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 if (inode->i_nlink) {
190 * When journalling data dirty buffers are tracked only in the
191 * journal. So although mm thinks everything is clean and
192 * ready for reaping the inode might still have some pages to
193 * write in the running transaction or waiting to be
194 * checkpointed. Thus calling jbd2_journal_invalidatepage()
195 * (via truncate_inode_pages()) to discard these buffers can
196 * cause data loss. Also even if we did not discard these
197 * buffers, we would have no way to find them after the inode
198 * is reaped and thus user could see stale data if he tries to
199 * read them before the transaction is checkpointed. So be
200 * careful and force everything to disk here... We use
201 * ei->i_datasync_tid to store the newest transaction
202 * containing inode's data.
204 * Note that directories do not have this problem because they
205 * don't use page cache.
207 if (ext4_should_journal_data(inode) &&
208 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
209 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
210 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
212 jbd2_log_start_commit(journal, commit_tid);
213 jbd2_log_wait_commit(journal, commit_tid);
214 filemap_write_and_wait(&inode->i_data);
216 truncate_inode_pages(&inode->i_data, 0);
217 ext4_ioend_shutdown(inode);
221 if (!is_bad_inode(inode))
222 dquot_initialize(inode);
224 if (ext4_should_order_data(inode))
225 ext4_begin_ordered_truncate(inode, 0);
226 truncate_inode_pages(&inode->i_data, 0);
227 ext4_ioend_shutdown(inode);
229 if (is_bad_inode(inode))
233 * Protect us against freezing - iput() caller didn't have to have any
234 * protection against it
236 sb_start_intwrite(inode->i_sb);
237 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
238 ext4_blocks_for_truncate(inode)+3);
239 if (IS_ERR(handle)) {
240 ext4_std_error(inode->i_sb, PTR_ERR(handle));
242 * If we're going to skip the normal cleanup, we still need to
243 * make sure that the in-core orphan linked list is properly
246 ext4_orphan_del(NULL, inode);
247 sb_end_intwrite(inode->i_sb);
252 ext4_handle_sync(handle);
254 err = ext4_mark_inode_dirty(handle, inode);
256 ext4_warning(inode->i_sb,
257 "couldn't mark inode dirty (err %d)", err);
261 ext4_truncate(inode);
264 * ext4_ext_truncate() doesn't reserve any slop when it
265 * restarts journal transactions; therefore there may not be
266 * enough credits left in the handle to remove the inode from
267 * the orphan list and set the dtime field.
269 if (!ext4_handle_has_enough_credits(handle, 3)) {
270 err = ext4_journal_extend(handle, 3);
272 err = ext4_journal_restart(handle, 3);
274 ext4_warning(inode->i_sb,
275 "couldn't extend journal (err %d)", err);
277 ext4_journal_stop(handle);
278 ext4_orphan_del(NULL, inode);
279 sb_end_intwrite(inode->i_sb);
285 * Kill off the orphan record which ext4_truncate created.
286 * AKPM: I think this can be inside the above `if'.
287 * Note that ext4_orphan_del() has to be able to cope with the
288 * deletion of a non-existent orphan - this is because we don't
289 * know if ext4_truncate() actually created an orphan record.
290 * (Well, we could do this if we need to, but heck - it works)
292 ext4_orphan_del(handle, inode);
293 EXT4_I(inode)->i_dtime = get_seconds();
296 * One subtle ordering requirement: if anything has gone wrong
297 * (transaction abort, IO errors, whatever), then we can still
298 * do these next steps (the fs will already have been marked as
299 * having errors), but we can't free the inode if the mark_dirty
302 if (ext4_mark_inode_dirty(handle, inode))
303 /* If that failed, just do the required in-core inode clear. */
304 ext4_clear_inode(inode);
306 ext4_free_inode(handle, inode);
307 ext4_journal_stop(handle);
308 sb_end_intwrite(inode->i_sb);
311 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
315 qsize_t *ext4_get_reserved_space(struct inode *inode)
317 return &EXT4_I(inode)->i_reserved_quota;
322 * Calculate the number of metadata blocks need to reserve
323 * to allocate a block located at @lblock
325 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
327 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
328 return ext4_ext_calc_metadata_amount(inode, lblock);
330 return ext4_ind_calc_metadata_amount(inode, lblock);
334 * Called with i_data_sem down, which is important since we can call
335 * ext4_discard_preallocations() from here.
337 void ext4_da_update_reserve_space(struct inode *inode,
338 int used, int quota_claim)
340 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
341 struct ext4_inode_info *ei = EXT4_I(inode);
343 spin_lock(&ei->i_block_reservation_lock);
344 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
345 if (unlikely(used > ei->i_reserved_data_blocks)) {
346 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
347 "with only %d reserved data blocks",
348 __func__, inode->i_ino, used,
349 ei->i_reserved_data_blocks);
351 used = ei->i_reserved_data_blocks;
354 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
355 ext4_warning(inode->i_sb, "ino %lu, allocated %d "
356 "with only %d reserved metadata blocks "
357 "(releasing %d blocks with reserved %d data blocks)",
358 inode->i_ino, ei->i_allocated_meta_blocks,
359 ei->i_reserved_meta_blocks, used,
360 ei->i_reserved_data_blocks);
362 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
365 /* Update per-inode reservations */
366 ei->i_reserved_data_blocks -= used;
367 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
368 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
369 used + ei->i_allocated_meta_blocks);
370 ei->i_allocated_meta_blocks = 0;
372 if (ei->i_reserved_data_blocks == 0) {
374 * We can release all of the reserved metadata blocks
375 * only when we have written all of the delayed
378 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
379 ei->i_reserved_meta_blocks);
380 ei->i_reserved_meta_blocks = 0;
381 ei->i_da_metadata_calc_len = 0;
383 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
385 /* Update quota subsystem for data blocks */
387 dquot_claim_block(inode, EXT4_C2B(sbi, used));
390 * We did fallocate with an offset that is already delayed
391 * allocated. So on delayed allocated writeback we should
392 * not re-claim the quota for fallocated blocks.
394 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
398 * If we have done all the pending block allocations and if
399 * there aren't any writers on the inode, we can discard the
400 * inode's preallocations.
402 if ((ei->i_reserved_data_blocks == 0) &&
403 (atomic_read(&inode->i_writecount) == 0))
404 ext4_discard_preallocations(inode);
407 static int __check_block_validity(struct inode *inode, const char *func,
409 struct ext4_map_blocks *map)
411 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
413 ext4_error_inode(inode, func, line, map->m_pblk,
414 "lblock %lu mapped to illegal pblock "
415 "(length %d)", (unsigned long) map->m_lblk,
422 #define check_block_validity(inode, map) \
423 __check_block_validity((inode), __func__, __LINE__, (map))
426 * Return the number of contiguous dirty pages in a given inode
427 * starting at page frame idx.
429 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
430 unsigned int max_pages)
432 struct address_space *mapping = inode->i_mapping;
436 int i, nr_pages, done = 0;
440 pagevec_init(&pvec, 0);
443 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
445 (pgoff_t)PAGEVEC_SIZE);
448 for (i = 0; i < nr_pages; i++) {
449 struct page *page = pvec.pages[i];
450 struct buffer_head *bh, *head;
453 if (unlikely(page->mapping != mapping) ||
455 PageWriteback(page) ||
456 page->index != idx) {
461 if (page_has_buffers(page)) {
462 bh = head = page_buffers(page);
464 if (!buffer_delay(bh) &&
465 !buffer_unwritten(bh))
467 bh = bh->b_this_page;
468 } while (!done && (bh != head));
475 if (num >= max_pages) {
480 pagevec_release(&pvec);
485 #ifdef ES_AGGRESSIVE_TEST
486 static void ext4_map_blocks_es_recheck(handle_t *handle,
488 struct ext4_map_blocks *es_map,
489 struct ext4_map_blocks *map,
496 * There is a race window that the result is not the same.
497 * e.g. xfstests #223 when dioread_nolock enables. The reason
498 * is that we lookup a block mapping in extent status tree with
499 * out taking i_data_sem. So at the time the unwritten extent
500 * could be converted.
502 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
503 down_read((&EXT4_I(inode)->i_data_sem));
504 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
505 retval = ext4_ext_map_blocks(handle, inode, map, flags &
506 EXT4_GET_BLOCKS_KEEP_SIZE);
508 retval = ext4_ind_map_blocks(handle, inode, map, flags &
509 EXT4_GET_BLOCKS_KEEP_SIZE);
511 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
512 up_read((&EXT4_I(inode)->i_data_sem));
514 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
515 * because it shouldn't be marked in es_map->m_flags.
517 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
520 * We don't check m_len because extent will be collpased in status
521 * tree. So the m_len might not equal.
523 if (es_map->m_lblk != map->m_lblk ||
524 es_map->m_flags != map->m_flags ||
525 es_map->m_pblk != map->m_pblk) {
526 printk("ES cache assertation failed for inode: %lu "
527 "es_cached ex [%d/%d/%llu/%x] != "
528 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
529 inode->i_ino, es_map->m_lblk, es_map->m_len,
530 es_map->m_pblk, es_map->m_flags, map->m_lblk,
531 map->m_len, map->m_pblk, map->m_flags,
535 #endif /* ES_AGGRESSIVE_TEST */
538 * The ext4_map_blocks() function tries to look up the requested blocks,
539 * and returns if the blocks are already mapped.
541 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
542 * and store the allocated blocks in the result buffer head and mark it
545 * If file type is extents based, it will call ext4_ext_map_blocks(),
546 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
549 * On success, it returns the number of blocks being mapped or allocate.
550 * if create==0 and the blocks are pre-allocated and uninitialized block,
551 * the result buffer head is unmapped. If the create ==1, it will make sure
552 * the buffer head is mapped.
554 * It returns 0 if plain look up failed (blocks have not been allocated), in
555 * that case, buffer head is unmapped
557 * It returns the error in case of allocation failure.
559 int ext4_map_blocks(handle_t *handle, struct inode *inode,
560 struct ext4_map_blocks *map, int flags)
562 struct extent_status es;
564 #ifdef ES_AGGRESSIVE_TEST
565 struct ext4_map_blocks orig_map;
567 memcpy(&orig_map, map, sizeof(*map));
571 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
572 "logical block %lu\n", inode->i_ino, flags, map->m_len,
573 (unsigned long) map->m_lblk);
575 /* Lookup extent status tree firstly */
576 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
577 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
578 map->m_pblk = ext4_es_pblock(&es) +
579 map->m_lblk - es.es_lblk;
580 map->m_flags |= ext4_es_is_written(&es) ?
581 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
582 retval = es.es_len - (map->m_lblk - es.es_lblk);
583 if (retval > map->m_len)
586 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
591 #ifdef ES_AGGRESSIVE_TEST
592 ext4_map_blocks_es_recheck(handle, inode, map,
599 * Try to see if we can get the block without requesting a new
602 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
603 down_read((&EXT4_I(inode)->i_data_sem));
604 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
605 retval = ext4_ext_map_blocks(handle, inode, map, flags &
606 EXT4_GET_BLOCKS_KEEP_SIZE);
608 retval = ext4_ind_map_blocks(handle, inode, map, flags &
609 EXT4_GET_BLOCKS_KEEP_SIZE);
613 unsigned long long status;
615 #ifdef ES_AGGRESSIVE_TEST
616 if (retval != map->m_len) {
617 printk("ES len assertation failed for inode: %lu "
618 "retval %d != map->m_len %d "
619 "in %s (lookup)\n", inode->i_ino, retval,
620 map->m_len, __func__);
624 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
625 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
626 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
627 ext4_find_delalloc_range(inode, map->m_lblk,
628 map->m_lblk + map->m_len - 1))
629 status |= EXTENT_STATUS_DELAYED;
630 ret = ext4_es_insert_extent(inode, map->m_lblk,
631 map->m_len, map->m_pblk, status);
635 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
636 up_read((&EXT4_I(inode)->i_data_sem));
639 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
640 int ret = check_block_validity(inode, map);
645 /* If it is only a block(s) look up */
646 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
650 * Returns if the blocks have already allocated
652 * Note that if blocks have been preallocated
653 * ext4_ext_get_block() returns the create = 0
654 * with buffer head unmapped.
656 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
660 * Here we clear m_flags because after allocating an new extent,
661 * it will be set again.
663 map->m_flags &= ~EXT4_MAP_FLAGS;
666 * New blocks allocate and/or writing to uninitialized extent
667 * will possibly result in updating i_data, so we take
668 * the write lock of i_data_sem, and call get_blocks()
669 * with create == 1 flag.
671 down_write((&EXT4_I(inode)->i_data_sem));
674 * if the caller is from delayed allocation writeout path
675 * we have already reserved fs blocks for allocation
676 * let the underlying get_block() function know to
677 * avoid double accounting
679 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
680 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
682 * We need to check for EXT4 here because migrate
683 * could have changed the inode type in between
685 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
686 retval = ext4_ext_map_blocks(handle, inode, map, flags);
688 retval = ext4_ind_map_blocks(handle, inode, map, flags);
690 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
692 * We allocated new blocks which will result in
693 * i_data's format changing. Force the migrate
694 * to fail by clearing migrate flags
696 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
700 * Update reserved blocks/metadata blocks after successful
701 * block allocation which had been deferred till now. We don't
702 * support fallocate for non extent files. So we can update
703 * reserve space here.
706 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
707 ext4_da_update_reserve_space(inode, retval, 1);
709 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
710 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
714 unsigned long long status;
716 #ifdef ES_AGGRESSIVE_TEST
717 if (retval != map->m_len) {
718 printk("ES len assertation failed for inode: %lu "
719 "retval %d != map->m_len %d "
720 "in %s (allocation)\n", inode->i_ino, retval,
721 map->m_len, __func__);
726 * If the extent has been zeroed out, we don't need to update
727 * extent status tree.
