2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51 struct ext4_inode_info *ei)
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 csum_lo = raw->i_checksum_lo;
59 raw->i_checksum_lo = 0;
60 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
61 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
62 csum_hi = raw->i_checksum_hi;
63 raw->i_checksum_hi = 0;
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
67 EXT4_INODE_SIZE(inode->i_sb));
69 raw->i_checksum_lo = csum_lo;
70 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
71 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
72 raw->i_checksum_hi = csum_hi;
77 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
78 struct ext4_inode_info *ei)
80 __u32 provided, calculated;
82 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
83 cpu_to_le32(EXT4_OS_LINUX) ||
84 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
85 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
88 provided = le16_to_cpu(raw->i_checksum_lo);
89 calculated = ext4_inode_csum(inode, raw, ei);
90 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
91 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
92 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
96 return provided == calculated;
99 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
100 struct ext4_inode_info *ei)
104 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
105 cpu_to_le32(EXT4_OS_LINUX) ||
106 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
107 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
110 csum = ext4_inode_csum(inode, raw, ei);
111 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
112 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
113 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
114 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
117 static inline int ext4_begin_ordered_truncate(struct inode *inode,
120 trace_ext4_begin_ordered_truncate(inode, new_size);
122 * If jinode is zero, then we never opened the file for
123 * writing, so there's no need to call
124 * jbd2_journal_begin_ordered_truncate() since there's no
125 * outstanding writes we need to flush.
127 if (!EXT4_I(inode)->jinode)
129 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
130 EXT4_I(inode)->jinode,
134 static void ext4_invalidatepage(struct page *page, unsigned long offset);
135 static int __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 ext4_ioend_wait(inode);
190 if (inode->i_nlink) {
192 * When journalling data dirty buffers are tracked only in the
193 * journal. So although mm thinks everything is clean and
194 * ready for reaping the inode might still have some pages to
195 * write in the running transaction or waiting to be
196 * checkpointed. Thus calling jbd2_journal_invalidatepage()
197 * (via truncate_inode_pages()) to discard these buffers can
198 * cause data loss. Also even if we did not discard these
199 * buffers, we would have no way to find them after the inode
200 * is reaped and thus user could see stale data if he tries to
201 * read them before the transaction is checkpointed. So be
202 * careful and force everything to disk here... We use
203 * ei->i_datasync_tid to store the newest transaction
204 * containing inode's data.
206 * Note that directories do not have this problem because they
207 * don't use page cache.
209 if (ext4_should_journal_data(inode) &&
210 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
211 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
212 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
214 jbd2_log_start_commit(journal, commit_tid);
215 jbd2_log_wait_commit(journal, commit_tid);
216 filemap_write_and_wait(&inode->i_data);
218 truncate_inode_pages(&inode->i_data, 0);
222 if (!is_bad_inode(inode))
223 dquot_initialize(inode);
225 if (ext4_should_order_data(inode))
226 ext4_begin_ordered_truncate(inode, 0);
227 truncate_inode_pages(&inode->i_data, 0);
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__);
725 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
726 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
727 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
728 ext4_find_delalloc_range(inode, map->m_lblk,
729 map->m_lblk + map->m_len - 1))
730 status |= EXTENT_STATUS_DELAYED;
731 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
732 map->m_pblk, status);
737 up_write((&EXT4_I(inode)->i_data_sem));
738 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
739 int ret = check_block_validity(inode, map);
746 /* Maximum number of blocks we map for direct IO at once. */
747 #define DIO_MAX_BLOCKS 4096
749 static int _ext4_get_block(struct inode *inode, sector_t iblock,
750 struct buffer_head *bh, int flags)
752 handle_t *handle = ext4_journal_current_handle();
753 struct ext4_map_blocks map;
754 int ret = 0, started = 0;
757 if (ext4_has_inline_data(inode))
761 map.m_len = bh->b_size >> inode->i_blkbits;
763 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
764 /* Direct IO write... */
765 if (map.m_len > DIO_MAX_BLOCKS)
766 map.m_len = DIO_MAX_BLOCKS;
767 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
768 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
770 if (IS_ERR(handle)) {
771 ret = PTR_ERR(handle);
777 ret = ext4_map_blocks(handle, inode, &map, flags);
779 map_bh(bh, inode->i_sb, map.m_pblk);
780 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
781 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
785 ext4_journal_stop(handle);
789 int ext4_get_block(struct inode *inode, sector_t iblock,
790 struct buffer_head *bh, int create)
792 return _ext4_get_block(inode, iblock, bh,
793 create ? EXT4_GET_BLOCKS_CREATE : 0);
797 * `handle' can be NULL if create is zero
799 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
800 ext4_lblk_t block, int create, int *errp)
802 struct ext4_map_blocks map;
803 struct buffer_head *bh;
806 J_ASSERT(handle != NULL || create == 0);
810 err = ext4_map_blocks(handle, inode, &map,
811 create ? EXT4_GET_BLOCKS_CREATE : 0);
813 /* ensure we send some value back into *errp */
816 if (create && err == 0)
817 err = -ENOSPC; /* should never happen */
823 bh = sb_getblk(inode->i_sb, map.m_pblk);
828 if (map.m_flags & EXT4_MAP_NEW) {
829 J_ASSERT(create != 0);
830 J_ASSERT(handle != NULL);
833 * Now that we do not always journal data, we should
834 * keep in mind whether this should always journal the
835 * new buffer as metadata. For now, regular file
836 * writes use ext4_get_block instead, so it's not a
840 BUFFER_TRACE(bh, "call get_create_access");
841 fatal = ext4_journal_get_create_access(handle, bh);
842 if (!fatal && !buffer_uptodate(bh)) {
843 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
844 set_buffer_uptodate(bh);
847 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
848 err = ext4_handle_dirty_metadata(handle, inode, bh);
852 BUFFER_TRACE(bh, "not a new buffer");
862 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
863 ext4_lblk_t block, int create, int *err)
865 struct buffer_head *bh;
867 bh = ext4_getblk(handle, inode, block, create, err);
870 if (buffer_uptodate(bh))
872 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
874 if (buffer_uptodate(bh))
881 int ext4_walk_page_buffers(handle_t *handle,
882 struct buffer_head *head,
886 int (*fn)(handle_t *handle,
887 struct buffer_head *bh))
889 struct buffer_head *bh;
890 unsigned block_start, block_end;
891 unsigned blocksize = head->b_size;
893 struct buffer_head *next;
895 for (bh = head, block_start = 0;
896 ret == 0 && (bh != head || !block_start);
897 block_start = block_end, bh = next) {
898 next = bh->b_this_page;
899 block_end = block_start + blocksize;
900 if (block_end <= from || block_start >= to) {
901 if (partial && !buffer_uptodate(bh))
905 err = (*fn)(handle, bh);
913 * To preserve ordering, it is essential that the hole instantiation and
914 * the data write be encapsulated in a single transaction. We cannot
915 * close off a transaction and start a new one between the ext4_get_block()
916 * and the commit_write(). So doing the jbd2_journal_start at the start of
917 * prepare_write() is the right place.
919 * Also, this function can nest inside ext4_writepage(). In that case, we
920 * *know* that ext4_writepage() has generated enough buffer credits to do the
921 * whole page. So we won't block on the journal in that case, which is good,
922 * because the caller may be PF_MEMALLOC.
924 * By accident, ext4 can be reentered when a transaction is open via
925 * quota file writes. If we were to commit the transaction while thus
926 * reentered, there can be a deadlock - we would be holding a quota
927 * lock, and the commit would never complete if another thread had a
928 * transaction open and was blocking on the quota lock - a ranking
931 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
932 * will _not_ run commit under these circumstances because handle->h_ref
933 * is elevated. We'll still have enough credits for the tiny quotafile
936 int do_journal_get_write_access(handle_t *handle,
937 struct buffer_head *bh)
939 int dirty = buffer_dirty(bh);
942 if (!buffer_mapped(bh) || buffer_freed(bh))
945 * __block_write_begin() could have dirtied some buffers. Clean
946 * the dirty bit as jbd2_journal_get_write_access() could complain
947 * otherwise about fs integrity issues. Setting of the dirty bit
948 * by __block_write_begin() isn't a real problem here as we clear
949 * the bit before releasing a page lock and thus writeback cannot
950 * ever write the buffer.
953 clear_buffer_dirty(bh);
954 ret = ext4_journal_get_write_access(handle, bh);
956 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
960 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
961 struct buffer_head *bh_result, int create);
962 static int ext4_write_begin(struct file *file, struct address_space *mapping,
963 loff_t pos, unsigned len, unsigned flags,
964 struct page **pagep, void **fsdata)
966 struct inode *inode = mapping->host;
967 int ret, needed_blocks;
974 trace_ext4_write_begin(inode, pos, len, flags);
976 * Reserve one block more for addition to orphan list in case
977 * we allocate blocks but write fails for some reason
979 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
980 index = pos >> PAGE_CACHE_SHIFT;
981 from = pos & (PAGE_CACHE_SIZE - 1);
984 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
985 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
994 * grab_cache_page_write_begin() can take a long time if the
995 * system is thrashing due to memory pressure, or if the page
996 * is being written back. So grab it first before we start
997 * the transaction handle. This also allows us to allocate
998 * the page (if needed) without using GFP_NOFS.
1001 page = grab_cache_page_write_begin(mapping, index, flags);
1007 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1008 if (IS_ERR(handle)) {
1009 page_cache_release(page);
1010 return PTR_ERR(handle);
1014 if (page->mapping != mapping) {
1015 /* The page got truncated from under us */
1017 page_cache_release(page);
1018 ext4_journal_stop(handle);
1021 wait_on_page_writeback(page);
1023 if (ext4_should_dioread_nolock(inode))
1024 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1026 ret = __block_write_begin(page, pos, len, ext4_get_block);
1028 if (!ret && ext4_should_journal_data(inode)) {
1029 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1031 do_journal_get_write_access);
1037 * __block_write_begin may have instantiated a few blocks
1038 * outside i_size. Trim these off again. Don't need
1039 * i_size_read because we hold i_mutex.
1041 * Add inode to orphan list in case we crash before
1044 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1045 ext4_orphan_add(handle, inode);
1047 ext4_journal_stop(handle);
1048 if (pos + len > inode->i_size) {
1049 ext4_truncate_failed_write(inode);
1051 * If truncate failed early the inode might
1052 * still be on the orphan list; we need to
1053 * make sure the inode is removed from the
1054 * orphan list in that case.
1057 ext4_orphan_del(NULL, inode);
1060 if (ret == -ENOSPC &&
1061 ext4_should_retry_alloc(inode->i_sb, &retries))
1063 page_cache_release(page);
1070 /* For write_end() in data=journal mode */
1071 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1073 if (!buffer_mapped(bh) || buffer_freed(bh))
1075 set_buffer_uptodate(bh);
1076 return ext4_handle_dirty_metadata(handle, NULL, bh);
1079 static int ext4_generic_write_end(struct file *file,
1080 struct address_space *mapping,
1081 loff_t pos, unsigned len, unsigned copied,
1082 struct page *page, void *fsdata)
1084 int i_size_changed = 0;
1085 struct inode *inode = mapping->host;
1086 handle_t *handle = ext4_journal_current_handle();
1088 if (ext4_has_inline_data(inode))
1089 copied = ext4_write_inline_data_end(inode, pos, len,
1092 copied = block_write_end(file, mapping, pos,
1093 len, copied, page, fsdata);
1096 * No need to use i_size_read() here, the i_size
1097 * cannot change under us because we hold i_mutex.
1099 * But it's important to update i_size while still holding page lock:
1100 * page writeout could otherwise come in and zero beyond i_size.
1102 if (pos + copied > inode->i_size) {
1103 i_size_write(inode, pos + copied);
1107 if (pos + copied > EXT4_I(inode)->i_disksize) {
1108 /* We need to mark inode dirty even if
1109 * new_i_size is less that inode->i_size
1110 * bu greater than i_disksize.(hint delalloc)
1112 ext4_update_i_disksize(inode, (pos + copied));
1116 page_cache_release(page);
1119 * Don't mark the inode dirty under page lock. First, it unnecessarily
1120 * makes the holding time of page lock longer. Second, it forces lock
1121 * ordering of page lock and transaction start for journaling
1125 ext4_mark_inode_dirty(handle, inode);
1131 * We need to pick up the new inode size which generic_commit_write gave us
1132 * `file' can be NULL - eg, when called from page_symlink().
1134 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1135 * buffers are managed internally.
1137 static int ext4_ordered_write_end(struct file *file,
1138 struct address_space *mapping,
1139 loff_t pos, unsigned len, unsigned copied,
1140 struct page *page, void *fsdata)
1142 handle_t *handle = ext4_journal_current_handle();
1143 struct inode *inode = mapping->host;
1146 trace_ext4_ordered_write_end(inode, pos, len, copied);
1147 ret = ext4_jbd2_file_inode(handle, inode);
1150 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1153 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1154 /* if we have allocated more blocks and copied
1155 * less. We will have blocks allocated outside
1156 * inode->i_size. So truncate them
1158 ext4_orphan_add(handle, inode);
1163 page_cache_release(page);
1166 ret2 = ext4_journal_stop(handle);
1170 if (pos + len > inode->i_size) {
1171 ext4_truncate_failed_write(inode);
1173 * If truncate failed early the inode might still be
1174 * on the orphan list; we need to make sure the inode
1175 * is removed from the orphan list in that case.
