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>
40 #include <linux/aio.h>
41 #include <linux/bitops.h>
43 #include "ext4_jbd2.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
52 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
53 struct ext4_inode_info *ei)
55 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
60 csum_lo = le16_to_cpu(raw->i_checksum_lo);
61 raw->i_checksum_lo = 0;
62 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
63 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
64 csum_hi = le16_to_cpu(raw->i_checksum_hi);
65 raw->i_checksum_hi = 0;
68 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
69 EXT4_INODE_SIZE(inode->i_sb));
71 raw->i_checksum_lo = cpu_to_le16(csum_lo);
72 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
73 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
74 raw->i_checksum_hi = cpu_to_le16(csum_hi);
79 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
80 struct ext4_inode_info *ei)
82 __u32 provided, calculated;
84 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
85 cpu_to_le32(EXT4_OS_LINUX) ||
86 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
87 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
90 provided = le16_to_cpu(raw->i_checksum_lo);
91 calculated = ext4_inode_csum(inode, raw, ei);
92 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
93 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
94 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
98 return provided == calculated;
101 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
102 struct ext4_inode_info *ei)
106 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
107 cpu_to_le32(EXT4_OS_LINUX) ||
108 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
109 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
112 csum = ext4_inode_csum(inode, raw, ei);
113 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
114 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
115 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
116 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
119 static inline int ext4_begin_ordered_truncate(struct inode *inode,
122 trace_ext4_begin_ordered_truncate(inode, new_size);
124 * If jinode is zero, then we never opened the file for
125 * writing, so there's no need to call
126 * jbd2_journal_begin_ordered_truncate() since there's no
127 * outstanding writes we need to flush.
129 if (!EXT4_I(inode)->jinode)
131 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
132 EXT4_I(inode)->jinode,
136 static void ext4_invalidatepage(struct page *page, unsigned long offset);
137 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
138 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
139 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
140 struct inode *inode, struct page *page, loff_t from,
141 loff_t length, int flags);
144 * Test whether an inode is a fast symlink.
146 static int ext4_inode_is_fast_symlink(struct inode *inode)
148 int ea_blocks = EXT4_I(inode)->i_file_acl ?
149 (inode->i_sb->s_blocksize >> 9) : 0;
151 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
155 * Restart the transaction associated with *handle. This does a commit,
156 * so before we call here everything must be consistently dirtied against
159 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
165 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
166 * moment, get_block can be called only for blocks inside i_size since
167 * page cache has been already dropped and writes are blocked by
168 * i_mutex. So we can safely drop the i_data_sem here.
170 BUG_ON(EXT4_JOURNAL(inode) == NULL);
171 jbd_debug(2, "restarting handle %p\n", handle);
172 up_write(&EXT4_I(inode)->i_data_sem);
173 ret = ext4_journal_restart(handle, nblocks);
174 down_write(&EXT4_I(inode)->i_data_sem);
175 ext4_discard_preallocations(inode);
181 * Called at the last iput() if i_nlink is zero.
183 void ext4_evict_inode(struct inode *inode)
188 trace_ext4_evict_inode(inode);
190 if (inode->i_nlink) {
192 * When journalling data dirty buffers are tracked only in the
193 * journal. So although mm thinks everything is clean and
194 * ready for reaping the inode might still have some pages to
195 * write in the running transaction or waiting to be
196 * checkpointed. Thus calling jbd2_journal_invalidatepage()
197 * (via truncate_inode_pages()) to discard these buffers can
198 * cause data loss. Also even if we did not discard these
199 * buffers, we would have no way to find them after the inode
200 * is reaped and thus user could see stale data if he tries to
201 * read them before the transaction is checkpointed. So be
202 * careful and force everything to disk here... We use
203 * ei->i_datasync_tid to store the newest transaction
204 * containing inode's data.
206 * Note that directories do not have this problem because they
207 * don't use page cache.
209 if (ext4_should_journal_data(inode) &&
210 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
211 inode->i_ino != EXT4_JOURNAL_INO) {
212 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
213 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
215 jbd2_complete_transaction(journal, commit_tid);
216 filemap_write_and_wait(&inode->i_data);
218 truncate_inode_pages(&inode->i_data, 0);
219 ext4_ioend_shutdown(inode);
223 if (!is_bad_inode(inode))
224 dquot_initialize(inode);
226 if (ext4_should_order_data(inode))
227 ext4_begin_ordered_truncate(inode, 0);
228 truncate_inode_pages(&inode->i_data, 0);
229 ext4_ioend_shutdown(inode);
231 if (is_bad_inode(inode))
235 * Protect us against freezing - iput() caller didn't have to have any
236 * protection against it
238 sb_start_intwrite(inode->i_sb);
239 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
240 ext4_blocks_for_truncate(inode)+3);
241 if (IS_ERR(handle)) {
242 ext4_std_error(inode->i_sb, PTR_ERR(handle));
244 * If we're going to skip the normal cleanup, we still need to
245 * make sure that the in-core orphan linked list is properly
248 ext4_orphan_del(NULL, inode);
249 sb_end_intwrite(inode->i_sb);
254 ext4_handle_sync(handle);
256 err = ext4_mark_inode_dirty(handle, inode);
258 ext4_warning(inode->i_sb,
259 "couldn't mark inode dirty (err %d)", err);
263 ext4_truncate(inode);
266 * ext4_ext_truncate() doesn't reserve any slop when it
267 * restarts journal transactions; therefore there may not be
268 * enough credits left in the handle to remove the inode from
269 * the orphan list and set the dtime field.
271 if (!ext4_handle_has_enough_credits(handle, 3)) {
272 err = ext4_journal_extend(handle, 3);
274 err = ext4_journal_restart(handle, 3);
276 ext4_warning(inode->i_sb,
277 "couldn't extend journal (err %d)", err);
279 ext4_journal_stop(handle);
280 ext4_orphan_del(NULL, inode);
281 sb_end_intwrite(inode->i_sb);
287 * Kill off the orphan record which ext4_truncate created.
288 * AKPM: I think this can be inside the above `if'.
289 * Note that ext4_orphan_del() has to be able to cope with the
290 * deletion of a non-existent orphan - this is because we don't
291 * know if ext4_truncate() actually created an orphan record.
292 * (Well, we could do this if we need to, but heck - it works)
294 ext4_orphan_del(handle, inode);
295 EXT4_I(inode)->i_dtime = get_seconds();
298 * One subtle ordering requirement: if anything has gone wrong
299 * (transaction abort, IO errors, whatever), then we can still
300 * do these next steps (the fs will already have been marked as
301 * having errors), but we can't free the inode if the mark_dirty
304 if (ext4_mark_inode_dirty(handle, inode))
305 /* If that failed, just do the required in-core inode clear. */
306 ext4_clear_inode(inode);
308 ext4_free_inode(handle, inode);
309 ext4_journal_stop(handle);
310 sb_end_intwrite(inode->i_sb);
313 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
317 qsize_t *ext4_get_reserved_space(struct inode *inode)
319 return &EXT4_I(inode)->i_reserved_quota;
324 * Calculate the number of metadata blocks need to reserve
325 * to allocate a block located at @lblock
327 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
329 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
330 return ext4_ext_calc_metadata_amount(inode, lblock);
332 return ext4_ind_calc_metadata_amount(inode, lblock);
336 * Called with i_data_sem down, which is important since we can call
337 * ext4_discard_preallocations() from here.
339 void ext4_da_update_reserve_space(struct inode *inode,
340 int used, int quota_claim)
342 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
343 struct ext4_inode_info *ei = EXT4_I(inode);
345 spin_lock(&ei->i_block_reservation_lock);
346 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
347 if (unlikely(used > ei->i_reserved_data_blocks)) {
348 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
349 "with only %d reserved data blocks",
350 __func__, inode->i_ino, used,
351 ei->i_reserved_data_blocks);
353 used = ei->i_reserved_data_blocks;
356 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
357 ext4_warning(inode->i_sb, "ino %lu, allocated %d "
358 "with only %d reserved metadata blocks "
359 "(releasing %d blocks with reserved %d data blocks)",
360 inode->i_ino, ei->i_allocated_meta_blocks,
361 ei->i_reserved_meta_blocks, used,
362 ei->i_reserved_data_blocks);
364 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
367 /* Update per-inode reservations */
368 ei->i_reserved_data_blocks -= used;
369 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
370 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
371 used + ei->i_allocated_meta_blocks);
372 ei->i_allocated_meta_blocks = 0;
374 if (ei->i_reserved_data_blocks == 0) {
376 * We can release all of the reserved metadata blocks
377 * only when we have written all of the delayed
380 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
381 ei->i_reserved_meta_blocks);
382 ei->i_reserved_meta_blocks = 0;
383 ei->i_da_metadata_calc_len = 0;
385 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
387 /* Update quota subsystem for data blocks */
389 dquot_claim_block(inode, EXT4_C2B(sbi, used));
392 * We did fallocate with an offset that is already delayed
393 * allocated. So on delayed allocated writeback we should
394 * not re-claim the quota for fallocated blocks.
396 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
400 * If we have done all the pending block allocations and if
401 * there aren't any writers on the inode, we can discard the
402 * inode's preallocations.
404 if ((ei->i_reserved_data_blocks == 0) &&
405 (atomic_read(&inode->i_writecount) == 0))
406 ext4_discard_preallocations(inode);
409 static int __check_block_validity(struct inode *inode, const char *func,
411 struct ext4_map_blocks *map)
413 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
415 ext4_error_inode(inode, func, line, map->m_pblk,
416 "lblock %lu mapped to illegal pblock "
417 "(length %d)", (unsigned long) map->m_lblk,
424 #define check_block_validity(inode, map) \
425 __check_block_validity((inode), __func__, __LINE__, (map))
428 * Return the number of contiguous dirty pages in a given inode
429 * starting at page frame idx.
431 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
432 unsigned int max_pages)
434 struct address_space *mapping = inode->i_mapping;
438 int i, nr_pages, done = 0;
442 pagevec_init(&pvec, 0);
445 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
447 (pgoff_t)PAGEVEC_SIZE);
450 for (i = 0; i < nr_pages; i++) {
451 struct page *page = pvec.pages[i];
452 struct buffer_head *bh, *head;
455 if (unlikely(page->mapping != mapping) ||
457 PageWriteback(page) ||
458 page->index != idx) {
463 if (page_has_buffers(page)) {
464 bh = head = page_buffers(page);
466 if (!buffer_delay(bh) &&
467 !buffer_unwritten(bh))
469 bh = bh->b_this_page;
470 } while (!done && (bh != head));
477 if (num >= max_pages) {
482 pagevec_release(&pvec);
487 #ifdef ES_AGGRESSIVE_TEST
488 static void ext4_map_blocks_es_recheck(handle_t *handle,
490 struct ext4_map_blocks *es_map,
491 struct ext4_map_blocks *map,
498 * There is a race window that the result is not the same.
