map)
1177 struct pagevec pvec;
1178 unsigned long index, end;
1179 int ret = 0, err, nr_pages, i;
1180 struct inode *inode = mpd->inode;
1181 struct address_space *mapping = inode->i_mapping;
1182 loff_t size = i_size_read(inode);
1183 unsigned int len, block_start;
1184 struct buffer_head *bh, *page_bufs = NULL;
1185 int journal_data = ext4_should_journal_data(inode);
1186 sector_t pblock = 0, cur_logical = 0;
1187 struct ext4_io_submit io_submit;
1189 BUG_ON(mpd->next_page <= mpd->first_page);
1190 memset(&io_submit, 0, sizeof(io_submit));
1192 * We need to start from the first_page to the next_page - 1
1193 * to make sure we also write the mapped dirty buffer_heads.
1194 * If we look at mpd->b_blocknr we would only be looking
1195 * at the currently mapped buffer_heads.
1197 index = mpd->first_page;
1198 end = mpd->next_page - 1;
1200 pagevec_init(&pvec, 0);
1201 while (index <= end) {
1202 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1205 for (i = 0; i < nr_pages; i++) {
1206 int commit_write = 0, skip_page = 0;
1207 struct page *page = pvec.pages[i];
1209 index = page->index;
1213 if (index == size >> PAGE_CACHE_SHIFT)
1214 len = size & ~PAGE_CACHE_MASK;
1216 len = PAGE_CACHE_SIZE;
1218 cur_logical = index << (PAGE_CACHE_SHIFT -
1220 pblock = map->m_pblk + (cur_logical -
1225 BUG_ON(!PageLocked(page));
1226 BUG_ON(PageWriteback(page));
1229 * If the page does not have buffers (for
1230 * whatever reason), try to create them using
1231 * __block_write_begin. If this fails,
1232 * skip the page and move on.
1234 if (!page_has_buffers(page)) {
1235 if (__block_write_begin(page, 0, len,
1236 noalloc_get_block_write)) {
1244 bh = page_bufs = page_buffers(page);
1249 if (map && (cur_logical >= map->m_lblk) &&
1250 (cur_logical <= (map->m_lblk +
1251 (map->m_len - 1)))) {
1252 if (buffer_delay(bh)) {
1253 clear_buffer_delay(bh);
1254 bh->b_blocknr = pblock;
1256 if (buffer_unwritten(bh) ||
1258 BUG_ON(bh->b_blocknr != pblock);
1259 if (map->m_flags & EXT4_MAP_UNINIT)
1260 set_buffer_uninit(bh);
1261 clear_buffer_unwritten(bh);
1264 /* skip page if block allocation undone */
1265 if (buffer_delay(bh) || buffer_unwritten(bh))
1267 bh = bh->b_this_page;
1268 block_start += bh->b_size;
1271 } while (bh != page_bufs);
1277 /* mark the buffer_heads as dirty & uptodate */
1278 block_commit_write(page, 0, len);
1280 clear_page_dirty_for_io(page);
1282 * Delalloc doesn't support data journalling,
1283 * but eventually maybe we'll lift this
1286 if (unlikely(journal_data && PageChecked(page)))
1287 err = __ext4_journalled_writepage(page, len);
1288 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1289 err = ext4_bio_write_page(&io_submit, page,
1291 else if (buffer_uninit(page_bufs)) {
1292 ext4_set_bh_endio(page_bufs, inode);
1293 err = block_write_full_page_endio(page,
1294 noalloc_get_block_write,
1295 mpd->wbc, ext4_end_io_buffer_write);
1297 err = block_write_full_page(page,
1298 noalloc_get_block_write, mpd->wbc);
1301 mpd->pages_written++;
1303 * In error case, we have to continue because
1304 * remaining pages are still locked
1309 pagevec_release(&pvec);
1311 ext4_io_submit(&io_submit);
1315 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1319 struct pagevec pvec;
1320 struct inode *inode = mpd->inode;
1321 struct address_space *mapping = inode->i_mapping;
1323 index = mpd->first_page;
1324 end = mpd->next_page - 1;
1325 while (index <= end) {
1326 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1329 for (i = 0; i < nr_pages; i++) {
1330 struct page *page = pvec.pages[i];
1331 if (page->index > end)
1333 BUG_ON(!PageLocked(page));
1334 BUG_ON(PageWriteback(page));
1335 block_invalidatepage(page, 0);
1336 ClearPageUptodate(page);
1339 index = pvec.pages[nr_pages - 1]->index + 1;
1340 pagevec_release(&pvec);
1345 static void ext4_print_free_blocks(struct inode *inode)
1347 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1348 printk(KERN_CRIT "Total free blocks count %lld\n",
1349 ext4_count_free_blocks(inode->i_sb));
1350 printk(KERN_CRIT "Free/Dirty block details\n");
1351 printk(KERN_CRIT "free_blocks=%lld\n",
1352 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
1353 printk(KERN_CRIT "dirty_blocks=%lld\n",
1354 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1355 printk(KERN_CRIT "Block reservation details\n");
1356 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
1357 EXT4_I(inode)->i_reserved_data_blocks);
1358 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
1359 EXT4_I(inode)->i_reserved_meta_blocks);
1364 * mpage_da_map_and_submit - go through given space, map them
1365 * if necessary, and then submit them for I/O
1367 * @mpd - bh describing space
1369 * The function skips space we know is already mapped to disk blocks.
1372 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1374 int err, blks, get_blocks_flags;
1375 struct ext4_map_blocks map, *mapp = NULL;
1376 sector_t next = mpd->b_blocknr;
1377 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1378 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1379 handle_t *handle = NULL;
1382 * If the blocks are mapped already, or we couldn't accumulate
1383 * any blocks, then proceed immediately to the submission stage.
1385 if ((mpd->b_size == 0) ||
1386 ((mpd->b_state & (1 << BH_Mapped)) &&
1387 !(mpd->b_state & (1 << BH_Delay)) &&
1388 !(mpd->b_state & (1 << BH_Unwritten))))
1391 handle = ext4_journal_current_handle();
1395 * Call ext4_map_blocks() to allocate any delayed allocation
1396 * blocks, or to convert an uninitialized extent to be
1397 * initialized (in the case where we have written into
1398 * one or more preallocated blocks).
1400 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1401 * indicate that we are on the delayed allocation path. This
1402 * affects functions in many different parts of the allocation
1403 * call path. This flag exists primarily because we don't
1404 * want to change *many* call functions, so ext4_map_blocks()
1405 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1406 * inode's allocation semaphore is taken.
1408 * If the blocks in questions were delalloc blocks, set
1409 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1410 * variables are updated after the blocks have been allocated.
1413 map.m_len = max_blocks;
1414 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1415 if (ext4_should_dioread_nolock(mpd->inode))
1416 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1417 if (mpd->b_state & (1 << BH_Delay))
1418 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1420 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1422 struct super_block *sb = mpd->inode->i_sb;
1426 * If get block returns EAGAIN or ENOSPC and there
1427 * appears to be free blocks we will just let
1428 * mpage_da_submit_io() unlock all of the pages.
1433 if (err == -ENOSPC &&
1434 ext4_count_free_blocks(sb)) {
1440 * get block failure will cause us to loop in
1441 * writepages, because a_ops->writepage won't be able
1442 * to make progress. The page will be redirtied by
1443 * writepage and writepages will again try to write
1446 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1447 ext4_msg(sb, KERN_CRIT,
1448 "delayed block allocation failed for inode %lu "
1449 "at logical offset %llu with max blocks %zd "
1450 "with error %d", mpd->inode->i_ino,
1451 (unsigned long long) next,
1452 mpd->b_size >> mpd->inode->i_blkbits, err);
1453 ext4_msg(sb, KERN_CRIT,
1454 "This should not happen!! Data will be lost\n");
1456 ext4_print_free_blocks(mpd->inode);
1458 /* invalidate all the pages */
1459 ext4_da_block_invalidatepages(mpd);
1461 /* Mark this page range as having been completed */
1468 if (map.m_flags & EXT4_MAP_NEW) {
1469 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1472 for (i = 0; i < map.m_len; i++)
1473 unmap_underlying_metadata(bdev, map.m_pblk + i);
1476 if (ext4_should_order_data(mpd->inode)) {
1477 err = ext4_jbd2_file_inode(handle, mpd->inode);
1479 /* This only happens if the journal is aborted */
1484 * Update on-disk size along with block allocation.
1486 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1487 if (disksize > i_size_read(mpd->inode))
1488 disksize = i_size_read(mpd->inode);
1489 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1490 ext4_update_i_disksize(mpd->inode, disksize);
1491 err = ext4_mark_inode_dirty(handle, mpd->inode);
1493 ext4_error(mpd->inode->i_sb,
1494 "Failed to mark inode %lu dirty",
1499 mpage_da_submit_io(mpd, mapp);
1503 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1504 (1 << BH_Delay) | (1 << BH_Unwritten))
1507 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1509 * @mpd->lbh - extent of blocks
1510 * @logical - logical number of the block in the file
1511 * @bh - bh of the block (used to access block's state)
1513 * the function is used to collect contig. blocks in same state
1515 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1516 sector_t logical, size_t b_size,
1517 unsigned long b_state)
1520 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1523 * XXX Don't go larger than mballoc is willing to allocate
1524 * This is a stopgap solution. We eventually need to fold
1525 * mpage_da_submit_io() into this function and then call
1526 * ext4_map_blocks() multiple times in a loop
1528 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1531 /* check if thereserved journal credits might overflow */
1532 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1533 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1535 * With non-extent format we are limited by the journal
1536 * credit available. Total credit needed to insert
1537 * nrblocks contiguous blocks is dependent on the
1538 * nrblocks. So limit nrblocks.
1541 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1542 EXT4_MAX_TRANS_DATA) {
1544 * Adding the new buffer_head would make it cross the
1545 * allowed limit for which we have journal credit
1546 * reserved. So limit the new bh->b_size
1548 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1549 mpd->inode->i_blkbits;
1550 /* we will do mpage_da_submit_io in the next loop */
1554 * First block in the extent
1556 if (mpd->b_size == 0) {
1557 mpd->b_blocknr = logical;
1558 mpd->b_size = b_size;
1559 mpd->b_state = b_state & BH_FLAGS;
1563 next = mpd->b_blocknr + nrblocks;
1565 * Can we merge the block to our big extent?
1567 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1568 mpd->b_size += b_size;
1574 * We couldn't merge the block to our extent, so we
1575 * need to flush current extent and start new one
1577 mpage_da_map_and_submit(mpd);
1581 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1583 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1587 * This is a special get_blocks_t callback which is used by
1588 * ext4_da_write_begin(). It will either return mapped block or
1589 * reserve space for a single block.
