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
22 #include <linux/time.h>
23 #include <linux/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51 struct ext4_inode_info *ei)
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 csum_lo = raw->i_checksum_lo;
59 raw->i_checksum_lo = 0;
60 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
61 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
62 csum_hi = raw->i_checksum_hi;
63 raw->i_checksum_hi = 0;
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
67 EXT4_INODE_SIZE(inode->i_sb));
69 raw->i_checksum_lo = csum_lo;
70 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
71 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
72 raw->i_checksum_hi = csum_hi;
77 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
78 struct ext4_inode_info *ei)
80 __u32 provided, calculated;
82 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
83 cpu_to_le32(EXT4_OS_LINUX) ||
84 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
85 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
88 provided = le16_to_cpu(raw->i_checksum_lo);
89 calculated = ext4_inode_csum(inode, raw, ei);
90 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
91 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
92 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
96 return provided == calculated;
99 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
100 struct ext4_inode_info *ei)
104 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
105 cpu_to_le32(EXT4_OS_LINUX) ||
106 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
107 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
110 csum = ext4_inode_csum(inode, raw, ei);
111 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
112 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
113 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
114 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
117 static inline int ext4_begin_ordered_truncate(struct inode *inode,
120 trace_ext4_begin_ordered_truncate(inode, new_size);
122 * If jinode is zero, then we never opened the file for
123 * writing, so there's no need to call
124 * jbd2_journal_begin_ordered_truncate() since there's no
125 * outstanding writes we need to flush.
127 if (!EXT4_I(inode)->jinode)
129 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
130 EXT4_I(inode)->jinode,
134 static void ext4_invalidatepage(struct page *page, unsigned long offset);
135 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
136 struct buffer_head *bh_result, int create);
137 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
138 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
139 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
140 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
141 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
142 struct inode *inode, struct page *page, loff_t from,
143 loff_t length, int flags);
146 * Test whether an inode is a fast symlink.
148 static int ext4_inode_is_fast_symlink(struct inode *inode)
150 int ea_blocks = EXT4_I(inode)->i_file_acl ?
151 (inode->i_sb->s_blocksize >> 9) : 0;
153 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
157 * Restart the transaction associated with *handle. This does a commit,
158 * so before we call here everything must be consistently dirtied against
161 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
167 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
168 * moment, get_block can be called only for blocks inside i_size since
169 * page cache has been already dropped and writes are blocked by
170 * i_mutex. So we can safely drop the i_data_sem here.
172 BUG_ON(EXT4_JOURNAL(inode) == NULL);
173 jbd_debug(2, "restarting handle %p\n", handle);
174 up_write(&EXT4_I(inode)->i_data_sem);
175 ret = ext4_journal_restart(handle, nblocks);
176 down_write(&EXT4_I(inode)->i_data_sem);
177 ext4_discard_preallocations(inode);
183 * Called at the last iput() if i_nlink is zero.
185 void ext4_evict_inode(struct inode *inode)
190 trace_ext4_evict_inode(inode);
192 ext4_ioend_wait(inode);
194 if (inode->i_nlink) {
196 * When journalling data dirty buffers are tracked only in the
197 * journal. So although mm thinks everything is clean and
198 * ready for reaping the inode might still have some pages to
199 * write in the running transaction or waiting to be
200 * checkpointed. Thus calling jbd2_journal_invalidatepage()
201 * (via truncate_inode_pages()) to discard these buffers can
202 * cause data loss. Also even if we did not discard these
203 * buffers, we would have no way to find them after the inode
204 * is reaped and thus user could see stale data if he tries to
205 * read them before the transaction is checkpointed. So be
206 * careful and force everything to disk here... We use
207 * ei->i_datasync_tid to store the newest transaction
208 * containing inode's data.
210 * Note that directories do not have this problem because they
211 * don't use page cache.
213 if (ext4_should_journal_data(inode) &&
214 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
215 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
216 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
218 jbd2_log_start_commit(journal, commit_tid);
219 jbd2_log_wait_commit(journal, commit_tid);
220 filemap_write_and_wait(&inode->i_data);
222 truncate_inode_pages(&inode->i_data, 0);
226 if (!is_bad_inode(inode))
227 dquot_initialize(inode);
229 if (ext4_should_order_data(inode))
230 ext4_begin_ordered_truncate(inode, 0);
231 truncate_inode_pages(&inode->i_data, 0);
233 if (is_bad_inode(inode))
237 * Protect us against freezing - iput() caller didn't have to have any
238 * protection against it
240 sb_start_intwrite(inode->i_sb);
241 handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
242 if (IS_ERR(handle)) {
243 ext4_std_error(inode->i_sb, PTR_ERR(handle));
245 * If we're going to skip the normal cleanup, we still need to
246 * make sure that the in-core orphan linked list is properly
249 ext4_orphan_del(NULL, inode);
250 sb_end_intwrite(inode->i_sb);
255 ext4_handle_sync(handle);
257 err = ext4_mark_inode_dirty(handle, inode);
259 ext4_warning(inode->i_sb,
260 "couldn't mark inode dirty (err %d)", err);
264 ext4_truncate(inode);
267 * ext4_ext_truncate() doesn't reserve any slop when it
268 * restarts journal transactions; therefore there may not be
269 * enough credits left in the handle to remove the inode from
270 * the orphan list and set the dtime field.
272 if (!ext4_handle_has_enough_credits(handle, 3)) {
273 err = ext4_journal_extend(handle, 3);
275 err = ext4_journal_restart(handle, 3);
277 ext4_warning(inode->i_sb,
278 "couldn't extend journal (err %d)", err);
280 ext4_journal_stop(handle);
281 ext4_orphan_del(NULL, inode);
282 sb_end_intwrite(inode->i_sb);
288 * Kill off the orphan record which ext4_truncate created.
289 * AKPM: I think this can be inside the above `if'.
290 * Note that ext4_orphan_del() has to be able to cope with the
291 * deletion of a non-existent orphan - this is because we don't
292 * know if ext4_truncate() actually created an orphan record.
293 * (Well, we could do this if we need to, but heck - it works)
295 ext4_orphan_del(handle, inode);
296 EXT4_I(inode)->i_dtime = get_seconds();
299 * One subtle ordering requirement: if anything has gone wrong
300 * (transaction abort, IO errors, whatever), then we can still
301 * do these next steps (the fs will already have been marked as
302 * having errors), but we can't free the inode if the mark_dirty
305 if (ext4_mark_inode_dirty(handle, inode))
306 /* If that failed, just do the required in-core inode clear. */
307 ext4_clear_inode(inode);
309 ext4_free_inode(handle, inode);
310 ext4_journal_stop(handle);
311 sb_end_intwrite(inode->i_sb);
314 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
318 qsize_t *ext4_get_reserved_space(struct inode *inode)
320 return &EXT4_I(inode)->i_reserved_quota;
325 * Calculate the number of metadata blocks need to reserve
326 * to allocate a block located at @lblock
328 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
330 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
331 return ext4_ext_calc_metadata_amount(inode, lblock);
333 return ext4_ind_calc_metadata_amount(inode, lblock);
337 * Called with i_data_sem down, which is important since we can call
338 * ext4_discard_preallocations() from here.
340 void ext4_da_update_reserve_space(struct inode *inode,
341 int used, int quota_claim)
343 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
344 struct ext4_inode_info *ei = EXT4_I(inode);
346 spin_lock(&ei->i_block_reservation_lock);
347 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
348 if (unlikely(used > ei->i_reserved_data_blocks)) {
349 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
350 "with only %d reserved data blocks",
351 __func__, inode->i_ino, used,
352 ei->i_reserved_data_blocks);
354 used = ei->i_reserved_data_blocks;
357 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
358 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, allocated %d "
359 "with only %d reserved metadata blocks\n", __func__,
360 inode->i_ino, ei->i_allocated_meta_blocks,
361 ei->i_reserved_meta_blocks);
363 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
366 /* Update per-inode reservations */
367 ei->i_reserved_data_blocks -= used;
368 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
369 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
370 used + ei->i_allocated_meta_blocks);
371 ei->i_allocated_meta_blocks = 0;
373 if (ei->i_reserved_data_blocks == 0) {
375 * We can release all of the reserved metadata blocks
376 * only when we have written all of the delayed
379 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
380 ei->i_reserved_meta_blocks);
381 ei->i_reserved_meta_blocks = 0;
382 ei->i_da_metadata_calc_len = 0;
384 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
386 /* Update quota subsystem for data blocks */
388 dquot_claim_block(inode, EXT4_C2B(sbi, used));
391 * We did fallocate with an offset that is already delayed
392 * allocated. So on delayed allocated writeback we should
393 * not re-claim the quota for fallocated blocks.
395 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
399 * If we have done all the pending block allocations and if
400 * there aren't any writers on the inode, we can discard the
401 * inode's preallocations.
403 if ((ei->i_reserved_data_blocks == 0) &&
404 (atomic_read(&inode->i_writecount) == 0))
405 ext4_discard_preallocations(inode);
408 static int __check_block_validity(struct inode *inode, const char *func,
410 struct ext4_map_blocks *map)
412 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
414 ext4_error_inode(inode, func, line, map->m_pblk,
415 "lblock %lu mapped to illegal pblock "
416 "(length %d)", (unsigned long) map->m_lblk,
423 #define check_block_validity(inode, map) \
424 __check_block_validity((inode), __func__, __LINE__, (map))
427 * Return the number of contiguous dirty pages in a given inode
428 * starting at page frame idx.
430 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
431 unsigned int max_pages)
433 struct address_space *mapping = inode->i_mapping;
437 int i, nr_pages, done = 0;
441 pagevec_init(&pvec, 0);
444 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
446 (pgoff_t)PAGEVEC_SIZE);
449 for (i = 0; i < nr_pages; i++) {
450 struct page *page = pvec.pages[i];
451 struct buffer_head *bh, *head;
454 if (unlikely(page->mapping != mapping) ||
456 PageWriteback(page) ||
457 page->index != idx) {
462 if (page_has_buffers(page)) {
463 bh = head = page_buffers(page);
465 if (!buffer_delay(bh) &&
466 !buffer_unwritten(bh))
468 bh = bh->b_this_page;
469 } while (!done && (bh != head));
476 if (num >= max_pages) {
481 pagevec_release(&pvec);
487 * The ext4_map_blocks() function tries to look up the requested blocks,
488 * and returns if the blocks are already mapped.
490 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
491 * and store the allocated blocks in the result buffer head and mark it
494 * If file type is extents based, it will call ext4_ext_map_blocks(),
495 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
498 * On success, it returns the number of blocks being mapped or allocate.
499 * if create==0 and the blocks are pre-allocated and uninitialized block,
500 * the result buffer head is unmapped. If the create ==1, it will make sure
501 * the buffer head is mapped.
503 * It returns 0 if plain look up failed (blocks have not been allocated), in
504 * that case, buffer head is unmapped
506 * It returns the error in case of allocation failure.
508 int ext4_map_blocks(handle_t *handle, struct inode *inode,
509 struct ext4_map_blocks *map, int flags)
514 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
515 "logical block %lu\n", inode->i_ino, flags, map->m_len,
516 (unsigned long) map->m_lblk);
518 * Try to see if we can get the block without requesting a new
521 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
522 down_read((&EXT4_I(inode)->i_data_sem));
523 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
524 retval = ext4_ext_map_blocks(handle, inode, map, flags &
525 EXT4_GET_BLOCKS_KEEP_SIZE);
527 retval = ext4_ind_map_blocks(handle, inode, map, flags &
528 EXT4_GET_BLOCKS_KEEP_SIZE);
530 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
531 up_read((&EXT4_I(inode)->i_data_sem));
533 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
535 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
536 /* delayed alloc may be allocated by fallocate and
537 * coverted to initialized by directIO.
538 * we need to handle delayed extent here.
540 down_write((&EXT4_I(inode)->i_data_sem));
543 ret = check_block_validity(inode, map);
548 /* If it is only a block(s) look up */
549 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
553 * Returns if the blocks have already allocated
555 * Note that if blocks have been preallocated
556 * ext4_ext_get_block() returns the create = 0
557 * with buffer head unmapped.
559 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
563 * When we call get_blocks without the create flag, the
564 * BH_Unwritten flag could have gotten set if the blocks
565 * requested were part of a uninitialized extent. We need to
566 * clear this flag now that we are committed to convert all or
567 * part of the uninitialized extent to be an initialized
568 * extent. This is because we need to avoid the combination
569 * of BH_Unwritten and BH_Mapped flags being simultaneously
570 * set on the buffer_head.
572 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
575 * New blocks allocate and/or writing to uninitialized extent
576 * will possibly result in updating i_data, so we take
577 * the write lock of i_data_sem, and call get_blocks()
578 * with create == 1 flag.
580 down_write((&EXT4_I(inode)->i_data_sem));
583 * if the caller is from delayed allocation writeout path
584 * we have already reserved fs blocks for allocation
585 * let the underlying get_block() function know to
586 * avoid double accounting
588 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
589 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
591 * We need to check for EXT4 here because migrate
592 * could have changed the inode type in between
594 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
595 retval = ext4_ext_map_blocks(handle, inode, map, flags);
597 retval = ext4_ind_map_blocks(handle, inode, map, flags);
599 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
601 * We allocated new blocks which will result in
602 * i_data's format changing. Force the migrate
603 * to fail by clearing migrate flags
605 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
609 * Update reserved blocks/metadata blocks after successful
610 * block allocation which had been deferred till now. We don't
611 * support fallocate for non extent files. So we can update
612 * reserve space here.