729 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
730 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
731 if (ext4_es_is_written(&es))
734 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
735 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
736 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
737 ext4_find_delalloc_range(inode, map->m_lblk,
738 map->m_lblk + map->m_len - 1))
739 status |= EXTENT_STATUS_DELAYED;
740 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
741 map->m_pblk, status);
747 up_write((&EXT4_I(inode)->i_data_sem));
748 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
749 int ret = check_block_validity(inode, map);
756 /* Maximum number of blocks we map for direct IO at once. */
757 #define DIO_MAX_BLOCKS 4096
759 static int _ext4_get_block(struct inode *inode, sector_t iblock,
760 struct buffer_head *bh, int flags)
762 handle_t *handle = ext4_journal_current_handle();
763 struct ext4_map_blocks map;
764 int ret = 0, started = 0;
767 if (ext4_has_inline_data(inode))
771 map.m_len = bh->b_size >> inode->i_blkbits;
773 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
774 /* Direct IO write... */
775 if (map.m_len > DIO_MAX_BLOCKS)
776 map.m_len = DIO_MAX_BLOCKS;
777 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
778 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
780 if (IS_ERR(handle)) {
781 ret = PTR_ERR(handle);
787 ret = ext4_map_blocks(handle, inode, &map, flags);
789 map_bh(bh, inode->i_sb, map.m_pblk);
790 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
791 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
795 ext4_journal_stop(handle);
799 int ext4_get_block(struct inode *inode, sector_t iblock,
800 struct buffer_head *bh, int create)
802 return _ext4_get_block(inode, iblock, bh,
803 create ? EXT4_GET_BLOCKS_CREATE : 0);
807 * `handle' can be NULL if create is zero
809 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
810 ext4_lblk_t block, int create, int *errp)
812 struct ext4_map_blocks map;
813 struct buffer_head *bh;
816 J_ASSERT(handle != NULL || create == 0);
820 err = ext4_map_blocks(handle, inode, &map,
821 create ? EXT4_GET_BLOCKS_CREATE : 0);
823 /* ensure we send some value back into *errp */
826 if (create && err == 0)
827 err = -ENOSPC; /* should never happen */
833 bh = sb_getblk(inode->i_sb, map.m_pblk);
838 if (map.m_flags & EXT4_MAP_NEW) {
839 J_ASSERT(create != 0);
840 J_ASSERT(handle != NULL);
843 * Now that we do not always journal data, we should
844 * keep in mind whether this should always journal the
845 * new buffer as metadata. For now, regular file
846 * writes use ext4_get_block instead, so it's not a
850 BUFFER_TRACE(bh, "call get_create_access");
851 fatal = ext4_journal_get_create_access(handle, bh);
852 if (!fatal && !buffer_uptodate(bh)) {
853 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
854 set_buffer_uptodate(bh);
857 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
858 err = ext4_handle_dirty_metadata(handle, inode, bh);
862 BUFFER_TRACE(bh, "not a new buffer");
872 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
873 ext4_lblk_t block, int create, int *err)
875 struct buffer_head *bh;
877 bh = ext4_getblk(handle, inode, block, create, err);
880 if (buffer_uptodate(bh))
882 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
884 if (buffer_uptodate(bh))
891 int ext4_walk_page_buffers(handle_t *handle,
892 struct buffer_head *head,
896 int (*fn)(handle_t *handle,
897 struct buffer_head *bh))
899 struct buffer_head *bh;
900 unsigned block_start, block_end;
901 unsigned blocksize = head->b_size;
903 struct buffer_head *next;
905 for (bh = head, block_start = 0;
906 ret == 0 && (bh != head || !block_start);
907 block_start = block_end, bh = next) {
908 next = bh->b_this_page;
909 block_end = block_start + blocksize;
910 if (block_end <= from || block_start >= to) {
911 if (partial && !buffer_uptodate(bh))
915 err = (*fn)(handle, bh);
923 * To preserve ordering, it is essential that the hole instantiation and
924 * the data write be encapsulated in a single transaction. We cannot
925 * close off a transaction and start a new one between the ext4_get_block()
926 * and the commit_write(). So doing the jbd2_journal_start at the start of
927 * prepare_write() is the right place.
929 * Also, this function can nest inside ext4_writepage(). In that case, we
930 * *know* that ext4_writepage() has generated enough buffer credits to do the
931 * whole page. So we won't block on the journal in that case, which is good,
932 * because the caller may be PF_MEMALLOC.
934 * By accident, ext4 can be reentered when a transaction is open via
935 * quota file writes. If we were to commit the transaction while thus
936 * reentered, there can be a deadlock - we would be holding a quota
937 * lock, and the commit would never complete if another thread had a
938 * transaction open and was blocking on the quota lock - a ranking
941 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
942 * will _not_ run commit under these circumstances because handle->h_ref
943 * is elevated. We'll still have enough credits for the tiny quotafile
946 int do_journal_get_write_access(handle_t *handle,
947 struct buffer_head *bh)
949 int dirty = buffer_dirty(bh);
952 if (!buffer_mapped(bh) || buffer_freed(bh))
955 * __block_write_begin() could have dirtied some buffers. Clean
956 * the dirty bit as jbd2_journal_get_write_access() could complain
957 * otherwise about fs integrity issues. Setting of the dirty bit
958 * by __block_write_begin() isn't a real problem here as we clear
959 * the bit before releasing a page lock and thus writeback cannot
960 * ever write the buffer.
963 clear_buffer_dirty(bh);
964 ret = ext4_journal_get_write_access(handle, bh);
966 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
970 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
971 struct buffer_head *bh_result, int create);
972 static int ext4_write_begin(struct file *file, struct address_space *mapping,
973 loff_t pos, unsigned len, unsigned flags,
974 struct page **pagep, void **fsdata)
976 struct inode *inode = mapping->host;
977 int ret, needed_blocks;
984 trace_ext4_write_begin(inode, pos, len, flags);
986 * Reserve one block more for addition to orphan list in case
987 * we allocate blocks but write fails for some reason
989 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
990 index = pos >> PAGE_CACHE_SHIFT;
991 from = pos & (PAGE_CACHE_SIZE - 1);
994 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
995 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1004 * grab_cache_page_write_begin() can take a long time if the
1005 * system is thrashing due to memory pressure, or if the page
1006 * is being written back. So grab it first before we start
1007 * the transaction handle. This also allows us to allocate
1008 * the page (if needed) without using GFP_NOFS.
1011 page = grab_cache_page_write_begin(mapping, index, flags);
1017 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1018 if (IS_ERR(handle)) {
1019 page_cache_release(page);
1020 return PTR_ERR(handle);
1024 if (page->mapping != mapping) {
1025 /* The page got truncated from under us */
1027 page_cache_release(page);
1028 ext4_journal_stop(handle);
1031 wait_on_page_writeback(page);
1033 if (ext4_should_dioread_nolock(inode))
1034 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1036 ret = __block_write_begin(page, pos, len, ext4_get_block);
1038 if (!ret && ext4_should_journal_data(inode)) {
1039 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1041 do_journal_get_write_access);
1047 * __block_write_begin may have instantiated a few blocks
1048 * outside i_size. Trim these off again. Don't need
1049 * i_size_read because we hold i_mutex.
1051 * Add inode to orphan list in case we crash before
1054 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1055 ext4_orphan_add(handle, inode);
1057 ext4_journal_stop(handle);
1058 if (pos + len > inode->i_size) {
1059 ext4_truncate_failed_write(inode);
1061 * If truncate failed early the inode might
1062 * still be on the orphan list; we need to
1063 * make sure the inode is removed from the
1064 * orphan list in that case.
1067 ext4_orphan_del(NULL, inode);
1070 if (ret == -ENOSPC &&
1071 ext4_should_retry_alloc(inode->i_sb, &retries))
1073 page_cache_release(page);
1080 /* For write_end() in data=journal mode */
1081 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1083 if (!buffer_mapped(bh) || buffer_freed(bh))
1085 set_buffer_uptodate(bh);
1086 return ext4_handle_dirty_metadata(handle, NULL, bh);
1089 static int ext4_generic_write_end(struct file *file,
1090 struct address_space *mapping,
1091 loff_t pos, unsigned len, unsigned copied,
1092 struct page *page, void *fsdata)
1094 int i_size_changed = 0;
1095 struct inode *inode = mapping->host;
1096 handle_t *handle = ext4_journal_current_handle();
1098 if (ext4_has_inline_data(inode))
1099 copied = ext4_write_inline_data_end(inode, pos, len,
1102 copied = block_write_end(file, mapping, pos,
1103 len, copied, page, fsdata);
1106 * No need to use i_size_read() here, the i_size
1107 * cannot change under us because we hold i_mutex.
1109 * But it's important to update i_size while still holding page lock:
1110 * page writeout could otherwise come in and zero beyond i_size.
1112 if (pos + copied > inode->i_size) {
1113 i_size_write(inode, pos + copied);
1117 if (pos + copied > EXT4_I(inode)->i_disksize) {
1118 /* We need to mark inode dirty even if
1119 * new_i_size is less that inode->i_size
1120 * bu greater than i_disksize.(hint delalloc)
1122 ext4_update_i_disksize(inode, (pos + copied));
1126 page_cache_release(page);
1129 * Don't mark the inode dirty under page lock. First, it unnecessarily
1130 * makes the holding time of page lock longer. Second, it forces lock
1131 * ordering of page lock and transaction start for journaling
1135 ext4_mark_inode_dirty(handle, inode);
1141 * We need to pick up the new inode size which generic_commit_write gave us
1142 * `file' can be NULL - eg, when called from page_symlink().
1144 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1145 * buffers are managed internally.
1147 static int ext4_ordered_write_end(struct file *file,
1148 struct address_space *mapping,
1149 loff_t pos, unsigned len, unsigned copied,
1150 struct page *page, void *fsdata)
1152 handle_t *handle = ext4_journal_current_handle();
1153 struct inode *inode = mapping->host;
1156 trace_ext4_ordered_write_end(inode, pos, len, copied);
1157 ret = ext4_jbd2_file_inode(handle, inode);
1160 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1163 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1164 /* if we have allocated more blocks and copied
1165 * less. We will have blocks allocated outside
1166 * inode->i_size. So truncate them
1168 ext4_orphan_add(handle, inode);
1173 page_cache_release(page);
1176 ret2 = ext4_journal_stop(handle);
1180 if (pos + len > inode->i_size) {
1181 ext4_truncate_failed_write(inode);
1183 * If truncate failed early the inode might still be
1184 * on the orphan list; we need to make sure the inode
1185 * is removed from the orphan list in that case.
1188 ext4_orphan_del(NULL, inode);
1192 return ret ? ret : copied;
1195 static int ext4_writeback_write_end(struct file *file,
1196 struct address_space *mapping,
1197 loff_t pos, unsigned len, unsigned copied,
1198 struct page *page, void *fsdata)
1200 handle_t *handle = ext4_journal_current_handle();
1201 struct inode *inode = mapping->host;
1204 trace_ext4_writeback_write_end(inode, pos, len, copied);
1205 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1208 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1209 /* if we have allocated more blocks and copied
1210 * less. We will have blocks allocated outside
1211 * inode->i_size. So truncate them
1213 ext4_orphan_add(handle, inode);
1218 ret2 = ext4_journal_stop(handle);
1222 if (pos + len > inode->i_size) {
1223 ext4_truncate_failed_write(inode);
1225 * If truncate failed early the inode might still be
1226 * on the orphan list; we need to make sure the inode
1227 * is removed from the orphan list in that case.
1230 ext4_orphan_del(NULL, inode);
1233 return ret ? ret : copied;
1236 static int ext4_journalled_write_end(struct file *file,
1237 struct address_space *mapping,
1238 loff_t pos, unsigned len, unsigned copied,
1239 struct page *page, void *fsdata)
1241 handle_t *handle = ext4_journal_current_handle();
1242 struct inode *inode = mapping->host;
1248 trace_ext4_journalled_write_end(inode, pos, len, copied);
1249 from = pos & (PAGE_CACHE_SIZE - 1);
1252 BUG_ON(!ext4_handle_valid(handle));
1254 if (ext4_has_inline_data(inode))
1255 copied = ext4_write_inline_data_end(inode, pos, len,
1259 if (!PageUptodate(page))
1261 page_zero_new_buffers(page, from+copied, to);
1264 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1265 to, &partial, write_end_fn);
1267 SetPageUptodate(page);
1269 new_i_size = pos + copied;
1270 if (new_i_size > inode->i_size)
1271 i_size_write(inode, pos+copied);
1272 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1273 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1274 if (new_i_size > EXT4_I(inode)->i_disksize) {
1275 ext4_update_i_disksize(inode, new_i_size);
1276 ret2 = ext4_mark_inode_dirty(handle, inode);
1282 page_cache_release(page);
1283 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1284 /* if we have allocated more blocks and copied
1285 * less. We will have blocks allocated outside
1286 * inode->i_size. So truncate them
1288 ext4_orphan_add(handle, inode);
1290 ret2 = ext4_journal_stop(handle);
1293 if (pos + len > inode->i_size) {
1294 ext4_truncate_failed_write(inode);
1296 * If truncate failed early the inode might still be
1297 * on the orphan list; we need to make sure the inode
1298 * is removed from the orphan list in that case.
1301 ext4_orphan_del(NULL, inode);
1304 return ret ? ret : copied;
1308 * Reserve a metadata for a single block located at lblock
1310 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1313 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1314 struct ext4_inode_info *ei = EXT4_I(inode);
1315 unsigned int md_needed;
1316 ext4_lblk_t save_last_lblock;
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, 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)) {
1350 ei->i_reserved_meta_blocks += md_needed;
1351 spin_unlock(&ei->i_block_reservation_lock);
1353 return 0; /* success */
1357 * Reserve a single cluster located at lblock
1359 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1362 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1363 struct ext4_inode_info *ei = EXT4_I(inode);
1364 unsigned int md_needed;
1366 ext4_lblk_t save_last_lblock;
1370 * We will charge metadata quota at writeout time; this saves
1371 * us from metadata over-estimation, though we may go over by
1372 * a small amount in the end. Here we just reserve for data.
1374 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1379 * recalculate the amount of metadata blocks to reserve
1380 * in order to allocate nrblocks
1381 * worse case is one extent per block
1384 spin_lock(&ei->i_block_reservation_lock);
1386 * ext4_calc_metadata_amount() has side effects, which we have
1387 * to be prepared undo if we fail to claim space.
1389 save_len = ei->i_da_metadata_calc_len;
1390 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1391 md_needed = EXT4_NUM_B2C(sbi,
1392 ext4_calc_metadata_amount(inode, lblock));
1393 trace_ext4_da_reserve_space(inode, md_needed);
1396 * We do still charge estimated metadata to the sb though;
1397 * we cannot afford to run out of free blocks.
1399 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1400 ei->i_da_metadata_calc_len = save_len;
1401 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1402 spin_unlock(&ei->i_block_reservation_lock);
1403 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1407 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1410 ei->i_reserved_data_blocks++;
1411 ei->i_reserved_meta_blocks += md_needed;
1412 spin_unlock(&ei->i_block_reservation_lock);
1414 return 0; /* success */
1417 static void ext4_da_release_space(struct inode *inode, int to_free)
1419 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1420 struct ext4_inode_info *ei = EXT4_I(inode);
1423 return; /* Nothing to release, exit */
1425 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1427 trace_ext4_da_release_space(inode, to_free);
1428 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1430 * if there aren't enough reserved blocks, then the
1431 * counter is messed up somewhere. Since this
1432 * function is called from invalidate page, it's
1433 * harmless to return without any action.
1435 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1436 "ino %lu, to_free %d with only %d reserved "
1437 "data blocks", inode->i_ino, to_free,
1438 ei->i_reserved_data_blocks);
1440 to_free = ei->i_reserved_data_blocks;
1442 ei->i_reserved_data_blocks -= to_free;
1444 if (ei->i_reserved_data_blocks == 0) {
1446 * We can release all of the reserved metadata blocks
1447 * only when we have written all of the delayed
1448 * allocation blocks.
1449 * Note that in case of bigalloc, i_reserved_meta_blocks,
1450 * i_reserved_data_blocks, etc. refer to number of clusters.