1591 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1592 * We also have b_blocknr = -1 and b_bdev initialized properly
1594 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1595 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1596 * initialized properly.
1598 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1599 struct buffer_head *bh, int create)
1601 struct ext4_map_blocks map;
1603 sector_t invalid_block = ~((sector_t) 0xffff);
1605 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1608 BUG_ON(create == 0);
1609 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1611 map.m_lblk = iblock;
1615 * first, we need to know whether the block is allocated already
1616 * preallocated blocks are unmapped but should treated
1617 * the same as allocated blocks.
1619 ret = ext4_map_blocks(NULL, inode, &map, 0);
1623 if (buffer_delay(bh))
1624 return 0; /* Not sure this could or should happen */
1626 * XXX: __block_write_begin() unmaps passed block, is it OK?
1628 ret = ext4_da_reserve_space(inode, iblock);
1630 /* not enough space to reserve */
1633 map_bh(bh, inode->i_sb, invalid_block);
1635 set_buffer_delay(bh);
1639 map_bh(bh, inode->i_sb, map.m_pblk);
1640 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1642 if (buffer_unwritten(bh)) {
1643 /* A delayed write to unwritten bh should be marked
1644 * new and mapped. Mapped ensures that we don't do
1645 * get_block multiple times when we write to the same
1646 * offset and new ensures that we do proper zero out
1647 * for partial write.
1650 set_buffer_mapped(bh);
1656 * This function is used as a standard get_block_t calback function
1657 * when there is no desire to allocate any blocks. It is used as a
1658 * callback function for block_write_begin() and block_write_full_page().
1659 * These functions should only try to map a single block at a time.
1661 * Since this function doesn't do block allocations even if the caller
1662 * requests it by passing in create=1, it is critically important that
1663 * any caller checks to make sure that any buffer heads are returned
1664 * by this function are either all already mapped or marked for
1665 * delayed allocation before calling block_write_full_page(). Otherwise,
1666 * b_blocknr could be left unitialized, and the page write functions will
1667 * be taken by surprise.
1669 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1670 struct buffer_head *bh_result, int create)
1672 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1673 return _ext4_get_block(inode, iblock, bh_result, 0);
1676 static int bget_one(handle_t *handle, struct buffer_head *bh)
1682 static int bput_one(handle_t *handle, struct buffer_head *bh)
1688 static int __ext4_journalled_writepage(struct page *page,
1691 struct address_space *mapping = page->mapping;
1692 struct inode *inode = mapping->host;
1693 struct buffer_head *page_bufs;
1694 handle_t *handle = NULL;
1698 ClearPageChecked(page);
1699 page_bufs = page_buffers(page);
1701 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1702 /* As soon as we unlock the page, it can go away, but we have
1703 * references to buffers so we are safe */
1706 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1707 if (IS_ERR(handle)) {
1708 ret = PTR_ERR(handle);
1712 BUG_ON(!ext4_handle_valid(handle));
1714 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1715 do_journal_get_write_access);
1717 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1721 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1722 err = ext4_journal_stop(handle);
1726 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
1727 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1732 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
1733 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
1736 * Note that we don't need to start a transaction unless we're journaling data
1737 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1738 * need to file the inode to the transaction's list in ordered mode because if
1739 * we are writing back data added by write(), the inode is already there and if
1740 * we are writing back data modified via mmap(), no one guarantees in which
1741 * transaction the data will hit the disk. In case we are journaling data, we
1742 * cannot start transaction directly because transaction start ranks above page
1743 * lock so we have to do some magic.
1745 * This function can get called via...
1746 * - ext4_da_writepages after taking page lock (have journal handle)
1747 * - journal_submit_inode_data_buffers (no journal handle)
1748 * - shrink_page_list via pdflush (no journal handle)
1749 * - grab_page_cache when doing write_begin (have journal handle)
1751 * We don't do any block allocation in this function. If we have page with
1752 * multiple blocks we need to write those buffer_heads that are mapped. This
1753 * is important for mmaped based write. So if we do with blocksize 1K
1754 * truncate(f, 1024);
1755 * a = mmap(f, 0, 4096);
1757 * truncate(f, 4096);
1758 * we have in the page first buffer_head mapped via page_mkwrite call back
1759 * but other bufer_heads would be unmapped but dirty(dirty done via the
1760 * do_wp_page). So writepage should write the first block. If we modify
1761 * the mmap area beyond 1024 we will again get a page_fault and the
1762 * page_mkwrite callback will do the block allocation and mark the
1763 * buffer_heads mapped.
1765 * We redirty the page if we have any buffer_heads that is either delay or
1766 * unwritten in the page.
1768 * We can get recursively called as show below.
1770 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1773 * But since we don't do any block allocation we should not deadlock.
1774 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1776 static int ext4_writepage(struct page *page,
1777 struct writeback_control *wbc)
1779 int ret = 0, commit_write = 0;
1782 struct buffer_head *page_bufs = NULL;
1783 struct inode *inode = page->mapping->host;
1785 trace_ext4_writepage(page);
1786 size = i_size_read(inode);
1787 if (page->index == size >> PAGE_CACHE_SHIFT)
1788 len = size & ~PAGE_CACHE_MASK;
1790 len = PAGE_CACHE_SIZE;
1793 * If the page does not have buffers (for whatever reason),
1794 * try to create them using __block_write_begin. If this
1795 * fails, redirty the page and move on.
1797 if (!page_has_buffers(page)) {
1798 if (__block_write_begin(page, 0, len,
1799 noalloc_get_block_write)) {
1801 redirty_page_for_writepage(wbc, page);
1807 page_bufs = page_buffers(page);
1808 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1809 ext4_bh_delay_or_unwritten)) {
1811 * We don't want to do block allocation, so redirty
1812 * the page and return. We may reach here when we do
1813 * a journal commit via journal_submit_inode_data_buffers.
1814 * We can also reach here via shrink_page_list but it
1815 * should never be for direct reclaim so warn if that
1818 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
1823 /* now mark the buffer_heads as dirty and uptodate */
1824 block_commit_write(page, 0, len);
1826 if (PageChecked(page) && ext4_should_journal_data(inode))
1828 * It's mmapped pagecache. Add buffers and journal it. There
1829 * doesn't seem much point in redirtying the page here.
1831 return __ext4_journalled_writepage(page, len);
1833 if (buffer_uninit(page_bufs)) {
1834 ext4_set_bh_endio(page_bufs, inode);
1835 ret = block_write_full_page_endio(page, noalloc_get_block_write,
1836 wbc, ext4_end_io_buffer_write);
1838 ret = block_write_full_page(page, noalloc_get_block_write,
1845 * This is called via ext4_da_writepages() to
1846 * calculate the total number of credits to reserve to fit
1847 * a single extent allocation into a single transaction,
1848 * ext4_da_writpeages() will loop calling this before
1849 * the block allocation.
1852 static int ext4_da_writepages_trans_blocks(struct inode *inode)
1854 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
1857 * With non-extent format the journal credit needed to
1858 * insert nrblocks contiguous block is dependent on
1859 * number of contiguous block. So we will limit
1860 * number of contiguous block to a sane value
1862 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
1863 (max_blocks > EXT4_MAX_TRANS_DATA))
1864 max_blocks = EXT4_MAX_TRANS_DATA;
1866 return ext4_chunk_trans_blocks(inode, max_blocks);
1870 * write_cache_pages_da - walk the list of dirty pages of the given
1871 * address space and accumulate pages that need writing, and call
1872 * mpage_da_map_and_submit to map a single contiguous memory region
1873 * and then write them.
1875 static int write_cache_pages_da(struct address_space *mapping,
1876 struct writeback_control *wbc,
1877 struct mpage_da_data *mpd,
1878 pgoff_t *done_index)
1880 struct buffer_head *bh, *head;
1881 struct inode *inode = mapping->host;
1882 struct pagevec pvec;
1883 unsigned int nr_pages;
1886 long nr_to_write = wbc->nr_to_write;
1887 int i, tag, ret = 0;
1889 memset(mpd, 0, sizeof(struct mpage_da_data));
1892 pagevec_init(&pvec, 0);
1893 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1894 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1896 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1897 tag = PAGECACHE_TAG_TOWRITE;
1899 tag = PAGECACHE_TAG_DIRTY;
1901 *done_index = index;
1902 while (index <= end) {
1903 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1904 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1908 for (i = 0; i < nr_pages; i++) {
1909 struct page *page = pvec.pages[i];
1912 * At this point, the page may be truncated or
1913 * invalidated (changing page->mapping to NULL), or
1914 * even swizzled back from swapper_space to tmpfs file
1915 * mapping. However, page->index will not change
1916 * because we have a reference on the page.
1918 if (page->index > end)
1921 *done_index = page->index + 1;
1924 * If we can't merge this page, and we have
1925 * accumulated an contiguous region, write it
1927 if ((mpd->next_page != page->index) &&
1928 (mpd->next_page != mpd->first_page)) {
1929 mpage_da_map_and_submit(mpd);
1930 goto ret_extent_tail;
1936 * If the page is no longer dirty, or its
1937 * mapping no longer corresponds to inode we
1938 * are writing (which means it has been
1939 * truncated or invalidated), or the page is
1940 * already under writeback and we are not
1941 * doing a data integrity writeback, skip the page
1943 if (!PageDirty(page) ||
1944 (PageWriteback(page) &&
1945 (wbc->sync_mode == WB_SYNC_NONE)) ||
1946 unlikely(page->mapping != mapping)) {
1951 wait_on_page_writeback(page);
1952 BUG_ON(PageWriteback(page));
1954 if (mpd->next_page != page->index)
1955 mpd->first_page = page->index;
1956 mpd->next_page = page->index + 1;
1957 logical = (sector_t) page->index <<
1958 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1960 if (!page_has_buffers(page)) {
1961 mpage_add_bh_to_extent(mpd, logical,
1963 (1 << BH_Dirty) | (1 << BH_Uptodate));
1965 goto ret_extent_tail;
1968 * Page with regular buffer heads,
1969 * just add all dirty ones
1971 head = page_buffers(page);
1974 BUG_ON(buffer_locked(bh));
1976 * We need to try to allocate
1977 * unmapped blocks in the same page.
