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 = le16_to_cpu(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 = le16_to_cpu(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 = cpu_to_le16(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 = cpu_to_le16(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 __ext4_journalled_writepage(struct page *page, unsigned int len);
136 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
137 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
138 struct inode *inode, struct page *page, loff_t from,
139 loff_t length, int flags);
142 * Test whether an inode is a fast symlink.
144 static int ext4_inode_is_fast_symlink(struct inode *inode)
146 int ea_blocks = EXT4_I(inode)->i_file_acl ?
147 (inode->i_sb->s_blocksize >> 9) : 0;
149 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
153 * Restart the transaction associated with *handle. This does a commit,
154 * so before we call here everything must be consistently dirtied against
157 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
163 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
164 * moment, get_block can be called only for blocks inside i_size since
165 * page cache has been already dropped and writes are blocked by
166 * i_mutex. So we can safely drop the i_data_sem here.
168 BUG_ON(EXT4_JOURNAL(inode) == NULL);
169 jbd_debug(2, "restarting handle %p\n", handle);
170 up_write(&EXT4_I(inode)->i_data_sem);
171 ret = ext4_journal_restart(handle, nblocks);
172 down_write(&EXT4_I(inode)->i_data_sem);
173 ext4_discard_preallocations(inode);
179 * Called at the last iput() if i_nlink is zero.
181 void ext4_evict_inode(struct inode *inode)
186 trace_ext4_evict_inode(inode);
188 if (inode->i_nlink) {
190 * When journalling data dirty buffers are tracked only in the
191 * journal. So although mm thinks everything is clean and
192 * ready for reaping the inode might still have some pages to
193 * write in the running transaction or waiting to be
194 * checkpointed. Thus calling jbd2_journal_invalidatepage()
195 * (via truncate_inode_pages()) to discard these buffers can
196 * cause data loss. Also even if we did not discard these
197 * buffers, we would have no way to find them after the inode
198 * is reaped and thus user could see stale data if he tries to
199 * read them before the transaction is checkpointed. So be
200 * careful and force everything to disk here... We use
201 * ei->i_datasync_tid to store the newest transaction
202 * containing inode's data.
204 * Note that directories do not have this problem because they
205 * don't use page cache.
207 if (ext4_should_journal_data(inode) &&
208 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
209 inode->i_ino != EXT4_JOURNAL_INO) {
210 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
211 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
213 jbd2_complete_transaction(journal, commit_tid);
214 filemap_write_and_wait(&inode->i_data);
216 truncate_inode_pages(&inode->i_data, 0);
217 ext4_ioend_shutdown(inode);
221 if (!is_bad_inode(inode))
222 dquot_initialize(inode);
224 if (ext4_should_order_data(inode))
225 ext4_begin_ordered_truncate(inode, 0);
226 truncate_inode_pages(&inode->i_data, 0);
227 ext4_ioend_shutdown(inode);
229 if (is_bad_inode(inode))
233 * Protect us against freezing - iput() caller didn't have to have any
234 * protection against it
236 sb_start_intwrite(inode->i_sb);
237 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
238 ext4_blocks_for_truncate(inode)+3);
239 if (IS_ERR(handle)) {
240 ext4_std_error(inode->i_sb, PTR_ERR(handle));
242 * If we're going to skip the normal cleanup, we still need to
243 * make sure that the in-core orphan linked list is properly
246 ext4_orphan_del(NULL, inode);
247 sb_end_intwrite(inode->i_sb);
252 ext4_handle_sync(handle);
254 err = ext4_mark_inode_dirty(handle, inode);
256 ext4_warning(inode->i_sb,
257 "couldn't mark inode dirty (err %d)", err);
261 ext4_truncate(inode);
264 * ext4_ext_truncate() doesn't reserve any slop when it
265 * restarts journal transactions; therefore there may not be
266 * enough credits left in the handle to remove the inode from
267 * the orphan list and set the dtime field.
269 if (!ext4_handle_has_enough_credits(handle, 3)) {
270 err = ext4_journal_extend(handle, 3);
272 err = ext4_journal_restart(handle, 3);
274 ext4_warning(inode->i_sb,
275 "couldn't extend journal (err %d)", err);
277 ext4_journal_stop(handle);
278 ext4_orphan_del(NULL, inode);
279 sb_end_intwrite(inode->i_sb);
285 * Kill off the orphan record which ext4_truncate created.
286 * AKPM: I think this can be inside the above `if'.
287 * Note that ext4_orphan_del() has to be able to cope with the
288 * deletion of a non-existent orphan - this is because we don't
289 * know if ext4_truncate() actually created an orphan record.
290 * (Well, we could do this if we need to, but heck - it works)
292 ext4_orphan_del(handle, inode);
293 EXT4_I(inode)->i_dtime = get_seconds();
296 * One subtle ordering requirement: if anything has gone wrong
297 * (transaction abort, IO errors, whatever), then we can still
298 * do these next steps (the fs will already have been marked as
299 * having errors), but we can't free the inode if the mark_dirty
302 if (ext4_mark_inode_dirty(handle, inode))
303 /* If that failed, just do the required in-core inode clear. */
304 ext4_clear_inode(inode);
306 ext4_free_inode(handle, inode);
307 ext4_journal_stop(handle);
308 sb_end_intwrite(inode->i_sb);
311 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
315 qsize_t *ext4_get_reserved_space(struct inode *inode)
317 return &EXT4_I(inode)->i_reserved_quota;
322 * Calculate the number of metadata blocks need to reserve
323 * to allocate a block located at @lblock
325 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
327 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
328 return ext4_ext_calc_metadata_amount(inode, lblock);
330 return ext4_ind_calc_metadata_amount(inode, lblock);
334 * Called with i_data_sem down, which is important since we can call
335 * ext4_discard_preallocations() from here.
337 void ext4_da_update_reserve_space(struct inode *inode,
338 int used, int quota_claim)
340 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
341 struct ext4_inode_info *ei = EXT4_I(inode);
343 spin_lock(&ei->i_block_reservation_lock);
344 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
345 if (unlikely(used > ei->i_reserved_data_blocks)) {
346 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
347 "with only %d reserved data blocks",
348 __func__, inode->i_ino, used,
349 ei->i_reserved_data_blocks);
351 used = ei->i_reserved_data_blocks;
354 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
355 ext4_warning(inode->i_sb, "ino %lu, allocated %d "
356 "with only %d reserved metadata blocks "
357 "(releasing %d blocks with reserved %d data blocks)",
358 inode->i_ino, ei->i_allocated_meta_blocks,
359 ei->i_reserved_meta_blocks, used,
360 ei->i_reserved_data_blocks);
362 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
365 /* Update per-inode reservations */
366 ei->i_reserved_data_blocks -= used;
367 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
368 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
369 used + ei->i_allocated_meta_blocks);
370 ei->i_allocated_meta_blocks = 0;
372 if (ei->i_reserved_data_blocks == 0) {
374 * We can release all of the reserved metadata blocks
375 * only when we have written all of the delayed
378 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
379 ei->i_reserved_meta_blocks);
380 ei->i_reserved_meta_blocks = 0;
381 ei->i_da_metadata_calc_len = 0;
383 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
385 /* Update quota subsystem for data blocks */
387 dquot_claim_block(inode, EXT4_C2B(sbi, used));
390 * We did fallocate with an offset that is already delayed
391 * allocated. So on delayed allocated writeback we should
392 * not re-claim the quota for fallocated blocks.
394 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
398 * If we have done all the pending block allocations and if
399 * there aren't any writers on the inode, we can discard the
400 * inode's preallocations.
402 if ((ei->i_reserved_data_blocks == 0) &&
403 (atomic_read(&inode->i_writecount) == 0))
404 ext4_discard_preallocations(inode);
407 static int __check_block_validity(struct inode *inode, const char *func,
409 struct ext4_map_blocks *map)
411 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
413 ext4_error_inode(inode, func, line, map->m_pblk,
414 "lblock %lu mapped to illegal pblock "
415 "(length %d)", (unsigned long) map->m_lblk,
422 #define check_block_validity(inode, map) \
423 __check_block_validity((inode), __func__, __LINE__, (map))
426 * Return the number of contiguous dirty pages in a given inode
427 * starting at page frame idx.
429 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
430 unsigned int max_pages)
432 struct address_space *mapping = inode->i_mapping;
436 int i, nr_pages, done = 0;
440 pagevec_init(&pvec, 0);
443 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
445 (pgoff_t)PAGEVEC_SIZE);
448 for (i = 0; i < nr_pages; i++) {
449 struct page *page = pvec.pages[i];
450 struct buffer_head *bh, *head;
453 if (unlikely(page->mapping != mapping) ||
455 PageWriteback(page) ||
456 page->index != idx) {
461 if (page_has_buffers(page)) {
462 bh = head = page_buffers(page);
464 if (!buffer_delay(bh) &&
465 !buffer_unwritten(bh))
467 bh = bh->b_this_page;
468 } while (!done && (bh != head));
475 if (num >= max_pages) {
480 pagevec_release(&pvec);
485 #ifdef ES_AGGRESSIVE_TEST
486 static void ext4_map_blocks_es_recheck(handle_t *handle,
488 struct ext4_map_blocks *es_map,
489 struct ext4_map_blocks *map,
496 * There is a race window that the result is not the same.
497 * e.g. xfstests #223 when dioread_nolock enables. The reason
498 * is that we lookup a block mapping in extent status tree with
499 * out taking i_data_sem. So at the time the unwritten extent
500 * could be converted.
502 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
503 down_read((&EXT4_I(inode)->i_data_sem));
504 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
505 retval = ext4_ext_map_blocks(handle, inode, map, flags &
506 EXT4_GET_BLOCKS_KEEP_SIZE);
508 retval = ext4_ind_map_blocks(handle, inode, map, flags &
509 EXT4_GET_BLOCKS_KEEP_SIZE);
511 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
512 up_read((&EXT4_I(inode)->i_data_sem));
514 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
515 * because it shouldn't be marked in es_map->m_flags.
517 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
520 * We don't check m_len because extent will be collpased in status
521 * tree. So the m_len might not equal.
523 if (es_map->m_lblk != map->m_lblk ||
524 es_map->m_flags != map->m_flags ||
525 es_map->m_pblk != map->m_pblk) {
526 printk("ES cache assertation failed for inode: %lu "
527 "es_cached ex [%d/%d/%llu/%x] != "
528 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
529 inode->i_ino, es_map->m_lblk, es_map->m_len,
530 es_map->m_pblk, es_map->m_flags, map->m_lblk,
531 map->m_len, map->m_pblk, map->m_flags,
535 #endif /* ES_AGGRESSIVE_TEST */
538 * The ext4_map_blocks() function tries to look up the requested blocks,
539 * and returns if the blocks are already mapped.
541 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
542 * and store the allocated blocks in the result buffer head and mark it
545 * If file type is extents based, it will call ext4_ext_map_blocks(),
546 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
549 * On success, it returns the number of blocks being mapped or allocate.
550 * if create==0 and the blocks are pre-allocated and uninitialized block,
551 * the result buffer head is unmapped. If the create ==1, it will make sure
552 * the buffer head is mapped.
554 * It returns 0 if plain look up failed (blocks have not been allocated), in
555 * that case, buffer head is unmapped
557 * It returns the error in case of allocation failure.
559 int ext4_map_blocks(handle_t *handle, struct inode *inode,
560 struct ext4_map_blocks *map, int flags)
562 struct extent_status es;
564 #ifdef ES_AGGRESSIVE_TEST
565 struct ext4_map_blocks orig_map;
567 memcpy(&orig_map, map, sizeof(*map));
571 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
572 "logical block %lu\n", inode->i_ino, flags, map->m_len,
573 (unsigned long) map->m_lblk);
575 /* Lookup extent status tree firstly */
576 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
577 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
578 map->m_pblk = ext4_es_pblock(&es) +
579 map->m_lblk - es.es_lblk;
580 map->m_flags |= ext4_es_is_written(&es) ?
581 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
582 retval = es.es_len - (map->m_lblk - es.es_lblk);
583 if (retval > map->m_len)
586 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
591 #ifdef ES_AGGRESSIVE_TEST
592 ext4_map_blocks_es_recheck(handle, inode, map,
599 * Try to see if we can get the block without requesting a new
602 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
603 down_read((&EXT4_I(inode)->i_data_sem));
604 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
605 retval = ext4_ext_map_blocks(handle, inode, map, flags &
606 EXT4_GET_BLOCKS_KEEP_SIZE);
608 retval = ext4_ind_map_blocks(handle, inode, map, flags &
609 EXT4_GET_BLOCKS_KEEP_SIZE);
613 unsigned long long status;
615 #ifdef ES_AGGRESSIVE_TEST
616 if (retval != map->m_len) {
617 printk("ES len assertation failed for inode: %lu "
618 "retval %d != map->m_len %d "
619 "in %s (lookup)\n", inode->i_ino, retval,
620 map->m_len, __func__);
624 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
625 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
626 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
627 ext4_find_delalloc_range(inode, map->m_lblk,
628 map->m_lblk + map->m_len - 1))
629 status |= EXTENT_STATUS_DELAYED;
630 ret = ext4_es_insert_extent(inode, map->m_lblk,
631 map->m_len, map->m_pblk, status);
635 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
636 up_read((&EXT4_I(inode)->i_data_sem));
639 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
640 int ret = check_block_validity(inode, map);
645 /* If it is only a block(s) look up */
646 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
650 * Returns if the blocks have already allocated
652 * Note that if blocks have been preallocated
653 * ext4_ext_get_block() returns the create = 0
654 * with buffer head unmapped.
656 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
660 * Here we clear m_flags because after allocating an new extent,
661 * it will be set again.
663 map->m_flags &= ~EXT4_MAP_FLAGS;
666 * New blocks allocate and/or writing to uninitialized extent
667 * will possibly result in updating i_data, so we take
668 * the write lock of i_data_sem, and call get_blocks()
669 * with create == 1 flag.
671 down_write((&EXT4_I(inode)->i_data_sem));
674 * if the caller is from delayed allocation writeout path
675 * we have already reserved fs blocks for allocation
676 * let the underlying get_block() function know to
677 * avoid double accounting
679 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
680 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
682 * We need to check for EXT4 here because migrate
683 * could have changed the inode type in between
685 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
686 retval = ext4_ext_map_blocks(handle, inode, map, flags);
688 retval = ext4_ind_map_blocks(handle, inode, map, flags);
690 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
692 * We allocated new blocks which will result in
693 * i_data's format changing. Force the migrate
694 * to fail by clearing migrate flags
696 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
700 * Update reserved blocks/metadata blocks after successful
701 * block allocation which had been deferred till now. We don't
702 * support fallocate for non extent files. So we can update
703 * reserve space here.
706 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
707 ext4_da_update_reserve_space(inode, retval, 1);
709 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
710 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
714 unsigned long long status;
716 #ifdef ES_AGGRESSIVE_TEST
717 if (retval != map->m_len) {
718 printk("ES len assertation failed for inode: %lu "
719 "retval %d != map->m_len %d "
720 "in %s (allocation)\n", inode->i_ino, retval,
721 map->m_len, __func__);
726 * If the extent has been zeroed out, we don't need to update
727 * extent status tree.
