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"
47 #include <trace/events/ext4.h>
49 #define MPAGE_DA_EXTENT_TAIL 0x01
51 static inline int ext4_begin_ordered_truncate(struct inode *inode,
54 trace_ext4_begin_ordered_truncate(inode, new_size);
56 * If jinode is zero, then we never opened the file for
57 * writing, so there's no need to call
58 * jbd2_journal_begin_ordered_truncate() since there's no
59 * outstanding writes we need to flush.
61 if (!EXT4_I(inode)->jinode)
63 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
64 EXT4_I(inode)->jinode,
68 static void ext4_invalidatepage(struct page *page, unsigned long offset);
69 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
70 struct buffer_head *bh_result, int create);
71 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
72 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
73 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
74 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
77 * Test whether an inode is a fast symlink.
79 static int ext4_inode_is_fast_symlink(struct inode *inode)
81 int ea_blocks = EXT4_I(inode)->i_file_acl ?
82 (inode->i_sb->s_blocksize >> 9) : 0;
84 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
88 * Restart the transaction associated with *handle. This does a commit,
89 * so before we call here everything must be consistently dirtied against
92 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
98 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
99 * moment, get_block can be called only for blocks inside i_size since
100 * page cache has been already dropped and writes are blocked by
101 * i_mutex. So we can safely drop the i_data_sem here.
103 BUG_ON(EXT4_JOURNAL(inode) == NULL);
104 jbd_debug(2, "restarting handle %p\n", handle);
105 up_write(&EXT4_I(inode)->i_data_sem);
106 ret = ext4_journal_restart(handle, nblocks);
107 down_write(&EXT4_I(inode)->i_data_sem);
108 ext4_discard_preallocations(inode);
114 * Called at the last iput() if i_nlink is zero.
116 void ext4_evict_inode(struct inode *inode)
121 trace_ext4_evict_inode(inode);
123 ext4_ioend_wait(inode);
125 if (inode->i_nlink) {
127 * When journalling data dirty buffers are tracked only in the
128 * journal. So although mm thinks everything is clean and
129 * ready for reaping the inode might still have some pages to
130 * write in the running transaction or waiting to be
131 * checkpointed. Thus calling jbd2_journal_invalidatepage()
132 * (via truncate_inode_pages()) to discard these buffers can
133 * cause data loss. Also even if we did not discard these
134 * buffers, we would have no way to find them after the inode
135 * is reaped and thus user could see stale data if he tries to
136 * read them before the transaction is checkpointed. So be
137 * careful and force everything to disk here... We use
138 * ei->i_datasync_tid to store the newest transaction
139 * containing inode's data.
141 * Note that directories do not have this problem because they
142 * don't use page cache.
144 if (ext4_should_journal_data(inode) &&
145 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
146 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
147 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
149 jbd2_log_start_commit(journal, commit_tid);
150 jbd2_log_wait_commit(journal, commit_tid);
151 filemap_write_and_wait(&inode->i_data);
153 truncate_inode_pages(&inode->i_data, 0);
157 if (!is_bad_inode(inode))
158 dquot_initialize(inode);
160 if (ext4_should_order_data(inode))
161 ext4_begin_ordered_truncate(inode, 0);
162 truncate_inode_pages(&inode->i_data, 0);
164 if (is_bad_inode(inode))
167 handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
168 if (IS_ERR(handle)) {
169 ext4_std_error(inode->i_sb, PTR_ERR(handle));
171 * If we're going to skip the normal cleanup, we still need to
172 * make sure that the in-core orphan linked list is properly
175 ext4_orphan_del(NULL, inode);
180 ext4_handle_sync(handle);
182 err = ext4_mark_inode_dirty(handle, inode);
184 ext4_warning(inode->i_sb,
185 "couldn't mark inode dirty (err %d)", err);
189 ext4_truncate(inode);
192 * ext4_ext_truncate() doesn't reserve any slop when it
193 * restarts journal transactions; therefore there may not be
194 * enough credits left in the handle to remove the inode from
195 * the orphan list and set the dtime field.
197 if (!ext4_handle_has_enough_credits(handle, 3)) {
198 err = ext4_journal_extend(handle, 3);
200 err = ext4_journal_restart(handle, 3);
202 ext4_warning(inode->i_sb,
203 "couldn't extend journal (err %d)", err);
205 ext4_journal_stop(handle);
206 ext4_orphan_del(NULL, inode);
212 * Kill off the orphan record which ext4_truncate created.
213 * AKPM: I think this can be inside the above `if'.
214 * Note that ext4_orphan_del() has to be able to cope with the
215 * deletion of a non-existent orphan - this is because we don't
216 * know if ext4_truncate() actually created an orphan record.
217 * (Well, we could do this if we need to, but heck - it works)
219 ext4_orphan_del(handle, inode);
220 EXT4_I(inode)->i_dtime = get_seconds();
223 * One subtle ordering requirement: if anything has gone wrong
224 * (transaction abort, IO errors, whatever), then we can still
225 * do these next steps (the fs will already have been marked as
226 * having errors), but we can't free the inode if the mark_dirty
229 if (ext4_mark_inode_dirty(handle, inode))
230 /* If that failed, just do the required in-core inode clear. */
231 ext4_clear_inode(inode);
233 ext4_free_inode(handle, inode);
234 ext4_journal_stop(handle);
237 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
241 qsize_t *ext4_get_reserved_space(struct inode *inode)
243 return &EXT4_I(inode)->i_reserved_quota;
248 * Calculate the number of metadata blocks need to reserve
249 * to allocate a block located at @lblock
251 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
253 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
254 return ext4_ext_calc_metadata_amount(inode, lblock);
256 return ext4_ind_calc_metadata_amount(inode, lblock);
260 * Called with i_data_sem down, which is important since we can call
261 * ext4_discard_preallocations() from here.
263 void ext4_da_update_reserve_space(struct inode *inode,
264 int used, int quota_claim)
266 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
267 struct ext4_inode_info *ei = EXT4_I(inode);
269 spin_lock(&ei->i_block_reservation_lock);
270 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
271 if (unlikely(used > ei->i_reserved_data_blocks)) {
272 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
273 "with only %d reserved data blocks\n",
274 __func__, inode->i_ino, used,
275 ei->i_reserved_data_blocks);
277 used = ei->i_reserved_data_blocks;
280 /* Update per-inode reservations */
281 ei->i_reserved_data_blocks -= used;
282 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
283 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
284 used + ei->i_allocated_meta_blocks);
285 ei->i_allocated_meta_blocks = 0;
287 if (ei->i_reserved_data_blocks == 0) {
289 * We can release all of the reserved metadata blocks
290 * only when we have written all of the delayed
293 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
294 ei->i_reserved_meta_blocks);
295 ei->i_reserved_meta_blocks = 0;
296 ei->i_da_metadata_calc_len = 0;
298 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
300 /* Update quota subsystem for data blocks */
302 dquot_claim_block(inode, EXT4_C2B(sbi, used));
305 * We did fallocate with an offset that is already delayed
306 * allocated. So on delayed allocated writeback we should
307 * not re-claim the quota for fallocated blocks.
309 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
313 * If we have done all the pending block allocations and if
314 * there aren't any writers on the inode, we can discard the
315 * inode's preallocations.
317 if ((ei->i_reserved_data_blocks == 0) &&
318 (atomic_read(&inode->i_writecount) == 0))
319 ext4_discard_preallocations(inode);
322 static int __check_block_validity(struct inode *inode, const char *func,
324 struct ext4_map_blocks *map)
326 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
328 ext4_error_inode(inode, func, line, map->m_pblk,
329 "lblock %lu mapped to illegal pblock "
330 "(length %d)", (unsigned long) map->m_lblk,
337 #define check_block_validity(inode, map) \
338 __check_block_validity((inode), __func__, __LINE__, (map))
341 * Return the number of contiguous dirty pages in a given inode
342 * starting at page frame idx.
344 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
345 unsigned int max_pages)
347 struct address_space *mapping = inode->i_mapping;
351 int i, nr_pages, done = 0;
355 pagevec_init(&pvec, 0);
358 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
360 (pgoff_t)PAGEVEC_SIZE);
363 for (i = 0; i < nr_pages; i++) {
364 struct page *page = pvec.pages[i];
365 struct buffer_head *bh, *head;
368 if (unlikely(page->mapping != mapping) ||
370 PageWriteback(page) ||
371 page->index != idx) {
376 if (page_has_buffers(page)) {
377 bh = head = page_buffers(page);
379 if (!buffer_delay(bh) &&
380 !buffer_unwritten(bh))
382 bh = bh->b_this_page;
383 } while (!done && (bh != head));
390 if (num >= max_pages) {
395 pagevec_release(&pvec);
401 * Sets the BH_Da_Mapped bit on the buffer heads corresponding to the given map.
403 static void set_buffers_da_mapped(struct inode *inode,
404 struct ext4_map_blocks *map)
406 struct address_space *mapping = inode->i_mapping;
411 index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
412 end = (map->m_lblk + map->m_len - 1) >>
413 (PAGE_CACHE_SHIFT - inode->i_blkbits);
415 pagevec_init(&pvec, 0);
416 while (index <= end) {
417 nr_pages = pagevec_lookup(&pvec, mapping, index,
419 (pgoff_t)PAGEVEC_SIZE));
422 for (i = 0; i < nr_pages; i++) {
423 struct page *page = pvec.pages[i];
424 struct buffer_head *bh, *head;
426 if (unlikely(page->mapping != mapping) ||
430 if (page_has_buffers(page)) {
431 bh = head = page_buffers(page);
433 set_buffer_da_mapped(bh);
434 bh = bh->b_this_page;
435 } while (bh != head);
439 pagevec_release(&pvec);
444 * The ext4_map_blocks() function tries to look up the requested blocks,
445 * and returns if the blocks are already mapped.
447 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
448 * and store the allocated blocks in the result buffer head and mark it
451 * If file type is extents based, it will call ext4_ext_map_blocks(),
452 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
455 * On success, it returns the number of blocks being mapped or allocate.
456 * if create==0 and the blocks are pre-allocated and uninitialized block,
457 * the result buffer head is unmapped. If the create ==1, it will make sure
458 * the buffer head is mapped.
460 * It returns 0 if plain look up failed (blocks have not been allocated), in
461 * that case, buffer head is unmapped
463 * It returns the error in case of allocation failure.
465 int ext4_map_blocks(handle_t *handle, struct inode *inode,
466 struct ext4_map_blocks *map, int flags)
471 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
472 "logical block %lu\n", inode->i_ino, flags, map->m_len,
473 (unsigned long) map->m_lblk);
475 * Try to see if we can get the block without requesting a new
478 down_read((&EXT4_I(inode)->i_data_sem));
479 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
480 retval = ext4_ext_map_blocks(handle, inode, map, 0);
482 retval = ext4_ind_map_blocks(handle, inode, map, 0);
484 up_read((&EXT4_I(inode)->i_data_sem));
486 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
487 int ret = check_block_validity(inode, map);
492 /* If it is only a block(s) look up */
493 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
497 * Returns if the blocks have already allocated
499 * Note that if blocks have been preallocated
500 * ext4_ext_get_block() returns the create = 0
501 * with buffer head unmapped.
503 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
507 * When we call get_blocks without the create flag, the
508 * BH_Unwritten flag could have gotten set if the blocks
509 * requested were part of a uninitialized extent. We need to
510 * clear this flag now that we are committed to convert all or
511 * part of the uninitialized extent to be an initialized
512 * extent. This is because we need to avoid the combination
513 * of BH_Unwritten and BH_Mapped flags being simultaneously
514 * set on the buffer_head.
516 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
519 * New blocks allocate and/or writing to uninitialized extent
520 * will possibly result in updating i_data, so we take
521 * the write lock of i_data_sem, and call get_blocks()
522 * with create == 1 flag.
524 down_write((&EXT4_I(inode)->i_data_sem));
527 * if the caller is from delayed allocation writeout path
528 * we have already reserved fs blocks for allocation
529 * let the underlying get_block() function know to
530 * avoid double accounting
532 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
533 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
535 * We need to check for EXT4 here because migrate
536 * could have changed the inode type in between
538 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
539 retval = ext4_ext_map_blocks(handle, inode, map, flags);
541 retval = ext4_ind_map_blocks(handle, inode, map, flags);
543 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
545 * We allocated new blocks which will result in
546 * i_data's format changing. Force the migrate
547 * to fail by clearing migrate flags
549 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
553 * Update reserved blocks/metadata blocks after successful
554 * block allocation which had been deferred till now. We don't
555 * support fallocate for non extent files. So we can update
556 * reserve space here.
559 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
560 ext4_da_update_reserve_space(inode, retval, 1);
562 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
563 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
565 /* If we have successfully mapped the delayed allocated blocks,
566 * set the BH_Da_Mapped bit on them. Its important to do this
567 * under the protection of i_data_sem.
569 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
570 set_buffers_da_mapped(inode, map);
573 up_write((&EXT4_I(inode)->i_data_sem));
574 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
575 int ret = check_block_validity(inode, map);
582 /* Maximum number of blocks we map for direct IO at once. */
583 #define DIO_MAX_BLOCKS 4096
585 static int _ext4_get_block(struct inode *inode, sector_t iblock,
586 struct buffer_head *bh, int flags)
588 handle_t *handle = ext4_journal_current_handle();
589 struct ext4_map_blocks map;
590 int ret = 0, started = 0;
594 map.m_len = bh->b_size >> inode->i_blkbits;
596 if (flags && !handle) {
597 /* Direct IO write... */
598 if (map.m_len > DIO_MAX_BLOCKS)
599 map.m_len = DIO_MAX_BLOCKS;
600 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
601 handle = ext4_journal_start(inode, dio_credits);
602 if (IS_ERR(handle)) {
603 ret = PTR_ERR(handle);
609 ret = ext4_map_blocks(handle, inode, &map, flags);
611 map_bh(bh, inode->i_sb, map.m_pblk);
612 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
613 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
617 ext4_journal_stop(handle);
621 int ext4_get_block(struct inode *inode, sector_t iblock,
622 struct buffer_head *bh, int create)
624 return _ext4_get_block(inode, iblock, bh,
625 create ? EXT4_GET_BLOCKS_CREATE : 0);
629 * `handle' can be NULL if create is zero
631 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
632 ext4_lblk_t block, int create, int *errp)
634 struct ext4_map_blocks map;
635 struct buffer_head *bh;
638 J_ASSERT(handle != NULL || create == 0);
642 err = ext4_map_blocks(handle, inode, &map,
643 create ? EXT4_GET_BLOCKS_CREATE : 0);
651 bh = sb_getblk(inode->i_sb, map.m_pblk);
656 if (map.m_flags & EXT4_MAP_NEW) {
657 J_ASSERT(create != 0);
658 J_ASSERT(handle != NULL);
661 * Now that we do not always journal data, we should
662 * keep in mind whether this should always journal the
663 * new buffer as metadata. For now, regular file
664 * writes use ext4_get_block instead, so it's not a
668 BUFFER_TRACE(bh, "call get_create_access");
669 fatal = ext4_journal_get_create_access(handle, bh);
670 if (!fatal && !buffer_uptodate(bh)) {
671 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
672 set_buffer_uptodate(bh);
675 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
676 err = ext4_handle_dirty_metadata(handle, inode, bh);
680 BUFFER_TRACE(bh, "not a new buffer");
690 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
691 ext4_lblk_t block, int create, int *err)
693 struct buffer_head *bh;
695 bh = ext4_getblk(handle, inode, block, create, err);
698 if (buffer_uptodate(bh))
700 ll_rw_block(READ_META, 1, &bh);
702 if (buffer_uptodate(bh))
709 static int walk_page_buffers(handle_t *handle,
710 struct buffer_head *head,
714 int (*fn)(handle_t *handle,
715 struct buffer_head *bh))
717 struct buffer_head *bh;
718 unsigned block_start, block_end;
719 unsigned blocksize = head->b_size;
721 struct buffer_head *next;
723 for (bh = head, block_start = 0;
724 ret == 0 && (bh != head || !block_start);
725 block_start = block_end, bh = next) {
726 next = bh->b_this_page;
727 block_end = block_start + blocksize;
728 if (block_end <= from || block_start >= to) {
729 if (partial && !buffer_uptodate(bh))
733 err = (*fn)(handle, bh);
741 * To preserve ordering, it is essential that the hole instantiation and
742 * the data write be encapsulated in a single transaction. We cannot
743 * close off a transaction and start a new one between the ext4_get_block()
744 * and the commit_write(). So doing the jbd2_journal_start at the start of
745 * prepare_write() is the right place.