1178 ext4_orphan_del(NULL, inode);
1182 return ret ? ret : copied;
1185 static int ext4_writeback_write_end(struct file *file,
1186 struct address_space *mapping,
1187 loff_t pos, unsigned len, unsigned copied,
1188 struct page *page, void *fsdata)
1190 handle_t *handle = ext4_journal_current_handle();
1191 struct inode *inode = mapping->host;
1194 trace_ext4_writeback_write_end(inode, pos, len, copied);
1195 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1198 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1199 /* if we have allocated more blocks and copied
1200 * less. We will have blocks allocated outside
1201 * inode->i_size. So truncate them
1203 ext4_orphan_add(handle, inode);
1208 ret2 = ext4_journal_stop(handle);
1212 if (pos + len > inode->i_size) {
1213 ext4_truncate_failed_write(inode);
1215 * If truncate failed early the inode might still be
1216 * on the orphan list; we need to make sure the inode
1217 * is removed from the orphan list in that case.
1220 ext4_orphan_del(NULL, inode);
1223 return ret ? ret : copied;
1226 static int ext4_journalled_write_end(struct file *file,
1227 struct address_space *mapping,
1228 loff_t pos, unsigned len, unsigned copied,
1229 struct page *page, void *fsdata)
1231 handle_t *handle = ext4_journal_current_handle();
1232 struct inode *inode = mapping->host;
1238 trace_ext4_journalled_write_end(inode, pos, len, copied);
1239 from = pos & (PAGE_CACHE_SIZE - 1);
1242 BUG_ON(!ext4_handle_valid(handle));
1244 if (ext4_has_inline_data(inode))
1245 copied = ext4_write_inline_data_end(inode, pos, len,
1249 if (!PageUptodate(page))
1251 page_zero_new_buffers(page, from+copied, to);
1254 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1255 to, &partial, write_end_fn);
1257 SetPageUptodate(page);
1259 new_i_size = pos + copied;
1260 if (new_i_size > inode->i_size)
1261 i_size_write(inode, pos+copied);
1262 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1263 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1264 if (new_i_size > EXT4_I(inode)->i_disksize) {
1265 ext4_update_i_disksize(inode, new_i_size);
1266 ret2 = ext4_mark_inode_dirty(handle, inode);
1272 page_cache_release(page);
1273 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1274 /* if we have allocated more blocks and copied
1275 * less. We will have blocks allocated outside
1276 * inode->i_size. So truncate them
1278 ext4_orphan_add(handle, inode);
1280 ret2 = ext4_journal_stop(handle);
1283 if (pos + len > inode->i_size) {
1284 ext4_truncate_failed_write(inode);
1286 * If truncate failed early the inode might still be
1287 * on the orphan list; we need to make sure the inode
1288 * is removed from the orphan list in that case.
1291 ext4_orphan_del(NULL, inode);
1294 return ret ? ret : copied;
1298 * Reserve a single cluster located at lblock
1300 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1303 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1304 struct ext4_inode_info *ei = EXT4_I(inode);
1305 unsigned int md_needed;
1307 ext4_lblk_t save_last_lblock;
1311 * We will charge metadata quota at writeout time; this saves
1312 * us from metadata over-estimation, though we may go over by
1313 * a small amount in the end. Here we just reserve for data.
1315 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1320 * recalculate the amount of metadata blocks to reserve
1321 * in order to allocate nrblocks
1322 * worse case is one extent per block
1325 spin_lock(&ei->i_block_reservation_lock);
1327 * ext4_calc_metadata_amount() has side effects, which we have
1328 * to be prepared undo if we fail to claim space.
1330 save_len = ei->i_da_metadata_calc_len;
1331 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1332 md_needed = EXT4_NUM_B2C(sbi,
1333 ext4_calc_metadata_amount(inode, lblock));
1334 trace_ext4_da_reserve_space(inode, md_needed);
1337 * We do still charge estimated metadata to the sb though;
1338 * we cannot afford to run out of free blocks.
1340 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1341 ei->i_da_metadata_calc_len = save_len;
1342 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1343 spin_unlock(&ei->i_block_reservation_lock);
1344 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1348 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1351 ei->i_reserved_data_blocks++;
1352 ei->i_reserved_meta_blocks += md_needed;
1353 spin_unlock(&ei->i_block_reservation_lock);
1355 return 0; /* success */
1358 static void ext4_da_release_space(struct inode *inode, int to_free)
1360 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1361 struct ext4_inode_info *ei = EXT4_I(inode);
1364 return; /* Nothing to release, exit */
1366 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1368 trace_ext4_da_release_space(inode, to_free);
1369 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1371 * if there aren't enough reserved blocks, then the
1372 * counter is messed up somewhere. Since this
1373 * function is called from invalidate page, it's
1374 * harmless to return without any action.
1376 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1377 "ino %lu, to_free %d with only %d reserved "
1378 "data blocks", inode->i_ino, to_free,
1379 ei->i_reserved_data_blocks);
1381 to_free = ei->i_reserved_data_blocks;
1383 ei->i_reserved_data_blocks -= to_free;
1385 if (ei->i_reserved_data_blocks == 0) {
1387 * We can release all of the reserved metadata blocks
1388 * only when we have written all of the delayed
1389 * allocation blocks.
1390 * Note that in case of bigalloc, i_reserved_meta_blocks,
1391 * i_reserved_data_blocks, etc. refer to number of clusters.
1393 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1394 ei->i_reserved_meta_blocks);
1395 ei->i_reserved_meta_blocks = 0;
1396 ei->i_da_metadata_calc_len = 0;
1399 /* update fs dirty data blocks counter */
1400 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1402 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1404 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1407 static void ext4_da_page_release_reservation(struct page *page,
1408 unsigned long offset)
1411 struct buffer_head *head, *bh;
1412 unsigned int curr_off = 0;
1413 struct inode *inode = page->mapping->host;
1414 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1418 head = page_buffers(page);
1421 unsigned int next_off = curr_off + bh->b_size;
1423 if ((offset <= curr_off) && (buffer_delay(bh))) {
1425 clear_buffer_delay(bh);
1427 curr_off = next_off;
1428 } while ((bh = bh->b_this_page) != head);
1431 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1432 ext4_es_remove_extent(inode, lblk, to_release);
1435 /* If we have released all the blocks belonging to a cluster, then we
1436 * need to release the reserved space for that cluster. */
1437 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1438 while (num_clusters > 0) {
1439 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1440 ((num_clusters - 1) << sbi->s_cluster_bits);
1441 if (sbi->s_cluster_ratio == 1 ||
1442 !ext4_find_delalloc_cluster(inode, lblk))
1443 ext4_da_release_space(inode, 1);
1450 * Delayed allocation stuff
1454 * mpage_da_submit_io - walks through extent of pages and try to write
1455 * them with writepage() call back
1457 * @mpd->inode: inode
1458 * @mpd->first_page: first page of the extent
1459 * @mpd->next_page: page after the last page of the extent
1461 * By the time mpage_da_submit_io() is called we expect all blocks
1462 * to be allocated. this may be wrong if allocation failed.
1464 * As pages are already locked by write_cache_pages(), we can't use it
1466 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1467 struct ext4_map_blocks *map)
1469 struct pagevec pvec;
1470 unsigned long index, end;
1471 int ret = 0, err, nr_pages, i;
1472 struct inode *inode = mpd->inode;
1473 struct address_space *mapping = inode->i_mapping;
1474 loff_t size = i_size_read(inode);
1475 unsigned int len, block_start;
1476 struct buffer_head *bh, *page_bufs = NULL;
1477 sector_t pblock = 0, cur_logical = 0;
1478 struct ext4_io_submit io_submit;
1480 BUG_ON(mpd->next_page <= mpd->first_page);
1481 memset(&io_submit, 0, sizeof(io_submit));
1483 * We need to start from the first_page to the next_page - 1
1484 * to make sure we also write the mapped dirty buffer_heads.
1485 * If we look at mpd->b_blocknr we would only be looking
1486 * at the currently mapped buffer_heads.
1488 index = mpd->first_page;
1489 end = mpd->next_page - 1;
1491 pagevec_init(&pvec, 0);
1492 while (index <= end) {
1493 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1496 for (i = 0; i < nr_pages; i++) {
1498 struct page *page = pvec.pages[i];
1500 index = page->index;
1504 if (index == size >> PAGE_CACHE_SHIFT)
1505 len = size & ~PAGE_CACHE_MASK;
1507 len = PAGE_CACHE_SIZE;
1509 cur_logical = index << (PAGE_CACHE_SHIFT -
1511 pblock = map->m_pblk + (cur_logical -
1516 BUG_ON(!PageLocked(page));
1517 BUG_ON(PageWriteback(page));
1519 bh = page_bufs = page_buffers(page);
1522 if (map && (cur_logical >= map->m_lblk) &&
1523 (cur_logical <= (map->m_lblk +
1524 (map->m_len - 1)))) {
1525 if (buffer_delay(bh)) {
1526 clear_buffer_delay(bh);
1527 bh->b_blocknr = pblock;
1529 if (buffer_unwritten(bh) ||
1531 BUG_ON(bh->b_blocknr != pblock);
1532 if (map->m_flags & EXT4_MAP_UNINIT)
1533 set_buffer_uninit(bh);
1534 clear_buffer_unwritten(bh);
1538 * skip page if block allocation undone and
1541 if (ext4_bh_delay_or_unwritten(NULL, bh))
1543 bh = bh->b_this_page;
1544 block_start += bh->b_size;
1547 } while (bh != page_bufs);
1554 clear_page_dirty_for_io(page);
1555 err = ext4_bio_write_page(&io_submit, page, len,
1558 mpd->pages_written++;
1560 * In error case, we have to continue because
1561 * remaining pages are still locked
1566 pagevec_release(&pvec);
1568 ext4_io_submit(&io_submit);
1572 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1576 struct pagevec pvec;
1577 struct inode *inode = mpd->inode;
1578 struct address_space *mapping = inode->i_mapping;
1579 ext4_lblk_t start, last;
1581 index = mpd->first_page;
1582 end = mpd->next_page - 1;
1584 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1585 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1586 ext4_es_remove_extent(inode, start, last - start + 1);
1588 pagevec_init(&pvec, 0);
1589 while (index <= end) {
1590 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1593 for (i = 0; i < nr_pages; i++) {
1594 struct page *page = pvec.pages[i];
1595 if (page->index > end)
1597 BUG_ON(!PageLocked(page));
1598 BUG_ON(PageWriteback(page));
1599 block_invalidatepage(page, 0);
1600 ClearPageUptodate(page);
1603 index = pvec.pages[nr_pages - 1]->index + 1;
1604 pagevec_release(&pvec);
1609 static void ext4_print_free_blocks(struct inode *inode)
1611 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1612 struct super_block *sb = inode->i_sb;
1614 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1615 EXT4_C2B(EXT4_SB(inode->i_sb),
1616 ext4_count_free_clusters(inode->i_sb)));
1617 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1618 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1619 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1620 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1621 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1622 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1623 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1624 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1625 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1626 EXT4_I(inode)->i_reserved_data_blocks);
1627 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1628 EXT4_I(inode)->i_reserved_meta_blocks);
1633 * mpage_da_map_and_submit - go through given space, map them
1634 * if necessary, and then submit them for I/O
1636 * @mpd - bh describing space
1638 * The function skips space we know is already mapped to disk blocks.
1641 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1643 int err, blks, get_blocks_flags;
1644 struct ext4_map_blocks map, *mapp = NULL;
1645 sector_t next = mpd->b_blocknr;
1646 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1647 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1648 handle_t *handle = NULL;
1651 * If the blocks are mapped already, or we couldn't accumulate
1652 * any blocks, then proceed immediately to the submission stage.
1654 if ((mpd->b_size == 0) ||
1655 ((mpd->b_state & (1 << BH_Mapped)) &&
1656 !(mpd->b_state & (1 << BH_Delay)) &&
1657 !(mpd->b_state & (1 << BH_Unwritten))))
1660 handle = ext4_journal_current_handle();
1664 * Call ext4_map_blocks() to allocate any delayed allocation
1665 * blocks, or to convert an uninitialized extent to be
1666 * initialized (in the case where we have written into
1667 * one or more preallocated blocks).
1669 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1670 * indicate that we are on the delayed allocation path. This
1671 * affects functions in many different parts of the allocation
1672 * call path. This flag exists primarily because we don't
1673 * want to change *many* call functions, so ext4_map_blocks()
1674 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1675 * inode's allocation semaphore is taken.
1677 * If the blocks in questions were delalloc blocks, set
1678 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1679 * variables are updated after the blocks have been allocated.
1682 map.m_len = max_blocks;
1683 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1684 if (ext4_should_dioread_nolock(mpd->inode))
1685 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1686 if (mpd->b_state & (1 << BH_Delay))
1687 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1689 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1691 struct super_block *sb = mpd->inode->i_sb;
1695 * If get block returns EAGAIN or ENOSPC and there
1696 * appears to be free blocks we will just let
1697 * mpage_da_submit_io() unlock all of the pages.
1702 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1708 * get block failure will cause us to loop in
1709 * writepages, because a_ops->writepage won't be able
1710 * to make progress. The page will be redirtied by
1711 * writepage and writepages will again try to write
1714 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1715 ext4_msg(sb, KERN_CRIT,
1716 "delayed block allocation failed for inode %lu "
1717 "at logical offset %llu with max blocks %zd "
1718 "with error %d", mpd->inode->i_ino,
1719 (unsigned long long) next,
1720 mpd->b_size >> mpd->inode->i_blkbits, err);
1721 ext4_msg(sb, KERN_CRIT,
1722 "This should not happen!! Data will be lost");
1724 ext4_print_free_blocks(mpd->inode);
1726 /* invalidate all the pages */
1727 ext4_da_block_invalidatepages(mpd);
1729 /* Mark this page range as having been completed */
1736 if (map.m_flags & EXT4_MAP_NEW) {
1737 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1740 for (i = 0; i < map.m_len; i++)
1741 unmap_underlying_metadata(bdev, map.m_pblk + i);
1745 * Update on-disk size along with block allocation.