729 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
730 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
731 if (ext4_es_is_written(&es))
734 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
735 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
736 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
737 ext4_find_delalloc_range(inode, map->m_lblk,
738 map->m_lblk + map->m_len - 1))
739 status |= EXTENT_STATUS_DELAYED;
740 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
741 map->m_pblk, status);
747 up_write((&EXT4_I(inode)->i_data_sem));
748 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
749 int ret = check_block_validity(inode, map);
756 /* Maximum number of blocks we map for direct IO at once. */
757 #define DIO_MAX_BLOCKS 4096
759 static int _ext4_get_block(struct inode *inode, sector_t iblock,
760 struct buffer_head *bh, int flags)
762 handle_t *handle = ext4_journal_current_handle();
763 struct ext4_map_blocks map;
764 int ret = 0, started = 0;
767 if (ext4_has_inline_data(inode))
771 map.m_len = bh->b_size >> inode->i_blkbits;
773 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
774 /* Direct IO write... */
775 if (map.m_len > DIO_MAX_BLOCKS)
776 map.m_len = DIO_MAX_BLOCKS;
777 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
778 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
780 if (IS_ERR(handle)) {
781 ret = PTR_ERR(handle);
787 ret = ext4_map_blocks(handle, inode, &map, flags);
789 map_bh(bh, inode->i_sb, map.m_pblk);
790 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
791 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
795 ext4_journal_stop(handle);
799 int ext4_get_block(struct inode *inode, sector_t iblock,
800 struct buffer_head *bh, int create)
802 return _ext4_get_block(inode, iblock, bh,
803 create ? EXT4_GET_BLOCKS_CREATE : 0);
807 * `handle' can be NULL if create is zero
809 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
810 ext4_lblk_t block, int create, int *errp)
812 struct ext4_map_blocks map;
813 struct buffer_head *bh;
816 J_ASSERT(handle != NULL || create == 0);
820 err = ext4_map_blocks(handle, inode, &map,
821 create ? EXT4_GET_BLOCKS_CREATE : 0);
823 /* ensure we send some value back into *errp */
826 if (create && err == 0)
827 err = -ENOSPC; /* should never happen */
833 bh = sb_getblk(inode->i_sb, map.m_pblk);
838 if (map.m_flags & EXT4_MAP_NEW) {
839 J_ASSERT(create != 0);
840 J_ASSERT(handle != NULL);
843 * Now that we do not always journal data, we should
844 * keep in mind whether this should always journal the
845 * new buffer as metadata. For now, regular file
846 * writes use ext4_get_block instead, so it's not a
850 BUFFER_TRACE(bh, "call get_create_access");
851 fatal = ext4_journal_get_create_access(handle, bh);
852 if (!fatal && !buffer_uptodate(bh)) {
853 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
854 set_buffer_uptodate(bh);
857 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
858 err = ext4_handle_dirty_metadata(handle, inode, bh);
862 BUFFER_TRACE(bh, "not a new buffer");
872 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
873 ext4_lblk_t block, int create, int *err)
875 struct buffer_head *bh;
877 bh = ext4_getblk(handle, inode, block, create, err);
880 if (buffer_uptodate(bh))
882 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
884 if (buffer_uptodate(bh))
891 int ext4_walk_page_buffers(handle_t *handle,
892 struct buffer_head *head,
896 int (*fn)(handle_t *handle,
897 struct buffer_head *bh))
899 struct buffer_head *bh;
900 unsigned block_start, block_end;
901 unsigned blocksize = head->b_size;
903 struct buffer_head *next;
905 for (bh = head, block_start = 0;
906 ret == 0 && (bh != head || !block_start);
907 block_start = block_end, bh = next) {
908 next = bh->b_this_page;
909 block_end = block_start + blocksize;
910 if (block_end <= from || block_start >= to) {
911 if (partial && !buffer_uptodate(bh))
915 err = (*fn)(handle, bh);
923 * To preserve ordering, it is essential that the hole instantiation and
924 * the data write be encapsulated in a single transaction. We cannot
925 * close off a transaction and start a new one between the ext4_get_block()
926 * and the commit_write(). So doing the jbd2_journal_start at the start of
927 * prepare_write() is the right place.
929 * Also, this function can nest inside ext4_writepage(). In that case, we
930 * *know* that ext4_writepage() has generated enough buffer credits to do the
931 * whole page. So we won't block on the journal in that case, which is good,
932 * because the caller may be PF_MEMALLOC.
934 * By accident, ext4 can be reentered when a transaction is open via
935 * quota file writes. If we were to commit the transaction while thus
936 * reentered, there can be a deadlock - we would be holding a quota
937 * lock, and the commit would never complete if another thread had a
938 * transaction open and was blocking on the quota lock - a ranking
941 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
942 * will _not_ run commit under these circumstances because handle->h_ref
943 * is elevated. We'll still have enough credits for the tiny quotafile
946 int do_journal_get_write_access(handle_t *handle,
947 struct buffer_head *bh)
949 int dirty = buffer_dirty(bh);
952 if (!buffer_mapped(bh) || buffer_freed(bh))
955 * __block_write_begin() could have dirtied some buffers. Clean
956 * the dirty bit as jbd2_journal_get_write_access() could complain
957 * otherwise about fs integrity issues. Setting of the dirty bit
958 * by __block_write_begin() isn't a real problem here as we clear
959 * the bit before releasing a page lock and thus writeback cannot
960 * ever write the buffer.
963 clear_buffer_dirty(bh);
964 ret = ext4_journal_get_write_access(handle, bh);
966 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
970 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
971 struct buffer_head *bh_result, int create);
972 static int ext4_write_begin(struct file *file, struct address_space *mapping,
973 loff_t pos, unsigned len, unsigned flags,
974 struct page **pagep, void **fsdata)
976 struct inode *inode = mapping->host;
977 int ret, needed_blocks;
984 trace_ext4_write_begin(inode, pos, len, flags);
986 * Reserve one block more for addition to orphan list in case
987 * we allocate blocks but write fails for some reason
989 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
990 index = pos >> PAGE_CACHE_SHIFT;
991 from = pos & (PAGE_CACHE_SIZE - 1);
994 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
995 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1004 * grab_cache_page_write_begin() can take a long time if the
1005 * system is thrashing due to memory pressure, or if the page
1006 * is being written back. So grab it first before we start
1007 * the transaction handle. This also allows us to allocate
1008 * the page (if needed) without using GFP_NOFS.
1011 page = grab_cache_page_write_begin(mapping, index, flags);
1017 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1018 if (IS_ERR(handle)) {
1019 page_cache_release(page);
1020 return PTR_ERR(handle);
1024 if (page->mapping != mapping) {
1025 /* The page got truncated from under us */
1027 page_cache_release(page);
1028 ext4_journal_stop(handle);
1031 wait_on_page_writeback(page);
1033 if (ext4_should_dioread_nolock(inode))
1034 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1036 ret = __block_write_begin(page, pos, len, ext4_get_block);
1038 if (!ret && ext4_should_journal_data(inode)) {
1039 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1041 do_journal_get_write_access);
1047 * __block_write_begin may have instantiated a few blocks
1048 * outside i_size. Trim these off again. Don't need
1049 * i_size_read because we hold i_mutex.
1051 * Add inode to orphan list in case we crash before
1054 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1055 ext4_orphan_add(handle, inode);
1057 ext4_journal_stop(handle);
1058 if (pos + len > inode->i_size) {
1059 ext4_truncate_failed_write(inode);
1061 * If truncate failed early the inode might
1062 * still be on the orphan list; we need to
1063 * make sure the inode is removed from the
1064 * orphan list in that case.
1067 ext4_orphan_del(NULL, inode);
1070 if (ret == -ENOSPC &&
1071 ext4_should_retry_alloc(inode->i_sb, &retries))
1073 page_cache_release(page);
1080 /* For write_end() in data=journal mode */
1081 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1083 if (!buffer_mapped(bh) || buffer_freed(bh))
1085 set_buffer_uptodate(bh);
1086 return ext4_handle_dirty_metadata(handle, NULL, bh);
1090 * We need to pick up the new inode size which generic_commit_write gave us
1091 * `file' can be NULL - eg, when called from page_symlink().
1093 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1094 * buffers are managed internally.
1096 static int ext4_write_end(struct file *file,
1097 struct address_space *mapping,
1098 loff_t pos, unsigned len, unsigned copied,
1099 struct page *page, void *fsdata)
1101 handle_t *handle = ext4_journal_current_handle();
1102 struct inode *inode = mapping->host;
1104 int i_size_changed = 0;
1106 trace_ext4_write_end(inode, pos, len, copied);
1107 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1108 ret = ext4_jbd2_file_inode(handle, inode);
1111 page_cache_release(page);
1116 if (ext4_has_inline_data(inode))
1117 copied = ext4_write_inline_data_end(inode, pos, len,
1120 copied = block_write_end(file, mapping, pos,
1121 len, copied, page, fsdata);
1124 * No need to use i_size_read() here, the i_size
1125 * cannot change under us because we hole i_mutex.
1127 * But it's important to update i_size while still holding page lock:
1128 * page writeout could otherwise come in and zero beyond i_size.
1130 if (pos + copied > inode->i_size) {
1131 i_size_write(inode, pos + copied);
1135 if (pos + copied > EXT4_I(inode)->i_disksize) {
1136 /* We need to mark inode dirty even if
1137 * new_i_size is less that inode->i_size
1138 * but greater than i_disksize. (hint delalloc)
1140 ext4_update_i_disksize(inode, (pos + copied));
1144 page_cache_release(page);
1147 * Don't mark the inode dirty under page lock. First, it unnecessarily
1148 * makes the holding time of page lock longer. Second, it forces lock
1149 * ordering of page lock and transaction start for journaling
1153 ext4_mark_inode_dirty(handle, inode);
1157 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1158 /* if we have allocated more blocks and copied
1159 * less. We will have blocks allocated outside
1160 * inode->i_size. So truncate them
1162 ext4_orphan_add(handle, inode);
1164 ret2 = ext4_journal_stop(handle);
1168 if (pos + len > inode->i_size) {
1169 ext4_truncate_failed_write(inode);
1171 * If truncate failed early the inode might still be
1172 * on the orphan list; we need to make sure the inode
1173 * is removed from the orphan list in that case.
1176 ext4_orphan_del(NULL, inode);
1179 return ret ? ret : copied;
1182 static int ext4_journalled_write_end(struct file *file,
1183 struct address_space *mapping,
1184 loff_t pos, unsigned len, unsigned copied,
1185 struct page *page, void *fsdata)
1187 handle_t *handle = ext4_journal_current_handle();
1188 struct inode *inode = mapping->host;
1194 trace_ext4_journalled_write_end(inode, pos, len, copied);
1195 from = pos & (PAGE_CACHE_SIZE - 1);
1198 BUG_ON(!ext4_handle_valid(handle));
1200 if (ext4_has_inline_data(inode))
1201 copied = ext4_write_inline_data_end(inode, pos, len,
1205 if (!PageUptodate(page))
1207 page_zero_new_buffers(page, from+copied, to);
1210 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1211 to, &partial, write_end_fn);
1213 SetPageUptodate(page);
1215 new_i_size = pos + copied;
1216 if (new_i_size > inode->i_size)
1217 i_size_write(inode, pos+copied);
1218 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1219 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1220 if (new_i_size > EXT4_I(inode)->i_disksize) {
1221 ext4_update_i_disksize(inode, new_i_size);
1222 ret2 = ext4_mark_inode_dirty(handle, inode);
1228 page_cache_release(page);
1229 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1230 /* if we have allocated more blocks and copied
1231 * less. We will have blocks allocated outside
1232 * inode->i_size. So truncate them
1234 ext4_orphan_add(handle, inode);
1236 ret2 = ext4_journal_stop(handle);
1239 if (pos + len > inode->i_size) {
1240 ext4_truncate_failed_write(inode);
1242 * If truncate failed early the inode might still be
1243 * on the orphan list; we need to make sure the inode
1244 * is removed from the orphan list in that case.
1247 ext4_orphan_del(NULL, inode);
1250 return ret ? ret : copied;
1254 * Reserve a metadata for a single block located at lblock
1256 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1259 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1260 struct ext4_inode_info *ei = EXT4_I(inode);
1261 unsigned int md_needed;
1262 ext4_lblk_t save_last_lblock;
1266 * recalculate the amount of metadata blocks to reserve
1267 * in order to allocate nrblocks
1268 * worse case is one extent per block
1271 spin_lock(&ei->i_block_reservation_lock);
1273 * ext4_calc_metadata_amount() has side effects, which we have
1274 * to be prepared undo if we fail to claim space.
1276 save_len = ei->i_da_metadata_calc_len;
1277 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1278 md_needed = EXT4_NUM_B2C(sbi,
1279 ext4_calc_metadata_amount(inode, lblock));
1280 trace_ext4_da_reserve_space(inode, md_needed);
1283 * We do still charge estimated metadata to the sb though;
1284 * we cannot afford to run out of free blocks.
1286 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1287 ei->i_da_metadata_calc_len = save_len;
1288 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1289 spin_unlock(&ei->i_block_reservation_lock);
1290 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1296 ei->i_reserved_meta_blocks += md_needed;
1297 spin_unlock(&ei->i_block_reservation_lock);
1299 return 0; /* success */
1303 * Reserve a single cluster located at lblock
1305 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1308 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1309 struct ext4_inode_info *ei = EXT4_I(inode);
1310 unsigned int md_needed;
1312 ext4_lblk_t save_last_lblock;
1316 * We will charge metadata quota at writeout time; this saves
1317 * us from metadata over-estimation, though we may go over by
1318 * a small amount in the end. Here we just reserve for data.
1320 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1325 * recalculate the amount of metadata blocks to reserve
1326 * in order to allocate nrblocks
1327 * worse case is one extent per block
1330 spin_lock(&ei->i_block_reservation_lock);
1332 * ext4_calc_metadata_amount() has side effects, which we have
1333 * to be prepared undo if we fail to claim space.
1335 save_len = ei->i_da_metadata_calc_len;
1336 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1337 md_needed = EXT4_NUM_B2C(sbi,
1338 ext4_calc_metadata_amount(inode, lblock));
1339 trace_ext4_da_reserve_space(inode, md_needed);
1342 * We do still charge estimated metadata to the sb though;
1343 * we cannot afford to run out of free blocks.
1345 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1346 ei->i_da_metadata_calc_len = save_len;
1347 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1348 spin_unlock(&ei->i_block_reservation_lock);
1349 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1353 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1356 ei->i_reserved_data_blocks++;
1357 ei->i_reserved_meta_blocks += md_needed;
1358 spin_unlock(&ei->i_block_reservation_lock);
1360 return 0; /* success */
1363 static void ext4_da_release_space(struct inode *inode, int to_free)
1365 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1366 struct ext4_inode_info *ei = EXT4_I(inode);
1369 return; /* Nothing to release, exit */
1371 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1373 trace_ext4_da_release_space(inode, to_free);
1374 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1376 * if there aren't enough reserved blocks, then the
1377 * counter is messed up somewhere. Since this
1378 * function is called from invalidate page, it's
1379 * harmless to return without any action.
1381 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1382 "ino %lu, to_free %d with only %d reserved "
1383 "data blocks", inode->i_ino, to_free,
1384 ei->i_reserved_data_blocks);
1386 to_free = ei->i_reserved_data_blocks;
1388 ei->i_reserved_data_blocks -= to_free;
1390 if (ei->i_reserved_data_blocks == 0) {
1392 * We can release all of the reserved metadata blocks
1393 * only when we have written all of the delayed
1394 * allocation blocks.
1395 * Note that in case of bigalloc, i_reserved_meta_blocks,
1396 * i_reserved_data_blocks, etc. refer to number of clusters.
1398 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1399 ei->i_reserved_meta_blocks);
1400 ei->i_reserved_meta_blocks = 0;
1401 ei->i_da_metadata_calc_len = 0;
1404 /* update fs dirty data blocks counter */
1405 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1407 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1409 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1412 static void ext4_da_page_release_reservation(struct page *page,
1413 unsigned long offset)
1416 struct buffer_head *head, *bh;
1417 unsigned int curr_off = 0;
1418 struct inode *inode = page->mapping->host;
1419 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1423 head = page_buffers(page);
1426 unsigned int next_off = curr_off + bh->b_size;
1428 if ((offset <= curr_off) && (buffer_delay(bh))) {
1430 clear_buffer_delay(bh);
1432 curr_off = next_off;
1433 } while ((bh = bh->b_this_page) != head);
1436 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1437 ext4_es_remove_extent(inode, lblk, to_release);
1440 /* If we have released all the blocks belonging to a cluster, then we
1441 * need to release the reserved space for that cluster. */
1442 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1443 while (num_clusters > 0) {
1444 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1445 ((num_clusters - 1) << sbi->s_cluster_bits);
1446 if (sbi->s_cluster_ratio == 1 ||
1447 !ext4_find_delalloc_cluster(inode, lblk))
1448 ext4_da_release_space(inode, 1);
1455 * Delayed allocation stuff
1459 * mpage_da_submit_io - walks through extent of pages and try to write
1460 * them with writepage() call back
1462 * @mpd->inode: inode
1463 * @mpd->first_page: first page of the extent
1464 * @mpd->next_page: page after the last page of the extent
1466 * By the time mpage_da_submit_io() is called we expect all blocks
1467 * to be allocated. this may be wrong if allocation failed.