747 * Also, this function can nest inside ext4_writepage() ->
748 * block_write_full_page(). In that case, we *know* that ext4_writepage()
749 * has generated enough buffer credits to do the whole page. So we won't
750 * block on the journal in that case, which is good, because the caller may
753 * By accident, ext4 can be reentered when a transaction is open via
754 * quota file writes. If we were to commit the transaction while thus
755 * reentered, there can be a deadlock - we would be holding a quota
756 * lock, and the commit would never complete if another thread had a
757 * transaction open and was blocking on the quota lock - a ranking
760 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
761 * will _not_ run commit under these circumstances because handle->h_ref
762 * is elevated. We'll still have enough credits for the tiny quotafile
765 static int do_journal_get_write_access(handle_t *handle,
766 struct buffer_head *bh)
768 int dirty = buffer_dirty(bh);
771 if (!buffer_mapped(bh) || buffer_freed(bh))
774 * __block_write_begin() could have dirtied some buffers. Clean
775 * the dirty bit as jbd2_journal_get_write_access() could complain
776 * otherwise about fs integrity issues. Setting of the dirty bit
777 * by __block_write_begin() isn't a real problem here as we clear
778 * the bit before releasing a page lock and thus writeback cannot
779 * ever write the buffer.
782 clear_buffer_dirty(bh);
783 ret = ext4_journal_get_write_access(handle, bh);
785 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
789 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
790 struct buffer_head *bh_result, int create);
791 static int ext4_write_begin(struct file *file, struct address_space *mapping,
792 loff_t pos, unsigned len, unsigned flags,
793 struct page **pagep, void **fsdata)
795 struct inode *inode = mapping->host;
796 int ret, needed_blocks;
803 trace_ext4_write_begin(inode, pos, len, flags);
805 * Reserve one block more for addition to orphan list in case
806 * we allocate blocks but write fails for some reason
808 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
809 index = pos >> PAGE_CACHE_SHIFT;
810 from = pos & (PAGE_CACHE_SIZE - 1);
814 handle = ext4_journal_start(inode, needed_blocks);
815 if (IS_ERR(handle)) {
816 ret = PTR_ERR(handle);
820 /* We cannot recurse into the filesystem as the transaction is already
822 flags |= AOP_FLAG_NOFS;
824 page = grab_cache_page_write_begin(mapping, index, flags);
826 ext4_journal_stop(handle);
832 if (ext4_should_dioread_nolock(inode))
833 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
835 ret = __block_write_begin(page, pos, len, ext4_get_block);
837 if (!ret && ext4_should_journal_data(inode)) {
838 ret = walk_page_buffers(handle, page_buffers(page),
839 from, to, NULL, do_journal_get_write_access);
844 page_cache_release(page);
846 * __block_write_begin may have instantiated a few blocks
847 * outside i_size. Trim these off again. Don't need
848 * i_size_read because we hold i_mutex.
850 * Add inode to orphan list in case we crash before
853 if (pos + len > inode->i_size && ext4_can_truncate(inode))
854 ext4_orphan_add(handle, inode);
856 ext4_journal_stop(handle);
857 if (pos + len > inode->i_size) {
858 ext4_truncate_failed_write(inode);
860 * If truncate failed early the inode might
861 * still be on the orphan list; we need to
862 * make sure the inode is removed from the
863 * orphan list in that case.
866 ext4_orphan_del(NULL, inode);
870 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
876 /* For write_end() in data=journal mode */
877 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
879 if (!buffer_mapped(bh) || buffer_freed(bh))
881 set_buffer_uptodate(bh);
882 return ext4_handle_dirty_metadata(handle, NULL, bh);
885 static int ext4_generic_write_end(struct file *file,
886 struct address_space *mapping,
887 loff_t pos, unsigned len, unsigned copied,
888 struct page *page, void *fsdata)
890 int i_size_changed = 0;
891 struct inode *inode = mapping->host;
892 handle_t *handle = ext4_journal_current_handle();
894 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
897 * No need to use i_size_read() here, the i_size
898 * cannot change under us because we hold i_mutex.
900 * But it's important to update i_size while still holding page lock:
901 * page writeout could otherwise come in and zero beyond i_size.
903 if (pos + copied > inode->i_size) {
904 i_size_write(inode, pos + copied);
908 if (pos + copied > EXT4_I(inode)->i_disksize) {
909 /* We need to mark inode dirty even if
910 * new_i_size is less that inode->i_size
911 * bu greater than i_disksize.(hint delalloc)
913 ext4_update_i_disksize(inode, (pos + copied));
917 page_cache_release(page);
920 * Don't mark the inode dirty under page lock. First, it unnecessarily
921 * makes the holding time of page lock longer. Second, it forces lock
922 * ordering of page lock and transaction start for journaling
926 ext4_mark_inode_dirty(handle, inode);
932 * We need to pick up the new inode size which generic_commit_write gave us
933 * `file' can be NULL - eg, when called from page_symlink().
935 * ext4 never places buffers on inode->i_mapping->private_list. metadata
936 * buffers are managed internally.
938 static int ext4_ordered_write_end(struct file *file,
939 struct address_space *mapping,
940 loff_t pos, unsigned len, unsigned copied,
941 struct page *page, void *fsdata)
943 handle_t *handle = ext4_journal_current_handle();
944 struct inode *inode = mapping->host;
947 trace_ext4_ordered_write_end(inode, pos, len, copied);
948 ret = ext4_jbd2_file_inode(handle, inode);
951 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
954 if (pos + len > inode->i_size && ext4_can_truncate(inode))
955 /* if we have allocated more blocks and copied
956 * less. We will have blocks allocated outside
957 * inode->i_size. So truncate them
959 ext4_orphan_add(handle, inode);
964 page_cache_release(page);
967 ret2 = ext4_journal_stop(handle);
971 if (pos + len > inode->i_size) {
972 ext4_truncate_failed_write(inode);
974 * If truncate failed early the inode might still be
975 * on the orphan list; we need to make sure the inode
976 * is removed from the orphan list in that case.
979 ext4_orphan_del(NULL, inode);
983 return ret ? ret : copied;
986 static int ext4_writeback_write_end(struct file *file,
987 struct address_space *mapping,
988 loff_t pos, unsigned len, unsigned copied,
989 struct page *page, void *fsdata)
991 handle_t *handle = ext4_journal_current_handle();
992 struct inode *inode = mapping->host;
995 trace_ext4_writeback_write_end(inode, pos, len, copied);
996 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
999 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1000 /* if we have allocated more blocks and copied
1001 * less. We will have blocks allocated outside
1002 * inode->i_size. So truncate them
1004 ext4_orphan_add(handle, inode);
1009 ret2 = ext4_journal_stop(handle);
1013 if (pos + len > inode->i_size) {
1014 ext4_truncate_failed_write(inode);
1016 * If truncate failed early the inode might still be
1017 * on the orphan list; we need to make sure the inode
1018 * is removed from the orphan list in that case.
1021 ext4_orphan_del(NULL, inode);
1024 return ret ? ret : copied;
1027 static int ext4_journalled_write_end(struct file *file,
1028 struct address_space *mapping,
1029 loff_t pos, unsigned len, unsigned copied,
1030 struct page *page, void *fsdata)
1032 handle_t *handle = ext4_journal_current_handle();
1033 struct inode *inode = mapping->host;
1039 trace_ext4_journalled_write_end(inode, pos, len, copied);
1040 from = pos & (PAGE_CACHE_SIZE - 1);
1043 BUG_ON(!ext4_handle_valid(handle));
1046 if (!PageUptodate(page))
1048 page_zero_new_buffers(page, from+copied, to);
1051 ret = walk_page_buffers(handle, page_buffers(page), from,
1052 to, &partial, write_end_fn);
1054 SetPageUptodate(page);
1055 new_i_size = pos + copied;
1056 if (new_i_size > inode->i_size)
1057 i_size_write(inode, pos+copied);
1058 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1059 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1060 if (new_i_size > EXT4_I(inode)->i_disksize) {
1061 ext4_update_i_disksize(inode, new_i_size);
1062 ret2 = ext4_mark_inode_dirty(handle, inode);
1068 page_cache_release(page);
1069 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1070 /* if we have allocated more blocks and copied
1071 * less. We will have blocks allocated outside
1072 * inode->i_size. So truncate them
1074 ext4_orphan_add(handle, inode);
1076 ret2 = ext4_journal_stop(handle);
1079 if (pos + len > inode->i_size) {
1080 ext4_truncate_failed_write(inode);
1082 * If truncate failed early the inode might still be
1083 * on the orphan list; we need to make sure the inode
1084 * is removed from the orphan list in that case.
1087 ext4_orphan_del(NULL, inode);
1090 return ret ? ret : copied;
1094 * Reserve a single cluster located at lblock
1096 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1099 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1100 struct ext4_inode_info *ei = EXT4_I(inode);
1101 unsigned int md_needed;
1105 * recalculate the amount of metadata blocks to reserve
1106 * in order to allocate nrblocks
1107 * worse case is one extent per block
1110 spin_lock(&ei->i_block_reservation_lock);
1111 md_needed = EXT4_NUM_B2C(sbi,
1112 ext4_calc_metadata_amount(inode, lblock));
1113 trace_ext4_da_reserve_space(inode, md_needed);
1114 spin_unlock(&ei->i_block_reservation_lock);
1117 * We will charge metadata quota at writeout time; this saves
1118 * us from metadata over-estimation, though we may go over by
1119 * a small amount in the end. Here we just reserve for data.
1121 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1125 * We do still charge estimated metadata to the sb though;
1126 * we cannot afford to run out of free blocks.
1128 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1129 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1130 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1136 spin_lock(&ei->i_block_reservation_lock);
1137 ei->i_reserved_data_blocks++;
1138 ei->i_reserved_meta_blocks += md_needed;
1139 spin_unlock(&ei->i_block_reservation_lock);
1141 return 0; /* success */
1144 static void ext4_da_release_space(struct inode *inode, int to_free)
1146 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1147 struct ext4_inode_info *ei = EXT4_I(inode);
1150 return; /* Nothing to release, exit */
1152 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1154 trace_ext4_da_release_space(inode, to_free);
1155 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1157 * if there aren't enough reserved blocks, then the
1158 * counter is messed up somewhere. Since this
1159 * function is called from invalidate page, it's
1160 * harmless to return without any action.
1162 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1163 "ino %lu, to_free %d with only %d reserved "
1164 "data blocks\n", inode->i_ino, to_free,
1165 ei->i_reserved_data_blocks);
1167 to_free = ei->i_reserved_data_blocks;
1169 ei->i_reserved_data_blocks -= to_free;
1171 if (ei->i_reserved_data_blocks == 0) {
1173 * We can release all of the reserved metadata blocks
1174 * only when we have written all of the delayed
1175 * allocation blocks.
1176 * Note that in case of bigalloc, i_reserved_meta_blocks,
1177 * i_reserved_data_blocks, etc. refer to number of clusters.
1179 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1180 ei->i_reserved_meta_blocks);
1181 ei->i_reserved_meta_blocks = 0;
1182 ei->i_da_metadata_calc_len = 0;
1185 /* update fs dirty data blocks counter */
1186 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1188 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1190 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1193 static void ext4_da_page_release_reservation(struct page *page,
1194 unsigned long offset)
1197 struct buffer_head *head, *bh;
1198 unsigned int curr_off = 0;
1199 struct inode *inode = page->mapping->host;
1200 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1203 head = page_buffers(page);
1206 unsigned int next_off = curr_off + bh->b_size;
1208 if ((offset <= curr_off) && (buffer_delay(bh))) {
1210 clear_buffer_delay(bh);
1211 clear_buffer_da_mapped(bh);
1213 curr_off = next_off;
1214 } while ((bh = bh->b_this_page) != head);
1216 /* If we have released all the blocks belonging to a cluster, then we
1217 * need to release the reserved space for that cluster. */
1218 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1219 while (num_clusters > 0) {
1221 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1222 ((num_clusters - 1) << sbi->s_cluster_bits);
1223 if (sbi->s_cluster_ratio == 1 ||
1224 !ext4_find_delalloc_cluster(inode, lblk, 1))
1225 ext4_da_release_space(inode, 1);
1232 * Delayed allocation stuff
1236 * mpage_da_submit_io - walks through extent of pages and try to write
1237 * them with writepage() call back
1239 * @mpd->inode: inode
1240 * @mpd->first_page: first page of the extent
1241 * @mpd->next_page: page after the last page of the extent
1243 * By the time mpage_da_submit_io() is called we expect all blocks
1244 * to be allocated. this may be wrong if allocation failed.
1246 * As pages are already locked by write_cache_pages(), we can't use it
1248 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1249 struct ext4_map_blocks *map)
1251 struct pagevec pvec;
1252 unsigned long index, end;
1253 int ret = 0, err, nr_pages, i;
1254 struct inode *inode = mpd->inode;
1255 struct address_space *mapping = inode->i_mapping;
1256 loff_t size = i_size_read(inode);
1257 unsigned int len, block_start;
1258 struct buffer_head *bh, *page_bufs = NULL;
1259 int journal_data = ext4_should_journal_data(inode);
1260 sector_t pblock = 0, cur_logical = 0;
1261 struct ext4_io_submit io_submit;
1263 BUG_ON(mpd->next_page <= mpd->first_page);
1264 memset(&io_submit, 0, sizeof(io_submit));
1266 * We need to start from the first_page to the next_page - 1
1267 * to make sure we also write the mapped dirty buffer_heads.