1978 * Otherwise we won't make progress
1979 * with the page in ext4_writepage
1981 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
1982 mpage_add_bh_to_extent(mpd, logical,
1986 goto ret_extent_tail;
1987 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
1989 * mapped dirty buffer. We need
1990 * to update the b_state
1991 * because we look at b_state
1992 * in mpage_da_map_blocks. We
1993 * don't update b_size because
1994 * if we find an unmapped
1995 * buffer_head later we need to
1996 * use the b_state flag of that
1999 if (mpd->b_size == 0)
2000 mpd->b_state = bh->b_state & BH_FLAGS;
2003 } while ((bh = bh->b_this_page) != head);
2006 if (nr_to_write > 0) {
2008 if (nr_to_write == 0 &&
2009 wbc->sync_mode == WB_SYNC_NONE)
2011 * We stop writing back only if we are
2012 * not doing integrity sync. In case of
2013 * integrity sync we have to keep going
2014 * because someone may be concurrently
2015 * dirtying pages, and we might have
2016 * synced a lot of newly appeared dirty
2017 * pages, but have not synced all of the
2023 pagevec_release(&pvec);
2028 ret = MPAGE_DA_EXTENT_TAIL;
2030 pagevec_release(&pvec);
2036 static int ext4_da_writepages(struct address_space *mapping,
2037 struct writeback_control *wbc)
2040 int range_whole = 0;
2041 handle_t *handle = NULL;
2042 struct mpage_da_data mpd;
2043 struct inode *inode = mapping->host;
2044 int pages_written = 0;
2045 unsigned int max_pages;
2046 int range_cyclic, cycled = 1, io_done = 0;
2047 int needed_blocks, ret = 0;
2048 long desired_nr_to_write, nr_to_writebump = 0;
2049 loff_t range_start = wbc->range_start;
2050 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2051 pgoff_t done_index = 0;
2054 trace_ext4_da_writepages(inode, wbc);
2057 * No pages to write? This is mainly a kludge to avoid starting
2058 * a transaction for special inodes like journal inode on last iput()
2059 * because that could violate lock ordering on umount
2061 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2065 * If the filesystem has aborted, it is read-only, so return
2066 * right away instead of dumping stack traces later on that
2067 * will obscure the real source of the problem. We test
2068 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2069 * the latter could be true if the filesystem is mounted
2070 * read-only, and in that case, ext4_da_writepages should
2071 * *never* be called, so if that ever happens, we would want
2074 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2077 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2080 range_cyclic = wbc->range_cyclic;
2081 if (wbc->range_cyclic) {
2082 index = mapping->writeback_index;
2085 wbc->range_start = index << PAGE_CACHE_SHIFT;
2086 wbc->range_end = LLONG_MAX;
2087 wbc->range_cyclic = 0;
2090 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2091 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2095 * This works around two forms of stupidity. The first is in
2096 * the writeback code, which caps the maximum number of pages
2097 * written to be 1024 pages. This is wrong on multiple
2098 * levels; different architectues have a different page size,
2099 * which changes the maximum amount of data which gets
2100 * written. Secondly, 4 megabytes is way too small. XFS
2101 * forces this value to be 16 megabytes by multiplying
2102 * nr_to_write parameter by four, and then relies on its
2103 * allocator to allocate larger extents to make them
2104 * contiguous. Unfortunately this brings us to the second
2105 * stupidity, which is that ext4's mballoc code only allocates
2106 * at most 2048 blocks. So we force contiguous writes up to
2107 * the number of dirty blocks in the inode, or
2108 * sbi->max_writeback_mb_bump whichever is smaller.
2110 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2111 if (!range_cyclic && range_whole) {
2112 if (wbc->nr_to_write == LONG_MAX)
2113 desired_nr_to_write = wbc->nr_to_write;
2115 desired_nr_to_write = wbc->nr_to_write * 8;
2117 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2119 if (desired_nr_to_write > max_pages)
2120 desired_nr_to_write = max_pages;
2122 if (wbc->nr_to_write < desired_nr_to_write) {
2123 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2124 wbc->nr_to_write = desired_nr_to_write;
2128 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2129 tag_pages_for_writeback(mapping, index, end);
2131 while (!ret && wbc->nr_to_write > 0) {
2134 * we insert one extent at a time. So we need
2135 * credit needed for single extent allocation.
2136 * journalled mode is currently not supported
2139 BUG_ON(ext4_should_journal_data(inode));
2140 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2142 /* start a new transaction*/
2143 handle = ext4_journal_start(inode, needed_blocks);
2144 if (IS_ERR(handle)) {
2145 ret = PTR_ERR(handle);
2146 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2147 "%ld pages, ino %lu; err %d", __func__,
2148 wbc->nr_to_write, inode->i_ino, ret);
2149 goto out_writepages;
2153 * Now call write_cache_pages_da() to find the next
2154 * contiguous region of logical blocks that need
2155 * blocks to be allocated by ext4 and submit them.
2157 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
2159 * If we have a contiguous extent of pages and we
2160 * haven't done the I/O yet, map the blocks and submit
2163 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2164 mpage_da_map_and_submit(&mpd);
2165 ret = MPAGE_DA_EXTENT_TAIL;
2167 trace_ext4_da_write_pages(inode, &mpd);
2168 wbc->nr_to_write -= mpd.pages_written;
2170 ext4_journal_stop(handle);
2172 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2173 /* commit the transaction which would
2174 * free blocks released in the transaction
2177 jbd2_journal_force_commit_nested(sbi->s_journal);
2179 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2181 * got one extent now try with
2184 pages_written += mpd.pages_written;
2187 } else if (wbc->nr_to_write)
2189 * There is no more writeout needed
2190 * or we requested for a noblocking writeout
2191 * and we found the device congested
2195 if (!io_done && !cycled) {
2198 wbc->range_start = index << PAGE_CACHE_SHIFT;
2199 wbc->range_end = mapping->writeback_index - 1;
2204 wbc->range_cyclic = range_cyclic;
2205 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2207 * set the writeback_index so that range_cyclic
2208 * mode will write it back later
2210 mapping->writeback_index = done_index;
2213 wbc->nr_to_write -= nr_to_writebump;
2214 wbc->range_start = range_start;
2215 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2219 #define FALL_BACK_TO_NONDELALLOC 1
2220 static int ext4_nonda_switch(struct super_block *sb)
2222 s64 free_blocks, dirty_blocks;
2223 struct ext4_sb_info *sbi = EXT4_SB(sb);
2226 * switch to non delalloc mode if we are running low
2227 * on free block. The free block accounting via percpu
2228 * counters can get slightly wrong with percpu_counter_batch getting
2229 * accumulated on each CPU without updating global counters
2230 * Delalloc need an accurate free block accounting. So switch
2231 * to non delalloc when we are near to error range.
2233 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2234 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2235 if (2 * free_blocks < 3 * dirty_blocks ||
2236 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2238 * free block count is less than 150% of dirty blocks
2239 * or free blocks is less than watermark
2244 * Even if we don't switch but are nearing capacity,
2245 * start pushing delalloc when 1/2 of free blocks are dirty.
2247 if (free_blocks < 2 * dirty_blocks)
2248 writeback_inodes_sb_if_idle(sb);
2253 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2254 loff_t pos, unsigned len, unsigned flags,
2255 struct page **pagep, void **fsdata)
2257 int ret, retries = 0;
2260 struct inode *inode = mapping->host;
2263 index = pos >> PAGE_CACHE_SHIFT;
2265 if (ext4_nonda_switch(inode->i_sb)) {
2266 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2267 return ext4_write_begin(file, mapping, pos,
2268 len, flags, pagep, fsdata);
2270 *fsdata = (void *)0;
2271 trace_ext4_da_write_begin(inode, pos, len, flags);
2274 * With delayed allocation, we don't log the i_disksize update
2275 * if there is delayed block allocation. But we still need
2276 * to journalling the i_disksize update if writes to the end
2277 * of file which has an already mapped buffer.
2279 handle = ext4_journal_start(inode, 1);
2280 if (IS_ERR(handle)) {
2281 ret = PTR_ERR(handle);
2284 /* We cannot recurse into the filesystem as the transaction is already
2286 flags |= AOP_FLAG_NOFS;
2288 page = grab_cache_page_write_begin(mapping, index, flags);
2290 ext4_journal_stop(handle);
2296 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2299 ext4_journal_stop(handle);
2300 page_cache_release(page);
2302 * block_write_begin may have instantiated a few blocks
2303 * outside i_size. Trim these off again. Don't need
2304 * i_size_read because we hold i_mutex.
2306 if (pos + len > inode->i_size)
2307 ext4_truncate_failed_write(inode);
2310 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2317 * Check if we should update i_disksize
2318 * when write to the end of file but not require block allocation
2320 static int ext4_da_should_update_i_disksize(struct page *page,
2321 unsigned long offset)
2323 struct buffer_head *bh;
2324 struct inode *inode = page->mapping->host;
2328 bh = page_buffers(page);
2329 idx = offset >> inode->i_blkbits;
2331 for (i = 0; i < idx; i++)
2332 bh = bh->b_this_page;
2334 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2339 static int ext4_da_write_end(struct file *file,
2340 struct address_space *mapping,
2341 loff_t pos, unsigned len, unsigned copied,
2342 struct page *page, void *fsdata)
2344 struct inode *inode = mapping->host;
2346 handle_t *handle = ext4_journal_current_handle();
2348 unsigned long start, end;
2349 int write_mode = (int)(unsigned long)fsdata;
2351 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2352 if (ext4_should_order_data(inode)) {
2353 return ext4_ordered_write_end(file, mapping, pos,
2354 len, copied, page, fsdata);
2355 } else if (ext4_should_writeback_data(inode)) {
2356 return ext4_writeback_write_end(file, mapping, pos,
2357 len, copied, page, fsdata);
2363 trace_ext4_da_write_end(inode, pos, len, copied);
2364 start = pos & (PAGE_CACHE_SIZE - 1);
2365 end = start + copied - 1;
2368 * generic_write_end() will run mark_inode_dirty() if i_size
2369 * changes. So let's piggyback the i_disksize mark_inode_dirty
2373 new_i_size = pos + copied;
2374 if (new_i_size > EXT4_I(inode)->i_disksize) {
2375 if (ext4_da_should_update_i_disksize(page, end)) {
2376 down_write(&EXT4_I(inode)->i_data_sem);
2377 if (new_i_size > EXT4_I(inode)->i_disksize) {
2379 * Updating i_disksize when extending file
2380 * without needing block allocation
2382 if (ext4_should_order_data(inode))
2383 ret = ext4_jbd2_file_inode(handle,
2386 EXT4_I(inode)->i_disksize = new_i_size;
2388 up_write(&EXT4_I(inode)->i_data_sem);
2389 /* We need to mark inode dirty even if
2390 * new_i_size is less that inode->i_size
2391 * bu greater than i_disksize.(hint delalloc)
2393 ext4_mark_inode_dirty(handle, inode);
2396 ret2 = generic_write_end(file, mapping, pos, len, copied,
2401 ret2 = ext4_journal_stop(handle);
2405 return ret ? ret : copied;
2408 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2411 * Drop reserved blocks
2413 BUG_ON(!PageLocked(page));
2414 if (!page_has_buffers(page))
2417 ext4_da_page_release_reservation(page, offset);
2420 ext4_invalidatepage(page, offset);
2426 * Force all delayed allocation blocks to be allocated for a given inode.