1747 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1748 if (disksize > i_size_read(mpd->inode))
1749 disksize = i_size_read(mpd->inode);
1750 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1751 ext4_update_i_disksize(mpd->inode, disksize);
1752 err = ext4_mark_inode_dirty(handle, mpd->inode);
1754 ext4_error(mpd->inode->i_sb,
1755 "Failed to mark inode %lu dirty",
1760 mpage_da_submit_io(mpd, mapp);
1764 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1765 (1 << BH_Delay) | (1 << BH_Unwritten))
1768 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1770 * @mpd->lbh - extent of blocks
1771 * @logical - logical number of the block in the file
1772 * @b_state - b_state of the buffer head added
1774 * the function is used to collect contig. blocks in same state
1776 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical,
1777 unsigned long b_state)
1780 int blkbits = mpd->inode->i_blkbits;
1781 int nrblocks = mpd->b_size >> blkbits;
1784 * XXX Don't go larger than mballoc is willing to allocate
1785 * This is a stopgap solution. We eventually need to fold
1786 * mpage_da_submit_io() into this function and then call
1787 * ext4_map_blocks() multiple times in a loop
1789 if (nrblocks >= (8*1024*1024 >> blkbits))
1792 /* check if the reserved journal credits might overflow */
1793 if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) {
1794 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1796 * With non-extent format we are limited by the journal
1797 * credit available. Total credit needed to insert
1798 * nrblocks contiguous blocks is dependent on the
1799 * nrblocks. So limit nrblocks.
1805 * First block in the extent
1807 if (mpd->b_size == 0) {
1808 mpd->b_blocknr = logical;
1809 mpd->b_size = 1 << blkbits;
1810 mpd->b_state = b_state & BH_FLAGS;
1814 next = mpd->b_blocknr + nrblocks;
1816 * Can we merge the block to our big extent?
1818 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1819 mpd->b_size += 1 << blkbits;
1825 * We couldn't merge the block to our extent, so we
1826 * need to flush current extent and start new one
1828 mpage_da_map_and_submit(mpd);
1832 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1834 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1838 * This function is grabs code from the very beginning of
1839 * ext4_map_blocks, but assumes that the caller is from delayed write
1840 * time. This function looks up the requested blocks and sets the
1841 * buffer delay bit under the protection of i_data_sem.
1843 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1844 struct ext4_map_blocks *map,
1845 struct buffer_head *bh)
1847 struct extent_status es;
1849 sector_t invalid_block = ~((sector_t) 0xffff);
1850 #ifdef ES_AGGRESSIVE_TEST
1851 struct ext4_map_blocks orig_map;
1853 memcpy(&orig_map, map, sizeof(*map));
1856 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1860 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1861 "logical block %lu\n", inode->i_ino, map->m_len,
1862 (unsigned long) map->m_lblk);
1864 /* Lookup extent status tree firstly */
1865 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1867 if (ext4_es_is_hole(&es)) {
1869 down_read((&EXT4_I(inode)->i_data_sem));
1874 * Delayed extent could be allocated by fallocate.
1875 * So we need to check it.
1877 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1878 map_bh(bh, inode->i_sb, invalid_block);
1880 set_buffer_delay(bh);
1884 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1885 retval = es.es_len - (iblock - es.es_lblk);
1886 if (retval > map->m_len)
1887 retval = map->m_len;
1888 map->m_len = retval;
1889 if (ext4_es_is_written(&es))
1890 map->m_flags |= EXT4_MAP_MAPPED;
1891 else if (ext4_es_is_unwritten(&es))
1892 map->m_flags |= EXT4_MAP_UNWRITTEN;
1896 #ifdef ES_AGGRESSIVE_TEST
1897 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1903 * Try to see if we can get the block without requesting a new
1904 * file system block.
1906 down_read((&EXT4_I(inode)->i_data_sem));
1907 if (ext4_has_inline_data(inode)) {
1909 * We will soon create blocks for this page, and let
1910 * us pretend as if the blocks aren't allocated yet.
1911 * In case of clusters, we have to handle the work
1912 * of mapping from cluster so that the reserved space
1913 * is calculated properly.
1915 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1916 ext4_find_delalloc_cluster(inode, map->m_lblk))
1917 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1919 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1920 retval = ext4_ext_map_blocks(NULL, inode, map,
1921 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1923 retval = ext4_ind_map_blocks(NULL, inode, map,
1924 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1930 * XXX: __block_prepare_write() unmaps passed block,
1933 /* If the block was allocated from previously allocated cluster,
1934 * then we dont need to reserve it again. */
1935 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1936 ret = ext4_da_reserve_space(inode, iblock);
1938 /* not enough space to reserve */
1944 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1945 ~0, EXTENT_STATUS_DELAYED);
1951 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1952 * and it should not appear on the bh->b_state.
1954 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1956 map_bh(bh, inode->i_sb, invalid_block);
1958 set_buffer_delay(bh);
1959 } else if (retval > 0) {
1961 unsigned long long status;
1963 #ifdef ES_AGGRESSIVE_TEST
1964 if (retval != map->m_len) {
1965 printk("ES len assertation failed for inode: %lu "
1966 "retval %d != map->m_len %d "
1967 "in %s (lookup)\n", inode->i_ino, retval,
1968 map->m_len, __func__);
1972 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1973 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1974 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1975 map->m_pblk, status);
1981 up_read((&EXT4_I(inode)->i_data_sem));
1987 * This is a special get_blocks_t callback which is used by
1988 * ext4_da_write_begin(). It will either return mapped block or
1989 * reserve space for a single block.
1991 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1992 * We also have b_blocknr = -1 and b_bdev initialized properly
1994 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1995 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1996 * initialized properly.
1998 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1999 struct buffer_head *bh, int create)
2001 struct ext4_map_blocks map;
2004 BUG_ON(create == 0);
2005 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2007 map.m_lblk = iblock;
2011 * first, we need to know whether the block is allocated already
2012 * preallocated blocks are unmapped but should treated
2013 * the same as allocated blocks.
2015 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
2019 map_bh(bh, inode->i_sb, map.m_pblk);
2020 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2022 if (buffer_unwritten(bh)) {
2023 /* A delayed write to unwritten bh should be marked
2024 * new and mapped. Mapped ensures that we don't do
2025 * get_block multiple times when we write to the same
2026 * offset and new ensures that we do proper zero out
2027 * for partial write.
2030 set_buffer_mapped(bh);
2035 static int bget_one(handle_t *handle, struct buffer_head *bh)
2041 static int bput_one(handle_t *handle, struct buffer_head *bh)
2047 static int __ext4_journalled_writepage(struct page *page,
2050 struct address_space *mapping = page->mapping;
2051 struct inode *inode = mapping->host;
2052 struct buffer_head *page_bufs = NULL;
2053 handle_t *handle = NULL;
2054 int ret = 0, err = 0;
2055 int inline_data = ext4_has_inline_data(inode);
2056 struct buffer_head *inode_bh = NULL;
2058 ClearPageChecked(page);
2061 BUG_ON(page->index != 0);
2062 BUG_ON(len > ext4_get_max_inline_size(inode));
2063 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2064 if (inode_bh == NULL)
2067 page_bufs = page_buffers(page);
2072 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2075 /* As soon as we unlock the page, it can go away, but we have
2076 * references to buffers so we are safe */
2079 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2080 ext4_writepage_trans_blocks(inode));
2081 if (IS_ERR(handle)) {
2082 ret = PTR_ERR(handle);
2086 BUG_ON(!ext4_handle_valid(handle));
2089 ret = ext4_journal_get_write_access(handle, inode_bh);
2091 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2094 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2095 do_journal_get_write_access);
2097 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2102 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2103 err = ext4_journal_stop(handle);
2107 if (!ext4_has_inline_data(inode))
2108 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2110 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2117 * Note that we don't need to start a transaction unless we're journaling data
2118 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2119 * need to file the inode to the transaction's list in ordered mode because if
2120 * we are writing back data added by write(), the inode is already there and if
2121 * we are writing back data modified via mmap(), no one guarantees in which
2122 * transaction the data will hit the disk. In case we are journaling data, we
2123 * cannot start transaction directly because transaction start ranks above page
2124 * lock so we have to do some magic.
2126 * This function can get called via...
2127 * - ext4_da_writepages after taking page lock (have journal handle)
2128 * - journal_submit_inode_data_buffers (no journal handle)
2129 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2130 * - grab_page_cache when doing write_begin (have journal handle)
2132 * We don't do any block allocation in this function. If we have page with
2133 * multiple blocks we need to write those buffer_heads that are mapped. This
2134 * is important for mmaped based write. So if we do with blocksize 1K
2135 * truncate(f, 1024);
2136 * a = mmap(f, 0, 4096);
2138 * truncate(f, 4096);
2139 * we have in the page first buffer_head mapped via page_mkwrite call back
2140 * but other buffer_heads would be unmapped but dirty (dirty done via the
2141 * do_wp_page). So writepage should write the first block. If we modify
2142 * the mmap area beyond 1024 we will again get a page_fault and the
2143 * page_mkwrite callback will do the block allocation and mark the
2144 * buffer_heads mapped.
2146 * We redirty the page if we have any buffer_heads that is either delay or
2147 * unwritten in the page.
2149 * We can get recursively called as show below.
2151 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2154 * But since we don't do any block allocation we should not deadlock.
2155 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2157 static int ext4_writepage(struct page *page,
2158 struct writeback_control *wbc)
2163 struct buffer_head *page_bufs = NULL;
2164 struct inode *inode = page->mapping->host;
2165 struct ext4_io_submit io_submit;
2167 trace_ext4_writepage(page);
2168 size = i_size_read(inode);
2169 if (page->index == size >> PAGE_CACHE_SHIFT)
2170 len = size & ~PAGE_CACHE_MASK;
2172 len = PAGE_CACHE_SIZE;
2174 page_bufs = page_buffers(page);
2176 * We cannot do block allocation or other extent handling in this
2177 * function. If there are buffers needing that, we have to redirty
2178 * the page. But we may reach here when we do a journal commit via
2179 * journal_submit_inode_data_buffers() and in that case we must write
2180 * allocated buffers to achieve data=ordered mode guarantees.
2182 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2183 ext4_bh_delay_or_unwritten)) {
2184 redirty_page_for_writepage(wbc, page);
2185 if (current->flags & PF_MEMALLOC) {
2187 * For memory cleaning there's no point in writing only
2188 * some buffers. So just bail out. Warn if we came here
2189 * from direct reclaim.
2191 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2198 if (PageChecked(page) && ext4_should_journal_data(inode))
2200 * It's mmapped pagecache. Add buffers and journal it. There
2201 * doesn't seem much point in redirtying the page here.
2203 return __ext4_journalled_writepage(page, len);
2205 memset(&io_submit, 0, sizeof(io_submit));
2206 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2207 ext4_io_submit(&io_submit);
2212 * This is called via ext4_da_writepages() to
2213 * calculate the total number of credits to reserve to fit
2214 * a single extent allocation into a single transaction,
2215 * ext4_da_writpeages() will loop calling this before
2216 * the block allocation.
2219 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2221 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2224 * With non-extent format the journal credit needed to
2225 * insert nrblocks contiguous block is dependent on
2226 * number of contiguous block. So we will limit
2227 * number of contiguous block to a sane value
2229 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2230 (max_blocks > EXT4_MAX_TRANS_DATA))
2231 max_blocks = EXT4_MAX_TRANS_DATA;
2233 return ext4_chunk_trans_blocks(inode, max_blocks);
2237 * write_cache_pages_da - walk the list of dirty pages of the given
2238 * address space and accumulate pages that need writing, and call
2239 * mpage_da_map_and_submit to map a single contiguous memory region
2240 * and then write them.
2242 static int write_cache_pages_da(handle_t *handle,
2243 struct address_space *mapping,
2244 struct writeback_control *wbc,
2245 struct mpage_da_data *mpd,
2246 pgoff_t *done_index)
2248 struct buffer_head *bh, *head;
2249 struct inode *inode = mapping->host;
2250 struct pagevec pvec;
2251 unsigned int nr_pages;
2254 long nr_to_write = wbc->nr_to_write;
2255 int i, tag, ret = 0;
2257 memset(mpd, 0, sizeof(struct mpage_da_data));
2260 pagevec_init(&pvec, 0);
2261 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2262 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2264 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2265 tag = PAGECACHE_TAG_TOWRITE;
2267 tag = PAGECACHE_TAG_DIRTY;
2269 *done_index = index;
2270 while (index <= end) {
2271 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2272 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2276 for (i = 0; i < nr_pages; i++) {
2277 struct page *page = pvec.pages[i];
2280 * At this point, the page may be truncated or
2281 * invalidated (changing page->mapping to NULL), or
2282 * even swizzled back from swapper_space to tmpfs file
2283 * mapping. However, page->index will not change
2284 * because we have a reference on the page.
2286 if (page->index > end)
2289 *done_index = page->index + 1;
2292 * If we can't merge this page, and we have
2293 * accumulated an contiguous region, write it
2295 if ((mpd->next_page != page->index) &&
2296 (mpd->next_page != mpd->first_page)) {
2297 mpage_da_map_and_submit(mpd);
2298 goto ret_extent_tail;
2304 * If the page is no longer dirty, or its
2305 * mapping no longer corresponds to inode we
2306 * are writing (which means it has been
2307 * truncated or invalidated), or the page is
2308 * already under writeback and we are not
2309 * doing a data integrity writeback, skip the page
2311 if (!PageDirty(page) ||
2312 (PageWriteback(page) &&
2313 (wbc->sync_mode == WB_SYNC_NONE)) ||
2314 unlikely(page->mapping != mapping)) {
2319 wait_on_page_writeback(page);
2320 BUG_ON(PageWriteback(page));
2323 * If we have inline data and arrive here, it means that
2324 * we will soon create the block for the 1st page, so
2325 * we'd better clear the inline data here.