1452 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1453 ei->i_reserved_meta_blocks);
1454 ei->i_reserved_meta_blocks = 0;
1455 ei->i_da_metadata_calc_len = 0;
1458 /* update fs dirty data blocks counter */
1459 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1461 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1463 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1466 static void ext4_da_page_release_reservation(struct page *page,
1467 unsigned long offset)
1470 struct buffer_head *head, *bh;
1471 unsigned int curr_off = 0;
1472 struct inode *inode = page->mapping->host;
1473 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1477 head = page_buffers(page);
1480 unsigned int next_off = curr_off + bh->b_size;
1482 if ((offset <= curr_off) && (buffer_delay(bh))) {
1484 clear_buffer_delay(bh);
1486 curr_off = next_off;
1487 } while ((bh = bh->b_this_page) != head);
1490 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1491 ext4_es_remove_extent(inode, lblk, to_release);
1494 /* If we have released all the blocks belonging to a cluster, then we
1495 * need to release the reserved space for that cluster. */
1496 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1497 while (num_clusters > 0) {
1498 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1499 ((num_clusters - 1) << sbi->s_cluster_bits);
1500 if (sbi->s_cluster_ratio == 1 ||
1501 !ext4_find_delalloc_cluster(inode, lblk))
1502 ext4_da_release_space(inode, 1);
1509 * Delayed allocation stuff
1513 * mpage_da_submit_io - walks through extent of pages and try to write
1514 * them with writepage() call back
1516 * @mpd->inode: inode
1517 * @mpd->first_page: first page of the extent
1518 * @mpd->next_page: page after the last page of the extent
1520 * By the time mpage_da_submit_io() is called we expect all blocks
1521 * to be allocated. this may be wrong if allocation failed.
1523 * As pages are already locked by write_cache_pages(), we can't use it
1525 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1526 struct ext4_map_blocks *map)
1528 struct pagevec pvec;
1529 unsigned long index, end;
1530 int ret = 0, err, nr_pages, i;
1531 struct inode *inode = mpd->inode;
1532 struct address_space *mapping = inode->i_mapping;
1533 loff_t size = i_size_read(inode);
1534 unsigned int len, block_start;
1535 struct buffer_head *bh, *page_bufs = NULL;
1536 sector_t pblock = 0, cur_logical = 0;
1537 struct ext4_io_submit io_submit;
1539 BUG_ON(mpd->next_page <= mpd->first_page);
1540 memset(&io_submit, 0, sizeof(io_submit));
1542 * We need to start from the first_page to the next_page - 1
1543 * to make sure we also write the mapped dirty buffer_heads.
1544 * If we look at mpd->b_blocknr we would only be looking
1545 * at the currently mapped buffer_heads.
1547 index = mpd->first_page;
1548 end = mpd->next_page - 1;
1550 pagevec_init(&pvec, 0);
1551 while (index <= end) {
1552 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1555 for (i = 0; i < nr_pages; i++) {
1557 struct page *page = pvec.pages[i];
1559 index = page->index;
1563 if (index == size >> PAGE_CACHE_SHIFT)
1564 len = size & ~PAGE_CACHE_MASK;
1566 len = PAGE_CACHE_SIZE;
1568 cur_logical = index << (PAGE_CACHE_SHIFT -
1570 pblock = map->m_pblk + (cur_logical -
1575 BUG_ON(!PageLocked(page));
1576 BUG_ON(PageWriteback(page));
1578 bh = page_bufs = page_buffers(page);
1581 if (map && (cur_logical >= map->m_lblk) &&
1582 (cur_logical <= (map->m_lblk +
1583 (map->m_len - 1)))) {
1584 if (buffer_delay(bh)) {
1585 clear_buffer_delay(bh);
1586 bh->b_blocknr = pblock;
1588 if (buffer_unwritten(bh) ||
1590 BUG_ON(bh->b_blocknr != pblock);
1591 if (map->m_flags & EXT4_MAP_UNINIT)
1592 set_buffer_uninit(bh);
1593 clear_buffer_unwritten(bh);
1597 * skip page if block allocation undone and
1600 if (ext4_bh_delay_or_unwritten(NULL, bh))
1602 bh = bh->b_this_page;
1603 block_start += bh->b_size;
1606 } while (bh != page_bufs);
1613 clear_page_dirty_for_io(page);
1614 err = ext4_bio_write_page(&io_submit, page, len,
1617 mpd->pages_written++;
1619 * In error case, we have to continue because
1620 * remaining pages are still locked
1625 pagevec_release(&pvec);
1627 ext4_io_submit(&io_submit);
1631 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1635 struct pagevec pvec;
1636 struct inode *inode = mpd->inode;
1637 struct address_space *mapping = inode->i_mapping;
1638 ext4_lblk_t start, last;
1640 index = mpd->first_page;
1641 end = mpd->next_page - 1;
1643 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1644 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1645 ext4_es_remove_extent(inode, start, last - start + 1);
1647 pagevec_init(&pvec, 0);
1648 while (index <= end) {
1649 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1652 for (i = 0; i < nr_pages; i++) {
1653 struct page *page = pvec.pages[i];
1654 if (page->index > end)
1656 BUG_ON(!PageLocked(page));
1657 BUG_ON(PageWriteback(page));
1658 block_invalidatepage(page, 0);
1659 ClearPageUptodate(page);
1662 index = pvec.pages[nr_pages - 1]->index + 1;
1663 pagevec_release(&pvec);
1668 static void ext4_print_free_blocks(struct inode *inode)
1670 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1671 struct super_block *sb = inode->i_sb;
1673 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1674 EXT4_C2B(EXT4_SB(inode->i_sb),
1675 ext4_count_free_clusters(inode->i_sb)));
1676 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1677 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1678 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1679 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1680 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1681 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1682 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1683 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1684 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1685 EXT4_I(inode)->i_reserved_data_blocks);
1686 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1687 EXT4_I(inode)->i_reserved_meta_blocks);
1692 * mpage_da_map_and_submit - go through given space, map them
1693 * if necessary, and then submit them for I/O
1695 * @mpd - bh describing space
1697 * The function skips space we know is already mapped to disk blocks.
1700 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1702 int err, blks, get_blocks_flags;
1703 struct ext4_map_blocks map, *mapp = NULL;
1704 sector_t next = mpd->b_blocknr;
1705 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1706 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1707 handle_t *handle = NULL;
1710 * If the blocks are mapped already, or we couldn't accumulate
1711 * any blocks, then proceed immediately to the submission stage.
1713 if ((mpd->b_size == 0) ||
1714 ((mpd->b_state & (1 << BH_Mapped)) &&
1715 !(mpd->b_state & (1 << BH_Delay)) &&
1716 !(mpd->b_state & (1 << BH_Unwritten))))
1719 handle = ext4_journal_current_handle();
1723 * Call ext4_map_blocks() to allocate any delayed allocation
1724 * blocks, or to convert an uninitialized extent to be
1725 * initialized (in the case where we have written into
1726 * one or more preallocated blocks).
1728 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1729 * indicate that we are on the delayed allocation path. This
1730 * affects functions in many different parts of the allocation
1731 * call path. This flag exists primarily because we don't
1732 * want to change *many* call functions, so ext4_map_blocks()
1733 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1734 * inode's allocation semaphore is taken.
1736 * If the blocks in questions were delalloc blocks, set
1737 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1738 * variables are updated after the blocks have been allocated.
1741 map.m_len = max_blocks;
1742 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1743 if (ext4_should_dioread_nolock(mpd->inode))
1744 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1745 if (mpd->b_state & (1 << BH_Delay))
1746 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1748 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1750 struct super_block *sb = mpd->inode->i_sb;
1754 * If get block returns EAGAIN or ENOSPC and there
1755 * appears to be free blocks we will just let
1756 * mpage_da_submit_io() unlock all of the pages.
1761 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1767 * get block failure will cause us to loop in
1768 * writepages, because a_ops->writepage won't be able
1769 * to make progress. The page will be redirtied by
1770 * writepage and writepages will again try to write
1773 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1774 ext4_msg(sb, KERN_CRIT,
1775 "delayed block allocation failed for inode %lu "
1776 "at logical offset %llu with max blocks %zd "
1777 "with error %d", mpd->inode->i_ino,
1778 (unsigned long long) next,
1779 mpd->b_size >> mpd->inode->i_blkbits, err);
1780 ext4_msg(sb, KERN_CRIT,
1781 "This should not happen!! Data will be lost");
1783 ext4_print_free_blocks(mpd->inode);
1785 /* invalidate all the pages */
1786 ext4_da_block_invalidatepages(mpd);
1788 /* Mark this page range as having been completed */
1795 if (map.m_flags & EXT4_MAP_NEW) {
1796 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1799 for (i = 0; i < map.m_len; i++)
1800 unmap_underlying_metadata(bdev, map.m_pblk + i);
1804 * Update on-disk size along with block allocation.
1806 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1807 if (disksize > i_size_read(mpd->inode))
1808 disksize = i_size_read(mpd->inode);
1809 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1810 ext4_update_i_disksize(mpd->inode, disksize);
1811 err = ext4_mark_inode_dirty(handle, mpd->inode);
1813 ext4_error(mpd->inode->i_sb,
1814 "Failed to mark inode %lu dirty",
1819 mpage_da_submit_io(mpd, mapp);
1823 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1824 (1 << BH_Delay) | (1 << BH_Unwritten))
1827 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1829 * @mpd->lbh - extent of blocks
1830 * @logical - logical number of the block in the file
1831 * @b_state - b_state of the buffer head added
1833 * the function is used to collect contig. blocks in same state
1835 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical,
1836 unsigned long b_state)
1839 int blkbits = mpd->inode->i_blkbits;
1840 int nrblocks = mpd->b_size >> blkbits;
1843 * XXX Don't go larger than mballoc is willing to allocate
1844 * This is a stopgap solution. We eventually need to fold
1845 * mpage_da_submit_io() into this function and then call
1846 * ext4_map_blocks() multiple times in a loop
1848 if (nrblocks >= (8*1024*1024 >> blkbits))
1851 /* check if the reserved journal credits might overflow */
1852 if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) {
1853 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1855 * With non-extent format we are limited by the journal
1856 * credit available. Total credit needed to insert
1857 * nrblocks contiguous blocks is dependent on the
1858 * nrblocks. So limit nrblocks.
1864 * First block in the extent
1866 if (mpd->b_size == 0) {
1867 mpd->b_blocknr = logical;
1868 mpd->b_size = 1 << blkbits;
1869 mpd->b_state = b_state & BH_FLAGS;
1873 next = mpd->b_blocknr + nrblocks;
1875 * Can we merge the block to our big extent?
1877 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1878 mpd->b_size += 1 << blkbits;
1884 * We couldn't merge the block to our extent, so we
1885 * need to flush current extent and start new one
1887 mpage_da_map_and_submit(mpd);
1891 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1893 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1897 * This function is grabs code from the very beginning of
1898 * ext4_map_blocks, but assumes that the caller is from delayed write
1899 * time. This function looks up the requested blocks and sets the
1900 * buffer delay bit under the protection of i_data_sem.
1902 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1903 struct ext4_map_blocks *map,
1904 struct buffer_head *bh)
1906 struct extent_status es;
1908 sector_t invalid_block = ~((sector_t) 0xffff);
1909 #ifdef ES_AGGRESSIVE_TEST
1910 struct ext4_map_blocks orig_map;
1912 memcpy(&orig_map, map, sizeof(*map));
1915 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1919 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1920 "logical block %lu\n", inode->i_ino, map->m_len,
1921 (unsigned long) map->m_lblk);
1923 /* Lookup extent status tree firstly */
1924 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1926 if (ext4_es_is_hole(&es)) {
1928 down_read((&EXT4_I(inode)->i_data_sem));
1933 * Delayed extent could be allocated by fallocate.
1934 * So we need to check it.
1936 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1937 map_bh(bh, inode->i_sb, invalid_block);
1939 set_buffer_delay(bh);
1943 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1944 retval = es.es_len - (iblock - es.es_lblk);
1945 if (retval > map->m_len)
1946 retval = map->m_len;
1947 map->m_len = retval;
1948 if (ext4_es_is_written(&es))
1949 map->m_flags |= EXT4_MAP_MAPPED;
1950 else if (ext4_es_is_unwritten(&es))
1951 map->m_flags |= EXT4_MAP_UNWRITTEN;
1955 #ifdef ES_AGGRESSIVE_TEST
1956 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1962 * Try to see if we can get the block without requesting a new
1963 * file system block.
1965 down_read((&EXT4_I(inode)->i_data_sem));
1966 if (ext4_has_inline_data(inode)) {
1968 * We will soon create blocks for this page, and let
1969 * us pretend as if the blocks aren't allocated yet.
1970 * In case of clusters, we have to handle the work
1971 * of mapping from cluster so that the reserved space
1972 * is calculated properly.
1974 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1975 ext4_find_delalloc_cluster(inode, map->m_lblk))
1976 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1978 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1979 retval = ext4_ext_map_blocks(NULL, inode, map,
1980 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1982 retval = ext4_ind_map_blocks(NULL, inode, map,
1983 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1989 * XXX: __block_prepare_write() unmaps passed block,
1993 * If the block was allocated from previously allocated cluster,
1994 * then we don't need to reserve it again. However we still need
1995 * to reserve metadata for every block we're going to write.
1997 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1998 ret = ext4_da_reserve_space(inode, iblock);
2000 /* not enough space to reserve */
2005 ret = ext4_da_reserve_metadata(inode, iblock);
2007 /* not enough space to reserve */
2013 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
2014 ~0, EXTENT_STATUS_DELAYED);
2020 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
2021 * and it should not appear on the bh->b_state.
2023 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
2025 map_bh(bh, inode->i_sb, invalid_block);
2027 set_buffer_delay(bh);
2028 } else if (retval > 0) {
2030 unsigned long long status;
2032 #ifdef ES_AGGRESSIVE_TEST
2033 if (retval != map->m_len) {
2034 printk("ES len assertation failed for inode: %lu "
2035 "retval %d != map->m_len %d "
2036 "in %s (lookup)\n", inode->i_ino, retval,
2037 map->m_len, __func__);
2041 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
2042 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
2043 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
2044 map->m_pblk, status);
2050 up_read((&EXT4_I(inode)->i_data_sem));
2056 * This is a special get_blocks_t callback which is used by
2057 * ext4_da_write_begin(). It will either return mapped block or
2058 * reserve space for a single block.
2060 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2061 * We also have b_blocknr = -1 and b_bdev initialized properly
2063 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2064 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2065 * initialized properly.
2067 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2068 struct buffer_head *bh, int create)
2070 struct ext4_map_blocks map;
2073 BUG_ON(create == 0);
2074 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2076 map.m_lblk = iblock;
2080 * first, we need to know whether the block is allocated already
2081 * preallocated blocks are unmapped but should treated
2082 * the same as allocated blocks.
2084 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
2088 map_bh(bh, inode->i_sb, map.m_pblk);
2089 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2091 if (buffer_unwritten(bh)) {
2092 /* A delayed write to unwritten bh should be marked
2093 * new and mapped. Mapped ensures that we don't do
2094 * get_block multiple times when we write to the same
2095 * offset and new ensures that we do proper zero out
2096 * for partial write.