499 * e.g. xfstests #223 when dioread_nolock enables. The reason
500 * is that we lookup a block mapping in extent status tree with
501 * out taking i_data_sem. So at the time the unwritten extent
502 * could be converted.
504 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
505 down_read((&EXT4_I(inode)->i_data_sem));
506 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
507 retval = ext4_ext_map_blocks(handle, inode, map, flags &
508 EXT4_GET_BLOCKS_KEEP_SIZE);
510 retval = ext4_ind_map_blocks(handle, inode, map, flags &
511 EXT4_GET_BLOCKS_KEEP_SIZE);
513 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
514 up_read((&EXT4_I(inode)->i_data_sem));
516 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
517 * because it shouldn't be marked in es_map->m_flags.
519 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
522 * We don't check m_len because extent will be collpased in status
523 * tree. So the m_len might not equal.
525 if (es_map->m_lblk != map->m_lblk ||
526 es_map->m_flags != map->m_flags ||
527 es_map->m_pblk != map->m_pblk) {
528 printk("ES cache assertation failed for inode: %lu "
529 "es_cached ex [%d/%d/%llu/%x] != "
530 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
531 inode->i_ino, es_map->m_lblk, es_map->m_len,
532 es_map->m_pblk, es_map->m_flags, map->m_lblk,
533 map->m_len, map->m_pblk, map->m_flags,
537 #endif /* ES_AGGRESSIVE_TEST */
540 * The ext4_map_blocks() function tries to look up the requested blocks,
541 * and returns if the blocks are already mapped.
543 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
544 * and store the allocated blocks in the result buffer head and mark it
547 * If file type is extents based, it will call ext4_ext_map_blocks(),
548 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
551 * On success, it returns the number of blocks being mapped or allocate.
552 * if create==0 and the blocks are pre-allocated and uninitialized block,
553 * the result buffer head is unmapped. If the create ==1, it will make sure
554 * the buffer head is mapped.
556 * It returns 0 if plain look up failed (blocks have not been allocated), in
557 * that case, buffer head is unmapped
559 * It returns the error in case of allocation failure.
561 int ext4_map_blocks(handle_t *handle, struct inode *inode,
562 struct ext4_map_blocks *map, int flags)
564 struct extent_status es;
566 #ifdef ES_AGGRESSIVE_TEST
567 struct ext4_map_blocks orig_map;
569 memcpy(&orig_map, map, sizeof(*map));
573 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
574 "logical block %lu\n", inode->i_ino, flags, map->m_len,
575 (unsigned long) map->m_lblk);
577 /* Lookup extent status tree firstly */
578 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
579 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
580 map->m_pblk = ext4_es_pblock(&es) +
581 map->m_lblk - es.es_lblk;
582 map->m_flags |= ext4_es_is_written(&es) ?
583 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
584 retval = es.es_len - (map->m_lblk - es.es_lblk);
585 if (retval > map->m_len)
588 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
593 #ifdef ES_AGGRESSIVE_TEST
594 ext4_map_blocks_es_recheck(handle, inode, map,
601 * Try to see if we can get the block without requesting a new
604 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
605 down_read((&EXT4_I(inode)->i_data_sem));
606 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
607 retval = ext4_ext_map_blocks(handle, inode, map, flags &
608 EXT4_GET_BLOCKS_KEEP_SIZE);
610 retval = ext4_ind_map_blocks(handle, inode, map, flags &
611 EXT4_GET_BLOCKS_KEEP_SIZE);
615 unsigned long long status;
617 #ifdef ES_AGGRESSIVE_TEST
618 if (retval != map->m_len) {
619 printk("ES len assertation failed for inode: %lu "
620 "retval %d != map->m_len %d "
621 "in %s (lookup)\n", inode->i_ino, retval,
622 map->m_len, __func__);
626 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
627 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
628 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
629 !(status & EXTENT_STATUS_WRITTEN) &&
630 ext4_find_delalloc_range(inode, map->m_lblk,
631 map->m_lblk + map->m_len - 1))
632 status |= EXTENT_STATUS_DELAYED;
633 ret = ext4_es_insert_extent(inode, map->m_lblk,
634 map->m_len, map->m_pblk, status);
638 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
639 up_read((&EXT4_I(inode)->i_data_sem));
642 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
643 int ret = check_block_validity(inode, map);
648 /* If it is only a block(s) look up */
649 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
653 * Returns if the blocks have already allocated
655 * Note that if blocks have been preallocated
656 * ext4_ext_get_block() returns the create = 0
657 * with buffer head unmapped.
659 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
663 * Here we clear m_flags because after allocating an new extent,
664 * it will be set again.
666 map->m_flags &= ~EXT4_MAP_FLAGS;
669 * New blocks allocate and/or writing to uninitialized extent
670 * will possibly result in updating i_data, so we take
671 * the write lock of i_data_sem, and call get_blocks()
672 * with create == 1 flag.
674 down_write((&EXT4_I(inode)->i_data_sem));
677 * if the caller is from delayed allocation writeout path
678 * we have already reserved fs blocks for allocation
679 * let the underlying get_block() function know to
680 * avoid double accounting
682 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
683 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
685 * We need to check for EXT4 here because migrate
686 * could have changed the inode type in between
688 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
689 retval = ext4_ext_map_blocks(handle, inode, map, flags);
691 retval = ext4_ind_map_blocks(handle, inode, map, flags);
693 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
695 * We allocated new blocks which will result in
696 * i_data's format changing. Force the migrate
697 * to fail by clearing migrate flags
699 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
703 * Update reserved blocks/metadata blocks after successful
704 * block allocation which had been deferred till now. We don't
705 * support fallocate for non extent files. So we can update
706 * reserve space here.
709 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
710 ext4_da_update_reserve_space(inode, retval, 1);
712 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
713 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
717 unsigned long long status;
719 #ifdef ES_AGGRESSIVE_TEST
720 if (retval != map->m_len) {
721 printk("ES len assertation failed for inode: %lu "
722 "retval %d != map->m_len %d "
723 "in %s (allocation)\n", inode->i_ino, retval,
724 map->m_len, __func__);
729 * If the extent has been zeroed out, we don't need to update
730 * extent status tree.
732 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
733 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
734 if (ext4_es_is_written(&es))
737 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
738 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
739 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
740 !(status & EXTENT_STATUS_WRITTEN) &&
741 ext4_find_delalloc_range(inode, map->m_lblk,
742 map->m_lblk + map->m_len - 1))
743 status |= EXTENT_STATUS_DELAYED;
744 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
745 map->m_pblk, status);
751 up_write((&EXT4_I(inode)->i_data_sem));
752 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
753 int ret = check_block_validity(inode, map);
760 /* Maximum number of blocks we map for direct IO at once. */
761 #define DIO_MAX_BLOCKS 4096
763 static int _ext4_get_block(struct inode *inode, sector_t iblock,
764 struct buffer_head *bh, int flags)
766 handle_t *handle = ext4_journal_current_handle();
767 struct ext4_map_blocks map;
768 int ret = 0, started = 0;
771 if (ext4_has_inline_data(inode))
775 map.m_len = bh->b_size >> inode->i_blkbits;
777 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
778 /* Direct IO write... */
779 if (map.m_len > DIO_MAX_BLOCKS)
780 map.m_len = DIO_MAX_BLOCKS;
781 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
782 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
784 if (IS_ERR(handle)) {
785 ret = PTR_ERR(handle);
791 ret = ext4_map_blocks(handle, inode, &map, flags);
793 map_bh(bh, inode->i_sb, map.m_pblk);
794 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
795 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
799 ext4_journal_stop(handle);
803 int ext4_get_block(struct inode *inode, sector_t iblock,
804 struct buffer_head *bh, int create)
806 return _ext4_get_block(inode, iblock, bh,
807 create ? EXT4_GET_BLOCKS_CREATE : 0);
811 * `handle' can be NULL if create is zero
813 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
814 ext4_lblk_t block, int create, int *errp)
816 struct ext4_map_blocks map;
817 struct buffer_head *bh;
820 J_ASSERT(handle != NULL || create == 0);
824 err = ext4_map_blocks(handle, inode, &map,
825 create ? EXT4_GET_BLOCKS_CREATE : 0);
827 /* ensure we send some value back into *errp */
830 if (create && err == 0)
831 err = -ENOSPC; /* should never happen */
837 bh = sb_getblk(inode->i_sb, map.m_pblk);
842 if (map.m_flags & EXT4_MAP_NEW) {
843 J_ASSERT(create != 0);
844 J_ASSERT(handle != NULL);
847 * Now that we do not always journal data, we should
848 * keep in mind whether this should always journal the
849 * new buffer as metadata. For now, regular file
850 * writes use ext4_get_block instead, so it's not a
854 BUFFER_TRACE(bh, "call get_create_access");
855 fatal = ext4_journal_get_create_access(handle, bh);
856 if (!fatal && !buffer_uptodate(bh)) {
857 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
858 set_buffer_uptodate(bh);
861 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
862 err = ext4_handle_dirty_metadata(handle, inode, bh);
866 BUFFER_TRACE(bh, "not a new buffer");
876 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
877 ext4_lblk_t block, int create, int *err)
879 struct buffer_head *bh;
881 bh = ext4_getblk(handle, inode, block, create, err);
884 if (buffer_uptodate(bh))
886 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
888 if (buffer_uptodate(bh))
895 int ext4_walk_page_buffers(handle_t *handle,
896 struct buffer_head *head,
900 int (*fn)(handle_t *handle,
901 struct buffer_head *bh))
903 struct buffer_head *bh;
904 unsigned block_start, block_end;
905 unsigned blocksize = head->b_size;
907 struct buffer_head *next;
909 for (bh = head, block_start = 0;
910 ret == 0 && (bh != head || !block_start);
911 block_start = block_end, bh = next) {
912 next = bh->b_this_page;
913 block_end = block_start + blocksize;
914 if (block_end <= from || block_start >= to) {
915 if (partial && !buffer_uptodate(bh))
919 err = (*fn)(handle, bh);
927 * To preserve ordering, it is essential that the hole instantiation and
928 * the data write be encapsulated in a single transaction. We cannot
929 * close off a transaction and start a new one between the ext4_get_block()
930 * and the commit_write(). So doing the jbd2_journal_start at the start of
931 * prepare_write() is the right place.