2327 if (ext4_has_inline_data(inode)) {
2328 BUG_ON(ext4_test_inode_state(inode,
2329 EXT4_STATE_MAY_INLINE_DATA));
2330 ext4_destroy_inline_data(handle, inode);
2333 if (mpd->next_page != page->index)
2334 mpd->first_page = page->index;
2335 mpd->next_page = page->index + 1;
2336 logical = (sector_t) page->index <<
2337 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2339 /* Add all dirty buffers to mpd */
2340 head = page_buffers(page);
2343 BUG_ON(buffer_locked(bh));
2345 * We need to try to allocate unmapped blocks
2346 * in the same page. Otherwise we won't make
2347 * progress with the page in ext4_writepage
2349 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2350 mpage_add_bh_to_extent(mpd, logical,
2353 goto ret_extent_tail;
2354 } else if (buffer_dirty(bh) &&
2355 buffer_mapped(bh)) {
2357 * mapped dirty buffer. We need to
2358 * update the b_state because we look
2359 * at b_state in mpage_da_map_blocks.
2360 * We don't update b_size because if we
2361 * find an unmapped buffer_head later
2362 * we need to use the b_state flag of
2365 if (mpd->b_size == 0)
2367 bh->b_state & BH_FLAGS;
2370 } while ((bh = bh->b_this_page) != head);
2372 if (nr_to_write > 0) {
2374 if (nr_to_write == 0 &&
2375 wbc->sync_mode == WB_SYNC_NONE)
2377 * We stop writing back only if we are
2378 * not doing integrity sync. In case of
2379 * integrity sync we have to keep going
2380 * because someone may be concurrently
2381 * dirtying pages, and we might have
2382 * synced a lot of newly appeared dirty
2383 * pages, but have not synced all of the
2389 pagevec_release(&pvec);
2394 ret = MPAGE_DA_EXTENT_TAIL;
2396 pagevec_release(&pvec);
2402 static int ext4_da_writepages(struct address_space *mapping,
2403 struct writeback_control *wbc)
2406 int range_whole = 0;
2407 handle_t *handle = NULL;
2408 struct mpage_da_data mpd;
2409 struct inode *inode = mapping->host;
2410 int pages_written = 0;
2411 unsigned int max_pages;
2412 int range_cyclic, cycled = 1, io_done = 0;
2413 int needed_blocks, ret = 0;
2414 long desired_nr_to_write, nr_to_writebump = 0;
2415 loff_t range_start = wbc->range_start;
2416 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2417 pgoff_t done_index = 0;
2419 struct blk_plug plug;
2421 trace_ext4_da_writepages(inode, wbc);
2424 * No pages to write? This is mainly a kludge to avoid starting
2425 * a transaction for special inodes like journal inode on last iput()
2426 * because that could violate lock ordering on umount
2428 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2432 * If the filesystem has aborted, it is read-only, so return
2433 * right away instead of dumping stack traces later on that
2434 * will obscure the real source of the problem. We test
2435 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2436 * the latter could be true if the filesystem is mounted
2437 * read-only, and in that case, ext4_da_writepages should
2438 * *never* be called, so if that ever happens, we would want
2441 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2444 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2447 range_cyclic = wbc->range_cyclic;
2448 if (wbc->range_cyclic) {
2449 index = mapping->writeback_index;
2452 wbc->range_start = index << PAGE_CACHE_SHIFT;
2453 wbc->range_end = LLONG_MAX;
2454 wbc->range_cyclic = 0;
2457 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2458 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2462 * This works around two forms of stupidity. The first is in
2463 * the writeback code, which caps the maximum number of pages
2464 * written to be 1024 pages. This is wrong on multiple
2465 * levels; different architectues have a different page size,
2466 * which changes the maximum amount of data which gets
2467 * written. Secondly, 4 megabytes is way too small. XFS
2468 * forces this value to be 16 megabytes by multiplying
2469 * nr_to_write parameter by four, and then relies on its
2470 * allocator to allocate larger extents to make them
2471 * contiguous. Unfortunately this brings us to the second
2472 * stupidity, which is that ext4's mballoc code only allocates
2473 * at most 2048 blocks. So we force contiguous writes up to
2474 * the number of dirty blocks in the inode, or
2475 * sbi->max_writeback_mb_bump whichever is smaller.
2477 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2478 if (!range_cyclic && range_whole) {
2479 if (wbc->nr_to_write == LONG_MAX)
2480 desired_nr_to_write = wbc->nr_to_write;
2482 desired_nr_to_write = wbc->nr_to_write * 8;
2484 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2486 if (desired_nr_to_write > max_pages)
2487 desired_nr_to_write = max_pages;
2489 if (wbc->nr_to_write < desired_nr_to_write) {
2490 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2491 wbc->nr_to_write = desired_nr_to_write;
2495 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2496 tag_pages_for_writeback(mapping, index, end);
2498 blk_start_plug(&plug);
2499 while (!ret && wbc->nr_to_write > 0) {
2502 * we insert one extent at a time. So we need
2503 * credit needed for single extent allocation.
2504 * journalled mode is currently not supported
2507 BUG_ON(ext4_should_journal_data(inode));
2508 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2510 /* start a new transaction*/
2511 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2513 if (IS_ERR(handle)) {
2514 ret = PTR_ERR(handle);
2515 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2516 "%ld pages, ino %lu; err %d", __func__,
2517 wbc->nr_to_write, inode->i_ino, ret);
2518 blk_finish_plug(&plug);
2519 goto out_writepages;
2523 * Now call write_cache_pages_da() to find the next
2524 * contiguous region of logical blocks that need
2525 * blocks to be allocated by ext4 and submit them.
2527 ret = write_cache_pages_da(handle, mapping,
2528 wbc, &mpd, &done_index);
2530 * If we have a contiguous extent of pages and we
2531 * haven't done the I/O yet, map the blocks and submit
2534 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2535 mpage_da_map_and_submit(&mpd);
2536 ret = MPAGE_DA_EXTENT_TAIL;
2538 trace_ext4_da_write_pages(inode, &mpd);
2539 wbc->nr_to_write -= mpd.pages_written;
2541 ext4_journal_stop(handle);
2543 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2544 /* commit the transaction which would
2545 * free blocks released in the transaction
2548 jbd2_journal_force_commit_nested(sbi->s_journal);
2550 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2552 * Got one extent now try with rest of the pages.
2553 * If mpd.retval is set -EIO, journal is aborted.
2554 * So we don't need to write any more.
2556 pages_written += mpd.pages_written;
2559 } else if (wbc->nr_to_write)
2561 * There is no more writeout needed
2562 * or we requested for a noblocking writeout
2563 * and we found the device congested
2567 blk_finish_plug(&plug);
2568 if (!io_done && !cycled) {
2571 wbc->range_start = index << PAGE_CACHE_SHIFT;
2572 wbc->range_end = mapping->writeback_index - 1;
2577 wbc->range_cyclic = range_cyclic;
2578 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2580 * set the writeback_index so that range_cyclic
2581 * mode will write it back later
2583 mapping->writeback_index = done_index;
2586 wbc->nr_to_write -= nr_to_writebump;
2587 wbc->range_start = range_start;
2588 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2592 static int ext4_nonda_switch(struct super_block *sb)
2594 s64 free_blocks, dirty_blocks;
2595 struct ext4_sb_info *sbi = EXT4_SB(sb);
2598 * switch to non delalloc mode if we are running low
2599 * on free block. The free block accounting via percpu
2600 * counters can get slightly wrong with percpu_counter_batch getting
2601 * accumulated on each CPU without updating global counters
2602 * Delalloc need an accurate free block accounting. So switch
2603 * to non delalloc when we are near to error range.
2605 free_blocks = EXT4_C2B(sbi,
2606 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2607 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2609 * Start pushing delalloc when 1/2 of free blocks are dirty.
2611 if (dirty_blocks && (free_blocks < 2 * dirty_blocks) &&
2612 !writeback_in_progress(sb->s_bdi) &&
2613 down_read_trylock(&sb->s_umount)) {
2614 writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2615 up_read(&sb->s_umount);
2618 if (2 * free_blocks < 3 * dirty_blocks ||
2619 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2621 * free block count is less than 150% of dirty blocks
2622 * or free blocks is less than watermark
2629 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2630 loff_t pos, unsigned len, unsigned flags,
2631 struct page **pagep, void **fsdata)
2633 int ret, retries = 0;
2636 struct inode *inode = mapping->host;
2639 index = pos >> PAGE_CACHE_SHIFT;
2641 if (ext4_nonda_switch(inode->i_sb)) {
2642 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2643 return ext4_write_begin(file, mapping, pos,
2644 len, flags, pagep, fsdata);
2646 *fsdata = (void *)0;
2647 trace_ext4_da_write_begin(inode, pos, len, flags);
2649 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2650 ret = ext4_da_write_inline_data_begin(mapping, inode,
2660 * grab_cache_page_write_begin() can take a long time if the
2661 * system is thrashing due to memory pressure, or if the page
2662 * is being written back. So grab it first before we start
2663 * the transaction handle. This also allows us to allocate
2664 * the page (if needed) without using GFP_NOFS.
2667 page = grab_cache_page_write_begin(mapping, index, flags);
2673 * With delayed allocation, we don't log the i_disksize update
2674 * if there is delayed block allocation. But we still need
2675 * to journalling the i_disksize update if writes to the end
2676 * of file which has an already mapped buffer.
2679 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2680 if (IS_ERR(handle)) {
2681 page_cache_release(page);
2682 return PTR_ERR(handle);
2686 if (page->mapping != mapping) {
2687 /* The page got truncated from under us */
2689 page_cache_release(page);
2690 ext4_journal_stop(handle);
2693 /* In case writeback began while the page was unlocked */
2694 wait_on_page_writeback(page);
2696 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2699 ext4_journal_stop(handle);
2701 * block_write_begin may have instantiated a few blocks
2702 * outside i_size. Trim these off again. Don't need
2703 * i_size_read because we hold i_mutex.
2705 if (pos + len > inode->i_size)
2706 ext4_truncate_failed_write(inode);
2708 if (ret == -ENOSPC &&
2709 ext4_should_retry_alloc(inode->i_sb, &retries))
2712 page_cache_release(page);
2721 * Check if we should update i_disksize
2722 * when write to the end of file but not require block allocation
2724 static int ext4_da_should_update_i_disksize(struct page *page,
2725 unsigned long offset)
2727 struct buffer_head *bh;
2728 struct inode *inode = page->mapping->host;
2732 bh = page_buffers(page);
2733 idx = offset >> inode->i_blkbits;
2735 for (i = 0; i < idx; i++)
2736 bh = bh->b_this_page;
2738 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2743 static int ext4_da_write_end(struct file *file,
2744 struct address_space *mapping,
2745 loff_t pos, unsigned len, unsigned copied,
2746 struct page *page, void *fsdata)
2748 struct inode *inode = mapping->host;
2750 handle_t *handle = ext4_journal_current_handle();
2752 unsigned long start, end;
2753 int write_mode = (int)(unsigned long)fsdata;
2755 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2756 switch (ext4_inode_journal_mode(inode)) {
2757 case EXT4_INODE_ORDERED_DATA_MODE:
2758 return ext4_ordered_write_end(file, mapping, pos,
2759 len, copied, page, fsdata);
2760 case EXT4_INODE_WRITEBACK_DATA_MODE:
2761 return ext4_writeback_write_end(file, mapping, pos,
2762 len, copied, page, fsdata);
2768 trace_ext4_da_write_end(inode, pos, len, copied);
2769 start = pos & (PAGE_CACHE_SIZE - 1);
2770 end = start + copied - 1;
2773 * generic_write_end() will run mark_inode_dirty() if i_size
2774 * changes. So let's piggyback the i_disksize mark_inode_dirty
2777 new_i_size = pos + copied;
2778 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2779 if (ext4_has_inline_data(inode) ||
2780 ext4_da_should_update_i_disksize(page, end)) {
2781 down_write(&EXT4_I(inode)->i_data_sem);
2782 if (new_i_size > EXT4_I(inode)->i_disksize)
2783 EXT4_I(inode)->i_disksize = new_i_size;
2784 up_write(&EXT4_I(inode)->i_data_sem);
2785 /* We need to mark inode dirty even if
2786 * new_i_size is less that inode->i_size
2787 * bu greater than i_disksize.(hint delalloc)
2789 ext4_mark_inode_dirty(handle, inode);
2793 if (write_mode != CONVERT_INLINE_DATA &&
2794 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2795 ext4_has_inline_data(inode))
2796 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2799 ret2 = generic_write_end(file, mapping, pos, len, copied,
2805 ret2 = ext4_journal_stop(handle);
2809 return ret ? ret : copied;
2812 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2815 * Drop reserved blocks
2817 BUG_ON(!PageLocked(page));
2818 if (!page_has_buffers(page))
2821 ext4_da_page_release_reservation(page, offset);
2824 ext4_invalidatepage(page, offset);
2830 * Force all delayed allocation blocks to be allocated for a given inode.
2832 int ext4_alloc_da_blocks(struct inode *inode)
2834 trace_ext4_alloc_da_blocks(inode);
2836 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2837 !EXT4_I(inode)->i_reserved_meta_blocks)
2841 * We do something simple for now. The filemap_flush() will
2842 * also start triggering a write of the data blocks, which is
2843 * not strictly speaking necessary (and for users of
2844 * laptop_mode, not even desirable). However, to do otherwise
2845 * would require replicating code paths in:
2847 * ext4_da_writepages() ->
2848 * write_cache_pages() ---> (via passed in callback function)
2849 * __mpage_da_writepage() -->
2850 * mpage_add_bh_to_extent()
2851 * mpage_da_map_blocks()
2853 * The problem is that write_cache_pages(), located in
2854 * mm/page-writeback.c, marks pages clean in preparation for
2855 * doing I/O, which is not desirable if we're not planning on
2858 * We could call write_cache_pages(), and then redirty all of
2859 * the pages by calling redirty_page_for_writepage() but that
2860 * would be ugly in the extreme. So instead we would need to
2861 * replicate parts of the code in the above functions,
2862 * simplifying them because we wouldn't actually intend to
2863 * write out the pages, but rather only collect contiguous
2864 * logical block extents, call the multi-block allocator, and
2865 * then update the buffer heads with the block allocations.