2428 int ext4_alloc_da_blocks(struct inode *inode)
2430 trace_ext4_alloc_da_blocks(inode);
2432 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2433 !EXT4_I(inode)->i_reserved_meta_blocks)
2437 * We do something simple for now. The filemap_flush() will
2438 * also start triggering a write of the data blocks, which is
2439 * not strictly speaking necessary (and for users of
2440 * laptop_mode, not even desirable). However, to do otherwise
2441 * would require replicating code paths in:
2443 * ext4_da_writepages() ->
2444 * write_cache_pages() ---> (via passed in callback function)
2445 * __mpage_da_writepage() -->
2446 * mpage_add_bh_to_extent()
2447 * mpage_da_map_blocks()
2449 * The problem is that write_cache_pages(), located in
2450 * mm/page-writeback.c, marks pages clean in preparation for
2451 * doing I/O, which is not desirable if we're not planning on
2454 * We could call write_cache_pages(), and then redirty all of
2455 * the pages by calling redirty_page_for_writepage() but that
2456 * would be ugly in the extreme. So instead we would need to
2457 * replicate parts of the code in the above functions,
2458 * simplifying them because we wouldn't actually intend to
2459 * write out the pages, but rather only collect contiguous
2460 * logical block extents, call the multi-block allocator, and
2461 * then update the buffer heads with the block allocations.
2463 * For now, though, we'll cheat by calling filemap_flush(),
2464 * which will map the blocks, and start the I/O, but not
2465 * actually wait for the I/O to complete.
2467 return filemap_flush(inode->i_mapping);
2471 * bmap() is special. It gets used by applications such as lilo and by
2472 * the swapper to find the on-disk block of a specific piece of data.
2474 * Naturally, this is dangerous if the block concerned is still in the
2475 * journal. If somebody makes a swapfile on an ext4 data-journaling
2476 * filesystem and enables swap, then they may get a nasty shock when the
2477 * data getting swapped to that swapfile suddenly gets overwritten by
2478 * the original zero's written out previously to the journal and
2479 * awaiting writeback in the kernel's buffer cache.
2481 * So, if we see any bmap calls here on a modified, data-journaled file,
2482 * take extra steps to flush any blocks which might be in the cache.
2484 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2486 struct inode *inode = mapping->host;
2490 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2491 test_opt(inode->i_sb, DELALLOC)) {
2493 * With delalloc we want to sync the file
2494 * so that we can make sure we allocate
2497 filemap_write_and_wait(mapping);
2500 if (EXT4_JOURNAL(inode) &&
2501 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2503 * This is a REALLY heavyweight approach, but the use of
2504 * bmap on dirty files is expected to be extremely rare:
2505 * only if we run lilo or swapon on a freshly made file
2506 * do we expect this to happen.
2508 * (bmap requires CAP_SYS_RAWIO so this does not
2509 * represent an unprivileged user DOS attack --- we'd be
2510 * in trouble if mortal users could trigger this path at
2513 * NB. EXT4_STATE_JDATA is not set on files other than
2514 * regular files. If somebody wants to bmap a directory
2515 * or symlink and gets confused because the buffer
2516 * hasn't yet been flushed to disk, they deserve
2517 * everything they get.
2520 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2521 journal = EXT4_JOURNAL(inode);
2522 jbd2_journal_lock_updates(journal);
2523 err = jbd2_journal_flush(journal);
2524 jbd2_journal_unlock_updates(journal);
2530 return generic_block_bmap(mapping, block, ext4_get_block);
2533 static int ext4_readpage(struct file *file, struct page *page)
2535 trace_ext4_readpage(page);
2536 return mpage_readpage(page, ext4_get_block);
2540 ext4_readpages(struct file *file, struct address_space *mapping,
2541 struct list_head *pages, unsigned nr_pages)
2543 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2546 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
2548 struct buffer_head *head, *bh;
2549 unsigned int curr_off = 0;
2551 if (!page_has_buffers(page))
2553 head = bh = page_buffers(page);
2555 if (offset <= curr_off && test_clear_buffer_uninit(bh)
2557 ext4_free_io_end(bh->b_private);
2558 bh->b_private = NULL;
2559 bh->b_end_io = NULL;
2561 curr_off = curr_off + bh->b_size;
2562 bh = bh->b_this_page;
2563 } while (bh != head);
2566 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2568 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2570 trace_ext4_invalidatepage(page, offset);
2573 * free any io_end structure allocated for buffers to be discarded
2575 if (ext4_should_dioread_nolock(page->mapping->host))
2576 ext4_invalidatepage_free_endio(page, offset);
2578 * If it's a full truncate we just forget about the pending dirtying
2581 ClearPageChecked(page);
2584 jbd2_journal_invalidatepage(journal, page, offset);
2586 block_invalidatepage(page, offset);
2589 static int ext4_releasepage(struct page *page, gfp_t wait)
2591 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2593 trace_ext4_releasepage(page);
2595 WARN_ON(PageChecked(page));
2596 if (!page_has_buffers(page))
2599 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2601 return try_to_free_buffers(page);
2605 * ext4_get_block used when preparing for a DIO write or buffer write.
2606 * We allocate an uinitialized extent if blocks haven't been allocated.
2607 * The extent will be converted to initialized after the IO is complete.
2609 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
2610 struct buffer_head *bh_result, int create)
2612 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2613 inode->i_ino, create);
2614 return _ext4_get_block(inode, iblock, bh_result,
2615 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2618 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2619 ssize_t size, void *private, int ret,
2622 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2623 ext4_io_end_t *io_end = iocb->private;
2624 struct workqueue_struct *wq;
2625 unsigned long flags;
2626 struct ext4_inode_info *ei;
2628 /* if not async direct IO or dio with 0 bytes write, just return */
2629 if (!io_end || !size)
2632 ext_debug("ext4_end_io_dio(): io_end 0x%p"
2633 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
2634 iocb->private, io_end->inode->i_ino, iocb, offset,
2637 /* if not aio dio with unwritten extents, just free io and return */
2638 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2639 ext4_free_io_end(io_end);
2640 iocb->private = NULL;
2643 aio_complete(iocb, ret, 0);
2644 inode_dio_done(inode);
2648 io_end->offset = offset;
2649 io_end->size = size;
2651 io_end->iocb = iocb;
2652 io_end->result = ret;
2654 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
2656 /* Add the io_end to per-inode completed aio dio list*/
2657 ei = EXT4_I(io_end->inode);
2658 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
2659 list_add_tail(&io_end->list, &ei->i_completed_io_list);
2660 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
2662 /* queue the work to convert unwritten extents to written */
2663 queue_work(wq, &io_end->work);
2664 iocb->private = NULL;
2666 /* XXX: probably should move into the real I/O completion handler */
2667 inode_dio_done(inode);
2670 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
2672 ext4_io_end_t *io_end = bh->b_private;
2673 struct workqueue_struct *wq;
2674 struct inode *inode;
2675 unsigned long flags;
2677 if (!test_clear_buffer_uninit(bh) || !io_end)
2680 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2681 printk("sb umounted, discard end_io request for inode %lu\n",
2682 io_end->inode->i_ino);
2683 ext4_free_io_end(io_end);
2688 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2689 * but being more careful is always safe for the future change.
2691 inode = io_end->inode;
2692 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2693 io_end->flag |= EXT4_IO_END_UNWRITTEN;
2694 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
2697 /* Add the io_end to per-inode completed io list*/
2698 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
2699 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
2700 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
2702 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
2703 /* queue the work to convert unwritten extents to written */
2704 queue_work(wq, &io_end->work);
2706 bh->b_private = NULL;
2707 bh->b_end_io = NULL;
2708 clear_buffer_uninit(bh);
2709 end_buffer_async_write(bh, uptodate);
2712 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2714 ext4_io_end_t *io_end;
2715 struct page *page = bh->b_page;
2716 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2717 size_t size = bh->b_size;
2720 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2722 pr_warn_ratelimited("%s: allocation fail\n", __func__);
2726 io_end->offset = offset;
2727 io_end->size = size;
2729 * We need to hold a reference to the page to make sure it
2730 * doesn't get evicted before ext4_end_io_work() has a chance
2731 * to convert the extent from written to unwritten.
2733 io_end->page = page;
2734 get_page(io_end->page);
2736 bh->b_private = io_end;
2737 bh->b_end_io = ext4_end_io_buffer_write;
2742 * For ext4 extent files, ext4 will do direct-io write to holes,
2743 * preallocated extents, and those write extend the file, no need to
2744 * fall back to buffered IO.
2746 * For holes, we fallocate those blocks, mark them as uninitialized
2747 * If those blocks were preallocated, we mark sure they are splited, but
2748 * still keep the range to write as uninitialized.
2750 * The unwrritten extents will be converted to written when DIO is completed.
2751 * For async direct IO, since the IO may still pending when return, we
2752 * set up an end_io call back function, which will do the conversion
2753 * when async direct IO completed.
2755 * If the O_DIRECT write will extend the file then add this inode to the
2756 * orphan list. So recovery will truncate it back to the original size
2757 * if the machine crashes during the write.
2760 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2761 const struct iovec *iov, loff_t offset,
2762 unsigned long nr_segs)
2764 struct file *file = iocb->ki_filp;
2765 struct inode *inode = file->f_mapping->host;
2767 size_t count = iov_length(iov, nr_segs);
2769 loff_t final_size = offset + count;
2770 if (rw == WRITE && final_size <= inode->i_size) {
2772 * We could direct write to holes and fallocate.
2774 * Allocated blocks to fill the hole are marked as uninitialized
2775 * to prevent parallel buffered read to expose the stale data
2776 * before DIO complete the data IO.
2778 * As to previously fallocated extents, ext4 get_block
2779 * will just simply mark the buffer mapped but still
2780 * keep the extents uninitialized.
2782 * for non AIO case, we will convert those unwritten extents
2783 * to written after return back from blockdev_direct_IO.
2785 * for async DIO, the conversion needs to be defered when
2786 * the IO is completed. The ext4 end_io callback function
2787 * will be called to take care of the conversion work.