933 * Also, this function can nest inside ext4_writepage(). In that case, we
934 * *know* that ext4_writepage() has generated enough buffer credits to do the
935 * whole page. So we won't block on the journal in that case, which is good,
936 * because the caller may be PF_MEMALLOC.
938 * By accident, ext4 can be reentered when a transaction is open via
939 * quota file writes. If we were to commit the transaction while thus
940 * reentered, there can be a deadlock - we would be holding a quota
941 * lock, and the commit would never complete if another thread had a
942 * transaction open and was blocking on the quota lock - a ranking
945 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
946 * will _not_ run commit under these circumstances because handle->h_ref
947 * is elevated. We'll still have enough credits for the tiny quotafile
950 int do_journal_get_write_access(handle_t *handle,
951 struct buffer_head *bh)
953 int dirty = buffer_dirty(bh);
956 if (!buffer_mapped(bh) || buffer_freed(bh))
959 * __block_write_begin() could have dirtied some buffers. Clean
960 * the dirty bit as jbd2_journal_get_write_access() could complain
961 * otherwise about fs integrity issues. Setting of the dirty bit
962 * by __block_write_begin() isn't a real problem here as we clear
963 * the bit before releasing a page lock and thus writeback cannot
964 * ever write the buffer.
967 clear_buffer_dirty(bh);
968 ret = ext4_journal_get_write_access(handle, bh);
970 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
974 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
975 struct buffer_head *bh_result, int create);
976 static int ext4_write_begin(struct file *file, struct address_space *mapping,
977 loff_t pos, unsigned len, unsigned flags,
978 struct page **pagep, void **fsdata)
980 struct inode *inode = mapping->host;
981 int ret, needed_blocks;
988 trace_ext4_write_begin(inode, pos, len, flags);
990 * Reserve one block more for addition to orphan list in case
991 * we allocate blocks but write fails for some reason
993 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
994 index = pos >> PAGE_CACHE_SHIFT;
995 from = pos & (PAGE_CACHE_SIZE - 1);
998 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
999 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1008 * grab_cache_page_write_begin() can take a long time if the
1009 * system is thrashing due to memory pressure, or if the page
1010 * is being written back. So grab it first before we start
1011 * the transaction handle. This also allows us to allocate
1012 * the page (if needed) without using GFP_NOFS.
1015 page = grab_cache_page_write_begin(mapping, index, flags);
1021 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1022 if (IS_ERR(handle)) {
1023 page_cache_release(page);
1024 return PTR_ERR(handle);
1028 if (page->mapping != mapping) {
1029 /* The page got truncated from under us */
1031 page_cache_release(page);
1032 ext4_journal_stop(handle);
1035 /* In case writeback began while the page was unlocked */
1036 wait_for_stable_page(page);
1038 if (ext4_should_dioread_nolock(inode))
1039 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1041 ret = __block_write_begin(page, pos, len, ext4_get_block);
1043 if (!ret && ext4_should_journal_data(inode)) {
1044 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1046 do_journal_get_write_access);
1052 * __block_write_begin may have instantiated a few blocks
1053 * outside i_size. Trim these off again. Don't need
1054 * i_size_read because we hold i_mutex.
1056 * Add inode to orphan list in case we crash before
1059 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1060 ext4_orphan_add(handle, inode);
1062 ext4_journal_stop(handle);
1063 if (pos + len > inode->i_size) {
1064 ext4_truncate_failed_write(inode);
1066 * If truncate failed early the inode might
1067 * still be on the orphan list; we need to
1068 * make sure the inode is removed from the
1069 * orphan list in that case.
1072 ext4_orphan_del(NULL, inode);
1075 if (ret == -ENOSPC &&
1076 ext4_should_retry_alloc(inode->i_sb, &retries))
1078 page_cache_release(page);
1085 /* For write_end() in data=journal mode */
1086 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1089 if (!buffer_mapped(bh) || buffer_freed(bh))
1091 set_buffer_uptodate(bh);
1092 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1093 clear_buffer_meta(bh);
1094 clear_buffer_prio(bh);
1099 * We need to pick up the new inode size which generic_commit_write gave us
1100 * `file' can be NULL - eg, when called from page_symlink().
1102 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1103 * buffers are managed internally.
1105 static int ext4_write_end(struct file *file,
1106 struct address_space *mapping,
1107 loff_t pos, unsigned len, unsigned copied,
1108 struct page *page, void *fsdata)
1110 handle_t *handle = ext4_journal_current_handle();
1111 struct inode *inode = mapping->host;
1113 int i_size_changed = 0;
1115 trace_ext4_write_end(inode, pos, len, copied);
1116 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1117 ret = ext4_jbd2_file_inode(handle, inode);
1120 page_cache_release(page);
1125 if (ext4_has_inline_data(inode)) {
1126 ret = ext4_write_inline_data_end(inode, pos, len,
1132 copied = block_write_end(file, mapping, pos,
1133 len, copied, page, fsdata);
1136 * No need to use i_size_read() here, the i_size
1137 * cannot change under us because we hole i_mutex.
1139 * But it's important to update i_size while still holding page lock:
1140 * page writeout could otherwise come in and zero beyond i_size.
1142 if (pos + copied > inode->i_size) {
1143 i_size_write(inode, pos + copied);
1147 if (pos + copied > EXT4_I(inode)->i_disksize) {
1148 /* We need to mark inode dirty even if
1149 * new_i_size is less that inode->i_size
1150 * but greater than i_disksize. (hint delalloc)
1152 ext4_update_i_disksize(inode, (pos + copied));
1156 page_cache_release(page);
1159 * Don't mark the inode dirty under page lock. First, it unnecessarily
1160 * makes the holding time of page lock longer. Second, it forces lock
1161 * ordering of page lock and transaction start for journaling
1165 ext4_mark_inode_dirty(handle, inode);
1169 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1170 /* if we have allocated more blocks and copied
1171 * less. We will have blocks allocated outside
1172 * inode->i_size. So truncate them
1174 ext4_orphan_add(handle, inode);
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);
1191 return ret ? ret : copied;
1194 static int ext4_journalled_write_end(struct file *file,
1195 struct address_space *mapping,
1196 loff_t pos, unsigned len, unsigned copied,
1197 struct page *page, void *fsdata)
1199 handle_t *handle = ext4_journal_current_handle();
1200 struct inode *inode = mapping->host;
1206 trace_ext4_journalled_write_end(inode, pos, len, copied);
1207 from = pos & (PAGE_CACHE_SIZE - 1);
1210 BUG_ON(!ext4_handle_valid(handle));
1212 if (ext4_has_inline_data(inode))
1213 copied = ext4_write_inline_data_end(inode, pos, len,
1217 if (!PageUptodate(page))
1219 page_zero_new_buffers(page, from+copied, to);
1222 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1223 to, &partial, write_end_fn);
1225 SetPageUptodate(page);
1227 new_i_size = pos + copied;
1228 if (new_i_size > inode->i_size)
1229 i_size_write(inode, pos+copied);
1230 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1231 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1232 if (new_i_size > EXT4_I(inode)->i_disksize) {
1233 ext4_update_i_disksize(inode, new_i_size);
1234 ret2 = ext4_mark_inode_dirty(handle, inode);
1240 page_cache_release(page);
1241 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1242 /* if we have allocated more blocks and copied
1243 * less. We will have blocks allocated outside
1244 * inode->i_size. So truncate them
1246 ext4_orphan_add(handle, inode);
1248 ret2 = ext4_journal_stop(handle);
1251 if (pos + len > inode->i_size) {
1252 ext4_truncate_failed_write(inode);
1254 * If truncate failed early the inode might still be
1255 * on the orphan list; we need to make sure the inode
1256 * is removed from the orphan list in that case.
1259 ext4_orphan_del(NULL, inode);
1262 return ret ? ret : copied;
1266 * Reserve a metadata for a single block located at lblock
1268 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1270 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1271 struct ext4_inode_info *ei = EXT4_I(inode);
1272 unsigned int md_needed;
1273 ext4_lblk_t save_last_lblock;
1277 * recalculate the amount of metadata blocks to reserve
1278 * in order to allocate nrblocks
1279 * worse case is one extent per block
1281 spin_lock(&ei->i_block_reservation_lock);
1283 * ext4_calc_metadata_amount() has side effects, which we have
1284 * to be prepared undo if we fail to claim space.
1286 save_len = ei->i_da_metadata_calc_len;
1287 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1288 md_needed = EXT4_NUM_B2C(sbi,
1289 ext4_calc_metadata_amount(inode, lblock));
1290 trace_ext4_da_reserve_space(inode, md_needed);
1293 * We do still charge estimated metadata to the sb though;
1294 * we cannot afford to run out of free blocks.
1296 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1297 ei->i_da_metadata_calc_len = save_len;
1298 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1299 spin_unlock(&ei->i_block_reservation_lock);
1302 ei->i_reserved_meta_blocks += md_needed;
1303 spin_unlock(&ei->i_block_reservation_lock);
1305 return 0; /* success */
1309 * Reserve a single cluster located at lblock
1311 static int ext4_da_reserve_space(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;
1317 ext4_lblk_t save_last_lblock;
1321 * We will charge metadata quota at writeout time; this saves
1322 * us from metadata over-estimation, though we may go over by
1323 * a small amount in the end. Here we just reserve for data.
1325 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1330 * recalculate the amount of metadata blocks to reserve
1331 * in order to allocate nrblocks
1332 * worse case is one extent per block
1334 spin_lock(&ei->i_block_reservation_lock);
1336 * ext4_calc_metadata_amount() has side effects, which we have
1337 * to be prepared undo if we fail to claim space.
1339 save_len = ei->i_da_metadata_calc_len;
1340 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1341 md_needed = EXT4_NUM_B2C(sbi,
1342 ext4_calc_metadata_amount(inode, lblock));
1343 trace_ext4_da_reserve_space(inode, md_needed);
1346 * We do still charge estimated metadata to the sb though;
1347 * we cannot afford to run out of free blocks.