1469 * As pages are already locked by write_cache_pages(), we can't use it
1471 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1472 struct ext4_map_blocks *map)
1474 struct pagevec pvec;
1475 unsigned long index, end;
1476 int ret = 0, err, nr_pages, i;
1477 struct inode *inode = mpd->inode;
1478 struct address_space *mapping = inode->i_mapping;
1479 loff_t size = i_size_read(inode);
1480 unsigned int len, block_start;
1481 struct buffer_head *bh, *page_bufs = NULL;
1482 sector_t pblock = 0, cur_logical = 0;
1483 struct ext4_io_submit io_submit;
1485 BUG_ON(mpd->next_page <= mpd->first_page);
1486 ext4_io_submit_init(&io_submit, mpd->wbc);
1487 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1488 if (!io_submit.io_end)
1491 * We need to start from the first_page to the next_page - 1
1492 * to make sure we also write the mapped dirty buffer_heads.
1493 * If we look at mpd->b_blocknr we would only be looking
1494 * at the currently mapped buffer_heads.
1496 index = mpd->first_page;
1497 end = mpd->next_page - 1;
1499 pagevec_init(&pvec, 0);
1500 while (index <= end) {
1501 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1504 for (i = 0; i < nr_pages; i++) {
1506 struct page *page = pvec.pages[i];
1508 index = page->index;
1512 if (index == size >> PAGE_CACHE_SHIFT)
1513 len = size & ~PAGE_CACHE_MASK;
1515 len = PAGE_CACHE_SIZE;
1517 cur_logical = index << (PAGE_CACHE_SHIFT -
1519 pblock = map->m_pblk + (cur_logical -
1524 BUG_ON(!PageLocked(page));
1525 BUG_ON(PageWriteback(page));
1527 bh = page_bufs = page_buffers(page);
1530 if (map && (cur_logical >= map->m_lblk) &&
1531 (cur_logical <= (map->m_lblk +
1532 (map->m_len - 1)))) {
1533 if (buffer_delay(bh)) {
1534 clear_buffer_delay(bh);
1535 bh->b_blocknr = pblock;
1537 if (buffer_unwritten(bh) ||
1539 BUG_ON(bh->b_blocknr != pblock);
1540 if (map->m_flags & EXT4_MAP_UNINIT)
1541 set_buffer_uninit(bh);
1542 clear_buffer_unwritten(bh);
1546 * skip page if block allocation undone and
1549 if (ext4_bh_delay_or_unwritten(NULL, bh))
1551 bh = bh->b_this_page;
1552 block_start += bh->b_size;
1555 } while (bh != page_bufs);
1562 clear_page_dirty_for_io(page);
1563 err = ext4_bio_write_page(&io_submit, page, len,
1566 mpd->pages_written++;
1568 * In error case, we have to continue because
1569 * remaining pages are still locked
1574 pagevec_release(&pvec);
1576 ext4_io_submit(&io_submit);
1577 /* Drop io_end reference we got from init */
1578 ext4_put_io_end_defer(io_submit.io_end);
1582 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1586 struct pagevec pvec;
1587 struct inode *inode = mpd->inode;
1588 struct address_space *mapping = inode->i_mapping;
1589 ext4_lblk_t start, last;
1591 index = mpd->first_page;
1592 end = mpd->next_page - 1;
1594 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1595 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1596 ext4_es_remove_extent(inode, start, last - start + 1);
1598 pagevec_init(&pvec, 0);
1599 while (index <= end) {
1600 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1603 for (i = 0; i < nr_pages; i++) {
1604 struct page *page = pvec.pages[i];
1605 if (page->index > end)
1607 BUG_ON(!PageLocked(page));
1608 BUG_ON(PageWriteback(page));
1609 block_invalidatepage(page, 0);
1610 ClearPageUptodate(page);
1613 index = pvec.pages[nr_pages - 1]->index + 1;
1614 pagevec_release(&pvec);
1619 static void ext4_print_free_blocks(struct inode *inode)
1621 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1622 struct super_block *sb = inode->i_sb;
1623 struct ext4_inode_info *ei = EXT4_I(inode);
1625 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1626 EXT4_C2B(EXT4_SB(inode->i_sb),
1627 ext4_count_free_clusters(sb)));
1628 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1629 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1630 (long long) EXT4_C2B(EXT4_SB(sb),
1631 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1632 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1633 (long long) EXT4_C2B(EXT4_SB(sb),
1634 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1635 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1636 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1637 ei->i_reserved_data_blocks);
1638 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1639 ei->i_reserved_meta_blocks);
1640 ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u",
1641 ei->i_allocated_meta_blocks);
1646 * mpage_da_map_and_submit - go through given space, map them
1647 * if necessary, and then submit them for I/O
1649 * @mpd - bh describing space
1651 * The function skips space we know is already mapped to disk blocks.
1654 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1656 int err, blks, get_blocks_flags;
1657 struct ext4_map_blocks map, *mapp = NULL;
1658 sector_t next = mpd->b_blocknr;
1659 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1660 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1661 handle_t *handle = NULL;
1664 * If the blocks are mapped already, or we couldn't accumulate
1665 * any blocks, then proceed immediately to the submission stage.
1667 if ((mpd->b_size == 0) ||
1668 ((mpd->b_state & (1 << BH_Mapped)) &&
1669 !(mpd->b_state & (1 << BH_Delay)) &&
1670 !(mpd->b_state & (1 << BH_Unwritten))))
1673 handle = ext4_journal_current_handle();
1677 * Call ext4_map_blocks() to allocate any delayed allocation
1678 * blocks, or to convert an uninitialized extent to be
1679 * initialized (in the case where we have written into
1680 * one or more preallocated blocks).
1682 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1683 * indicate that we are on the delayed allocation path. This
1684 * affects functions in many different parts of the allocation
1685 * call path. This flag exists primarily because we don't
1686 * want to change *many* call functions, so ext4_map_blocks()
1687 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1688 * inode's allocation semaphore is taken.
1690 * If the blocks in questions were delalloc blocks, set
1691 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1692 * variables are updated after the blocks have been allocated.
1695 map.m_len = max_blocks;
1697 * We're in delalloc path and it is possible that we're going to
1698 * need more metadata blocks than previously reserved. However
1699 * we must not fail because we're in writeback and there is
1700 * nothing we can do about it so it might result in data loss.
1701 * So use reserved blocks to allocate metadata if possible.
1703 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
1704 EXT4_GET_BLOCKS_METADATA_NOFAIL;
1705 if (ext4_should_dioread_nolock(mpd->inode))
1706 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1707 if (mpd->b_state & (1 << BH_Delay))
1708 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1711 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1713 struct super_block *sb = mpd->inode->i_sb;
1717 * If get block returns EAGAIN or ENOSPC and there
1718 * appears to be free blocks we will just let
1719 * mpage_da_submit_io() unlock all of the pages.
1724 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1730 * get block failure will cause us to loop in
1731 * writepages, because a_ops->writepage won't be able
1732 * to make progress. The page will be redirtied by
1733 * writepage and writepages will again try to write
1736 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1737 ext4_msg(sb, KERN_CRIT,
1738 "delayed block allocation failed for inode %lu "
1739 "at logical offset %llu with max blocks %zd "
1740 "with error %d", mpd->inode->i_ino,
1741 (unsigned long long) next,
1742 mpd->b_size >> mpd->inode->i_blkbits, err);
1743 ext4_msg(sb, KERN_CRIT,
1744 "This should not happen!! Data will be lost");
1746 ext4_print_free_blocks(mpd->inode);
1748 /* invalidate all the pages */
1749 ext4_da_block_invalidatepages(mpd);
1751 /* Mark this page range as having been completed */
1758 if (map.m_flags & EXT4_MAP_NEW) {
1759 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1762 for (i = 0; i < map.m_len; i++)
1763 unmap_underlying_metadata(bdev, map.m_pblk + i);
1767 * Update on-disk size along with block allocation.
1769 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1770 if (disksize > i_size_read(mpd->inode))
1771 disksize = i_size_read(mpd->inode);
1772 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1773 ext4_update_i_disksize(mpd->inode, disksize);
1774 err = ext4_mark_inode_dirty(handle, mpd->inode);
1776 ext4_error(mpd->inode->i_sb,
1777 "Failed to mark inode %lu dirty",
1782 mpage_da_submit_io(mpd, mapp);
1786 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1787 (1 << BH_Delay) | (1 << BH_Unwritten))
1790 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1792 * @mpd->lbh - extent of blocks
1793 * @logical - logical number of the block in the file
1794 * @b_state - b_state of the buffer head added
1796 * the function is used to collect contig. blocks in same state
1798 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical,
1799 unsigned long b_state)
1802 int blkbits = mpd->inode->i_blkbits;
1803 int nrblocks = mpd->b_size >> blkbits;
1806 * XXX Don't go larger than mballoc is willing to allocate
1807 * This is a stopgap solution. We eventually need to fold
1808 * mpage_da_submit_io() into this function and then call
1809 * ext4_map_blocks() multiple times in a loop
1811 if (nrblocks >= (8*1024*1024 >> blkbits))
1814 /* check if the reserved journal credits might overflow */
1815 if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) {
1816 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1818 * With non-extent format we are limited by the journal
1819 * credit available. Total credit needed to insert
1820 * nrblocks contiguous blocks is dependent on the
1821 * nrblocks. So limit nrblocks.
1827 * First block in the extent
1829 if (mpd->b_size == 0) {
1830 mpd->b_blocknr = logical;
1831 mpd->b_size = 1 << blkbits;
1832 mpd->b_state = b_state & BH_FLAGS;
1836 next = mpd->b_blocknr + nrblocks;
1838 * Can we merge the block to our big extent?
1840 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1841 mpd->b_size += 1 << blkbits;
1847 * We couldn't merge the block to our extent, so we
1848 * need to flush current extent and start new one
1850 mpage_da_map_and_submit(mpd);
1854 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1856 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1860 * This function is grabs code from the very beginning of
1861 * ext4_map_blocks, but assumes that the caller is from delayed write
1862 * time. This function looks up the requested blocks and sets the
1863 * buffer delay bit under the protection of i_data_sem.
1865 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1866 struct ext4_map_blocks *map,
1867 struct buffer_head *bh)
1869 struct extent_status es;
1871 sector_t invalid_block = ~((sector_t) 0xffff);
1872 #ifdef ES_AGGRESSIVE_TEST
1873 struct ext4_map_blocks orig_map;
1875 memcpy(&orig_map, map, sizeof(*map));
1878 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1882 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1883 "logical block %lu\n", inode->i_ino, map->m_len,
1884 (unsigned long) map->m_lblk);
1886 /* Lookup extent status tree firstly */
1887 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1889 if (ext4_es_is_hole(&es)) {
1891 down_read((&EXT4_I(inode)->i_data_sem));
1896 * Delayed extent could be allocated by fallocate.
1897 * So we need to check it.
1899 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1900 map_bh(bh, inode->i_sb, invalid_block);
1902 set_buffer_delay(bh);
1906 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1907 retval = es.es_len - (iblock - es.es_lblk);
1908 if (retval > map->m_len)
1909 retval = map->m_len;
1910 map->m_len = retval;
1911 if (ext4_es_is_written(&es))
1912 map->m_flags |= EXT4_MAP_MAPPED;
1913 else if (ext4_es_is_unwritten(&es))
1914 map->m_flags |= EXT4_MAP_UNWRITTEN;
1918 #ifdef ES_AGGRESSIVE_TEST
1919 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1925 * Try to see if we can get the block without requesting a new
1926 * file system block.
1928 down_read((&EXT4_I(inode)->i_data_sem));
1929 if (ext4_has_inline_data(inode)) {
1931 * We will soon create blocks for this page, and let
1932 * us pretend as if the blocks aren't allocated yet.
1933 * In case of clusters, we have to handle the work
1934 * of mapping from cluster so that the reserved space
1935 * is calculated properly.
1937 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1938 ext4_find_delalloc_cluster(inode, map->m_lblk))
1939 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1941 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1942 retval = ext4_ext_map_blocks(NULL, inode, map,
1943 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1945 retval = ext4_ind_map_blocks(NULL, inode, map,
1946 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1952 * XXX: __block_prepare_write() unmaps passed block,
1956 * If the block was allocated from previously allocated cluster,
1957 * then we don't need to reserve it again. However we still need
1958 * to reserve metadata for every block we're going to write.
1960 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1961 ret = ext4_da_reserve_space(inode, iblock);
1963 /* not enough space to reserve */
1968 ret = ext4_da_reserve_metadata(inode, iblock);
1970 /* not enough space to reserve */
1976 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1977 ~0, EXTENT_STATUS_DELAYED);
1983 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1984 * and it should not appear on the bh->b_state.
1986 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1988 map_bh(bh, inode->i_sb, invalid_block);
1990 set_buffer_delay(bh);
1991 } else if (retval > 0) {
1993 unsigned long long status;
1995 #ifdef ES_AGGRESSIVE_TEST
1996 if (retval != map->m_len) {
1997 printk("ES len assertation failed for inode: %lu "
1998 "retval %d != map->m_len %d "
1999 "in %s (lookup)\n", inode->i_ino, retval,
2000 map->m_len, __func__);
2004 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
2005 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
2006 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
2007 map->m_pblk, status);
2013 up_read((&EXT4_I(inode)->i_data_sem));
2019 * This is a special get_blocks_t callback which is used by
2020 * ext4_da_write_begin(). It will either return mapped block or
2021 * reserve space for a single block.
2023 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2024 * We also have b_blocknr = -1 and b_bdev initialized properly
2026 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2027 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2028 * initialized properly.
2030 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2031 struct buffer_head *bh, int create)
2033 struct ext4_map_blocks map;
2036 BUG_ON(create == 0);
2037 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2039 map.m_lblk = iblock;
2043 * first, we need to know whether the block is allocated already
2044 * preallocated blocks are unmapped but should treated
2045 * the same as allocated blocks.
2047 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
2051 map_bh(bh, inode->i_sb, map.m_pblk);
2052 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2054 if (buffer_unwritten(bh)) {
2055 /* A delayed write to unwritten bh should be marked
2056 * new and mapped. Mapped ensures that we don't do
2057 * get_block multiple times when we write to the same
2058 * offset and new ensures that we do proper zero out
2059 * for partial write.
2062 set_buffer_mapped(bh);
2067 static int bget_one(handle_t *handle, struct buffer_head *bh)
2073 static int bput_one(handle_t *handle, struct buffer_head *bh)
2079 static int __ext4_journalled_writepage(struct page *page,
2082 struct address_space *mapping = page->mapping;
2083 struct inode *inode = mapping->host;
2084 struct buffer_head *page_bufs = NULL;
2085 handle_t *handle = NULL;
2086 int ret = 0, err = 0;
2087 int inline_data = ext4_has_inline_data(inode);
2088 struct buffer_head *inode_bh = NULL;
2090 ClearPageChecked(page);
2093 BUG_ON(page->index != 0);
2094 BUG_ON(len > ext4_get_max_inline_size(inode));
2095 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2096 if (inode_bh == NULL)
2099 page_bufs = page_buffers(page);
2104 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2107 /* As soon as we unlock the page, it can go away, but we have
2108 * references to buffers so we are safe */
2111 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2112 ext4_writepage_trans_blocks(inode));
2113 if (IS_ERR(handle)) {
2114 ret = PTR_ERR(handle);
2118 BUG_ON(!ext4_handle_valid(handle));
2121 ret = ext4_journal_get_write_access(handle, inode_bh);
2123 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2126 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2127 do_journal_get_write_access);
2129 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2134 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2135 err = ext4_journal_stop(handle);
2139 if (!ext4_has_inline_data(inode))
2140 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2142 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2149 * Note that we don't need to start a transaction unless we're journaling data
2150 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2151 * need to file the inode to the transaction's list in ordered mode because if
2152 * we are writing back data added by write(), the inode is already there and if
2153 * we are writing back data modified via mmap(), no one guarantees in which
2154 * transaction the data will hit the disk. In case we are journaling data, we
2155 * cannot start transaction directly because transaction start ranks above page
2156 * lock so we have to do some magic.