2099 set_buffer_mapped(bh);
2104 static int bget_one(handle_t *handle, struct buffer_head *bh)
2110 static int bput_one(handle_t *handle, struct buffer_head *bh)
2116 static int __ext4_journalled_writepage(struct page *page,
2119 struct address_space *mapping = page->mapping;
2120 struct inode *inode = mapping->host;
2121 struct buffer_head *page_bufs = NULL;
2122 handle_t *handle = NULL;
2123 int ret = 0, err = 0;
2124 int inline_data = ext4_has_inline_data(inode);
2125 struct buffer_head *inode_bh = NULL;
2127 ClearPageChecked(page);
2130 BUG_ON(page->index != 0);
2131 BUG_ON(len > ext4_get_max_inline_size(inode));
2132 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2133 if (inode_bh == NULL)
2136 page_bufs = page_buffers(page);
2141 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2144 /* As soon as we unlock the page, it can go away, but we have
2145 * references to buffers so we are safe */
2148 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2149 ext4_writepage_trans_blocks(inode));
2150 if (IS_ERR(handle)) {
2151 ret = PTR_ERR(handle);
2155 BUG_ON(!ext4_handle_valid(handle));
2158 ret = ext4_journal_get_write_access(handle, inode_bh);
2160 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2163 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2164 do_journal_get_write_access);
2166 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2171 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2172 err = ext4_journal_stop(handle);
2176 if (!ext4_has_inline_data(inode))
2177 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2179 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2186 * Note that we don't need to start a transaction unless we're journaling data
2187 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2188 * need to file the inode to the transaction's list in ordered mode because if
2189 * we are writing back data added by write(), the inode is already there and if
2190 * we are writing back data modified via mmap(), no one guarantees in which
2191 * transaction the data will hit the disk. In case we are journaling data, we
2192 * cannot start transaction directly because transaction start ranks above page
2193 * lock so we have to do some magic.
2195 * This function can get called via...
2196 * - ext4_da_writepages after taking page lock (have journal handle)
2197 * - journal_submit_inode_data_buffers (no journal handle)
2198 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2199 * - grab_page_cache when doing write_begin (have journal handle)
2201 * We don't do any block allocation in this function. If we have page with
2202 * multiple blocks we need to write those buffer_heads that are mapped. This
2203 * is important for mmaped based write. So if we do with blocksize 1K
2204 * truncate(f, 1024);
2205 * a = mmap(f, 0, 4096);
2207 * truncate(f, 4096);
2208 * we have in the page first buffer_head mapped via page_mkwrite call back
2209 * but other buffer_heads would be unmapped but dirty (dirty done via the
2210 * do_wp_page). So writepage should write the first block. If we modify
2211 * the mmap area beyond 1024 we will again get a page_fault and the
2212 * page_mkwrite callback will do the block allocation and mark the
2213 * buffer_heads mapped.
2215 * We redirty the page if we have any buffer_heads that is either delay or
2216 * unwritten in the page.
2218 * We can get recursively called as show below.
2220 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2223 * But since we don't do any block allocation we should not deadlock.
2224 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2226 static int ext4_writepage(struct page *page,
2227 struct writeback_control *wbc)
2232 struct buffer_head *page_bufs = NULL;
2233 struct inode *inode = page->mapping->host;
2234 struct ext4_io_submit io_submit;
2236 trace_ext4_writepage(page);
2237 size = i_size_read(inode);
2238 if (page->index == size >> PAGE_CACHE_SHIFT)
2239 len = size & ~PAGE_CACHE_MASK;
2241 len = PAGE_CACHE_SIZE;
2243 page_bufs = page_buffers(page);
2245 * We cannot do block allocation or other extent handling in this
2246 * function. If there are buffers needing that, we have to redirty
2247 * the page. But we may reach here when we do a journal commit via
2248 * journal_submit_inode_data_buffers() and in that case we must write
2249 * allocated buffers to achieve data=ordered mode guarantees.
2251 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2252 ext4_bh_delay_or_unwritten)) {
2253 redirty_page_for_writepage(wbc, page);
2254 if (current->flags & PF_MEMALLOC) {
2256 * For memory cleaning there's no point in writing only
2257 * some buffers. So just bail out. Warn if we came here
2258 * from direct reclaim.
2260 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2267 if (PageChecked(page) && ext4_should_journal_data(inode))
2269 * It's mmapped pagecache. Add buffers and journal it. There
2270 * doesn't seem much point in redirtying the page here.
2272 return __ext4_journalled_writepage(page, len);
2274 memset(&io_submit, 0, sizeof(io_submit));
2275 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2276 ext4_io_submit(&io_submit);
2281 * This is called via ext4_da_writepages() to
2282 * calculate the total number of credits to reserve to fit
2283 * a single extent allocation into a single transaction,
2284 * ext4_da_writpeages() will loop calling this before
2285 * the block allocation.
2288 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2290 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2293 * With non-extent format the journal credit needed to
2294 * insert nrblocks contiguous block is dependent on
2295 * number of contiguous block. So we will limit
2296 * number of contiguous block to a sane value
2298 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2299 (max_blocks > EXT4_MAX_TRANS_DATA))
2300 max_blocks = EXT4_MAX_TRANS_DATA;
2302 return ext4_chunk_trans_blocks(inode, max_blocks);
2306 * write_cache_pages_da - walk the list of dirty pages of the given
2307 * address space and accumulate pages that need writing, and call
2308 * mpage_da_map_and_submit to map a single contiguous memory region
2309 * and then write them.
2311 static int write_cache_pages_da(handle_t *handle,
2312 struct address_space *mapping,
2313 struct writeback_control *wbc,
2314 struct mpage_da_data *mpd,
2315 pgoff_t *done_index)
2317 struct buffer_head *bh, *head;
2318 struct inode *inode = mapping->host;
2319 struct pagevec pvec;
2320 unsigned int nr_pages;
2323 long nr_to_write = wbc->nr_to_write;
2324 int i, tag, ret = 0;
2326 memset(mpd, 0, sizeof(struct mpage_da_data));
2329 pagevec_init(&pvec, 0);
2330 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2331 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2333 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2334 tag = PAGECACHE_TAG_TOWRITE;
2336 tag = PAGECACHE_TAG_DIRTY;
2338 *done_index = index;
2339 while (index <= end) {
2340 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2341 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2345 for (i = 0; i < nr_pages; i++) {
2346 struct page *page = pvec.pages[i];
2349 * At this point, the page may be truncated or
2350 * invalidated (changing page->mapping to NULL), or
2351 * even swizzled back from swapper_space to tmpfs file
2352 * mapping. However, page->index will not change
2353 * because we have a reference on the page.
2355 if (page->index > end)
2358 *done_index = page->index + 1;
2361 * If we can't merge this page, and we have
2362 * accumulated an contiguous region, write it
2364 if ((mpd->next_page != page->index) &&
2365 (mpd->next_page != mpd->first_page)) {
2366 mpage_da_map_and_submit(mpd);
2367 goto ret_extent_tail;
2373 * If the page is no longer dirty, or its
2374 * mapping no longer corresponds to inode we
2375 * are writing (which means it has been
2376 * truncated or invalidated), or the page is
2377 * already under writeback and we are not
2378 * doing a data integrity writeback, skip the page
2380 if (!PageDirty(page) ||
2381 (PageWriteback(page) &&
2382 (wbc->sync_mode == WB_SYNC_NONE)) ||
2383 unlikely(page->mapping != mapping)) {
2388 wait_on_page_writeback(page);
2389 BUG_ON(PageWriteback(page));
2392 * If we have inline data and arrive here, it means that
2393 * we will soon create the block for the 1st page, so
2394 * we'd better clear the inline data here.
2396 if (ext4_has_inline_data(inode)) {
2397 BUG_ON(ext4_test_inode_state(inode,
2398 EXT4_STATE_MAY_INLINE_DATA));
2399 ext4_destroy_inline_data(handle, inode);
2402 if (mpd->next_page != page->index)
2403 mpd->first_page = page->index;
2404 mpd->next_page = page->index + 1;
2405 logical = (sector_t) page->index <<
2406 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2408 /* Add all dirty buffers to mpd */
2409 head = page_buffers(page);
2412 BUG_ON(buffer_locked(bh));
2414 * We need to try to allocate unmapped blocks
2415 * in the same page. Otherwise we won't make
2416 * progress with the page in ext4_writepage
2418 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2419 mpage_add_bh_to_extent(mpd, logical,
2422 goto ret_extent_tail;
2423 } else if (buffer_dirty(bh) &&
2424 buffer_mapped(bh)) {
2426 * mapped dirty buffer. We need to
2427 * update the b_state because we look
2428 * at b_state in mpage_da_map_blocks.
2429 * We don't update b_size because if we
2430 * find an unmapped buffer_head later
2431 * we need to use the b_state flag of
2434 if (mpd->b_size == 0)
2436 bh->b_state & BH_FLAGS;
2439 } while ((bh = bh->b_this_page) != head);
2441 if (nr_to_write > 0) {
2443 if (nr_to_write == 0 &&
2444 wbc->sync_mode == WB_SYNC_NONE)
2446 * We stop writing back only if we are
2447 * not doing integrity sync. In case of
2448 * integrity sync we have to keep going
2449 * because someone may be concurrently
2450 * dirtying pages, and we might have
2451 * synced a lot of newly appeared dirty
2452 * pages, but have not synced all of the
2458 pagevec_release(&pvec);
2463 ret = MPAGE_DA_EXTENT_TAIL;
2465 pagevec_release(&pvec);
2471 static int ext4_da_writepages(struct address_space *mapping,
2472 struct writeback_control *wbc)
2475 int range_whole = 0;
2476 handle_t *handle = NULL;
2477 struct mpage_da_data mpd;
2478 struct inode *inode = mapping->host;
2479 int pages_written = 0;
2480 unsigned int max_pages;
2481 int range_cyclic, cycled = 1, io_done = 0;
2482 int needed_blocks, ret = 0;
2483 long desired_nr_to_write, nr_to_writebump = 0;
2484 loff_t range_start = wbc->range_start;
2485 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2486 pgoff_t done_index = 0;
2488 struct blk_plug plug;
2490 trace_ext4_da_writepages(inode, wbc);
2493 * No pages to write? This is mainly a kludge to avoid starting
2494 * a transaction for special inodes like journal inode on last iput()
2495 * because that could violate lock ordering on umount
2497 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2501 * If the filesystem has aborted, it is read-only, so return
2502 * right away instead of dumping stack traces later on that
2503 * will obscure the real source of the problem. We test
2504 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2505 * the latter could be true if the filesystem is mounted
2506 * read-only, and in that case, ext4_da_writepages should
2507 * *never* be called, so if that ever happens, we would want
2510 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2513 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2516 range_cyclic = wbc->range_cyclic;
2517 if (wbc->range_cyclic) {
2518 index = mapping->writeback_index;
2521 wbc->range_start = index << PAGE_CACHE_SHIFT;
2522 wbc->range_end = LLONG_MAX;
2523 wbc->range_cyclic = 0;
2526 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2527 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2531 * This works around two forms of stupidity. The first is in
2532 * the writeback code, which caps the maximum number of pages
2533 * written to be 1024 pages. This is wrong on multiple
2534 * levels; different architectues have a different page size,
2535 * which changes the maximum amount of data which gets
2536 * written. Secondly, 4 megabytes is way too small. XFS
2537 * forces this value to be 16 megabytes by multiplying
2538 * nr_to_write parameter by four, and then relies on its
2539 * allocator to allocate larger extents to make them
2540 * contiguous. Unfortunately this brings us to the second
2541 * stupidity, which is that ext4's mballoc code only allocates
2542 * at most 2048 blocks. So we force contiguous writes up to
2543 * the number of dirty blocks in the inode, or
2544 * sbi->max_writeback_mb_bump whichever is smaller.
2546 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2547 if (!range_cyclic && range_whole) {
2548 if (wbc->nr_to_write == LONG_MAX)
2549 desired_nr_to_write = wbc->nr_to_write;
2551 desired_nr_to_write = wbc->nr_to_write * 8;
2553 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2555 if (desired_nr_to_write > max_pages)
2556 desired_nr_to_write = max_pages;
2558 if (wbc->nr_to_write < desired_nr_to_write) {
2559 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2560 wbc->nr_to_write = desired_nr_to_write;
2564 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2565 tag_pages_for_writeback(mapping, index, end);
2567 blk_start_plug(&plug);
2568 while (!ret && wbc->nr_to_write > 0) {
2571 * we insert one extent at a time. So we need
2572 * credit needed for single extent allocation.
2573 * journalled mode is currently not supported
2576 BUG_ON(ext4_should_journal_data(inode));
2577 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2579 /* start a new transaction*/
2580 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2582 if (IS_ERR(handle)) {
2583 ret = PTR_ERR(handle);
2584 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2585 "%ld pages, ino %lu; err %d", __func__,
2586 wbc->nr_to_write, inode->i_ino, ret);
2587 blk_finish_plug(&plug);
2588 goto out_writepages;
2592 * Now call write_cache_pages_da() to find the next
2593 * contiguous region of logical blocks that need
2594 * blocks to be allocated by ext4 and submit them.
2596 ret = write_cache_pages_da(handle, mapping,
2597 wbc, &mpd, &done_index);
2599 * If we have a contiguous extent of pages and we
2600 * haven't done the I/O yet, map the blocks and submit
2603 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2604 mpage_da_map_and_submit(&mpd);
2605 ret = MPAGE_DA_EXTENT_TAIL;
2607 trace_ext4_da_write_pages(inode, &mpd);
2608 wbc->nr_to_write -= mpd.pages_written;
2610 ext4_journal_stop(handle);
2612 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2613 /* commit the transaction which would
2614 * free blocks released in the transaction
2617 jbd2_journal_force_commit_nested(sbi->s_journal);
2619 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2621 * Got one extent now try with rest of the pages.
2622 * If mpd.retval is set -EIO, journal is aborted.
2623 * So we don't need to write any more.
2625 pages_written += mpd.pages_written;
2628 } else if (wbc->nr_to_write)
2630 * There is no more writeout needed
2631 * or we requested for a noblocking writeout
2632 * and we found the device congested
2636 blk_finish_plug(&plug);
2637 if (!io_done && !cycled) {
2640 wbc->range_start = index << PAGE_CACHE_SHIFT;
2641 wbc->range_end = mapping->writeback_index - 1;
2646 wbc->range_cyclic = range_cyclic;
2647 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2649 * set the writeback_index so that range_cyclic
2650 * mode will write it back later
2652 mapping->writeback_index = done_index;
2655 wbc->nr_to_write -= nr_to_writebump;
2656 wbc->range_start = range_start;
2657 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2661 static int ext4_nonda_switch(struct super_block *sb)
2663 s64 free_blocks, dirty_blocks;
2664 struct ext4_sb_info *sbi = EXT4_SB(sb);
2667 * switch to non delalloc mode if we are running low
2668 * on free block. The free block accounting via percpu
2669 * counters can get slightly wrong with percpu_counter_batch getting
2670 * accumulated on each CPU without updating global counters
2671 * Delalloc need an accurate free block accounting. So switch
2672 * to non delalloc when we are near to error range.
2674 free_blocks = EXT4_C2B(sbi,
2675 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2676 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2678 * Start pushing delalloc when 1/2 of free blocks are dirty.