1349 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1350 ei->i_da_metadata_calc_len = save_len;
1351 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1352 spin_unlock(&ei->i_block_reservation_lock);
1353 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1356 ei->i_reserved_data_blocks++;
1357 ei->i_reserved_meta_blocks += md_needed;
1358 spin_unlock(&ei->i_block_reservation_lock);
1360 return 0; /* success */
1363 static void ext4_da_release_space(struct inode *inode, int to_free)
1365 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1366 struct ext4_inode_info *ei = EXT4_I(inode);
1369 return; /* Nothing to release, exit */
1371 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1373 trace_ext4_da_release_space(inode, to_free);
1374 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1376 * if there aren't enough reserved blocks, then the
1377 * counter is messed up somewhere. Since this
1378 * function is called from invalidate page, it's
1379 * harmless to return without any action.
1381 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1382 "ino %lu, to_free %d with only %d reserved "
1383 "data blocks", inode->i_ino, to_free,
1384 ei->i_reserved_data_blocks);
1386 to_free = ei->i_reserved_data_blocks;
1388 ei->i_reserved_data_blocks -= to_free;
1390 if (ei->i_reserved_data_blocks == 0) {
1392 * We can release all of the reserved metadata blocks
1393 * only when we have written all of the delayed
1394 * allocation blocks.
1395 * Note that in case of bigalloc, i_reserved_meta_blocks,
1396 * i_reserved_data_blocks, etc. refer to number of clusters.
1398 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1399 ei->i_reserved_meta_blocks);
1400 ei->i_reserved_meta_blocks = 0;
1401 ei->i_da_metadata_calc_len = 0;
1404 /* update fs dirty data blocks counter */
1405 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1407 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1409 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1412 static void ext4_da_page_release_reservation(struct page *page,
1413 unsigned long offset)
1416 struct buffer_head *head, *bh;
1417 unsigned int curr_off = 0;
1418 struct inode *inode = page->mapping->host;
1419 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1423 head = page_buffers(page);
1426 unsigned int next_off = curr_off + bh->b_size;
1428 if ((offset <= curr_off) && (buffer_delay(bh))) {
1430 clear_buffer_delay(bh);
1432 curr_off = next_off;
1433 } while ((bh = bh->b_this_page) != head);
1436 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1437 ext4_es_remove_extent(inode, lblk, to_release);
1440 /* If we have released all the blocks belonging to a cluster, then we
1441 * need to release the reserved space for that cluster. */
1442 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1443 while (num_clusters > 0) {
1444 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1445 ((num_clusters - 1) << sbi->s_cluster_bits);
1446 if (sbi->s_cluster_ratio == 1 ||
1447 !ext4_find_delalloc_cluster(inode, lblk))
1448 ext4_da_release_space(inode, 1);
1455 * Delayed allocation stuff
1459 * mpage_da_submit_io - walks through extent of pages and try to write
1460 * them with writepage() call back
1462 * @mpd->inode: inode
1463 * @mpd->first_page: first page of the extent
1464 * @mpd->next_page: page after the last page of the extent
1466 * By the time mpage_da_submit_io() is called we expect all blocks
1467 * to be allocated. this may be wrong if allocation failed.
1469 * As pages are already locked by write_cache_pages(), we can't use it
1471 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1472 struct ext4_map_blocks *map)
1474 struct pagevec pvec;
1475 unsigned long index, end;
1476 int ret = 0, err, nr_pages, i;
1477 struct inode *inode = mpd->inode;
1478 struct address_space *mapping = inode->i_mapping;
1479 loff_t size = i_size_read(inode);
1480 unsigned int len, block_start;
1481 struct buffer_head *bh, *page_bufs = NULL;
1482 sector_t pblock = 0, cur_logical = 0;
1483 struct ext4_io_submit io_submit;
1485 BUG_ON(mpd->next_page <= mpd->first_page);
1486 memset(&io_submit, 0, sizeof(io_submit));
1488 * We need to start from the first_page to the next_page - 1
1489 * to make sure we also write the mapped dirty buffer_heads.
1490 * If we look at mpd->b_blocknr we would only be looking
1491 * at the currently mapped buffer_heads.
1493 index = mpd->first_page;
1494 end = mpd->next_page - 1;
1496 pagevec_init(&pvec, 0);
1497 while (index <= end) {
1498 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1501 for (i = 0; i < nr_pages; i++) {
1503 struct page *page = pvec.pages[i];
1505 index = page->index;
1509 if (index == size >> PAGE_CACHE_SHIFT)
1510 len = size & ~PAGE_CACHE_MASK;
1512 len = PAGE_CACHE_SIZE;
1514 cur_logical = index << (PAGE_CACHE_SHIFT -
1516 pblock = map->m_pblk + (cur_logical -
1521 BUG_ON(!PageLocked(page));
1522 BUG_ON(PageWriteback(page));
1524 bh = page_bufs = page_buffers(page);
1527 if (map && (cur_logical >= map->m_lblk) &&
1528 (cur_logical <= (map->m_lblk +
1529 (map->m_len - 1)))) {
1530 if (buffer_delay(bh)) {
1531 clear_buffer_delay(bh);
1532 bh->b_blocknr = pblock;
1534 if (buffer_unwritten(bh) ||
1536 BUG_ON(bh->b_blocknr != pblock);
1537 if (map->m_flags & EXT4_MAP_UNINIT)
1538 set_buffer_uninit(bh);
1539 clear_buffer_unwritten(bh);
1543 * skip page if block allocation undone and
1546 if (ext4_bh_delay_or_unwritten(NULL, bh))
1548 bh = bh->b_this_page;
1549 block_start += bh->b_size;
1552 } while (bh != page_bufs);
1559 clear_page_dirty_for_io(page);
1560 err = ext4_bio_write_page(&io_submit, page, len,
1563 mpd->pages_written++;
1565 * In error case, we have to continue because
1566 * remaining pages are still locked
1571 pagevec_release(&pvec);
1573 ext4_io_submit(&io_submit);
1577 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1581 struct pagevec pvec;
1582 struct inode *inode = mpd->inode;
1583 struct address_space *mapping = inode->i_mapping;
1584 ext4_lblk_t start, last;
1586 index = mpd->first_page;
1587 end = mpd->next_page - 1;
1589 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1590 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1591 ext4_es_remove_extent(inode, start, last - start + 1);
1593 pagevec_init(&pvec, 0);
1594 while (index <= end) {
1595 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1598 for (i = 0; i < nr_pages; i++) {
1599 struct page *page = pvec.pages[i];
1600 if (page->index > end)
1602 BUG_ON(!PageLocked(page));
1603 BUG_ON(PageWriteback(page));
1604 block_invalidatepage(page, 0);
1605 ClearPageUptodate(page);
1608 index = pvec.pages[nr_pages - 1]->index + 1;
1609 pagevec_release(&pvec);
1614 static void ext4_print_free_blocks(struct inode *inode)
1616 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1617 struct super_block *sb = inode->i_sb;
1618 struct ext4_inode_info *ei = EXT4_I(inode);
1620 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1621 EXT4_C2B(EXT4_SB(inode->i_sb),
1622 ext4_count_free_clusters(sb)));
1623 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1624 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1625 (long long) EXT4_C2B(EXT4_SB(sb),
1626 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1627 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1628 (long long) EXT4_C2B(EXT4_SB(sb),
1629 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1630 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1631 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1632 ei->i_reserved_data_blocks);
1633 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1634 ei->i_reserved_meta_blocks);
1635 ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u",
1636 ei->i_allocated_meta_blocks);
1641 * mpage_da_map_and_submit - go through given space, map them
1642 * if necessary, and then submit them for I/O
1644 * @mpd - bh describing space
1646 * The function skips space we know is already mapped to disk blocks.
1649 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1651 int err, blks, get_blocks_flags;
1652 struct ext4_map_blocks map, *mapp = NULL;
1653 sector_t next = mpd->b_blocknr;
1654 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1655 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1656 handle_t *handle = NULL;
1659 * If the blocks are mapped already, or we couldn't accumulate
1660 * any blocks, then proceed immediately to the submission stage.
1662 if ((mpd->b_size == 0) ||
1663 ((mpd->b_state & (1 << BH_Mapped)) &&
1664 !(mpd->b_state & (1 << BH_Delay)) &&
1665 !(mpd->b_state & (1 << BH_Unwritten))))
1668 handle = ext4_journal_current_handle();
1672 * Call ext4_map_blocks() to allocate any delayed allocation
1673 * blocks, or to convert an uninitialized extent to be
1674 * initialized (in the case where we have written into
1675 * one or more preallocated blocks).
1677 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1678 * indicate that we are on the delayed allocation path. This
1679 * affects functions in many different parts of the allocation
1680 * call path. This flag exists primarily because we don't
1681 * want to change *many* call functions, so ext4_map_blocks()
1682 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1683 * inode's allocation semaphore is taken.
1685 * If the blocks in questions were delalloc blocks, set
1686 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1687 * variables are updated after the blocks have been allocated.
1690 map.m_len = max_blocks;
1692 * We're in delalloc path and it is possible that we're going to
1693 * need more metadata blocks than previously reserved. However
1694 * we must not fail because we're in writeback and there is
1695 * nothing we can do about it so it might result in data loss.
1696 * So use reserved blocks to allocate metadata if possible.
1698 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
1699 EXT4_GET_BLOCKS_METADATA_NOFAIL;
1700 if (ext4_should_dioread_nolock(mpd->inode))
1701 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1702 if (mpd->b_state & (1 << BH_Delay))
1703 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1706 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1708 struct super_block *sb = mpd->inode->i_sb;
1712 * If get block returns EAGAIN or ENOSPC and there
1713 * appears to be free blocks we will just let
1714 * mpage_da_submit_io() unlock all of the pages.
1719 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1725 * get block failure will cause us to loop in
1726 * writepages, because a_ops->writepage won't be able
1727 * to make progress. The page will be redirtied by
1728 * writepage and writepages will again try to write
1731 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1732 ext4_msg(sb, KERN_CRIT,
1733 "delayed block allocation failed for inode %lu "
1734 "at logical offset %llu with max blocks %zd "
1735 "with error %d", mpd->inode->i_ino,
1736 (unsigned long long) next,
1737 mpd->b_size >> mpd->inode->i_blkbits, err);
1738 ext4_msg(sb, KERN_CRIT,
1739 "This should not happen!! Data will be lost");
1741 ext4_print_free_blocks(mpd->inode);
1743 /* invalidate all the pages */
1744 ext4_da_block_invalidatepages(mpd);
1746 /* Mark this page range as having been completed */
1753 if (map.m_flags & EXT4_MAP_NEW) {
1754 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1757 for (i = 0; i < map.m_len; i++)
1758 unmap_underlying_metadata(bdev, map.m_pblk + i);
1762 * Update on-disk size along with block allocation.