2158 * This function can get called via...
2159 * - ext4_da_writepages after taking page lock (have journal handle)
2160 * - journal_submit_inode_data_buffers (no journal handle)
2161 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2162 * - grab_page_cache when doing write_begin (have journal handle)
2164 * We don't do any block allocation in this function. If we have page with
2165 * multiple blocks we need to write those buffer_heads that are mapped. This
2166 * is important for mmaped based write. So if we do with blocksize 1K
2167 * truncate(f, 1024);
2168 * a = mmap(f, 0, 4096);
2170 * truncate(f, 4096);
2171 * we have in the page first buffer_head mapped via page_mkwrite call back
2172 * but other buffer_heads would be unmapped but dirty (dirty done via the
2173 * do_wp_page). So writepage should write the first block. If we modify
2174 * the mmap area beyond 1024 we will again get a page_fault and the
2175 * page_mkwrite callback will do the block allocation and mark the
2176 * buffer_heads mapped.
2178 * We redirty the page if we have any buffer_heads that is either delay or
2179 * unwritten in the page.
2181 * We can get recursively called as show below.
2183 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2186 * But since we don't do any block allocation we should not deadlock.
2187 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2189 static int ext4_writepage(struct page *page,
2190 struct writeback_control *wbc)
2195 struct buffer_head *page_bufs = NULL;
2196 struct inode *inode = page->mapping->host;
2197 struct ext4_io_submit io_submit;
2199 trace_ext4_writepage(page);
2200 size = i_size_read(inode);
2201 if (page->index == size >> PAGE_CACHE_SHIFT)
2202 len = size & ~PAGE_CACHE_MASK;
2204 len = PAGE_CACHE_SIZE;
2206 page_bufs = page_buffers(page);
2208 * We cannot do block allocation or other extent handling in this
2209 * function. If there are buffers needing that, we have to redirty
2210 * the page. But we may reach here when we do a journal commit via
2211 * journal_submit_inode_data_buffers() and in that case we must write
2212 * allocated buffers to achieve data=ordered mode guarantees.
2214 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2215 ext4_bh_delay_or_unwritten)) {
2216 redirty_page_for_writepage(wbc, page);
2217 if (current->flags & PF_MEMALLOC) {
2219 * For memory cleaning there's no point in writing only
2220 * some buffers. So just bail out. Warn if we came here
2221 * from direct reclaim.
2223 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2230 if (PageChecked(page) && ext4_should_journal_data(inode))
2232 * It's mmapped pagecache. Add buffers and journal it. There
2233 * doesn't seem much point in redirtying the page here.
2235 return __ext4_journalled_writepage(page, len);
2237 ext4_io_submit_init(&io_submit, wbc);
2238 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2239 if (!io_submit.io_end) {
2240 redirty_page_for_writepage(wbc, page);
2243 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2244 ext4_io_submit(&io_submit);
2245 /* Drop io_end reference we got from init */
2246 ext4_put_io_end_defer(io_submit.io_end);
2251 * This is called via ext4_da_writepages() to
2252 * calculate the total number of credits to reserve to fit
2253 * a single extent allocation into a single transaction,
2254 * ext4_da_writpeages() will loop calling this before
2255 * the block allocation.
2258 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2260 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2263 * With non-extent format the journal credit needed to
2264 * insert nrblocks contiguous block is dependent on
2265 * number of contiguous block. So we will limit
2266 * number of contiguous block to a sane value
2268 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2269 (max_blocks > EXT4_MAX_TRANS_DATA))
2270 max_blocks = EXT4_MAX_TRANS_DATA;
2272 return ext4_chunk_trans_blocks(inode, max_blocks);
2276 * write_cache_pages_da - walk the list of dirty pages of the given
2277 * address space and accumulate pages that need writing, and call
2278 * mpage_da_map_and_submit to map a single contiguous memory region
2279 * and then write them.
2281 static int write_cache_pages_da(handle_t *handle,
2282 struct address_space *mapping,
2283 struct writeback_control *wbc,
2284 struct mpage_da_data *mpd,
2285 pgoff_t *done_index)
2287 struct buffer_head *bh, *head;
2288 struct inode *inode = mapping->host;
2289 struct pagevec pvec;
2290 unsigned int nr_pages;
2293 long nr_to_write = wbc->nr_to_write;
2294 int i, tag, ret = 0;
2296 memset(mpd, 0, sizeof(struct mpage_da_data));
2299 pagevec_init(&pvec, 0);
2300 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2301 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2303 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2304 tag = PAGECACHE_TAG_TOWRITE;
2306 tag = PAGECACHE_TAG_DIRTY;
2308 *done_index = index;
2309 while (index <= end) {
2310 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2311 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2315 for (i = 0; i < nr_pages; i++) {
2316 struct page *page = pvec.pages[i];
2319 * At this point, the page may be truncated or
2320 * invalidated (changing page->mapping to NULL), or
2321 * even swizzled back from swapper_space to tmpfs file
2322 * mapping. However, page->index will not change
2323 * because we have a reference on the page.
2325 if (page->index > end)
2328 *done_index = page->index + 1;
2331 * If we can't merge this page, and we have
2332 * accumulated an contiguous region, write it
2334 if ((mpd->next_page != page->index) &&
2335 (mpd->next_page != mpd->first_page)) {
2336 mpage_da_map_and_submit(mpd);
2337 goto ret_extent_tail;
2343 * If the page is no longer dirty, or its
2344 * mapping no longer corresponds to inode we
2345 * are writing (which means it has been
2346 * truncated or invalidated), or the page is
2347 * already under writeback and we are not
2348 * doing a data integrity writeback, skip the page
2350 if (!PageDirty(page) ||
2351 (PageWriteback(page) &&
2352 (wbc->sync_mode == WB_SYNC_NONE)) ||
2353 unlikely(page->mapping != mapping)) {
2358 wait_on_page_writeback(page);
2359 BUG_ON(PageWriteback(page));
2362 * If we have inline data and arrive here, it means that
2363 * we will soon create the block for the 1st page, so
2364 * we'd better clear the inline data here.
2366 if (ext4_has_inline_data(inode)) {
2367 BUG_ON(ext4_test_inode_state(inode,
2368 EXT4_STATE_MAY_INLINE_DATA));
2369 ext4_destroy_inline_data(handle, inode);
2372 if (mpd->next_page != page->index)
2373 mpd->first_page = page->index;
2374 mpd->next_page = page->index + 1;
2375 logical = (sector_t) page->index <<
2376 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2378 /* Add all dirty buffers to mpd */
2379 head = page_buffers(page);
2382 BUG_ON(buffer_locked(bh));
2384 * We need to try to allocate unmapped blocks
2385 * in the same page. Otherwise we won't make
2386 * progress with the page in ext4_writepage
2388 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2389 mpage_add_bh_to_extent(mpd, logical,
2392 goto ret_extent_tail;
2393 } else if (buffer_dirty(bh) &&
2394 buffer_mapped(bh)) {
2396 * mapped dirty buffer. We need to
2397 * update the b_state because we look
2398 * at b_state in mpage_da_map_blocks.
2399 * We don't update b_size because if we
2400 * find an unmapped buffer_head later
2401 * we need to use the b_state flag of
2404 if (mpd->b_size == 0)
2406 bh->b_state & BH_FLAGS;
2409 } while ((bh = bh->b_this_page) != head);
2411 if (nr_to_write > 0) {
2413 if (nr_to_write == 0 &&
2414 wbc->sync_mode == WB_SYNC_NONE)
2416 * We stop writing back only if we are
2417 * not doing integrity sync. In case of
2418 * integrity sync we have to keep going
2419 * because someone may be concurrently
2420 * dirtying pages, and we might have
2421 * synced a lot of newly appeared dirty
2422 * pages, but have not synced all of the
2428 pagevec_release(&pvec);
2433 ret = MPAGE_DA_EXTENT_TAIL;
2435 pagevec_release(&pvec);
2441 static int ext4_da_writepages(struct address_space *mapping,
2442 struct writeback_control *wbc)
2445 int range_whole = 0;
2446 handle_t *handle = NULL;
2447 struct mpage_da_data mpd;
2448 struct inode *inode = mapping->host;
2449 int pages_written = 0;
2450 unsigned int max_pages;
2451 int range_cyclic, cycled = 1, io_done = 0;
2452 int needed_blocks, ret = 0;
2453 long desired_nr_to_write, nr_to_writebump = 0;
2454 loff_t range_start = wbc->range_start;
2455 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2456 pgoff_t done_index = 0;
2458 struct blk_plug plug;
2460 trace_ext4_da_writepages(inode, wbc);
2463 * No pages to write? This is mainly a kludge to avoid starting
2464 * a transaction for special inodes like journal inode on last iput()
2465 * because that could violate lock ordering on umount
2467 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2471 * If the filesystem has aborted, it is read-only, so return
2472 * right away instead of dumping stack traces later on that
2473 * will obscure the real source of the problem. We test
2474 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2475 * the latter could be true if the filesystem is mounted
2476 * read-only, and in that case, ext4_da_writepages should
2477 * *never* be called, so if that ever happens, we would want
2480 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2483 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2486 range_cyclic = wbc->range_cyclic;
2487 if (wbc->range_cyclic) {
2488 index = mapping->writeback_index;
2491 wbc->range_start = index << PAGE_CACHE_SHIFT;
2492 wbc->range_end = LLONG_MAX;
2493 wbc->range_cyclic = 0;
2496 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2497 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2501 * This works around two forms of stupidity. The first is in
2502 * the writeback code, which caps the maximum number of pages
2503 * written to be 1024 pages. This is wrong on multiple
2504 * levels; different architectues have a different page size,
2505 * which changes the maximum amount of data which gets
2506 * written. Secondly, 4 megabytes is way too small. XFS
2507 * forces this value to be 16 megabytes by multiplying
2508 * nr_to_write parameter by four, and then relies on its
2509 * allocator to allocate larger extents to make them
2510 * contiguous. Unfortunately this brings us to the second
2511 * stupidity, which is that ext4's mballoc code only allocates
2512 * at most 2048 blocks. So we force contiguous writes up to
2513 * the number of dirty blocks in the inode, or
2514 * sbi->max_writeback_mb_bump whichever is smaller.
2516 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2517 if (!range_cyclic && range_whole) {
2518 if (wbc->nr_to_write == LONG_MAX)
2519 desired_nr_to_write = wbc->nr_to_write;
2521 desired_nr_to_write = wbc->nr_to_write * 8;
2523 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2525 if (desired_nr_to_write > max_pages)
2526 desired_nr_to_write = max_pages;
2528 if (wbc->nr_to_write < desired_nr_to_write) {
2529 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2530 wbc->nr_to_write = desired_nr_to_write;
2534 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2535 tag_pages_for_writeback(mapping, index, end);
2537 blk_start_plug(&plug);
2538 while (!ret && wbc->nr_to_write > 0) {
2541 * we insert one extent at a time. So we need
2542 * credit needed for single extent allocation.
2543 * journalled mode is currently not supported
2546 BUG_ON(ext4_should_journal_data(inode));
2547 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2549 /* start a new transaction*/
2550 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2552 if (IS_ERR(handle)) {
2553 ret = PTR_ERR(handle);
2554 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2555 "%ld pages, ino %lu; err %d", __func__,
2556 wbc->nr_to_write, inode->i_ino, ret);
2557 blk_finish_plug(&plug);
2558 goto out_writepages;
2562 * Now call write_cache_pages_da() to find the next
2563 * contiguous region of logical blocks that need
2564 * blocks to be allocated by ext4 and submit them.
2566 ret = write_cache_pages_da(handle, mapping,
2567 wbc, &mpd, &done_index);
2569 * If we have a contiguous extent of pages and we
2570 * haven't done the I/O yet, map the blocks and submit
2573 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2574 mpage_da_map_and_submit(&mpd);
2575 ret = MPAGE_DA_EXTENT_TAIL;
2577 trace_ext4_da_write_pages(inode, &mpd);
2578 wbc->nr_to_write -= mpd.pages_written;
2580 ext4_journal_stop(handle);
2582 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2583 /* commit the transaction which would
2584 * free blocks released in the transaction
2587 jbd2_journal_force_commit_nested(sbi->s_journal);
2589 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2591 * Got one extent now try with rest of the pages.
2592 * If mpd.retval is set -EIO, journal is aborted.
2593 * So we don't need to write any more.
2595 pages_written += mpd.pages_written;
2598 } else if (wbc->nr_to_write)
2600 * There is no more writeout needed
2601 * or we requested for a noblocking writeout
2602 * and we found the device congested
2606 blk_finish_plug(&plug);
2607 if (!io_done && !cycled) {
2610 wbc->range_start = index << PAGE_CACHE_SHIFT;
2611 wbc->range_end = mapping->writeback_index - 1;
2616 wbc->range_cyclic = range_cyclic;
2617 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2619 * set the writeback_index so that range_cyclic
2620 * mode will write it back later
2622 mapping->writeback_index = done_index;
2625 wbc->nr_to_write -= nr_to_writebump;
2626 wbc->range_start = range_start;
2627 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2631 static int ext4_nonda_switch(struct super_block *sb)
2633 s64 free_clusters, dirty_clusters;
2634 struct ext4_sb_info *sbi = EXT4_SB(sb);
2637 * switch to non delalloc mode if we are running low
2638 * on free block. The free block accounting via percpu
2639 * counters can get slightly wrong with percpu_counter_batch getting
2640 * accumulated on each CPU without updating global counters
2641 * Delalloc need an accurate free block accounting. So switch
2642 * to non delalloc when we are near to error range.
2645 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2647 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2649 * Start pushing delalloc when 1/2 of free blocks are dirty.
2651 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2652 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2654 if (2 * free_clusters < 3 * dirty_clusters ||
2655 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2657 * free block count is less than 150% of dirty blocks
2658 * or free blocks is less than watermark
2665 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2666 loff_t pos, unsigned len, unsigned flags,
2667 struct page **pagep, void **fsdata)
2669 int ret, retries = 0;
2672 struct inode *inode = mapping->host;
2675 index = pos >> PAGE_CACHE_SHIFT;
2677 if (ext4_nonda_switch(inode->i_sb)) {
2678 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2679 return ext4_write_begin(file, mapping, pos,
2680 len, flags, pagep, fsdata);
2682 *fsdata = (void *)0;
2683 trace_ext4_da_write_begin(inode, pos, len, flags);
2685 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2686 ret = ext4_da_write_inline_data_begin(mapping, inode,
2696 * grab_cache_page_write_begin() can take a long time if the
2697 * system is thrashing due to memory pressure, or if the page
2698 * is being written back. So grab it first before we start
2699 * the transaction handle. This also allows us to allocate
2700 * the page (if needed) without using GFP_NOFS.
2703 page = grab_cache_page_write_begin(mapping, index, flags);
2709 * With delayed allocation, we don't log the i_disksize update
2710 * if there is delayed block allocation. But we still need
2711 * to journalling the i_disksize update if writes to the end
2712 * of file which has an already mapped buffer.
2715 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2716 if (IS_ERR(handle)) {
2717 page_cache_release(page);
2718 return PTR_ERR(handle);
2722 if (page->mapping != mapping) {
2723 /* The page got truncated from under us */
2725 page_cache_release(page);
2726 ext4_journal_stop(handle);
2729 /* In case writeback began while the page was unlocked */
2730 wait_on_page_writeback(page);
2732 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2735 ext4_journal_stop(handle);
2737 * block_write_begin may have instantiated a few blocks
2738 * outside i_size. Trim these off again. Don't need
2739 * i_size_read because we hold i_mutex.