2680 if (dirty_blocks && (free_blocks < 2 * dirty_blocks) &&
2681 !writeback_in_progress(sb->s_bdi) &&
2682 down_read_trylock(&sb->s_umount)) {
2683 writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2684 up_read(&sb->s_umount);
2687 if (2 * free_blocks < 3 * dirty_blocks ||
2688 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2690 * free block count is less than 150% of dirty blocks
2691 * or free blocks is less than watermark
2698 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2699 loff_t pos, unsigned len, unsigned flags,
2700 struct page **pagep, void **fsdata)
2702 int ret, retries = 0;
2705 struct inode *inode = mapping->host;
2708 index = pos >> PAGE_CACHE_SHIFT;
2710 if (ext4_nonda_switch(inode->i_sb)) {
2711 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2712 return ext4_write_begin(file, mapping, pos,
2713 len, flags, pagep, fsdata);
2715 *fsdata = (void *)0;
2716 trace_ext4_da_write_begin(inode, pos, len, flags);
2718 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2719 ret = ext4_da_write_inline_data_begin(mapping, inode,
2729 * grab_cache_page_write_begin() can take a long time if the
2730 * system is thrashing due to memory pressure, or if the page
2731 * is being written back. So grab it first before we start
2732 * the transaction handle. This also allows us to allocate
2733 * the page (if needed) without using GFP_NOFS.
2736 page = grab_cache_page_write_begin(mapping, index, flags);
2742 * With delayed allocation, we don't log the i_disksize update
2743 * if there is delayed block allocation. But we still need
2744 * to journalling the i_disksize update if writes to the end
2745 * of file which has an already mapped buffer.
2748 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2749 if (IS_ERR(handle)) {
2750 page_cache_release(page);
2751 return PTR_ERR(handle);
2755 if (page->mapping != mapping) {
2756 /* The page got truncated from under us */
2758 page_cache_release(page);
2759 ext4_journal_stop(handle);
2762 /* In case writeback began while the page was unlocked */
2763 wait_on_page_writeback(page);
2765 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2768 ext4_journal_stop(handle);
2770 * block_write_begin may have instantiated a few blocks
2771 * outside i_size. Trim these off again. Don't need
2772 * i_size_read because we hold i_mutex.
2774 if (pos + len > inode->i_size)
2775 ext4_truncate_failed_write(inode);
2777 if (ret == -ENOSPC &&
2778 ext4_should_retry_alloc(inode->i_sb, &retries))
2781 page_cache_release(page);
2790 * Check if we should update i_disksize
2791 * when write to the end of file but not require block allocation
2793 static int ext4_da_should_update_i_disksize(struct page *page,
2794 unsigned long offset)
2796 struct buffer_head *bh;
2797 struct inode *inode = page->mapping->host;
2801 bh = page_buffers(page);
2802 idx = offset >> inode->i_blkbits;
2804 for (i = 0; i < idx; i++)
2805 bh = bh->b_this_page;
2807 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2812 static int ext4_da_write_end(struct file *file,
2813 struct address_space *mapping,
2814 loff_t pos, unsigned len, unsigned copied,
2815 struct page *page, void *fsdata)
2817 struct inode *inode = mapping->host;
2819 handle_t *handle = ext4_journal_current_handle();
2821 unsigned long start, end;
2822 int write_mode = (int)(unsigned long)fsdata;
2824 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2825 switch (ext4_inode_journal_mode(inode)) {
2826 case EXT4_INODE_ORDERED_DATA_MODE:
2827 return ext4_ordered_write_end(file, mapping, pos,
2828 len, copied, page, fsdata);
2829 case EXT4_INODE_WRITEBACK_DATA_MODE:
2830 return ext4_writeback_write_end(file, mapping, pos,
2831 len, copied, page, fsdata);
2837 trace_ext4_da_write_end(inode, pos, len, copied);
2838 start = pos & (PAGE_CACHE_SIZE - 1);
2839 end = start + copied - 1;
2842 * generic_write_end() will run mark_inode_dirty() if i_size
2843 * changes. So let's piggyback the i_disksize mark_inode_dirty
2846 new_i_size = pos + copied;
2847 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2848 if (ext4_has_inline_data(inode) ||
2849 ext4_da_should_update_i_disksize(page, end)) {
2850 down_write(&EXT4_I(inode)->i_data_sem);
2851 if (new_i_size > EXT4_I(inode)->i_disksize)
2852 EXT4_I(inode)->i_disksize = new_i_size;
2853 up_write(&EXT4_I(inode)->i_data_sem);
2854 /* We need to mark inode dirty even if
2855 * new_i_size is less that inode->i_size
2856 * bu greater than i_disksize.(hint delalloc)
2858 ext4_mark_inode_dirty(handle, inode);
2862 if (write_mode != CONVERT_INLINE_DATA &&
2863 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2864 ext4_has_inline_data(inode))
2865 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2868 ret2 = generic_write_end(file, mapping, pos, len, copied,
2874 ret2 = ext4_journal_stop(handle);
2878 return ret ? ret : copied;
2881 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2884 * Drop reserved blocks
2886 BUG_ON(!PageLocked(page));
2887 if (!page_has_buffers(page))
2890 ext4_da_page_release_reservation(page, offset);
2893 ext4_invalidatepage(page, offset);
2899 * Force all delayed allocation blocks to be allocated for a given inode.
2901 int ext4_alloc_da_blocks(struct inode *inode)
2903 trace_ext4_alloc_da_blocks(inode);
2905 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2906 !EXT4_I(inode)->i_reserved_meta_blocks)
2910 * We do something simple for now. The filemap_flush() will
2911 * also start triggering a write of the data blocks, which is
2912 * not strictly speaking necessary (and for users of
2913 * laptop_mode, not even desirable). However, to do otherwise
2914 * would require replicating code paths in:
2916 * ext4_da_writepages() ->
2917 * write_cache_pages() ---> (via passed in callback function)
2918 * __mpage_da_writepage() -->
2919 * mpage_add_bh_to_extent()
2920 * mpage_da_map_blocks()
2922 * The problem is that write_cache_pages(), located in
2923 * mm/page-writeback.c, marks pages clean in preparation for
2924 * doing I/O, which is not desirable if we're not planning on
2927 * We could call write_cache_pages(), and then redirty all of
2928 * the pages by calling redirty_page_for_writepage() but that
2929 * would be ugly in the extreme. So instead we would need to
2930 * replicate parts of the code in the above functions,
2931 * simplifying them because we wouldn't actually intend to
2932 * write out the pages, but rather only collect contiguous
2933 * logical block extents, call the multi-block allocator, and
2934 * then update the buffer heads with the block allocations.
2936 * For now, though, we'll cheat by calling filemap_flush(),
2937 * which will map the blocks, and start the I/O, but not
2938 * actually wait for the I/O to complete.
2940 return filemap_flush(inode->i_mapping);
2944 * bmap() is special. It gets used by applications such as lilo and by
2945 * the swapper to find the on-disk block of a specific piece of data.
2947 * Naturally, this is dangerous if the block concerned is still in the
2948 * journal. If somebody makes a swapfile on an ext4 data-journaling
2949 * filesystem and enables swap, then they may get a nasty shock when the
2950 * data getting swapped to that swapfile suddenly gets overwritten by
2951 * the original zero's written out previously to the journal and
2952 * awaiting writeback in the kernel's buffer cache.
2954 * So, if we see any bmap calls here on a modified, data-journaled file,
2955 * take extra steps to flush any blocks which might be in the cache.
2957 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2959 struct inode *inode = mapping->host;
2964 * We can get here for an inline file via the FIBMAP ioctl
2966 if (ext4_has_inline_data(inode))
2969 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2970 test_opt(inode->i_sb, DELALLOC)) {
2972 * With delalloc we want to sync the file
2973 * so that we can make sure we allocate
2976 filemap_write_and_wait(mapping);
2979 if (EXT4_JOURNAL(inode) &&
2980 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2982 * This is a REALLY heavyweight approach, but the use of
2983 * bmap on dirty files is expected to be extremely rare:
2984 * only if we run lilo or swapon on a freshly made file
2985 * do we expect this to happen.
2987 * (bmap requires CAP_SYS_RAWIO so this does not
2988 * represent an unprivileged user DOS attack --- we'd be
2989 * in trouble if mortal users could trigger this path at
2992 * NB. EXT4_STATE_JDATA is not set on files other than
2993 * regular files. If somebody wants to bmap a directory
2994 * or symlink and gets confused because the buffer
2995 * hasn't yet been flushed to disk, they deserve
2996 * everything they get.
2999 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3000 journal = EXT4_JOURNAL(inode);
3001 jbd2_journal_lock_updates(journal);
3002 err = jbd2_journal_flush(journal);
3003 jbd2_journal_unlock_updates(journal);
3009 return generic_block_bmap(mapping, block, ext4_get_block);
3012 static int ext4_readpage(struct file *file, struct page *page)
3015 struct inode *inode = page->mapping->host;
3017 trace_ext4_readpage(page);
3019 if (ext4_has_inline_data(inode))
3020 ret = ext4_readpage_inline(inode, page);
3023 return mpage_readpage(page, ext4_get_block);
3029 ext4_readpages(struct file *file, struct address_space *mapping,
3030 struct list_head *pages, unsigned nr_pages)
3032 struct inode *inode = mapping->host;
3034 /* If the file has inline data, no need to do readpages. */
3035 if (ext4_has_inline_data(inode))
3038 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3041 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3043 trace_ext4_invalidatepage(page, offset);
3045 /* No journalling happens on data buffers when this function is used */
3046 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3048 block_invalidatepage(page, offset);
3051 static int __ext4_journalled_invalidatepage(struct page *page,
3052 unsigned long offset)
3054 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3056 trace_ext4_journalled_invalidatepage(page, offset);
3059 * If it's a full truncate we just forget about the pending dirtying
3062 ClearPageChecked(page);
3064 return jbd2_journal_invalidatepage(journal, page, offset);
3067 /* Wrapper for aops... */
3068 static void ext4_journalled_invalidatepage(struct page *page,
3069 unsigned long offset)
3071 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
3074 static int ext4_releasepage(struct page *page, gfp_t wait)
3076 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3078 trace_ext4_releasepage(page);
3080 /* Page has dirty journalled data -> cannot release */
3081 if (PageChecked(page))
3084 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3086 return try_to_free_buffers(page);
3090 * ext4_get_block used when preparing for a DIO write or buffer write.
3091 * We allocate an uinitialized extent if blocks haven't been allocated.
3092 * The extent will be converted to initialized after the IO is complete.
3094 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3095 struct buffer_head *bh_result, int create)
3097 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3098 inode->i_ino, create);
3099 return _ext4_get_block(inode, iblock, bh_result,
3100 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3103 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3104 struct buffer_head *bh_result, int create)
3106 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3107 inode->i_ino, create);
3108 return _ext4_get_block(inode, iblock, bh_result,
3109 EXT4_GET_BLOCKS_NO_LOCK);
3112 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3113 ssize_t size, void *private, int ret,
3116 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
3117 ext4_io_end_t *io_end = iocb->private;
3119 /* if not async direct IO or dio with 0 bytes write, just return */
3120 if (!io_end || !size)
3123 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3124 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3125 iocb->private, io_end->inode->i_ino, iocb, offset,
3128 iocb->private = NULL;
3130 /* if not aio dio with unwritten extents, just free io and return */
3131 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3132 ext4_free_io_end(io_end);
3134 inode_dio_done(inode);
3136 aio_complete(iocb, ret, 0);
3140 io_end->offset = offset;
3141 io_end->size = size;
3143 io_end->iocb = iocb;
3144 io_end->result = ret;
3147 ext4_add_complete_io(io_end);
3151 * For ext4 extent files, ext4 will do direct-io write to holes,
3152 * preallocated extents, and those write extend the file, no need to
3153 * fall back to buffered IO.
3155 * For holes, we fallocate those blocks, mark them as uninitialized
3156 * If those blocks were preallocated, we mark sure they are split, but
3157 * still keep the range to write as uninitialized.
3159 * The unwritten extents will be converted to written when DIO is completed.
3160 * For async direct IO, since the IO may still pending when return, we
3161 * set up an end_io call back function, which will do the conversion
3162 * when async direct IO completed.
3164 * If the O_DIRECT write will extend the file then add this inode to the
3165 * orphan list. So recovery will truncate it back to the original size
3166 * if the machine crashes during the write.
3169 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3170 const struct iovec *iov, loff_t offset,
3171 unsigned long nr_segs)
3173 struct file *file = iocb->ki_filp;
3174 struct inode *inode = file->f_mapping->host;
3176 size_t count = iov_length(iov, nr_segs);
3178 get_block_t *get_block_func = NULL;
3180 loff_t final_size = offset + count;
3182 /* Use the old path for reads and writes beyond i_size. */
3183 if (rw != WRITE || final_size > inode->i_size)
3184 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3186 BUG_ON(iocb->private == NULL);
3188 /* If we do a overwrite dio, i_mutex locking can be released */
3189 overwrite = *((int *)iocb->private);
3192 atomic_inc(&inode->i_dio_count);
3193 down_read(&EXT4_I(inode)->i_data_sem);
3194 mutex_unlock(&inode->i_mutex);
3198 * We could direct write to holes and fallocate.
3200 * Allocated blocks to fill the hole are marked as
3201 * uninitialized to prevent parallel buffered read to expose
3202 * the stale data before DIO complete the data IO.
3204 * As to previously fallocated extents, ext4 get_block will
3205 * just simply mark the buffer mapped but still keep the
3206 * extents uninitialized.
3208 * For non AIO case, we will convert those unwritten extents
3209 * to written after return back from blockdev_direct_IO.
3211 * For async DIO, the conversion needs to be deferred when the
3212 * IO is completed. The ext4 end_io callback function will be
3213 * called to take care of the conversion work. Here for async
3214 * case, we allocate an io_end structure to hook to the iocb.
3216 iocb->private = NULL;
3217 ext4_inode_aio_set(inode, NULL);
3218 if (!is_sync_kiocb(iocb)) {
3219 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
3224 io_end->flag |= EXT4_IO_END_DIRECT;
3225 iocb->private = io_end;
3227 * we save the io structure for current async direct
3228 * IO, so that later ext4_map_blocks() could flag the
3229 * io structure whether there is a unwritten extents
3230 * needs to be converted when IO is completed.
3232 ext4_inode_aio_set(inode, io_end);
3236 get_block_func = ext4_get_block_write_nolock;
3238 get_block_func = ext4_get_block_write;
3239 dio_flags = DIO_LOCKING;
3241 ret = __blockdev_direct_IO(rw, iocb, inode,
3242 inode->i_sb->s_bdev, iov,
3250 ext4_inode_aio_set(inode, NULL);
3252 * The io_end structure takes a reference to the inode, that
3253 * structure needs to be destroyed and the reference to the
3254 * inode need to be dropped, when IO is complete, even with 0
3255 * byte write, or failed.
3257 * In the successful AIO DIO case, the io_end structure will
3258 * be destroyed and the reference to the inode will be dropped
3259 * after the end_io call back function is called.
3261 * In the case there is 0 byte write, or error case, since VFS
3262 * direct IO won't invoke the end_io call back function, we
3263 * need to free the end_io structure here.