1764 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1765 if (disksize > i_size_read(mpd->inode))
1766 disksize = i_size_read(mpd->inode);
1767 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1768 ext4_update_i_disksize(mpd->inode, disksize);
1769 err = ext4_mark_inode_dirty(handle, mpd->inode);
1771 ext4_error(mpd->inode->i_sb,
1772 "Failed to mark inode %lu dirty",
1777 mpage_da_submit_io(mpd, mapp);
1781 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1782 (1 << BH_Delay) | (1 << BH_Unwritten))
1785 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1787 * @mpd->lbh - extent of blocks
1788 * @logical - logical number of the block in the file
1789 * @b_state - b_state of the buffer head added
1791 * the function is used to collect contig. blocks in same state
1793 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical,
1794 unsigned long b_state)
1797 int blkbits = mpd->inode->i_blkbits;
1798 int nrblocks = mpd->b_size >> blkbits;
1801 * XXX Don't go larger than mballoc is willing to allocate
1802 * This is a stopgap solution. We eventually need to fold
1803 * mpage_da_submit_io() into this function and then call
1804 * ext4_map_blocks() multiple times in a loop
1806 if (nrblocks >= (8*1024*1024 >> blkbits))
1809 /* check if the reserved journal credits might overflow */
1810 if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) {
1811 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1813 * With non-extent format we are limited by the journal
1814 * credit available. Total credit needed to insert
1815 * nrblocks contiguous blocks is dependent on the
1816 * nrblocks. So limit nrblocks.
1822 * First block in the extent
1824 if (mpd->b_size == 0) {
1825 mpd->b_blocknr = logical;
1826 mpd->b_size = 1 << blkbits;
1827 mpd->b_state = b_state & BH_FLAGS;
1831 next = mpd->b_blocknr + nrblocks;
1833 * Can we merge the block to our big extent?
1835 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1836 mpd->b_size += 1 << blkbits;
1842 * We couldn't merge the block to our extent, so we
1843 * need to flush current extent and start new one
1845 mpage_da_map_and_submit(mpd);
1849 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1851 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1855 * This function is grabs code from the very beginning of
1856 * ext4_map_blocks, but assumes that the caller is from delayed write
1857 * time. This function looks up the requested blocks and sets the
1858 * buffer delay bit under the protection of i_data_sem.
1860 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1861 struct ext4_map_blocks *map,
1862 struct buffer_head *bh)
1864 struct extent_status es;
1866 sector_t invalid_block = ~((sector_t) 0xffff);
1867 #ifdef ES_AGGRESSIVE_TEST
1868 struct ext4_map_blocks orig_map;
1870 memcpy(&orig_map, map, sizeof(*map));
1873 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1877 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1878 "logical block %lu\n", inode->i_ino, map->m_len,
1879 (unsigned long) map->m_lblk);
1881 /* Lookup extent status tree firstly */
1882 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1884 if (ext4_es_is_hole(&es)) {
1886 down_read((&EXT4_I(inode)->i_data_sem));
1891 * Delayed extent could be allocated by fallocate.
1892 * So we need to check it.
1894 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1895 map_bh(bh, inode->i_sb, invalid_block);
1897 set_buffer_delay(bh);
1901 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1902 retval = es.es_len - (iblock - es.es_lblk);
1903 if (retval > map->m_len)
1904 retval = map->m_len;
1905 map->m_len = retval;
1906 if (ext4_es_is_written(&es))
1907 map->m_flags |= EXT4_MAP_MAPPED;
1908 else if (ext4_es_is_unwritten(&es))
1909 map->m_flags |= EXT4_MAP_UNWRITTEN;
1913 #ifdef ES_AGGRESSIVE_TEST
1914 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1920 * Try to see if we can get the block without requesting a new
1921 * file system block.
1923 down_read((&EXT4_I(inode)->i_data_sem));
1924 if (ext4_has_inline_data(inode)) {
1926 * We will soon create blocks for this page, and let
1927 * us pretend as if the blocks aren't allocated yet.
1928 * In case of clusters, we have to handle the work
1929 * of mapping from cluster so that the reserved space
1930 * is calculated properly.
1932 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1933 ext4_find_delalloc_cluster(inode, map->m_lblk))
1934 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1936 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1937 retval = ext4_ext_map_blocks(NULL, inode, map,
1938 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1940 retval = ext4_ind_map_blocks(NULL, inode, map,
1941 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1947 * XXX: __block_prepare_write() unmaps passed block,
1951 * If the block was allocated from previously allocated cluster,
1952 * then we don't need to reserve it again. However we still need
1953 * to reserve metadata for every block we're going to write.
1955 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1956 ret = ext4_da_reserve_space(inode, iblock);
1958 /* not enough space to reserve */
1963 ret = ext4_da_reserve_metadata(inode, iblock);
1965 /* not enough space to reserve */
1971 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1972 ~0, EXTENT_STATUS_DELAYED);
1978 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1979 * and it should not appear on the bh->b_state.
1981 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1983 map_bh(bh, inode->i_sb, invalid_block);
1985 set_buffer_delay(bh);
1986 } else if (retval > 0) {
1988 unsigned long long status;
1990 #ifdef ES_AGGRESSIVE_TEST
1991 if (retval != map->m_len) {
1992 printk("ES len assertation failed for inode: %lu "
1993 "retval %d != map->m_len %d "
1994 "in %s (lookup)\n", inode->i_ino, retval,
1995 map->m_len, __func__);
1999 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
2000 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
2001 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
2002 map->m_pblk, status);
2008 up_read((&EXT4_I(inode)->i_data_sem));
2014 * This is a special get_blocks_t callback which is used by
2015 * ext4_da_write_begin(). It will either return mapped block or
2016 * reserve space for a single block.
2018 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2019 * We also have b_blocknr = -1 and b_bdev initialized properly
2021 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2022 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2023 * initialized properly.
2025 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2026 struct buffer_head *bh, int create)
2028 struct ext4_map_blocks map;
2031 BUG_ON(create == 0);
2032 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2034 map.m_lblk = iblock;
2038 * first, we need to know whether the block is allocated already
2039 * preallocated blocks are unmapped but should treated
2040 * the same as allocated blocks.
2042 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
2046 map_bh(bh, inode->i_sb, map.m_pblk);
2047 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2049 if (buffer_unwritten(bh)) {
2050 /* A delayed write to unwritten bh should be marked
2051 * new and mapped. Mapped ensures that we don't do
2052 * get_block multiple times when we write to the same
2053 * offset and new ensures that we do proper zero out
2054 * for partial write.
2057 set_buffer_mapped(bh);
2062 static int bget_one(handle_t *handle, struct buffer_head *bh)
2068 static int bput_one(handle_t *handle, struct buffer_head *bh)
2074 static int __ext4_journalled_writepage(struct page *page,
2077 struct address_space *mapping = page->mapping;
2078 struct inode *inode = mapping->host;
2079 struct buffer_head *page_bufs = NULL;
2080 handle_t *handle = NULL;
2081 int ret = 0, err = 0;
2082 int inline_data = ext4_has_inline_data(inode);
2083 struct buffer_head *inode_bh = NULL;
2085 ClearPageChecked(page);
2088 BUG_ON(page->index != 0);
2089 BUG_ON(len > ext4_get_max_inline_size(inode));
2090 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2091 if (inode_bh == NULL)
2094 page_bufs = page_buffers(page);
2099 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2102 /* As soon as we unlock the page, it can go away, but we have
2103 * references to buffers so we are safe */
2106 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2107 ext4_writepage_trans_blocks(inode));
2108 if (IS_ERR(handle)) {
2109 ret = PTR_ERR(handle);
2113 BUG_ON(!ext4_handle_valid(handle));
2116 ret = ext4_journal_get_write_access(handle, inode_bh);
2118 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2121 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2122 do_journal_get_write_access);
2124 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2129 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2130 err = ext4_journal_stop(handle);
2134 if (!ext4_has_inline_data(inode))
2135 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2137 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2144 * Note that we don't need to start a transaction unless we're journaling data
2145 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2146 * need to file the inode to the transaction's list in ordered mode because if
2147 * we are writing back data added by write(), the inode is already there and if
2148 * we are writing back data modified via mmap(), no one guarantees in which
2149 * transaction the data will hit the disk. In case we are journaling data, we
2150 * cannot start transaction directly because transaction start ranks above page
2151 * lock so we have to do some magic.
2153 * This function can get called via...
2154 * - ext4_da_writepages after taking page lock (have journal handle)
2155 * - journal_submit_inode_data_buffers (no journal handle)
2156 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2157 * - grab_page_cache when doing write_begin (have journal handle)
2159 * We don't do any block allocation in this function. If we have page with
2160 * multiple blocks we need to write those buffer_heads that are mapped. This
2161 * is important for mmaped based write. So if we do with blocksize 1K
2162 * truncate(f, 1024);
2163 * a = mmap(f, 0, 4096);
2165 * truncate(f, 4096);
2166 * we have in the page first buffer_head mapped via page_mkwrite call back
2167 * but other buffer_heads would be unmapped but dirty (dirty done via the
2168 * do_wp_page). So writepage should write the first block. If we modify
2169 * the mmap area beyond 1024 we will again get a page_fault and the
2170 * page_mkwrite callback will do the block allocation and mark the
2171 * buffer_heads mapped.
2173 * We redirty the page if we have any buffer_heads that is either delay or
2174 * unwritten in the page.
2176 * We can get recursively called as show below.
2178 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2181 * But since we don't do any block allocation we should not deadlock.
2182 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2184 static int ext4_writepage(struct page *page,
2185 struct writeback_control *wbc)
2190 struct buffer_head *page_bufs = NULL;
2191 struct inode *inode = page->mapping->host;
2192 struct ext4_io_submit io_submit;
2194 trace_ext4_writepage(page);
2195 size = i_size_read(inode);
2196 if (page->index == size >> PAGE_CACHE_SHIFT)
2197 len = size & ~PAGE_CACHE_MASK;
2199 len = PAGE_CACHE_SIZE;
2201 page_bufs = page_buffers(page);
2203 * We cannot do block allocation or other extent handling in this
2204 * function. If there are buffers needing that, we have to redirty
2205 * the page. But we may reach here when we do a journal commit via
2206 * journal_submit_inode_data_buffers() and in that case we must write
2207 * allocated buffers to achieve data=ordered mode guarantees.
2209 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2210 ext4_bh_delay_or_unwritten)) {
2211 redirty_page_for_writepage(wbc, page);
2212 if (current->flags & PF_MEMALLOC) {
2214 * For memory cleaning there's no point in writing only
2215 * some buffers. So just bail out. Warn if we came here
2216 * from direct reclaim.