615 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
616 ext4_da_update_reserve_space(inode, retval, 1);
618 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
619 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
621 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
624 /* delayed allocation blocks has been allocated */
625 ret = ext4_es_remove_extent(inode, map->m_lblk,
632 up_write((&EXT4_I(inode)->i_data_sem));
633 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
634 int ret = check_block_validity(inode, map);
641 /* Maximum number of blocks we map for direct IO at once. */
642 #define DIO_MAX_BLOCKS 4096
644 static int _ext4_get_block(struct inode *inode, sector_t iblock,
645 struct buffer_head *bh, int flags)
647 handle_t *handle = ext4_journal_current_handle();
648 struct ext4_map_blocks map;
649 int ret = 0, started = 0;
652 if (ext4_has_inline_data(inode))
656 map.m_len = bh->b_size >> inode->i_blkbits;
658 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
659 /* Direct IO write... */
660 if (map.m_len > DIO_MAX_BLOCKS)
661 map.m_len = DIO_MAX_BLOCKS;
662 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
663 handle = ext4_journal_start(inode, dio_credits);
664 if (IS_ERR(handle)) {
665 ret = PTR_ERR(handle);
671 ret = ext4_map_blocks(handle, inode, &map, flags);
673 map_bh(bh, inode->i_sb, map.m_pblk);
674 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
675 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
679 ext4_journal_stop(handle);
683 int ext4_get_block(struct inode *inode, sector_t iblock,
684 struct buffer_head *bh, int create)
686 return _ext4_get_block(inode, iblock, bh,
687 create ? EXT4_GET_BLOCKS_CREATE : 0);
691 * `handle' can be NULL if create is zero
693 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
694 ext4_lblk_t block, int create, int *errp)
696 struct ext4_map_blocks map;
697 struct buffer_head *bh;
700 J_ASSERT(handle != NULL || create == 0);
704 err = ext4_map_blocks(handle, inode, &map,
705 create ? EXT4_GET_BLOCKS_CREATE : 0);
707 /* ensure we send some value back into *errp */
715 bh = sb_getblk(inode->i_sb, map.m_pblk);
720 if (map.m_flags & EXT4_MAP_NEW) {
721 J_ASSERT(create != 0);
722 J_ASSERT(handle != NULL);
725 * Now that we do not always journal data, we should
726 * keep in mind whether this should always journal the
727 * new buffer as metadata. For now, regular file
728 * writes use ext4_get_block instead, so it's not a
732 BUFFER_TRACE(bh, "call get_create_access");
733 fatal = ext4_journal_get_create_access(handle, bh);
734 if (!fatal && !buffer_uptodate(bh)) {
735 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
736 set_buffer_uptodate(bh);
739 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
740 err = ext4_handle_dirty_metadata(handle, inode, bh);
744 BUFFER_TRACE(bh, "not a new buffer");
754 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
755 ext4_lblk_t block, int create, int *err)
757 struct buffer_head *bh;
759 bh = ext4_getblk(handle, inode, block, create, err);
762 if (buffer_uptodate(bh))
764 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
766 if (buffer_uptodate(bh))
773 int ext4_walk_page_buffers(handle_t *handle,
774 struct buffer_head *head,
778 int (*fn)(handle_t *handle,
779 struct buffer_head *bh))
781 struct buffer_head *bh;
782 unsigned block_start, block_end;
783 unsigned blocksize = head->b_size;
785 struct buffer_head *next;
787 for (bh = head, block_start = 0;
788 ret == 0 && (bh != head || !block_start);
789 block_start = block_end, bh = next) {
790 next = bh->b_this_page;
791 block_end = block_start + blocksize;
792 if (block_end <= from || block_start >= to) {
793 if (partial && !buffer_uptodate(bh))
797 err = (*fn)(handle, bh);
805 * To preserve ordering, it is essential that the hole instantiation and
806 * the data write be encapsulated in a single transaction. We cannot
807 * close off a transaction and start a new one between the ext4_get_block()
808 * and the commit_write(). So doing the jbd2_journal_start at the start of
809 * prepare_write() is the right place.
811 * Also, this function can nest inside ext4_writepage() ->
812 * block_write_full_page(). In that case, we *know* that ext4_writepage()
813 * has generated enough buffer credits to do the whole page. So we won't
814 * block on the journal in that case, which is good, because the caller may
817 * By accident, ext4 can be reentered when a transaction is open via
818 * quota file writes. If we were to commit the transaction while thus
819 * reentered, there can be a deadlock - we would be holding a quota
820 * lock, and the commit would never complete if another thread had a
821 * transaction open and was blocking on the quota lock - a ranking
824 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
825 * will _not_ run commit under these circumstances because handle->h_ref
826 * is elevated. We'll still have enough credits for the tiny quotafile
829 int do_journal_get_write_access(handle_t *handle,
830 struct buffer_head *bh)
832 int dirty = buffer_dirty(bh);
835 if (!buffer_mapped(bh) || buffer_freed(bh))
838 * __block_write_begin() could have dirtied some buffers. Clean
839 * the dirty bit as jbd2_journal_get_write_access() could complain
840 * otherwise about fs integrity issues. Setting of the dirty bit
841 * by __block_write_begin() isn't a real problem here as we clear
842 * the bit before releasing a page lock and thus writeback cannot
843 * ever write the buffer.
846 clear_buffer_dirty(bh);
847 ret = ext4_journal_get_write_access(handle, bh);
849 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
853 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
854 struct buffer_head *bh_result, int create);
855 static int ext4_write_begin(struct file *file, struct address_space *mapping,
856 loff_t pos, unsigned len, unsigned flags,
857 struct page **pagep, void **fsdata)
859 struct inode *inode = mapping->host;
860 int ret, needed_blocks;
867 trace_ext4_write_begin(inode, pos, len, flags);
869 * Reserve one block more for addition to orphan list in case
870 * we allocate blocks but write fails for some reason
872 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
873 index = pos >> PAGE_CACHE_SHIFT;
874 from = pos & (PAGE_CACHE_SIZE - 1);
877 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
878 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
889 handle = ext4_journal_start(inode, needed_blocks);
890 if (IS_ERR(handle)) {
891 ret = PTR_ERR(handle);
895 /* We cannot recurse into the filesystem as the transaction is already
897 flags |= AOP_FLAG_NOFS;
899 page = grab_cache_page_write_begin(mapping, index, flags);
901 ext4_journal_stop(handle);
908 if (ext4_should_dioread_nolock(inode))
909 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
911 ret = __block_write_begin(page, pos, len, ext4_get_block);
913 if (!ret && ext4_should_journal_data(inode)) {
914 ret = ext4_walk_page_buffers(handle, page_buffers(page),
916 do_journal_get_write_access);
921 page_cache_release(page);
923 * __block_write_begin may have instantiated a few blocks
924 * outside i_size. Trim these off again. Don't need
925 * i_size_read because we hold i_mutex.
927 * Add inode to orphan list in case we crash before
930 if (pos + len > inode->i_size && ext4_can_truncate(inode))
931 ext4_orphan_add(handle, inode);
933 ext4_journal_stop(handle);
934 if (pos + len > inode->i_size) {
935 ext4_truncate_failed_write(inode);
937 * If truncate failed early the inode might
938 * still be on the orphan list; we need to
939 * make sure the inode is removed from the
940 * orphan list in that case.
943 ext4_orphan_del(NULL, inode);
947 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
953 /* For write_end() in data=journal mode */
954 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
956 if (!buffer_mapped(bh) || buffer_freed(bh))
958 set_buffer_uptodate(bh);
959 return ext4_handle_dirty_metadata(handle, NULL, bh);
962 static int ext4_generic_write_end(struct file *file,
963 struct address_space *mapping,
964 loff_t pos, unsigned len, unsigned copied,
965 struct page *page, void *fsdata)
967 int i_size_changed = 0;
968 struct inode *inode = mapping->host;
969 handle_t *handle = ext4_journal_current_handle();
971 if (ext4_has_inline_data(inode))
972 copied = ext4_write_inline_data_end(inode, pos, len,
975 copied = block_write_end(file, mapping, pos,
976 len, copied, page, fsdata);
979 * No need to use i_size_read() here, the i_size
980 * cannot change under us because we hold i_mutex.
982 * But it's important to update i_size while still holding page lock:
983 * page writeout could otherwise come in and zero beyond i_size.
985 if (pos + copied > inode->i_size) {
986 i_size_write(inode, pos + copied);
990 if (pos + copied > EXT4_I(inode)->i_disksize) {
991 /* We need to mark inode dirty even if
992 * new_i_size is less that inode->i_size
993 * bu greater than i_disksize.(hint delalloc)
995 ext4_update_i_disksize(inode, (pos + copied));
999 page_cache_release(page);
1002 * Don't mark the inode dirty under page lock. First, it unnecessarily
1003 * makes the holding time of page lock longer. Second, it forces lock
1004 * ordering of page lock and transaction start for journaling
1008 ext4_mark_inode_dirty(handle, inode);
1014 * We need to pick up the new inode size which generic_commit_write gave us
1015 * `file' can be NULL - eg, when called from page_symlink().
1017 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1018 * buffers are managed internally.
1020 static int ext4_ordered_write_end(struct file *file,
1021 struct address_space *mapping,
1022 loff_t pos, unsigned len, unsigned copied,
1023 struct page *page, void *fsdata)
1025 handle_t *handle = ext4_journal_current_handle();
1026 struct inode *inode = mapping->host;
1029 trace_ext4_ordered_write_end(inode, pos, len, copied);
1030 ret = ext4_jbd2_file_inode(handle, inode);
1033 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1036 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1037 /* if we have allocated more blocks and copied
1038 * less. We will have blocks allocated outside
1039 * inode->i_size. So truncate them
1041 ext4_orphan_add(handle, inode);
1046 page_cache_release(page);
1049 ret2 = ext4_journal_stop(handle);
1053 if (pos + len > inode->i_size) {
1054 ext4_truncate_failed_write(inode);
1056 * If truncate failed early the inode might still be
1057 * on the orphan list; we need to make sure the inode
1058 * is removed from the orphan list in that case.
1061 ext4_orphan_del(NULL, inode);
1065 return ret ? ret : copied;
1068 static int ext4_writeback_write_end(struct file *file,
1069 struct address_space *mapping,
1070 loff_t pos, unsigned len, unsigned copied,
1071 struct page *page, void *fsdata)
1073 handle_t *handle = ext4_journal_current_handle();
1074 struct inode *inode = mapping->host;
1077 trace_ext4_writeback_write_end(inode, pos, len, copied);
1078 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1081 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1082 /* if we have allocated more blocks and copied
1083 * less. We will have blocks allocated outside
1084 * inode->i_size. So truncate them
1086 ext4_orphan_add(handle, inode);
1091 ret2 = ext4_journal_stop(handle);
1095 if (pos + len > inode->i_size) {
1096 ext4_truncate_failed_write(inode);
1098 * If truncate failed early the inode might still be
1099 * on the orphan list; we need to make sure the inode
1100 * is removed from the orphan list in that case.
1103 ext4_orphan_del(NULL, inode);
1106 return ret ? ret : copied;
1109 static int ext4_journalled_write_end(struct file *file,
1110 struct address_space *mapping,
1111 loff_t pos, unsigned len, unsigned copied,
1112 struct page *page, void *fsdata)
1114 handle_t *handle = ext4_journal_current_handle();
1115 struct inode *inode = mapping->host;
1121 trace_ext4_journalled_write_end(inode, pos, len, copied);
1122 from = pos & (PAGE_CACHE_SIZE - 1);
1125 BUG_ON(!ext4_handle_valid(handle));
1127 if (ext4_has_inline_data(inode))
1128 copied = ext4_write_inline_data_end(inode, pos, len,
1132 if (!PageUptodate(page))
1134 page_zero_new_buffers(page, from+copied, to);
1137 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1138 to, &partial, write_end_fn);
1140 SetPageUptodate(page);
1142 new_i_size = pos + copied;
1143 if (new_i_size > inode->i_size)
1144 i_size_write(inode, pos+copied);
1145 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1146 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1147 if (new_i_size > EXT4_I(inode)->i_disksize) {
1148 ext4_update_i_disksize(inode, new_i_size);
1149 ret2 = ext4_mark_inode_dirty(handle, inode);
1155 page_cache_release(page);
1156 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1157 /* if we have allocated more blocks and copied
1158 * less. We will have blocks allocated outside
1159 * inode->i_size. So truncate them
1161 ext4_orphan_add(handle, inode);
1163 ret2 = ext4_journal_stop(handle);
1166 if (pos + len > inode->i_size) {
1167 ext4_truncate_failed_write(inode);
1169 * If truncate failed early the inode might still be
1170 * on the orphan list; we need to make sure the inode
1171 * is removed from the orphan list in that case.
1174 ext4_orphan_del(NULL, inode);
1177 return ret ? ret : copied;
1181 * Reserve a single cluster located at lblock
1183 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1186 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1187 struct ext4_inode_info *ei = EXT4_I(inode);
1188 unsigned int md_needed;
1190 ext4_lblk_t save_last_lblock;
1194 * We will charge metadata quota at writeout time; this saves
1195 * us from metadata over-estimation, though we may go over by
1196 * a small amount in the end. Here we just reserve for data.
1198 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1203 * recalculate the amount of metadata blocks to reserve
1204 * in order to allocate nrblocks
1205 * worse case is one extent per block
1208 spin_lock(&ei->i_block_reservation_lock);
1210 * ext4_calc_metadata_amount() has side effects, which we have
1211 * to be prepared undo if we fail to claim space.
1213 save_len = ei->i_da_metadata_calc_len;
1214 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1215 md_needed = EXT4_NUM_B2C(sbi,
1216 ext4_calc_metadata_amount(inode, lblock));
1217 trace_ext4_da_reserve_space(inode, md_needed);
1220 * We do still charge estimated metadata to the sb though;
1221 * we cannot afford to run out of free blocks.
1223 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1224 ei->i_da_metadata_calc_len = save_len;
1225 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1226 spin_unlock(&ei->i_block_reservation_lock);
1227 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1231 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1234 ei->i_reserved_data_blocks++;
1235 ei->i_reserved_meta_blocks += md_needed;
1236 spin_unlock(&ei->i_block_reservation_lock);
1238 return 0; /* success */
1241 static void ext4_da_release_space(struct inode *inode, int to_free)
1243 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1244 struct ext4_inode_info *ei = EXT4_I(inode);
1247 return; /* Nothing to release, exit */
1249 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1251 trace_ext4_da_release_space(inode, to_free);
1252 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1254 * if there aren't enough reserved blocks, then the
1255 * counter is messed up somewhere. Since this
1256 * function is called from invalidate page, it's
1257 * harmless to return without any action.
1259 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1260 "ino %lu, to_free %d with only %d reserved "
1261 "data blocks", inode->i_ino, to_free,
1262 ei->i_reserved_data_blocks);
1264 to_free = ei->i_reserved_data_blocks;
1266 ei->i_reserved_data_blocks -= to_free;
1268 if (ei->i_reserved_data_blocks == 0) {
1270 * We can release all of the reserved metadata blocks
1271 * only when we have written all of the delayed
1272 * allocation blocks.
1273 * Note that in case of bigalloc, i_reserved_meta_blocks,
1274 * i_reserved_data_blocks, etc. refer to number of clusters.
1276 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1277 ei->i_reserved_meta_blocks);
1278 ei->i_reserved_meta_blocks = 0;
1279 ei->i_da_metadata_calc_len = 0;
1282 /* update fs dirty data blocks counter */
1283 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1285 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1287 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1290 static void ext4_da_page_release_reservation(struct page *page,
1291 unsigned long offset)
1294 struct buffer_head *head, *bh;
1295 unsigned int curr_off = 0;
1296 struct inode *inode = page->mapping->host;
1297 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1301 head = page_buffers(page);
1304 unsigned int next_off = curr_off + bh->b_size;
1306 if ((offset <= curr_off) && (buffer_delay(bh))) {
1308 clear_buffer_delay(bh);
1310 curr_off = next_off;
1311 } while ((bh = bh->b_this_page) != head);
1314 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1315 ext4_es_remove_extent(inode, lblk, to_release);
1318 /* If we have released all the blocks belonging to a cluster, then we
1319 * need to release the reserved space for that cluster. */
1320 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1321 while (num_clusters > 0) {
1322 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1323 ((num_clusters - 1) << sbi->s_cluster_bits);
1324 if (sbi->s_cluster_ratio == 1 ||
1325 !ext4_find_delalloc_cluster(inode, lblk))
1326 ext4_da_release_space(inode, 1);
1333 * Delayed allocation stuff
1337 * mpage_da_submit_io - walks through extent of pages and try to write
1338 * them with writepage() call back
1340 * @mpd->inode: inode
1341 * @mpd->first_page: first page of the extent
1342 * @mpd->next_page: page after the last page of the extent
1344 * By the time mpage_da_submit_io() is called we expect all blocks
1345 * to be allocated. this may be wrong if allocation failed.