2741 if (pos + len > inode->i_size)
2742 ext4_truncate_failed_write(inode);
2744 if (ret == -ENOSPC &&
2745 ext4_should_retry_alloc(inode->i_sb, &retries))
2748 page_cache_release(page);
2757 * Check if we should update i_disksize
2758 * when write to the end of file but not require block allocation
2760 static int ext4_da_should_update_i_disksize(struct page *page,
2761 unsigned long offset)
2763 struct buffer_head *bh;
2764 struct inode *inode = page->mapping->host;
2768 bh = page_buffers(page);
2769 idx = offset >> inode->i_blkbits;
2771 for (i = 0; i < idx; i++)
2772 bh = bh->b_this_page;
2774 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2779 static int ext4_da_write_end(struct file *file,
2780 struct address_space *mapping,
2781 loff_t pos, unsigned len, unsigned copied,
2782 struct page *page, void *fsdata)
2784 struct inode *inode = mapping->host;
2786 handle_t *handle = ext4_journal_current_handle();
2788 unsigned long start, end;
2789 int write_mode = (int)(unsigned long)fsdata;
2791 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2792 return ext4_write_end(file, mapping, pos,
2793 len, copied, page, fsdata);
2795 trace_ext4_da_write_end(inode, pos, len, copied);
2796 start = pos & (PAGE_CACHE_SIZE - 1);
2797 end = start + copied - 1;
2800 * generic_write_end() will run mark_inode_dirty() if i_size
2801 * changes. So let's piggyback the i_disksize mark_inode_dirty
2804 new_i_size = pos + copied;
2805 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2806 if (ext4_has_inline_data(inode) ||
2807 ext4_da_should_update_i_disksize(page, end)) {
2808 down_write(&EXT4_I(inode)->i_data_sem);
2809 if (new_i_size > EXT4_I(inode)->i_disksize)
2810 EXT4_I(inode)->i_disksize = new_i_size;
2811 up_write(&EXT4_I(inode)->i_data_sem);
2812 /* We need to mark inode dirty even if
2813 * new_i_size is less that inode->i_size
2814 * bu greater than i_disksize.(hint delalloc)
2816 ext4_mark_inode_dirty(handle, inode);
2820 if (write_mode != CONVERT_INLINE_DATA &&
2821 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2822 ext4_has_inline_data(inode))
2823 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2826 ret2 = generic_write_end(file, mapping, pos, len, copied,
2832 ret2 = ext4_journal_stop(handle);
2836 return ret ? ret : copied;
2839 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2842 * Drop reserved blocks
2844 BUG_ON(!PageLocked(page));
2845 if (!page_has_buffers(page))
2848 ext4_da_page_release_reservation(page, offset);
2851 ext4_invalidatepage(page, offset);
2857 * Force all delayed allocation blocks to be allocated for a given inode.
2859 int ext4_alloc_da_blocks(struct inode *inode)
2861 trace_ext4_alloc_da_blocks(inode);
2863 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2864 !EXT4_I(inode)->i_reserved_meta_blocks)
2868 * We do something simple for now. The filemap_flush() will
2869 * also start triggering a write of the data blocks, which is
2870 * not strictly speaking necessary (and for users of
2871 * laptop_mode, not even desirable). However, to do otherwise
2872 * would require replicating code paths in:
2874 * ext4_da_writepages() ->
2875 * write_cache_pages() ---> (via passed in callback function)
2876 * __mpage_da_writepage() -->
2877 * mpage_add_bh_to_extent()
2878 * mpage_da_map_blocks()
2880 * The problem is that write_cache_pages(), located in
2881 * mm/page-writeback.c, marks pages clean in preparation for
2882 * doing I/O, which is not desirable if we're not planning on
2885 * We could call write_cache_pages(), and then redirty all of
2886 * the pages by calling redirty_page_for_writepage() but that
2887 * would be ugly in the extreme. So instead we would need to
2888 * replicate parts of the code in the above functions,
2889 * simplifying them because we wouldn't actually intend to
2890 * write out the pages, but rather only collect contiguous
2891 * logical block extents, call the multi-block allocator, and
2892 * then update the buffer heads with the block allocations.
2894 * For now, though, we'll cheat by calling filemap_flush(),
2895 * which will map the blocks, and start the I/O, but not
2896 * actually wait for the I/O to complete.
2898 return filemap_flush(inode->i_mapping);
2902 * bmap() is special. It gets used by applications such as lilo and by
2903 * the swapper to find the on-disk block of a specific piece of data.
2905 * Naturally, this is dangerous if the block concerned is still in the
2906 * journal. If somebody makes a swapfile on an ext4 data-journaling
2907 * filesystem and enables swap, then they may get a nasty shock when the
2908 * data getting swapped to that swapfile suddenly gets overwritten by
2909 * the original zero's written out previously to the journal and
2910 * awaiting writeback in the kernel's buffer cache.
2912 * So, if we see any bmap calls here on a modified, data-journaled file,
2913 * take extra steps to flush any blocks which might be in the cache.
2915 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2917 struct inode *inode = mapping->host;
2922 * We can get here for an inline file via the FIBMAP ioctl
2924 if (ext4_has_inline_data(inode))
2927 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2928 test_opt(inode->i_sb, DELALLOC)) {
2930 * With delalloc we want to sync the file
2931 * so that we can make sure we allocate
2934 filemap_write_and_wait(mapping);
2937 if (EXT4_JOURNAL(inode) &&
2938 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2940 * This is a REALLY heavyweight approach, but the use of
2941 * bmap on dirty files is expected to be extremely rare:
2942 * only if we run lilo or swapon on a freshly made file
2943 * do we expect this to happen.
2945 * (bmap requires CAP_SYS_RAWIO so this does not
2946 * represent an unprivileged user DOS attack --- we'd be
2947 * in trouble if mortal users could trigger this path at
2950 * NB. EXT4_STATE_JDATA is not set on files other than
2951 * regular files. If somebody wants to bmap a directory
2952 * or symlink and gets confused because the buffer
2953 * hasn't yet been flushed to disk, they deserve
2954 * everything they get.
2957 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2958 journal = EXT4_JOURNAL(inode);
2959 jbd2_journal_lock_updates(journal);
2960 err = jbd2_journal_flush(journal);
2961 jbd2_journal_unlock_updates(journal);
2967 return generic_block_bmap(mapping, block, ext4_get_block);
2970 static int ext4_readpage(struct file *file, struct page *page)
2973 struct inode *inode = page->mapping->host;
2975 trace_ext4_readpage(page);
2977 if (ext4_has_inline_data(inode))
2978 ret = ext4_readpage_inline(inode, page);
2981 return mpage_readpage(page, ext4_get_block);
2987 ext4_readpages(struct file *file, struct address_space *mapping,
2988 struct list_head *pages, unsigned nr_pages)
2990 struct inode *inode = mapping->host;
2992 /* If the file has inline data, no need to do readpages. */
2993 if (ext4_has_inline_data(inode))
2996 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2999 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3001 trace_ext4_invalidatepage(page, offset);
3003 /* No journalling happens on data buffers when this function is used */
3004 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3006 block_invalidatepage(page, offset);
3009 static int __ext4_journalled_invalidatepage(struct page *page,
3010 unsigned long offset)
3012 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3014 trace_ext4_journalled_invalidatepage(page, offset);
3017 * If it's a full truncate we just forget about the pending dirtying
3020 ClearPageChecked(page);
3022 return jbd2_journal_invalidatepage(journal, page, offset);
3025 /* Wrapper for aops... */
3026 static void ext4_journalled_invalidatepage(struct page *page,
3027 unsigned long offset)
3029 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
3032 static int ext4_releasepage(struct page *page, gfp_t wait)
3034 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3036 trace_ext4_releasepage(page);
3038 /* Page has dirty journalled data -> cannot release */
3039 if (PageChecked(page))
3042 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3044 return try_to_free_buffers(page);
3048 * ext4_get_block used when preparing for a DIO write or buffer write.
3049 * We allocate an uinitialized extent if blocks haven't been allocated.
3050 * The extent will be converted to initialized after the IO is complete.
3052 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3053 struct buffer_head *bh_result, int create)
3055 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3056 inode->i_ino, create);
3057 return _ext4_get_block(inode, iblock, bh_result,
3058 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3061 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3062 struct buffer_head *bh_result, int create)
3064 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3065 inode->i_ino, create);
3066 return _ext4_get_block(inode, iblock, bh_result,
3067 EXT4_GET_BLOCKS_NO_LOCK);
3070 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3071 ssize_t size, void *private, int ret,
3074 struct inode *inode = file_inode(iocb->ki_filp);
3075 ext4_io_end_t *io_end = iocb->private;
3077 /* if not async direct IO just return */
3079 inode_dio_done(inode);
3081 aio_complete(iocb, ret, 0);
3085 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3086 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3087 iocb->private, io_end->inode->i_ino, iocb, offset,
3090 iocb->private = NULL;
3091 io_end->offset = offset;
3092 io_end->size = size;
3094 io_end->iocb = iocb;
3095 io_end->result = ret;
3097 ext4_put_io_end_defer(io_end);
3101 * For ext4 extent files, ext4 will do direct-io write to holes,
3102 * preallocated extents, and those write extend the file, no need to
3103 * fall back to buffered IO.
3105 * For holes, we fallocate those blocks, mark them as uninitialized
3106 * If those blocks were preallocated, we mark sure they are split, but
3107 * still keep the range to write as uninitialized.
3109 * The unwritten extents will be converted to written when DIO is completed.
3110 * For async direct IO, since the IO may still pending when return, we
3111 * set up an end_io call back function, which will do the conversion
3112 * when async direct IO completed.
3114 * If the O_DIRECT write will extend the file then add this inode to the
3115 * orphan list. So recovery will truncate it back to the original size
3116 * if the machine crashes during the write.
3119 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3120 const struct iovec *iov, loff_t offset,
3121 unsigned long nr_segs)
3123 struct file *file = iocb->ki_filp;
3124 struct inode *inode = file->f_mapping->host;
3126 size_t count = iov_length(iov, nr_segs);
3128 get_block_t *get_block_func = NULL;
3130 loff_t final_size = offset + count;
3131 ext4_io_end_t *io_end = NULL;
3133 /* Use the old path for reads and writes beyond i_size. */
3134 if (rw != WRITE || final_size > inode->i_size)
3135 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3137 BUG_ON(iocb->private == NULL);
3139 /* If we do a overwrite dio, i_mutex locking can be released */
3140 overwrite = *((int *)iocb->private);
3143 atomic_inc(&inode->i_dio_count);
3144 down_read(&EXT4_I(inode)->i_data_sem);
3145 mutex_unlock(&inode->i_mutex);
3149 * We could direct write to holes and fallocate.
3151 * Allocated blocks to fill the hole are marked as
3152 * uninitialized to prevent parallel buffered read to expose
3153 * the stale data before DIO complete the data IO.
3155 * As to previously fallocated extents, ext4 get_block will
3156 * just simply mark the buffer mapped but still keep the
3157 * extents uninitialized.
3159 * For non AIO case, we will convert those unwritten extents
3160 * to written after return back from blockdev_direct_IO.
3162 * For async DIO, the conversion needs to be deferred when the
3163 * IO is completed. The ext4 end_io callback function will be
3164 * called to take care of the conversion work. Here for async
3165 * case, we allocate an io_end structure to hook to the iocb.
3167 iocb->private = NULL;
3168 ext4_inode_aio_set(inode, NULL);
3169 if (!is_sync_kiocb(iocb)) {
3170 io_end = ext4_init_io_end(inode, GFP_NOFS);
3175 io_end->flag |= EXT4_IO_END_DIRECT;
3177 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3179 iocb->private = ext4_get_io_end(io_end);
3181 * we save the io structure for current async direct
3182 * IO, so that later ext4_map_blocks() could flag the
3183 * io structure whether there is a unwritten extents
3184 * needs to be converted when IO is completed.
3186 ext4_inode_aio_set(inode, io_end);
3190 get_block_func = ext4_get_block_write_nolock;
3192 get_block_func = ext4_get_block_write;
3193 dio_flags = DIO_LOCKING;
3195 ret = __blockdev_direct_IO(rw, iocb, inode,
3196 inode->i_sb->s_bdev, iov,
3204 * Put our reference to io_end. This can free the io_end structure e.g.
3205 * in sync IO case or in case of error. It can even perform extent
3206 * conversion if all bios we submitted finished before we got here.
3207 * Note that in that case iocb->private can be already set to NULL
3211 ext4_inode_aio_set(inode, NULL);
3212 ext4_put_io_end(io_end);
3214 * In case of error or no write ext4_end_io_dio() was not
3215 * called so we have to put iocb's reference.
3217 if (ret <= 0 && ret != -EIOCBQUEUED) {
3218 WARN_ON(iocb->private != io_end);
3219 ext4_put_io_end(io_end);
3220 iocb->private = NULL;
3223 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3224 EXT4_STATE_DIO_UNWRITTEN)) {
3227 * for non AIO case, since the IO is already
3228 * completed, we could do the conversion right here
3230 err = ext4_convert_unwritten_extents(inode,
3234 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3238 /* take i_mutex locking again if we do a ovewrite dio */
3240 inode_dio_done(inode);
3241 up_read(&EXT4_I(inode)->i_data_sem);
3242 mutex_lock(&inode->i_mutex);
3248 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3249 const struct iovec *iov, loff_t offset,
3250 unsigned long nr_segs)
3252 struct file *file = iocb->ki_filp;
3253 struct inode *inode = file->f_mapping->host;
3257 * If we are doing data journalling we don't support O_DIRECT
3259 if (ext4_should_journal_data(inode))
3262 /* Let buffer I/O handle the inline data case. */
3263 if (ext4_has_inline_data(inode))
3266 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3267 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3268 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3270 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3271 trace_ext4_direct_IO_exit(inode, offset,
3272 iov_length(iov, nr_segs), rw, ret);
3277 * Pages can be marked dirty completely asynchronously from ext4's journalling
3278 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3279 * much here because ->set_page_dirty is called under VFS locks. The page is
3280 * not necessarily locked.
3282 * We cannot just dirty the page and leave attached buffers clean, because the
3283 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3284 * or jbddirty because all the journalling code will explode.
3286 * So what we do is to mark the page "pending dirty" and next time writepage
3287 * is called, propagate that into the buffers appropriately.
3289 static int ext4_journalled_set_page_dirty(struct page *page)
3291 SetPageChecked(page);
3292 return __set_page_dirty_nobuffers(page);
3295 static const struct address_space_operations ext4_aops = {
3296 .readpage = ext4_readpage,
3297 .readpages = ext4_readpages,
3298 .writepage = ext4_writepage,
3299 .write_begin = ext4_write_begin,
3300 .write_end = ext4_write_end,
3302 .invalidatepage = ext4_invalidatepage,
3303 .releasepage = ext4_releasepage,
3304 .direct_IO = ext4_direct_IO,
3305 .migratepage = buffer_migrate_page,
3306 .is_partially_uptodate = block_is_partially_uptodate,
3307 .error_remove_page = generic_error_remove_page,
3310 static const struct address_space_operations ext4_journalled_aops = {
3311 .readpage = ext4_readpage,
3312 .readpages = ext4_readpages,
3313 .writepage = ext4_writepage,
3314 .write_begin = ext4_write_begin,
3315 .write_end = ext4_journalled_write_end,
3316 .set_page_dirty = ext4_journalled_set_page_dirty,
3318 .invalidatepage = ext4_journalled_invalidatepage,
3319 .releasepage = ext4_releasepage,
3320 .direct_IO = ext4_direct_IO,
3321 .is_partially_uptodate = block_is_partially_uptodate,
3322 .error_remove_page = generic_error_remove_page,
3325 static const struct address_space_operations ext4_da_aops = {
3326 .readpage = ext4_readpage,
3327 .readpages = ext4_readpages,
3328 .writepage = ext4_writepage,
3329 .writepages = ext4_da_writepages,
3330 .write_begin = ext4_da_write_begin,
3331 .write_end = ext4_da_write_end,
3333 .invalidatepage = ext4_da_invalidatepage,
3334 .releasepage = ext4_releasepage,
3335 .direct_IO = ext4_direct_IO,
3336 .migratepage = buffer_migrate_page,
3337 .is_partially_uptodate = block_is_partially_uptodate,
3338 .error_remove_page = generic_error_remove_page,
3341 void ext4_set_aops(struct inode *inode)
3343 switch (ext4_inode_journal_mode(inode)) {
3344 case EXT4_INODE_ORDERED_DATA_MODE:
3345 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3347 case EXT4_INODE_WRITEBACK_DATA_MODE:
3348 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3350 case EXT4_INODE_JOURNAL_DATA_MODE:
3351 inode->i_mapping->a_ops = &ext4_journalled_aops;
3356 if (test_opt(inode->i_sb, DELALLOC))
3357 inode->i_mapping->a_ops = &ext4_da_aops;
3359 inode->i_mapping->a_ops = &ext4_aops;
3364 * ext4_discard_partial_page_buffers()
3365 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3366 * This function finds and locks the page containing the offset
3367 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3368 * Calling functions that already have the page locked should call
3369 * ext4_discard_partial_page_buffers_no_lock directly.