2867 * For now, though, we'll cheat by calling filemap_flush(),
2868 * which will map the blocks, and start the I/O, but not
2869 * actually wait for the I/O to complete.
2871 return filemap_flush(inode->i_mapping);
2875 * bmap() is special. It gets used by applications such as lilo and by
2876 * the swapper to find the on-disk block of a specific piece of data.
2878 * Naturally, this is dangerous if the block concerned is still in the
2879 * journal. If somebody makes a swapfile on an ext4 data-journaling
2880 * filesystem and enables swap, then they may get a nasty shock when the
2881 * data getting swapped to that swapfile suddenly gets overwritten by
2882 * the original zero's written out previously to the journal and
2883 * awaiting writeback in the kernel's buffer cache.
2885 * So, if we see any bmap calls here on a modified, data-journaled file,
2886 * take extra steps to flush any blocks which might be in the cache.
2888 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2890 struct inode *inode = mapping->host;
2895 * We can get here for an inline file via the FIBMAP ioctl
2897 if (ext4_has_inline_data(inode))
2900 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2901 test_opt(inode->i_sb, DELALLOC)) {
2903 * With delalloc we want to sync the file
2904 * so that we can make sure we allocate
2907 filemap_write_and_wait(mapping);
2910 if (EXT4_JOURNAL(inode) &&
2911 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2913 * This is a REALLY heavyweight approach, but the use of
2914 * bmap on dirty files is expected to be extremely rare:
2915 * only if we run lilo or swapon on a freshly made file
2916 * do we expect this to happen.
2918 * (bmap requires CAP_SYS_RAWIO so this does not
2919 * represent an unprivileged user DOS attack --- we'd be
2920 * in trouble if mortal users could trigger this path at
2923 * NB. EXT4_STATE_JDATA is not set on files other than
2924 * regular files. If somebody wants to bmap a directory
2925 * or symlink and gets confused because the buffer
2926 * hasn't yet been flushed to disk, they deserve
2927 * everything they get.
2930 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2931 journal = EXT4_JOURNAL(inode);
2932 jbd2_journal_lock_updates(journal);
2933 err = jbd2_journal_flush(journal);
2934 jbd2_journal_unlock_updates(journal);
2940 return generic_block_bmap(mapping, block, ext4_get_block);
2943 static int ext4_readpage(struct file *file, struct page *page)
2946 struct inode *inode = page->mapping->host;
2948 trace_ext4_readpage(page);
2950 if (ext4_has_inline_data(inode))
2951 ret = ext4_readpage_inline(inode, page);
2954 return mpage_readpage(page, ext4_get_block);
2960 ext4_readpages(struct file *file, struct address_space *mapping,
2961 struct list_head *pages, unsigned nr_pages)
2963 struct inode *inode = mapping->host;
2965 /* If the file has inline data, no need to do readpages. */
2966 if (ext4_has_inline_data(inode))
2969 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2972 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2974 trace_ext4_invalidatepage(page, offset);
2976 /* No journalling happens on data buffers when this function is used */
2977 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2979 block_invalidatepage(page, offset);
2982 static int __ext4_journalled_invalidatepage(struct page *page,
2983 unsigned long offset)
2985 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2987 trace_ext4_journalled_invalidatepage(page, offset);
2990 * If it's a full truncate we just forget about the pending dirtying
2993 ClearPageChecked(page);
2995 return jbd2_journal_invalidatepage(journal, page, offset);
2998 /* Wrapper for aops... */
2999 static void ext4_journalled_invalidatepage(struct page *page,
3000 unsigned long offset)
3002 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
3005 static int ext4_releasepage(struct page *page, gfp_t wait)
3007 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3009 trace_ext4_releasepage(page);
3011 WARN_ON(PageChecked(page));
3012 if (!page_has_buffers(page))
3015 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3017 return try_to_free_buffers(page);
3021 * ext4_get_block used when preparing for a DIO write or buffer write.
3022 * We allocate an uinitialized extent if blocks haven't been allocated.
3023 * The extent will be converted to initialized after the IO is complete.
3025 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3026 struct buffer_head *bh_result, int create)
3028 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3029 inode->i_ino, create);
3030 return _ext4_get_block(inode, iblock, bh_result,
3031 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3034 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3035 struct buffer_head *bh_result, int create)
3037 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3038 inode->i_ino, create);
3039 return _ext4_get_block(inode, iblock, bh_result,
3040 EXT4_GET_BLOCKS_NO_LOCK);
3043 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3044 ssize_t size, void *private, int ret,
3047 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
3048 ext4_io_end_t *io_end = iocb->private;
3050 /* if not async direct IO or dio with 0 bytes write, just return */
3051 if (!io_end || !size)
3054 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3055 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3056 iocb->private, io_end->inode->i_ino, iocb, offset,
3059 iocb->private = NULL;
3061 /* if not aio dio with unwritten extents, just free io and return */
3062 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3063 ext4_free_io_end(io_end);
3065 inode_dio_done(inode);
3067 aio_complete(iocb, ret, 0);
3071 io_end->offset = offset;
3072 io_end->size = size;
3074 io_end->iocb = iocb;
3075 io_end->result = ret;
3078 ext4_add_complete_io(io_end);
3082 * For ext4 extent files, ext4 will do direct-io write to holes,
3083 * preallocated extents, and those write extend the file, no need to
3084 * fall back to buffered IO.
3086 * For holes, we fallocate those blocks, mark them as uninitialized
3087 * If those blocks were preallocated, we mark sure they are split, but
3088 * still keep the range to write as uninitialized.
3090 * The unwritten extents will be converted to written when DIO is completed.
3091 * For async direct IO, since the IO may still pending when return, we
3092 * set up an end_io call back function, which will do the conversion
3093 * when async direct IO completed.
3095 * If the O_DIRECT write will extend the file then add this inode to the
3096 * orphan list. So recovery will truncate it back to the original size
3097 * if the machine crashes during the write.
3100 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3101 const struct iovec *iov, loff_t offset,
3102 unsigned long nr_segs)
3104 struct file *file = iocb->ki_filp;
3105 struct inode *inode = file->f_mapping->host;
3107 size_t count = iov_length(iov, nr_segs);
3109 get_block_t *get_block_func = NULL;
3111 loff_t final_size = offset + count;
3113 /* Use the old path for reads and writes beyond i_size. */
3114 if (rw != WRITE || final_size > inode->i_size)
3115 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3117 BUG_ON(iocb->private == NULL);
3119 /* If we do a overwrite dio, i_mutex locking can be released */
3120 overwrite = *((int *)iocb->private);
3123 atomic_inc(&inode->i_dio_count);
3124 down_read(&EXT4_I(inode)->i_data_sem);
3125 mutex_unlock(&inode->i_mutex);
3129 * We could direct write to holes and fallocate.
3131 * Allocated blocks to fill the hole are marked as
3132 * uninitialized to prevent parallel buffered read to expose
3133 * the stale data before DIO complete the data IO.
3135 * As to previously fallocated extents, ext4 get_block will
3136 * just simply mark the buffer mapped but still keep the
3137 * extents uninitialized.
3139 * For non AIO case, we will convert those unwritten extents
3140 * to written after return back from blockdev_direct_IO.
3142 * For async DIO, the conversion needs to be deferred when the
3143 * IO is completed. The ext4 end_io callback function will be
3144 * called to take care of the conversion work. Here for async
3145 * case, we allocate an io_end structure to hook to the iocb.
3147 iocb->private = NULL;
3148 ext4_inode_aio_set(inode, NULL);
3149 if (!is_sync_kiocb(iocb)) {
3150 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
3155 io_end->flag |= EXT4_IO_END_DIRECT;
3156 iocb->private = io_end;
3158 * we save the io structure for current async direct
3159 * IO, so that later ext4_map_blocks() could flag the
3160 * io structure whether there is a unwritten extents
3161 * needs to be converted when IO is completed.
3163 ext4_inode_aio_set(inode, io_end);
3167 get_block_func = ext4_get_block_write_nolock;
3169 get_block_func = ext4_get_block_write;
3170 dio_flags = DIO_LOCKING;
3172 ret = __blockdev_direct_IO(rw, iocb, inode,
3173 inode->i_sb->s_bdev, iov,
3181 ext4_inode_aio_set(inode, NULL);
3183 * The io_end structure takes a reference to the inode, that
3184 * structure needs to be destroyed and the reference to the
3185 * inode need to be dropped, when IO is complete, even with 0
3186 * byte write, or failed.
3188 * In the successful AIO DIO case, the io_end structure will
3189 * be destroyed and the reference to the inode will be dropped
3190 * after the end_io call back function is called.
3192 * In the case there is 0 byte write, or error case, since VFS
3193 * direct IO won't invoke the end_io call back function, we
3194 * need to free the end_io structure here.
3196 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3197 ext4_free_io_end(iocb->private);
3198 iocb->private = NULL;
3199 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3200 EXT4_STATE_DIO_UNWRITTEN)) {
3203 * for non AIO case, since the IO is already
3204 * completed, we could do the conversion right here
3206 err = ext4_convert_unwritten_extents(inode,
3210 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3214 /* take i_mutex locking again if we do a ovewrite dio */
3216 inode_dio_done(inode);
3217 up_read(&EXT4_I(inode)->i_data_sem);
3218 mutex_lock(&inode->i_mutex);
3224 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3225 const struct iovec *iov, loff_t offset,
3226 unsigned long nr_segs)
3228 struct file *file = iocb->ki_filp;
3229 struct inode *inode = file->f_mapping->host;
3233 * If we are doing data journalling we don't support O_DIRECT
3235 if (ext4_should_journal_data(inode))
3238 /* Let buffer I/O handle the inline data case. */
3239 if (ext4_has_inline_data(inode))
3242 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3243 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3244 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3246 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3247 trace_ext4_direct_IO_exit(inode, offset,
3248 iov_length(iov, nr_segs), rw, ret);
3253 * Pages can be marked dirty completely asynchronously from ext4's journalling
3254 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3255 * much here because ->set_page_dirty is called under VFS locks. The page is
3256 * not necessarily locked.
3258 * We cannot just dirty the page and leave attached buffers clean, because the
3259 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3260 * or jbddirty because all the journalling code will explode.
3262 * So what we do is to mark the page "pending dirty" and next time writepage
3263 * is called, propagate that into the buffers appropriately.
3265 static int ext4_journalled_set_page_dirty(struct page *page)
3267 SetPageChecked(page);
3268 return __set_page_dirty_nobuffers(page);
3271 static const struct address_space_operations ext4_ordered_aops = {
3272 .readpage = ext4_readpage,
3273 .readpages = ext4_readpages,
3274 .writepage = ext4_writepage,
3275 .write_begin = ext4_write_begin,
3276 .write_end = ext4_ordered_write_end,
3278 .invalidatepage = ext4_invalidatepage,
3279 .releasepage = ext4_releasepage,
3280 .direct_IO = ext4_direct_IO,
3281 .migratepage = buffer_migrate_page,
3282 .is_partially_uptodate = block_is_partially_uptodate,
3283 .error_remove_page = generic_error_remove_page,
3286 static const struct address_space_operations ext4_writeback_aops = {
3287 .readpage = ext4_readpage,
3288 .readpages = ext4_readpages,
3289 .writepage = ext4_writepage,
3290 .write_begin = ext4_write_begin,
3291 .write_end = ext4_writeback_write_end,
3293 .invalidatepage = ext4_invalidatepage,
3294 .releasepage = ext4_releasepage,
3295 .direct_IO = ext4_direct_IO,
3296 .migratepage = buffer_migrate_page,
3297 .is_partially_uptodate = block_is_partially_uptodate,
3298 .error_remove_page = generic_error_remove_page,
3301 static const struct address_space_operations ext4_journalled_aops = {
3302 .readpage = ext4_readpage,
3303 .readpages = ext4_readpages,
3304 .writepage = ext4_writepage,
3305 .write_begin = ext4_write_begin,
3306 .write_end = ext4_journalled_write_end,
3307 .set_page_dirty = ext4_journalled_set_page_dirty,
3309 .invalidatepage = ext4_journalled_invalidatepage,
3310 .releasepage = ext4_releasepage,
3311 .direct_IO = ext4_direct_IO,
3312 .is_partially_uptodate = block_is_partially_uptodate,
3313 .error_remove_page = generic_error_remove_page,
3316 static const struct address_space_operations ext4_da_aops = {
3317 .readpage = ext4_readpage,
3318 .readpages = ext4_readpages,
3319 .writepage = ext4_writepage,
3320 .writepages = ext4_da_writepages,
3321 .write_begin = ext4_da_write_begin,
3322 .write_end = ext4_da_write_end,
3324 .invalidatepage = ext4_da_invalidatepage,
3325 .releasepage = ext4_releasepage,
3326 .direct_IO = ext4_direct_IO,
3327 .migratepage = buffer_migrate_page,
3328 .is_partially_uptodate = block_is_partially_uptodate,
3329 .error_remove_page = generic_error_remove_page,
3332 void ext4_set_aops(struct inode *inode)
3334 switch (ext4_inode_journal_mode(inode)) {
3335 case EXT4_INODE_ORDERED_DATA_MODE:
3336 if (test_opt(inode->i_sb, DELALLOC))
3337 inode->i_mapping->a_ops = &ext4_da_aops;
3339 inode->i_mapping->a_ops = &ext4_ordered_aops;
3341 case EXT4_INODE_WRITEBACK_DATA_MODE:
3342 if (test_opt(inode->i_sb, DELALLOC))
3343 inode->i_mapping->a_ops = &ext4_da_aops;
3345 inode->i_mapping->a_ops = &ext4_writeback_aops;
3347 case EXT4_INODE_JOURNAL_DATA_MODE:
3348 inode->i_mapping->a_ops = &ext4_journalled_aops;
3357 * ext4_discard_partial_page_buffers()
3358 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3359 * This function finds and locks the page containing the offset
3360 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3361 * Calling functions that already have the page locked should call
3362 * ext4_discard_partial_page_buffers_no_lock directly.