1347 * As pages are already locked by write_cache_pages(), we can't use it
1349 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1350 struct ext4_map_blocks *map)
1352 struct pagevec pvec;
1353 unsigned long index, end;
1354 int ret = 0, err, nr_pages, i;
1355 struct inode *inode = mpd->inode;
1356 struct address_space *mapping = inode->i_mapping;
1357 loff_t size = i_size_read(inode);
1358 unsigned int len, block_start;
1359 struct buffer_head *bh, *page_bufs = NULL;
1360 int journal_data = ext4_should_journal_data(inode);
1361 sector_t pblock = 0, cur_logical = 0;
1362 struct ext4_io_submit io_submit;
1364 BUG_ON(mpd->next_page <= mpd->first_page);
1365 memset(&io_submit, 0, sizeof(io_submit));
1367 * We need to start from the first_page to the next_page - 1
1368 * to make sure we also write the mapped dirty buffer_heads.
1369 * If we look at mpd->b_blocknr we would only be looking
1370 * at the currently mapped buffer_heads.
1372 index = mpd->first_page;
1373 end = mpd->next_page - 1;
1375 pagevec_init(&pvec, 0);
1376 while (index <= end) {
1377 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1380 for (i = 0; i < nr_pages; i++) {
1381 int commit_write = 0, skip_page = 0;
1382 struct page *page = pvec.pages[i];
1384 index = page->index;
1388 if (index == size >> PAGE_CACHE_SHIFT)
1389 len = size & ~PAGE_CACHE_MASK;
1391 len = PAGE_CACHE_SIZE;
1393 cur_logical = index << (PAGE_CACHE_SHIFT -
1395 pblock = map->m_pblk + (cur_logical -
1400 BUG_ON(!PageLocked(page));
1401 BUG_ON(PageWriteback(page));
1404 * If the page does not have buffers (for
1405 * whatever reason), try to create them using
1406 * __block_write_begin. If this fails,
1407 * skip the page and move on.
1409 if (!page_has_buffers(page)) {
1410 if (__block_write_begin(page, 0, len,
1411 noalloc_get_block_write)) {
1419 bh = page_bufs = page_buffers(page);
1424 if (map && (cur_logical >= map->m_lblk) &&
1425 (cur_logical <= (map->m_lblk +
1426 (map->m_len - 1)))) {
1427 if (buffer_delay(bh)) {
1428 clear_buffer_delay(bh);
1429 bh->b_blocknr = pblock;
1431 if (buffer_unwritten(bh) ||
1433 BUG_ON(bh->b_blocknr != pblock);
1434 if (map->m_flags & EXT4_MAP_UNINIT)
1435 set_buffer_uninit(bh);
1436 clear_buffer_unwritten(bh);
1440 * skip page if block allocation undone and
1443 if (ext4_bh_delay_or_unwritten(NULL, bh))
1445 bh = bh->b_this_page;
1446 block_start += bh->b_size;
1449 } while (bh != page_bufs);
1455 /* mark the buffer_heads as dirty & uptodate */
1456 block_commit_write(page, 0, len);
1458 clear_page_dirty_for_io(page);
1460 * Delalloc doesn't support data journalling,
1461 * but eventually maybe we'll lift this
1464 if (unlikely(journal_data && PageChecked(page)))
1465 err = __ext4_journalled_writepage(page, len);
1466 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1467 err = ext4_bio_write_page(&io_submit, page,
1469 else if (buffer_uninit(page_bufs)) {
1470 ext4_set_bh_endio(page_bufs, inode);
1471 err = block_write_full_page_endio(page,
1472 noalloc_get_block_write,
1473 mpd->wbc, ext4_end_io_buffer_write);
1475 err = block_write_full_page(page,
1476 noalloc_get_block_write, mpd->wbc);
1479 mpd->pages_written++;
1481 * In error case, we have to continue because
1482 * remaining pages are still locked
1487 pagevec_release(&pvec);
1489 ext4_io_submit(&io_submit);
1493 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1497 struct pagevec pvec;
1498 struct inode *inode = mpd->inode;
1499 struct address_space *mapping = inode->i_mapping;
1500 ext4_lblk_t start, last;
1502 index = mpd->first_page;
1503 end = mpd->next_page - 1;
1505 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1506 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1507 ext4_es_remove_extent(inode, start, last - start + 1);
1509 pagevec_init(&pvec, 0);
1510 while (index <= end) {
1511 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1514 for (i = 0; i < nr_pages; i++) {
1515 struct page *page = pvec.pages[i];
1516 if (page->index > end)
1518 BUG_ON(!PageLocked(page));
1519 BUG_ON(PageWriteback(page));
1520 block_invalidatepage(page, 0);
1521 ClearPageUptodate(page);
1524 index = pvec.pages[nr_pages - 1]->index + 1;
1525 pagevec_release(&pvec);
1530 static void ext4_print_free_blocks(struct inode *inode)
1532 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1533 struct super_block *sb = inode->i_sb;
1535 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1536 EXT4_C2B(EXT4_SB(inode->i_sb),
1537 ext4_count_free_clusters(inode->i_sb)));
1538 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1539 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1540 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1541 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1542 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1543 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1544 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1545 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1546 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1547 EXT4_I(inode)->i_reserved_data_blocks);
1548 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1549 EXT4_I(inode)->i_reserved_meta_blocks);
1554 * mpage_da_map_and_submit - go through given space, map them
1555 * if necessary, and then submit them for I/O
1557 * @mpd - bh describing space
1559 * The function skips space we know is already mapped to disk blocks.
1562 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1564 int err, blks, get_blocks_flags;
1565 struct ext4_map_blocks map, *mapp = NULL;
1566 sector_t next = mpd->b_blocknr;
1567 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1568 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1569 handle_t *handle = NULL;
1572 * If the blocks are mapped already, or we couldn't accumulate
1573 * any blocks, then proceed immediately to the submission stage.
1575 if ((mpd->b_size == 0) ||
1576 ((mpd->b_state & (1 << BH_Mapped)) &&
1577 !(mpd->b_state & (1 << BH_Delay)) &&
1578 !(mpd->b_state & (1 << BH_Unwritten))))
1581 handle = ext4_journal_current_handle();
1585 * Call ext4_map_blocks() to allocate any delayed allocation
1586 * blocks, or to convert an uninitialized extent to be
1587 * initialized (in the case where we have written into
1588 * one or more preallocated blocks).
1590 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1591 * indicate that we are on the delayed allocation path. This
1592 * affects functions in many different parts of the allocation
1593 * call path. This flag exists primarily because we don't
1594 * want to change *many* call functions, so ext4_map_blocks()
1595 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1596 * inode's allocation semaphore is taken.
1598 * If the blocks in questions were delalloc blocks, set
1599 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1600 * variables are updated after the blocks have been allocated.
1603 map.m_len = max_blocks;
1604 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1605 if (ext4_should_dioread_nolock(mpd->inode))
1606 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1607 if (mpd->b_state & (1 << BH_Delay))
1608 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1610 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1612 struct super_block *sb = mpd->inode->i_sb;
1616 * If get block returns EAGAIN or ENOSPC and there
1617 * appears to be free blocks we will just let
1618 * mpage_da_submit_io() unlock all of the pages.
1623 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1629 * get block failure will cause us to loop in
1630 * writepages, because a_ops->writepage won't be able
1631 * to make progress. The page will be redirtied by
1632 * writepage and writepages will again try to write
1635 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1636 ext4_msg(sb, KERN_CRIT,
1637 "delayed block allocation failed for inode %lu "
1638 "at logical offset %llu with max blocks %zd "
1639 "with error %d", mpd->inode->i_ino,
1640 (unsigned long long) next,
1641 mpd->b_size >> mpd->inode->i_blkbits, err);
1642 ext4_msg(sb, KERN_CRIT,
1643 "This should not happen!! Data will be lost\n");
1645 ext4_print_free_blocks(mpd->inode);
1647 /* invalidate all the pages */
1648 ext4_da_block_invalidatepages(mpd);
1650 /* Mark this page range as having been completed */
1657 if (map.m_flags & EXT4_MAP_NEW) {
1658 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1661 for (i = 0; i < map.m_len; i++)
1662 unmap_underlying_metadata(bdev, map.m_pblk + i);
1666 * Update on-disk size along with block allocation.
1668 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1669 if (disksize > i_size_read(mpd->inode))
1670 disksize = i_size_read(mpd->inode);
1671 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1672 ext4_update_i_disksize(mpd->inode, disksize);
1673 err = ext4_mark_inode_dirty(handle, mpd->inode);
1675 ext4_error(mpd->inode->i_sb,
1676 "Failed to mark inode %lu dirty",
1681 mpage_da_submit_io(mpd, mapp);
1685 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1686 (1 << BH_Delay) | (1 << BH_Unwritten))
1689 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1691 * @mpd->lbh - extent of blocks
1692 * @logical - logical number of the block in the file
1693 * @bh - bh of the block (used to access block's state)
1695 * the function is used to collect contig. blocks in same state
1697 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1698 sector_t logical, size_t b_size,
1699 unsigned long b_state)
1702 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1705 * XXX Don't go larger than mballoc is willing to allocate
1706 * This is a stopgap solution. We eventually need to fold
1707 * mpage_da_submit_io() into this function and then call
1708 * ext4_map_blocks() multiple times in a loop
1710 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1713 /* check if thereserved journal credits might overflow */
1714 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1715 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1717 * With non-extent format we are limited by the journal
1718 * credit available. Total credit needed to insert
1719 * nrblocks contiguous blocks is dependent on the
1720 * nrblocks. So limit nrblocks.
1723 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1724 EXT4_MAX_TRANS_DATA) {
1726 * Adding the new buffer_head would make it cross the
1727 * allowed limit for which we have journal credit
1728 * reserved. So limit the new bh->b_size
1730 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1731 mpd->inode->i_blkbits;
1732 /* we will do mpage_da_submit_io in the next loop */
1736 * First block in the extent
1738 if (mpd->b_size == 0) {
1739 mpd->b_blocknr = logical;
1740 mpd->b_size = b_size;
1741 mpd->b_state = b_state & BH_FLAGS;
1745 next = mpd->b_blocknr + nrblocks;
1747 * Can we merge the block to our big extent?
1749 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1750 mpd->b_size += b_size;
1756 * We couldn't merge the block to our extent, so we
1757 * need to flush current extent and start new one
1759 mpage_da_map_and_submit(mpd);
1763 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1765 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1769 * This function is grabs code from the very beginning of
1770 * ext4_map_blocks, but assumes that the caller is from delayed write
1771 * time. This function looks up the requested blocks and sets the
1772 * buffer delay bit under the protection of i_data_sem.
1774 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1775 struct ext4_map_blocks *map,
1776 struct buffer_head *bh)
1779 sector_t invalid_block = ~((sector_t) 0xffff);
1781 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1785 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1786 "logical block %lu\n", inode->i_ino, map->m_len,
1787 (unsigned long) map->m_lblk);
1789 * Try to see if we can get the block without requesting a new
1790 * file system block.
1792 down_read((&EXT4_I(inode)->i_data_sem));
1793 if (ext4_has_inline_data(inode)) {
1795 * We will soon create blocks for this page, and let
1796 * us pretend as if the blocks aren't allocated yet.
1797 * In case of clusters, we have to handle the work
1798 * of mapping from cluster so that the reserved space
1799 * is calculated properly.
1801 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1802 ext4_find_delalloc_cluster(inode, map->m_lblk))
1803 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1805 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1806 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1808 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1812 * XXX: __block_prepare_write() unmaps passed block,
1815 /* If the block was allocated from previously allocated cluster,
1816 * then we dont need to reserve it again. */
1817 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1818 retval = ext4_da_reserve_space(inode, iblock);
1820 /* not enough space to reserve */
1824 retval = ext4_es_insert_extent(inode, map->m_lblk, map->m_len);
1828 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1829 * and it should not appear on the bh->b_state.
1831 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1833 map_bh(bh, inode->i_sb, invalid_block);
1835 set_buffer_delay(bh);
1839 up_read((&EXT4_I(inode)->i_data_sem));
1845 * This is a special get_blocks_t callback which is used by
1846 * ext4_da_write_begin(). It will either return mapped block or
1847 * reserve space for a single block.
1849 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1850 * We also have b_blocknr = -1 and b_bdev initialized properly
1852 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1853 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1854 * initialized properly.
1856 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1857 struct buffer_head *bh, int create)
1859 struct ext4_map_blocks map;
1862 BUG_ON(create == 0);
1863 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1865 map.m_lblk = iblock;
1869 * first, we need to know whether the block is allocated already
1870 * preallocated blocks are unmapped but should treated
1871 * the same as allocated blocks.
1873 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1877 map_bh(bh, inode->i_sb, map.m_pblk);
1878 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1880 if (buffer_unwritten(bh)) {
1881 /* A delayed write to unwritten bh should be marked
1882 * new and mapped. Mapped ensures that we don't do
1883 * get_block multiple times when we write to the same
1884 * offset and new ensures that we do proper zero out
1885 * for partial write.
1888 set_buffer_mapped(bh);
1894 * This function is used as a standard get_block_t calback function
1895 * when there is no desire to allocate any blocks. It is used as a
1896 * callback function for block_write_begin() and block_write_full_page().
1897 * These functions should only try to map a single block at a time.
1899 * Since this function doesn't do block allocations even if the caller
1900 * requests it by passing in create=1, it is critically important that
1901 * any caller checks to make sure that any buffer heads are returned
1902 * by this function are either all already mapped or marked for
1903 * delayed allocation before calling block_write_full_page(). Otherwise,
1904 * b_blocknr could be left unitialized, and the page write functions will
1905 * be taken by surprise.
1907 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1908 struct buffer_head *bh_result, int create)
1910 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1911 return _ext4_get_block(inode, iblock, bh_result, 0);
1914 static int bget_one(handle_t *handle, struct buffer_head *bh)
1920 static int bput_one(handle_t *handle, struct buffer_head *bh)
1926 static int __ext4_journalled_writepage(struct page *page,
1929 struct address_space *mapping = page->mapping;
1930 struct inode *inode = mapping->host;
1931 struct buffer_head *page_bufs = NULL;
1932 handle_t *handle = NULL;
1933 int ret = 0, err = 0;
1934 int inline_data = ext4_has_inline_data(inode);
1935 struct buffer_head *inode_bh = NULL;
1937 ClearPageChecked(page);
1940 BUG_ON(page->index != 0);
1941 BUG_ON(len > ext4_get_max_inline_size(inode));
1942 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1943 if (inode_bh == NULL)
1946 page_bufs = page_buffers(page);
1951 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1954 /* As soon as we unlock the page, it can go away, but we have
1955 * references to buffers so we are safe */
1958 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1959 if (IS_ERR(handle)) {
1960 ret = PTR_ERR(handle);
1964 BUG_ON(!ext4_handle_valid(handle));
1967 ret = ext4_journal_get_write_access(handle, inode_bh);
1969 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1972 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1973 do_journal_get_write_access);
1975 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1980 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1981 err = ext4_journal_stop(handle);
1985 if (!ext4_has_inline_data(inode))
1986 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1988 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1995 * Note that we don't need to start a transaction unless we're journaling data
1996 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1997 * need to file the inode to the transaction's list in ordered mode because if
1998 * we are writing back data added by write(), the inode is already there and if
1999 * we are writing back data modified via mmap(), no one guarantees in which
2000 * transaction the data will hit the disk. In case we are journaling data, we
2001 * cannot start transaction directly because transaction start ranks above page
2002 * lock so we have to do some magic.