2218 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2225 if (PageChecked(page) && ext4_should_journal_data(inode))
2227 * It's mmapped pagecache. Add buffers and journal it. There
2228 * doesn't seem much point in redirtying the page here.
2230 return __ext4_journalled_writepage(page, len);
2232 memset(&io_submit, 0, sizeof(io_submit));
2233 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2234 ext4_io_submit(&io_submit);
2239 * This is called via ext4_da_writepages() to
2240 * calculate the total number of credits to reserve to fit
2241 * a single extent allocation into a single transaction,
2242 * ext4_da_writpeages() will loop calling this before
2243 * the block allocation.
2246 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2248 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2251 * With non-extent format the journal credit needed to
2252 * insert nrblocks contiguous block is dependent on
2253 * number of contiguous block. So we will limit
2254 * number of contiguous block to a sane value
2256 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2257 (max_blocks > EXT4_MAX_TRANS_DATA))
2258 max_blocks = EXT4_MAX_TRANS_DATA;
2260 return ext4_chunk_trans_blocks(inode, max_blocks);
2264 * write_cache_pages_da - walk the list of dirty pages of the given
2265 * address space and accumulate pages that need writing, and call
2266 * mpage_da_map_and_submit to map a single contiguous memory region
2267 * and then write them.
2269 static int write_cache_pages_da(handle_t *handle,
2270 struct address_space *mapping,
2271 struct writeback_control *wbc,
2272 struct mpage_da_data *mpd,
2273 pgoff_t *done_index)
2275 struct buffer_head *bh, *head;
2276 struct inode *inode = mapping->host;
2277 struct pagevec pvec;
2278 unsigned int nr_pages;
2281 long nr_to_write = wbc->nr_to_write;
2282 int i, tag, ret = 0;
2284 memset(mpd, 0, sizeof(struct mpage_da_data));
2287 pagevec_init(&pvec, 0);
2288 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2289 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2291 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2292 tag = PAGECACHE_TAG_TOWRITE;
2294 tag = PAGECACHE_TAG_DIRTY;
2296 *done_index = index;
2297 while (index <= end) {
2298 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2299 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2303 for (i = 0; i < nr_pages; i++) {
2304 struct page *page = pvec.pages[i];
2307 * At this point, the page may be truncated or
2308 * invalidated (changing page->mapping to NULL), or
2309 * even swizzled back from swapper_space to tmpfs file
2310 * mapping. However, page->index will not change
2311 * because we have a reference on the page.
2313 if (page->index > end)
2316 *done_index = page->index + 1;
2319 * If we can't merge this page, and we have
2320 * accumulated an contiguous region, write it
2322 if ((mpd->next_page != page->index) &&
2323 (mpd->next_page != mpd->first_page)) {
2324 mpage_da_map_and_submit(mpd);
2325 goto ret_extent_tail;
2331 * If the page is no longer dirty, or its
2332 * mapping no longer corresponds to inode we
2333 * are writing (which means it has been
2334 * truncated or invalidated), or the page is
2335 * already under writeback and we are not
2336 * doing a data integrity writeback, skip the page
2338 if (!PageDirty(page) ||
2339 (PageWriteback(page) &&
2340 (wbc->sync_mode == WB_SYNC_NONE)) ||
2341 unlikely(page->mapping != mapping)) {
2346 wait_on_page_writeback(page);
2347 BUG_ON(PageWriteback(page));
2350 * If we have inline data and arrive here, it means that
2351 * we will soon create the block for the 1st page, so
2352 * we'd better clear the inline data here.
2354 if (ext4_has_inline_data(inode)) {
2355 BUG_ON(ext4_test_inode_state(inode,
2356 EXT4_STATE_MAY_INLINE_DATA));
2357 ext4_destroy_inline_data(handle, inode);
2360 if (mpd->next_page != page->index)
2361 mpd->first_page = page->index;
2362 mpd->next_page = page->index + 1;
2363 logical = (sector_t) page->index <<
2364 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2366 /* Add all dirty buffers to mpd */
2367 head = page_buffers(page);
2370 BUG_ON(buffer_locked(bh));
2372 * We need to try to allocate unmapped blocks
2373 * in the same page. Otherwise we won't make
2374 * progress with the page in ext4_writepage
2376 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2377 mpage_add_bh_to_extent(mpd, logical,
2380 goto ret_extent_tail;
2381 } else if (buffer_dirty(bh) &&
2382 buffer_mapped(bh)) {
2384 * mapped dirty buffer. We need to
2385 * update the b_state because we look
2386 * at b_state in mpage_da_map_blocks.
2387 * We don't update b_size because if we
2388 * find an unmapped buffer_head later
2389 * we need to use the b_state flag of
2392 if (mpd->b_size == 0)
2394 bh->b_state & BH_FLAGS;
2397 } while ((bh = bh->b_this_page) != head);
2399 if (nr_to_write > 0) {
2401 if (nr_to_write == 0 &&
2402 wbc->sync_mode == WB_SYNC_NONE)
2404 * We stop writing back only if we are
2405 * not doing integrity sync. In case of
2406 * integrity sync we have to keep going
2407 * because someone may be concurrently
2408 * dirtying pages, and we might have
2409 * synced a lot of newly appeared dirty
2410 * pages, but have not synced all of the
2416 pagevec_release(&pvec);
2421 ret = MPAGE_DA_EXTENT_TAIL;
2423 pagevec_release(&pvec);
2429 static int ext4_da_writepages(struct address_space *mapping,
2430 struct writeback_control *wbc)
2433 int range_whole = 0;
2434 handle_t *handle = NULL;
2435 struct mpage_da_data mpd;
2436 struct inode *inode = mapping->host;
2437 int pages_written = 0;
2438 unsigned int max_pages;
2439 int range_cyclic, cycled = 1, io_done = 0;
2440 int needed_blocks, ret = 0;
2441 long desired_nr_to_write, nr_to_writebump = 0;
2442 loff_t range_start = wbc->range_start;
2443 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2444 pgoff_t done_index = 0;
2446 struct blk_plug plug;
2448 trace_ext4_da_writepages(inode, wbc);
2451 * No pages to write? This is mainly a kludge to avoid starting
2452 * a transaction for special inodes like journal inode on last iput()
2453 * because that could violate lock ordering on umount
2455 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2459 * If the filesystem has aborted, it is read-only, so return
2460 * right away instead of dumping stack traces later on that
2461 * will obscure the real source of the problem. We test
2462 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2463 * the latter could be true if the filesystem is mounted
2464 * read-only, and in that case, ext4_da_writepages should
2465 * *never* be called, so if that ever happens, we would want
2468 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2471 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2474 range_cyclic = wbc->range_cyclic;
2475 if (wbc->range_cyclic) {
2476 index = mapping->writeback_index;
2479 wbc->range_start = index << PAGE_CACHE_SHIFT;
2480 wbc->range_end = LLONG_MAX;
2481 wbc->range_cyclic = 0;
2484 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2485 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2489 * This works around two forms of stupidity. The first is in
2490 * the writeback code, which caps the maximum number of pages
2491 * written to be 1024 pages. This is wrong on multiple
2492 * levels; different architectues have a different page size,
2493 * which changes the maximum amount of data which gets
2494 * written. Secondly, 4 megabytes is way too small. XFS
2495 * forces this value to be 16 megabytes by multiplying
2496 * nr_to_write parameter by four, and then relies on its
2497 * allocator to allocate larger extents to make them
2498 * contiguous. Unfortunately this brings us to the second
2499 * stupidity, which is that ext4's mballoc code only allocates
2500 * at most 2048 blocks. So we force contiguous writes up to
2501 * the number of dirty blocks in the inode, or
2502 * sbi->max_writeback_mb_bump whichever is smaller.
2504 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2505 if (!range_cyclic && range_whole) {
2506 if (wbc->nr_to_write == LONG_MAX)
2507 desired_nr_to_write = wbc->nr_to_write;
2509 desired_nr_to_write = wbc->nr_to_write * 8;
2511 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2513 if (desired_nr_to_write > max_pages)
2514 desired_nr_to_write = max_pages;
2516 if (wbc->nr_to_write < desired_nr_to_write) {
2517 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2518 wbc->nr_to_write = desired_nr_to_write;
2522 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2523 tag_pages_for_writeback(mapping, index, end);
2525 blk_start_plug(&plug);
2526 while (!ret && wbc->nr_to_write > 0) {
2529 * we insert one extent at a time. So we need
2530 * credit needed for single extent allocation.
2531 * journalled mode is currently not supported
2534 BUG_ON(ext4_should_journal_data(inode));
2535 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2537 /* start a new transaction*/
2538 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2540 if (IS_ERR(handle)) {
2541 ret = PTR_ERR(handle);
2542 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2543 "%ld pages, ino %lu; err %d", __func__,
2544 wbc->nr_to_write, inode->i_ino, ret);
2545 blk_finish_plug(&plug);
2546 goto out_writepages;
2550 * Now call write_cache_pages_da() to find the next
2551 * contiguous region of logical blocks that need
2552 * blocks to be allocated by ext4 and submit them.
2554 ret = write_cache_pages_da(handle, mapping,
2555 wbc, &mpd, &done_index);
2557 * If we have a contiguous extent of pages and we
2558 * haven't done the I/O yet, map the blocks and submit
2561 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2562 mpage_da_map_and_submit(&mpd);
2563 ret = MPAGE_DA_EXTENT_TAIL;
2565 trace_ext4_da_write_pages(inode, &mpd);
2566 wbc->nr_to_write -= mpd.pages_written;
2568 ext4_journal_stop(handle);
2570 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2571 /* commit the transaction which would
2572 * free blocks released in the transaction
2575 jbd2_journal_force_commit_nested(sbi->s_journal);
2577 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2579 * Got one extent now try with rest of the pages.
2580 * If mpd.retval is set -EIO, journal is aborted.
2581 * So we don't need to write any more.