1268 * If we look at mpd->b_blocknr we would only be looking
1269 * at the currently mapped buffer_heads.
1271 index = mpd->first_page;
1272 end = mpd->next_page - 1;
1274 pagevec_init(&pvec, 0);
1275 while (index <= end) {
1276 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1279 for (i = 0; i < nr_pages; i++) {
1280 int commit_write = 0, skip_page = 0;
1281 struct page *page = pvec.pages[i];
1283 index = page->index;
1287 if (index == size >> PAGE_CACHE_SHIFT)
1288 len = size & ~PAGE_CACHE_MASK;
1290 len = PAGE_CACHE_SIZE;
1292 cur_logical = index << (PAGE_CACHE_SHIFT -
1294 pblock = map->m_pblk + (cur_logical -
1299 BUG_ON(!PageLocked(page));
1300 BUG_ON(PageWriteback(page));
1303 * If the page does not have buffers (for
1304 * whatever reason), try to create them using
1305 * __block_write_begin. If this fails,
1306 * skip the page and move on.
1308 if (!page_has_buffers(page)) {
1309 if (__block_write_begin(page, 0, len,
1310 noalloc_get_block_write)) {
1318 bh = page_bufs = page_buffers(page);
1323 if (map && (cur_logical >= map->m_lblk) &&
1324 (cur_logical <= (map->m_lblk +
1325 (map->m_len - 1)))) {
1326 if (buffer_delay(bh)) {
1327 clear_buffer_delay(bh);
1328 bh->b_blocknr = pblock;
1330 if (buffer_da_mapped(bh))
1331 clear_buffer_da_mapped(bh);
1332 if (buffer_unwritten(bh) ||
1334 BUG_ON(bh->b_blocknr != pblock);
1335 if (map->m_flags & EXT4_MAP_UNINIT)
1336 set_buffer_uninit(bh);
1337 clear_buffer_unwritten(bh);
1340 /* skip page if block allocation undone */
1341 if (buffer_delay(bh) || buffer_unwritten(bh))
1343 bh = bh->b_this_page;
1344 block_start += bh->b_size;
1347 } while (bh != page_bufs);
1353 /* mark the buffer_heads as dirty & uptodate */
1354 block_commit_write(page, 0, len);
1356 clear_page_dirty_for_io(page);
1358 * Delalloc doesn't support data journalling,
1359 * but eventually maybe we'll lift this
1362 if (unlikely(journal_data && PageChecked(page)))
1363 err = __ext4_journalled_writepage(page, len);
1364 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1365 err = ext4_bio_write_page(&io_submit, page,
1367 else if (buffer_uninit(page_bufs)) {
1368 ext4_set_bh_endio(page_bufs, inode);
1369 err = block_write_full_page_endio(page,
1370 noalloc_get_block_write,
1371 mpd->wbc, ext4_end_io_buffer_write);
1373 err = block_write_full_page(page,
1374 noalloc_get_block_write, mpd->wbc);
1377 mpd->pages_written++;
1379 * In error case, we have to continue because
1380 * remaining pages are still locked
1385 pagevec_release(&pvec);
1387 ext4_io_submit(&io_submit);
1391 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1395 struct pagevec pvec;
1396 struct inode *inode = mpd->inode;
1397 struct address_space *mapping = inode->i_mapping;
1399 index = mpd->first_page;
1400 end = mpd->next_page - 1;
1401 while (index <= end) {
1402 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1405 for (i = 0; i < nr_pages; i++) {
1406 struct page *page = pvec.pages[i];
1407 if (page->index > end)
1409 BUG_ON(!PageLocked(page));
1410 BUG_ON(PageWriteback(page));
1411 block_invalidatepage(page, 0);
1412 ClearPageUptodate(page);
1415 index = pvec.pages[nr_pages - 1]->index + 1;
1416 pagevec_release(&pvec);
1421 static void ext4_print_free_blocks(struct inode *inode)
1423 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1424 printk(KERN_CRIT "Total free blocks count %lld\n",
1425 EXT4_C2B(EXT4_SB(inode->i_sb),
1426 ext4_count_free_clusters(inode->i_sb)));
1427 printk(KERN_CRIT "Free/Dirty block details\n");
1428 printk(KERN_CRIT "free_blocks=%lld\n",
1429 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1430 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1431 printk(KERN_CRIT "dirty_blocks=%lld\n",
1432 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1433 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1434 printk(KERN_CRIT "Block reservation details\n");
1435 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
1436 EXT4_I(inode)->i_reserved_data_blocks);
1437 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
1438 EXT4_I(inode)->i_reserved_meta_blocks);
1443 * mpage_da_map_and_submit - go through given space, map them
1444 * if necessary, and then submit them for I/O
1446 * @mpd - bh describing space
1448 * The function skips space we know is already mapped to disk blocks.
1451 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1453 int err, blks, get_blocks_flags;
1454 struct ext4_map_blocks map, *mapp = NULL;
1455 sector_t next = mpd->b_blocknr;
1456 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1457 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1458 handle_t *handle = NULL;
1461 * If the blocks are mapped already, or we couldn't accumulate
1462 * any blocks, then proceed immediately to the submission stage.
1464 if ((mpd->b_size == 0) ||
1465 ((mpd->b_state & (1 << BH_Mapped)) &&
1466 !(mpd->b_state & (1 << BH_Delay)) &&
1467 !(mpd->b_state & (1 << BH_Unwritten))))
1470 handle = ext4_journal_current_handle();
1474 * Call ext4_map_blocks() to allocate any delayed allocation
1475 * blocks, or to convert an uninitialized extent to be
1476 * initialized (in the case where we have written into
1477 * one or more preallocated blocks).
1479 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1480 * indicate that we are on the delayed allocation path. This
1481 * affects functions in many different parts of the allocation
1482 * call path. This flag exists primarily because we don't
1483 * want to change *many* call functions, so ext4_map_blocks()
1484 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1485 * inode's allocation semaphore is taken.
1487 * If the blocks in questions were delalloc blocks, set
1488 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1489 * variables are updated after the blocks have been allocated.
1492 map.m_len = max_blocks;
1493 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1494 if (ext4_should_dioread_nolock(mpd->inode))
1495 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1496 if (mpd->b_state & (1 << BH_Delay))
1497 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1499 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1501 struct super_block *sb = mpd->inode->i_sb;
1505 * If get block returns EAGAIN or ENOSPC and there
1506 * appears to be free blocks we will just let
1507 * mpage_da_submit_io() unlock all of the pages.
1512 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1518 * get block failure will cause us to loop in
1519 * writepages, because a_ops->writepage won't be able
1520 * to make progress. The page will be redirtied by
1521 * writepage and writepages will again try to write
1524 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1525 ext4_msg(sb, KERN_CRIT,
1526 "delayed block allocation failed for inode %lu "
1527 "at logical offset %llu with max blocks %zd "
1528 "with error %d", mpd->inode->i_ino,
1529 (unsigned long long) next,
1530 mpd->b_size >> mpd->inode->i_blkbits, err);
1531 ext4_msg(sb, KERN_CRIT,
1532 "This should not happen!! Data will be lost\n");
1534 ext4_print_free_blocks(mpd->inode);
1536 /* invalidate all the pages */
1537 ext4_da_block_invalidatepages(mpd);
1539 /* Mark this page range as having been completed */
1546 if (map.m_flags & EXT4_MAP_NEW) {
1547 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1550 for (i = 0; i < map.m_len; i++)
1551 unmap_underlying_metadata(bdev, map.m_pblk + i);
1553 if (ext4_should_order_data(mpd->inode)) {
1554 err = ext4_jbd2_file_inode(handle, mpd->inode);
1556 /* Only if the journal is aborted */
1564 * Update on-disk size along with block allocation.
1566 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1567 if (disksize > i_size_read(mpd->inode))
1568 disksize = i_size_read(mpd->inode);
1569 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1570 ext4_update_i_disksize(mpd->inode, disksize);
1571 err = ext4_mark_inode_dirty(handle, mpd->inode);
1573 ext4_error(mpd->inode->i_sb,
1574 "Failed to mark inode %lu dirty",
1579 mpage_da_submit_io(mpd, mapp);
1583 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1584 (1 << BH_Delay) | (1 << BH_Unwritten))
1587 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1589 * @mpd->lbh - extent of blocks
1590 * @logical - logical number of the block in the file
1591 * @bh - bh of the block (used to access block's state)
1593 * the function is used to collect contig. blocks in same state
1595 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1596 sector_t logical, size_t b_size,
1597 unsigned long b_state)
1600 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1603 * XXX Don't go larger than mballoc is willing to allocate
1604 * This is a stopgap solution. We eventually need to fold
1605 * mpage_da_submit_io() into this function and then call
1606 * ext4_map_blocks() multiple times in a loop
1608 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1611 /* check if thereserved journal credits might overflow */
1612 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1613 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1615 * With non-extent format we are limited by the journal
1616 * credit available. Total credit needed to insert
1617 * nrblocks contiguous blocks is dependent on the
1618 * nrblocks. So limit nrblocks.
1621 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1622 EXT4_MAX_TRANS_DATA) {
1624 * Adding the new buffer_head would make it cross the
1625 * allowed limit for which we have journal credit
1626 * reserved. So limit the new bh->b_size
1628 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1629 mpd->inode->i_blkbits;
1630 /* we will do mpage_da_submit_io in the next loop */
1634 * First block in the extent
1636 if (mpd->b_size == 0) {
1637 mpd->b_blocknr = logical;
1638 mpd->b_size = b_size;
1639 mpd->b_state = b_state & BH_FLAGS;
1643 next = mpd->b_blocknr + nrblocks;
1645 * Can we merge the block to our big extent?
1647 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1648 mpd->b_size += b_size;
1654 * We couldn't merge the block to our extent, so we
1655 * need to flush current extent and start new one
1657 mpage_da_map_and_submit(mpd);
1661 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1663 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1667 * This function is grabs code from the very beginning of
1668 * ext4_map_blocks, but assumes that the caller is from delayed write
1669 * time. This function looks up the requested blocks and sets the
1670 * buffer delay bit under the protection of i_data_sem.
1672 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1673 struct ext4_map_blocks *map,
1674 struct buffer_head *bh)
1677 sector_t invalid_block = ~((sector_t) 0xffff);
1679 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1683 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1684 "logical block %lu\n", inode->i_ino, map->m_len,
1685 (unsigned long) map->m_lblk);
1687 * Try to see if we can get the block without requesting a new
1688 * file system block.
1690 down_read((&EXT4_I(inode)->i_data_sem));
1691 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1692 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1694 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1698 * XXX: __block_prepare_write() unmaps passed block,
1701 /* If the block was allocated from previously allocated cluster,
1702 * then we dont need to reserve it again. */
1703 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1704 retval = ext4_da_reserve_space(inode, iblock);
1706 /* not enough space to reserve */
1710 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1711 * and it should not appear on the bh->b_state.
1713 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1715 map_bh(bh, inode->i_sb, invalid_block);
1717 set_buffer_delay(bh);
1721 up_read((&EXT4_I(inode)->i_data_sem));
1727 * This is a special get_blocks_t callback which is used by
1728 * ext4_da_write_begin(). It will either return mapped block or
1729 * reserve space for a single block.
1731 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1732 * We also have b_blocknr = -1 and b_bdev initialized properly
1734 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1735 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1736 * initialized properly.
1738 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1739 struct buffer_head *bh, int create)
1741 struct ext4_map_blocks map;
1744 BUG_ON(create == 0);
1745 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1747 map.m_lblk = iblock;
1751 * first, we need to know whether the block is allocated already
1752 * preallocated blocks are unmapped but should treated
1753 * the same as allocated blocks.
1755 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1759 map_bh(bh, inode->i_sb, map.m_pblk);
1760 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1762 if (buffer_unwritten(bh)) {
1763 /* A delayed write to unwritten bh should be marked
1764 * new and mapped. Mapped ensures that we don't do
1765 * get_block multiple times when we write to the same
1766 * offset and new ensures that we do proper zero out
1767 * for partial write.
1770 set_buffer_mapped(bh);
1776 * This function is used as a standard get_block_t calback function
1777 * when there is no desire to allocate any blocks. It is used as a
1778 * callback function for block_write_begin() and block_write_full_page().
1779 * These functions should only try to map a single block at a time.
1781 * Since this function doesn't do block allocations even if the caller
1782 * requests it by passing in create=1, it is critically important that
1783 * any caller checks to make sure that any buffer heads are returned
1784 * by this function are either all already mapped or marked for
1785 * delayed allocation before calling block_write_full_page(). Otherwise,
1786 * b_blocknr could be left unitialized, and the page write functions will
1787 * be taken by surprise.
1789 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1790 struct buffer_head *bh_result, int create)
1792 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1793 return _ext4_get_block(inode, iblock, bh_result, 0);
1796 static int bget_one(handle_t *handle, struct buffer_head *bh)
1802 static int bput_one(handle_t *handle, struct buffer_head *bh)
1808 static int __ext4_journalled_writepage(struct page *page,
1811 struct address_space *mapping = page->mapping;
1812 struct inode *inode = mapping->host;
1813 struct buffer_head *page_bufs;
1814 handle_t *handle = NULL;
1818 ClearPageChecked(page);
1819 page_bufs = page_buffers(page);
1821 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1822 /* As soon as we unlock the page, it can go away, but we have
1823 * references to buffers so we are safe */
1826 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1827 if (IS_ERR(handle)) {
1828 ret = PTR_ERR(handle);
1832 BUG_ON(!ext4_handle_valid(handle));
1834 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1835 do_journal_get_write_access);
1837 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1841 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1842 err = ext4_journal_stop(handle);
1846 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
1847 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1852 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
1853 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
1856 * Note that we don't need to start a transaction unless we're journaling data
1857 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1858 * need to file the inode to the transaction's list in ordered mode because if
1859 * we are writing back data added by write(), the inode is already there and if
1860 * we are writing back data modified via mmap(), no one guarantees in which
1861 * transaction the data will hit the disk. In case we are journaling data, we
1862 * cannot start transaction directly because transaction start ranks above page
1863 * lock so we have to do some magic.
1865 * This function can get called via...
1866 * - ext4_da_writepages after taking page lock (have journal handle)
1867 * - journal_submit_inode_data_buffers (no journal handle)
1868 * - shrink_page_list via pdflush (no journal handle)
1869 * - grab_page_cache when doing write_begin (have journal handle)
1871 * We don't do any block allocation in this function. If we have page with
1872 * multiple blocks we need to write those buffer_heads that are mapped. This
1873 * is important for mmaped based write. So if we do with blocksize 1K
1874 * truncate(f, 1024);
1875 * a = mmap(f, 0, 4096);
1877 * truncate(f, 4096);
1878 * we have in the page first buffer_head mapped via page_mkwrite call back
1879 * but other bufer_heads would be unmapped but dirty(dirty done via the
1880 * do_wp_page). So writepage should write the first block. If we modify
1881 * the mmap area beyond 1024 we will again get a page_fault and the
1882 * page_mkwrite callback will do the block allocation and mark the
1883 * buffer_heads mapped.