2004 * This function can get called via...
2005 * - ext4_da_writepages after taking page lock (have journal handle)
2006 * - journal_submit_inode_data_buffers (no journal handle)
2007 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2008 * - grab_page_cache when doing write_begin (have journal handle)
2010 * We don't do any block allocation in this function. If we have page with
2011 * multiple blocks we need to write those buffer_heads that are mapped. This
2012 * is important for mmaped based write. So if we do with blocksize 1K
2013 * truncate(f, 1024);
2014 * a = mmap(f, 0, 4096);
2016 * truncate(f, 4096);
2017 * we have in the page first buffer_head mapped via page_mkwrite call back
2018 * but other buffer_heads would be unmapped but dirty (dirty done via the
2019 * do_wp_page). So writepage should write the first block. If we modify
2020 * the mmap area beyond 1024 we will again get a page_fault and the
2021 * page_mkwrite callback will do the block allocation and mark the
2022 * buffer_heads mapped.
2024 * We redirty the page if we have any buffer_heads that is either delay or
2025 * unwritten in the page.
2027 * We can get recursively called as show below.
2029 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2032 * But since we don't do any block allocation we should not deadlock.
2033 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2035 static int ext4_writepage(struct page *page,
2036 struct writeback_control *wbc)
2038 int ret = 0, commit_write = 0;
2041 struct buffer_head *page_bufs = NULL;
2042 struct inode *inode = page->mapping->host;
2044 trace_ext4_writepage(page);
2045 size = i_size_read(inode);
2046 if (page->index == size >> PAGE_CACHE_SHIFT)
2047 len = size & ~PAGE_CACHE_MASK;
2049 len = PAGE_CACHE_SIZE;
2052 * If the page does not have buffers (for whatever reason),
2053 * try to create them using __block_write_begin. If this
2054 * fails, redirty the page and move on.
2056 if (!page_has_buffers(page)) {
2057 if (__block_write_begin(page, 0, len,
2058 noalloc_get_block_write)) {
2060 redirty_page_for_writepage(wbc, page);
2066 page_bufs = page_buffers(page);
2067 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2068 ext4_bh_delay_or_unwritten)) {
2070 * We don't want to do block allocation, so redirty
2071 * the page and return. We may reach here when we do
2072 * a journal commit via journal_submit_inode_data_buffers.
2073 * We can also reach here via shrink_page_list but it
2074 * should never be for direct reclaim so warn if that
2077 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
2082 /* now mark the buffer_heads as dirty and uptodate */
2083 block_commit_write(page, 0, len);
2085 if (PageChecked(page) && ext4_should_journal_data(inode))
2087 * It's mmapped pagecache. Add buffers and journal it. There
2088 * doesn't seem much point in redirtying the page here.
2090 return __ext4_journalled_writepage(page, len);
2092 if (buffer_uninit(page_bufs)) {
2093 ext4_set_bh_endio(page_bufs, inode);
2094 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2095 wbc, ext4_end_io_buffer_write);
2097 ret = block_write_full_page(page, noalloc_get_block_write,
2104 * This is called via ext4_da_writepages() to
2105 * calculate the total number of credits to reserve to fit
2106 * a single extent allocation into a single transaction,
2107 * ext4_da_writpeages() will loop calling this before
2108 * the block allocation.
2111 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2113 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2116 * With non-extent format the journal credit needed to
2117 * insert nrblocks contiguous block is dependent on
2118 * number of contiguous block. So we will limit
2119 * number of contiguous block to a sane value
2121 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2122 (max_blocks > EXT4_MAX_TRANS_DATA))
2123 max_blocks = EXT4_MAX_TRANS_DATA;
2125 return ext4_chunk_trans_blocks(inode, max_blocks);
2129 * write_cache_pages_da - walk the list of dirty pages of the given
2130 * address space and accumulate pages that need writing, and call
2131 * mpage_da_map_and_submit to map a single contiguous memory region
2132 * and then write them.
2134 static int write_cache_pages_da(handle_t *handle,
2135 struct address_space *mapping,
2136 struct writeback_control *wbc,
2137 struct mpage_da_data *mpd,
2138 pgoff_t *done_index)
2140 struct buffer_head *bh, *head;
2141 struct inode *inode = mapping->host;
2142 struct pagevec pvec;
2143 unsigned int nr_pages;
2146 long nr_to_write = wbc->nr_to_write;
2147 int i, tag, ret = 0;
2149 memset(mpd, 0, sizeof(struct mpage_da_data));
2152 pagevec_init(&pvec, 0);
2153 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2154 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2156 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2157 tag = PAGECACHE_TAG_TOWRITE;
2159 tag = PAGECACHE_TAG_DIRTY;
2161 *done_index = index;
2162 while (index <= end) {
2163 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2164 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2168 for (i = 0; i < nr_pages; i++) {
2169 struct page *page = pvec.pages[i];
2172 * At this point, the page may be truncated or
2173 * invalidated (changing page->mapping to NULL), or
2174 * even swizzled back from swapper_space to tmpfs file
2175 * mapping. However, page->index will not change
2176 * because we have a reference on the page.
2178 if (page->index > end)
2181 *done_index = page->index + 1;
2184 * If we can't merge this page, and we have
2185 * accumulated an contiguous region, write it
2187 if ((mpd->next_page != page->index) &&
2188 (mpd->next_page != mpd->first_page)) {
2189 mpage_da_map_and_submit(mpd);
2190 goto ret_extent_tail;
2196 * If the page is no longer dirty, or its
2197 * mapping no longer corresponds to inode we
2198 * are writing (which means it has been
2199 * truncated or invalidated), or the page is
2200 * already under writeback and we are not
2201 * doing a data integrity writeback, skip the page
2203 if (!PageDirty(page) ||
2204 (PageWriteback(page) &&
2205 (wbc->sync_mode == WB_SYNC_NONE)) ||
2206 unlikely(page->mapping != mapping)) {
2211 wait_on_page_writeback(page);
2212 BUG_ON(PageWriteback(page));
2215 * If we have inline data and arrive here, it means that
2216 * we will soon create the block for the 1st page, so
2217 * we'd better clear the inline data here.
2219 if (ext4_has_inline_data(inode)) {
2220 BUG_ON(ext4_test_inode_state(inode,
2221 EXT4_STATE_MAY_INLINE_DATA));
2222 ext4_destroy_inline_data(handle, inode);
2225 if (mpd->next_page != page->index)
2226 mpd->first_page = page->index;
2227 mpd->next_page = page->index + 1;
2228 logical = (sector_t) page->index <<
2229 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2231 if (!page_has_buffers(page)) {
2232 mpage_add_bh_to_extent(mpd, logical,
2234 (1 << BH_Dirty) | (1 << BH_Uptodate));
2236 goto ret_extent_tail;
2239 * Page with regular buffer heads,
2240 * just add all dirty ones
2242 head = page_buffers(page);
2245 BUG_ON(buffer_locked(bh));
2247 * We need to try to allocate
2248 * unmapped blocks in the same page.
2249 * Otherwise we won't make progress
2250 * with the page in ext4_writepage
2252 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2253 mpage_add_bh_to_extent(mpd, logical,
2257 goto ret_extent_tail;
2258 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2260 * mapped dirty buffer. We need
2261 * to update the b_state
2262 * because we look at b_state
2263 * in mpage_da_map_blocks. We
2264 * don't update b_size because
2265 * if we find an unmapped
2266 * buffer_head later we need to
2267 * use the b_state flag of that
2270 if (mpd->b_size == 0)
2271 mpd->b_state = bh->b_state & BH_FLAGS;
2274 } while ((bh = bh->b_this_page) != head);
2277 if (nr_to_write > 0) {
2279 if (nr_to_write == 0 &&
2280 wbc->sync_mode == WB_SYNC_NONE)
2282 * We stop writing back only if we are
2283 * not doing integrity sync. In case of
2284 * integrity sync we have to keep going
2285 * because someone may be concurrently
2286 * dirtying pages, and we might have
2287 * synced a lot of newly appeared dirty
2288 * pages, but have not synced all of the
2294 pagevec_release(&pvec);
2299 ret = MPAGE_DA_EXTENT_TAIL;
2301 pagevec_release(&pvec);
2307 static int ext4_da_writepages(struct address_space *mapping,
2308 struct writeback_control *wbc)
2311 int range_whole = 0;
2312 handle_t *handle = NULL;
2313 struct mpage_da_data mpd;
2314 struct inode *inode = mapping->host;
2315 int pages_written = 0;
2316 unsigned int max_pages;
2317 int range_cyclic, cycled = 1, io_done = 0;
2318 int needed_blocks, ret = 0;
2319 long desired_nr_to_write, nr_to_writebump = 0;
2320 loff_t range_start = wbc->range_start;
2321 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2322 pgoff_t done_index = 0;
2324 struct blk_plug plug;
2326 trace_ext4_da_writepages(inode, wbc);
2329 * No pages to write? This is mainly a kludge to avoid starting
2330 * a transaction for special inodes like journal inode on last iput()
2331 * because that could violate lock ordering on umount
2333 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2337 * If the filesystem has aborted, it is read-only, so return
2338 * right away instead of dumping stack traces later on that
2339 * will obscure the real source of the problem. We test
2340 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2341 * the latter could be true if the filesystem is mounted
2342 * read-only, and in that case, ext4_da_writepages should
2343 * *never* be called, so if that ever happens, we would want
2346 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2349 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2352 range_cyclic = wbc->range_cyclic;
2353 if (wbc->range_cyclic) {
2354 index = mapping->writeback_index;
2357 wbc->range_start = index << PAGE_CACHE_SHIFT;
2358 wbc->range_end = LLONG_MAX;
2359 wbc->range_cyclic = 0;
2362 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2363 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2367 * This works around two forms of stupidity. The first is in
2368 * the writeback code, which caps the maximum number of pages
2369 * written to be 1024 pages. This is wrong on multiple
2370 * levels; different architectues have a different page size,
2371 * which changes the maximum amount of data which gets
2372 * written. Secondly, 4 megabytes is way too small. XFS
2373 * forces this value to be 16 megabytes by multiplying
2374 * nr_to_write parameter by four, and then relies on its
2375 * allocator to allocate larger extents to make them
2376 * contiguous. Unfortunately this brings us to the second
2377 * stupidity, which is that ext4's mballoc code only allocates
2378 * at most 2048 blocks. So we force contiguous writes up to
2379 * the number of dirty blocks in the inode, or
2380 * sbi->max_writeback_mb_bump whichever is smaller.
2382 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2383 if (!range_cyclic && range_whole) {
2384 if (wbc->nr_to_write == LONG_MAX)
2385 desired_nr_to_write = wbc->nr_to_write;
2387 desired_nr_to_write = wbc->nr_to_write * 8;
2389 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2391 if (desired_nr_to_write > max_pages)
2392 desired_nr_to_write = max_pages;
2394 if (wbc->nr_to_write < desired_nr_to_write) {
2395 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2396 wbc->nr_to_write = desired_nr_to_write;
2400 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2401 tag_pages_for_writeback(mapping, index, end);
2403 blk_start_plug(&plug);
2404 while (!ret && wbc->nr_to_write > 0) {
2407 * we insert one extent at a time. So we need
2408 * credit needed for single extent allocation.
2409 * journalled mode is currently not supported
2412 BUG_ON(ext4_should_journal_data(inode));
2413 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2415 /* start a new transaction*/
2416 handle = ext4_journal_start(inode, needed_blocks);
2417 if (IS_ERR(handle)) {
2418 ret = PTR_ERR(handle);
2419 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2420 "%ld pages, ino %lu; err %d", __func__,
2421 wbc->nr_to_write, inode->i_ino, ret);
2422 blk_finish_plug(&plug);
2423 goto out_writepages;
2427 * Now call write_cache_pages_da() to find the next
2428 * contiguous region of logical blocks that need
2429 * blocks to be allocated by ext4 and submit them.
2431 ret = write_cache_pages_da(handle, mapping,
2432 wbc, &mpd, &done_index);
2434 * If we have a contiguous extent of pages and we
2435 * haven't done the I/O yet, map the blocks and submit
2438 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2439 mpage_da_map_and_submit(&mpd);
2440 ret = MPAGE_DA_EXTENT_TAIL;
2442 trace_ext4_da_write_pages(inode, &mpd);
2443 wbc->nr_to_write -= mpd.pages_written;
2445 ext4_journal_stop(handle);
2447 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2448 /* commit the transaction which would
2449 * free blocks released in the transaction
2452 jbd2_journal_force_commit_nested(sbi->s_journal);
2454 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2456 * Got one extent now try with rest of the pages.
2457 * If mpd.retval is set -EIO, journal is aborted.
2458 * So we don't need to write any more.
2460 pages_written += mpd.pages_written;
2463 } else if (wbc->nr_to_write)
2465 * There is no more writeout needed
2466 * or we requested for a noblocking writeout
2467 * and we found the device congested
2471 blk_finish_plug(&plug);
2472 if (!io_done && !cycled) {
2475 wbc->range_start = index << PAGE_CACHE_SHIFT;
2476 wbc->range_end = mapping->writeback_index - 1;
2481 wbc->range_cyclic = range_cyclic;
2482 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2484 * set the writeback_index so that range_cyclic
2485 * mode will write it back later
2487 mapping->writeback_index = done_index;
2490 wbc->nr_to_write -= nr_to_writebump;
2491 wbc->range_start = range_start;
2492 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2496 static int ext4_nonda_switch(struct super_block *sb)
2498 s64 free_blocks, dirty_blocks;
2499 struct ext4_sb_info *sbi = EXT4_SB(sb);
2502 * switch to non delalloc mode if we are running low
2503 * on free block. The free block accounting via percpu
2504 * counters can get slightly wrong with percpu_counter_batch getting
2505 * accumulated on each CPU without updating global counters
2506 * Delalloc need an accurate free block accounting. So switch
2507 * to non delalloc when we are near to error range.
2509 free_blocks = EXT4_C2B(sbi,
2510 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2511 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2513 * Start pushing delalloc when 1/2 of free blocks are dirty.
2515 if (dirty_blocks && (free_blocks < 2 * dirty_blocks))
2516 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2518 if (2 * free_blocks < 3 * dirty_blocks ||
2519 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2521 * free block count is less than 150% of dirty blocks
2522 * or free blocks is less than watermark
2529 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2530 loff_t pos, unsigned len, unsigned flags,
2531 struct page **pagep, void **fsdata)
2533 int ret, retries = 0;
2536 struct inode *inode = mapping->host;
2539 index = pos >> PAGE_CACHE_SHIFT;
2541 if (ext4_nonda_switch(inode->i_sb)) {
2542 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2543 return ext4_write_begin(file, mapping, pos,
2544 len, flags, pagep, fsdata);
2546 *fsdata = (void *)0;
2547 trace_ext4_da_write_begin(inode, pos, len, flags);
2549 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2550 ret = ext4_da_write_inline_data_begin(mapping, inode,
2563 * With delayed allocation, we don't log the i_disksize update
2564 * if there is delayed block allocation. But we still need
2565 * to journalling the i_disksize update if writes to the end
2566 * of file which has an already mapped buffer.