3371 int ext4_discard_partial_page_buffers(handle_t *handle,
3372 struct address_space *mapping, loff_t from,
3373 loff_t length, int flags)
3375 struct inode *inode = mapping->host;
3379 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3380 mapping_gfp_mask(mapping) & ~__GFP_FS);
3384 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3385 from, length, flags);
3388 page_cache_release(page);
3393 * ext4_discard_partial_page_buffers_no_lock()
3394 * Zeros a page range of length 'length' starting from offset 'from'.
3395 * Buffer heads that correspond to the block aligned regions of the
3396 * zeroed range will be unmapped. Unblock aligned regions
3397 * will have the corresponding buffer head mapped if needed so that
3398 * that region of the page can be updated with the partial zero out.
3400 * This function assumes that the page has already been locked. The
3401 * The range to be discarded must be contained with in the given page.
3402 * If the specified range exceeds the end of the page it will be shortened
3403 * to the end of the page that corresponds to 'from'. This function is
3404 * appropriate for updating a page and it buffer heads to be unmapped and
3405 * zeroed for blocks that have been either released, or are going to be
3408 * handle: The journal handle
3409 * inode: The files inode
3410 * page: A locked page that contains the offset "from"
3411 * from: The starting byte offset (from the beginning of the file)
3412 * to begin discarding
3413 * len: The length of bytes to discard
3414 * flags: Optional flags that may be used:
3416 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3417 * Only zero the regions of the page whose buffer heads
3418 * have already been unmapped. This flag is appropriate
3419 * for updating the contents of a page whose blocks may
3420 * have already been released, and we only want to zero
3421 * out the regions that correspond to those released blocks.
3423 * Returns zero on success or negative on failure.
3425 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3426 struct inode *inode, struct page *page, loff_t from,
3427 loff_t length, int flags)
3429 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3430 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3431 unsigned int blocksize, max, pos;
3433 struct buffer_head *bh;
3436 blocksize = inode->i_sb->s_blocksize;
3437 max = PAGE_CACHE_SIZE - offset;
3439 if (index != page->index)
3443 * correct length if it does not fall between
3444 * 'from' and the end of the page
3446 if (length > max || length < 0)
3449 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3451 if (!page_has_buffers(page))
3452 create_empty_buffers(page, blocksize, 0);
3454 /* Find the buffer that contains "offset" */
3455 bh = page_buffers(page);
3457 while (offset >= pos) {
3458 bh = bh->b_this_page;
3464 while (pos < offset + length) {
3465 unsigned int end_of_block, range_to_discard;
3469 /* The length of space left to zero and unmap */
3470 range_to_discard = offset + length - pos;
3472 /* The length of space until the end of the block */
3473 end_of_block = blocksize - (pos & (blocksize-1));
3476 * Do not unmap or zero past end of block
3477 * for this buffer head
3479 if (range_to_discard > end_of_block)
3480 range_to_discard = end_of_block;
3484 * Skip this buffer head if we are only zeroing unampped
3485 * regions of the page
3487 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3491 /* If the range is block aligned, unmap */
3492 if (range_to_discard == blocksize) {
3493 clear_buffer_dirty(bh);
3495 clear_buffer_mapped(bh);
3496 clear_buffer_req(bh);
3497 clear_buffer_new(bh);
3498 clear_buffer_delay(bh);
3499 clear_buffer_unwritten(bh);
3500 clear_buffer_uptodate(bh);
3501 zero_user(page, pos, range_to_discard);
3502 BUFFER_TRACE(bh, "Buffer discarded");
3507 * If this block is not completely contained in the range
3508 * to be discarded, then it is not going to be released. Because
3509 * we need to keep this block, we need to make sure this part
3510 * of the page is uptodate before we modify it by writeing
3511 * partial zeros on it.
3513 if (!buffer_mapped(bh)) {
3515 * Buffer head must be mapped before we can read
3518 BUFFER_TRACE(bh, "unmapped");
3519 ext4_get_block(inode, iblock, bh, 0);
3520 /* unmapped? It's a hole - nothing to do */
3521 if (!buffer_mapped(bh)) {
3522 BUFFER_TRACE(bh, "still unmapped");
3527 /* Ok, it's mapped. Make sure it's up-to-date */
3528 if (PageUptodate(page))
3529 set_buffer_uptodate(bh);
3531 if (!buffer_uptodate(bh)) {
3533 ll_rw_block(READ, 1, &bh);
3535 /* Uhhuh. Read error. Complain and punt.*/
3536 if (!buffer_uptodate(bh))
3540 if (ext4_should_journal_data(inode)) {
3541 BUFFER_TRACE(bh, "get write access");
3542 err = ext4_journal_get_write_access(handle, bh);
3547 zero_user(page, pos, range_to_discard);
3550 if (ext4_should_journal_data(inode)) {
3551 err = ext4_handle_dirty_metadata(handle, inode, bh);
3553 mark_buffer_dirty(bh);
3555 BUFFER_TRACE(bh, "Partial buffer zeroed");
3557 bh = bh->b_this_page;
3559 pos += range_to_discard;
3565 int ext4_can_truncate(struct inode *inode)
3567 if (S_ISREG(inode->i_mode))
3569 if (S_ISDIR(inode->i_mode))
3571 if (S_ISLNK(inode->i_mode))
3572 return !ext4_inode_is_fast_symlink(inode);
3577 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3578 * associated with the given offset and length
3580 * @inode: File inode
3581 * @offset: The offset where the hole will begin
3582 * @len: The length of the hole
3584 * Returns: 0 on success or negative on failure
3587 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3589 struct inode *inode = file_inode(file);
3590 struct super_block *sb = inode->i_sb;
3591 ext4_lblk_t first_block, stop_block;
3592 struct address_space *mapping = inode->i_mapping;
3593 loff_t first_page, last_page, page_len;
3594 loff_t first_page_offset, last_page_offset;
3596 unsigned int credits;
3599 if (!S_ISREG(inode->i_mode))
3602 if (EXT4_SB(sb)->s_cluster_ratio > 1) {
3603 /* TODO: Add support for bigalloc file systems */
3607 trace_ext4_punch_hole(inode, offset, length);
3610 * Write out all dirty pages to avoid race conditions
3611 * Then release them.
3613 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3614 ret = filemap_write_and_wait_range(mapping, offset,
3615 offset + length - 1);
3620 mutex_lock(&inode->i_mutex);
3621 /* It's not possible punch hole on append only file */
3622 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
3626 if (IS_SWAPFILE(inode)) {
3631 /* No need to punch hole beyond i_size */
3632 if (offset >= inode->i_size)
3636 * If the hole extends beyond i_size, set the hole
3637 * to end after the page that contains i_size
3639 if (offset + length > inode->i_size) {
3640 length = inode->i_size +
3641 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3645 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
3646 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
3648 first_page_offset = first_page << PAGE_CACHE_SHIFT;
3649 last_page_offset = last_page << PAGE_CACHE_SHIFT;
3651 /* Now release the pages */
3652 if (last_page_offset > first_page_offset) {
3653 truncate_pagecache_range(inode, first_page_offset,
3654 last_page_offset - 1);
3657 /* Wait all existing dio workers, newcomers will block on i_mutex */
3658 ext4_inode_block_unlocked_dio(inode);
3659 ret = ext4_flush_unwritten_io(inode);
3662 inode_dio_wait(inode);
3664 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3665 credits = ext4_writepage_trans_blocks(inode);
3667 credits = ext4_blocks_for_truncate(inode);
3668 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3669 if (IS_ERR(handle)) {
3670 ret = PTR_ERR(handle);
3671 ext4_std_error(sb, ret);
3676 * Now we need to zero out the non-page-aligned data in the
3677 * pages at the start and tail of the hole, and unmap the
3678 * buffer heads for the block aligned regions of the page that
3679 * were completely zeroed.
3681 if (first_page > last_page) {
3683 * If the file space being truncated is contained
3684 * within a page just zero out and unmap the middle of
3687 ret = ext4_discard_partial_page_buffers(handle,
3688 mapping, offset, length, 0);
3694 * zero out and unmap the partial page that contains
3695 * the start of the hole
3697 page_len = first_page_offset - offset;
3699 ret = ext4_discard_partial_page_buffers(handle, mapping,
3700 offset, page_len, 0);
3706 * zero out and unmap the partial page that contains
3707 * the end of the hole
3709 page_len = offset + length - last_page_offset;
3711 ret = ext4_discard_partial_page_buffers(handle, mapping,
3712 last_page_offset, page_len, 0);
3719 * If i_size is contained in the last page, we need to
3720 * unmap and zero the partial page after i_size
3722 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
3723 inode->i_size % PAGE_CACHE_SIZE != 0) {
3724 page_len = PAGE_CACHE_SIZE -
3725 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3728 ret = ext4_discard_partial_page_buffers(handle,
3729 mapping, inode->i_size, page_len, 0);
3736 first_block = (offset + sb->s_blocksize - 1) >>
3737 EXT4_BLOCK_SIZE_BITS(sb);
3738 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3740 /* If there are no blocks to remove, return now */
3741 if (first_block >= stop_block)
3744 down_write(&EXT4_I(inode)->i_data_sem);
3745 ext4_discard_preallocations(inode);
3747 ret = ext4_es_remove_extent(inode, first_block,
3748 stop_block - first_block);
3750 up_write(&EXT4_I(inode)->i_data_sem);
3754 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3755 ret = ext4_ext_remove_space(inode, first_block,
3758 ret = ext4_free_hole_blocks(handle, inode, first_block,
3761 ext4_discard_preallocations(inode);
3762 up_write(&EXT4_I(inode)->i_data_sem);
3764 ext4_handle_sync(handle);
3765 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3766 ext4_mark_inode_dirty(handle, inode);
3768 ext4_journal_stop(handle);
3770 ext4_inode_resume_unlocked_dio(inode);
3772 mutex_unlock(&inode->i_mutex);
3779 * We block out ext4_get_block() block instantiations across the entire
3780 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3781 * simultaneously on behalf of the same inode.
3783 * As we work through the truncate and commit bits of it to the journal there
3784 * is one core, guiding principle: the file's tree must always be consistent on
3785 * disk. We must be able to restart the truncate after a crash.
3787 * The file's tree may be transiently inconsistent in memory (although it
3788 * probably isn't), but whenever we close off and commit a journal transaction,
3789 * the contents of (the filesystem + the journal) must be consistent and
3790 * restartable. It's pretty simple, really: bottom up, right to left (although
3791 * left-to-right works OK too).
3793 * Note that at recovery time, journal replay occurs *before* the restart of
3794 * truncate against the orphan inode list.
3796 * The committed inode has the new, desired i_size (which is the same as
3797 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3798 * that this inode's truncate did not complete and it will again call
3799 * ext4_truncate() to have another go. So there will be instantiated blocks
3800 * to the right of the truncation point in a crashed ext4 filesystem. But
3801 * that's fine - as long as they are linked from the inode, the post-crash
3802 * ext4_truncate() run will find them and release them.
3804 void ext4_truncate(struct inode *inode)
3806 struct ext4_inode_info *ei = EXT4_I(inode);
3807 unsigned int credits;
3809 struct address_space *mapping = inode->i_mapping;
3813 * There is a possibility that we're either freeing the inode
3814 * or it completely new indode. In those cases we might not
3815 * have i_mutex locked because it's not necessary.
3817 if (!(inode->i_state & (I_NEW|I_FREEING)))
3818 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3819 trace_ext4_truncate_enter(inode);
3821 if (!ext4_can_truncate(inode))
3824 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3826 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3827 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3829 if (ext4_has_inline_data(inode)) {
3832 ext4_inline_data_truncate(inode, &has_inline);
3838 * finish any pending end_io work so we won't run the risk of
3839 * converting any truncated blocks to initialized later
3841 ext4_flush_unwritten_io(inode);
3843 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3844 credits = ext4_writepage_trans_blocks(inode);
3846 credits = ext4_blocks_for_truncate(inode);
3848 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3849 if (IS_ERR(handle)) {
3850 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3854 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
3855 page_len = PAGE_CACHE_SIZE -
3856 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3858 if (ext4_discard_partial_page_buffers(handle,
3859 mapping, inode->i_size, page_len, 0))
3864 * We add the inode to the orphan list, so that if this
3865 * truncate spans multiple transactions, and we crash, we will
3866 * resume the truncate when the filesystem recovers. It also
3867 * marks the inode dirty, to catch the new size.
3869 * Implication: the file must always be in a sane, consistent
3870 * truncatable state while each transaction commits.
3872 if (ext4_orphan_add(handle, inode))
3875 down_write(&EXT4_I(inode)->i_data_sem);
3877 ext4_discard_preallocations(inode);
3879 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3880 ext4_ext_truncate(handle, inode);
3882 ext4_ind_truncate(handle, inode);
3884 up_write(&ei->i_data_sem);
3887 ext4_handle_sync(handle);
3891 * If this was a simple ftruncate() and the file will remain alive,
3892 * then we need to clear up the orphan record which we created above.
3893 * However, if this was a real unlink then we were called by
3894 * ext4_delete_inode(), and we allow that function to clean up the
3895 * orphan info for us.
3898 ext4_orphan_del(handle, inode);
3900 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3901 ext4_mark_inode_dirty(handle, inode);
3902 ext4_journal_stop(handle);
3904 trace_ext4_truncate_exit(inode);
3908 * ext4_get_inode_loc returns with an extra refcount against the inode's
3909 * underlying buffer_head on success. If 'in_mem' is true, we have all
3910 * data in memory that is needed to recreate the on-disk version of this
3913 static int __ext4_get_inode_loc(struct inode *inode,
3914 struct ext4_iloc *iloc, int in_mem)
3916 struct ext4_group_desc *gdp;
3917 struct buffer_head *bh;
3918 struct super_block *sb = inode->i_sb;
3920 int inodes_per_block, inode_offset;
3923 if (!ext4_valid_inum(sb, inode->i_ino))
3926 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3927 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3932 * Figure out the offset within the block group inode table
3934 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3935 inode_offset = ((inode->i_ino - 1) %
3936 EXT4_INODES_PER_GROUP(sb));
3937 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3938 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3940 bh = sb_getblk(sb, block);
3943 if (!buffer_uptodate(bh)) {
3947 * If the buffer has the write error flag, we have failed
3948 * to write out another inode in the same block. In this
3949 * case, we don't have to read the block because we may
3950 * read the old inode data successfully.