3364 int ext4_discard_partial_page_buffers(handle_t *handle,
3365 struct address_space *mapping, loff_t from,
3366 loff_t length, int flags)
3368 struct inode *inode = mapping->host;
3372 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3373 mapping_gfp_mask(mapping) & ~__GFP_FS);
3377 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3378 from, length, flags);
3381 page_cache_release(page);
3386 * ext4_discard_partial_page_buffers_no_lock()
3387 * Zeros a page range of length 'length' starting from offset 'from'.
3388 * Buffer heads that correspond to the block aligned regions of the
3389 * zeroed range will be unmapped. Unblock aligned regions
3390 * will have the corresponding buffer head mapped if needed so that
3391 * that region of the page can be updated with the partial zero out.
3393 * This function assumes that the page has already been locked. The
3394 * The range to be discarded must be contained with in the given page.
3395 * If the specified range exceeds the end of the page it will be shortened
3396 * to the end of the page that corresponds to 'from'. This function is
3397 * appropriate for updating a page and it buffer heads to be unmapped and
3398 * zeroed for blocks that have been either released, or are going to be
3401 * handle: The journal handle
3402 * inode: The files inode
3403 * page: A locked page that contains the offset "from"
3404 * from: The starting byte offset (from the beginning of the file)
3405 * to begin discarding
3406 * len: The length of bytes to discard
3407 * flags: Optional flags that may be used:
3409 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3410 * Only zero the regions of the page whose buffer heads
3411 * have already been unmapped. This flag is appropriate
3412 * for updating the contents of a page whose blocks may
3413 * have already been released, and we only want to zero
3414 * out the regions that correspond to those released blocks.
3416 * Returns zero on success or negative on failure.
3418 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3419 struct inode *inode, struct page *page, loff_t from,
3420 loff_t length, int flags)
3422 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3423 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3424 unsigned int blocksize, max, pos;
3426 struct buffer_head *bh;
3429 blocksize = inode->i_sb->s_blocksize;
3430 max = PAGE_CACHE_SIZE - offset;
3432 if (index != page->index)
3436 * correct length if it does not fall between
3437 * 'from' and the end of the page
3439 if (length > max || length < 0)
3442 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3444 if (!page_has_buffers(page))
3445 create_empty_buffers(page, blocksize, 0);
3447 /* Find the buffer that contains "offset" */
3448 bh = page_buffers(page);
3450 while (offset >= pos) {
3451 bh = bh->b_this_page;
3457 while (pos < offset + length) {
3458 unsigned int end_of_block, range_to_discard;
3462 /* The length of space left to zero and unmap */
3463 range_to_discard = offset + length - pos;
3465 /* The length of space until the end of the block */
3466 end_of_block = blocksize - (pos & (blocksize-1));
3469 * Do not unmap or zero past end of block
3470 * for this buffer head
3472 if (range_to_discard > end_of_block)
3473 range_to_discard = end_of_block;
3477 * Skip this buffer head if we are only zeroing unampped
3478 * regions of the page
3480 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3484 /* If the range is block aligned, unmap */
3485 if (range_to_discard == blocksize) {
3486 clear_buffer_dirty(bh);
3488 clear_buffer_mapped(bh);
3489 clear_buffer_req(bh);
3490 clear_buffer_new(bh);
3491 clear_buffer_delay(bh);
3492 clear_buffer_unwritten(bh);
3493 clear_buffer_uptodate(bh);
3494 zero_user(page, pos, range_to_discard);
3495 BUFFER_TRACE(bh, "Buffer discarded");
3500 * If this block is not completely contained in the range
3501 * to be discarded, then it is not going to be released. Because
3502 * we need to keep this block, we need to make sure this part
3503 * of the page is uptodate before we modify it by writeing
3504 * partial zeros on it.
3506 if (!buffer_mapped(bh)) {
3508 * Buffer head must be mapped before we can read
3511 BUFFER_TRACE(bh, "unmapped");
3512 ext4_get_block(inode, iblock, bh, 0);
3513 /* unmapped? It's a hole - nothing to do */
3514 if (!buffer_mapped(bh)) {
3515 BUFFER_TRACE(bh, "still unmapped");
3520 /* Ok, it's mapped. Make sure it's up-to-date */
3521 if (PageUptodate(page))
3522 set_buffer_uptodate(bh);
3524 if (!buffer_uptodate(bh)) {
3526 ll_rw_block(READ, 1, &bh);
3528 /* Uhhuh. Read error. Complain and punt.*/
3529 if (!buffer_uptodate(bh))
3533 if (ext4_should_journal_data(inode)) {
3534 BUFFER_TRACE(bh, "get write access");
3535 err = ext4_journal_get_write_access(handle, bh);
3540 zero_user(page, pos, range_to_discard);
3543 if (ext4_should_journal_data(inode)) {
3544 err = ext4_handle_dirty_metadata(handle, inode, bh);
3546 mark_buffer_dirty(bh);
3548 BUFFER_TRACE(bh, "Partial buffer zeroed");
3550 bh = bh->b_this_page;
3552 pos += range_to_discard;
3558 int ext4_can_truncate(struct inode *inode)
3560 if (S_ISREG(inode->i_mode))
3562 if (S_ISDIR(inode->i_mode))
3564 if (S_ISLNK(inode->i_mode))
3565 return !ext4_inode_is_fast_symlink(inode);
3570 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3571 * associated with the given offset and length
3573 * @inode: File inode
3574 * @offset: The offset where the hole will begin
3575 * @len: The length of the hole
3577 * Returns: 0 on success or negative on failure
3580 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3582 struct inode *inode = file->f_path.dentry->d_inode;
3583 if (!S_ISREG(inode->i_mode))
3586 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3587 return ext4_ind_punch_hole(file, offset, length);
3589 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3590 /* TODO: Add support for bigalloc file systems */
3594 trace_ext4_punch_hole(inode, offset, length);
3596 return ext4_ext_punch_hole(file, offset, length);
3602 * We block out ext4_get_block() block instantiations across the entire
3603 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3604 * simultaneously on behalf of the same inode.
3606 * As we work through the truncate and commit bits of it to the journal there
3607 * is one core, guiding principle: the file's tree must always be consistent on
3608 * disk. We must be able to restart the truncate after a crash.
3610 * The file's tree may be transiently inconsistent in memory (although it
3611 * probably isn't), but whenever we close off and commit a journal transaction,
3612 * the contents of (the filesystem + the journal) must be consistent and
3613 * restartable. It's pretty simple, really: bottom up, right to left (although
3614 * left-to-right works OK too).
3616 * Note that at recovery time, journal replay occurs *before* the restart of
3617 * truncate against the orphan inode list.
3619 * The committed inode has the new, desired i_size (which is the same as
3620 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3621 * that this inode's truncate did not complete and it will again call
3622 * ext4_truncate() to have another go. So there will be instantiated blocks
3623 * to the right of the truncation point in a crashed ext4 filesystem. But
3624 * that's fine - as long as they are linked from the inode, the post-crash
3625 * ext4_truncate() run will find them and release them.
3627 void ext4_truncate(struct inode *inode)
3629 trace_ext4_truncate_enter(inode);
3631 if (!ext4_can_truncate(inode))
3634 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3636 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3637 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3639 if (ext4_has_inline_data(inode)) {
3642 ext4_inline_data_truncate(inode, &has_inline);
3647 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3648 ext4_ext_truncate(inode);
3650 ext4_ind_truncate(inode);
3652 trace_ext4_truncate_exit(inode);
3656 * ext4_get_inode_loc returns with an extra refcount against the inode's
3657 * underlying buffer_head on success. If 'in_mem' is true, we have all
3658 * data in memory that is needed to recreate the on-disk version of this
3661 static int __ext4_get_inode_loc(struct inode *inode,
3662 struct ext4_iloc *iloc, int in_mem)
3664 struct ext4_group_desc *gdp;
3665 struct buffer_head *bh;
3666 struct super_block *sb = inode->i_sb;
3668 int inodes_per_block, inode_offset;
3671 if (!ext4_valid_inum(sb, inode->i_ino))
3674 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3675 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3680 * Figure out the offset within the block group inode table
3682 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3683 inode_offset = ((inode->i_ino - 1) %
3684 EXT4_INODES_PER_GROUP(sb));
3685 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3686 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3688 bh = sb_getblk(sb, block);
3691 if (!buffer_uptodate(bh)) {
3695 * If the buffer has the write error flag, we have failed
3696 * to write out another inode in the same block. In this
3697 * case, we don't have to read the block because we may
3698 * read the old inode data successfully.
3700 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3701 set_buffer_uptodate(bh);
3703 if (buffer_uptodate(bh)) {
3704 /* someone brought it uptodate while we waited */
3710 * If we have all information of the inode in memory and this
3711 * is the only valid inode in the block, we need not read the
3715 struct buffer_head *bitmap_bh;
3718 start = inode_offset & ~(inodes_per_block - 1);
3720 /* Is the inode bitmap in cache? */
3721 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3722 if (unlikely(!bitmap_bh))
3726 * If the inode bitmap isn't in cache then the
3727 * optimisation may end up performing two reads instead
3728 * of one, so skip it.
3730 if (!buffer_uptodate(bitmap_bh)) {
3734 for (i = start; i < start + inodes_per_block; i++) {
3735 if (i == inode_offset)
3737 if (ext4_test_bit(i, bitmap_bh->b_data))
3741 if (i == start + inodes_per_block) {
3742 /* all other inodes are free, so skip I/O */
3743 memset(bh->b_data, 0, bh->b_size);
3744 set_buffer_uptodate(bh);
3752 * If we need to do any I/O, try to pre-readahead extra
3753 * blocks from the inode table.
3755 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3756 ext4_fsblk_t b, end, table;
3759 table = ext4_inode_table(sb, gdp);
3760 /* s_inode_readahead_blks is always a power of 2 */
3761 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3764 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3765 num = EXT4_INODES_PER_GROUP(sb);
3766 if (ext4_has_group_desc_csum(sb))
3767 num -= ext4_itable_unused_count(sb, gdp);
3768 table += num / inodes_per_block;
3772 sb_breadahead(sb, b++);
3776 * There are other valid inodes in the buffer, this inode
3777 * has in-inode xattrs, or we don't have this inode in memory.
3778 * Read the block from disk.
3780 trace_ext4_load_inode(inode);
3782 bh->b_end_io = end_buffer_read_sync;
3783 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3785 if (!buffer_uptodate(bh)) {
3786 EXT4_ERROR_INODE_BLOCK(inode, block,
3787 "unable to read itable block");
3797 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3799 /* We have all inode data except xattrs in memory here. */
3800 return __ext4_get_inode_loc(inode, iloc,
3801 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3804 void ext4_set_inode_flags(struct inode *inode)
3806 unsigned int flags = EXT4_I(inode)->i_flags;
3808 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3809 if (flags & EXT4_SYNC_FL)
3810 inode->i_flags |= S_SYNC;
3811 if (flags & EXT4_APPEND_FL)
3812 inode->i_flags |= S_APPEND;
3813 if (flags & EXT4_IMMUTABLE_FL)
3814 inode->i_flags |= S_IMMUTABLE;
3815 if (flags & EXT4_NOATIME_FL)
3816 inode->i_flags |= S_NOATIME;
3817 if (flags & EXT4_DIRSYNC_FL)
3818 inode->i_flags |= S_DIRSYNC;
3821 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3822 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3824 unsigned int vfs_fl;
3825 unsigned long old_fl, new_fl;
3828 vfs_fl = ei->vfs_inode.i_flags;
3829 old_fl = ei->i_flags;
3830 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3831 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3833 if (vfs_fl & S_SYNC)
3834 new_fl |= EXT4_SYNC_FL;
3835 if (vfs_fl & S_APPEND)
3836 new_fl |= EXT4_APPEND_FL;
3837 if (vfs_fl & S_IMMUTABLE)
3838 new_fl |= EXT4_IMMUTABLE_FL;
3839 if (vfs_fl & S_NOATIME)
3840 new_fl |= EXT4_NOATIME_FL;
3841 if (vfs_fl & S_DIRSYNC)
3842 new_fl |= EXT4_DIRSYNC_FL;
3843 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3846 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3847 struct ext4_inode_info *ei)
3850 struct inode *inode = &(ei->vfs_inode);
3851 struct super_block *sb = inode->i_sb;
3853 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3854 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3855 /* we are using combined 48 bit field */
3856 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3857 le32_to_cpu(raw_inode->i_blocks_lo);
3858 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3859 /* i_blocks represent file system block size */
3860 return i_blocks << (inode->i_blkbits - 9);
3865 return le32_to_cpu(raw_inode->i_blocks_lo);
3869 static inline void ext4_iget_extra_inode(struct inode *inode,
3870 struct ext4_inode *raw_inode,
3871 struct ext4_inode_info *ei)
3873 __le32 *magic = (void *)raw_inode +
3874 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3875 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3876 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3877 ext4_find_inline_data_nolock(inode);
3879 EXT4_I(inode)->i_inline_off = 0;
3882 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3884 struct ext4_iloc iloc;
3885 struct ext4_inode *raw_inode;
3886 struct ext4_inode_info *ei;
3887 struct inode *inode;
3888 journal_t *journal = EXT4_SB(sb)->s_journal;
3894 inode = iget_locked(sb, ino);
3896 return ERR_PTR(-ENOMEM);
3897 if (!(inode->i_state & I_NEW))
3903 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3906 raw_inode = ext4_raw_inode(&iloc);
3908 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3909 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3910 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3911 EXT4_INODE_SIZE(inode->i_sb)) {
3912 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3913 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3914 EXT4_INODE_SIZE(inode->i_sb));
3919 ei->i_extra_isize = 0;
3921 /* Precompute checksum seed for inode metadata */
3922 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3923 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3924 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3926 __le32 inum = cpu_to_le32(inode->i_ino);
3927 __le32 gen = raw_inode->i_generation;
3928 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3930 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3934 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3935 EXT4_ERROR_INODE(inode, "checksum invalid");
3940 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3941 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3942 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3943 if (!(test_opt(inode->i_sb, NO_UID32))) {
3944 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3945 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3947 i_uid_write(inode, i_uid);
3948 i_gid_write(inode, i_gid);
3949 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3951 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3952 ei->i_inline_off = 0;
3953 ei->i_dir_start_lookup = 0;
3954 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3955 /* We now have enough fields to check if the inode was active or not.