2568 handle = ext4_journal_start(inode, 1);
2569 if (IS_ERR(handle)) {
2570 ret = PTR_ERR(handle);
2573 /* We cannot recurse into the filesystem as the transaction is already
2575 flags |= AOP_FLAG_NOFS;
2577 page = grab_cache_page_write_begin(mapping, index, flags);
2579 ext4_journal_stop(handle);
2585 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2588 ext4_journal_stop(handle);
2589 page_cache_release(page);
2591 * block_write_begin may have instantiated a few blocks
2592 * outside i_size. Trim these off again. Don't need
2593 * i_size_read because we hold i_mutex.
2595 if (pos + len > inode->i_size)
2596 ext4_truncate_failed_write(inode);
2599 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2606 * Check if we should update i_disksize
2607 * when write to the end of file but not require block allocation
2609 static int ext4_da_should_update_i_disksize(struct page *page,
2610 unsigned long offset)
2612 struct buffer_head *bh;
2613 struct inode *inode = page->mapping->host;
2617 bh = page_buffers(page);
2618 idx = offset >> inode->i_blkbits;
2620 for (i = 0; i < idx; i++)
2621 bh = bh->b_this_page;
2623 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2628 static int ext4_da_write_end(struct file *file,
2629 struct address_space *mapping,
2630 loff_t pos, unsigned len, unsigned copied,
2631 struct page *page, void *fsdata)
2633 struct inode *inode = mapping->host;
2635 handle_t *handle = ext4_journal_current_handle();
2637 unsigned long start, end;
2638 int write_mode = (int)(unsigned long)fsdata;
2640 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2641 switch (ext4_inode_journal_mode(inode)) {
2642 case EXT4_INODE_ORDERED_DATA_MODE:
2643 return ext4_ordered_write_end(file, mapping, pos,
2644 len, copied, page, fsdata);
2645 case EXT4_INODE_WRITEBACK_DATA_MODE:
2646 return ext4_writeback_write_end(file, mapping, pos,
2647 len, copied, page, fsdata);
2653 trace_ext4_da_write_end(inode, pos, len, copied);
2654 start = pos & (PAGE_CACHE_SIZE - 1);
2655 end = start + copied - 1;
2658 * generic_write_end() will run mark_inode_dirty() if i_size
2659 * changes. So let's piggyback the i_disksize mark_inode_dirty
2662 new_i_size = pos + copied;
2663 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2664 if (ext4_has_inline_data(inode) ||
2665 ext4_da_should_update_i_disksize(page, end)) {
2666 down_write(&EXT4_I(inode)->i_data_sem);
2667 if (new_i_size > EXT4_I(inode)->i_disksize)
2668 EXT4_I(inode)->i_disksize = new_i_size;
2669 up_write(&EXT4_I(inode)->i_data_sem);
2670 /* We need to mark inode dirty even if
2671 * new_i_size is less that inode->i_size
2672 * bu greater than i_disksize.(hint delalloc)
2674 ext4_mark_inode_dirty(handle, inode);
2678 if (write_mode != CONVERT_INLINE_DATA &&
2679 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2680 ext4_has_inline_data(inode))
2681 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2684 ret2 = generic_write_end(file, mapping, pos, len, copied,
2690 ret2 = ext4_journal_stop(handle);
2694 return ret ? ret : copied;
2697 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2700 * Drop reserved blocks
2702 BUG_ON(!PageLocked(page));
2703 if (!page_has_buffers(page))
2706 ext4_da_page_release_reservation(page, offset);
2709 ext4_invalidatepage(page, offset);
2715 * Force all delayed allocation blocks to be allocated for a given inode.
2717 int ext4_alloc_da_blocks(struct inode *inode)
2719 trace_ext4_alloc_da_blocks(inode);
2721 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2722 !EXT4_I(inode)->i_reserved_meta_blocks)
2726 * We do something simple for now. The filemap_flush() will
2727 * also start triggering a write of the data blocks, which is
2728 * not strictly speaking necessary (and for users of
2729 * laptop_mode, not even desirable). However, to do otherwise
2730 * would require replicating code paths in:
2732 * ext4_da_writepages() ->
2733 * write_cache_pages() ---> (via passed in callback function)
2734 * __mpage_da_writepage() -->
2735 * mpage_add_bh_to_extent()
2736 * mpage_da_map_blocks()
2738 * The problem is that write_cache_pages(), located in
2739 * mm/page-writeback.c, marks pages clean in preparation for
2740 * doing I/O, which is not desirable if we're not planning on
2743 * We could call write_cache_pages(), and then redirty all of
2744 * the pages by calling redirty_page_for_writepage() but that
2745 * would be ugly in the extreme. So instead we would need to
2746 * replicate parts of the code in the above functions,
2747 * simplifying them because we wouldn't actually intend to
2748 * write out the pages, but rather only collect contiguous
2749 * logical block extents, call the multi-block allocator, and
2750 * then update the buffer heads with the block allocations.
2752 * For now, though, we'll cheat by calling filemap_flush(),
2753 * which will map the blocks, and start the I/O, but not
2754 * actually wait for the I/O to complete.
2756 return filemap_flush(inode->i_mapping);
2760 * bmap() is special. It gets used by applications such as lilo and by
2761 * the swapper to find the on-disk block of a specific piece of data.
2763 * Naturally, this is dangerous if the block concerned is still in the
2764 * journal. If somebody makes a swapfile on an ext4 data-journaling
2765 * filesystem and enables swap, then they may get a nasty shock when the
2766 * data getting swapped to that swapfile suddenly gets overwritten by
2767 * the original zero's written out previously to the journal and
2768 * awaiting writeback in the kernel's buffer cache.
2770 * So, if we see any bmap calls here on a modified, data-journaled file,
2771 * take extra steps to flush any blocks which might be in the cache.
2773 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2775 struct inode *inode = mapping->host;
2780 * We can get here for an inline file via the FIBMAP ioctl
2782 if (ext4_has_inline_data(inode))
2785 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2786 test_opt(inode->i_sb, DELALLOC)) {
2788 * With delalloc we want to sync the file
2789 * so that we can make sure we allocate
2792 filemap_write_and_wait(mapping);
2795 if (EXT4_JOURNAL(inode) &&
2796 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2798 * This is a REALLY heavyweight approach, but the use of
2799 * bmap on dirty files is expected to be extremely rare:
2800 * only if we run lilo or swapon on a freshly made file
2801 * do we expect this to happen.
2803 * (bmap requires CAP_SYS_RAWIO so this does not
2804 * represent an unprivileged user DOS attack --- we'd be
2805 * in trouble if mortal users could trigger this path at
2808 * NB. EXT4_STATE_JDATA is not set on files other than
2809 * regular files. If somebody wants to bmap a directory
2810 * or symlink and gets confused because the buffer
2811 * hasn't yet been flushed to disk, they deserve
2812 * everything they get.
2815 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2816 journal = EXT4_JOURNAL(inode);
2817 jbd2_journal_lock_updates(journal);
2818 err = jbd2_journal_flush(journal);
2819 jbd2_journal_unlock_updates(journal);
2825 return generic_block_bmap(mapping, block, ext4_get_block);
2828 static int ext4_readpage(struct file *file, struct page *page)
2831 struct inode *inode = page->mapping->host;
2833 trace_ext4_readpage(page);
2835 if (ext4_has_inline_data(inode))
2836 ret = ext4_readpage_inline(inode, page);
2839 return mpage_readpage(page, ext4_get_block);
2845 ext4_readpages(struct file *file, struct address_space *mapping,
2846 struct list_head *pages, unsigned nr_pages)
2848 struct inode *inode = mapping->host;
2850 /* If the file has inline data, no need to do readpages. */
2851 if (ext4_has_inline_data(inode))
2854 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2857 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
2859 struct buffer_head *head, *bh;
2860 unsigned int curr_off = 0;
2862 if (!page_has_buffers(page))
2864 head = bh = page_buffers(page);
2866 if (offset <= curr_off && test_clear_buffer_uninit(bh)
2868 ext4_free_io_end(bh->b_private);
2869 bh->b_private = NULL;
2870 bh->b_end_io = NULL;
2872 curr_off = curr_off + bh->b_size;
2873 bh = bh->b_this_page;
2874 } while (bh != head);
2877 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2879 trace_ext4_invalidatepage(page, offset);
2882 * free any io_end structure allocated for buffers to be discarded
2884 if (ext4_should_dioread_nolock(page->mapping->host))
2885 ext4_invalidatepage_free_endio(page, offset);
2887 /* No journalling happens on data buffers when this function is used */
2888 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2890 block_invalidatepage(page, offset);
2893 static int __ext4_journalled_invalidatepage(struct page *page,
2894 unsigned long offset)
2896 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2898 trace_ext4_journalled_invalidatepage(page, offset);
2901 * If it's a full truncate we just forget about the pending dirtying
2904 ClearPageChecked(page);
2906 return jbd2_journal_invalidatepage(journal, page, offset);
2909 /* Wrapper for aops... */
2910 static void ext4_journalled_invalidatepage(struct page *page,
2911 unsigned long offset)
2913 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
2916 static int ext4_releasepage(struct page *page, gfp_t wait)
2918 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2920 trace_ext4_releasepage(page);
2922 WARN_ON(PageChecked(page));
2923 if (!page_has_buffers(page))
2926 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2928 return try_to_free_buffers(page);
2932 * ext4_get_block used when preparing for a DIO write or buffer write.
2933 * We allocate an uinitialized extent if blocks haven't been allocated.
2934 * The extent will be converted to initialized after the IO is complete.
2936 int ext4_get_block_write(struct inode *inode, sector_t iblock,
2937 struct buffer_head *bh_result, int create)
2939 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2940 inode->i_ino, create);
2941 return _ext4_get_block(inode, iblock, bh_result,
2942 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2945 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2946 struct buffer_head *bh_result, int create)
2948 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2949 inode->i_ino, create);
2950 return _ext4_get_block(inode, iblock, bh_result,
2951 EXT4_GET_BLOCKS_NO_LOCK);
2954 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2955 ssize_t size, void *private, int ret,
2958 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2959 ext4_io_end_t *io_end = iocb->private;
2961 /* if not async direct IO or dio with 0 bytes write, just return */
2962 if (!io_end || !size)
2965 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2966 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2967 iocb->private, io_end->inode->i_ino, iocb, offset,
2970 iocb->private = NULL;
2972 /* if not aio dio with unwritten extents, just free io and return */
2973 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2974 ext4_free_io_end(io_end);
2977 aio_complete(iocb, ret, 0);
2978 inode_dio_done(inode);
2982 io_end->offset = offset;
2983 io_end->size = size;
2985 io_end->iocb = iocb;
2986 io_end->result = ret;
2989 ext4_add_complete_io(io_end);
2992 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
2994 ext4_io_end_t *io_end = bh->b_private;
2995 struct inode *inode;
2997 if (!test_clear_buffer_uninit(bh) || !io_end)
3000 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3001 ext4_msg(io_end->inode->i_sb, KERN_INFO,
3002 "sb umounted, discard end_io request for inode %lu",
3003 io_end->inode->i_ino);
3004 ext4_free_io_end(io_end);
3009 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
3010 * but being more careful is always safe for the future change.
3012 inode = io_end->inode;
3013 ext4_set_io_unwritten_flag(inode, io_end);
3014 ext4_add_complete_io(io_end);
3016 bh->b_private = NULL;
3017 bh->b_end_io = NULL;
3018 clear_buffer_uninit(bh);
3019 end_buffer_async_write(bh, uptodate);
3022 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3024 ext4_io_end_t *io_end;
3025 struct page *page = bh->b_page;
3026 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3027 size_t size = bh->b_size;
3030 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3032 pr_warn_ratelimited("%s: allocation fail\n", __func__);
3036 io_end->offset = offset;
3037 io_end->size = size;
3039 * We need to hold a reference to the page to make sure it
3040 * doesn't get evicted before ext4_end_io_work() has a chance
3041 * to convert the extent from written to unwritten.
3043 io_end->page = page;
3044 get_page(io_end->page);
3046 bh->b_private = io_end;
3047 bh->b_end_io = ext4_end_io_buffer_write;
3052 * For ext4 extent files, ext4 will do direct-io write to holes,
3053 * preallocated extents, and those write extend the file, no need to
3054 * fall back to buffered IO.
3056 * For holes, we fallocate those blocks, mark them as uninitialized
3057 * If those blocks were preallocated, we mark sure they are split, but
3058 * still keep the range to write as uninitialized.
3060 * The unwritten extents will be converted to written when DIO is completed.
3061 * For async direct IO, since the IO may still pending when return, we
3062 * set up an end_io call back function, which will do the conversion
3063 * when async direct IO completed.
3065 * If the O_DIRECT write will extend the file then add this inode to the
3066 * orphan list. So recovery will truncate it back to the original size
3067 * if the machine crashes during the write.
3070 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3071 const struct iovec *iov, loff_t offset,
3072 unsigned long nr_segs)
3074 struct file *file = iocb->ki_filp;
3075 struct inode *inode = file->f_mapping->host;
3077 size_t count = iov_length(iov, nr_segs);
3079 get_block_t *get_block_func = NULL;
3081 loff_t final_size = offset + count;
3083 /* Use the old path for reads and writes beyond i_size. */
3084 if (rw != WRITE || final_size > inode->i_size)
3085 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3087 BUG_ON(iocb->private == NULL);
3089 /* If we do a overwrite dio, i_mutex locking can be released */
3090 overwrite = *((int *)iocb->private);
3093 atomic_inc(&inode->i_dio_count);
3094 down_read(&EXT4_I(inode)->i_data_sem);
3095 mutex_unlock(&inode->i_mutex);
3099 * We could direct write to holes and fallocate.
3101 * Allocated blocks to fill the hole are marked as
3102 * uninitialized to prevent parallel buffered read to expose
3103 * the stale data before DIO complete the data IO.
3105 * As to previously fallocated extents, ext4 get_block will
3106 * just simply mark the buffer mapped but still keep the
3107 * extents uninitialized.
3109 * For non AIO case, we will convert those unwritten extents
3110 * to written after return back from blockdev_direct_IO.
3112 * For async DIO, the conversion needs to be deferred when the
3113 * IO is completed. The ext4 end_io callback function will be
3114 * called to take care of the conversion work. Here for async
3115 * case, we allocate an io_end structure to hook to the iocb.
3117 iocb->private = NULL;
3118 ext4_inode_aio_set(inode, NULL);
3119 if (!is_sync_kiocb(iocb)) {
3120 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
3125 io_end->flag |= EXT4_IO_END_DIRECT;
3126 iocb->private = io_end;
3128 * we save the io structure for current async direct
3129 * IO, so that later ext4_map_blocks() could flag the
3130 * io structure whether there is a unwritten extents
3131 * needs to be converted when IO is completed.
3133 ext4_inode_aio_set(inode, io_end);
3137 get_block_func = ext4_get_block_write_nolock;
3139 get_block_func = ext4_get_block_write;
3140 dio_flags = DIO_LOCKING;
3142 ret = __blockdev_direct_IO(rw, iocb, inode,
3143 inode->i_sb->s_bdev, iov,
3151 ext4_inode_aio_set(inode, NULL);
3153 * The io_end structure takes a reference to the inode, that
3154 * structure needs to be destroyed and the reference to the
3155 * inode need to be dropped, when IO is complete, even with 0
3156 * byte write, or failed.
3158 * In the successful AIO DIO case, the io_end structure will
3159 * be destroyed and the reference to the inode will be dropped
3160 * after the end_io call back function is called.