3265 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3266 ext4_free_io_end(iocb->private);
3267 iocb->private = NULL;
3268 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3269 EXT4_STATE_DIO_UNWRITTEN)) {
3272 * for non AIO case, since the IO is already
3273 * completed, we could do the conversion right here
3275 err = ext4_convert_unwritten_extents(inode,
3279 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3283 /* take i_mutex locking again if we do a ovewrite dio */
3285 inode_dio_done(inode);
3286 up_read(&EXT4_I(inode)->i_data_sem);
3287 mutex_lock(&inode->i_mutex);
3293 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3294 const struct iovec *iov, loff_t offset,
3295 unsigned long nr_segs)
3297 struct file *file = iocb->ki_filp;
3298 struct inode *inode = file->f_mapping->host;
3302 * If we are doing data journalling we don't support O_DIRECT
3304 if (ext4_should_journal_data(inode))
3307 /* Let buffer I/O handle the inline data case. */
3308 if (ext4_has_inline_data(inode))
3311 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3312 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3313 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3315 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3316 trace_ext4_direct_IO_exit(inode, offset,
3317 iov_length(iov, nr_segs), rw, ret);
3322 * Pages can be marked dirty completely asynchronously from ext4's journalling
3323 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3324 * much here because ->set_page_dirty is called under VFS locks. The page is
3325 * not necessarily locked.
3327 * We cannot just dirty the page and leave attached buffers clean, because the
3328 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3329 * or jbddirty because all the journalling code will explode.
3331 * So what we do is to mark the page "pending dirty" and next time writepage
3332 * is called, propagate that into the buffers appropriately.
3334 static int ext4_journalled_set_page_dirty(struct page *page)
3336 SetPageChecked(page);
3337 return __set_page_dirty_nobuffers(page);
3340 static const struct address_space_operations ext4_ordered_aops = {
3341 .readpage = ext4_readpage,
3342 .readpages = ext4_readpages,
3343 .writepage = ext4_writepage,
3344 .write_begin = ext4_write_begin,
3345 .write_end = ext4_ordered_write_end,
3347 .invalidatepage = ext4_invalidatepage,
3348 .releasepage = ext4_releasepage,
3349 .direct_IO = ext4_direct_IO,
3350 .migratepage = buffer_migrate_page,
3351 .is_partially_uptodate = block_is_partially_uptodate,
3352 .error_remove_page = generic_error_remove_page,
3355 static const struct address_space_operations ext4_writeback_aops = {
3356 .readpage = ext4_readpage,
3357 .readpages = ext4_readpages,
3358 .writepage = ext4_writepage,
3359 .write_begin = ext4_write_begin,
3360 .write_end = ext4_writeback_write_end,
3362 .invalidatepage = ext4_invalidatepage,
3363 .releasepage = ext4_releasepage,
3364 .direct_IO = ext4_direct_IO,
3365 .migratepage = buffer_migrate_page,
3366 .is_partially_uptodate = block_is_partially_uptodate,
3367 .error_remove_page = generic_error_remove_page,
3370 static const struct address_space_operations ext4_journalled_aops = {
3371 .readpage = ext4_readpage,
3372 .readpages = ext4_readpages,
3373 .writepage = ext4_writepage,
3374 .write_begin = ext4_write_begin,
3375 .write_end = ext4_journalled_write_end,
3376 .set_page_dirty = ext4_journalled_set_page_dirty,
3378 .invalidatepage = ext4_journalled_invalidatepage,
3379 .releasepage = ext4_releasepage,
3380 .direct_IO = ext4_direct_IO,
3381 .is_partially_uptodate = block_is_partially_uptodate,
3382 .error_remove_page = generic_error_remove_page,
3385 static const struct address_space_operations ext4_da_aops = {
3386 .readpage = ext4_readpage,
3387 .readpages = ext4_readpages,
3388 .writepage = ext4_writepage,
3389 .writepages = ext4_da_writepages,
3390 .write_begin = ext4_da_write_begin,
3391 .write_end = ext4_da_write_end,
3393 .invalidatepage = ext4_da_invalidatepage,
3394 .releasepage = ext4_releasepage,
3395 .direct_IO = ext4_direct_IO,
3396 .migratepage = buffer_migrate_page,
3397 .is_partially_uptodate = block_is_partially_uptodate,
3398 .error_remove_page = generic_error_remove_page,
3401 void ext4_set_aops(struct inode *inode)
3403 switch (ext4_inode_journal_mode(inode)) {
3404 case EXT4_INODE_ORDERED_DATA_MODE:
3405 if (test_opt(inode->i_sb, DELALLOC))
3406 inode->i_mapping->a_ops = &ext4_da_aops;
3408 inode->i_mapping->a_ops = &ext4_ordered_aops;
3410 case EXT4_INODE_WRITEBACK_DATA_MODE:
3411 if (test_opt(inode->i_sb, DELALLOC))
3412 inode->i_mapping->a_ops = &ext4_da_aops;
3414 inode->i_mapping->a_ops = &ext4_writeback_aops;
3416 case EXT4_INODE_JOURNAL_DATA_MODE:
3417 inode->i_mapping->a_ops = &ext4_journalled_aops;
3426 * ext4_discard_partial_page_buffers()
3427 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3428 * This function finds and locks the page containing the offset
3429 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3430 * Calling functions that already have the page locked should call
3431 * ext4_discard_partial_page_buffers_no_lock directly.
3433 int ext4_discard_partial_page_buffers(handle_t *handle,
3434 struct address_space *mapping, loff_t from,
3435 loff_t length, int flags)
3437 struct inode *inode = mapping->host;
3441 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3442 mapping_gfp_mask(mapping) & ~__GFP_FS);
3446 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3447 from, length, flags);
3450 page_cache_release(page);
3455 * ext4_discard_partial_page_buffers_no_lock()
3456 * Zeros a page range of length 'length' starting from offset 'from'.
3457 * Buffer heads that correspond to the block aligned regions of the
3458 * zeroed range will be unmapped. Unblock aligned regions
3459 * will have the corresponding buffer head mapped if needed so that
3460 * that region of the page can be updated with the partial zero out.
3462 * This function assumes that the page has already been locked. The
3463 * The range to be discarded must be contained with in the given page.
3464 * If the specified range exceeds the end of the page it will be shortened
3465 * to the end of the page that corresponds to 'from'. This function is
3466 * appropriate for updating a page and it buffer heads to be unmapped and
3467 * zeroed for blocks that have been either released, or are going to be
3470 * handle: The journal handle
3471 * inode: The files inode
3472 * page: A locked page that contains the offset "from"
3473 * from: The starting byte offset (from the beginning of the file)
3474 * to begin discarding
3475 * len: The length of bytes to discard
3476 * flags: Optional flags that may be used:
3478 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3479 * Only zero the regions of the page whose buffer heads
3480 * have already been unmapped. This flag is appropriate
3481 * for updating the contents of a page whose blocks may
3482 * have already been released, and we only want to zero
3483 * out the regions that correspond to those released blocks.
3485 * Returns zero on success or negative on failure.
3487 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3488 struct inode *inode, struct page *page, loff_t from,
3489 loff_t length, int flags)
3491 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3492 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3493 unsigned int blocksize, max, pos;
3495 struct buffer_head *bh;
3498 blocksize = inode->i_sb->s_blocksize;
3499 max = PAGE_CACHE_SIZE - offset;
3501 if (index != page->index)
3505 * correct length if it does not fall between
3506 * 'from' and the end of the page
3508 if (length > max || length < 0)
3511 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3513 if (!page_has_buffers(page))
3514 create_empty_buffers(page, blocksize, 0);
3516 /* Find the buffer that contains "offset" */
3517 bh = page_buffers(page);
3519 while (offset >= pos) {
3520 bh = bh->b_this_page;
3526 while (pos < offset + length) {
3527 unsigned int end_of_block, range_to_discard;
3531 /* The length of space left to zero and unmap */
3532 range_to_discard = offset + length - pos;
3534 /* The length of space until the end of the block */
3535 end_of_block = blocksize - (pos & (blocksize-1));
3538 * Do not unmap or zero past end of block
3539 * for this buffer head
3541 if (range_to_discard > end_of_block)
3542 range_to_discard = end_of_block;
3546 * Skip this buffer head if we are only zeroing unampped
3547 * regions of the page
3549 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3553 /* If the range is block aligned, unmap */
3554 if (range_to_discard == blocksize) {
3555 clear_buffer_dirty(bh);
3557 clear_buffer_mapped(bh);
3558 clear_buffer_req(bh);
3559 clear_buffer_new(bh);
3560 clear_buffer_delay(bh);
3561 clear_buffer_unwritten(bh);
3562 clear_buffer_uptodate(bh);
3563 zero_user(page, pos, range_to_discard);
3564 BUFFER_TRACE(bh, "Buffer discarded");
3569 * If this block is not completely contained in the range
3570 * to be discarded, then it is not going to be released. Because
3571 * we need to keep this block, we need to make sure this part
3572 * of the page is uptodate before we modify it by writeing
3573 * partial zeros on it.
3575 if (!buffer_mapped(bh)) {
3577 * Buffer head must be mapped before we can read
3580 BUFFER_TRACE(bh, "unmapped");
3581 ext4_get_block(inode, iblock, bh, 0);
3582 /* unmapped? It's a hole - nothing to do */
3583 if (!buffer_mapped(bh)) {
3584 BUFFER_TRACE(bh, "still unmapped");
3589 /* Ok, it's mapped. Make sure it's up-to-date */
3590 if (PageUptodate(page))
3591 set_buffer_uptodate(bh);
3593 if (!buffer_uptodate(bh)) {
3595 ll_rw_block(READ, 1, &bh);
3597 /* Uhhuh. Read error. Complain and punt.*/
3598 if (!buffer_uptodate(bh))
3602 if (ext4_should_journal_data(inode)) {
3603 BUFFER_TRACE(bh, "get write access");
3604 err = ext4_journal_get_write_access(handle, bh);
3609 zero_user(page, pos, range_to_discard);
3612 if (ext4_should_journal_data(inode)) {
3613 err = ext4_handle_dirty_metadata(handle, inode, bh);
3615 mark_buffer_dirty(bh);
3617 BUFFER_TRACE(bh, "Partial buffer zeroed");
3619 bh = bh->b_this_page;
3621 pos += range_to_discard;
3627 int ext4_can_truncate(struct inode *inode)
3629 if (S_ISREG(inode->i_mode))
3631 if (S_ISDIR(inode->i_mode))
3633 if (S_ISLNK(inode->i_mode))
3634 return !ext4_inode_is_fast_symlink(inode);
3639 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3640 * associated with the given offset and length
3642 * @inode: File inode
3643 * @offset: The offset where the hole will begin
3644 * @len: The length of the hole
3646 * Returns: 0 on success or negative on failure
3649 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3651 struct inode *inode = file->f_path.dentry->d_inode;
3652 if (!S_ISREG(inode->i_mode))
3655 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3656 return ext4_ind_punch_hole(file, offset, length);
3658 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3659 /* TODO: Add support for bigalloc file systems */
3663 trace_ext4_punch_hole(inode, offset, length);
3665 return ext4_ext_punch_hole(file, offset, length);
3671 * We block out ext4_get_block() block instantiations across the entire
3672 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3673 * simultaneously on behalf of the same inode.
3675 * As we work through the truncate and commit bits of it to the journal there
3676 * is one core, guiding principle: the file's tree must always be consistent on
3677 * disk. We must be able to restart the truncate after a crash.
3679 * The file's tree may be transiently inconsistent in memory (although it
3680 * probably isn't), but whenever we close off and commit a journal transaction,
3681 * the contents of (the filesystem + the journal) must be consistent and
3682 * restartable. It's pretty simple, really: bottom up, right to left (although
3683 * left-to-right works OK too).
3685 * Note that at recovery time, journal replay occurs *before* the restart of
3686 * truncate against the orphan inode list.
3688 * The committed inode has the new, desired i_size (which is the same as
3689 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3690 * that this inode's truncate did not complete and it will again call
3691 * ext4_truncate() to have another go. So there will be instantiated blocks
3692 * to the right of the truncation point in a crashed ext4 filesystem. But
3693 * that's fine - as long as they are linked from the inode, the post-crash
3694 * ext4_truncate() run will find them and release them.
3696 void ext4_truncate(struct inode *inode)
3698 trace_ext4_truncate_enter(inode);
3700 if (!ext4_can_truncate(inode))
3703 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3705 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3706 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3708 if (ext4_has_inline_data(inode)) {
3711 ext4_inline_data_truncate(inode, &has_inline);
3716 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3717 ext4_ext_truncate(inode);
3719 ext4_ind_truncate(inode);
3721 trace_ext4_truncate_exit(inode);
3725 * ext4_get_inode_loc returns with an extra refcount against the inode's
3726 * underlying buffer_head on success. If 'in_mem' is true, we have all
3727 * data in memory that is needed to recreate the on-disk version of this
3730 static int __ext4_get_inode_loc(struct inode *inode,
3731 struct ext4_iloc *iloc, int in_mem)
3733 struct ext4_group_desc *gdp;
3734 struct buffer_head *bh;
3735 struct super_block *sb = inode->i_sb;
3737 int inodes_per_block, inode_offset;
3740 if (!ext4_valid_inum(sb, inode->i_ino))
3743 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3744 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3749 * Figure out the offset within the block group inode table
3751 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3752 inode_offset = ((inode->i_ino - 1) %
3753 EXT4_INODES_PER_GROUP(sb));
3754 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3755 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3757 bh = sb_getblk(sb, block);
3760 if (!buffer_uptodate(bh)) {
3764 * If the buffer has the write error flag, we have failed
3765 * to write out another inode in the same block. In this
3766 * case, we don't have to read the block because we may
3767 * read the old inode data successfully.
3769 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3770 set_buffer_uptodate(bh);
3772 if (buffer_uptodate(bh)) {
3773 /* someone brought it uptodate while we waited */
3779 * If we have all information of the inode in memory and this
3780 * is the only valid inode in the block, we need not read the
3784 struct buffer_head *bitmap_bh;
3787 start = inode_offset & ~(inodes_per_block - 1);
3789 /* Is the inode bitmap in cache? */
3790 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3791 if (unlikely(!bitmap_bh))
3795 * If the inode bitmap isn't in cache then the
3796 * optimisation may end up performing two reads instead
3797 * of one, so skip it.
3799 if (!buffer_uptodate(bitmap_bh)) {
3803 for (i = start; i < start + inodes_per_block; i++) {
3804 if (i == inode_offset)
3806 if (ext4_test_bit(i, bitmap_bh->b_data))
3810 if (i == start + inodes_per_block) {
3811 /* all other inodes are free, so skip I/O */
3812 memset(bh->b_data, 0, bh->b_size);
3813 set_buffer_uptodate(bh);
3821 * If we need to do any I/O, try to pre-readahead extra
3822 * blocks from the inode table.
3824 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3825 ext4_fsblk_t b, end, table;
3828 table = ext4_inode_table(sb, gdp);
3829 /* s_inode_readahead_blks is always a power of 2 */
3830 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3833 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3834 num = EXT4_INODES_PER_GROUP(sb);
3835 if (ext4_has_group_desc_csum(sb))
3836 num -= ext4_itable_unused_count(sb, gdp);
3837 table += num / inodes_per_block;
3841 sb_breadahead(sb, b++);
3845 * There are other valid inodes in the buffer, this inode
3846 * has in-inode xattrs, or we don't have this inode in memory.