3952 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3953 set_buffer_uptodate(bh);
3955 if (buffer_uptodate(bh)) {
3956 /* someone brought it uptodate while we waited */
3962 * If we have all information of the inode in memory and this
3963 * is the only valid inode in the block, we need not read the
3967 struct buffer_head *bitmap_bh;
3970 start = inode_offset & ~(inodes_per_block - 1);
3972 /* Is the inode bitmap in cache? */
3973 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3974 if (unlikely(!bitmap_bh))
3978 * If the inode bitmap isn't in cache then the
3979 * optimisation may end up performing two reads instead
3980 * of one, so skip it.
3982 if (!buffer_uptodate(bitmap_bh)) {
3986 for (i = start; i < start + inodes_per_block; i++) {
3987 if (i == inode_offset)
3989 if (ext4_test_bit(i, bitmap_bh->b_data))
3993 if (i == start + inodes_per_block) {
3994 /* all other inodes are free, so skip I/O */
3995 memset(bh->b_data, 0, bh->b_size);
3996 set_buffer_uptodate(bh);
4004 * If we need to do any I/O, try to pre-readahead extra
4005 * blocks from the inode table.
4007 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4008 ext4_fsblk_t b, end, table;
4011 table = ext4_inode_table(sb, gdp);
4012 /* s_inode_readahead_blks is always a power of 2 */
4013 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4016 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4017 num = EXT4_INODES_PER_GROUP(sb);
4018 if (ext4_has_group_desc_csum(sb))
4019 num -= ext4_itable_unused_count(sb, gdp);
4020 table += num / inodes_per_block;
4024 sb_breadahead(sb, b++);
4028 * There are other valid inodes in the buffer, this inode
4029 * has in-inode xattrs, or we don't have this inode in memory.
4030 * Read the block from disk.
4032 trace_ext4_load_inode(inode);
4034 bh->b_end_io = end_buffer_read_sync;
4035 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4037 if (!buffer_uptodate(bh)) {
4038 EXT4_ERROR_INODE_BLOCK(inode, block,
4039 "unable to read itable block");
4049 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4051 /* We have all inode data except xattrs in memory here. */
4052 return __ext4_get_inode_loc(inode, iloc,
4053 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4056 void ext4_set_inode_flags(struct inode *inode)
4058 unsigned int flags = EXT4_I(inode)->i_flags;
4060 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4061 if (flags & EXT4_SYNC_FL)
4062 inode->i_flags |= S_SYNC;
4063 if (flags & EXT4_APPEND_FL)
4064 inode->i_flags |= S_APPEND;
4065 if (flags & EXT4_IMMUTABLE_FL)
4066 inode->i_flags |= S_IMMUTABLE;
4067 if (flags & EXT4_NOATIME_FL)
4068 inode->i_flags |= S_NOATIME;
4069 if (flags & EXT4_DIRSYNC_FL)
4070 inode->i_flags |= S_DIRSYNC;
4073 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4074 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4076 unsigned int vfs_fl;
4077 unsigned long old_fl, new_fl;
4080 vfs_fl = ei->vfs_inode.i_flags;
4081 old_fl = ei->i_flags;
4082 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4083 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4085 if (vfs_fl & S_SYNC)
4086 new_fl |= EXT4_SYNC_FL;
4087 if (vfs_fl & S_APPEND)
4088 new_fl |= EXT4_APPEND_FL;
4089 if (vfs_fl & S_IMMUTABLE)
4090 new_fl |= EXT4_IMMUTABLE_FL;
4091 if (vfs_fl & S_NOATIME)
4092 new_fl |= EXT4_NOATIME_FL;
4093 if (vfs_fl & S_DIRSYNC)
4094 new_fl |= EXT4_DIRSYNC_FL;
4095 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4098 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4099 struct ext4_inode_info *ei)
4102 struct inode *inode = &(ei->vfs_inode);
4103 struct super_block *sb = inode->i_sb;
4105 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4106 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4107 /* we are using combined 48 bit field */
4108 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4109 le32_to_cpu(raw_inode->i_blocks_lo);
4110 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4111 /* i_blocks represent file system block size */
4112 return i_blocks << (inode->i_blkbits - 9);
4117 return le32_to_cpu(raw_inode->i_blocks_lo);
4121 static inline void ext4_iget_extra_inode(struct inode *inode,
4122 struct ext4_inode *raw_inode,
4123 struct ext4_inode_info *ei)
4125 __le32 *magic = (void *)raw_inode +
4126 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4127 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4128 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4129 ext4_find_inline_data_nolock(inode);
4131 EXT4_I(inode)->i_inline_off = 0;
4134 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4136 struct ext4_iloc iloc;
4137 struct ext4_inode *raw_inode;
4138 struct ext4_inode_info *ei;
4139 struct inode *inode;
4140 journal_t *journal = EXT4_SB(sb)->s_journal;
4146 inode = iget_locked(sb, ino);
4148 return ERR_PTR(-ENOMEM);
4149 if (!(inode->i_state & I_NEW))
4155 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4158 raw_inode = ext4_raw_inode(&iloc);
4160 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4161 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4162 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4163 EXT4_INODE_SIZE(inode->i_sb)) {
4164 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4165 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4166 EXT4_INODE_SIZE(inode->i_sb));
4171 ei->i_extra_isize = 0;
4173 /* Precompute checksum seed for inode metadata */
4174 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4175 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
4176 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4178 __le32 inum = cpu_to_le32(inode->i_ino);
4179 __le32 gen = raw_inode->i_generation;
4180 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4182 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4186 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4187 EXT4_ERROR_INODE(inode, "checksum invalid");
4192 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4193 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4194 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4195 if (!(test_opt(inode->i_sb, NO_UID32))) {
4196 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4197 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4199 i_uid_write(inode, i_uid);
4200 i_gid_write(inode, i_gid);
4201 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4203 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4204 ei->i_inline_off = 0;
4205 ei->i_dir_start_lookup = 0;
4206 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4207 /* We now have enough fields to check if the inode was active or not.
4208 * This is needed because nfsd might try to access dead inodes
4209 * the test is that same one that e2fsck uses
4210 * NeilBrown 1999oct15
4212 if (inode->i_nlink == 0) {
4213 if ((inode->i_mode == 0 ||
4214 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4215 ino != EXT4_BOOT_LOADER_INO) {
4216 /* this inode is deleted */
4220 /* The only unlinked inodes we let through here have
4221 * valid i_mode and are being read by the orphan
4222 * recovery code: that's fine, we're about to complete
4223 * the process of deleting those.
4224 * OR it is the EXT4_BOOT_LOADER_INO which is
4225 * not initialized on a new filesystem. */
4227 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4228 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4229 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4230 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4232 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4233 inode->i_size = ext4_isize(raw_inode);
4234 ei->i_disksize = inode->i_size;
4236 ei->i_reserved_quota = 0;
4238 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4239 ei->i_block_group = iloc.block_group;
4240 ei->i_last_alloc_group = ~0;
4242 * NOTE! The in-memory inode i_data array is in little-endian order
4243 * even on big-endian machines: we do NOT byteswap the block numbers!
4245 for (block = 0; block < EXT4_N_BLOCKS; block++)
4246 ei->i_data[block] = raw_inode->i_block[block];
4247 INIT_LIST_HEAD(&ei->i_orphan);
4250 * Set transaction id's of transactions that have to be committed
4251 * to finish f[data]sync. We set them to currently running transaction
4252 * as we cannot be sure that the inode or some of its metadata isn't
4253 * part of the transaction - the inode could have been reclaimed and
4254 * now it is reread from disk.
4257 transaction_t *transaction;
4260 read_lock(&journal->j_state_lock);
4261 if (journal->j_running_transaction)
4262 transaction = journal->j_running_transaction;
4264 transaction = journal->j_committing_transaction;
4266 tid = transaction->t_tid;
4268 tid = journal->j_commit_sequence;
4269 read_unlock(&journal->j_state_lock);
4270 ei->i_sync_tid = tid;
4271 ei->i_datasync_tid = tid;
4274 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4275 if (ei->i_extra_isize == 0) {
4276 /* The extra space is currently unused. Use it. */
4277 ei->i_extra_isize = sizeof(struct ext4_inode) -
4278 EXT4_GOOD_OLD_INODE_SIZE;
4280 ext4_iget_extra_inode(inode, raw_inode, ei);
4284 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4285 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4286 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4287 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4289 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4290 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4291 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4293 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4297 if (ei->i_file_acl &&
4298 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4299 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4303 } else if (!ext4_has_inline_data(inode)) {
4304 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4305 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4306 (S_ISLNK(inode->i_mode) &&
4307 !ext4_inode_is_fast_symlink(inode))))
4308 /* Validate extent which is part of inode */
4309 ret = ext4_ext_check_inode(inode);
4310 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4311 (S_ISLNK(inode->i_mode) &&
4312 !ext4_inode_is_fast_symlink(inode))) {
4313 /* Validate block references which are part of inode */
4314 ret = ext4_ind_check_inode(inode);
4320 if (S_ISREG(inode->i_mode)) {
4321 inode->i_op = &ext4_file_inode_operations;
4322 inode->i_fop = &ext4_file_operations;
4323 ext4_set_aops(inode);
4324 } else if (S_ISDIR(inode->i_mode)) {
4325 inode->i_op = &ext4_dir_inode_operations;
4326 inode->i_fop = &ext4_dir_operations;
4327 } else if (S_ISLNK(inode->i_mode)) {
4328 if (ext4_inode_is_fast_symlink(inode)) {
4329 inode->i_op = &ext4_fast_symlink_inode_operations;
4330 nd_terminate_link(ei->i_data, inode->i_size,
4331 sizeof(ei->i_data) - 1);
4333 inode->i_op = &ext4_symlink_inode_operations;
4334 ext4_set_aops(inode);
4336 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4337 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4338 inode->i_op = &ext4_special_inode_operations;
4339 if (raw_inode->i_block[0])
4340 init_special_inode(inode, inode->i_mode,
4341 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4343 init_special_inode(inode, inode->i_mode,
4344 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4345 } else if (ino == EXT4_BOOT_LOADER_INO) {
4346 make_bad_inode(inode);
4349 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4353 ext4_set_inode_flags(inode);
4354 unlock_new_inode(inode);
4360 return ERR_PTR(ret);
4363 static int ext4_inode_blocks_set(handle_t *handle,
4364 struct ext4_inode *raw_inode,
4365 struct ext4_inode_info *ei)
4367 struct inode *inode = &(ei->vfs_inode);
4368 u64 i_blocks = inode->i_blocks;
4369 struct super_block *sb = inode->i_sb;
4371 if (i_blocks <= ~0U) {
4373 * i_blocks can be represented in a 32 bit variable
4374 * as multiple of 512 bytes
4376 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4377 raw_inode->i_blocks_high = 0;
4378 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4381 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4384 if (i_blocks <= 0xffffffffffffULL) {
4386 * i_blocks can be represented in a 48 bit variable
4387 * as multiple of 512 bytes
4389 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4390 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4391 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4393 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4394 /* i_block is stored in file system block size */
4395 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4396 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4397 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4403 * Post the struct inode info into an on-disk inode location in the
4404 * buffer-cache. This gobbles the caller's reference to the
4405 * buffer_head in the inode location struct.
4407 * The caller must have write access to iloc->bh.
4409 static int ext4_do_update_inode(handle_t *handle,
4410 struct inode *inode,
4411 struct ext4_iloc *iloc)
4413 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4414 struct ext4_inode_info *ei = EXT4_I(inode);
4415 struct buffer_head *bh = iloc->bh;
4416 int err = 0, rc, block;
4417 int need_datasync = 0;
4421 /* For fields not not tracking in the in-memory inode,
4422 * initialise them to zero for new inodes. */
4423 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4424 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4426 ext4_get_inode_flags(ei);
4427 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4428 i_uid = i_uid_read(inode);
4429 i_gid = i_gid_read(inode);
4430 if (!(test_opt(inode->i_sb, NO_UID32))) {
4431 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4432 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4434 * Fix up interoperability with old kernels. Otherwise, old inodes get
4435 * re-used with the upper 16 bits of the uid/gid intact
4438 raw_inode->i_uid_high =
4439 cpu_to_le16(high_16_bits(i_uid));
4440 raw_inode->i_gid_high =
4441 cpu_to_le16(high_16_bits(i_gid));
4443 raw_inode->i_uid_high = 0;
4444 raw_inode->i_gid_high = 0;
4447 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4448 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4449 raw_inode->i_uid_high = 0;
4450 raw_inode->i_gid_high = 0;
4452 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4454 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4455 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4456 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4457 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4459 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4461 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4462 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4463 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4464 cpu_to_le32(EXT4_OS_HURD))
4465 raw_inode->i_file_acl_high =
4466 cpu_to_le16(ei->i_file_acl >> 32);
4467 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4468 if (ei->i_disksize != ext4_isize(raw_inode)) {
4469 ext4_isize_set(raw_inode, ei->i_disksize);
4472 if (ei->i_disksize > 0x7fffffffULL) {
4473 struct super_block *sb = inode->i_sb;
4474 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4475 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4476 EXT4_SB(sb)->s_es->s_rev_level ==
4477 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4478 /* If this is the first large file
4479 * created, add a flag to the superblock.
4481 err = ext4_journal_get_write_access(handle,
4482 EXT4_SB(sb)->s_sbh);
4485 ext4_update_dynamic_rev(sb);
4486 EXT4_SET_RO_COMPAT_FEATURE(sb,
4487 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4488 ext4_handle_sync(handle);
4489 err = ext4_handle_dirty_super(handle, sb);
4492 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4493 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4494 if (old_valid_dev(inode->i_rdev)) {
4495 raw_inode->i_block[0] =
4496 cpu_to_le32(old_encode_dev(inode->i_rdev));
4497 raw_inode->i_block[1] = 0;
4499 raw_inode->i_block[0] = 0;
4500 raw_inode->i_block[1] =
4501 cpu_to_le32(new_encode_dev(inode->i_rdev));
4502 raw_inode->i_block[2] = 0;
4504 } else if (!ext4_has_inline_data(inode)) {
4505 for (block = 0; block < EXT4_N_BLOCKS; block++)
4506 raw_inode->i_block[block] = ei->i_data[block];
4509 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4510 if (ei->i_extra_isize) {
4511 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4512 raw_inode->i_version_hi =
4513 cpu_to_le32(inode->i_version >> 32);
4514 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4517 ext4_inode_csum_set(inode, raw_inode, ei);
4519 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4520 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4523 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4525 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4528 ext4_std_error(inode->i_sb, err);
4533 * ext4_write_inode()
4535 * We are called from a few places:
4537 * - Within generic_file_write() for O_SYNC files.
4538 * Here, there will be no transaction running. We wait for any running
4539 * transaction to commit.
4541 * - Within sys_sync(), kupdate and such.
4542 * We wait on commit, if tol to.
4544 * - Within prune_icache() (PF_MEMALLOC == true)
4545 * Here we simply return. We can't afford to block kswapd on the
4548 * In all cases it is actually safe for us to return without doing anything,
4549 * because the inode has been copied into a raw inode buffer in
4550 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4553 * Note that we are absolutely dependent upon all inode dirtiers doing the
4554 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4555 * which we are interested.
4557 * It would be a bug for them to not do this. The code:
4559 * mark_inode_dirty(inode)
4561 * inode->i_size = expr;
4563 * is in error because a kswapd-driven write_inode() could occur while
4564 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4565 * will no longer be on the superblock's dirty inode list.
4567 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4571 if (current->flags & PF_MEMALLOC)
4574 if (EXT4_SB(inode->i_sb)->s_journal) {
4575 if (ext4_journal_current_handle()) {
4576 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4581 if (wbc->sync_mode != WB_SYNC_ALL)
4584 err = ext4_force_commit(inode->i_sb);
4586 struct ext4_iloc iloc;
4588 err = __ext4_get_inode_loc(inode, &iloc, 0);
4591 if (wbc->sync_mode == WB_SYNC_ALL)
4592 sync_dirty_buffer(iloc.bh);
4593 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4594 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4595 "IO error syncing inode");
4604 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4605 * buffers that are attached to a page stradding i_size and are undergoing
4606 * commit. In that case we have to wait for commit to finish and try again.