1885 * We redirty the page if we have any buffer_heads that is either delay or
1886 * unwritten in the page.
1888 * We can get recursively called as show below.
1890 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1893 * But since we don't do any block allocation we should not deadlock.
1894 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1896 static int ext4_writepage(struct page *page,
1897 struct writeback_control *wbc)
1899 int ret = 0, commit_write = 0;
1902 struct buffer_head *page_bufs = NULL;
1903 struct inode *inode = page->mapping->host;
1905 trace_ext4_writepage(page);
1906 size = i_size_read(inode);
1907 if (page->index == size >> PAGE_CACHE_SHIFT)
1908 len = size & ~PAGE_CACHE_MASK;
1910 len = PAGE_CACHE_SIZE;
1913 * If the page does not have buffers (for whatever reason),
1914 * try to create them using __block_write_begin. If this
1915 * fails, redirty the page and move on.
1917 if (!page_has_buffers(page)) {
1918 if (__block_write_begin(page, 0, len,
1919 noalloc_get_block_write)) {
1921 redirty_page_for_writepage(wbc, page);
1927 page_bufs = page_buffers(page);
1928 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1929 ext4_bh_delay_or_unwritten)) {
1931 * We don't want to do block allocation, so redirty
1932 * the page and return. We may reach here when we do
1933 * a journal commit via journal_submit_inode_data_buffers.
1934 * We can also reach here via shrink_page_list
1939 /* now mark the buffer_heads as dirty and uptodate */
1940 block_commit_write(page, 0, len);
1942 if (PageChecked(page) && ext4_should_journal_data(inode))
1944 * It's mmapped pagecache. Add buffers and journal it. There
1945 * doesn't seem much point in redirtying the page here.
1947 return __ext4_journalled_writepage(page, len);
1949 if (buffer_uninit(page_bufs)) {
1950 ext4_set_bh_endio(page_bufs, inode);
1951 ret = block_write_full_page_endio(page, noalloc_get_block_write,
1952 wbc, ext4_end_io_buffer_write);
1954 ret = block_write_full_page(page, noalloc_get_block_write,
1961 * This is called via ext4_da_writepages() to
1962 * calculate the total number of credits to reserve to fit
1963 * a single extent allocation into a single transaction,
1964 * ext4_da_writpeages() will loop calling this before
1965 * the block allocation.
1968 static int ext4_da_writepages_trans_blocks(struct inode *inode)
1970 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
1973 * With non-extent format the journal credit needed to
1974 * insert nrblocks contiguous block is dependent on
1975 * number of contiguous block. So we will limit
1976 * number of contiguous block to a sane value
1978 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
1979 (max_blocks > EXT4_MAX_TRANS_DATA))
1980 max_blocks = EXT4_MAX_TRANS_DATA;
1982 return ext4_chunk_trans_blocks(inode, max_blocks);
1986 * write_cache_pages_da - walk the list of dirty pages of the given
1987 * address space and accumulate pages that need writing, and call
1988 * mpage_da_map_and_submit to map a single contiguous memory region
1989 * and then write them.
1991 static int write_cache_pages_da(struct address_space *mapping,
1992 struct writeback_control *wbc,
1993 struct mpage_da_data *mpd,
1994 pgoff_t *done_index)
1996 struct buffer_head *bh, *head;
1997 struct inode *inode = mapping->host;
1998 struct pagevec pvec;
1999 unsigned int nr_pages;
2002 long nr_to_write = wbc->nr_to_write;
2003 int i, tag, ret = 0;
2005 memset(mpd, 0, sizeof(struct mpage_da_data));
2008 pagevec_init(&pvec, 0);
2009 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2010 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2012 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2013 tag = PAGECACHE_TAG_TOWRITE;
2015 tag = PAGECACHE_TAG_DIRTY;
2017 *done_index = index;
2018 while (index <= end) {
2019 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2020 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2024 for (i = 0; i < nr_pages; i++) {
2025 struct page *page = pvec.pages[i];
2028 * At this point, the page may be truncated or
2029 * invalidated (changing page->mapping to NULL), or
2030 * even swizzled back from swapper_space to tmpfs file
2031 * mapping. However, page->index will not change
2032 * because we have a reference on the page.
2034 if (page->index > end)
2037 *done_index = page->index + 1;
2040 * If we can't merge this page, and we have
2041 * accumulated an contiguous region, write it
2043 if ((mpd->next_page != page->index) &&
2044 (mpd->next_page != mpd->first_page)) {
2045 mpage_da_map_and_submit(mpd);
2046 goto ret_extent_tail;
2052 * If the page is no longer dirty, or its
2053 * mapping no longer corresponds to inode we
2054 * are writing (which means it has been
2055 * truncated or invalidated), or the page is
2056 * already under writeback and we are not
2057 * doing a data integrity writeback, skip the page
2059 if (!PageDirty(page) ||
2060 (PageWriteback(page) &&
2061 (wbc->sync_mode == WB_SYNC_NONE)) ||
2062 unlikely(page->mapping != mapping)) {
2067 wait_on_page_writeback(page);
2068 BUG_ON(PageWriteback(page));
2070 if (mpd->next_page != page->index)
2071 mpd->first_page = page->index;
2072 mpd->next_page = page->index + 1;
2073 logical = (sector_t) page->index <<
2074 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2076 if (!page_has_buffers(page)) {
2077 mpage_add_bh_to_extent(mpd, logical,
2079 (1 << BH_Dirty) | (1 << BH_Uptodate));
2081 goto ret_extent_tail;
2084 * Page with regular buffer heads,
2085 * just add all dirty ones
2087 head = page_buffers(page);
2090 BUG_ON(buffer_locked(bh));
2092 * We need to try to allocate
2093 * unmapped blocks in the same page.
2094 * Otherwise we won't make progress
2095 * with the page in ext4_writepage
2097 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2098 mpage_add_bh_to_extent(mpd, logical,
2102 goto ret_extent_tail;
2103 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2105 * mapped dirty buffer. We need
2106 * to update the b_state
2107 * because we look at b_state
2108 * in mpage_da_map_blocks. We
2109 * don't update b_size because
2110 * if we find an unmapped
2111 * buffer_head later we need to
2112 * use the b_state flag of that
2115 if (mpd->b_size == 0)
2116 mpd->b_state = bh->b_state & BH_FLAGS;
2119 } while ((bh = bh->b_this_page) != head);
2122 if (nr_to_write > 0) {
2124 if (nr_to_write == 0 &&
2125 wbc->sync_mode == WB_SYNC_NONE)
2127 * We stop writing back only if we are
2128 * not doing integrity sync. In case of
2129 * integrity sync we have to keep going
2130 * because someone may be concurrently
2131 * dirtying pages, and we might have
2132 * synced a lot of newly appeared dirty
2133 * pages, but have not synced all of the
2139 pagevec_release(&pvec);
2144 ret = MPAGE_DA_EXTENT_TAIL;
2146 pagevec_release(&pvec);
2152 static int ext4_da_writepages(struct address_space *mapping,
2153 struct writeback_control *wbc)
2156 int range_whole = 0;
2157 handle_t *handle = NULL;
2158 struct mpage_da_data mpd;
2159 struct inode *inode = mapping->host;
2160 int pages_written = 0;
2161 unsigned int max_pages;
2162 int range_cyclic, cycled = 1, io_done = 0;
2163 int needed_blocks, ret = 0;
2164 long desired_nr_to_write, nr_to_writebump = 0;
2165 loff_t range_start = wbc->range_start;
2166 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2167 pgoff_t done_index = 0;
2169 struct blk_plug plug;
2171 trace_ext4_da_writepages(inode, wbc);
2174 * No pages to write? This is mainly a kludge to avoid starting
2175 * a transaction for special inodes like journal inode on last iput()
2176 * because that could violate lock ordering on umount
2178 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2182 * If the filesystem has aborted, it is read-only, so return
2183 * right away instead of dumping stack traces later on that
2184 * will obscure the real source of the problem. We test
2185 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2186 * the latter could be true if the filesystem is mounted
2187 * read-only, and in that case, ext4_da_writepages should
2188 * *never* be called, so if that ever happens, we would want
2191 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2194 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2197 range_cyclic = wbc->range_cyclic;
2198 if (wbc->range_cyclic) {
2199 index = mapping->writeback_index;
2202 wbc->range_start = index << PAGE_CACHE_SHIFT;
2203 wbc->range_end = LLONG_MAX;
2204 wbc->range_cyclic = 0;
2207 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2208 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2212 * This works around two forms of stupidity. The first is in
2213 * the writeback code, which caps the maximum number of pages
2214 * written to be 1024 pages. This is wrong on multiple
2215 * levels; different architectues have a different page size,
2216 * which changes the maximum amount of data which gets
2217 * written. Secondly, 4 megabytes is way too small. XFS
2218 * forces this value to be 16 megabytes by multiplying
2219 * nr_to_write parameter by four, and then relies on its
2220 * allocator to allocate larger extents to make them
2221 * contiguous. Unfortunately this brings us to the second
2222 * stupidity, which is that ext4's mballoc code only allocates
2223 * at most 2048 blocks. So we force contiguous writes up to
2224 * the number of dirty blocks in the inode, or
2225 * sbi->max_writeback_mb_bump whichever is smaller.
2227 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2228 if (!range_cyclic && range_whole) {
2229 if (wbc->nr_to_write == LONG_MAX)
2230 desired_nr_to_write = wbc->nr_to_write;
2232 desired_nr_to_write = wbc->nr_to_write * 8;
2234 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2236 if (desired_nr_to_write > max_pages)
2237 desired_nr_to_write = max_pages;
2239 if (wbc->nr_to_write < desired_nr_to_write) {
2240 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2241 wbc->nr_to_write = desired_nr_to_write;
2245 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2246 tag_pages_for_writeback(mapping, index, end);
2248 blk_start_plug(&plug);
2249 while (!ret && wbc->nr_to_write > 0) {
2252 * we insert one extent at a time. So we need
2253 * credit needed for single extent allocation.
2254 * journalled mode is currently not supported
2257 BUG_ON(ext4_should_journal_data(inode));
2258 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2260 /* start a new transaction*/
2261 handle = ext4_journal_start(inode, needed_blocks);
2262 if (IS_ERR(handle)) {
2263 ret = PTR_ERR(handle);
2264 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2265 "%ld pages, ino %lu; err %d", __func__,
2266 wbc->nr_to_write, inode->i_ino, ret);
2267 goto out_writepages;
2271 * Now call write_cache_pages_da() to find the next
2272 * contiguous region of logical blocks that need
2273 * blocks to be allocated by ext4 and submit them.
2275 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
2277 * If we have a contiguous extent of pages and we
2278 * haven't done the I/O yet, map the blocks and submit
2281 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2282 mpage_da_map_and_submit(&mpd);
2283 ret = MPAGE_DA_EXTENT_TAIL;
2285 trace_ext4_da_write_pages(inode, &mpd);
2286 wbc->nr_to_write -= mpd.pages_written;
2288 ext4_journal_stop(handle);
2290 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2291 /* commit the transaction which would
2292 * free blocks released in the transaction
2295 jbd2_journal_force_commit_nested(sbi->s_journal);
2297 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2299 * Got one extent now try with rest of the pages.
2300 * If mpd.retval is set -EIO, journal is aborted.
2301 * So we don't need to write any more.
2303 pages_written += mpd.pages_written;
2306 } else if (wbc->nr_to_write)
2308 * There is no more writeout needed
2309 * or we requested for a noblocking writeout
2310 * and we found the device congested
2314 blk_finish_plug(&plug);
2315 if (!io_done && !cycled) {
2318 wbc->range_start = index << PAGE_CACHE_SHIFT;
2319 wbc->range_end = mapping->writeback_index - 1;
2324 wbc->range_cyclic = range_cyclic;
2325 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2327 * set the writeback_index so that range_cyclic
2328 * mode will write it back later
2330 mapping->writeback_index = done_index;
2333 wbc->nr_to_write -= nr_to_writebump;
2334 wbc->range_start = range_start;
2335 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2339 #define FALL_BACK_TO_NONDELALLOC 1
2340 static int ext4_nonda_switch(struct super_block *sb)
2342 s64 free_blocks, dirty_blocks;
2343 struct ext4_sb_info *sbi = EXT4_SB(sb);
2346 * switch to non delalloc mode if we are running low
2347 * on free block. The free block accounting via percpu
2348 * counters can get slightly wrong with percpu_counter_batch getting
2349 * accumulated on each CPU without updating global counters
2350 * Delalloc need an accurate free block accounting. So switch
2351 * to non delalloc when we are near to error range.
2353 free_blocks = EXT4_C2B(sbi,
2354 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2355 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2356 if (2 * free_blocks < 3 * dirty_blocks ||
2357 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2359 * free block count is less than 150% of dirty blocks
2360 * or free blocks is less than watermark
2365 * Even if we don't switch but are nearing capacity,
2366 * start pushing delalloc when 1/2 of free blocks are dirty.
2368 if (free_blocks < 2 * dirty_blocks)
2369 writeback_inodes_sb_if_idle(sb);
2374 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2375 loff_t pos, unsigned len, unsigned flags,
2376 struct page **pagep, void **fsdata)
2378 int ret, retries = 0;
2381 struct inode *inode = mapping->host;
2385 index = pos >> PAGE_CACHE_SHIFT;
2387 if (ext4_nonda_switch(inode->i_sb)) {
2388 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2389 return ext4_write_begin(file, mapping, pos,
2390 len, flags, pagep, fsdata);
2392 *fsdata = (void *)0;
2393 trace_ext4_da_write_begin(inode, pos, len, flags);
2396 * With delayed allocation, we don't log the i_disksize update
2397 * if there is delayed block allocation. But we still need
2398 * to journalling the i_disksize update if writes to the end
2399 * of file which has an already mapped buffer.
2401 handle = ext4_journal_start(inode, 1);
2402 if (IS_ERR(handle)) {
2403 ret = PTR_ERR(handle);
2406 /* We cannot recurse into the filesystem as the transaction is already
2408 flags |= AOP_FLAG_NOFS;
2410 page = grab_cache_page_write_begin(mapping, index, flags);
2412 ext4_journal_stop(handle);
2418 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2421 ext4_journal_stop(handle);
2422 page_cache_release(page);
2424 * block_write_begin may have instantiated a few blocks
2425 * outside i_size. Trim these off again. Don't need
2426 * i_size_read because we hold i_mutex.