2788 * Here for async case, we allocate an io_end structure to
2791 iocb->private = NULL;
2792 EXT4_I(inode)->cur_aio_dio = NULL;
2793 if (!is_sync_kiocb(iocb)) {
2794 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
2798 * we save the io structure for current async
2799 * direct IO, so that later ext4_map_blocks()
2800 * could flag the io structure whether there
2801 * is a unwritten extents needs to be converted
2802 * when IO is completed.
2804 EXT4_I(inode)->cur_aio_dio = iocb->private;
2807 ret = __blockdev_direct_IO(rw, iocb, inode,
2808 inode->i_sb->s_bdev, iov,
2810 ext4_get_block_write,
2813 DIO_LOCKING | DIO_SKIP_HOLES);
2815 EXT4_I(inode)->cur_aio_dio = NULL;
2817 * The io_end structure takes a reference to the inode,
2818 * that structure needs to be destroyed and the
2819 * reference to the inode need to be dropped, when IO is
2820 * complete, even with 0 byte write, or failed.
2822 * In the successful AIO DIO case, the io_end structure will be
2823 * desctroyed and the reference to the inode will be dropped
2824 * after the end_io call back function is called.
2826 * In the case there is 0 byte write, or error case, since
2827 * VFS direct IO won't invoke the end_io call back function,
2828 * we need to free the end_io structure here.
2830 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
2831 ext4_free_io_end(iocb->private);
2832 iocb->private = NULL;
2833 } else if (ret > 0 && ext4_test_inode_state(inode,
2834 EXT4_STATE_DIO_UNWRITTEN)) {
2837 * for non AIO case, since the IO is already
2838 * completed, we could do the conversion right here
2840 err = ext4_convert_unwritten_extents(inode,
2844 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
2849 /* for write the the end of file case, we fall back to old way */
2850 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
2853 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
2854 const struct iovec *iov, loff_t offset,
2855 unsigned long nr_segs)
2857 struct file *file = iocb->ki_filp;
2858 struct inode *inode = file->f_mapping->host;
2861 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
2862 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
2863 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
2865 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
2866 trace_ext4_direct_IO_exit(inode, offset,
2867 iov_length(iov, nr_segs), rw, ret);
2872 * Pages can be marked dirty completely asynchronously from ext4's journalling
2873 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
2874 * much here because ->set_page_dirty is called under VFS locks. The page is
2875 * not necessarily locked.
2877 * We cannot just dirty the page and leave attached buffers clean, because the
2878 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
2879 * or jbddirty because all the journalling code will explode.
2881 * So what we do is to mark the page "pending dirty" and next time writepage
2882 * is called, propagate that into the buffers appropriately.
2884 static int ext4_journalled_set_page_dirty(struct page *page)
2886 SetPageChecked(page);
2887 return __set_page_dirty_nobuffers(page);
2890 static const struct address_space_operations ext4_ordered_aops = {
2891 .readpage = ext4_readpage,
2892 .readpages = ext4_readpages,
2893 .writepage = ext4_writepage,
2894 .write_begin = ext4_write_begin,
2895 .write_end = ext4_ordered_write_end,
2897 .invalidatepage = ext4_invalidatepage,
2898 .releasepage = ext4_releasepage,
2899 .direct_IO = ext4_direct_IO,
2900 .migratepage = buffer_migrate_page,
2901 .is_partially_uptodate = block_is_partially_uptodate,
2902 .error_remove_page = generic_error_remove_page,
2905 static const struct address_space_operations ext4_writeback_aops = {
2906 .readpage = ext4_readpage,
2907 .readpages = ext4_readpages,
2908 .writepage = ext4_writepage,
2909 .write_begin = ext4_write_begin,
2910 .write_end = ext4_writeback_write_end,
2912 .invalidatepage = ext4_invalidatepage,
2913 .releasepage = ext4_releasepage,
2914 .direct_IO = ext4_direct_IO,
2915 .migratepage = buffer_migrate_page,
2916 .is_partially_uptodate = block_is_partially_uptodate,
2917 .error_remove_page = generic_error_remove_page,
2920 static const struct address_space_operations ext4_journalled_aops = {
2921 .readpage = ext4_readpage,
2922 .readpages = ext4_readpages,
2923 .writepage = ext4_writepage,
2924 .write_begin = ext4_write_begin,
2925 .write_end = ext4_journalled_write_end,
2926 .set_page_dirty = ext4_journalled_set_page_dirty,
2928 .invalidatepage = ext4_invalidatepage,
2929 .releasepage = ext4_releasepage,
2930 .is_partially_uptodate = block_is_partially_uptodate,
2931 .error_remove_page = generic_error_remove_page,
2934 static const struct address_space_operations ext4_da_aops = {
2935 .readpage = ext4_readpage,
2936 .readpages = ext4_readpages,
2937 .writepage = ext4_writepage,
2938 .writepages = ext4_da_writepages,
2939 .write_begin = ext4_da_write_begin,
2940 .write_end = ext4_da_write_end,
2942 .invalidatepage = ext4_da_invalidatepage,
2943 .releasepage = ext4_releasepage,
2944 .direct_IO = ext4_direct_IO,
2945 .migratepage = buffer_migrate_page,
2946 .is_partially_uptodate = block_is_partially_uptodate,
2947 .error_remove_page = generic_error_remove_page,
2950 void ext4_set_aops(struct inode *inode)
2952 if (ext4_should_order_data(inode) &&
2953 test_opt(inode->i_sb, DELALLOC))
2954 inode->i_mapping->a_ops = &ext4_da_aops;
2955 else if (ext4_should_order_data(inode))
2956 inode->i_mapping->a_ops = &ext4_ordered_aops;
2957 else if (ext4_should_writeback_data(inode) &&
2958 test_opt(inode->i_sb, DELALLOC))
2959 inode->i_mapping->a_ops = &ext4_da_aops;
2960 else if (ext4_should_writeback_data(inode))
2961 inode->i_mapping->a_ops = &ext4_writeback_aops;
2963 inode->i_mapping->a_ops = &ext4_journalled_aops;
2967 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
2968 * up to the end of the block which corresponds to `from'.
2969 * This required during truncate. We need to physically zero the tail end
2970 * of that block so it doesn't yield old data if the file is later grown.
2972 int ext4_block_truncate_page(handle_t *handle,
2973 struct address_space *mapping, loff_t from)
2975 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2978 struct inode *inode = mapping->host;
2980 blocksize = inode->i_sb->s_blocksize;
2981 length = blocksize - (offset & (blocksize - 1));
2983 return ext4_block_zero_page_range(handle, mapping, from, length);
2987 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
2988 * starting from file offset 'from'. The range to be zero'd must
2989 * be contained with in one block. If the specified range exceeds
2990 * the end of the block it will be shortened to end of the block
2991 * that cooresponds to 'from'
2993 int ext4_block_zero_page_range(handle_t *handle,
2994 struct address_space *mapping, loff_t from, loff_t length)
2996 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
2997 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2998 unsigned blocksize, max, pos;
3000 struct inode *inode = mapping->host;
3001 struct buffer_head *bh;
3005 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3006 mapping_gfp_mask(mapping) & ~__GFP_FS);
3010 blocksize = inode->i_sb->s_blocksize;
3011 max = blocksize - (offset & (blocksize - 1));
3014 * correct length if it does not fall between
3015 * 'from' and the end of the block
3017 if (length > max || length < 0)
3020 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3022 if (!page_has_buffers(page))
3023 create_empty_buffers(page, blocksize, 0);
3025 /* Find the buffer that contains "offset" */
3026 bh = page_buffers(page);
3028 while (offset >= pos) {
3029 bh = bh->b_this_page;
3035 if (buffer_freed(bh)) {
3036 BUFFER_TRACE(bh, "freed: skip");
3040 if (!buffer_mapped(bh)) {
3041 BUFFER_TRACE(bh, "unmapped");
3042 ext4_get_block(inode, iblock, bh, 0);
3043 /* unmapped? It's a hole - nothing to do */
3044 if (!buffer_mapped(bh)) {
3045 BUFFER_TRACE(bh, "still unmapped");
3050 /* Ok, it's mapped. Make sure it's up-to-date */
3051 if (PageUptodate(page))
3052 set_buffer_uptodate(bh);
3054 if (!buffer_uptodate(bh)) {
3056 ll_rw_block(READ, 1, &bh);
3058 /* Uhhuh. Read error. Complain and punt. */
3059 if (!buffer_uptodate(bh))
3063 if (ext4_should_journal_data(inode)) {
3064 BUFFER_TRACE(bh, "get write access");
3065 err = ext4_journal_get_write_access(handle, bh);
3070 zero_user(page, offset, length);
3072 BUFFER_TRACE(bh, "zeroed end of block");
3075 if (ext4_should_journal_data(inode)) {
3076 err = ext4_handle_dirty_metadata(handle, inode, bh);
3078 if (ext4_should_order_data(inode) && EXT4_I(inode)->jinode)
3079 err = ext4_jbd2_file_inode(handle, inode);
3080 mark_buffer_dirty(bh);
3085 page_cache_release(page);
3089 int ext4_can_truncate(struct inode *inode)
3091 if (S_ISREG(inode->i_mode))
3093 if (S_ISDIR(inode->i_mode))
3095 if (S_ISLNK(inode->i_mode))
3096 return !ext4_inode_is_fast_symlink(inode);
3101 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3102 * associated with the given offset and length
3104 * @inode: File inode
3105 * @offset: The offset where the hole will begin
3106 * @len: The length of the hole
3108 * Returns: 0 on sucess or negative on failure
3111 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3113 struct inode *inode = file->f_path.dentry->d_inode;
3114 if (!S_ISREG(inode->i_mode))
3117 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3118 /* TODO: Add support for non extent hole punching */
3122 return ext4_ext_punch_hole(file, offset, length);
3128 * We block out ext4_get_block() block instantiations across the entire
3129 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3130 * simultaneously on behalf of the same inode.
3132 * As we work through the truncate and commmit bits of it to the journal there
3133 * is one core, guiding principle: the file's tree must always be consistent on
3134 * disk. We must be able to restart the truncate after a crash.
3136 * The file's tree may be transiently inconsistent in memory (although it
3137 * probably isn't), but whenever we close off and commit a journal transaction,
3138 * the contents of (the filesystem + the journal) must be consistent and
3139 * restartable. It's pretty simple, really: bottom up, right to left (although
3140 * left-to-right works OK too).
3142 * Note that at recovery time, journal replay occurs *before* the restart of
3143 * truncate against the orphan inode list.