3956 * This is needed because nfsd might try to access dead inodes
3957 * the test is that same one that e2fsck uses
3958 * NeilBrown 1999oct15
3960 if (inode->i_nlink == 0) {
3961 if (inode->i_mode == 0 ||
3962 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3963 /* this inode is deleted */
3967 /* The only unlinked inodes we let through here have
3968 * valid i_mode and are being read by the orphan
3969 * recovery code: that's fine, we're about to complete
3970 * the process of deleting those. */
3972 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3973 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3974 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3975 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3977 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3978 inode->i_size = ext4_isize(raw_inode);
3979 ei->i_disksize = inode->i_size;
3981 ei->i_reserved_quota = 0;
3983 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3984 ei->i_block_group = iloc.block_group;
3985 ei->i_last_alloc_group = ~0;
3987 * NOTE! The in-memory inode i_data array is in little-endian order
3988 * even on big-endian machines: we do NOT byteswap the block numbers!
3990 for (block = 0; block < EXT4_N_BLOCKS; block++)
3991 ei->i_data[block] = raw_inode->i_block[block];
3992 INIT_LIST_HEAD(&ei->i_orphan);
3995 * Set transaction id's of transactions that have to be committed
3996 * to finish f[data]sync. We set them to currently running transaction
3997 * as we cannot be sure that the inode or some of its metadata isn't
3998 * part of the transaction - the inode could have been reclaimed and
3999 * now it is reread from disk.
4002 transaction_t *transaction;
4005 read_lock(&journal->j_state_lock);
4006 if (journal->j_running_transaction)
4007 transaction = journal->j_running_transaction;
4009 transaction = journal->j_committing_transaction;
4011 tid = transaction->t_tid;
4013 tid = journal->j_commit_sequence;
4014 read_unlock(&journal->j_state_lock);
4015 ei->i_sync_tid = tid;
4016 ei->i_datasync_tid = tid;
4019 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4020 if (ei->i_extra_isize == 0) {
4021 /* The extra space is currently unused. Use it. */
4022 ei->i_extra_isize = sizeof(struct ext4_inode) -
4023 EXT4_GOOD_OLD_INODE_SIZE;
4025 ext4_iget_extra_inode(inode, raw_inode, ei);
4029 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4030 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4031 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4032 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4034 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4035 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4036 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4038 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4042 if (ei->i_file_acl &&
4043 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4044 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4048 } else if (!ext4_has_inline_data(inode)) {
4049 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4050 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4051 (S_ISLNK(inode->i_mode) &&
4052 !ext4_inode_is_fast_symlink(inode))))
4053 /* Validate extent which is part of inode */
4054 ret = ext4_ext_check_inode(inode);
4055 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4056 (S_ISLNK(inode->i_mode) &&
4057 !ext4_inode_is_fast_symlink(inode))) {
4058 /* Validate block references which are part of inode */
4059 ret = ext4_ind_check_inode(inode);
4065 if (S_ISREG(inode->i_mode)) {
4066 inode->i_op = &ext4_file_inode_operations;
4067 inode->i_fop = &ext4_file_operations;
4068 ext4_set_aops(inode);
4069 } else if (S_ISDIR(inode->i_mode)) {
4070 inode->i_op = &ext4_dir_inode_operations;
4071 inode->i_fop = &ext4_dir_operations;
4072 } else if (S_ISLNK(inode->i_mode)) {
4073 if (ext4_inode_is_fast_symlink(inode)) {
4074 inode->i_op = &ext4_fast_symlink_inode_operations;
4075 nd_terminate_link(ei->i_data, inode->i_size,
4076 sizeof(ei->i_data) - 1);
4078 inode->i_op = &ext4_symlink_inode_operations;
4079 ext4_set_aops(inode);
4081 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4082 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4083 inode->i_op = &ext4_special_inode_operations;
4084 if (raw_inode->i_block[0])
4085 init_special_inode(inode, inode->i_mode,
4086 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4088 init_special_inode(inode, inode->i_mode,
4089 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4092 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4096 ext4_set_inode_flags(inode);
4097 unlock_new_inode(inode);
4103 return ERR_PTR(ret);
4106 static int ext4_inode_blocks_set(handle_t *handle,
4107 struct ext4_inode *raw_inode,
4108 struct ext4_inode_info *ei)
4110 struct inode *inode = &(ei->vfs_inode);
4111 u64 i_blocks = inode->i_blocks;
4112 struct super_block *sb = inode->i_sb;
4114 if (i_blocks <= ~0U) {
4116 * i_blocks can be represented in a 32 bit variable
4117 * as multiple of 512 bytes
4119 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4120 raw_inode->i_blocks_high = 0;
4121 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4124 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4127 if (i_blocks <= 0xffffffffffffULL) {
4129 * i_blocks can be represented in a 48 bit variable
4130 * as multiple of 512 bytes
4132 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4133 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4134 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4136 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4137 /* i_block is stored in file system block size */
4138 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4139 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4140 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4146 * Post the struct inode info into an on-disk inode location in the
4147 * buffer-cache. This gobbles the caller's reference to the
4148 * buffer_head in the inode location struct.
4150 * The caller must have write access to iloc->bh.
4152 static int ext4_do_update_inode(handle_t *handle,
4153 struct inode *inode,
4154 struct ext4_iloc *iloc)
4156 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4157 struct ext4_inode_info *ei = EXT4_I(inode);
4158 struct buffer_head *bh = iloc->bh;
4159 int err = 0, rc, block;
4160 int need_datasync = 0;
4164 /* For fields not not tracking in the in-memory inode,
4165 * initialise them to zero for new inodes. */
4166 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4167 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4169 ext4_get_inode_flags(ei);
4170 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4171 i_uid = i_uid_read(inode);
4172 i_gid = i_gid_read(inode);
4173 if (!(test_opt(inode->i_sb, NO_UID32))) {
4174 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4175 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4177 * Fix up interoperability with old kernels. Otherwise, old inodes get
4178 * re-used with the upper 16 bits of the uid/gid intact
4181 raw_inode->i_uid_high =
4182 cpu_to_le16(high_16_bits(i_uid));
4183 raw_inode->i_gid_high =
4184 cpu_to_le16(high_16_bits(i_gid));
4186 raw_inode->i_uid_high = 0;
4187 raw_inode->i_gid_high = 0;
4190 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4191 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4192 raw_inode->i_uid_high = 0;
4193 raw_inode->i_gid_high = 0;
4195 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4197 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4198 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4199 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4200 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4202 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4204 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4205 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4206 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4207 cpu_to_le32(EXT4_OS_HURD))
4208 raw_inode->i_file_acl_high =
4209 cpu_to_le16(ei->i_file_acl >> 32);
4210 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4211 if (ei->i_disksize != ext4_isize(raw_inode)) {
4212 ext4_isize_set(raw_inode, ei->i_disksize);
4215 if (ei->i_disksize > 0x7fffffffULL) {
4216 struct super_block *sb = inode->i_sb;
4217 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4218 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4219 EXT4_SB(sb)->s_es->s_rev_level ==
4220 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4221 /* If this is the first large file
4222 * created, add a flag to the superblock.
4224 err = ext4_journal_get_write_access(handle,
4225 EXT4_SB(sb)->s_sbh);
4228 ext4_update_dynamic_rev(sb);
4229 EXT4_SET_RO_COMPAT_FEATURE(sb,
4230 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4231 ext4_handle_sync(handle);
4232 err = ext4_handle_dirty_super(handle, sb);
4235 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4236 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4237 if (old_valid_dev(inode->i_rdev)) {
4238 raw_inode->i_block[0] =
4239 cpu_to_le32(old_encode_dev(inode->i_rdev));
4240 raw_inode->i_block[1] = 0;
4242 raw_inode->i_block[0] = 0;
4243 raw_inode->i_block[1] =
4244 cpu_to_le32(new_encode_dev(inode->i_rdev));
4245 raw_inode->i_block[2] = 0;
4247 } else if (!ext4_has_inline_data(inode)) {
4248 for (block = 0; block < EXT4_N_BLOCKS; block++)
4249 raw_inode->i_block[block] = ei->i_data[block];
4252 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4253 if (ei->i_extra_isize) {
4254 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4255 raw_inode->i_version_hi =
4256 cpu_to_le32(inode->i_version >> 32);
4257 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4260 ext4_inode_csum_set(inode, raw_inode, ei);
4262 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4263 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4266 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4268 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4271 ext4_std_error(inode->i_sb, err);
4276 * ext4_write_inode()
4278 * We are called from a few places:
4280 * - Within generic_file_write() for O_SYNC files.
4281 * Here, there will be no transaction running. We wait for any running
4282 * transaction to commit.
4284 * - Within sys_sync(), kupdate and such.
4285 * We wait on commit, if tol to.
4287 * - Within prune_icache() (PF_MEMALLOC == true)
4288 * Here we simply return. We can't afford to block kswapd on the
4291 * In all cases it is actually safe for us to return without doing anything,
4292 * because the inode has been copied into a raw inode buffer in
4293 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4296 * Note that we are absolutely dependent upon all inode dirtiers doing the
4297 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4298 * which we are interested.
4300 * It would be a bug for them to not do this. The code:
4302 * mark_inode_dirty(inode)
4304 * inode->i_size = expr;
4306 * is in error because a kswapd-driven write_inode() could occur while
4307 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4308 * will no longer be on the superblock's dirty inode list.
4310 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4314 if (current->flags & PF_MEMALLOC)
4317 if (EXT4_SB(inode->i_sb)->s_journal) {
4318 if (ext4_journal_current_handle()) {
4319 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4324 if (wbc->sync_mode != WB_SYNC_ALL)
4327 err = ext4_force_commit(inode->i_sb);
4329 struct ext4_iloc iloc;
4331 err = __ext4_get_inode_loc(inode, &iloc, 0);
4334 if (wbc->sync_mode == WB_SYNC_ALL)
4335 sync_dirty_buffer(iloc.bh);
4336 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4337 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4338 "IO error syncing inode");
4347 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4348 * buffers that are attached to a page stradding i_size and are undergoing
4349 * commit. In that case we have to wait for commit to finish and try again.
4351 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4355 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4356 tid_t commit_tid = 0;
4359 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4361 * All buffers in the last page remain valid? Then there's nothing to
4362 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4365 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4368 page = find_lock_page(inode->i_mapping,
4369 inode->i_size >> PAGE_CACHE_SHIFT);
4372 ret = __ext4_journalled_invalidatepage(page, offset);
4374 page_cache_release(page);
4378 read_lock(&journal->j_state_lock);
4379 if (journal->j_committing_transaction)
4380 commit_tid = journal->j_committing_transaction->t_tid;
4381 read_unlock(&journal->j_state_lock);
4383 jbd2_log_wait_commit(journal, commit_tid);
4390 * Called from notify_change.
4392 * We want to trap VFS attempts to truncate the file as soon as
4393 * possible. In particular, we want to make sure that when the VFS
4394 * shrinks i_size, we put the inode on the orphan list and modify
4395 * i_disksize immediately, so that during the subsequent flushing of
4396 * dirty pages and freeing of disk blocks, we can guarantee that any
4397 * commit will leave the blocks being flushed in an unused state on
4398 * disk. (On recovery, the inode will get truncated and the blocks will
4399 * be freed, so we have a strong guarantee that no future commit will
4400 * leave these blocks visible to the user.)
4402 * Another thing we have to assure is that if we are in ordered mode
4403 * and inode is still attached to the committing transaction, we must
4404 * we start writeout of all the dirty pages which are being truncated.
4405 * This way we are sure that all the data written in the previous
4406 * transaction are already on disk (truncate waits for pages under
4409 * Called with inode->i_mutex down.