2428 if (pos + len > inode->i_size)
2429 ext4_truncate_failed_write(inode);
2431 page_len = pos & (PAGE_CACHE_SIZE - 1);
2433 ret = ext4_discard_partial_page_buffers_no_lock(handle,
2434 inode, page, pos - page_len, page_len,
2435 EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED);
2439 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2446 * Check if we should update i_disksize
2447 * when write to the end of file but not require block allocation
2449 static int ext4_da_should_update_i_disksize(struct page *page,
2450 unsigned long offset)
2452 struct buffer_head *bh;
2453 struct inode *inode = page->mapping->host;
2457 bh = page_buffers(page);
2458 idx = offset >> inode->i_blkbits;
2460 for (i = 0; i < idx; i++)
2461 bh = bh->b_this_page;
2463 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2468 static int ext4_da_write_end(struct file *file,
2469 struct address_space *mapping,
2470 loff_t pos, unsigned len, unsigned copied,
2471 struct page *page, void *fsdata)
2473 struct inode *inode = mapping->host;
2475 handle_t *handle = ext4_journal_current_handle();
2477 unsigned long start, end;
2478 int write_mode = (int)(unsigned long)fsdata;
2481 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2482 if (ext4_should_order_data(inode)) {
2483 return ext4_ordered_write_end(file, mapping, pos,
2484 len, copied, page, fsdata);
2485 } else if (ext4_should_writeback_data(inode)) {
2486 return ext4_writeback_write_end(file, mapping, pos,
2487 len, copied, page, fsdata);
2493 trace_ext4_da_write_end(inode, pos, len, copied);
2494 start = pos & (PAGE_CACHE_SIZE - 1);
2495 end = start + copied - 1;
2498 * generic_write_end() will run mark_inode_dirty() if i_size
2499 * changes. So let's piggyback the i_disksize mark_inode_dirty
2503 new_i_size = pos + copied;
2504 if (new_i_size > EXT4_I(inode)->i_disksize) {
2505 if (ext4_da_should_update_i_disksize(page, end)) {
2506 down_write(&EXT4_I(inode)->i_data_sem);
2507 if (new_i_size > EXT4_I(inode)->i_disksize) {
2509 * Updating i_disksize when extending file
2510 * without needing block allocation
2512 if (ext4_should_order_data(inode))
2513 ret = ext4_jbd2_file_inode(handle,
2516 EXT4_I(inode)->i_disksize = new_i_size;
2518 up_write(&EXT4_I(inode)->i_data_sem);
2519 /* We need to mark inode dirty even if
2520 * new_i_size is less that inode->i_size
2521 * bu greater than i_disksize.(hint delalloc)
2523 ext4_mark_inode_dirty(handle, inode);
2526 ret2 = generic_write_end(file, mapping, pos, len, copied,
2529 page_len = PAGE_CACHE_SIZE -
2530 ((pos + copied - 1) & (PAGE_CACHE_SIZE - 1));
2533 ret = ext4_discard_partial_page_buffers_no_lock(handle,
2534 inode, page, pos + copied - 1, page_len,
2535 EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED);
2541 ret2 = ext4_journal_stop(handle);
2545 return ret ? ret : copied;
2548 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2551 * Drop reserved blocks
2553 BUG_ON(!PageLocked(page));
2554 if (!page_has_buffers(page))
2557 ext4_da_page_release_reservation(page, offset);
2560 ext4_invalidatepage(page, offset);
2566 * Force all delayed allocation blocks to be allocated for a given inode.
2568 int ext4_alloc_da_blocks(struct inode *inode)
2570 trace_ext4_alloc_da_blocks(inode);
2572 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2573 !EXT4_I(inode)->i_reserved_meta_blocks)
2577 * We do something simple for now. The filemap_flush() will
2578 * also start triggering a write of the data blocks, which is
2579 * not strictly speaking necessary (and for users of
2580 * laptop_mode, not even desirable). However, to do otherwise
2581 * would require replicating code paths in:
2583 * ext4_da_writepages() ->
2584 * write_cache_pages() ---> (via passed in callback function)
2585 * __mpage_da_writepage() -->
2586 * mpage_add_bh_to_extent()
2587 * mpage_da_map_blocks()
2589 * The problem is that write_cache_pages(), located in
2590 * mm/page-writeback.c, marks pages clean in preparation for
2591 * doing I/O, which is not desirable if we're not planning on
2594 * We could call write_cache_pages(), and then redirty all of
2595 * the pages by calling redirty_page_for_writepage() but that
2596 * would be ugly in the extreme. So instead we would need to
2597 * replicate parts of the code in the above functions,
2598 * simplifying them because we wouldn't actually intend to
2599 * write out the pages, but rather only collect contiguous
2600 * logical block extents, call the multi-block allocator, and
2601 * then update the buffer heads with the block allocations.
2603 * For now, though, we'll cheat by calling filemap_flush(),
2604 * which will map the blocks, and start the I/O, but not
2605 * actually wait for the I/O to complete.
2607 return filemap_flush(inode->i_mapping);
2611 * bmap() is special. It gets used by applications such as lilo and by
2612 * the swapper to find the on-disk block of a specific piece of data.
2614 * Naturally, this is dangerous if the block concerned is still in the
2615 * journal. If somebody makes a swapfile on an ext4 data-journaling
2616 * filesystem and enables swap, then they may get a nasty shock when the
2617 * data getting swapped to that swapfile suddenly gets overwritten by
2618 * the original zero's written out previously to the journal and
2619 * awaiting writeback in the kernel's buffer cache.
2621 * So, if we see any bmap calls here on a modified, data-journaled file,
2622 * take extra steps to flush any blocks which might be in the cache.
2624 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2626 struct inode *inode = mapping->host;
2630 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2631 test_opt(inode->i_sb, DELALLOC)) {
2633 * With delalloc we want to sync the file
2634 * so that we can make sure we allocate
2637 filemap_write_and_wait(mapping);
2640 if (EXT4_JOURNAL(inode) &&
2641 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2643 * This is a REALLY heavyweight approach, but the use of
2644 * bmap on dirty files is expected to be extremely rare:
2645 * only if we run lilo or swapon on a freshly made file
2646 * do we expect this to happen.
2648 * (bmap requires CAP_SYS_RAWIO so this does not
2649 * represent an unprivileged user DOS attack --- we'd be
2650 * in trouble if mortal users could trigger this path at
2653 * NB. EXT4_STATE_JDATA is not set on files other than
2654 * regular files. If somebody wants to bmap a directory
2655 * or symlink and gets confused because the buffer
2656 * hasn't yet been flushed to disk, they deserve
2657 * everything they get.
2660 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2661 journal = EXT4_JOURNAL(inode);
2662 jbd2_journal_lock_updates(journal);
2663 err = jbd2_journal_flush(journal);
2664 jbd2_journal_unlock_updates(journal);
2670 return generic_block_bmap(mapping, block, ext4_get_block);
2673 static int ext4_readpage(struct file *file, struct page *page)
2675 trace_ext4_readpage(page);
2676 return mpage_readpage(page, ext4_get_block);
2680 ext4_readpages(struct file *file, struct address_space *mapping,
2681 struct list_head *pages, unsigned nr_pages)
2683 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2686 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
2688 struct buffer_head *head, *bh;
2689 unsigned int curr_off = 0;
2691 if (!page_has_buffers(page))
2693 head = bh = page_buffers(page);
2695 if (offset <= curr_off && test_clear_buffer_uninit(bh)
2697 ext4_free_io_end(bh->b_private);
2698 bh->b_private = NULL;
2699 bh->b_end_io = NULL;
2701 curr_off = curr_off + bh->b_size;
2702 bh = bh->b_this_page;
2703 } while (bh != head);
2706 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2708 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2710 trace_ext4_invalidatepage(page, offset);
2713 * free any io_end structure allocated for buffers to be discarded
2715 if (ext4_should_dioread_nolock(page->mapping->host))
2716 ext4_invalidatepage_free_endio(page, offset);
2718 * If it's a full truncate we just forget about the pending dirtying
2721 ClearPageChecked(page);
2724 jbd2_journal_invalidatepage(journal, page, offset);
2726 block_invalidatepage(page, offset);
2729 static int ext4_releasepage(struct page *page, gfp_t wait)
2731 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2733 trace_ext4_releasepage(page);
2735 WARN_ON(PageChecked(page));
2736 if (!page_has_buffers(page))
2739 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2741 return try_to_free_buffers(page);
2745 * ext4_get_block used when preparing for a DIO write or buffer write.
2746 * We allocate an uinitialized extent if blocks haven't been allocated.
2747 * The extent will be converted to initialized after the IO is complete.
2749 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
2750 struct buffer_head *bh_result, int create)
2752 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2753 inode->i_ino, create);
2754 return _ext4_get_block(inode, iblock, bh_result,
2755 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2758 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2759 ssize_t size, void *private, int ret,
2762 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2763 ext4_io_end_t *io_end = iocb->private;
2764 struct workqueue_struct *wq;
2765 unsigned long flags;
2766 struct ext4_inode_info *ei;
2768 /* if not async direct IO or dio with 0 bytes write, just return */
2769 if (!io_end || !size)
2772 ext_debug("ext4_end_io_dio(): io_end 0x%p"
2773 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
2774 iocb->private, io_end->inode->i_ino, iocb, offset,
2777 /* if not aio dio with unwritten extents, just free io and return */
2778 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2779 ext4_free_io_end(io_end);
2780 iocb->private = NULL;
2783 aio_complete(iocb, ret, 0);
2784 inode_dio_done(inode);
2788 io_end->offset = offset;
2789 io_end->size = size;
2791 io_end->iocb = iocb;
2792 io_end->result = ret;
2794 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
2796 /* Add the io_end to per-inode completed aio dio list*/
2797 ei = EXT4_I(io_end->inode);
2798 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
2799 list_add_tail(&io_end->list, &ei->i_completed_io_list);
2800 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
2802 /* queue the work to convert unwritten extents to written */
2803 queue_work(wq, &io_end->work);
2804 iocb->private = NULL;
2806 /* XXX: probably should move into the real I/O completion handler */
2807 inode_dio_done(inode);
2810 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
2812 ext4_io_end_t *io_end = bh->b_private;
2813 struct workqueue_struct *wq;
2814 struct inode *inode;
2815 unsigned long flags;
2817 if (!test_clear_buffer_uninit(bh) || !io_end)
2820 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2821 printk("sb umounted, discard end_io request for inode %lu\n",
2822 io_end->inode->i_ino);
2823 ext4_free_io_end(io_end);
2828 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2829 * but being more careful is always safe for the future change.
2831 inode = io_end->inode;
2832 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2833 io_end->flag |= EXT4_IO_END_UNWRITTEN;
2834 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
2837 /* Add the io_end to per-inode completed io list*/
2838 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
2839 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
2840 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
2842 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
2843 /* queue the work to convert unwritten extents to written */
2844 queue_work(wq, &io_end->work);
2846 bh->b_private = NULL;
2847 bh->b_end_io = NULL;
2848 clear_buffer_uninit(bh);
2849 end_buffer_async_write(bh, uptodate);
2852 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2854 ext4_io_end_t *io_end;
2855 struct page *page = bh->b_page;
2856 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2857 size_t size = bh->b_size;
2860 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2862 pr_warn_ratelimited("%s: allocation fail\n", __func__);
2866 io_end->offset = offset;
2867 io_end->size = size;
2869 * We need to hold a reference to the page to make sure it
2870 * doesn't get evicted before ext4_end_io_work() has a chance
2871 * to convert the extent from written to unwritten.
2873 io_end->page = page;
2874 get_page(io_end->page);
2876 bh->b_private = io_end;
2877 bh->b_end_io = ext4_end_io_buffer_write;
2882 * For ext4 extent files, ext4 will do direct-io write to holes,
2883 * preallocated extents, and those write extend the file, no need to
2884 * fall back to buffered IO.
2886 * For holes, we fallocate those blocks, mark them as uninitialized
2887 * If those blocks were preallocated, we mark sure they are splited, but
2888 * still keep the range to write as uninitialized.
2890 * The unwrritten extents will be converted to written when DIO is completed.
2891 * For async direct IO, since the IO may still pending when return, we
2892 * set up an end_io call back function, which will do the conversion
2893 * when async direct IO completed.
2895 * If the O_DIRECT write will extend the file then add this inode to the
2896 * orphan list. So recovery will truncate it back to the original size
2897 * if the machine crashes during the write.
2900 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2901 const struct iovec *iov, loff_t offset,
2902 unsigned long nr_segs)
2904 struct file *file = iocb->ki_filp;
2905 struct inode *inode = file->f_mapping->host;
2907 size_t count = iov_length(iov, nr_segs);
2909 loff_t final_size = offset + count;
2910 if (rw == WRITE && final_size <= inode->i_size) {
2912 * We could direct write to holes and fallocate.
2914 * Allocated blocks to fill the hole are marked as uninitialized
2915 * to prevent parallel buffered read to expose the stale data
2916 * before DIO complete the data IO.
2918 * As to previously fallocated extents, ext4 get_block
2919 * will just simply mark the buffer mapped but still
2920 * keep the extents uninitialized.
2922 * for non AIO case, we will convert those unwritten extents
2923 * to written after return back from blockdev_direct_IO.
2925 * for async DIO, the conversion needs to be defered when
2926 * the IO is completed. The ext4 end_io callback function
2927 * will be called to take care of the conversion work.
2928 * Here for async case, we allocate an io_end structure to
2931 iocb->private = NULL;
2932 EXT4_I(inode)->cur_aio_dio = NULL;
2933 if (!is_sync_kiocb(iocb)) {
2934 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
2938 * we save the io structure for current async
2939 * direct IO, so that later ext4_map_blocks()
2940 * could flag the io structure whether there
2941 * is a unwritten extents needs to be converted
2942 * when IO is completed.
2944 EXT4_I(inode)->cur_aio_dio = iocb->private;
2947 ret = __blockdev_direct_IO(rw, iocb, inode,
2948 inode->i_sb->s_bdev, iov,
2950 ext4_get_block_write,
2953 DIO_LOCKING | DIO_SKIP_HOLES);
2955 EXT4_I(inode)->cur_aio_dio = NULL;
2957 * The io_end structure takes a reference to the inode,
2958 * that structure needs to be destroyed and the
2959 * reference to the inode need to be dropped, when IO is
2960 * complete, even with 0 byte write, or failed.
2962 * In the successful AIO DIO case, the io_end structure will be
2963 * desctroyed and the reference to the inode will be dropped
2964 * after the end_io call back function is called.
2966 * In the case there is 0 byte write, or error case, since
2967 * VFS direct IO won't invoke the end_io call back function,
2968 * we need to free the end_io structure here.
2970 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
2971 ext4_free_io_end(iocb->private);
2972 iocb->private = NULL;
2973 } else if (ret > 0 && ext4_test_inode_state(inode,
2974 EXT4_STATE_DIO_UNWRITTEN)) {
2977 * for non AIO case, since the IO is already
2978 * completed, we could do the conversion right here
2980 err = ext4_convert_unwritten_extents(inode,
2984 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
2989 /* for write the the end of file case, we fall back to old way */
2990 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
2993 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
2994 const struct iovec *iov, loff_t offset,
2995 unsigned long nr_segs)
2997 struct file *file = iocb->ki_filp;
2998 struct inode *inode = file->f_mapping->host;
3002 * If we are doing data journalling we don't support O_DIRECT
3004 if (ext4_should_journal_data(inode))
3007 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3008 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3009 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3011 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3012 trace_ext4_direct_IO_exit(inode, offset,
3013 iov_length(iov, nr_segs), rw, ret);
3018 * Pages can be marked dirty completely asynchronously from ext4's journalling
3019 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3020 * much here because ->set_page_dirty is called under VFS locks. The page is
3021 * not necessarily locked.