3145 * The committed inode has the new, desired i_size (which is the same as
3146 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3147 * that this inode's truncate did not complete and it will again call
3148 * ext4_truncate() to have another go. So there will be instantiated blocks
3149 * to the right of the truncation point in a crashed ext4 filesystem. But
3150 * that's fine - as long as they are linked from the inode, the post-crash
3151 * ext4_truncate() run will find them and release them.
3153 void ext4_truncate(struct inode *inode)
3155 trace_ext4_truncate_enter(inode);
3157 if (!ext4_can_truncate(inode))
3160 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3162 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3163 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3165 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3166 ext4_ext_truncate(inode);
3168 ext4_ind_truncate(inode);
3170 trace_ext4_truncate_exit(inode);
3174 * ext4_get_inode_loc returns with an extra refcount against the inode's
3175 * underlying buffer_head on success. If 'in_mem' is true, we have all
3176 * data in memory that is needed to recreate the on-disk version of this
3179 static int __ext4_get_inode_loc(struct inode *inode,
3180 struct ext4_iloc *iloc, int in_mem)
3182 struct ext4_group_desc *gdp;
3183 struct buffer_head *bh;
3184 struct super_block *sb = inode->i_sb;
3186 int inodes_per_block, inode_offset;
3189 if (!ext4_valid_inum(sb, inode->i_ino))
3192 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3193 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3198 * Figure out the offset within the block group inode table
3200 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3201 inode_offset = ((inode->i_ino - 1) %
3202 EXT4_INODES_PER_GROUP(sb));
3203 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3204 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3206 bh = sb_getblk(sb, block);
3208 EXT4_ERROR_INODE_BLOCK(inode, block,
3209 "unable to read itable block");
3212 if (!buffer_uptodate(bh)) {
3216 * If the buffer has the write error flag, we have failed
3217 * to write out another inode in the same block. In this
3218 * case, we don't have to read the block because we may
3219 * read the old inode data successfully.
3221 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3222 set_buffer_uptodate(bh);
3224 if (buffer_uptodate(bh)) {
3225 /* someone brought it uptodate while we waited */
3231 * If we have all information of the inode in memory and this
3232 * is the only valid inode in the block, we need not read the
3236 struct buffer_head *bitmap_bh;
3239 start = inode_offset & ~(inodes_per_block - 1);
3241 /* Is the inode bitmap in cache? */
3242 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3247 * If the inode bitmap isn't in cache then the
3248 * optimisation may end up performing two reads instead
3249 * of one, so skip it.
3251 if (!buffer_uptodate(bitmap_bh)) {
3255 for (i = start; i < start + inodes_per_block; i++) {
3256 if (i == inode_offset)
3258 if (ext4_test_bit(i, bitmap_bh->b_data))
3262 if (i == start + inodes_per_block) {
3263 /* all other inodes are free, so skip I/O */
3264 memset(bh->b_data, 0, bh->b_size);
3265 set_buffer_uptodate(bh);
3273 * If we need to do any I/O, try to pre-readahead extra
3274 * blocks from the inode table.
3276 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3277 ext4_fsblk_t b, end, table;
3280 table = ext4_inode_table(sb, gdp);
3281 /* s_inode_readahead_blks is always a power of 2 */
3282 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3285 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3286 num = EXT4_INODES_PER_GROUP(sb);
3287 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3288 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3289 num -= ext4_itable_unused_count(sb, gdp);
3290 table += num / inodes_per_block;
3294 sb_breadahead(sb, b++);
3298 * There are other valid inodes in the buffer, this inode
3299 * has in-inode xattrs, or we don't have this inode in memory.
3300 * Read the block from disk.
3302 trace_ext4_load_inode(inode);
3304 bh->b_end_io = end_buffer_read_sync;
3305 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3307 if (!buffer_uptodate(bh)) {
3308 EXT4_ERROR_INODE_BLOCK(inode, block,
3309 "unable to read itable block");
3319 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3321 /* We have all inode data except xattrs in memory here. */
3322 return __ext4_get_inode_loc(inode, iloc,
3323 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3326 void ext4_set_inode_flags(struct inode *inode)
3328 unsigned int flags = EXT4_I(inode)->i_flags;
3330 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3331 if (flags & EXT4_SYNC_FL)
3332 inode->i_flags |= S_SYNC;
3333 if (flags & EXT4_APPEND_FL)
3334 inode->i_flags |= S_APPEND;
3335 if (flags & EXT4_IMMUTABLE_FL)
3336 inode->i_flags |= S_IMMUTABLE;
3337 if (flags & EXT4_NOATIME_FL)
3338 inode->i_flags |= S_NOATIME;
3339 if (flags & EXT4_DIRSYNC_FL)
3340 inode->i_flags |= S_DIRSYNC;
3343 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3344 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3346 unsigned int vfs_fl;
3347 unsigned long old_fl, new_fl;
3350 vfs_fl = ei->vfs_inode.i_flags;
3351 old_fl = ei->i_flags;
3352 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3353 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3355 if (vfs_fl & S_SYNC)
3356 new_fl |= EXT4_SYNC_FL;
3357 if (vfs_fl & S_APPEND)
3358 new_fl |= EXT4_APPEND_FL;
3359 if (vfs_fl & S_IMMUTABLE)
3360 new_fl |= EXT4_IMMUTABLE_FL;
3361 if (vfs_fl & S_NOATIME)
3362 new_fl |= EXT4_NOATIME_FL;
3363 if (vfs_fl & S_DIRSYNC)
3364 new_fl |= EXT4_DIRSYNC_FL;
3365 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3368 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3369 struct ext4_inode_info *ei)
3372 struct inode *inode = &(ei->vfs_inode);
3373 struct super_block *sb = inode->i_sb;
3375 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3376 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3377 /* we are using combined 48 bit field */
3378 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3379 le32_to_cpu(raw_inode->i_blocks_lo);
3380 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3381 /* i_blocks represent file system block size */
3382 return i_blocks << (inode->i_blkbits - 9);
3387 return le32_to_cpu(raw_inode->i_blocks_lo);
3391 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3393 struct ext4_iloc iloc;
3394 struct ext4_inode *raw_inode;
3395 struct ext4_inode_info *ei;
3396 struct inode *inode;
3397 journal_t *journal = EXT4_SB(sb)->s_journal;
3401 inode = iget_locked(sb, ino);
3403 return ERR_PTR(-ENOMEM);
3404 if (!(inode->i_state & I_NEW))
3410 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3413 raw_inode = ext4_raw_inode(&iloc);
3414 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3415 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3416 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3417 if (!(test_opt(inode->i_sb, NO_UID32))) {
3418 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3419 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3421 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
3423 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3424 ei->i_dir_start_lookup = 0;
3425 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3426 /* We now have enough fields to check if the inode was active or not.
3427 * This is needed because nfsd might try to access dead inodes
3428 * the test is that same one that e2fsck uses
3429 * NeilBrown 1999oct15
3431 if (inode->i_nlink == 0) {
3432 if (inode->i_mode == 0 ||
3433 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3434 /* this inode is deleted */
3438 /* The only unlinked inodes we let through here have
3439 * valid i_mode and are being read by the orphan
3440 * recovery code: that's fine, we're about to complete
3441 * the process of deleting those. */
3443 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3444 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3445 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3446 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3448 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3449 inode->i_size = ext4_isize(raw_inode);
3450 ei->i_disksize = inode->i_size;
3452 ei->i_reserved_quota = 0;
3454 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3455 ei->i_block_group = iloc.block_group;
3456 ei->i_last_alloc_group = ~0;
3458 * NOTE! The in-memory inode i_data array is in little-endian order
3459 * even on big-endian machines: we do NOT byteswap the block numbers!
3461 for (block = 0; block < EXT4_N_BLOCKS; block++)
3462 ei->i_data[block] = raw_inode->i_block[block];
3463 INIT_LIST_HEAD(&ei->i_orphan);
3466 * Set transaction id's of transactions that have to be committed
3467 * to finish f[data]sync. We set them to currently running transaction
3468 * as we cannot be sure that the inode or some of its metadata isn't
3469 * part of the transaction - the inode could have been reclaimed and
3470 * now it is reread from disk.
3473 transaction_t *transaction;
3476 read_lock(&journal->j_state_lock);
3477 if (journal->j_running_transaction)
3478 transaction = journal->j_running_transaction;
3480 transaction = journal->j_committing_transaction;
3482 tid = transaction->t_tid;
3484 tid = journal->j_commit_sequence;
3485 read_unlock(&journal->j_state_lock);
3486 ei->i_sync_tid = tid;
3487 ei->i_datasync_tid = tid;
3490 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3491 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3492 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3493 EXT4_INODE_SIZE(inode->i_sb)) {
3497 if (ei->i_extra_isize == 0) {
3498 /* The extra space is currently unused. Use it. */
3499 ei->i_extra_isize = sizeof(struct ext4_inode) -
3500 EXT4_GOOD_OLD_INODE_SIZE;
3502 __le32 *magic = (void *)raw_inode +
3503 EXT4_GOOD_OLD_INODE_SIZE +
3505 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3506 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3509 ei->i_extra_isize = 0;
3511 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3512 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3513 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3514 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3516 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3517 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3518 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3520 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3524 if (ei->i_file_acl &&
3525 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3526 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
3530 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3531 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3532 (S_ISLNK(inode->i_mode) &&
3533 !ext4_inode_is_fast_symlink(inode)))
3534 /* Validate extent which is part of inode */
3535 ret = ext4_ext_check_inode(inode);
3536 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3537 (S_ISLNK(inode->i_mode) &&
3538 !ext4_inode_is_fast_symlink(inode))) {
3539 /* Validate block references which are part of inode */
3540 ret = ext4_ind_check_inode(inode);
3545 if (S_ISREG(inode->i_mode)) {
3546 inode->i_op = &ext4_file_inode_operations;
3547 inode->i_fop = &ext4_file_operations;
3548 ext4_set_aops(inode);
3549 } else if (S_ISDIR(inode->i_mode)) {
3550 inode->i_op = &ext4_dir_inode_operations;
3551 inode->i_fop = &ext4_dir_operations;
3552 } else if (S_ISLNK(inode->i_mode)) {
3553 if (ext4_inode_is_fast_symlink(inode)) {
3554 inode->i_op = &ext4_fast_symlink_inode_operations;
3555 nd_terminate_link(ei->i_data, inode->i_size,
3556 sizeof(ei->i_data) - 1);
3558 inode->i_op = &ext4_symlink_inode_operations;
3559 ext4_set_aops(inode);
3561 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3562 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3563 inode->i_op = &ext4_special_inode_operations;
3564 if (raw_inode->i_block[0])
3565 init_special_inode(inode, inode->i_mode,
3566 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3568 init_special_inode(inode, inode->i_mode,
3569 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3572 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
3576 ext4_set_inode_flags(inode);
3577 unlock_new_inode(inode);
3583 return ERR_PTR(ret);
3586 static int ext4_inode_blocks_set(handle_t *handle,
3587 struct ext4_inode *raw_inode,
3588 struct ext4_inode_info *ei)
3590 struct inode *inode = &(ei->vfs_inode);
3591 u64 i_blocks = inode->i_blocks;
3592 struct super_block *sb = inode->i_sb;
3594 if (i_blocks <= ~0U) {
3596 * i_blocks can be represnted in a 32 bit variable
3597 * as multiple of 512 bytes
3599 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3600 raw_inode->i_blocks_high = 0;
3601 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3604 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
3607 if (i_blocks <= 0xffffffffffffULL) {
3609 * i_blocks can be represented in a 48 bit variable
3610 * as multiple of 512 bytes
3612 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3613 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3614 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3616 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3617 /* i_block is stored in file system block size */
3618 i_blocks = i_blocks >> (inode->i_blkbits - 9);
3619 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3620 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3626 * Post the struct inode info into an on-disk inode location in the
3627 * buffer-cache. This gobbles the caller's reference to the
3628 * buffer_head in the inode location struct.