3847 * Read the block from disk.
3849 trace_ext4_load_inode(inode);
3851 bh->b_end_io = end_buffer_read_sync;
3852 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3854 if (!buffer_uptodate(bh)) {
3855 EXT4_ERROR_INODE_BLOCK(inode, block,
3856 "unable to read itable block");
3866 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3868 /* We have all inode data except xattrs in memory here. */
3869 return __ext4_get_inode_loc(inode, iloc,
3870 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3873 void ext4_set_inode_flags(struct inode *inode)
3875 unsigned int flags = EXT4_I(inode)->i_flags;
3877 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3878 if (flags & EXT4_SYNC_FL)
3879 inode->i_flags |= S_SYNC;
3880 if (flags & EXT4_APPEND_FL)
3881 inode->i_flags |= S_APPEND;
3882 if (flags & EXT4_IMMUTABLE_FL)
3883 inode->i_flags |= S_IMMUTABLE;
3884 if (flags & EXT4_NOATIME_FL)
3885 inode->i_flags |= S_NOATIME;
3886 if (flags & EXT4_DIRSYNC_FL)
3887 inode->i_flags |= S_DIRSYNC;
3890 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3891 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3893 unsigned int vfs_fl;
3894 unsigned long old_fl, new_fl;
3897 vfs_fl = ei->vfs_inode.i_flags;
3898 old_fl = ei->i_flags;
3899 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3900 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3902 if (vfs_fl & S_SYNC)
3903 new_fl |= EXT4_SYNC_FL;
3904 if (vfs_fl & S_APPEND)
3905 new_fl |= EXT4_APPEND_FL;
3906 if (vfs_fl & S_IMMUTABLE)
3907 new_fl |= EXT4_IMMUTABLE_FL;
3908 if (vfs_fl & S_NOATIME)
3909 new_fl |= EXT4_NOATIME_FL;
3910 if (vfs_fl & S_DIRSYNC)
3911 new_fl |= EXT4_DIRSYNC_FL;
3912 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3915 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3916 struct ext4_inode_info *ei)
3919 struct inode *inode = &(ei->vfs_inode);
3920 struct super_block *sb = inode->i_sb;
3922 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3923 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3924 /* we are using combined 48 bit field */
3925 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3926 le32_to_cpu(raw_inode->i_blocks_lo);
3927 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3928 /* i_blocks represent file system block size */
3929 return i_blocks << (inode->i_blkbits - 9);
3934 return le32_to_cpu(raw_inode->i_blocks_lo);
3938 static inline void ext4_iget_extra_inode(struct inode *inode,
3939 struct ext4_inode *raw_inode,
3940 struct ext4_inode_info *ei)
3942 __le32 *magic = (void *)raw_inode +
3943 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3944 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3945 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3946 ext4_find_inline_data_nolock(inode);
3948 EXT4_I(inode)->i_inline_off = 0;
3951 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3953 struct ext4_iloc iloc;
3954 struct ext4_inode *raw_inode;
3955 struct ext4_inode_info *ei;
3956 struct inode *inode;
3957 journal_t *journal = EXT4_SB(sb)->s_journal;
3963 inode = iget_locked(sb, ino);
3965 return ERR_PTR(-ENOMEM);
3966 if (!(inode->i_state & I_NEW))
3972 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3975 raw_inode = ext4_raw_inode(&iloc);
3977 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3978 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3979 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3980 EXT4_INODE_SIZE(inode->i_sb)) {
3981 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3982 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3983 EXT4_INODE_SIZE(inode->i_sb));
3988 ei->i_extra_isize = 0;
3990 /* Precompute checksum seed for inode metadata */
3991 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3992 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3993 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3995 __le32 inum = cpu_to_le32(inode->i_ino);
3996 __le32 gen = raw_inode->i_generation;
3997 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3999 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4003 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4004 EXT4_ERROR_INODE(inode, "checksum invalid");
4009 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4010 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4011 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4012 if (!(test_opt(inode->i_sb, NO_UID32))) {
4013 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4014 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4016 i_uid_write(inode, i_uid);
4017 i_gid_write(inode, i_gid);
4018 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4020 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4021 ei->i_inline_off = 0;
4022 ei->i_dir_start_lookup = 0;
4023 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4024 /* We now have enough fields to check if the inode was active or not.
4025 * This is needed because nfsd might try to access dead inodes
4026 * the test is that same one that e2fsck uses
4027 * NeilBrown 1999oct15
4029 if (inode->i_nlink == 0) {
4030 if (inode->i_mode == 0 ||
4031 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4032 /* this inode is deleted */
4036 /* The only unlinked inodes we let through here have
4037 * valid i_mode and are being read by the orphan
4038 * recovery code: that's fine, we're about to complete
4039 * the process of deleting those. */
4041 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4042 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4043 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4044 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4046 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4047 inode->i_size = ext4_isize(raw_inode);
4048 ei->i_disksize = inode->i_size;
4050 ei->i_reserved_quota = 0;
4052 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4053 ei->i_block_group = iloc.block_group;
4054 ei->i_last_alloc_group = ~0;
4056 * NOTE! The in-memory inode i_data array is in little-endian order
4057 * even on big-endian machines: we do NOT byteswap the block numbers!
4059 for (block = 0; block < EXT4_N_BLOCKS; block++)
4060 ei->i_data[block] = raw_inode->i_block[block];
4061 INIT_LIST_HEAD(&ei->i_orphan);
4064 * Set transaction id's of transactions that have to be committed
4065 * to finish f[data]sync. We set them to currently running transaction
4066 * as we cannot be sure that the inode or some of its metadata isn't
4067 * part of the transaction - the inode could have been reclaimed and
4068 * now it is reread from disk.
4071 transaction_t *transaction;
4074 read_lock(&journal->j_state_lock);
4075 if (journal->j_running_transaction)
4076 transaction = journal->j_running_transaction;
4078 transaction = journal->j_committing_transaction;
4080 tid = transaction->t_tid;
4082 tid = journal->j_commit_sequence;
4083 read_unlock(&journal->j_state_lock);
4084 ei->i_sync_tid = tid;
4085 ei->i_datasync_tid = tid;
4088 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4089 if (ei->i_extra_isize == 0) {
4090 /* The extra space is currently unused. Use it. */
4091 ei->i_extra_isize = sizeof(struct ext4_inode) -
4092 EXT4_GOOD_OLD_INODE_SIZE;
4094 ext4_iget_extra_inode(inode, raw_inode, ei);
4098 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4099 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4100 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4101 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4103 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4104 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4105 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4107 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4111 if (ei->i_file_acl &&
4112 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4113 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4117 } else if (!ext4_has_inline_data(inode)) {
4118 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4119 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4120 (S_ISLNK(inode->i_mode) &&
4121 !ext4_inode_is_fast_symlink(inode))))
4122 /* Validate extent which is part of inode */
4123 ret = ext4_ext_check_inode(inode);
4124 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4125 (S_ISLNK(inode->i_mode) &&
4126 !ext4_inode_is_fast_symlink(inode))) {
4127 /* Validate block references which are part of inode */
4128 ret = ext4_ind_check_inode(inode);
4134 if (S_ISREG(inode->i_mode)) {
4135 inode->i_op = &ext4_file_inode_operations;
4136 inode->i_fop = &ext4_file_operations;
4137 ext4_set_aops(inode);
4138 } else if (S_ISDIR(inode->i_mode)) {
4139 inode->i_op = &ext4_dir_inode_operations;
4140 inode->i_fop = &ext4_dir_operations;
4141 } else if (S_ISLNK(inode->i_mode)) {
4142 if (ext4_inode_is_fast_symlink(inode)) {
4143 inode->i_op = &ext4_fast_symlink_inode_operations;
4144 nd_terminate_link(ei->i_data, inode->i_size,
4145 sizeof(ei->i_data) - 1);
4147 inode->i_op = &ext4_symlink_inode_operations;
4148 ext4_set_aops(inode);
4150 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4151 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4152 inode->i_op = &ext4_special_inode_operations;
4153 if (raw_inode->i_block[0])
4154 init_special_inode(inode, inode->i_mode,
4155 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4157 init_special_inode(inode, inode->i_mode,
4158 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4161 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4165 ext4_set_inode_flags(inode);
4166 unlock_new_inode(inode);
4172 return ERR_PTR(ret);
4175 static int ext4_inode_blocks_set(handle_t *handle,
4176 struct ext4_inode *raw_inode,
4177 struct ext4_inode_info *ei)
4179 struct inode *inode = &(ei->vfs_inode);
4180 u64 i_blocks = inode->i_blocks;
4181 struct super_block *sb = inode->i_sb;
4183 if (i_blocks <= ~0U) {
4185 * i_blocks can be represented in a 32 bit variable
4186 * as multiple of 512 bytes
4188 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4189 raw_inode->i_blocks_high = 0;
4190 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4193 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4196 if (i_blocks <= 0xffffffffffffULL) {
4198 * i_blocks can be represented in a 48 bit variable
4199 * as multiple of 512 bytes
4201 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4202 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4203 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4205 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4206 /* i_block is stored in file system block size */
4207 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4208 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4209 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4215 * Post the struct inode info into an on-disk inode location in the
4216 * buffer-cache. This gobbles the caller's reference to the
4217 * buffer_head in the inode location struct.
4219 * The caller must have write access to iloc->bh.
4221 static int ext4_do_update_inode(handle_t *handle,
4222 struct inode *inode,
4223 struct ext4_iloc *iloc)
4225 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4226 struct ext4_inode_info *ei = EXT4_I(inode);
4227 struct buffer_head *bh = iloc->bh;
4228 int err = 0, rc, block;
4229 int need_datasync = 0;
4233 /* For fields not not tracking in the in-memory inode,
4234 * initialise them to zero for new inodes. */
4235 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4236 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4238 ext4_get_inode_flags(ei);
4239 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4240 i_uid = i_uid_read(inode);
4241 i_gid = i_gid_read(inode);
4242 if (!(test_opt(inode->i_sb, NO_UID32))) {
4243 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4244 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4246 * Fix up interoperability with old kernels. Otherwise, old inodes get
4247 * re-used with the upper 16 bits of the uid/gid intact
4250 raw_inode->i_uid_high =
4251 cpu_to_le16(high_16_bits(i_uid));
4252 raw_inode->i_gid_high =
4253 cpu_to_le16(high_16_bits(i_gid));
4255 raw_inode->i_uid_high = 0;
4256 raw_inode->i_gid_high = 0;
4259 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4260 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4261 raw_inode->i_uid_high = 0;
4262 raw_inode->i_gid_high = 0;
4264 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4266 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4267 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4268 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4269 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4271 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4273 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4274 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4275 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4276 cpu_to_le32(EXT4_OS_HURD))
4277 raw_inode->i_file_acl_high =
4278 cpu_to_le16(ei->i_file_acl >> 32);
4279 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4280 if (ei->i_disksize != ext4_isize(raw_inode)) {
4281 ext4_isize_set(raw_inode, ei->i_disksize);
4284 if (ei->i_disksize > 0x7fffffffULL) {
4285 struct super_block *sb = inode->i_sb;
4286 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4287 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4288 EXT4_SB(sb)->s_es->s_rev_level ==
4289 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4290 /* If this is the first large file
4291 * created, add a flag to the superblock.
4293 err = ext4_journal_get_write_access(handle,
4294 EXT4_SB(sb)->s_sbh);
4297 ext4_update_dynamic_rev(sb);
4298 EXT4_SET_RO_COMPAT_FEATURE(sb,
4299 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4300 ext4_handle_sync(handle);
4301 err = ext4_handle_dirty_super(handle, sb);
4304 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4305 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4306 if (old_valid_dev(inode->i_rdev)) {
4307 raw_inode->i_block[0] =
4308 cpu_to_le32(old_encode_dev(inode->i_rdev));
4309 raw_inode->i_block[1] = 0;
4311 raw_inode->i_block[0] = 0;
4312 raw_inode->i_block[1] =
4313 cpu_to_le32(new_encode_dev(inode->i_rdev));
4314 raw_inode->i_block[2] = 0;
4316 } else if (!ext4_has_inline_data(inode)) {
4317 for (block = 0; block < EXT4_N_BLOCKS; block++)
4318 raw_inode->i_block[block] = ei->i_data[block];
4321 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4322 if (ei->i_extra_isize) {
4323 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4324 raw_inode->i_version_hi =
4325 cpu_to_le32(inode->i_version >> 32);
4326 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4329 ext4_inode_csum_set(inode, raw_inode, ei);
4331 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4332 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4335 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4337 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4340 ext4_std_error(inode->i_sb, err);
4345 * ext4_write_inode()
4347 * We are called from a few places:
4349 * - Within generic_file_write() for O_SYNC files.
4350 * Here, there will be no transaction running. We wait for any running
4351 * transaction to commit.
4353 * - Within sys_sync(), kupdate and such.
4354 * We wait on commit, if tol to.
4356 * - Within prune_icache() (PF_MEMALLOC == true)
4357 * Here we simply return. We can't afford to block kswapd on the
4360 * In all cases it is actually safe for us to return without doing anything,
4361 * because the inode has been copied into a raw inode buffer in
4362 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4365 * Note that we are absolutely dependent upon all inode dirtiers doing the
4366 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4367 * which we are interested.
4369 * It would be a bug for them to not do this. The code:
4371 * mark_inode_dirty(inode)
4373 * inode->i_size = expr;
4375 * is in error because a kswapd-driven write_inode() could occur while
4376 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4377 * will no longer be on the superblock's dirty inode list.
4379 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4383 if (current->flags & PF_MEMALLOC)
4386 if (EXT4_SB(inode->i_sb)->s_journal) {
4387 if (ext4_journal_current_handle()) {
4388 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4393 if (wbc->sync_mode != WB_SYNC_ALL)
4396 err = ext4_force_commit(inode->i_sb);
4398 struct ext4_iloc iloc;
4400 err = __ext4_get_inode_loc(inode, &iloc, 0);
4403 if (wbc->sync_mode == WB_SYNC_ALL)
4404 sync_dirty_buffer(iloc.bh);
4405 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4406 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4407 "IO error syncing inode");
4416 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4417 * buffers that are attached to a page stradding i_size and are undergoing
4418 * commit. In that case we have to wait for commit to finish and try again.
4420 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4424 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4425 tid_t commit_tid = 0;
4428 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4430 * All buffers in the last page remain valid? Then there's nothing to
4431 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4434 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4437 page = find_lock_page(inode->i_mapping,
4438 inode->i_size >> PAGE_CACHE_SHIFT);
4441 ret = __ext4_journalled_invalidatepage(page, offset);
4443 page_cache_release(page);
4447 read_lock(&journal->j_state_lock);
4448 if (journal->j_committing_transaction)
4449 commit_tid = journal->j_committing_transaction->t_tid;
4450 read_unlock(&journal->j_state_lock);
4452 jbd2_log_wait_commit(journal, commit_tid);
4459 * Called from notify_change.