4411 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4413 struct inode *inode = dentry->d_inode;
4416 const unsigned int ia_valid = attr->ia_valid;
4418 error = inode_change_ok(inode, attr);
4422 if (is_quota_modification(inode, attr))
4423 dquot_initialize(inode);
4424 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4425 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4428 /* (user+group)*(old+new) structure, inode write (sb,
4429 * inode block, ? - but truncate inode update has it) */
4430 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4431 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4432 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4433 if (IS_ERR(handle)) {
4434 error = PTR_ERR(handle);
4437 error = dquot_transfer(inode, attr);
4439 ext4_journal_stop(handle);
4442 /* Update corresponding info in inode so that everything is in
4443 * one transaction */
4444 if (attr->ia_valid & ATTR_UID)
4445 inode->i_uid = attr->ia_uid;
4446 if (attr->ia_valid & ATTR_GID)
4447 inode->i_gid = attr->ia_gid;
4448 error = ext4_mark_inode_dirty(handle, inode);
4449 ext4_journal_stop(handle);
4452 if (attr->ia_valid & ATTR_SIZE) {
4454 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4455 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4457 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4462 if (S_ISREG(inode->i_mode) &&
4463 attr->ia_valid & ATTR_SIZE &&
4464 (attr->ia_size < inode->i_size)) {
4467 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4468 if (IS_ERR(handle)) {
4469 error = PTR_ERR(handle);
4472 if (ext4_handle_valid(handle)) {
4473 error = ext4_orphan_add(handle, inode);
4476 EXT4_I(inode)->i_disksize = attr->ia_size;
4477 rc = ext4_mark_inode_dirty(handle, inode);
4480 ext4_journal_stop(handle);
4482 if (ext4_should_order_data(inode)) {
4483 error = ext4_begin_ordered_truncate(inode,
4486 /* Do as much error cleanup as possible */
4487 handle = ext4_journal_start(inode,
4489 if (IS_ERR(handle)) {
4490 ext4_orphan_del(NULL, inode);
4493 ext4_orphan_del(handle, inode);
4495 ext4_journal_stop(handle);
4501 if (attr->ia_valid & ATTR_SIZE) {
4502 if (attr->ia_size != inode->i_size) {
4503 loff_t oldsize = inode->i_size;
4505 i_size_write(inode, attr->ia_size);
4507 * Blocks are going to be removed from the inode. Wait
4508 * for dio in flight. Temporarily disable
4509 * dioread_nolock to prevent livelock.
4512 if (!ext4_should_journal_data(inode)) {
4513 ext4_inode_block_unlocked_dio(inode);
4514 inode_dio_wait(inode);
4515 ext4_inode_resume_unlocked_dio(inode);
4517 ext4_wait_for_tail_page_commit(inode);
4520 * Truncate pagecache after we've waited for commit
4521 * in data=journal mode to make pages freeable.
4523 truncate_pagecache(inode, oldsize, inode->i_size);
4525 ext4_truncate(inode);
4529 setattr_copy(inode, attr);
4530 mark_inode_dirty(inode);
4534 * If the call to ext4_truncate failed to get a transaction handle at
4535 * all, we need to clean up the in-core orphan list manually.
4537 if (orphan && inode->i_nlink)
4538 ext4_orphan_del(NULL, inode);
4540 if (!rc && (ia_valid & ATTR_MODE))
4541 rc = ext4_acl_chmod(inode);
4544 ext4_std_error(inode->i_sb, error);
4550 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4553 struct inode *inode;
4554 unsigned long delalloc_blocks;
4556 inode = dentry->d_inode;
4557 generic_fillattr(inode, stat);
4560 * We can't update i_blocks if the block allocation is delayed
4561 * otherwise in the case of system crash before the real block
4562 * allocation is done, we will have i_blocks inconsistent with
4563 * on-disk file blocks.
4564 * We always keep i_blocks updated together with real
4565 * allocation. But to not confuse with user, stat
4566 * will return the blocks that include the delayed allocation
4567 * blocks for this file.
4569 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4570 EXT4_I(inode)->i_reserved_data_blocks);
4572 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4576 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4578 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4579 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4580 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4584 * Account for index blocks, block groups bitmaps and block group
4585 * descriptor blocks if modify datablocks and index blocks
4586 * worse case, the indexs blocks spread over different block groups
4588 * If datablocks are discontiguous, they are possible to spread over
4589 * different block groups too. If they are contiguous, with flexbg,
4590 * they could still across block group boundary.
4592 * Also account for superblock, inode, quota and xattr blocks
4594 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4596 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4602 * How many index blocks need to touch to modify nrblocks?
4603 * The "Chunk" flag indicating whether the nrblocks is
4604 * physically contiguous on disk
4606 * For Direct IO and fallocate, they calls get_block to allocate
4607 * one single extent at a time, so they could set the "Chunk" flag
4609 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4614 * Now let's see how many group bitmaps and group descriptors need
4624 if (groups > ngroups)
4626 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4627 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4629 /* bitmaps and block group descriptor blocks */
4630 ret += groups + gdpblocks;
4632 /* Blocks for super block, inode, quota and xattr blocks */
4633 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4639 * Calculate the total number of credits to reserve to fit
4640 * the modification of a single pages into a single transaction,
4641 * which may include multiple chunks of block allocations.
4643 * This could be called via ext4_write_begin()
4645 * We need to consider the worse case, when
4646 * one new block per extent.
4648 int ext4_writepage_trans_blocks(struct inode *inode)
4650 int bpp = ext4_journal_blocks_per_page(inode);
4653 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4655 /* Account for data blocks for journalled mode */
4656 if (ext4_should_journal_data(inode))
4662 * Calculate the journal credits for a chunk of data modification.
4664 * This is called from DIO, fallocate or whoever calling
4665 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4667 * journal buffers for data blocks are not included here, as DIO
4668 * and fallocate do no need to journal data buffers.
4670 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4672 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4676 * The caller must have previously called ext4_reserve_inode_write().
4677 * Give this, we know that the caller already has write access to iloc->bh.
4679 int ext4_mark_iloc_dirty(handle_t *handle,
4680 struct inode *inode, struct ext4_iloc *iloc)
4684 if (IS_I_VERSION(inode))
4685 inode_inc_iversion(inode);
4687 /* the do_update_inode consumes one bh->b_count */
4690 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4691 err = ext4_do_update_inode(handle, inode, iloc);
4697 * On success, We end up with an outstanding reference count against
4698 * iloc->bh. This _must_ be cleaned up later.
4702 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4703 struct ext4_iloc *iloc)
4707 err = ext4_get_inode_loc(inode, iloc);
4709 BUFFER_TRACE(iloc->bh, "get_write_access");
4710 err = ext4_journal_get_write_access(handle, iloc->bh);
4716 ext4_std_error(inode->i_sb, err);
4721 * Expand an inode by new_extra_isize bytes.
4722 * Returns 0 on success or negative error number on failure.
4724 static int ext4_expand_extra_isize(struct inode *inode,
4725 unsigned int new_extra_isize,
4726 struct ext4_iloc iloc,
4729 struct ext4_inode *raw_inode;
4730 struct ext4_xattr_ibody_header *header;
4732 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4735 raw_inode = ext4_raw_inode(&iloc);
4737 header = IHDR(inode, raw_inode);
4739 /* No extended attributes present */
4740 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4741 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4742 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4744 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4748 /* try to expand with EAs present */
4749 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4754 * What we do here is to mark the in-core inode as clean with respect to inode
4755 * dirtiness (it may still be data-dirty).
4756 * This means that the in-core inode may be reaped by prune_icache
4757 * without having to perform any I/O. This is a very good thing,
4758 * because *any* task may call prune_icache - even ones which
4759 * have a transaction open against a different journal.
4761 * Is this cheating? Not really. Sure, we haven't written the
4762 * inode out, but prune_icache isn't a user-visible syncing function.
4763 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4764 * we start and wait on commits.
4766 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4768 struct ext4_iloc iloc;
4769 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4770 static unsigned int mnt_count;
4774 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4775 err = ext4_reserve_inode_write(handle, inode, &iloc);
4776 if (ext4_handle_valid(handle) &&
4777 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4778 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4780 * We need extra buffer credits since we may write into EA block
4781 * with this same handle. If journal_extend fails, then it will
4782 * only result in a minor loss of functionality for that inode.
4783 * If this is felt to be critical, then e2fsck should be run to
4784 * force a large enough s_min_extra_isize.
4786 if ((jbd2_journal_extend(handle,
4787 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4788 ret = ext4_expand_extra_isize(inode,
4789 sbi->s_want_extra_isize,
4792 ext4_set_inode_state(inode,
4793 EXT4_STATE_NO_EXPAND);
4795 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4796 ext4_warning(inode->i_sb,
4797 "Unable to expand inode %lu. Delete"
4798 " some EAs or run e2fsck.",
4801 le16_to_cpu(sbi->s_es->s_mnt_count);
4807 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4812 * ext4_dirty_inode() is called from __mark_inode_dirty()
4814 * We're really interested in the case where a file is being extended.
4815 * i_size has been changed by generic_commit_write() and we thus need
4816 * to include the updated inode in the current transaction.
4818 * Also, dquot_alloc_block() will always dirty the inode when blocks
4819 * are allocated to the file.
4821 * If the inode is marked synchronous, we don't honour that here - doing
4822 * so would cause a commit on atime updates, which we don't bother doing.
4823 * We handle synchronous inodes at the highest possible level.
4825 void ext4_dirty_inode(struct inode *inode, int flags)
4829 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
4833 ext4_mark_inode_dirty(handle, inode);
4835 ext4_journal_stop(handle);
4842 * Bind an inode's backing buffer_head into this transaction, to prevent
4843 * it from being flushed to disk early. Unlike
4844 * ext4_reserve_inode_write, this leaves behind no bh reference and
4845 * returns no iloc structure, so the caller needs to repeat the iloc
4846 * lookup to mark the inode dirty later.
4848 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4850 struct ext4_iloc iloc;
4854 err = ext4_get_inode_loc(inode, &iloc);
4856 BUFFER_TRACE(iloc.bh, "get_write_access");
4857 err = jbd2_journal_get_write_access(handle, iloc.bh);
4859 err = ext4_handle_dirty_metadata(handle,
4865 ext4_std_error(inode->i_sb, err);
4870 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4877 * We have to be very careful here: changing a data block's
4878 * journaling status dynamically is dangerous. If we write a
4879 * data block to the journal, change the status and then delete
4880 * that block, we risk forgetting to revoke the old log record
4881 * from the journal and so a subsequent replay can corrupt data.
4882 * So, first we make sure that the journal is empty and that
4883 * nobody is changing anything.
4886 journal = EXT4_JOURNAL(inode);
4889 if (is_journal_aborted(journal))
4891 /* We have to allocate physical blocks for delalloc blocks
4892 * before flushing journal. otherwise delalloc blocks can not
4893 * be allocated any more. even more truncate on delalloc blocks
4894 * could trigger BUG by flushing delalloc blocks in journal.
4895 * There is no delalloc block in non-journal data mode.
4897 if (val && test_opt(inode->i_sb, DELALLOC)) {
4898 err = ext4_alloc_da_blocks(inode);
4903 /* Wait for all existing dio workers */
4904 ext4_inode_block_unlocked_dio(inode);
4905 inode_dio_wait(inode);
4907 jbd2_journal_lock_updates(journal);
4910 * OK, there are no updates running now, and all cached data is
4911 * synced to disk. We are now in a completely consistent state
4912 * which doesn't have anything in the journal, and we know that
4913 * no filesystem updates are running, so it is safe to modify
4914 * the inode's in-core data-journaling state flag now.
4918 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4920 jbd2_journal_flush(journal);
4921 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4923 ext4_set_aops(inode);
4925 jbd2_journal_unlock_updates(journal);
4926 ext4_inode_resume_unlocked_dio(inode);
4928 /* Finally we can mark the inode as dirty. */
4930 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
4932 return PTR_ERR(handle);
4934 err = ext4_mark_inode_dirty(handle, inode);
4935 ext4_handle_sync(handle);
4936 ext4_journal_stop(handle);
4937 ext4_std_error(inode->i_sb, err);
4942 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4944 return !buffer_mapped(bh);
4947 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4949 struct page *page = vmf->page;
4953 struct file *file = vma->vm_file;
4954 struct inode *inode = file->f_path.dentry->d_inode;
4955 struct address_space *mapping = inode->i_mapping;
4957 get_block_t *get_block;
4960 sb_start_pagefault(inode->i_sb);
4961 file_update_time(vma->vm_file);
4962 /* Delalloc case is easy... */
4963 if (test_opt(inode->i_sb, DELALLOC) &&
4964 !ext4_should_journal_data(inode) &&
4965 !ext4_nonda_switch(inode->i_sb)) {
4967 ret = __block_page_mkwrite(vma, vmf,
4968 ext4_da_get_block_prep);
4969 } while (ret == -ENOSPC &&
4970 ext4_should_retry_alloc(inode->i_sb, &retries));
4975 size = i_size_read(inode);
4976 /* Page got truncated from under us? */
4977 if (page->mapping != mapping || page_offset(page) > size) {
4979 ret = VM_FAULT_NOPAGE;
4983 if (page->index == size >> PAGE_CACHE_SHIFT)
4984 len = size & ~PAGE_CACHE_MASK;
4986 len = PAGE_CACHE_SIZE;
4988 * Return if we have all the buffers mapped. This avoids the need to do
4989 * journal_start/journal_stop which can block and take a long time
4991 if (page_has_buffers(page)) {
4992 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
4994 ext4_bh_unmapped)) {
4995 /* Wait so that we don't change page under IO */
4996 wait_on_page_writeback(page);
4997 ret = VM_FAULT_LOCKED;
5002 /* OK, we need to fill the hole... */
5003 if (ext4_should_dioread_nolock(inode))
5004 get_block = ext4_get_block_write;
5006 get_block = ext4_get_block;
5008 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5009 ext4_writepage_trans_blocks(inode));
5010 if (IS_ERR(handle)) {
5011 ret = VM_FAULT_SIGBUS;
5014 ret = __block_page_mkwrite(vma, vmf, get_block);
5015 if (!ret && ext4_should_journal_data(inode)) {
5016 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5017 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5019 ret = VM_FAULT_SIGBUS;
5020 ext4_journal_stop(handle);
5023 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5025 ext4_journal_stop(handle);
5026 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5029 ret = block_page_mkwrite_return(ret);
5031 sb_end_pagefault(inode->i_sb);