2583 pages_written += mpd.pages_written;
2586 } else if (wbc->nr_to_write)
2588 * There is no more writeout needed
2589 * or we requested for a noblocking writeout
2590 * and we found the device congested
2594 blk_finish_plug(&plug);
2595 if (!io_done && !cycled) {
2598 wbc->range_start = index << PAGE_CACHE_SHIFT;
2599 wbc->range_end = mapping->writeback_index - 1;
2604 wbc->range_cyclic = range_cyclic;
2605 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2607 * set the writeback_index so that range_cyclic
2608 * mode will write it back later
2610 mapping->writeback_index = done_index;
2613 wbc->nr_to_write -= nr_to_writebump;
2614 wbc->range_start = range_start;
2615 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2619 static int ext4_nonda_switch(struct super_block *sb)
2621 s64 free_clusters, dirty_clusters;
2622 struct ext4_sb_info *sbi = EXT4_SB(sb);
2625 * switch to non delalloc mode if we are running low
2626 * on free block. The free block accounting via percpu
2627 * counters can get slightly wrong with percpu_counter_batch getting
2628 * accumulated on each CPU without updating global counters
2629 * Delalloc need an accurate free block accounting. So switch
2630 * to non delalloc when we are near to error range.
2633 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2635 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2637 * Start pushing delalloc when 1/2 of free blocks are dirty.
2639 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2640 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2642 if (2 * free_clusters < 3 * dirty_clusters ||
2643 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2645 * free block count is less than 150% of dirty blocks
2646 * or free blocks is less than watermark
2653 /* We always reserve for an inode update; the superblock could be there too */
2654 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2656 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
2657 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2660 if (pos + len <= 0x7fffffffULL)
2663 /* We might need to update the superblock to set LARGE_FILE */
2667 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2668 loff_t pos, unsigned len, unsigned flags,
2669 struct page **pagep, void **fsdata)
2671 int ret, retries = 0;
2674 struct inode *inode = mapping->host;
2677 index = pos >> PAGE_CACHE_SHIFT;
2679 if (ext4_nonda_switch(inode->i_sb)) {
2680 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2681 return ext4_write_begin(file, mapping, pos,
2682 len, flags, pagep, fsdata);
2684 *fsdata = (void *)0;
2685 trace_ext4_da_write_begin(inode, pos, len, flags);
2687 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2688 ret = ext4_da_write_inline_data_begin(mapping, inode,
2698 * grab_cache_page_write_begin() can take a long time if the
2699 * system is thrashing due to memory pressure, or if the page
2700 * is being written back. So grab it first before we start
2701 * the transaction handle. This also allows us to allocate
2702 * the page (if needed) without using GFP_NOFS.
2705 page = grab_cache_page_write_begin(mapping, index, flags);
2711 * With delayed allocation, we don't log the i_disksize update
2712 * if there is delayed block allocation. But we still need
2713 * to journalling the i_disksize update if writes to the end
2714 * of file which has an already mapped buffer.
2717 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2718 ext4_da_write_credits(inode, pos, len));
2719 if (IS_ERR(handle)) {
2720 page_cache_release(page);
2721 return PTR_ERR(handle);
2725 if (page->mapping != mapping) {
2726 /* The page got truncated from under us */
2728 page_cache_release(page);
2729 ext4_journal_stop(handle);
2732 /* In case writeback began while the page was unlocked */
2733 wait_for_stable_page(page);
2735 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2738 ext4_journal_stop(handle);
2740 * block_write_begin may have instantiated a few blocks
2741 * outside i_size. Trim these off again. Don't need
2742 * i_size_read because we hold i_mutex.
2744 if (pos + len > inode->i_size)
2745 ext4_truncate_failed_write(inode);
2747 if (ret == -ENOSPC &&
2748 ext4_should_retry_alloc(inode->i_sb, &retries))
2751 page_cache_release(page);
2760 * Check if we should update i_disksize
2761 * when write to the end of file but not require block allocation
2763 static int ext4_da_should_update_i_disksize(struct page *page,
2764 unsigned long offset)
2766 struct buffer_head *bh;
2767 struct inode *inode = page->mapping->host;
2771 bh = page_buffers(page);
2772 idx = offset >> inode->i_blkbits;
2774 for (i = 0; i < idx; i++)
2775 bh = bh->b_this_page;
2777 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2782 static int ext4_da_write_end(struct file *file,
2783 struct address_space *mapping,
2784 loff_t pos, unsigned len, unsigned copied,
2785 struct page *page, void *fsdata)
2787 struct inode *inode = mapping->host;
2789 handle_t *handle = ext4_journal_current_handle();
2791 unsigned long start, end;
2792 int write_mode = (int)(unsigned long)fsdata;
2794 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2795 return ext4_write_end(file, mapping, pos,
2796 len, copied, page, fsdata);
2798 trace_ext4_da_write_end(inode, pos, len, copied);
2799 start = pos & (PAGE_CACHE_SIZE - 1);
2800 end = start + copied - 1;
2803 * generic_write_end() will run mark_inode_dirty() if i_size
2804 * changes. So let's piggyback the i_disksize mark_inode_dirty
2807 new_i_size = pos + copied;
2808 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2809 if (ext4_has_inline_data(inode) ||
2810 ext4_da_should_update_i_disksize(page, end)) {
2811 down_write(&EXT4_I(inode)->i_data_sem);
2812 if (new_i_size > EXT4_I(inode)->i_disksize)
2813 EXT4_I(inode)->i_disksize = new_i_size;
2814 up_write(&EXT4_I(inode)->i_data_sem);
2815 /* We need to mark inode dirty even if
2816 * new_i_size is less that inode->i_size
2817 * bu greater than i_disksize.(hint delalloc)
2819 ext4_mark_inode_dirty(handle, inode);
2823 if (write_mode != CONVERT_INLINE_DATA &&
2824 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2825 ext4_has_inline_data(inode))
2826 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2829 ret2 = generic_write_end(file, mapping, pos, len, copied,
2835 ret2 = ext4_journal_stop(handle);
2839 return ret ? ret : copied;
2842 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2845 * Drop reserved blocks
2847 BUG_ON(!PageLocked(page));
2848 if (!page_has_buffers(page))
2851 ext4_da_page_release_reservation(page, offset);
2854 ext4_invalidatepage(page, offset);
2860 * Force all delayed allocation blocks to be allocated for a given inode.
2862 int ext4_alloc_da_blocks(struct inode *inode)
2864 trace_ext4_alloc_da_blocks(inode);
2866 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2867 !EXT4_I(inode)->i_reserved_meta_blocks)
2871 * We do something simple for now. The filemap_flush() will
2872 * also start triggering a write of the data blocks, which is
2873 * not strictly speaking necessary (and for users of
2874 * laptop_mode, not even desirable). However, to do otherwise
2875 * would require replicating code paths in:
2877 * ext4_da_writepages() ->
2878 * write_cache_pages() ---> (via passed in callback function)
2879 * __mpage_da_writepage() -->
2880 * mpage_add_bh_to_extent()
2881 * mpage_da_map_blocks()
2883 * The problem is that write_cache_pages(), located in
2884 * mm/page-writeback.c, marks pages clean in preparation for
2885 * doing I/O, which is not desirable if we're not planning on
2888 * We could call write_cache_pages(), and then redirty all of
2889 * the pages by calling redirty_page_for_writepage() but that
2890 * would be ugly in the extreme. So instead we would need to
2891 * replicate parts of the code in the above functions,
2892 * simplifying them because we wouldn't actually intend to
2893 * write out the pages, but rather only collect contiguous
2894 * logical block extents, call the multi-block allocator, and
2895 * then update the buffer heads with the block allocations.
2897 * For now, though, we'll cheat by calling filemap_flush(),
2898 * which will map the blocks, and start the I/O, but not
2899 * actually wait for the I/O to complete.
2901 return filemap_flush(inode->i_mapping);
2905 * bmap() is special. It gets used by applications such as lilo and by
2906 * the swapper to find the on-disk block of a specific piece of data.
2908 * Naturally, this is dangerous if the block concerned is still in the
2909 * journal. If somebody makes a swapfile on an ext4 data-journaling
2910 * filesystem and enables swap, then they may get a nasty shock when the
2911 * data getting swapped to that swapfile suddenly gets overwritten by
2912 * the original zero's written out previously to the journal and
2913 * awaiting writeback in the kernel's buffer cache.
2915 * So, if we see any bmap calls here on a modified, data-journaled file,
2916 * take extra steps to flush any blocks which might be in the cache.
2918 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2920 struct inode *inode = mapping->host;
2925 * We can get here for an inline file via the FIBMAP ioctl
2927 if (ext4_has_inline_data(inode))
2930 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2931 test_opt(inode->i_sb, DELALLOC)) {
2933 * With delalloc we want to sync the file
2934 * so that we can make sure we allocate
2937 filemap_write_and_wait(mapping);
2940 if (EXT4_JOURNAL(inode) &&
2941 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2943 * This is a REALLY heavyweight approach, but the use of
2944 * bmap on dirty files is expected to be extremely rare:
2945 * only if we run lilo or swapon on a freshly made file
2946 * do we expect this to happen.
2948 * (bmap requires CAP_SYS_RAWIO so this does not
2949 * represent an unprivileged user DOS attack --- we'd be
2950 * in trouble if mortal users could trigger this path at
2953 * NB. EXT4_STATE_JDATA is not set on files other than
2954 * regular files. If somebody wants to bmap a directory
2955 * or symlink and gets confused because the buffer
2956 * hasn't yet been flushed to disk, they deserve
2957 * everything they get.
2960 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2961 journal = EXT4_JOURNAL(inode);
2962 jbd2_journal_lock_updates(journal);
2963 err = jbd2_journal_flush(journal);
2964 jbd2_journal_unlock_updates(journal);
2970 return generic_block_bmap(mapping, block, ext4_get_block);
2973 static int ext4_readpage(struct file *file, struct page *page)
2976 struct inode *inode = page->mapping->host;
2978 trace_ext4_readpage(page);
2980 if (ext4_has_inline_data(inode))
2981 ret = ext4_readpage_inline(inode, page);
2984 return mpage_readpage(page, ext4_get_block);
2990 ext4_readpages(struct file *file, struct address_space *mapping,
2991 struct list_head *pages, unsigned nr_pages)
2993 struct inode *inode = mapping->host;
2995 /* If the file has inline data, no need to do readpages. */
2996 if (ext4_has_inline_data(inode))
2999 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3002 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3004 trace_ext4_invalidatepage(page, offset);
3006 /* No journalling happens on data buffers when this function is used */
3007 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3009 block_invalidatepage(page, offset);
3012 static int __ext4_journalled_invalidatepage(struct page *page,
3013 unsigned long offset)
3015 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3017 trace_ext4_journalled_invalidatepage(page, offset);
3020 * If it's a full truncate we just forget about the pending dirtying
3023 ClearPageChecked(page);
3025 return jbd2_journal_invalidatepage(journal, page, offset);
3028 /* Wrapper for aops... */
3029 static void ext4_journalled_invalidatepage(struct page *page,
3030 unsigned long offset)
3032 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
3035 static int ext4_releasepage(struct page *page, gfp_t wait)
3037 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3039 trace_ext4_releasepage(page);
3041 /* Page has dirty journalled data -> cannot release */
3042 if (PageChecked(page))
3045 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3047 return try_to_free_buffers(page);
3051 * ext4_get_block used when preparing for a DIO write or buffer write.