4461 * We want to trap VFS attempts to truncate the file as soon as
4462 * possible. In particular, we want to make sure that when the VFS
4463 * shrinks i_size, we put the inode on the orphan list and modify
4464 * i_disksize immediately, so that during the subsequent flushing of
4465 * dirty pages and freeing of disk blocks, we can guarantee that any
4466 * commit will leave the blocks being flushed in an unused state on
4467 * disk. (On recovery, the inode will get truncated and the blocks will
4468 * be freed, so we have a strong guarantee that no future commit will
4469 * leave these blocks visible to the user.)
4471 * Another thing we have to assure is that if we are in ordered mode
4472 * and inode is still attached to the committing transaction, we must
4473 * we start writeout of all the dirty pages which are being truncated.
4474 * This way we are sure that all the data written in the previous
4475 * transaction are already on disk (truncate waits for pages under
4478 * Called with inode->i_mutex down.
4480 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4482 struct inode *inode = dentry->d_inode;
4485 const unsigned int ia_valid = attr->ia_valid;
4487 error = inode_change_ok(inode, attr);
4491 if (is_quota_modification(inode, attr))
4492 dquot_initialize(inode);
4493 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4494 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4497 /* (user+group)*(old+new) structure, inode write (sb,
4498 * inode block, ? - but truncate inode update has it) */
4499 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4500 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4501 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4502 if (IS_ERR(handle)) {
4503 error = PTR_ERR(handle);
4506 error = dquot_transfer(inode, attr);
4508 ext4_journal_stop(handle);
4511 /* Update corresponding info in inode so that everything is in
4512 * one transaction */
4513 if (attr->ia_valid & ATTR_UID)
4514 inode->i_uid = attr->ia_uid;
4515 if (attr->ia_valid & ATTR_GID)
4516 inode->i_gid = attr->ia_gid;
4517 error = ext4_mark_inode_dirty(handle, inode);
4518 ext4_journal_stop(handle);
4521 if (attr->ia_valid & ATTR_SIZE) {
4523 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4524 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4526 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4531 if (S_ISREG(inode->i_mode) &&
4532 attr->ia_valid & ATTR_SIZE &&
4533 (attr->ia_size < inode->i_size)) {
4536 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4537 if (IS_ERR(handle)) {
4538 error = PTR_ERR(handle);
4541 if (ext4_handle_valid(handle)) {
4542 error = ext4_orphan_add(handle, inode);
4545 EXT4_I(inode)->i_disksize = attr->ia_size;
4546 rc = ext4_mark_inode_dirty(handle, inode);
4549 ext4_journal_stop(handle);
4551 if (ext4_should_order_data(inode)) {
4552 error = ext4_begin_ordered_truncate(inode,
4555 /* Do as much error cleanup as possible */
4556 handle = ext4_journal_start(inode,
4558 if (IS_ERR(handle)) {
4559 ext4_orphan_del(NULL, inode);
4562 ext4_orphan_del(handle, inode);
4564 ext4_journal_stop(handle);
4570 if (attr->ia_valid & ATTR_SIZE) {
4571 if (attr->ia_size != inode->i_size) {
4572 loff_t oldsize = inode->i_size;
4574 i_size_write(inode, attr->ia_size);
4576 * Blocks are going to be removed from the inode. Wait
4577 * for dio in flight. Temporarily disable
4578 * dioread_nolock to prevent livelock.
4581 if (!ext4_should_journal_data(inode)) {
4582 ext4_inode_block_unlocked_dio(inode);
4583 inode_dio_wait(inode);
4584 ext4_inode_resume_unlocked_dio(inode);
4586 ext4_wait_for_tail_page_commit(inode);
4589 * Truncate pagecache after we've waited for commit
4590 * in data=journal mode to make pages freeable.
4592 truncate_pagecache(inode, oldsize, inode->i_size);
4594 ext4_truncate(inode);
4598 setattr_copy(inode, attr);
4599 mark_inode_dirty(inode);
4603 * If the call to ext4_truncate failed to get a transaction handle at
4604 * all, we need to clean up the in-core orphan list manually.
4606 if (orphan && inode->i_nlink)
4607 ext4_orphan_del(NULL, inode);
4609 if (!rc && (ia_valid & ATTR_MODE))
4610 rc = ext4_acl_chmod(inode);
4613 ext4_std_error(inode->i_sb, error);
4619 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4622 struct inode *inode;
4623 unsigned long delalloc_blocks;
4625 inode = dentry->d_inode;
4626 generic_fillattr(inode, stat);
4629 * We can't update i_blocks if the block allocation is delayed
4630 * otherwise in the case of system crash before the real block
4631 * allocation is done, we will have i_blocks inconsistent with
4632 * on-disk file blocks.
4633 * We always keep i_blocks updated together with real
4634 * allocation. But to not confuse with user, stat
4635 * will return the blocks that include the delayed allocation
4636 * blocks for this file.
4638 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4639 EXT4_I(inode)->i_reserved_data_blocks);
4641 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4645 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4647 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4648 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4649 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4653 * Account for index blocks, block groups bitmaps and block group
4654 * descriptor blocks if modify datablocks and index blocks
4655 * worse case, the indexs blocks spread over different block groups
4657 * If datablocks are discontiguous, they are possible to spread over
4658 * different block groups too. If they are contiguous, with flexbg,
4659 * they could still across block group boundary.
4661 * Also account for superblock, inode, quota and xattr blocks
4663 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4665 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4671 * How many index blocks need to touch to modify nrblocks?
4672 * The "Chunk" flag indicating whether the nrblocks is
4673 * physically contiguous on disk
4675 * For Direct IO and fallocate, they calls get_block to allocate
4676 * one single extent at a time, so they could set the "Chunk" flag
4678 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4683 * Now let's see how many group bitmaps and group descriptors need
4693 if (groups > ngroups)
4695 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4696 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4698 /* bitmaps and block group descriptor blocks */
4699 ret += groups + gdpblocks;
4701 /* Blocks for super block, inode, quota and xattr blocks */
4702 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4708 * Calculate the total number of credits to reserve to fit
4709 * the modification of a single pages into a single transaction,
4710 * which may include multiple chunks of block allocations.
4712 * This could be called via ext4_write_begin()
4714 * We need to consider the worse case, when
4715 * one new block per extent.
4717 int ext4_writepage_trans_blocks(struct inode *inode)
4719 int bpp = ext4_journal_blocks_per_page(inode);
4722 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4724 /* Account for data blocks for journalled mode */
4725 if (ext4_should_journal_data(inode))
4731 * Calculate the journal credits for a chunk of data modification.
4733 * This is called from DIO, fallocate or whoever calling
4734 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4736 * journal buffers for data blocks are not included here, as DIO
4737 * and fallocate do no need to journal data buffers.
4739 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4741 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4745 * The caller must have previously called ext4_reserve_inode_write().
4746 * Give this, we know that the caller already has write access to iloc->bh.
4748 int ext4_mark_iloc_dirty(handle_t *handle,
4749 struct inode *inode, struct ext4_iloc *iloc)
4753 if (IS_I_VERSION(inode))
4754 inode_inc_iversion(inode);
4756 /* the do_update_inode consumes one bh->b_count */
4759 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4760 err = ext4_do_update_inode(handle, inode, iloc);
4766 * On success, We end up with an outstanding reference count against
4767 * iloc->bh. This _must_ be cleaned up later.
4771 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4772 struct ext4_iloc *iloc)
4776 err = ext4_get_inode_loc(inode, iloc);
4778 BUFFER_TRACE(iloc->bh, "get_write_access");
4779 err = ext4_journal_get_write_access(handle, iloc->bh);
4785 ext4_std_error(inode->i_sb, err);
4790 * Expand an inode by new_extra_isize bytes.
4791 * Returns 0 on success or negative error number on failure.
4793 static int ext4_expand_extra_isize(struct inode *inode,
4794 unsigned int new_extra_isize,
4795 struct ext4_iloc iloc,
4798 struct ext4_inode *raw_inode;
4799 struct ext4_xattr_ibody_header *header;
4801 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4804 raw_inode = ext4_raw_inode(&iloc);
4806 header = IHDR(inode, raw_inode);
4808 /* No extended attributes present */
4809 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4810 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4811 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4813 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4817 /* try to expand with EAs present */
4818 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4823 * What we do here is to mark the in-core inode as clean with respect to inode
4824 * dirtiness (it may still be data-dirty).
4825 * This means that the in-core inode may be reaped by prune_icache
4826 * without having to perform any I/O. This is a very good thing,
4827 * because *any* task may call prune_icache - even ones which
4828 * have a transaction open against a different journal.
4830 * Is this cheating? Not really. Sure, we haven't written the
4831 * inode out, but prune_icache isn't a user-visible syncing function.
4832 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4833 * we start and wait on commits.
4835 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4837 struct ext4_iloc iloc;
4838 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4839 static unsigned int mnt_count;
4843 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4844 err = ext4_reserve_inode_write(handle, inode, &iloc);
4845 if (ext4_handle_valid(handle) &&
4846 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4847 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4849 * We need extra buffer credits since we may write into EA block
4850 * with this same handle. If journal_extend fails, then it will
4851 * only result in a minor loss of functionality for that inode.
4852 * If this is felt to be critical, then e2fsck should be run to
4853 * force a large enough s_min_extra_isize.
4855 if ((jbd2_journal_extend(handle,
4856 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4857 ret = ext4_expand_extra_isize(inode,
4858 sbi->s_want_extra_isize,
4861 ext4_set_inode_state(inode,
4862 EXT4_STATE_NO_EXPAND);
4864 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4865 ext4_warning(inode->i_sb,
4866 "Unable to expand inode %lu. Delete"
4867 " some EAs or run e2fsck.",
4870 le16_to_cpu(sbi->s_es->s_mnt_count);
4876 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4881 * ext4_dirty_inode() is called from __mark_inode_dirty()
4883 * We're really interested in the case where a file is being extended.
4884 * i_size has been changed by generic_commit_write() and we thus need
4885 * to include the updated inode in the current transaction.
4887 * Also, dquot_alloc_block() will always dirty the inode when blocks
4888 * are allocated to the file.
4890 * If the inode is marked synchronous, we don't honour that here - doing
4891 * so would cause a commit on atime updates, which we don't bother doing.
4892 * We handle synchronous inodes at the highest possible level.
4894 void ext4_dirty_inode(struct inode *inode, int flags)
4898 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
4902 ext4_mark_inode_dirty(handle, inode);
4904 ext4_journal_stop(handle);
4911 * Bind an inode's backing buffer_head into this transaction, to prevent
4912 * it from being flushed to disk early. Unlike
4913 * ext4_reserve_inode_write, this leaves behind no bh reference and
4914 * returns no iloc structure, so the caller needs to repeat the iloc
4915 * lookup to mark the inode dirty later.
4917 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4919 struct ext4_iloc iloc;
4923 err = ext4_get_inode_loc(inode, &iloc);
4925 BUFFER_TRACE(iloc.bh, "get_write_access");
4926 err = jbd2_journal_get_write_access(handle, iloc.bh);
4928 err = ext4_handle_dirty_metadata(handle,
4934 ext4_std_error(inode->i_sb, err);
4939 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4946 * We have to be very careful here: changing a data block's
4947 * journaling status dynamically is dangerous. If we write a
4948 * data block to the journal, change the status and then delete
4949 * that block, we risk forgetting to revoke the old log record
4950 * from the journal and so a subsequent replay can corrupt data.
4951 * So, first we make sure that the journal is empty and that
4952 * nobody is changing anything.
4955 journal = EXT4_JOURNAL(inode);
4958 if (is_journal_aborted(journal))
4960 /* We have to allocate physical blocks for delalloc blocks
4961 * before flushing journal. otherwise delalloc blocks can not
4962 * be allocated any more. even more truncate on delalloc blocks
4963 * could trigger BUG by flushing delalloc blocks in journal.
4964 * There is no delalloc block in non-journal data mode.
4966 if (val && test_opt(inode->i_sb, DELALLOC)) {
4967 err = ext4_alloc_da_blocks(inode);
4972 /* Wait for all existing dio workers */
4973 ext4_inode_block_unlocked_dio(inode);
4974 inode_dio_wait(inode);
4976 jbd2_journal_lock_updates(journal);
4979 * OK, there are no updates running now, and all cached data is
4980 * synced to disk. We are now in a completely consistent state
4981 * which doesn't have anything in the journal, and we know that
4982 * no filesystem updates are running, so it is safe to modify
4983 * the inode's in-core data-journaling state flag now.
4987 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4989 jbd2_journal_flush(journal);
4990 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4992 ext4_set_aops(inode);
4994 jbd2_journal_unlock_updates(journal);
4995 ext4_inode_resume_unlocked_dio(inode);
4997 /* Finally we can mark the inode as dirty. */
4999 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5001 return PTR_ERR(handle);
5003 err = ext4_mark_inode_dirty(handle, inode);
5004 ext4_handle_sync(handle);
5005 ext4_journal_stop(handle);
5006 ext4_std_error(inode->i_sb, err);
5011 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5013 return !buffer_mapped(bh);
5016 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5018 struct page *page = vmf->page;
5022 struct file *file = vma->vm_file;
5023 struct inode *inode = file->f_path.dentry->d_inode;
5024 struct address_space *mapping = inode->i_mapping;
5026 get_block_t *get_block;
5029 sb_start_pagefault(inode->i_sb);
5030 file_update_time(vma->vm_file);
5031 /* Delalloc case is easy... */
5032 if (test_opt(inode->i_sb, DELALLOC) &&
5033 !ext4_should_journal_data(inode) &&
5034 !ext4_nonda_switch(inode->i_sb)) {
5036 ret = __block_page_mkwrite(vma, vmf,
5037 ext4_da_get_block_prep);
5038 } while (ret == -ENOSPC &&
5039 ext4_should_retry_alloc(inode->i_sb, &retries));
5044 size = i_size_read(inode);
5045 /* Page got truncated from under us? */
5046 if (page->mapping != mapping || page_offset(page) > size) {
5048 ret = VM_FAULT_NOPAGE;
5052 if (page->index == size >> PAGE_CACHE_SHIFT)
5053 len = size & ~PAGE_CACHE_MASK;
5055 len = PAGE_CACHE_SIZE;
5057 * Return if we have all the buffers mapped. This avoids the need to do
5058 * journal_start/journal_stop which can block and take a long time
5060 if (page_has_buffers(page)) {
5061 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5063 ext4_bh_unmapped)) {
5064 /* Wait so that we don't change page under IO */
5065 wait_on_page_writeback(page);
5066 ret = VM_FAULT_LOCKED;
5071 /* OK, we need to fill the hole... */
5072 if (ext4_should_dioread_nolock(inode))
5073 get_block = ext4_get_block_write;
5075 get_block = ext4_get_block;
5077 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5078 ext4_writepage_trans_blocks(inode));
5079 if (IS_ERR(handle)) {
5080 ret = VM_FAULT_SIGBUS;
5083 ret = __block_page_mkwrite(vma, vmf, get_block);
5084 if (!ret && ext4_should_journal_data(inode)) {
5085 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5086 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5088 ret = VM_FAULT_SIGBUS;
5089 ext4_journal_stop(handle);
5092 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5094 ext4_journal_stop(handle);
5095 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5098 ret = block_page_mkwrite_return(ret);
5100 sb_end_pagefault(inode->i_sb);