3162 * In the case there is 0 byte write, or error case, since VFS
3163 * direct IO won't invoke the end_io call back function, we
3164 * need to free the end_io structure here.
3166 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3167 ext4_free_io_end(iocb->private);
3168 iocb->private = NULL;
3169 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3170 EXT4_STATE_DIO_UNWRITTEN)) {
3173 * for non AIO case, since the IO is already
3174 * completed, we could do the conversion right here
3176 err = ext4_convert_unwritten_extents(inode,
3180 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3184 /* take i_mutex locking again if we do a ovewrite dio */
3186 inode_dio_done(inode);
3187 up_read(&EXT4_I(inode)->i_data_sem);
3188 mutex_lock(&inode->i_mutex);
3194 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3195 const struct iovec *iov, loff_t offset,
3196 unsigned long nr_segs)
3198 struct file *file = iocb->ki_filp;
3199 struct inode *inode = file->f_mapping->host;
3203 * If we are doing data journalling we don't support O_DIRECT
3205 if (ext4_should_journal_data(inode))
3208 /* Let buffer I/O handle the inline data case. */
3209 if (ext4_has_inline_data(inode))
3212 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3213 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3214 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3216 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3217 trace_ext4_direct_IO_exit(inode, offset,
3218 iov_length(iov, nr_segs), rw, ret);
3223 * Pages can be marked dirty completely asynchronously from ext4's journalling
3224 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3225 * much here because ->set_page_dirty is called under VFS locks. The page is
3226 * not necessarily locked.
3228 * We cannot just dirty the page and leave attached buffers clean, because the
3229 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3230 * or jbddirty because all the journalling code will explode.
3232 * So what we do is to mark the page "pending dirty" and next time writepage
3233 * is called, propagate that into the buffers appropriately.
3235 static int ext4_journalled_set_page_dirty(struct page *page)
3237 SetPageChecked(page);
3238 return __set_page_dirty_nobuffers(page);
3241 static const struct address_space_operations ext4_ordered_aops = {
3242 .readpage = ext4_readpage,
3243 .readpages = ext4_readpages,
3244 .writepage = ext4_writepage,
3245 .write_begin = ext4_write_begin,
3246 .write_end = ext4_ordered_write_end,
3248 .invalidatepage = ext4_invalidatepage,
3249 .releasepage = ext4_releasepage,
3250 .direct_IO = ext4_direct_IO,
3251 .migratepage = buffer_migrate_page,
3252 .is_partially_uptodate = block_is_partially_uptodate,
3253 .error_remove_page = generic_error_remove_page,
3256 static const struct address_space_operations ext4_writeback_aops = {
3257 .readpage = ext4_readpage,
3258 .readpages = ext4_readpages,
3259 .writepage = ext4_writepage,
3260 .write_begin = ext4_write_begin,
3261 .write_end = ext4_writeback_write_end,
3263 .invalidatepage = ext4_invalidatepage,
3264 .releasepage = ext4_releasepage,
3265 .direct_IO = ext4_direct_IO,
3266 .migratepage = buffer_migrate_page,
3267 .is_partially_uptodate = block_is_partially_uptodate,
3268 .error_remove_page = generic_error_remove_page,
3271 static const struct address_space_operations ext4_journalled_aops = {
3272 .readpage = ext4_readpage,
3273 .readpages = ext4_readpages,
3274 .writepage = ext4_writepage,
3275 .write_begin = ext4_write_begin,
3276 .write_end = ext4_journalled_write_end,
3277 .set_page_dirty = ext4_journalled_set_page_dirty,
3279 .invalidatepage = ext4_journalled_invalidatepage,
3280 .releasepage = ext4_releasepage,
3281 .direct_IO = ext4_direct_IO,
3282 .is_partially_uptodate = block_is_partially_uptodate,
3283 .error_remove_page = generic_error_remove_page,
3286 static const struct address_space_operations ext4_da_aops = {
3287 .readpage = ext4_readpage,
3288 .readpages = ext4_readpages,
3289 .writepage = ext4_writepage,
3290 .writepages = ext4_da_writepages,
3291 .write_begin = ext4_da_write_begin,
3292 .write_end = ext4_da_write_end,
3294 .invalidatepage = ext4_da_invalidatepage,
3295 .releasepage = ext4_releasepage,
3296 .direct_IO = ext4_direct_IO,
3297 .migratepage = buffer_migrate_page,
3298 .is_partially_uptodate = block_is_partially_uptodate,
3299 .error_remove_page = generic_error_remove_page,
3302 void ext4_set_aops(struct inode *inode)
3304 switch (ext4_inode_journal_mode(inode)) {
3305 case EXT4_INODE_ORDERED_DATA_MODE:
3306 if (test_opt(inode->i_sb, DELALLOC))
3307 inode->i_mapping->a_ops = &ext4_da_aops;
3309 inode->i_mapping->a_ops = &ext4_ordered_aops;
3311 case EXT4_INODE_WRITEBACK_DATA_MODE:
3312 if (test_opt(inode->i_sb, DELALLOC))
3313 inode->i_mapping->a_ops = &ext4_da_aops;
3315 inode->i_mapping->a_ops = &ext4_writeback_aops;
3317 case EXT4_INODE_JOURNAL_DATA_MODE:
3318 inode->i_mapping->a_ops = &ext4_journalled_aops;
3327 * ext4_discard_partial_page_buffers()
3328 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3329 * This function finds and locks the page containing the offset
3330 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3331 * Calling functions that already have the page locked should call
3332 * ext4_discard_partial_page_buffers_no_lock directly.
3334 int ext4_discard_partial_page_buffers(handle_t *handle,
3335 struct address_space *mapping, loff_t from,
3336 loff_t length, int flags)
3338 struct inode *inode = mapping->host;
3342 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3343 mapping_gfp_mask(mapping) & ~__GFP_FS);
3347 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3348 from, length, flags);
3351 page_cache_release(page);
3356 * ext4_discard_partial_page_buffers_no_lock()
3357 * Zeros a page range of length 'length' starting from offset 'from'.
3358 * Buffer heads that correspond to the block aligned regions of the
3359 * zeroed range will be unmapped. Unblock aligned regions
3360 * will have the corresponding buffer head mapped if needed so that
3361 * that region of the page can be updated with the partial zero out.
3363 * This function assumes that the page has already been locked. The
3364 * The range to be discarded must be contained with in the given page.
3365 * If the specified range exceeds the end of the page it will be shortened
3366 * to the end of the page that corresponds to 'from'. This function is
3367 * appropriate for updating a page and it buffer heads to be unmapped and
3368 * zeroed for blocks that have been either released, or are going to be
3371 * handle: The journal handle
3372 * inode: The files inode
3373 * page: A locked page that contains the offset "from"
3374 * from: The starting byte offset (from the beginning of the file)
3375 * to begin discarding
3376 * len: The length of bytes to discard
3377 * flags: Optional flags that may be used:
3379 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3380 * Only zero the regions of the page whose buffer heads
3381 * have already been unmapped. This flag is appropriate
3382 * for updating the contents of a page whose blocks may
3383 * have already been released, and we only want to zero
3384 * out the regions that correspond to those released blocks.
3386 * Returns zero on success or negative on failure.
3388 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3389 struct inode *inode, struct page *page, loff_t from,
3390 loff_t length, int flags)
3392 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3393 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3394 unsigned int blocksize, max, pos;
3396 struct buffer_head *bh;
3399 blocksize = inode->i_sb->s_blocksize;
3400 max = PAGE_CACHE_SIZE - offset;
3402 if (index != page->index)
3406 * correct length if it does not fall between
3407 * 'from' and the end of the page
3409 if (length > max || length < 0)
3412 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3414 if (!page_has_buffers(page))
3415 create_empty_buffers(page, blocksize, 0);
3417 /* Find the buffer that contains "offset" */
3418 bh = page_buffers(page);
3420 while (offset >= pos) {
3421 bh = bh->b_this_page;
3427 while (pos < offset + length) {
3428 unsigned int end_of_block, range_to_discard;
3432 /* The length of space left to zero and unmap */
3433 range_to_discard = offset + length - pos;
3435 /* The length of space until the end of the block */
3436 end_of_block = blocksize - (pos & (blocksize-1));
3439 * Do not unmap or zero past end of block
3440 * for this buffer head
3442 if (range_to_discard > end_of_block)
3443 range_to_discard = end_of_block;
3447 * Skip this buffer head if we are only zeroing unampped
3448 * regions of the page
3450 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3454 /* If the range is block aligned, unmap */
3455 if (range_to_discard == blocksize) {
3456 clear_buffer_dirty(bh);
3458 clear_buffer_mapped(bh);
3459 clear_buffer_req(bh);
3460 clear_buffer_new(bh);
3461 clear_buffer_delay(bh);
3462 clear_buffer_unwritten(bh);
3463 clear_buffer_uptodate(bh);
3464 zero_user(page, pos, range_to_discard);
3465 BUFFER_TRACE(bh, "Buffer discarded");
3470 * If this block is not completely contained in the range
3471 * to be discarded, then it is not going to be released. Because
3472 * we need to keep this block, we need to make sure this part
3473 * of the page is uptodate before we modify it by writeing
3474 * partial zeros on it.
3476 if (!buffer_mapped(bh)) {
3478 * Buffer head must be mapped before we can read
3481 BUFFER_TRACE(bh, "unmapped");
3482 ext4_get_block(inode, iblock, bh, 0);
3483 /* unmapped? It's a hole - nothing to do */
3484 if (!buffer_mapped(bh)) {
3485 BUFFER_TRACE(bh, "still unmapped");
3490 /* Ok, it's mapped. Make sure it's up-to-date */
3491 if (PageUptodate(page))
3492 set_buffer_uptodate(bh);
3494 if (!buffer_uptodate(bh)) {
3496 ll_rw_block(READ, 1, &bh);
3498 /* Uhhuh. Read error. Complain and punt.*/
3499 if (!buffer_uptodate(bh))
3503 if (ext4_should_journal_data(inode)) {
3504 BUFFER_TRACE(bh, "get write access");
3505 err = ext4_journal_get_write_access(handle, bh);
3510 zero_user(page, pos, range_to_discard);
3513 if (ext4_should_journal_data(inode)) {
3514 err = ext4_handle_dirty_metadata(handle, inode, bh);
3516 mark_buffer_dirty(bh);
3518 BUFFER_TRACE(bh, "Partial buffer zeroed");
3520 bh = bh->b_this_page;
3522 pos += range_to_discard;
3528 int ext4_can_truncate(struct inode *inode)
3530 if (S_ISREG(inode->i_mode))
3532 if (S_ISDIR(inode->i_mode))
3534 if (S_ISLNK(inode->i_mode))
3535 return !ext4_inode_is_fast_symlink(inode);
3540 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3541 * associated with the given offset and length
3543 * @inode: File inode
3544 * @offset: The offset where the hole will begin
3545 * @len: The length of the hole
3547 * Returns: 0 on success or negative on failure
3550 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3552 struct inode *inode = file->f_path.dentry->d_inode;
3553 if (!S_ISREG(inode->i_mode))
3556 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3557 /* TODO: Add support for non extent hole punching */
3561 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3562 /* TODO: Add support for bigalloc file systems */
3566 return ext4_ext_punch_hole(file, offset, length);
3572 * We block out ext4_get_block() block instantiations across the entire
3573 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3574 * simultaneously on behalf of the same inode.
3576 * As we work through the truncate and commit bits of it to the journal there
3577 * is one core, guiding principle: the file's tree must always be consistent on
3578 * disk. We must be able to restart the truncate after a crash.
3580 * The file's tree may be transiently inconsistent in memory (although it
3581 * probably isn't), but whenever we close off and commit a journal transaction,
3582 * the contents of (the filesystem + the journal) must be consistent and
3583 * restartable. It's pretty simple, really: bottom up, right to left (although
3584 * left-to-right works OK too).
3586 * Note that at recovery time, journal replay occurs *before* the restart of
3587 * truncate against the orphan inode list.
3589 * The committed inode has the new, desired i_size (which is the same as
3590 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3591 * that this inode's truncate did not complete and it will again call
3592 * ext4_truncate() to have another go. So there will be instantiated blocks
3593 * to the right of the truncation point in a crashed ext4 filesystem. But
3594 * that's fine - as long as they are linked from the inode, the post-crash
3595 * ext4_truncate() run will find them and release them.
3597 void ext4_truncate(struct inode *inode)
3599 trace_ext4_truncate_enter(inode);
3601 if (!ext4_can_truncate(inode))
3604 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3606 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3607 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3609 if (ext4_has_inline_data(inode)) {
3612 ext4_inline_data_truncate(inode, &has_inline);
3617 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3618 ext4_ext_truncate(inode);
3620 ext4_ind_truncate(inode);
3622 trace_ext4_truncate_exit(inode);
3626 * ext4_get_inode_loc returns with an extra refcount against the inode's
3627 * underlying buffer_head on success. If 'in_mem' is true, we have all
3628 * data in memory that is needed to recreate the on-disk version of this
3631 static int __ext4_get_inode_loc(struct inode *inode,
3632 struct ext4_iloc *iloc, int in_mem)
3634 struct ext4_group_desc *gdp;
3635 struct buffer_head *bh;
3636 struct super_block *sb = inode->i_sb;
3638 int inodes_per_block, inode_offset;
3641 if (!ext4_valid_inum(sb, inode->i_ino))
3644 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3645 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3650 * Figure out the offset within the block group inode table
3652 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3653 inode_offset = ((inode->i_ino - 1) %
3654 EXT4_INODES_PER_GROUP(sb));
3655 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3656 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3658 bh = sb_getblk(sb, block);
3660 EXT4_ERROR_INODE_BLOCK(inode, block,
3661 "unable to read itable block");
3664 if (!buffer_uptodate(bh)) {
3668 * If the buffer has the write error flag, we have failed
3669 * to write out another inode in the same block. In this
3670 * case, we don't have to read the block because we may
3671 * read the old inode data successfully.
3673 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3674 set_buffer_uptodate(bh);
3676 if (buffer_uptodate(bh)) {
3677 /* someone brought it uptodate while we waited */
3683 * If we have all information of the inode in memory and this
3684 * is the only valid inode in the block, we need not read the
3688 struct buffer_head *bitmap_bh;
3691 start = inode_offset & ~(inodes_per_block - 1);
3693 /* Is the inode bitmap in cache? */
3694 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3699 * If the inode bitmap isn't in cache then the
3700 * optimisation may end up performing two reads instead
3701 * of one, so skip it.
3703 if (!buffer_uptodate(bitmap_bh)) {
3707 for (i = start; i < start + inodes_per_block; i++) {
3708 if (i == inode_offset)
3710 if (ext4_test_bit(i, bitmap_bh->b_data))
3714 if (i == start + inodes_per_block) {
3715 /* all other inodes are free, so skip I/O */
3716 memset(bh->b_data, 0, bh->b_size);
3717 set_buffer_uptodate(bh);
3725 * If we need to do any I/O, try to pre-readahead extra
3726 * blocks from the inode table.