3630 * The caller must have write access to iloc->bh.
3632 static int ext4_do_update_inode(handle_t *handle,
3633 struct inode *inode,
3634 struct ext4_iloc *iloc)
3636 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
3637 struct ext4_inode_info *ei = EXT4_I(inode);
3638 struct buffer_head *bh = iloc->bh;
3639 int err = 0, rc, block;
3641 /* For fields not not tracking in the in-memory inode,
3642 * initialise them to zero for new inodes. */
3643 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
3644 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
3646 ext4_get_inode_flags(ei);
3647 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
3648 if (!(test_opt(inode->i_sb, NO_UID32))) {
3649 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
3650 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
3652 * Fix up interoperability with old kernels. Otherwise, old inodes get
3653 * re-used with the upper 16 bits of the uid/gid intact
3656 raw_inode->i_uid_high =
3657 cpu_to_le16(high_16_bits(inode->i_uid));
3658 raw_inode->i_gid_high =
3659 cpu_to_le16(high_16_bits(inode->i_gid));
3661 raw_inode->i_uid_high = 0;
3662 raw_inode->i_gid_high = 0;
3665 raw_inode->i_uid_low =
3666 cpu_to_le16(fs_high2lowuid(inode->i_uid));
3667 raw_inode->i_gid_low =
3668 cpu_to_le16(fs_high2lowgid(inode->i_gid));
3669 raw_inode->i_uid_high = 0;
3670 raw_inode->i_gid_high = 0;
3672 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
3674 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
3675 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
3676 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
3677 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
3679 if (ext4_inode_blocks_set(handle, raw_inode, ei))
3681 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3682 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
3683 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
3684 cpu_to_le32(EXT4_OS_HURD))
3685 raw_inode->i_file_acl_high =
3686 cpu_to_le16(ei->i_file_acl >> 32);
3687 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
3688 ext4_isize_set(raw_inode, ei->i_disksize);
3689 if (ei->i_disksize > 0x7fffffffULL) {
3690 struct super_block *sb = inode->i_sb;
3691 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
3692 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
3693 EXT4_SB(sb)->s_es->s_rev_level ==
3694 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
3695 /* If this is the first large file
3696 * created, add a flag to the superblock.
3698 err = ext4_journal_get_write_access(handle,
3699 EXT4_SB(sb)->s_sbh);
3702 ext4_update_dynamic_rev(sb);
3703 EXT4_SET_RO_COMPAT_FEATURE(sb,
3704 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
3706 ext4_handle_sync(handle);
3707 err = ext4_handle_dirty_metadata(handle, NULL,
3708 EXT4_SB(sb)->s_sbh);
3711 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3712 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3713 if (old_valid_dev(inode->i_rdev)) {
3714 raw_inode->i_block[0] =
3715 cpu_to_le32(old_encode_dev(inode->i_rdev));
3716 raw_inode->i_block[1] = 0;
3718 raw_inode->i_block[0] = 0;
3719 raw_inode->i_block[1] =
3720 cpu_to_le32(new_encode_dev(inode->i_rdev));
3721 raw_inode->i_block[2] = 0;
3724 for (block = 0; block < EXT4_N_BLOCKS; block++)
3725 raw_inode->i_block[block] = ei->i_data[block];
3727 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
3728 if (ei->i_extra_isize) {
3729 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3730 raw_inode->i_version_hi =
3731 cpu_to_le32(inode->i_version >> 32);
3732 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
3735 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
3736 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
3739 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
3741 ext4_update_inode_fsync_trans(handle, inode, 0);
3744 ext4_std_error(inode->i_sb, err);
3749 * ext4_write_inode()
3751 * We are called from a few places:
3753 * - Within generic_file_write() for O_SYNC files.
3754 * Here, there will be no transaction running. We wait for any running
3755 * trasnaction to commit.
3757 * - Within sys_sync(), kupdate and such.
3758 * We wait on commit, if tol to.
3760 * - Within prune_icache() (PF_MEMALLOC == true)
3761 * Here we simply return. We can't afford to block kswapd on the
3764 * In all cases it is actually safe for us to return without doing anything,
3765 * because the inode has been copied into a raw inode buffer in
3766 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3769 * Note that we are absolutely dependent upon all inode dirtiers doing the
3770 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3771 * which we are interested.
3773 * It would be a bug for them to not do this. The code:
3775 * mark_inode_dirty(inode)
3777 * inode->i_size = expr;
3779 * is in error because a kswapd-driven write_inode() could occur while
3780 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3781 * will no longer be on the superblock's dirty inode list.
3783 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
3787 if (current->flags & PF_MEMALLOC)
3790 if (EXT4_SB(inode->i_sb)->s_journal) {
3791 if (ext4_journal_current_handle()) {
3792 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3797 if (wbc->sync_mode != WB_SYNC_ALL)
3800 err = ext4_force_commit(inode->i_sb);
3802 struct ext4_iloc iloc;
3804 err = __ext4_get_inode_loc(inode, &iloc, 0);
3807 if (wbc->sync_mode == WB_SYNC_ALL)
3808 sync_dirty_buffer(iloc.bh);
3809 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
3810 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
3811 "IO error syncing inode");
3822 * Called from notify_change.
3824 * We want to trap VFS attempts to truncate the file as soon as
3825 * possible. In particular, we want to make sure that when the VFS
3826 * shrinks i_size, we put the inode on the orphan list and modify
3827 * i_disksize immediately, so that during the subsequent flushing of
3828 * dirty pages and freeing of disk blocks, we can guarantee that any
3829 * commit will leave the blocks being flushed in an unused state on
3830 * disk. (On recovery, the inode will get truncated and the blocks will
3831 * be freed, so we have a strong guarantee that no future commit will
3832 * leave these blocks visible to the user.)
3834 * Another thing we have to assure is that if we are in ordered mode
3835 * and inode is still attached to the committing transaction, we must
3836 * we start writeout of all the dirty pages which are being truncated.
3837 * This way we are sure that all the data written in the previous
3838 * transaction are already on disk (truncate waits for pages under
3841 * Called with inode->i_mutex down.
3843 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
3845 struct inode *inode = dentry->d_inode;
3848 const unsigned int ia_valid = attr->ia_valid;
3850 error = inode_change_ok(inode, attr);
3854 if (is_quota_modification(inode, attr))
3855 dquot_initialize(inode);
3856 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3857 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3860 /* (user+group)*(old+new) structure, inode write (sb,
3861 * inode block, ? - but truncate inode update has it) */
3862 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
3863 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
3864 if (IS_ERR(handle)) {
3865 error = PTR_ERR(handle);
3868 error = dquot_transfer(inode, attr);
3870 ext4_journal_stop(handle);
3873 /* Update corresponding info in inode so that everything is in
3874 * one transaction */
3875 if (attr->ia_valid & ATTR_UID)
3876 inode->i_uid = attr->ia_uid;
3877 if (attr->ia_valid & ATTR_GID)
3878 inode->i_gid = attr->ia_gid;
3879 error = ext4_mark_inode_dirty(handle, inode);
3880 ext4_journal_stop(handle);
3883 if (attr->ia_valid & ATTR_SIZE) {
3884 inode_dio_wait(inode);
3886 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3887 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3889 if (attr->ia_size > sbi->s_bitmap_maxbytes)
3894 if (S_ISREG(inode->i_mode) &&
3895 attr->ia_valid & ATTR_SIZE &&
3896 (attr->ia_size < inode->i_size)) {
3899 handle = ext4_journal_start(inode, 3);
3900 if (IS_ERR(handle)) {
3901 error = PTR_ERR(handle);
3904 if (ext4_handle_valid(handle)) {
3905 error = ext4_orphan_add(handle, inode);
3908 EXT4_I(inode)->i_disksize = attr->ia_size;
3909 rc = ext4_mark_inode_dirty(handle, inode);
3912 ext4_journal_stop(handle);
3914 if (ext4_should_order_data(inode)) {
3915 error = ext4_begin_ordered_truncate(inode,
3918 /* Do as much error cleanup as possible */
3919 handle = ext4_journal_start(inode, 3);
3920 if (IS_ERR(handle)) {
3921 ext4_orphan_del(NULL, inode);
3924 ext4_orphan_del(handle, inode);
3926 ext4_journal_stop(handle);
3932 if (attr->ia_valid & ATTR_SIZE) {
3933 if (attr->ia_size != i_size_read(inode)) {
3934 truncate_setsize(inode, attr->ia_size);
3935 ext4_truncate(inode);
3936 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3937 ext4_truncate(inode);
3941 setattr_copy(inode, attr);
3942 mark_inode_dirty(inode);
3946 * If the call to ext4_truncate failed to get a transaction handle at
3947 * all, we need to clean up the in-core orphan list manually.
3949 if (orphan && inode->i_nlink)
3950 ext4_orphan_del(NULL, inode);
3952 if (!rc && (ia_valid & ATTR_MODE))
3953 rc = ext4_acl_chmod(inode);
3956 ext4_std_error(inode->i_sb, error);
3962 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
3965 struct inode *inode;
3966 unsigned long delalloc_blocks;
3968 inode = dentry->d_inode;
3969 generic_fillattr(inode, stat);
3972 * We can't update i_blocks if the block allocation is delayed
3973 * otherwise in the case of system crash before the real block
3974 * allocation is done, we will have i_blocks inconsistent with
3975 * on-disk file blocks.