4608 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4612 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4613 tid_t commit_tid = 0;
4616 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4618 * All buffers in the last page remain valid? Then there's nothing to
4619 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4622 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4625 page = find_lock_page(inode->i_mapping,
4626 inode->i_size >> PAGE_CACHE_SHIFT);
4629 ret = __ext4_journalled_invalidatepage(page, offset);
4631 page_cache_release(page);
4635 read_lock(&journal->j_state_lock);
4636 if (journal->j_committing_transaction)
4637 commit_tid = journal->j_committing_transaction->t_tid;
4638 read_unlock(&journal->j_state_lock);
4640 jbd2_log_wait_commit(journal, commit_tid);
4647 * Called from notify_change.
4649 * We want to trap VFS attempts to truncate the file as soon as
4650 * possible. In particular, we want to make sure that when the VFS
4651 * shrinks i_size, we put the inode on the orphan list and modify
4652 * i_disksize immediately, so that during the subsequent flushing of
4653 * dirty pages and freeing of disk blocks, we can guarantee that any
4654 * commit will leave the blocks being flushed in an unused state on
4655 * disk. (On recovery, the inode will get truncated and the blocks will
4656 * be freed, so we have a strong guarantee that no future commit will
4657 * leave these blocks visible to the user.)
4659 * Another thing we have to assure is that if we are in ordered mode
4660 * and inode is still attached to the committing transaction, we must
4661 * we start writeout of all the dirty pages which are being truncated.
4662 * This way we are sure that all the data written in the previous
4663 * transaction are already on disk (truncate waits for pages under
4666 * Called with inode->i_mutex down.
4668 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4670 struct inode *inode = dentry->d_inode;
4673 const unsigned int ia_valid = attr->ia_valid;
4675 error = inode_change_ok(inode, attr);
4679 if (is_quota_modification(inode, attr))
4680 dquot_initialize(inode);
4681 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4682 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4685 /* (user+group)*(old+new) structure, inode write (sb,
4686 * inode block, ? - but truncate inode update has it) */
4687 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4688 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4689 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4690 if (IS_ERR(handle)) {
4691 error = PTR_ERR(handle);
4694 error = dquot_transfer(inode, attr);
4696 ext4_journal_stop(handle);
4699 /* Update corresponding info in inode so that everything is in
4700 * one transaction */
4701 if (attr->ia_valid & ATTR_UID)
4702 inode->i_uid = attr->ia_uid;
4703 if (attr->ia_valid & ATTR_GID)
4704 inode->i_gid = attr->ia_gid;
4705 error = ext4_mark_inode_dirty(handle, inode);
4706 ext4_journal_stop(handle);
4709 if (attr->ia_valid & ATTR_SIZE) {
4711 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4712 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4714 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4719 if (S_ISREG(inode->i_mode) &&
4720 attr->ia_valid & ATTR_SIZE &&
4721 (attr->ia_size < inode->i_size)) {
4724 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4725 if (IS_ERR(handle)) {
4726 error = PTR_ERR(handle);
4729 if (ext4_handle_valid(handle)) {
4730 error = ext4_orphan_add(handle, inode);
4733 EXT4_I(inode)->i_disksize = attr->ia_size;
4734 rc = ext4_mark_inode_dirty(handle, inode);
4737 ext4_journal_stop(handle);
4739 if (ext4_should_order_data(inode)) {
4740 error = ext4_begin_ordered_truncate(inode,
4743 /* Do as much error cleanup as possible */
4744 handle = ext4_journal_start(inode,
4746 if (IS_ERR(handle)) {
4747 ext4_orphan_del(NULL, inode);
4750 ext4_orphan_del(handle, inode);
4752 ext4_journal_stop(handle);
4758 if (attr->ia_valid & ATTR_SIZE) {
4759 if (attr->ia_size != inode->i_size) {
4760 loff_t oldsize = inode->i_size;
4762 i_size_write(inode, attr->ia_size);
4764 * Blocks are going to be removed from the inode. Wait
4765 * for dio in flight. Temporarily disable
4766 * dioread_nolock to prevent livelock.
4769 if (!ext4_should_journal_data(inode)) {
4770 ext4_inode_block_unlocked_dio(inode);
4771 inode_dio_wait(inode);
4772 ext4_inode_resume_unlocked_dio(inode);
4774 ext4_wait_for_tail_page_commit(inode);
4777 * Truncate pagecache after we've waited for commit
4778 * in data=journal mode to make pages freeable.
4780 truncate_pagecache(inode, oldsize, inode->i_size);
4782 ext4_truncate(inode);
4786 setattr_copy(inode, attr);
4787 mark_inode_dirty(inode);
4791 * If the call to ext4_truncate failed to get a transaction handle at
4792 * all, we need to clean up the in-core orphan list manually.
4794 if (orphan && inode->i_nlink)
4795 ext4_orphan_del(NULL, inode);
4797 if (!rc && (ia_valid & ATTR_MODE))
4798 rc = ext4_acl_chmod(inode);
4801 ext4_std_error(inode->i_sb, error);
4807 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4810 struct inode *inode;
4811 unsigned long delalloc_blocks;
4813 inode = dentry->d_inode;
4814 generic_fillattr(inode, stat);
4817 * We can't update i_blocks if the block allocation is delayed
4818 * otherwise in the case of system crash before the real block
4819 * allocation is done, we will have i_blocks inconsistent with
4820 * on-disk file blocks.
4821 * We always keep i_blocks updated together with real
4822 * allocation. But to not confuse with user, stat
4823 * will return the blocks that include the delayed allocation
4824 * blocks for this file.
4826 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4827 EXT4_I(inode)->i_reserved_data_blocks);
4829 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4833 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4835 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4836 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4837 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4841 * Account for index blocks, block groups bitmaps and block group
4842 * descriptor blocks if modify datablocks and index blocks
4843 * worse case, the indexs blocks spread over different block groups
4845 * If datablocks are discontiguous, they are possible to spread over
4846 * different block groups too. If they are contiguous, with flexbg,
4847 * they could still across block group boundary.
4849 * Also account for superblock, inode, quota and xattr blocks
4851 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4853 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4859 * How many index blocks need to touch to modify nrblocks?
4860 * The "Chunk" flag indicating whether the nrblocks is
4861 * physically contiguous on disk
4863 * For Direct IO and fallocate, they calls get_block to allocate
4864 * one single extent at a time, so they could set the "Chunk" flag
4866 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4871 * Now let's see how many group bitmaps and group descriptors need
4881 if (groups > ngroups)
4883 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4884 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4886 /* bitmaps and block group descriptor blocks */
4887 ret += groups + gdpblocks;
4889 /* Blocks for super block, inode, quota and xattr blocks */
4890 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4896 * Calculate the total number of credits to reserve to fit
4897 * the modification of a single pages into a single transaction,
4898 * which may include multiple chunks of block allocations.
4900 * This could be called via ext4_write_begin()
4902 * We need to consider the worse case, when
4903 * one new block per extent.
4905 int ext4_writepage_trans_blocks(struct inode *inode)
4907 int bpp = ext4_journal_blocks_per_page(inode);
4910 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4912 /* Account for data blocks for journalled mode */
4913 if (ext4_should_journal_data(inode))
4919 * Calculate the journal credits for a chunk of data modification.
4921 * This is called from DIO, fallocate or whoever calling
4922 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4924 * journal buffers for data blocks are not included here, as DIO
4925 * and fallocate do no need to journal data buffers.
4927 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4929 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4933 * The caller must have previously called ext4_reserve_inode_write().
4934 * Give this, we know that the caller already has write access to iloc->bh.
4936 int ext4_mark_iloc_dirty(handle_t *handle,
4937 struct inode *inode, struct ext4_iloc *iloc)
4941 if (IS_I_VERSION(inode))
4942 inode_inc_iversion(inode);
4944 /* the do_update_inode consumes one bh->b_count */
4947 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4948 err = ext4_do_update_inode(handle, inode, iloc);
4954 * On success, We end up with an outstanding reference count against
4955 * iloc->bh. This _must_ be cleaned up later.
4959 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4960 struct ext4_iloc *iloc)
4964 err = ext4_get_inode_loc(inode, iloc);
4966 BUFFER_TRACE(iloc->bh, "get_write_access");
4967 err = ext4_journal_get_write_access(handle, iloc->bh);
4973 ext4_std_error(inode->i_sb, err);
4978 * Expand an inode by new_extra_isize bytes.
4979 * Returns 0 on success or negative error number on failure.
4981 static int ext4_expand_extra_isize(struct inode *inode,
4982 unsigned int new_extra_isize,
4983 struct ext4_iloc iloc,
4986 struct ext4_inode *raw_inode;
4987 struct ext4_xattr_ibody_header *header;
4989 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4992 raw_inode = ext4_raw_inode(&iloc);
4994 header = IHDR(inode, raw_inode);
4996 /* No extended attributes present */
4997 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4998 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4999 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5001 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5005 /* try to expand with EAs present */
5006 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5011 * What we do here is to mark the in-core inode as clean with respect to inode
5012 * dirtiness (it may still be data-dirty).
5013 * This means that the in-core inode may be reaped by prune_icache
5014 * without having to perform any I/O. This is a very good thing,
5015 * because *any* task may call prune_icache - even ones which
5016 * have a transaction open against a different journal.
5018 * Is this cheating? Not really. Sure, we haven't written the
5019 * inode out, but prune_icache isn't a user-visible syncing function.
5020 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5021 * we start and wait on commits.
5023 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5025 struct ext4_iloc iloc;
5026 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5027 static unsigned int mnt_count;
5031 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5032 err = ext4_reserve_inode_write(handle, inode, &iloc);
5033 if (ext4_handle_valid(handle) &&
5034 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5035 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5037 * We need extra buffer credits since we may write into EA block
5038 * with this same handle. If journal_extend fails, then it will
5039 * only result in a minor loss of functionality for that inode.
5040 * If this is felt to be critical, then e2fsck should be run to
5041 * force a large enough s_min_extra_isize.
5043 if ((jbd2_journal_extend(handle,
5044 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5045 ret = ext4_expand_extra_isize(inode,
5046 sbi->s_want_extra_isize,
5049 ext4_set_inode_state(inode,
5050 EXT4_STATE_NO_EXPAND);
5052 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5053 ext4_warning(inode->i_sb,
5054 "Unable to expand inode %lu. Delete"
5055 " some EAs or run e2fsck.",
5058 le16_to_cpu(sbi->s_es->s_mnt_count);
5064 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5069 * ext4_dirty_inode() is called from __mark_inode_dirty()
5071 * We're really interested in the case where a file is being extended.
5072 * i_size has been changed by generic_commit_write() and we thus need
5073 * to include the updated inode in the current transaction.
5075 * Also, dquot_alloc_block() will always dirty the inode when blocks
5076 * are allocated to the file.
5078 * If the inode is marked synchronous, we don't honour that here - doing
5079 * so would cause a commit on atime updates, which we don't bother doing.
5080 * We handle synchronous inodes at the highest possible level.
5082 void ext4_dirty_inode(struct inode *inode, int flags)
5086 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5090 ext4_mark_inode_dirty(handle, inode);
5092 ext4_journal_stop(handle);
5099 * Bind an inode's backing buffer_head into this transaction, to prevent
5100 * it from being flushed to disk early. Unlike
5101 * ext4_reserve_inode_write, this leaves behind no bh reference and
5102 * returns no iloc structure, so the caller needs to repeat the iloc
5103 * lookup to mark the inode dirty later.
5105 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5107 struct ext4_iloc iloc;
5111 err = ext4_get_inode_loc(inode, &iloc);
5113 BUFFER_TRACE(iloc.bh, "get_write_access");
5114 err = jbd2_journal_get_write_access(handle, iloc.bh);
5116 err = ext4_handle_dirty_metadata(handle,
5122 ext4_std_error(inode->i_sb, err);
5127 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5134 * We have to be very careful here: changing a data block's
5135 * journaling status dynamically is dangerous. If we write a
5136 * data block to the journal, change the status and then delete
5137 * that block, we risk forgetting to revoke the old log record
5138 * from the journal and so a subsequent replay can corrupt data.
5139 * So, first we make sure that the journal is empty and that
5140 * nobody is changing anything.
5143 journal = EXT4_JOURNAL(inode);
5146 if (is_journal_aborted(journal))
5148 /* We have to allocate physical blocks for delalloc blocks
5149 * before flushing journal. otherwise delalloc blocks can not
5150 * be allocated any more. even more truncate on delalloc blocks
5151 * could trigger BUG by flushing delalloc blocks in journal.
5152 * There is no delalloc block in non-journal data mode.
5154 if (val && test_opt(inode->i_sb, DELALLOC)) {
5155 err = ext4_alloc_da_blocks(inode);
5160 /* Wait for all existing dio workers */
5161 ext4_inode_block_unlocked_dio(inode);
5162 inode_dio_wait(inode);
5164 jbd2_journal_lock_updates(journal);
5167 * OK, there are no updates running now, and all cached data is
5168 * synced to disk. We are now in a completely consistent state
5169 * which doesn't have anything in the journal, and we know that
5170 * no filesystem updates are running, so it is safe to modify
5171 * the inode's in-core data-journaling state flag now.
5175 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5177 jbd2_journal_flush(journal);
5178 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5180 ext4_set_aops(inode);
5182 jbd2_journal_unlock_updates(journal);
5183 ext4_inode_resume_unlocked_dio(inode);
5185 /* Finally we can mark the inode as dirty. */
5187 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5189 return PTR_ERR(handle);
5191 err = ext4_mark_inode_dirty(handle, inode);
5192 ext4_handle_sync(handle);
5193 ext4_journal_stop(handle);
5194 ext4_std_error(inode->i_sb, err);
5199 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5201 return !buffer_mapped(bh);
5204 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5206 struct page *page = vmf->page;
5210 struct file *file = vma->vm_file;
5211 struct inode *inode = file_inode(file);
5212 struct address_space *mapping = inode->i_mapping;
5214 get_block_t *get_block;
5217 sb_start_pagefault(inode->i_sb);
5218 file_update_time(vma->vm_file);
5219 /* Delalloc case is easy... */
5220 if (test_opt(inode->i_sb, DELALLOC) &&
5221 !ext4_should_journal_data(inode) &&
5222 !ext4_nonda_switch(inode->i_sb)) {
5224 ret = __block_page_mkwrite(vma, vmf,
5225 ext4_da_get_block_prep);
5226 } while (ret == -ENOSPC &&
5227 ext4_should_retry_alloc(inode->i_sb, &retries));
5232 size = i_size_read(inode);
5233 /* Page got truncated from under us? */
5234 if (page->mapping != mapping || page_offset(page) > size) {
5236 ret = VM_FAULT_NOPAGE;
5240 if (page->index == size >> PAGE_CACHE_SHIFT)
5241 len = size & ~PAGE_CACHE_MASK;
5243 len = PAGE_CACHE_SIZE;
5245 * Return if we have all the buffers mapped. This avoids the need to do
5246 * journal_start/journal_stop which can block and take a long time
5248 if (page_has_buffers(page)) {
5249 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5251 ext4_bh_unmapped)) {
5252 /* Wait so that we don't change page under IO */
5253 wait_for_stable_page(page);
5254 ret = VM_FAULT_LOCKED;
5259 /* OK, we need to fill the hole... */
5260 if (ext4_should_dioread_nolock(inode))
5261 get_block = ext4_get_block_write;
5263 get_block = ext4_get_block;
5265 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5266 ext4_writepage_trans_blocks(inode));
5267 if (IS_ERR(handle)) {
5268 ret = VM_FAULT_SIGBUS;
5271 ret = __block_page_mkwrite(vma, vmf, get_block);
5272 if (!ret && ext4_should_journal_data(inode)) {
5273 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5274 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5276 ret = VM_FAULT_SIGBUS;
5277 ext4_journal_stop(handle);
5280 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5282 ext4_journal_stop(handle);
5283 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5286 ret = block_page_mkwrite_return(ret);
5288 sb_end_pagefault(inode->i_sb);