3023 * We cannot just dirty the page and leave attached buffers clean, because the
3024 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3025 * or jbddirty because all the journalling code will explode.
3027 * So what we do is to mark the page "pending dirty" and next time writepage
3028 * is called, propagate that into the buffers appropriately.
3030 static int ext4_journalled_set_page_dirty(struct page *page)
3032 SetPageChecked(page);
3033 return __set_page_dirty_nobuffers(page);
3036 static const struct address_space_operations ext4_ordered_aops = {
3037 .readpage = ext4_readpage,
3038 .readpages = ext4_readpages,
3039 .writepage = ext4_writepage,
3040 .write_begin = ext4_write_begin,
3041 .write_end = ext4_ordered_write_end,
3043 .invalidatepage = ext4_invalidatepage,
3044 .releasepage = ext4_releasepage,
3045 .direct_IO = ext4_direct_IO,
3046 .migratepage = buffer_migrate_page,
3047 .is_partially_uptodate = block_is_partially_uptodate,
3048 .error_remove_page = generic_error_remove_page,
3051 static const struct address_space_operations ext4_writeback_aops = {
3052 .readpage = ext4_readpage,
3053 .readpages = ext4_readpages,
3054 .writepage = ext4_writepage,
3055 .write_begin = ext4_write_begin,
3056 .write_end = ext4_writeback_write_end,
3058 .invalidatepage = ext4_invalidatepage,
3059 .releasepage = ext4_releasepage,
3060 .direct_IO = ext4_direct_IO,
3061 .migratepage = buffer_migrate_page,
3062 .is_partially_uptodate = block_is_partially_uptodate,
3063 .error_remove_page = generic_error_remove_page,
3066 static const struct address_space_operations ext4_journalled_aops = {
3067 .readpage = ext4_readpage,
3068 .readpages = ext4_readpages,
3069 .writepage = ext4_writepage,
3070 .write_begin = ext4_write_begin,
3071 .write_end = ext4_journalled_write_end,
3072 .set_page_dirty = ext4_journalled_set_page_dirty,
3074 .invalidatepage = ext4_invalidatepage,
3075 .releasepage = ext4_releasepage,
3076 .direct_IO = ext4_direct_IO,
3077 .is_partially_uptodate = block_is_partially_uptodate,
3078 .error_remove_page = generic_error_remove_page,
3081 static const struct address_space_operations ext4_da_aops = {
3082 .readpage = ext4_readpage,
3083 .readpages = ext4_readpages,
3084 .writepage = ext4_writepage,
3085 .writepages = ext4_da_writepages,
3086 .write_begin = ext4_da_write_begin,
3087 .write_end = ext4_da_write_end,
3089 .invalidatepage = ext4_da_invalidatepage,
3090 .releasepage = ext4_releasepage,
3091 .direct_IO = ext4_direct_IO,
3092 .migratepage = buffer_migrate_page,
3093 .is_partially_uptodate = block_is_partially_uptodate,
3094 .error_remove_page = generic_error_remove_page,
3097 void ext4_set_aops(struct inode *inode)
3099 if (ext4_should_order_data(inode) &&
3100 test_opt(inode->i_sb, DELALLOC))
3101 inode->i_mapping->a_ops = &ext4_da_aops;
3102 else if (ext4_should_order_data(inode))
3103 inode->i_mapping->a_ops = &ext4_ordered_aops;
3104 else if (ext4_should_writeback_data(inode) &&
3105 test_opt(inode->i_sb, DELALLOC))
3106 inode->i_mapping->a_ops = &ext4_da_aops;
3107 else if (ext4_should_writeback_data(inode))
3108 inode->i_mapping->a_ops = &ext4_writeback_aops;
3110 inode->i_mapping->a_ops = &ext4_journalled_aops;
3115 * ext4_discard_partial_page_buffers()
3116 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3117 * This function finds and locks the page containing the offset
3118 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3119 * Calling functions that already have the page locked should call
3120 * ext4_discard_partial_page_buffers_no_lock directly.
3122 int ext4_discard_partial_page_buffers(handle_t *handle,
3123 struct address_space *mapping, loff_t from,
3124 loff_t length, int flags)
3126 struct inode *inode = mapping->host;
3130 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3131 mapping_gfp_mask(mapping) & ~__GFP_FS);
3135 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3136 from, length, flags);
3139 page_cache_release(page);
3144 * ext4_discard_partial_page_buffers_no_lock()
3145 * Zeros a page range of length 'length' starting from offset 'from'.
3146 * Buffer heads that correspond to the block aligned regions of the
3147 * zeroed range will be unmapped. Unblock aligned regions
3148 * will have the corresponding buffer head mapped if needed so that
3149 * that region of the page can be updated with the partial zero out.
3151 * This function assumes that the page has already been locked. The
3152 * The range to be discarded must be contained with in the given page.
3153 * If the specified range exceeds the end of the page it will be shortened
3154 * to the end of the page that corresponds to 'from'. This function is
3155 * appropriate for updating a page and it buffer heads to be unmapped and
3156 * zeroed for blocks that have been either released, or are going to be
3159 * handle: The journal handle
3160 * inode: The files inode
3161 * page: A locked page that contains the offset "from"
3162 * from: The starting byte offset (from the begining of the file)
3163 * to begin discarding
3164 * len: The length of bytes to discard
3165 * flags: Optional flags that may be used:
3167 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3168 * Only zero the regions of the page whose buffer heads
3169 * have already been unmapped. This flag is appropriate
3170 * for updateing the contents of a page whose blocks may
3171 * have already been released, and we only want to zero
3172 * out the regions that correspond to those released blocks.
3174 * Returns zero on sucess or negative on failure.
3176 int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3177 struct inode *inode, struct page *page, loff_t from,
3178 loff_t length, int flags)
3180 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3181 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3182 unsigned int blocksize, max, pos;
3183 unsigned int end_of_block, range_to_discard;
3185 struct buffer_head *bh;
3188 blocksize = inode->i_sb->s_blocksize;
3189 max = PAGE_CACHE_SIZE - offset;
3191 if (index != page->index)
3195 * correct length if it does not fall between
3196 * 'from' and the end of the page
3198 if (length > max || length < 0)
3201 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3203 if (!page_has_buffers(page)) {
3205 * If the range to be discarded covers a partial block
3206 * we need to get the page buffers. This is because
3207 * partial blocks cannot be released and the page needs
3208 * to be updated with the contents of the block before
3209 * we write the zeros on top of it.
3211 if (!(from & (blocksize - 1)) ||
3212 !((from + length) & (blocksize - 1))) {
3213 create_empty_buffers(page, blocksize, 0);
3216 * If there are no partial blocks,
3217 * there is nothing to update,
3218 * so we can return now
3224 /* Find the buffer that contains "offset" */
3225 bh = page_buffers(page);
3227 while (offset >= pos) {
3228 bh = bh->b_this_page;
3234 while (pos < offset + length) {
3237 /* The length of space left to zero and unmap */
3238 range_to_discard = offset + length - pos;
3240 /* The length of space until the end of the block */
3241 end_of_block = blocksize - (pos & (blocksize-1));
3244 * Do not unmap or zero past end of block
3245 * for this buffer head
3247 if (range_to_discard > end_of_block)
3248 range_to_discard = end_of_block;
3252 * Skip this buffer head if we are only zeroing unampped
3253 * regions of the page
3255 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3259 /* If the range is block aligned, unmap */
3260 if (range_to_discard == blocksize) {
3261 clear_buffer_dirty(bh);
3263 clear_buffer_mapped(bh);
3264 clear_buffer_req(bh);
3265 clear_buffer_new(bh);
3266 clear_buffer_delay(bh);
3267 clear_buffer_unwritten(bh);
3268 clear_buffer_uptodate(bh);
3269 zero_user(page, pos, range_to_discard);
3270 BUFFER_TRACE(bh, "Buffer discarded");
3275 * If this block is not completely contained in the range
3276 * to be discarded, then it is not going to be released. Because
3277 * we need to keep this block, we need to make sure this part
3278 * of the page is uptodate before we modify it by writeing
3279 * partial zeros on it.
3281 if (!buffer_mapped(bh)) {
3283 * Buffer head must be mapped before we can read
3286 BUFFER_TRACE(bh, "unmapped");
3287 ext4_get_block(inode, iblock, bh, 0);
3288 /* unmapped? It's a hole - nothing to do */
3289 if (!buffer_mapped(bh)) {
3290 BUFFER_TRACE(bh, "still unmapped");
3295 /* Ok, it's mapped. Make sure it's up-to-date */
3296 if (PageUptodate(page))
3297 set_buffer_uptodate(bh);
3299 if (!buffer_uptodate(bh)) {
3301 ll_rw_block(READ, 1, &bh);
3303 /* Uhhuh. Read error. Complain and punt.*/
3304 if (!buffer_uptodate(bh))
3308 if (ext4_should_journal_data(inode)) {
3309 BUFFER_TRACE(bh, "get write access");
3310 err = ext4_journal_get_write_access(handle, bh);
3315 zero_user(page, pos, range_to_discard);
3318 if (ext4_should_journal_data(inode)) {
3319 err = ext4_handle_dirty_metadata(handle, inode, bh);
3321 mark_buffer_dirty(bh);
3323 BUFFER_TRACE(bh, "Partial buffer zeroed");
3325 bh = bh->b_this_page;
3327 pos += range_to_discard;
3334 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3335 * up to the end of the block which corresponds to `from'.
3336 * This required during truncate. We need to physically zero the tail end
3337 * of that block so it doesn't yield old data if the file is later grown.
3339 int ext4_block_truncate_page(handle_t *handle,
3340 struct address_space *mapping, loff_t from)
3342 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3345 struct inode *inode = mapping->host;
3347 blocksize = inode->i_sb->s_blocksize;
3348 length = blocksize - (offset & (blocksize - 1));
3350 return ext4_block_zero_page_range(handle, mapping, from, length);
3354 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3355 * starting from file offset 'from'. The range to be zero'd must
3356 * be contained with in one block. If the specified range exceeds
3357 * the end of the block it will be shortened to end of the block
3358 * that cooresponds to 'from'
3360 int ext4_block_zero_page_range(handle_t *handle,
3361 struct address_space *mapping, loff_t from, loff_t length)
3363 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3364 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3365 unsigned blocksize, max, pos;
3367 struct inode *inode = mapping->host;
3368 struct buffer_head *bh;
3372 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3373 mapping_gfp_mask(mapping) & ~__GFP_FS);
3377 blocksize = inode->i_sb->s_blocksize;
3378 max = blocksize - (offset & (blocksize - 1));
3381 * correct length if it does not fall between
3382 * 'from' and the end of the block
3384 if (length > max || length < 0)
3387 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3389 if (!page_has_buffers(page))
3390 create_empty_buffers(page, blocksize, 0);
3392 /* Find the buffer that contains "offset" */
3393 bh = page_buffers(page);
3395 while (offset >= pos) {
3396 bh = bh->b_this_page;
3402 if (buffer_freed(bh)) {
3403 BUFFER_TRACE(bh, "freed: skip");
3407 if (!buffer_mapped(bh)) {
3408 BUFFER_TRACE(bh, "unmapped");
3409 ext4_get_block(inode, iblock, bh, 0);
3410 /* unmapped? It's a hole - nothing to do */
3411 if (!buffer_mapped(bh)) {
3412 BUFFER_TRACE(bh, "still unmapped");
3417 /* Ok, it's mapped. Make sure it's up-to-date */
3418 if (PageUptodate(page))
3419 set_buffer_uptodate(bh);
3421 if (!buffer_uptodate(bh)) {
3423 ll_rw_block(READ, 1, &bh);
3425 /* Uhhuh. Read error. Complain and punt. */
3426 if (!buffer_uptodate(bh))
3430 if (ext4_should_journal_data(inode)) {
3431 BUFFER_TRACE(bh, "get write access");
3432 err = ext4_journal_get_write_access(handle, bh);
3437 zero_user(page, offset, length);
3439 BUFFER_TRACE(bh, "zeroed end of block");
3442 if (ext4_should_journal_data(inode)) {
3443 err = ext4_handle_dirty_metadata(handle, inode, bh);
3445 mark_buffer_dirty(bh);
3449 page_cache_release(page);
3453 int ext4_can_truncate(struct inode *inode)
3455 if (S_ISREG(inode->i_mode))
3457 if (S_ISDIR(inode->i_mode))
3459 if (S_ISLNK(inode->i_mode))
3460 return !ext4_inode_is_fast_symlink(inode);
3465 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3466 * associated with the given offset and length
3468 * @inode: File inode
3469 * @offset: The offset where the hole will begin
3470 * @len: The length of the hole
3472 * Returns: 0 on sucess or negative on failure
3475 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3477 struct inode *inode = file->f_path.dentry->d_inode;
3478 if (!S_ISREG(inode->i_mode))
3481 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3482 /* TODO: Add support for non extent hole punching */
3486 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3487 /* TODO: Add support for bigalloc file systems */
3491 return ext4_ext_punch_hole(file, offset, length);
3497 * We block out ext4_get_block() block instantiations across the entire
3498 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3499 * simultaneously on behalf of the same inode.
3501 * As we work through the truncate and commmit bits of it to the journal there
3502 * is one core, guiding principle: the file's tree must always be consistent on
3503 * disk. We must be able to restart the truncate after a crash.
3505 * The file's tree may be transiently inconsistent in memory (although it
3506 * probably isn't), but whenever we close off and commit a journal transaction,
3507 * the contents of (the filesystem + the journal) must be consistent and
3508 * restartable. It's pretty simple, really: bottom up, right to left (although
3509 * left-to-right works OK too).
3511 * Note that at recovery time, journal replay occurs *before* the restart of
3512 * truncate against the orphan inode list.
3514 * The committed inode has the new, desired i_size (which is the same as
3515 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3516 * that this inode's truncate did not complete and it will again call
3517 * ext4_truncate() to have another go. So there will be instantiated blocks
3518 * to the right of the truncation point in a crashed ext4 filesystem. But
3519 * that's fine - as long as they are linked from the inode, the post-crash
3520 * ext4_truncate() run will find them and release them.
3522 void ext4_truncate(struct inode *inode)
3524 trace_ext4_truncate_enter(inode);
3526 if (!ext4_can_truncate(inode))
3529 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3531 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3532 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3534 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3535 ext4_ext_truncate(inode);
3537 ext4_ind_truncate(inode);
3539 trace_ext4_truncate_exit(inode);
3543 * ext4_get_inode_loc returns with an extra refcount against the inode's
3544 * underlying buffer_head on success. If 'in_mem' is true, we have all
3545 * data in memory that is needed to recreate the on-disk version of this
3548 static int __ext4_get_inode_loc(struct inode *inode,
3549 struct ext4_iloc *iloc, int in_mem)
3551 struct ext4_group_desc *gdp;
3552 struct buffer_head *bh;
3553 struct super_block *sb = inode->i_sb;
3555 int inodes_per_block, inode_offset;
3558 if (!ext4_valid_inum(sb, inode->i_ino))
3561 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3562 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3567 * Figure out the offset within the block group inode table
3569 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3570 inode_offset = ((inode->i_ino - 1) %
3571 EXT4_INODES_PER_GROUP(sb));
3572 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3573 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3575 bh = sb_getblk(sb, block);
3577 EXT4_ERROR_INODE_BLOCK(inode, block,
3578 "unable to read itable block");
3581 if (!buffer_uptodate(bh)) {
3585 * If the buffer has the write error flag, we have failed
3586 * to write out another inode in the same block. In this
3587 * case, we don't have to read the block because we may
3588 * read the old inode data successfully.