3976 * We always keep i_blocks updated together with real
3977 * allocation. But to not confuse with user, stat
3978 * will return the blocks that include the delayed allocation
3979 * blocks for this file.
3981 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
3983 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
3987 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
3989 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3990 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
3991 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
3995 * Account for index blocks, block groups bitmaps and block group
3996 * descriptor blocks if modify datablocks and index blocks
3997 * worse case, the indexs blocks spread over different block groups
3999 * If datablocks are discontiguous, they are possible to spread over
4000 * different block groups too. If they are contiuguous, with flexbg,
4001 * they could still across block group boundary.
4003 * Also account for superblock, inode, quota and xattr blocks
4005 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4007 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4013 * How many index blocks need to touch to modify nrblocks?
4014 * The "Chunk" flag indicating whether the nrblocks is
4015 * physically contiguous on disk
4017 * For Direct IO and fallocate, they calls get_block to allocate
4018 * one single extent at a time, so they could set the "Chunk" flag
4020 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4025 * Now let's see how many group bitmaps and group descriptors need
4035 if (groups > ngroups)
4037 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4038 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4040 /* bitmaps and block group descriptor blocks */
4041 ret += groups + gdpblocks;
4043 /* Blocks for super block, inode, quota and xattr blocks */
4044 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4050 * Calculate the total number of credits to reserve to fit
4051 * the modification of a single pages into a single transaction,
4052 * which may include multiple chunks of block allocations.
4054 * This could be called via ext4_write_begin()
4056 * We need to consider the worse case, when
4057 * one new block per extent.
4059 int ext4_writepage_trans_blocks(struct inode *inode)
4061 int bpp = ext4_journal_blocks_per_page(inode);
4064 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4066 /* Account for data blocks for journalled mode */
4067 if (ext4_should_journal_data(inode))
4073 * Calculate the journal credits for a chunk of data modification.
4075 * This is called from DIO, fallocate or whoever calling
4076 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4078 * journal buffers for data blocks are not included here, as DIO
4079 * and fallocate do no need to journal data buffers.
4081 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4083 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4087 * The caller must have previously called ext4_reserve_inode_write().
4088 * Give this, we know that the caller already has write access to iloc->bh.
4090 int ext4_mark_iloc_dirty(handle_t *handle,
4091 struct inode *inode, struct ext4_iloc *iloc)
4095 if (test_opt(inode->i_sb, I_VERSION))
4096 inode_inc_iversion(inode);
4098 /* the do_update_inode consumes one bh->b_count */
4101 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4102 err = ext4_do_update_inode(handle, inode, iloc);
4108 * On success, We end up with an outstanding reference count against
4109 * iloc->bh. This _must_ be cleaned up later.
4113 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4114 struct ext4_iloc *iloc)
4118 err = ext4_get_inode_loc(inode, iloc);
4120 BUFFER_TRACE(iloc->bh, "get_write_access");
4121 err = ext4_journal_get_write_access(handle, iloc->bh);
4127 ext4_std_error(inode->i_sb, err);
4132 * Expand an inode by new_extra_isize bytes.
4133 * Returns 0 on success or negative error number on failure.
4135 static int ext4_expand_extra_isize(struct inode *inode,
4136 unsigned int new_extra_isize,
4137 struct ext4_iloc iloc,
4140 struct ext4_inode *raw_inode;
4141 struct ext4_xattr_ibody_header *header;
4143 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4146 raw_inode = ext4_raw_inode(&iloc);
4148 header = IHDR(inode, raw_inode);
4150 /* No extended attributes present */
4151 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4152 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4153 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4155 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4159 /* try to expand with EAs present */
4160 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4165 * What we do here is to mark the in-core inode as clean with respect to inode
4166 * dirtiness (it may still be data-dirty).
4167 * This means that the in-core inode may be reaped by prune_icache
4168 * without having to perform any I/O. This is a very good thing,
4169 * because *any* task may call prune_icache - even ones which
4170 * have a transaction open against a different journal.
4172 * Is this cheating? Not really. Sure, we haven't written the
4173 * inode out, but prune_icache isn't a user-visible syncing function.
4174 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4175 * we start and wait on commits.
4177 * Is this efficient/effective? Well, we're being nice to the system
4178 * by cleaning up our inodes proactively so they can be reaped
4179 * without I/O. But we are potentially leaving up to five seconds'
4180 * worth of inodes floating about which prune_icache wants us to
4181 * write out. One way to fix that would be to get prune_icache()
4182 * to do a write_super() to free up some memory. It has the desired
4185 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4187 struct ext4_iloc iloc;
4188 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4189 static unsigned int mnt_count;
4193 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4194 err = ext4_reserve_inode_write(handle, inode, &iloc);
4195 if (ext4_handle_valid(handle) &&
4196 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4197 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4199 * We need extra buffer credits since we may write into EA block
4200 * with this same handle. If journal_extend fails, then it will
4201 * only result in a minor loss of functionality for that inode.
4202 * If this is felt to be critical, then e2fsck should be run to
4203 * force a large enough s_min_extra_isize.
4205 if ((jbd2_journal_extend(handle,
4206 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4207 ret = ext4_expand_extra_isize(inode,
4208 sbi->s_want_extra_isize,
4211 ext4_set_inode_state(inode,
4212 EXT4_STATE_NO_EXPAND);
4214 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4215 ext4_warning(inode->i_sb,
4216 "Unable to expand inode %lu. Delete"
4217 " some EAs or run e2fsck.",
4220 le16_to_cpu(sbi->s_es->s_mnt_count);
4226 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4231 * ext4_dirty_inode() is called from __mark_inode_dirty()
4233 * We're really interested in the case where a file is being extended.
4234 * i_size has been changed by generic_commit_write() and we thus need
4235 * to include the updated inode in the current transaction.
4237 * Also, dquot_alloc_block() will always dirty the inode when blocks
4238 * are allocated to the file.
4240 * If the inode is marked synchronous, we don't honour that here - doing
4241 * so would cause a commit on atime updates, which we don't bother doing.
4242 * We handle synchronous inodes at the highest possible level.
4244 void ext4_dirty_inode(struct inode *inode, int flags)
4248 handle = ext4_journal_start(inode, 2);
4252 ext4_mark_inode_dirty(handle, inode);
4254 ext4_journal_stop(handle);
4261 * Bind an inode's backing buffer_head into this transaction, to prevent
4262 * it from being flushed to disk early. Unlike
4263 * ext4_reserve_inode_write, this leaves behind no bh reference and
4264 * returns no iloc structure, so the caller needs to repeat the iloc
4265 * lookup to mark the inode dirty later.
4267 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4269 struct ext4_iloc iloc;
4273 err = ext4_get_inode_loc(inode, &iloc);
4275 BUFFER_TRACE(iloc.bh, "get_write_access");
4276 err = jbd2_journal_get_write_access(handle, iloc.bh);
4278 err = ext4_handle_dirty_metadata(handle,
4284 ext4_std_error(inode->i_sb, err);
4289 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4296 * We have to be very careful here: changing a data block's
4297 * journaling status dynamically is dangerous. If we write a
4298 * data block to the journal, change the status and then delete
4299 * that block, we risk forgetting to revoke the old log record
4300 * from the journal and so a subsequent replay can corrupt data.
4301 * So, first we make sure that the journal is empty and that
4302 * nobody is changing anything.
4305 journal = EXT4_JOURNAL(inode);
4308 if (is_journal_aborted(journal))
4311 jbd2_journal_lock_updates(journal);
4312 jbd2_journal_flush(journal);
4315 * OK, there are no updates running now, and all cached data is
4316 * synced to disk. We are now in a completely consistent state
4317 * which doesn't have anything in the journal, and we know that
4318 * no filesystem updates are running, so it is safe to modify
4319 * the inode's in-core data-journaling state flag now.
4323 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4325 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4326 ext4_set_aops(inode);
4328 jbd2_journal_unlock_updates(journal);
4330 /* Finally we can mark the inode as dirty. */
4332 handle = ext4_journal_start(inode, 1);
4334 return PTR_ERR(handle);
4336 err = ext4_mark_inode_dirty(handle, inode);
4337 ext4_handle_sync(handle);
4338 ext4_journal_stop(handle);
4339 ext4_std_error(inode->i_sb, err);
4344 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4346 return !buffer_mapped(bh);
4349 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4351 struct page *page = vmf->page;
4355 struct file *file = vma->vm_file;
4356 struct inode *inode = file->f_path.dentry->d_inode;
4357 struct address_space *mapping = inode->i_mapping;
4359 get_block_t *get_block;
4363 * This check is racy but catches the common case. We rely on
4364 * __block_page_mkwrite() to do a reliable check.
4366 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
4367 /* Delalloc case is easy... */
4368 if (test_opt(inode->i_sb, DELALLOC) &&
4369 !ext4_should_journal_data(inode) &&
4370 !ext4_nonda_switch(inode->i_sb)) {
4372 ret = __block_page_mkwrite(vma, vmf,
4373 ext4_da_get_block_prep);
4374 } while (ret == -ENOSPC &&
4375 ext4_should_retry_alloc(inode->i_sb, &retries));
4380 size = i_size_read(inode);
4381 /* Page got truncated from under us? */
4382 if (page->mapping != mapping || page_offset(page) > size) {
4384 ret = VM_FAULT_NOPAGE;
4388 if (page->index == size >> PAGE_CACHE_SHIFT)
4389 len = size & ~PAGE_CACHE_MASK;
4391 len = PAGE_CACHE_SIZE;
4393 * Return if we have all the buffers mapped. This avoids the need to do
4394 * journal_start/journal_stop which can block and take a long time
4396 if (page_has_buffers(page)) {
4397 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4398 ext4_bh_unmapped)) {
4399 /* Wait so that we don't change page under IO */
4400 wait_on_page_writeback(page);
4401 ret = VM_FAULT_LOCKED;
4406 /* OK, we need to fill the hole... */
4407 if (ext4_should_dioread_nolock(inode))
4408 get_block = ext4_get_block_write;
4410 get_block = ext4_get_block;
4412 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4413 if (IS_ERR(handle)) {
4414 ret = VM_FAULT_SIGBUS;
4417 ret = __block_page_mkwrite(vma, vmf, get_block);
4418 if (!ret && ext4_should_journal_data(inode)) {
4419 if (walk_page_buffers(handle, page_buffers(page), 0,
4420 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4422 ret = VM_FAULT_SIGBUS;
4425 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4427 ext4_journal_stop(handle);
4428 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4431 ret = block_page_mkwrite_return(ret);