3728 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3729 ext4_fsblk_t b, end, table;
3732 table = ext4_inode_table(sb, gdp);
3733 /* s_inode_readahead_blks is always a power of 2 */
3734 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3737 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3738 num = EXT4_INODES_PER_GROUP(sb);
3739 if (ext4_has_group_desc_csum(sb))
3740 num -= ext4_itable_unused_count(sb, gdp);
3741 table += num / inodes_per_block;
3745 sb_breadahead(sb, b++);
3749 * There are other valid inodes in the buffer, this inode
3750 * has in-inode xattrs, or we don't have this inode in memory.
3751 * Read the block from disk.
3753 trace_ext4_load_inode(inode);
3755 bh->b_end_io = end_buffer_read_sync;
3756 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3758 if (!buffer_uptodate(bh)) {
3759 EXT4_ERROR_INODE_BLOCK(inode, block,
3760 "unable to read itable block");
3770 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3772 /* We have all inode data except xattrs in memory here. */
3773 return __ext4_get_inode_loc(inode, iloc,
3774 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3777 void ext4_set_inode_flags(struct inode *inode)
3779 unsigned int flags = EXT4_I(inode)->i_flags;
3781 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3782 if (flags & EXT4_SYNC_FL)
3783 inode->i_flags |= S_SYNC;
3784 if (flags & EXT4_APPEND_FL)
3785 inode->i_flags |= S_APPEND;
3786 if (flags & EXT4_IMMUTABLE_FL)
3787 inode->i_flags |= S_IMMUTABLE;
3788 if (flags & EXT4_NOATIME_FL)
3789 inode->i_flags |= S_NOATIME;
3790 if (flags & EXT4_DIRSYNC_FL)
3791 inode->i_flags |= S_DIRSYNC;
3794 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3795 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3797 unsigned int vfs_fl;
3798 unsigned long old_fl, new_fl;
3801 vfs_fl = ei->vfs_inode.i_flags;
3802 old_fl = ei->i_flags;
3803 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3804 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3806 if (vfs_fl & S_SYNC)
3807 new_fl |= EXT4_SYNC_FL;
3808 if (vfs_fl & S_APPEND)
3809 new_fl |= EXT4_APPEND_FL;
3810 if (vfs_fl & S_IMMUTABLE)
3811 new_fl |= EXT4_IMMUTABLE_FL;
3812 if (vfs_fl & S_NOATIME)
3813 new_fl |= EXT4_NOATIME_FL;
3814 if (vfs_fl & S_DIRSYNC)
3815 new_fl |= EXT4_DIRSYNC_FL;
3816 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3819 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3820 struct ext4_inode_info *ei)
3823 struct inode *inode = &(ei->vfs_inode);
3824 struct super_block *sb = inode->i_sb;
3826 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3827 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3828 /* we are using combined 48 bit field */
3829 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3830 le32_to_cpu(raw_inode->i_blocks_lo);
3831 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3832 /* i_blocks represent file system block size */
3833 return i_blocks << (inode->i_blkbits - 9);
3838 return le32_to_cpu(raw_inode->i_blocks_lo);
3842 static inline void ext4_iget_extra_inode(struct inode *inode,
3843 struct ext4_inode *raw_inode,
3844 struct ext4_inode_info *ei)
3846 __le32 *magic = (void *)raw_inode +
3847 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3848 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3849 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3850 ext4_find_inline_data_nolock(inode);
3852 EXT4_I(inode)->i_inline_off = 0;
3855 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3857 struct ext4_iloc iloc;
3858 struct ext4_inode *raw_inode;
3859 struct ext4_inode_info *ei;
3860 struct inode *inode;
3861 journal_t *journal = EXT4_SB(sb)->s_journal;
3867 inode = iget_locked(sb, ino);
3869 return ERR_PTR(-ENOMEM);
3870 if (!(inode->i_state & I_NEW))
3876 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3879 raw_inode = ext4_raw_inode(&iloc);
3881 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3882 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3883 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3884 EXT4_INODE_SIZE(inode->i_sb)) {
3885 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3886 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3887 EXT4_INODE_SIZE(inode->i_sb));
3892 ei->i_extra_isize = 0;
3894 /* Precompute checksum seed for inode metadata */
3895 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3896 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3897 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3899 __le32 inum = cpu_to_le32(inode->i_ino);
3900 __le32 gen = raw_inode->i_generation;
3901 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3903 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3907 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3908 EXT4_ERROR_INODE(inode, "checksum invalid");
3913 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3914 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3915 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3916 if (!(test_opt(inode->i_sb, NO_UID32))) {
3917 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3918 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3920 i_uid_write(inode, i_uid);
3921 i_gid_write(inode, i_gid);
3922 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3924 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3925 ei->i_inline_off = 0;
3926 ei->i_dir_start_lookup = 0;
3927 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3928 /* We now have enough fields to check if the inode was active or not.
3929 * This is needed because nfsd might try to access dead inodes
3930 * the test is that same one that e2fsck uses
3931 * NeilBrown 1999oct15
3933 if (inode->i_nlink == 0) {
3934 if (inode->i_mode == 0 ||
3935 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3936 /* this inode is deleted */
3940 /* The only unlinked inodes we let through here have
3941 * valid i_mode and are being read by the orphan
3942 * recovery code: that's fine, we're about to complete
3943 * the process of deleting those. */
3945 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3946 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3947 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3948 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3950 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3951 inode->i_size = ext4_isize(raw_inode);
3952 ei->i_disksize = inode->i_size;
3954 ei->i_reserved_quota = 0;
3956 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3957 ei->i_block_group = iloc.block_group;
3958 ei->i_last_alloc_group = ~0;
3960 * NOTE! The in-memory inode i_data array is in little-endian order
3961 * even on big-endian machines: we do NOT byteswap the block numbers!
3963 for (block = 0; block < EXT4_N_BLOCKS; block++)
3964 ei->i_data[block] = raw_inode->i_block[block];
3965 INIT_LIST_HEAD(&ei->i_orphan);
3968 * Set transaction id's of transactions that have to be committed
3969 * to finish f[data]sync. We set them to currently running transaction
3970 * as we cannot be sure that the inode or some of its metadata isn't
3971 * part of the transaction - the inode could have been reclaimed and
3972 * now it is reread from disk.
3975 transaction_t *transaction;
3978 read_lock(&journal->j_state_lock);
3979 if (journal->j_running_transaction)
3980 transaction = journal->j_running_transaction;
3982 transaction = journal->j_committing_transaction;
3984 tid = transaction->t_tid;
3986 tid = journal->j_commit_sequence;
3987 read_unlock(&journal->j_state_lock);
3988 ei->i_sync_tid = tid;
3989 ei->i_datasync_tid = tid;
3992 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3993 if (ei->i_extra_isize == 0) {
3994 /* The extra space is currently unused. Use it. */
3995 ei->i_extra_isize = sizeof(struct ext4_inode) -
3996 EXT4_GOOD_OLD_INODE_SIZE;
3998 ext4_iget_extra_inode(inode, raw_inode, ei);
4002 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4003 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4004 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4005 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4007 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4008 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4009 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4011 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4015 if (ei->i_file_acl &&
4016 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4017 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4021 } else if (!ext4_has_inline_data(inode)) {
4022 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4023 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4024 (S_ISLNK(inode->i_mode) &&
4025 !ext4_inode_is_fast_symlink(inode))))
4026 /* Validate extent which is part of inode */
4027 ret = ext4_ext_check_inode(inode);
4028 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4029 (S_ISLNK(inode->i_mode) &&
4030 !ext4_inode_is_fast_symlink(inode))) {
4031 /* Validate block references which are part of inode */
4032 ret = ext4_ind_check_inode(inode);
4038 if (S_ISREG(inode->i_mode)) {
4039 inode->i_op = &ext4_file_inode_operations;
4040 inode->i_fop = &ext4_file_operations;
4041 ext4_set_aops(inode);
4042 } else if (S_ISDIR(inode->i_mode)) {
4043 inode->i_op = &ext4_dir_inode_operations;
4044 inode->i_fop = &ext4_dir_operations;
4045 } else if (S_ISLNK(inode->i_mode)) {
4046 if (ext4_inode_is_fast_symlink(inode)) {
4047 inode->i_op = &ext4_fast_symlink_inode_operations;
4048 nd_terminate_link(ei->i_data, inode->i_size,
4049 sizeof(ei->i_data) - 1);
4051 inode->i_op = &ext4_symlink_inode_operations;
4052 ext4_set_aops(inode);
4054 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4055 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4056 inode->i_op = &ext4_special_inode_operations;
4057 if (raw_inode->i_block[0])
4058 init_special_inode(inode, inode->i_mode,
4059 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4061 init_special_inode(inode, inode->i_mode,
4062 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4065 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4069 ext4_set_inode_flags(inode);
4070 unlock_new_inode(inode);
4076 return ERR_PTR(ret);
4079 static int ext4_inode_blocks_set(handle_t *handle,
4080 struct ext4_inode *raw_inode,
4081 struct ext4_inode_info *ei)
4083 struct inode *inode = &(ei->vfs_inode);
4084 u64 i_blocks = inode->i_blocks;
4085 struct super_block *sb = inode->i_sb;
4087 if (i_blocks <= ~0U) {
4089 * i_blocks can be represented in a 32 bit variable
4090 * as multiple of 512 bytes
4092 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4093 raw_inode->i_blocks_high = 0;
4094 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4097 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4100 if (i_blocks <= 0xffffffffffffULL) {
4102 * i_blocks can be represented in a 48 bit variable
4103 * as multiple of 512 bytes
4105 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4106 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4107 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4109 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4110 /* i_block is stored in file system block size */
4111 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4112 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4113 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4119 * Post the struct inode info into an on-disk inode location in the
4120 * buffer-cache. This gobbles the caller's reference to the
4121 * buffer_head in the inode location struct.
4123 * The caller must have write access to iloc->bh.
4125 static int ext4_do_update_inode(handle_t *handle,
4126 struct inode *inode,
4127 struct ext4_iloc *iloc)
4129 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4130 struct ext4_inode_info *ei = EXT4_I(inode);
4131 struct buffer_head *bh = iloc->bh;
4132 int err = 0, rc, block;
4133 int need_datasync = 0;
4137 /* For fields not not tracking in the in-memory inode,
4138 * initialise them to zero for new inodes. */
4139 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4140 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4142 ext4_get_inode_flags(ei);
4143 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4144 i_uid = i_uid_read(inode);
4145 i_gid = i_gid_read(inode);
4146 if (!(test_opt(inode->i_sb, NO_UID32))) {
4147 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4148 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4150 * Fix up interoperability with old kernels. Otherwise, old inodes get
4151 * re-used with the upper 16 bits of the uid/gid intact
4154 raw_inode->i_uid_high =
4155 cpu_to_le16(high_16_bits(i_uid));
4156 raw_inode->i_gid_high =
4157 cpu_to_le16(high_16_bits(i_gid));
4159 raw_inode->i_uid_high = 0;
4160 raw_inode->i_gid_high = 0;
4163 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4164 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4165 raw_inode->i_uid_high = 0;
4166 raw_inode->i_gid_high = 0;
4168 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4170 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4171 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4172 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4173 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4175 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4177 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4178 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4179 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4180 cpu_to_le32(EXT4_OS_HURD))
4181 raw_inode->i_file_acl_high =
4182 cpu_to_le16(ei->i_file_acl >> 32);
4183 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4184 if (ei->i_disksize != ext4_isize(raw_inode)) {
4185 ext4_isize_set(raw_inode, ei->i_disksize);
4188 if (ei->i_disksize > 0x7fffffffULL) {
4189 struct super_block *sb = inode->i_sb;
4190 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4191 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4192 EXT4_SB(sb)->s_es->s_rev_level ==
4193 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4194 /* If this is the first large file
4195 * created, add a flag to the superblock.
4197 err = ext4_journal_get_write_access(handle,
4198 EXT4_SB(sb)->s_sbh);
4201 ext4_update_dynamic_rev(sb);
4202 EXT4_SET_RO_COMPAT_FEATURE(sb,
4203 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4204 ext4_handle_sync(handle);
4205 err = ext4_handle_dirty_super(handle, sb);
4208 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4209 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4210 if (old_valid_dev(inode->i_rdev)) {
4211 raw_inode->i_block[0] =
4212 cpu_to_le32(old_encode_dev(inode->i_rdev));
4213 raw_inode->i_block[1] = 0;
4215 raw_inode->i_block[0] = 0;
4216 raw_inode->i_block[1] =
4217 cpu_to_le32(new_encode_dev(inode->i_rdev));
4218 raw_inode->i_block[2] = 0;
4220 } else if (!ext4_has_inline_data(inode)) {
4221 for (block = 0; block < EXT4_N_BLOCKS; block++)
4222 raw_inode->i_block[block] = ei->i_data[block];
4225 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4226 if (ei->i_extra_isize) {
4227 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4228 raw_inode->i_version_hi =
4229 cpu_to_le32(inode->i_version >> 32);
4230 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4233 ext4_inode_csum_set(inode, raw_inode, ei);
4235 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4236 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4239 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4241 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4244 ext4_std_error(inode->i_sb, err);
4249 * ext4_write_inode()
4251 * We are called from a few places:
4253 * - Within generic_file_write() for O_SYNC files.
4254 * Here, there will be no transaction running. We wait for any running
4255 * transaction to commit.
4257 * - Within sys_sync(), kupdate and such.
4258 * We wait on commit, if tol to.
4260 * - Within prune_icache() (PF_MEMALLOC == true)
4261 * Here we simply return. We can't afford to block kswapd on the
4264 * In all cases it is actually safe for us to return without doing anything,
4265 * because the inode has been copied into a raw inode buffer in
4266 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4269 * Note that we are absolutely dependent upon all inode dirtiers doing the
4270 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4271 * which we are interested.
4273 * It would be a bug for them to not do this. The code:
4275 * mark_inode_dirty(inode)
4277 * inode->i_size = expr;
4279 * is in error because a kswapd-driven write_inode() could occur while
4280 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4281 * will no longer be on the superblock's dirty inode list.
4283 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4287 if (current->flags & PF_MEMALLOC)
4290 if (EXT4_SB(inode->i_sb)->s_journal) {
4291 if (ext4_journal_current_handle()) {
4292 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4297 if (wbc->sync_mode != WB_SYNC_ALL)
4300 err = ext4_force_commit(inode->i_sb);
4302 struct ext4_iloc iloc;
4304 err = __ext4_get_inode_loc(inode, &iloc, 0);
4307 if (wbc->sync_mode == WB_SYNC_ALL)
4308 sync_dirty_buffer(iloc.bh);
4309 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4310 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4311 "IO error syncing inode");
4320 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4321 * buffers that are attached to a page stradding i_size and are undergoing
4322 * commit. In that case we have to wait for commit to finish and try again.
4324 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4328 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4329 tid_t commit_tid = 0;
4332 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4334 * All buffers in the last page remain valid? Then there's nothing to
4335 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4338 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4341 page = find_lock_page(inode->i_mapping,
4342 inode->i_size >> PAGE_CACHE_SHIFT);
4345 ret = __ext4_journalled_invalidatepage(page, offset);
4347 page_cache_release(page);
4351 read_lock(&journal->j_state_lock);
4352 if (journal->j_committing_transaction)
4353 commit_tid = journal->j_committing_transaction->t_tid;
4354 read_unlock(&journal->j_state_lock);
4356 jbd2_log_wait_commit(journal, commit_tid);
4363 * Called from notify_change.