3590 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3591 set_buffer_uptodate(bh);
3593 if (buffer_uptodate(bh)) {
3594 /* someone brought it uptodate while we waited */
3600 * If we have all information of the inode in memory and this
3601 * is the only valid inode in the block, we need not read the
3605 struct buffer_head *bitmap_bh;
3608 start = inode_offset & ~(inodes_per_block - 1);
3610 /* Is the inode bitmap in cache? */
3611 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3616 * If the inode bitmap isn't in cache then the
3617 * optimisation may end up performing two reads instead
3618 * of one, so skip it.
3620 if (!buffer_uptodate(bitmap_bh)) {
3624 for (i = start; i < start + inodes_per_block; i++) {
3625 if (i == inode_offset)
3627 if (ext4_test_bit(i, bitmap_bh->b_data))
3631 if (i == start + inodes_per_block) {
3632 /* all other inodes are free, so skip I/O */
3633 memset(bh->b_data, 0, bh->b_size);
3634 set_buffer_uptodate(bh);
3642 * If we need to do any I/O, try to pre-readahead extra
3643 * blocks from the inode table.
3645 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3646 ext4_fsblk_t b, end, table;
3649 table = ext4_inode_table(sb, gdp);
3650 /* s_inode_readahead_blks is always a power of 2 */
3651 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3654 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3655 num = EXT4_INODES_PER_GROUP(sb);
3656 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3657 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3658 num -= ext4_itable_unused_count(sb, gdp);
3659 table += num / inodes_per_block;
3663 sb_breadahead(sb, b++);
3667 * There are other valid inodes in the buffer, this inode
3668 * has in-inode xattrs, or we don't have this inode in memory.
3669 * Read the block from disk.
3671 trace_ext4_load_inode(inode);
3673 bh->b_end_io = end_buffer_read_sync;
3674 submit_bh(READ_META, bh);
3676 if (!buffer_uptodate(bh)) {
3677 EXT4_ERROR_INODE_BLOCK(inode, block,
3678 "unable to read itable block");
3688 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3690 /* We have all inode data except xattrs in memory here. */
3691 return __ext4_get_inode_loc(inode, iloc,
3692 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3695 void ext4_set_inode_flags(struct inode *inode)
3697 unsigned int flags = EXT4_I(inode)->i_flags;
3699 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3700 if (flags & EXT4_SYNC_FL)
3701 inode->i_flags |= S_SYNC;
3702 if (flags & EXT4_APPEND_FL)
3703 inode->i_flags |= S_APPEND;
3704 if (flags & EXT4_IMMUTABLE_FL)
3705 inode->i_flags |= S_IMMUTABLE;
3706 if (flags & EXT4_NOATIME_FL)
3707 inode->i_flags |= S_NOATIME;
3708 if (flags & EXT4_DIRSYNC_FL)
3709 inode->i_flags |= S_DIRSYNC;
3712 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3713 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3715 unsigned int vfs_fl;
3716 unsigned long old_fl, new_fl;
3719 vfs_fl = ei->vfs_inode.i_flags;
3720 old_fl = ei->i_flags;
3721 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3722 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3724 if (vfs_fl & S_SYNC)
3725 new_fl |= EXT4_SYNC_FL;
3726 if (vfs_fl & S_APPEND)
3727 new_fl |= EXT4_APPEND_FL;
3728 if (vfs_fl & S_IMMUTABLE)
3729 new_fl |= EXT4_IMMUTABLE_FL;
3730 if (vfs_fl & S_NOATIME)
3731 new_fl |= EXT4_NOATIME_FL;
3732 if (vfs_fl & S_DIRSYNC)
3733 new_fl |= EXT4_DIRSYNC_FL;
3734 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3737 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3738 struct ext4_inode_info *ei)
3741 struct inode *inode = &(ei->vfs_inode);
3742 struct super_block *sb = inode->i_sb;
3744 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3745 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3746 /* we are using combined 48 bit field */
3747 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3748 le32_to_cpu(raw_inode->i_blocks_lo);
3749 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3750 /* i_blocks represent file system block size */
3751 return i_blocks << (inode->i_blkbits - 9);
3756 return le32_to_cpu(raw_inode->i_blocks_lo);
3760 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3762 struct ext4_iloc iloc;
3763 struct ext4_inode *raw_inode;
3764 struct ext4_inode_info *ei;
3765 struct inode *inode;
3766 journal_t *journal = EXT4_SB(sb)->s_journal;
3770 inode = iget_locked(sb, ino);
3772 return ERR_PTR(-ENOMEM);
3773 if (!(inode->i_state & I_NEW))
3779 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3782 raw_inode = ext4_raw_inode(&iloc);
3783 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3784 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3785 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3786 if (!(test_opt(inode->i_sb, NO_UID32))) {
3787 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3788 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3790 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
3792 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3793 ei->i_dir_start_lookup = 0;
3794 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3795 /* We now have enough fields to check if the inode was active or not.
3796 * This is needed because nfsd might try to access dead inodes
3797 * the test is that same one that e2fsck uses
3798 * NeilBrown 1999oct15
3800 if (inode->i_nlink == 0) {
3801 if (inode->i_mode == 0 ||
3802 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3803 /* this inode is deleted */
3807 /* The only unlinked inodes we let through here have
3808 * valid i_mode and are being read by the orphan
3809 * recovery code: that's fine, we're about to complete
3810 * the process of deleting those. */
3812 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3813 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3814 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3815 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3817 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3818 inode->i_size = ext4_isize(raw_inode);
3819 ei->i_disksize = inode->i_size;
3821 ei->i_reserved_quota = 0;
3823 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3824 ei->i_block_group = iloc.block_group;
3825 ei->i_last_alloc_group = ~0;
3827 * NOTE! The in-memory inode i_data array is in little-endian order
3828 * even on big-endian machines: we do NOT byteswap the block numbers!
3830 for (block = 0; block < EXT4_N_BLOCKS; block++)
3831 ei->i_data[block] = raw_inode->i_block[block];
3832 INIT_LIST_HEAD(&ei->i_orphan);
3835 * Set transaction id's of transactions that have to be committed
3836 * to finish f[data]sync. We set them to currently running transaction
3837 * as we cannot be sure that the inode or some of its metadata isn't
3838 * part of the transaction - the inode could have been reclaimed and
3839 * now it is reread from disk.
3842 transaction_t *transaction;
3845 read_lock(&journal->j_state_lock);
3846 if (journal->j_running_transaction)
3847 transaction = journal->j_running_transaction;
3849 transaction = journal->j_committing_transaction;
3851 tid = transaction->t_tid;
3853 tid = journal->j_commit_sequence;
3854 read_unlock(&journal->j_state_lock);
3855 ei->i_sync_tid = tid;
3856 ei->i_datasync_tid = tid;
3859 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3860 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3861 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3862 EXT4_INODE_SIZE(inode->i_sb)) {
3866 if (ei->i_extra_isize == 0) {
3867 /* The extra space is currently unused. Use it. */
3868 ei->i_extra_isize = sizeof(struct ext4_inode) -
3869 EXT4_GOOD_OLD_INODE_SIZE;
3871 __le32 *magic = (void *)raw_inode +
3872 EXT4_GOOD_OLD_INODE_SIZE +
3874 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3875 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3878 ei->i_extra_isize = 0;
3880 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3881 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3882 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3883 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3885 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3886 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3887 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3889 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3893 if (ei->i_file_acl &&
3894 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3895 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
3899 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3900 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3901 (S_ISLNK(inode->i_mode) &&
3902 !ext4_inode_is_fast_symlink(inode)))
3903 /* Validate extent which is part of inode */
3904 ret = ext4_ext_check_inode(inode);
3905 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3906 (S_ISLNK(inode->i_mode) &&
3907 !ext4_inode_is_fast_symlink(inode))) {
3908 /* Validate block references which are part of inode */
3909 ret = ext4_ind_check_inode(inode);
3914 if (S_ISREG(inode->i_mode)) {
3915 inode->i_op = &ext4_file_inode_operations;
3916 inode->i_fop = &ext4_file_operations;
3917 ext4_set_aops(inode);
3918 } else if (S_ISDIR(inode->i_mode)) {
3919 inode->i_op = &ext4_dir_inode_operations;
3920 inode->i_fop = &ext4_dir_operations;
3921 } else if (S_ISLNK(inode->i_mode)) {
3922 if (ext4_inode_is_fast_symlink(inode)) {
3923 inode->i_op = &ext4_fast_symlink_inode_operations;
3924 nd_terminate_link(ei->i_data, inode->i_size,
3925 sizeof(ei->i_data) - 1);
3927 inode->i_op = &ext4_symlink_inode_operations;
3928 ext4_set_aops(inode);
3930 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3931 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3932 inode->i_op = &ext4_special_inode_operations;
3933 if (raw_inode->i_block[0])
3934 init_special_inode(inode, inode->i_mode,
3935 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3937 init_special_inode(inode, inode->i_mode,
3938 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3941 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
3945 ext4_set_inode_flags(inode);
3946 unlock_new_inode(inode);
3952 return ERR_PTR(ret);
3955 static int ext4_inode_blocks_set(handle_t *handle,
3956 struct ext4_inode *raw_inode,
3957 struct ext4_inode_info *ei)
3959 struct inode *inode = &(ei->vfs_inode);
3960 u64 i_blocks = inode->i_blocks;
3961 struct super_block *sb = inode->i_sb;
3963 if (i_blocks <= ~0U) {
3965 * i_blocks can be represnted in a 32 bit variable
3966 * as multiple of 512 bytes
3968 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3969 raw_inode->i_blocks_high = 0;
3970 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3973 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
3976 if (i_blocks <= 0xffffffffffffULL) {
3978 * i_blocks can be represented in a 48 bit variable
3979 * as multiple of 512 bytes
3981 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3982 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3983 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3985 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3986 /* i_block is stored in file system block size */
3987 i_blocks = i_blocks >> (inode->i_blkbits - 9);
3988 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3989 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3995 * Post the struct inode info into an on-disk inode location in the
3996 * buffer-cache. This gobbles the caller's reference to the
3997 * buffer_head in the inode location struct.
3999 * The caller must have write access to iloc->bh.
4001 static int ext4_do_update_inode(handle_t *handle,
4002 struct inode *inode,
4003 struct ext4_iloc *iloc)
4005 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4006 struct ext4_inode_info *ei = EXT4_I(inode);
4007 struct buffer_head *bh = iloc->bh;
4008 int err = 0, rc, block;
4010 /* For fields not not tracking in the in-memory inode,
4011 * initialise them to zero for new inodes. */
4012 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4013 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4015 ext4_get_inode_flags(ei);
4016 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4017 if (!(test_opt(inode->i_sb, NO_UID32))) {
4018 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4019 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4021 * Fix up interoperability with old kernels. Otherwise, old inodes get
4022 * re-used with the upper 16 bits of the uid/gid intact
4025 raw_inode->i_uid_high =
4026 cpu_to_le16(high_16_bits(inode->i_uid));
4027 raw_inode->i_gid_high =
4028 cpu_to_le16(high_16_bits(inode->i_gid));
4030 raw_inode->i_uid_high = 0;
4031 raw_inode->i_gid_high = 0;
4034 raw_inode->i_uid_low =
4035 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4036 raw_inode->i_gid_low =
4037 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4038 raw_inode->i_uid_high = 0;
4039 raw_inode->i_gid_high = 0;
4041 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4043 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4044 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4045 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4046 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4048 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4050 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4051 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4052 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4053 cpu_to_le32(EXT4_OS_HURD))
4054 raw_inode->i_file_acl_high =
4055 cpu_to_le16(ei->i_file_acl >> 32);
4056 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4057 ext4_isize_set(raw_inode, ei->i_disksize);
4058 if (ei->i_disksize > 0x7fffffffULL) {
4059 struct super_block *sb = inode->i_sb;
4060 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4061 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4062 EXT4_SB(sb)->s_es->s_rev_level ==
4063 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4064 /* If this is the first large file
4065 * created, add a flag to the superblock.
4067 err = ext4_journal_get_write_access(handle,
4068 EXT4_SB(sb)->s_sbh);
4071 ext4_update_dynamic_rev(sb);
4072 EXT4_SET_RO_COMPAT_FEATURE(sb,
4073 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4075 ext4_handle_sync(handle);
4076 err = ext4_handle_dirty_metadata(handle, NULL,
4077 EXT4_SB(sb)->s_sbh);
4080 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4081 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4082 if (old_valid_dev(inode->i_rdev)) {
4083 raw_inode->i_block[0] =
4084 cpu_to_le32(old_encode_dev(inode->i_rdev));
4085 raw_inode->i_block[1] = 0;
4087 raw_inode->i_block[0] = 0;
4088 raw_inode->i_block[1] =
4089 cpu_to_le32(new_encode_dev(inode->i_rdev));
4090 raw_inode->i_block[2] = 0;
4093 for (block = 0; block < EXT4_N_BLOCKS; block++)
4094 raw_inode->i_block[block] = ei->i_data[block];
4096 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4097 if (ei->i_extra_isize) {
4098 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4099 raw_inode->i_version_hi =
4100 cpu_to_le32(inode->i_version >> 32);
4101 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4104 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4105 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4108 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4110 ext4_update_inode_fsync_trans(handle, inode, 0);
4113 ext4_std_error(inode->i_sb, err);
4118 * ext4_write_inode()
4120 * We are called from a few places:
4122 * - Within generic_file_write() for O_SYNC files.
4123 * Here, there will be no transaction running. We wait for any running
4124 * trasnaction to commit.
4126 * - Within sys_sync(), kupdate and such.
4127 * We wait on commit, if tol to.
4129 * - Within prune_icache() (PF_MEMALLOC == true)
4130 * Here we simply return. We can't afford to block kswapd on the
4133 * In all cases it is actually safe for us to return without doing anything,
4134 * because the inode has been copied into a raw inode buffer in
4135 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4138 * Note that we are absolutely dependent upon all inode dirtiers doing the
4139 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4140 * which we are interested.
4142 * It would be a bug for them to not do this. The code:
4144 * mark_inode_dirty(inode)
4146 * inode->i_size = expr;
4148 * is in error because a kswapd-driven write_inode() could occur while
4149 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4150 * will no longer be on the superblock's dirty inode list.