4365 * We want to trap VFS attempts to truncate the file as soon as
4366 * possible. In particular, we want to make sure that when the VFS
4367 * shrinks i_size, we put the inode on the orphan list and modify
4368 * i_disksize immediately, so that during the subsequent flushing of
4369 * dirty pages and freeing of disk blocks, we can guarantee that any
4370 * commit will leave the blocks being flushed in an unused state on
4371 * disk. (On recovery, the inode will get truncated and the blocks will
4372 * be freed, so we have a strong guarantee that no future commit will
4373 * leave these blocks visible to the user.)
4375 * Another thing we have to assure is that if we are in ordered mode
4376 * and inode is still attached to the committing transaction, we must
4377 * we start writeout of all the dirty pages which are being truncated.
4378 * This way we are sure that all the data written in the previous
4379 * transaction are already on disk (truncate waits for pages under
4382 * Called with inode->i_mutex down.
4384 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4386 struct inode *inode = dentry->d_inode;
4389 const unsigned int ia_valid = attr->ia_valid;
4391 error = inode_change_ok(inode, attr);
4395 if (is_quota_modification(inode, attr))
4396 dquot_initialize(inode);
4397 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4398 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4401 /* (user+group)*(old+new) structure, inode write (sb,
4402 * inode block, ? - but truncate inode update has it) */
4403 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
4404 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
4405 if (IS_ERR(handle)) {
4406 error = PTR_ERR(handle);
4409 error = dquot_transfer(inode, attr);
4411 ext4_journal_stop(handle);
4414 /* Update corresponding info in inode so that everything is in
4415 * one transaction */
4416 if (attr->ia_valid & ATTR_UID)
4417 inode->i_uid = attr->ia_uid;
4418 if (attr->ia_valid & ATTR_GID)
4419 inode->i_gid = attr->ia_gid;
4420 error = ext4_mark_inode_dirty(handle, inode);
4421 ext4_journal_stop(handle);
4424 if (attr->ia_valid & ATTR_SIZE) {
4426 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4427 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4429 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4434 if (S_ISREG(inode->i_mode) &&
4435 attr->ia_valid & ATTR_SIZE &&
4436 (attr->ia_size < inode->i_size)) {
4439 handle = ext4_journal_start(inode, 3);
4440 if (IS_ERR(handle)) {
4441 error = PTR_ERR(handle);
4444 if (ext4_handle_valid(handle)) {
4445 error = ext4_orphan_add(handle, inode);
4448 EXT4_I(inode)->i_disksize = attr->ia_size;
4449 rc = ext4_mark_inode_dirty(handle, inode);
4452 ext4_journal_stop(handle);
4454 if (ext4_should_order_data(inode)) {
4455 error = ext4_begin_ordered_truncate(inode,
4458 /* Do as much error cleanup as possible */
4459 handle = ext4_journal_start(inode, 3);
4460 if (IS_ERR(handle)) {
4461 ext4_orphan_del(NULL, inode);
4464 ext4_orphan_del(handle, inode);
4466 ext4_journal_stop(handle);
4472 if (attr->ia_valid & ATTR_SIZE) {
4473 if (attr->ia_size != inode->i_size) {
4474 loff_t oldsize = inode->i_size;
4476 i_size_write(inode, attr->ia_size);
4478 * Blocks are going to be removed from the inode. Wait
4479 * for dio in flight. Temporarily disable
4480 * dioread_nolock to prevent livelock.
4483 if (!ext4_should_journal_data(inode)) {
4484 ext4_inode_block_unlocked_dio(inode);
4485 inode_dio_wait(inode);
4486 ext4_inode_resume_unlocked_dio(inode);
4488 ext4_wait_for_tail_page_commit(inode);
4491 * Truncate pagecache after we've waited for commit
4492 * in data=journal mode to make pages freeable.
4494 truncate_pagecache(inode, oldsize, inode->i_size);
4496 ext4_truncate(inode);
4500 setattr_copy(inode, attr);
4501 mark_inode_dirty(inode);
4505 * If the call to ext4_truncate failed to get a transaction handle at
4506 * all, we need to clean up the in-core orphan list manually.
4508 if (orphan && inode->i_nlink)
4509 ext4_orphan_del(NULL, inode);
4511 if (!rc && (ia_valid & ATTR_MODE))
4512 rc = ext4_acl_chmod(inode);
4515 ext4_std_error(inode->i_sb, error);
4521 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4524 struct inode *inode;
4525 unsigned long delalloc_blocks;
4527 inode = dentry->d_inode;
4528 generic_fillattr(inode, stat);
4531 * We can't update i_blocks if the block allocation is delayed
4532 * otherwise in the case of system crash before the real block
4533 * allocation is done, we will have i_blocks inconsistent with
4534 * on-disk file blocks.
4535 * We always keep i_blocks updated together with real
4536 * allocation. But to not confuse with user, stat
4537 * will return the blocks that include the delayed allocation
4538 * blocks for this file.
4540 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4541 EXT4_I(inode)->i_reserved_data_blocks);
4543 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4547 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4549 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4550 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4551 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4555 * Account for index blocks, block groups bitmaps and block group
4556 * descriptor blocks if modify datablocks and index blocks
4557 * worse case, the indexs blocks spread over different block groups
4559 * If datablocks are discontiguous, they are possible to spread over
4560 * different block groups too. If they are contiguous, with flexbg,
4561 * they could still across block group boundary.
4563 * Also account for superblock, inode, quota and xattr blocks
4565 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4567 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4573 * How many index blocks need to touch to modify nrblocks?
4574 * The "Chunk" flag indicating whether the nrblocks is
4575 * physically contiguous on disk
4577 * For Direct IO and fallocate, they calls get_block to allocate
4578 * one single extent at a time, so they could set the "Chunk" flag
4580 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4585 * Now let's see how many group bitmaps and group descriptors need
4595 if (groups > ngroups)
4597 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4598 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4600 /* bitmaps and block group descriptor blocks */
4601 ret += groups + gdpblocks;
4603 /* Blocks for super block, inode, quota and xattr blocks */
4604 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4610 * Calculate the total number of credits to reserve to fit
4611 * the modification of a single pages into a single transaction,
4612 * which may include multiple chunks of block allocations.
4614 * This could be called via ext4_write_begin()
4616 * We need to consider the worse case, when
4617 * one new block per extent.
4619 int ext4_writepage_trans_blocks(struct inode *inode)
4621 int bpp = ext4_journal_blocks_per_page(inode);
4624 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4626 /* Account for data blocks for journalled mode */
4627 if (ext4_should_journal_data(inode))
4633 * Calculate the journal credits for a chunk of data modification.
4635 * This is called from DIO, fallocate or whoever calling
4636 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4638 * journal buffers for data blocks are not included here, as DIO
4639 * and fallocate do no need to journal data buffers.
4641 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4643 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4647 * The caller must have previously called ext4_reserve_inode_write().
4648 * Give this, we know that the caller already has write access to iloc->bh.
4650 int ext4_mark_iloc_dirty(handle_t *handle,
4651 struct inode *inode, struct ext4_iloc *iloc)
4655 if (IS_I_VERSION(inode))
4656 inode_inc_iversion(inode);
4658 /* the do_update_inode consumes one bh->b_count */
4661 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4662 err = ext4_do_update_inode(handle, inode, iloc);
4668 * On success, We end up with an outstanding reference count against
4669 * iloc->bh. This _must_ be cleaned up later.
4673 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4674 struct ext4_iloc *iloc)
4678 err = ext4_get_inode_loc(inode, iloc);
4680 BUFFER_TRACE(iloc->bh, "get_write_access");
4681 err = ext4_journal_get_write_access(handle, iloc->bh);
4687 ext4_std_error(inode->i_sb, err);
4692 * Expand an inode by new_extra_isize bytes.
4693 * Returns 0 on success or negative error number on failure.
4695 static int ext4_expand_extra_isize(struct inode *inode,
4696 unsigned int new_extra_isize,
4697 struct ext4_iloc iloc,
4700 struct ext4_inode *raw_inode;
4701 struct ext4_xattr_ibody_header *header;
4703 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4706 raw_inode = ext4_raw_inode(&iloc);
4708 header = IHDR(inode, raw_inode);
4710 /* No extended attributes present */
4711 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4712 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4713 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4715 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4719 /* try to expand with EAs present */
4720 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4725 * What we do here is to mark the in-core inode as clean with respect to inode
4726 * dirtiness (it may still be data-dirty).
4727 * This means that the in-core inode may be reaped by prune_icache
4728 * without having to perform any I/O. This is a very good thing,
4729 * because *any* task may call prune_icache - even ones which
4730 * have a transaction open against a different journal.
4732 * Is this cheating? Not really. Sure, we haven't written the
4733 * inode out, but prune_icache isn't a user-visible syncing function.
4734 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4735 * we start and wait on commits.
4737 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4739 struct ext4_iloc iloc;
4740 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4741 static unsigned int mnt_count;
4745 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4746 err = ext4_reserve_inode_write(handle, inode, &iloc);
4747 if (ext4_handle_valid(handle) &&
4748 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4749 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4751 * We need extra buffer credits since we may write into EA block
4752 * with this same handle. If journal_extend fails, then it will
4753 * only result in a minor loss of functionality for that inode.
4754 * If this is felt to be critical, then e2fsck should be run to
4755 * force a large enough s_min_extra_isize.
4757 if ((jbd2_journal_extend(handle,
4758 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4759 ret = ext4_expand_extra_isize(inode,
4760 sbi->s_want_extra_isize,
4763 ext4_set_inode_state(inode,
4764 EXT4_STATE_NO_EXPAND);
4766 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4767 ext4_warning(inode->i_sb,
4768 "Unable to expand inode %lu. Delete"
4769 " some EAs or run e2fsck.",
4772 le16_to_cpu(sbi->s_es->s_mnt_count);
4778 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4783 * ext4_dirty_inode() is called from __mark_inode_dirty()
4785 * We're really interested in the case where a file is being extended.
4786 * i_size has been changed by generic_commit_write() and we thus need
4787 * to include the updated inode in the current transaction.
4789 * Also, dquot_alloc_block() will always dirty the inode when blocks
4790 * are allocated to the file.
4792 * If the inode is marked synchronous, we don't honour that here - doing
4793 * so would cause a commit on atime updates, which we don't bother doing.
4794 * We handle synchronous inodes at the highest possible level.
4796 void ext4_dirty_inode(struct inode *inode, int flags)
4800 handle = ext4_journal_start(inode, 2);
4804 ext4_mark_inode_dirty(handle, inode);
4806 ext4_journal_stop(handle);
4813 * Bind an inode's backing buffer_head into this transaction, to prevent
4814 * it from being flushed to disk early. Unlike
4815 * ext4_reserve_inode_write, this leaves behind no bh reference and
4816 * returns no iloc structure, so the caller needs to repeat the iloc
4817 * lookup to mark the inode dirty later.
4819 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4821 struct ext4_iloc iloc;
4825 err = ext4_get_inode_loc(inode, &iloc);
4827 BUFFER_TRACE(iloc.bh, "get_write_access");
4828 err = jbd2_journal_get_write_access(handle, iloc.bh);
4830 err = ext4_handle_dirty_metadata(handle,
4836 ext4_std_error(inode->i_sb, err);
4841 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4848 * We have to be very careful here: changing a data block's
4849 * journaling status dynamically is dangerous. If we write a
4850 * data block to the journal, change the status and then delete
4851 * that block, we risk forgetting to revoke the old log record
4852 * from the journal and so a subsequent replay can corrupt data.
4853 * So, first we make sure that the journal is empty and that
4854 * nobody is changing anything.
4857 journal = EXT4_JOURNAL(inode);
4860 if (is_journal_aborted(journal))
4862 /* We have to allocate physical blocks for delalloc blocks
4863 * before flushing journal. otherwise delalloc blocks can not
4864 * be allocated any more. even more truncate on delalloc blocks
4865 * could trigger BUG by flushing delalloc blocks in journal.
4866 * There is no delalloc block in non-journal data mode.
4868 if (val && test_opt(inode->i_sb, DELALLOC)) {
4869 err = ext4_alloc_da_blocks(inode);
4874 /* Wait for all existing dio workers */
4875 ext4_inode_block_unlocked_dio(inode);
4876 inode_dio_wait(inode);
4878 jbd2_journal_lock_updates(journal);
4881 * OK, there are no updates running now, and all cached data is
4882 * synced to disk. We are now in a completely consistent state
4883 * which doesn't have anything in the journal, and we know that
4884 * no filesystem updates are running, so it is safe to modify
4885 * the inode's in-core data-journaling state flag now.
4889 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4891 jbd2_journal_flush(journal);
4892 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4894 ext4_set_aops(inode);
4896 jbd2_journal_unlock_updates(journal);
4897 ext4_inode_resume_unlocked_dio(inode);
4899 /* Finally we can mark the inode as dirty. */
4901 handle = ext4_journal_start(inode, 1);
4903 return PTR_ERR(handle);
4905 err = ext4_mark_inode_dirty(handle, inode);
4906 ext4_handle_sync(handle);
4907 ext4_journal_stop(handle);
4908 ext4_std_error(inode->i_sb, err);
4913 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4915 return !buffer_mapped(bh);
4918 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4920 struct page *page = vmf->page;
4924 struct file *file = vma->vm_file;
4925 struct inode *inode = file->f_path.dentry->d_inode;
4926 struct address_space *mapping = inode->i_mapping;
4928 get_block_t *get_block;
4931 sb_start_pagefault(inode->i_sb);
4932 file_update_time(vma->vm_file);
4933 /* Delalloc case is easy... */
4934 if (test_opt(inode->i_sb, DELALLOC) &&
4935 !ext4_should_journal_data(inode) &&
4936 !ext4_nonda_switch(inode->i_sb)) {
4938 ret = __block_page_mkwrite(vma, vmf,
4939 ext4_da_get_block_prep);
4940 } while (ret == -ENOSPC &&
4941 ext4_should_retry_alloc(inode->i_sb, &retries));
4946 size = i_size_read(inode);
4947 /* Page got truncated from under us? */
4948 if (page->mapping != mapping || page_offset(page) > size) {
4950 ret = VM_FAULT_NOPAGE;
4954 if (page->index == size >> PAGE_CACHE_SHIFT)
4955 len = size & ~PAGE_CACHE_MASK;
4957 len = PAGE_CACHE_SIZE;
4959 * Return if we have all the buffers mapped. This avoids the need to do
4960 * journal_start/journal_stop which can block and take a long time
4962 if (page_has_buffers(page)) {
4963 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
4965 ext4_bh_unmapped)) {
4966 /* Wait so that we don't change page under IO */
4967 wait_on_page_writeback(page);
4968 ret = VM_FAULT_LOCKED;
4973 /* OK, we need to fill the hole... */
4974 if (ext4_should_dioread_nolock(inode))
4975 get_block = ext4_get_block_write;
4977 get_block = ext4_get_block;
4979 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4980 if (IS_ERR(handle)) {
4981 ret = VM_FAULT_SIGBUS;
4984 ret = __block_page_mkwrite(vma, vmf, get_block);
4985 if (!ret && ext4_should_journal_data(inode)) {
4986 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
4987 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4989 ret = VM_FAULT_SIGBUS;
4990 ext4_journal_stop(handle);
4993 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4995 ext4_journal_stop(handle);
4996 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4999 ret = block_page_mkwrite_return(ret);
5001 sb_end_pagefault(inode->i_sb);