4152 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4156 if (current->flags & PF_MEMALLOC)
4159 if (EXT4_SB(inode->i_sb)->s_journal) {
4160 if (ext4_journal_current_handle()) {
4161 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4166 if (wbc->sync_mode != WB_SYNC_ALL)
4169 err = ext4_force_commit(inode->i_sb);
4171 struct ext4_iloc iloc;
4173 err = __ext4_get_inode_loc(inode, &iloc, 0);
4176 if (wbc->sync_mode == WB_SYNC_ALL)
4177 sync_dirty_buffer(iloc.bh);
4178 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4179 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4180 "IO error syncing inode");
4191 * Called from notify_change.
4193 * We want to trap VFS attempts to truncate the file as soon as
4194 * possible. In particular, we want to make sure that when the VFS
4195 * shrinks i_size, we put the inode on the orphan list and modify
4196 * i_disksize immediately, so that during the subsequent flushing of
4197 * dirty pages and freeing of disk blocks, we can guarantee that any
4198 * commit will leave the blocks being flushed in an unused state on
4199 * disk. (On recovery, the inode will get truncated and the blocks will
4200 * be freed, so we have a strong guarantee that no future commit will
4201 * leave these blocks visible to the user.)
4203 * Another thing we have to assure is that if we are in ordered mode
4204 * and inode is still attached to the committing transaction, we must
4205 * we start writeout of all the dirty pages which are being truncated.
4206 * This way we are sure that all the data written in the previous
4207 * transaction are already on disk (truncate waits for pages under
4210 * Called with inode->i_mutex down.
4212 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4214 struct inode *inode = dentry->d_inode;
4217 const unsigned int ia_valid = attr->ia_valid;
4219 error = inode_change_ok(inode, attr);
4223 if (is_quota_modification(inode, attr))
4224 dquot_initialize(inode);
4225 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4226 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4229 /* (user+group)*(old+new) structure, inode write (sb,
4230 * inode block, ? - but truncate inode update has it) */
4231 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
4232 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
4233 if (IS_ERR(handle)) {
4234 error = PTR_ERR(handle);
4237 error = dquot_transfer(inode, attr);
4239 ext4_journal_stop(handle);
4242 /* Update corresponding info in inode so that everything is in
4243 * one transaction */
4244 if (attr->ia_valid & ATTR_UID)
4245 inode->i_uid = attr->ia_uid;
4246 if (attr->ia_valid & ATTR_GID)
4247 inode->i_gid = attr->ia_gid;
4248 error = ext4_mark_inode_dirty(handle, inode);
4249 ext4_journal_stop(handle);
4252 if (attr->ia_valid & ATTR_SIZE) {
4253 inode_dio_wait(inode);
4255 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4256 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4258 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4263 if (S_ISREG(inode->i_mode) &&
4264 attr->ia_valid & ATTR_SIZE &&
4265 (attr->ia_size < inode->i_size)) {
4268 handle = ext4_journal_start(inode, 3);
4269 if (IS_ERR(handle)) {
4270 error = PTR_ERR(handle);
4273 if (ext4_handle_valid(handle)) {
4274 error = ext4_orphan_add(handle, inode);
4277 EXT4_I(inode)->i_disksize = attr->ia_size;
4278 rc = ext4_mark_inode_dirty(handle, inode);
4281 ext4_journal_stop(handle);
4283 if (ext4_should_order_data(inode)) {
4284 error = ext4_begin_ordered_truncate(inode,
4287 /* Do as much error cleanup as possible */
4288 handle = ext4_journal_start(inode, 3);
4289 if (IS_ERR(handle)) {
4290 ext4_orphan_del(NULL, inode);
4293 ext4_orphan_del(handle, inode);
4295 ext4_journal_stop(handle);
4301 if (attr->ia_valid & ATTR_SIZE) {
4302 if (attr->ia_size != i_size_read(inode)) {
4303 truncate_setsize(inode, attr->ia_size);
4304 ext4_truncate(inode);
4305 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
4306 ext4_truncate(inode);
4310 setattr_copy(inode, attr);
4311 mark_inode_dirty(inode);
4315 * If the call to ext4_truncate failed to get a transaction handle at
4316 * all, we need to clean up the in-core orphan list manually.
4318 if (orphan && inode->i_nlink)
4319 ext4_orphan_del(NULL, inode);
4321 if (!rc && (ia_valid & ATTR_MODE))
4322 rc = ext4_acl_chmod(inode);
4325 ext4_std_error(inode->i_sb, error);
4331 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4334 struct inode *inode;
4335 unsigned long delalloc_blocks;
4337 inode = dentry->d_inode;
4338 generic_fillattr(inode, stat);
4341 * We can't update i_blocks if the block allocation is delayed
4342 * otherwise in the case of system crash before the real block
4343 * allocation is done, we will have i_blocks inconsistent with
4344 * on-disk file blocks.
4345 * We always keep i_blocks updated together with real
4346 * allocation. But to not confuse with user, stat
4347 * will return the blocks that include the delayed allocation
4348 * blocks for this file.
4350 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4352 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4356 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4358 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4359 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4360 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4364 * Account for index blocks, block groups bitmaps and block group
4365 * descriptor blocks if modify datablocks and index blocks
4366 * worse case, the indexs blocks spread over different block groups
4368 * If datablocks are discontiguous, they are possible to spread over
4369 * different block groups too. If they are contiuguous, with flexbg,
4370 * they could still across block group boundary.
4372 * Also account for superblock, inode, quota and xattr blocks
4374 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4376 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4382 * How many index blocks need to touch to modify nrblocks?
4383 * The "Chunk" flag indicating whether the nrblocks is
4384 * physically contiguous on disk
4386 * For Direct IO and fallocate, they calls get_block to allocate
4387 * one single extent at a time, so they could set the "Chunk" flag
4389 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4394 * Now let's see how many group bitmaps and group descriptors need
4404 if (groups > ngroups)
4406 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4407 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4409 /* bitmaps and block group descriptor blocks */
4410 ret += groups + gdpblocks;
4412 /* Blocks for super block, inode, quota and xattr blocks */
4413 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4419 * Calculate the total number of credits to reserve to fit
4420 * the modification of a single pages into a single transaction,
4421 * which may include multiple chunks of block allocations.
4423 * This could be called via ext4_write_begin()
4425 * We need to consider the worse case, when
4426 * one new block per extent.
4428 int ext4_writepage_trans_blocks(struct inode *inode)
4430 int bpp = ext4_journal_blocks_per_page(inode);
4433 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4435 /* Account for data blocks for journalled mode */
4436 if (ext4_should_journal_data(inode))
4442 * Calculate the journal credits for a chunk of data modification.
4444 * This is called from DIO, fallocate or whoever calling
4445 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4447 * journal buffers for data blocks are not included here, as DIO
4448 * and fallocate do no need to journal data buffers.
4450 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4452 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4456 * The caller must have previously called ext4_reserve_inode_write().
4457 * Give this, we know that the caller already has write access to iloc->bh.
4459 int ext4_mark_iloc_dirty(handle_t *handle,
4460 struct inode *inode, struct ext4_iloc *iloc)
4464 if (test_opt(inode->i_sb, I_VERSION))
4465 inode_inc_iversion(inode);
4467 /* the do_update_inode consumes one bh->b_count */
4470 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4471 err = ext4_do_update_inode(handle, inode, iloc);
4477 * On success, We end up with an outstanding reference count against
4478 * iloc->bh. This _must_ be cleaned up later.
4482 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4483 struct ext4_iloc *iloc)
4487 err = ext4_get_inode_loc(inode, iloc);
4489 BUFFER_TRACE(iloc->bh, "get_write_access");
4490 err = ext4_journal_get_write_access(handle, iloc->bh);
4496 ext4_std_error(inode->i_sb, err);
4501 * Expand an inode by new_extra_isize bytes.
4502 * Returns 0 on success or negative error number on failure.
4504 static int ext4_expand_extra_isize(struct inode *inode,
4505 unsigned int new_extra_isize,
4506 struct ext4_iloc iloc,
4509 struct ext4_inode *raw_inode;
4510 struct ext4_xattr_ibody_header *header;
4512 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4515 raw_inode = ext4_raw_inode(&iloc);
4517 header = IHDR(inode, raw_inode);
4519 /* No extended attributes present */
4520 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4521 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4522 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4524 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4528 /* try to expand with EAs present */
4529 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4534 * What we do here is to mark the in-core inode as clean with respect to inode
4535 * dirtiness (it may still be data-dirty).
4536 * This means that the in-core inode may be reaped by prune_icache
4537 * without having to perform any I/O. This is a very good thing,
4538 * because *any* task may call prune_icache - even ones which
4539 * have a transaction open against a different journal.
4541 * Is this cheating? Not really. Sure, we haven't written the
4542 * inode out, but prune_icache isn't a user-visible syncing function.
4543 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4544 * we start and wait on commits.
4546 * Is this efficient/effective? Well, we're being nice to the system
4547 * by cleaning up our inodes proactively so they can be reaped
4548 * without I/O. But we are potentially leaving up to five seconds'
4549 * worth of inodes floating about which prune_icache wants us to
4550 * write out. One way to fix that would be to get prune_icache()
4551 * to do a write_super() to free up some memory. It has the desired
4554 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4556 struct ext4_iloc iloc;
4557 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4558 static unsigned int mnt_count;
4562 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4563 err = ext4_reserve_inode_write(handle, inode, &iloc);
4564 if (ext4_handle_valid(handle) &&
4565 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4566 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4568 * We need extra buffer credits since we may write into EA block
4569 * with this same handle. If journal_extend fails, then it will
4570 * only result in a minor loss of functionality for that inode.
4571 * If this is felt to be critical, then e2fsck should be run to
4572 * force a large enough s_min_extra_isize.
4574 if ((jbd2_journal_extend(handle,
4575 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4576 ret = ext4_expand_extra_isize(inode,
4577 sbi->s_want_extra_isize,
4580 ext4_set_inode_state(inode,
4581 EXT4_STATE_NO_EXPAND);
4583 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4584 ext4_warning(inode->i_sb,
4585 "Unable to expand inode %lu. Delete"
4586 " some EAs or run e2fsck.",
4589 le16_to_cpu(sbi->s_es->s_mnt_count);
4595 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4600 * ext4_dirty_inode() is called from __mark_inode_dirty()
4602 * We're really interested in the case where a file is being extended.
4603 * i_size has been changed by generic_commit_write() and we thus need
4604 * to include the updated inode in the current transaction.
4606 * Also, dquot_alloc_block() will always dirty the inode when blocks
4607 * are allocated to the file.
4609 * If the inode is marked synchronous, we don't honour that here - doing
4610 * so would cause a commit on atime updates, which we don't bother doing.
4611 * We handle synchronous inodes at the highest possible level.
4613 void ext4_dirty_inode(struct inode *inode, int flags)
4617 handle = ext4_journal_start(inode, 2);
4621 ext4_mark_inode_dirty(handle, inode);
4623 ext4_journal_stop(handle);
4630 * Bind an inode's backing buffer_head into this transaction, to prevent
4631 * it from being flushed to disk early. Unlike
4632 * ext4_reserve_inode_write, this leaves behind no bh reference and
4633 * returns no iloc structure, so the caller needs to repeat the iloc
4634 * lookup to mark the inode dirty later.
4636 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4638 struct ext4_iloc iloc;
4642 err = ext4_get_inode_loc(inode, &iloc);
4644 BUFFER_TRACE(iloc.bh, "get_write_access");
4645 err = jbd2_journal_get_write_access(handle, iloc.bh);
4647 err = ext4_handle_dirty_metadata(handle,
4653 ext4_std_error(inode->i_sb, err);
4658 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4665 * We have to be very careful here: changing a data block's
4666 * journaling status dynamically is dangerous. If we write a
4667 * data block to the journal, change the status and then delete
4668 * that block, we risk forgetting to revoke the old log record
4669 * from the journal and so a subsequent replay can corrupt data.
4670 * So, first we make sure that the journal is empty and that
4671 * nobody is changing anything.
4674 journal = EXT4_JOURNAL(inode);
4677 if (is_journal_aborted(journal))
4680 jbd2_journal_lock_updates(journal);
4681 jbd2_journal_flush(journal);
4684 * OK, there are no updates running now, and all cached data is
4685 * synced to disk. We are now in a completely consistent state
4686 * which doesn't have anything in the journal, and we know that
4687 * no filesystem updates are running, so it is safe to modify
4688 * the inode's in-core data-journaling state flag now.
4692 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4694 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4695 ext4_set_aops(inode);
4697 jbd2_journal_unlock_updates(journal);
4699 /* Finally we can mark the inode as dirty. */
4701 handle = ext4_journal_start(inode, 1);
4703 return PTR_ERR(handle);
4705 err = ext4_mark_inode_dirty(handle, inode);
4706 ext4_handle_sync(handle);
4707 ext4_journal_stop(handle);
4708 ext4_std_error(inode->i_sb, err);
4713 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4715 return !buffer_mapped(bh);
4718 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4720 struct page *page = vmf->page;
4724 struct file *file = vma->vm_file;
4725 struct inode *inode = file->f_path.dentry->d_inode;
4726 struct address_space *mapping = inode->i_mapping;
4728 get_block_t *get_block;
4732 * This check is racy but catches the common case. We rely on
4733 * __block_page_mkwrite() to do a reliable check.
4735 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
4736 /* Delalloc case is easy... */
4737 if (test_opt(inode->i_sb, DELALLOC) &&
4738 !ext4_should_journal_data(inode) &&
4739 !ext4_nonda_switch(inode->i_sb)) {
4741 ret = __block_page_mkwrite(vma, vmf,
4742 ext4_da_get_block_prep);
4743 } while (ret == -ENOSPC &&
4744 ext4_should_retry_alloc(inode->i_sb, &retries));
4749 size = i_size_read(inode);
4750 /* Page got truncated from under us? */
4751 if (page->mapping != mapping || page_offset(page) > size) {
4753 ret = VM_FAULT_NOPAGE;
4757 if (page->index == size >> PAGE_CACHE_SHIFT)
4758 len = size & ~PAGE_CACHE_MASK;
4760 len = PAGE_CACHE_SIZE;
4762 * Return if we have all the buffers mapped. This avoids the need to do
4763 * journal_start/journal_stop which can block and take a long time
4765 if (page_has_buffers(page)) {
4766 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4767 ext4_bh_unmapped)) {
4768 /* Wait so that we don't change page under IO */
4769 wait_on_page_writeback(page);
4770 ret = VM_FAULT_LOCKED;
4775 /* OK, we need to fill the hole... */
4776 if (ext4_should_dioread_nolock(inode))
4777 get_block = ext4_get_block_write;
4779 get_block = ext4_get_block;
4781 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4782 if (IS_ERR(handle)) {
4783 ret = VM_FAULT_SIGBUS;
4786 ret = __block_page_mkwrite(vma, vmf, get_block);
4787 if (!ret && ext4_should_journal_data(inode)) {
4788 if (walk_page_buffers(handle, page_buffers(page), 0,
4789 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4791 ret = VM_FAULT_SIGBUS;
4794 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4796 ext4_journal_stop(handle);
4797 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4800 ret = block_page_mkwrite_return(ret);