3052 * We allocate an uinitialized extent if blocks haven't been allocated.
3053 * The extent will be converted to initialized after the IO is complete.
3055 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3056 struct buffer_head *bh_result, int create)
3058 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3059 inode->i_ino, create);
3060 return _ext4_get_block(inode, iblock, bh_result,
3061 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3064 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3065 struct buffer_head *bh_result, int create)
3067 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3068 inode->i_ino, create);
3069 return _ext4_get_block(inode, iblock, bh_result,
3070 EXT4_GET_BLOCKS_NO_LOCK);
3073 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3074 ssize_t size, void *private, int ret,
3077 struct inode *inode = file_inode(iocb->ki_filp);
3078 ext4_io_end_t *io_end = iocb->private;
3080 /* if not async direct IO or dio with 0 bytes write, just return */
3081 if (!io_end || !size)
3084 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3085 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3086 iocb->private, io_end->inode->i_ino, iocb, offset,
3089 iocb->private = NULL;
3091 /* if not aio dio with unwritten extents, just free io and return */
3092 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3093 ext4_free_io_end(io_end);
3095 inode_dio_done(inode);
3097 aio_complete(iocb, ret, 0);
3101 io_end->offset = offset;
3102 io_end->size = size;
3104 io_end->iocb = iocb;
3105 io_end->result = ret;
3108 ext4_add_complete_io(io_end);
3112 * For ext4 extent files, ext4 will do direct-io write to holes,
3113 * preallocated extents, and those write extend the file, no need to
3114 * fall back to buffered IO.
3116 * For holes, we fallocate those blocks, mark them as uninitialized
3117 * If those blocks were preallocated, we mark sure they are split, but
3118 * still keep the range to write as uninitialized.
3120 * The unwritten extents will be converted to written when DIO is completed.
3121 * For async direct IO, since the IO may still pending when return, we
3122 * set up an end_io call back function, which will do the conversion
3123 * when async direct IO completed.
3125 * If the O_DIRECT write will extend the file then add this inode to the
3126 * orphan list. So recovery will truncate it back to the original size
3127 * if the machine crashes during the write.
3130 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3131 const struct iovec *iov, loff_t offset,
3132 unsigned long nr_segs)
3134 struct file *file = iocb->ki_filp;
3135 struct inode *inode = file->f_mapping->host;
3137 size_t count = iov_length(iov, nr_segs);
3139 get_block_t *get_block_func = NULL;
3141 loff_t final_size = offset + count;
3143 /* Use the old path for reads and writes beyond i_size. */
3144 if (rw != WRITE || final_size > inode->i_size)
3145 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3147 BUG_ON(iocb->private == NULL);
3149 /* If we do a overwrite dio, i_mutex locking can be released */
3150 overwrite = *((int *)iocb->private);
3153 atomic_inc(&inode->i_dio_count);
3154 down_read(&EXT4_I(inode)->i_data_sem);
3155 mutex_unlock(&inode->i_mutex);
3159 * We could direct write to holes and fallocate.
3161 * Allocated blocks to fill the hole are marked as
3162 * uninitialized to prevent parallel buffered read to expose
3163 * the stale data before DIO complete the data IO.
3165 * As to previously fallocated extents, ext4 get_block will
3166 * just simply mark the buffer mapped but still keep the
3167 * extents uninitialized.
3169 * For non AIO case, we will convert those unwritten extents
3170 * to written after return back from blockdev_direct_IO.
3172 * For async DIO, the conversion needs to be deferred when the
3173 * IO is completed. The ext4 end_io callback function will be
3174 * called to take care of the conversion work. Here for async
3175 * case, we allocate an io_end structure to hook to the iocb.
3177 iocb->private = NULL;
3178 ext4_inode_aio_set(inode, NULL);
3179 if (!is_sync_kiocb(iocb)) {
3180 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
3185 io_end->flag |= EXT4_IO_END_DIRECT;
3186 iocb->private = io_end;
3188 * we save the io structure for current async direct
3189 * IO, so that later ext4_map_blocks() could flag the
3190 * io structure whether there is a unwritten extents
3191 * needs to be converted when IO is completed.
3193 ext4_inode_aio_set(inode, io_end);
3197 get_block_func = ext4_get_block_write_nolock;
3199 get_block_func = ext4_get_block_write;
3200 dio_flags = DIO_LOCKING;
3202 ret = __blockdev_direct_IO(rw, iocb, inode,
3203 inode->i_sb->s_bdev, iov,
3211 ext4_inode_aio_set(inode, NULL);
3213 * The io_end structure takes a reference to the inode, that
3214 * structure needs to be destroyed and the reference to the
3215 * inode need to be dropped, when IO is complete, even with 0
3216 * byte write, or failed.
3218 * In the successful AIO DIO case, the io_end structure will
3219 * be destroyed and the reference to the inode will be dropped
3220 * after the end_io call back function is called.
3222 * In the case there is 0 byte write, or error case, since VFS
3223 * direct IO won't invoke the end_io call back function, we
3224 * need to free the end_io structure here.
3226 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3227 ext4_free_io_end(iocb->private);
3228 iocb->private = NULL;
3229 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3230 EXT4_STATE_DIO_UNWRITTEN)) {
3233 * for non AIO case, since the IO is already
3234 * completed, we could do the conversion right here
3236 err = ext4_convert_unwritten_extents(inode,
3240 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3244 /* take i_mutex locking again if we do a ovewrite dio */
3246 inode_dio_done(inode);
3247 up_read(&EXT4_I(inode)->i_data_sem);
3248 mutex_lock(&inode->i_mutex);
3254 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3255 const struct iovec *iov, loff_t offset,
3256 unsigned long nr_segs)
3258 struct file *file = iocb->ki_filp;
3259 struct inode *inode = file->f_mapping->host;
3263 * If we are doing data journalling we don't support O_DIRECT
3265 if (ext4_should_journal_data(inode))
3268 /* Let buffer I/O handle the inline data case. */
3269 if (ext4_has_inline_data(inode))
3272 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3273 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3274 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3276 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3277 trace_ext4_direct_IO_exit(inode, offset,
3278 iov_length(iov, nr_segs), rw, ret);
3283 * Pages can be marked dirty completely asynchronously from ext4's journalling
3284 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3285 * much here because ->set_page_dirty is called under VFS locks. The page is
3286 * not necessarily locked.
3288 * We cannot just dirty the page and leave attached buffers clean, because the
3289 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3290 * or jbddirty because all the journalling code will explode.
3292 * So what we do is to mark the page "pending dirty" and next time writepage
3293 * is called, propagate that into the buffers appropriately.
3295 static int ext4_journalled_set_page_dirty(struct page *page)
3297 SetPageChecked(page);
3298 return __set_page_dirty_nobuffers(page);
3301 static const struct address_space_operations ext4_aops = {
3302 .readpage = ext4_readpage,
3303 .readpages = ext4_readpages,
3304 .writepage = ext4_writepage,
3305 .write_begin = ext4_write_begin,
3306 .write_end = ext4_write_end,
3308 .invalidatepage = ext4_invalidatepage,
3309 .releasepage = ext4_releasepage,
3310 .direct_IO = ext4_direct_IO,
3311 .migratepage = buffer_migrate_page,
3312 .is_partially_uptodate = block_is_partially_uptodate,
3313 .error_remove_page = generic_error_remove_page,
3316 static const struct address_space_operations ext4_journalled_aops = {
3317 .readpage = ext4_readpage,
3318 .readpages = ext4_readpages,
3319 .writepage = ext4_writepage,
3320 .write_begin = ext4_write_begin,
3321 .write_end = ext4_journalled_write_end,
3322 .set_page_dirty = ext4_journalled_set_page_dirty,
3324 .invalidatepage = ext4_journalled_invalidatepage,
3325 .releasepage = ext4_releasepage,
3326 .direct_IO = ext4_direct_IO,
3327 .is_partially_uptodate = block_is_partially_uptodate,
3328 .error_remove_page = generic_error_remove_page,
3331 static const struct address_space_operations ext4_da_aops = {
3332 .readpage = ext4_readpage,
3333 .readpages = ext4_readpages,
3334 .writepage = ext4_writepage,
3335 .writepages = ext4_da_writepages,
3336 .write_begin = ext4_da_write_begin,
3337 .write_end = ext4_da_write_end,
3339 .invalidatepage = ext4_da_invalidatepage,
3340 .releasepage = ext4_releasepage,
3341 .direct_IO = ext4_direct_IO,
3342 .migratepage = buffer_migrate_page,
3343 .is_partially_uptodate = block_is_partially_uptodate,
3344 .error_remove_page = generic_error_remove_page,
3347 void ext4_set_aops(struct inode *inode)
3349 switch (ext4_inode_journal_mode(inode)) {
3350 case EXT4_INODE_ORDERED_DATA_MODE:
3351 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3353 case EXT4_INODE_WRITEBACK_DATA_MODE:
3354 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3356 case EXT4_INODE_JOURNAL_DATA_MODE:
3357 inode->i_mapping->a_ops = &ext4_journalled_aops;
3362 if (test_opt(inode->i_sb, DELALLOC))
3363 inode->i_mapping->a_ops = &ext4_da_aops;
3365 inode->i_mapping->a_ops = &ext4_aops;
3370 * ext4_discard_partial_page_buffers()
3371 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3372 * This function finds and locks the page containing the offset
3373 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3374 * Calling functions that already have the page locked should call
3375 * ext4_discard_partial_page_buffers_no_lock directly.
3377 int ext4_discard_partial_page_buffers(handle_t *handle,
3378 struct address_space *mapping, loff_t from,
3379 loff_t length, int flags)
3381 struct inode *inode = mapping->host;
3385 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3386 mapping_gfp_mask(mapping) & ~__GFP_FS);
3390 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3391 from, length, flags);
3394 page_cache_release(page);
3399 * ext4_discard_partial_page_buffers_no_lock()
3400 * Zeros a page range of length 'length' starting from offset 'from'.
3401 * Buffer heads that correspond to the block aligned regions of the
3402 * zeroed range will be unmapped. Unblock aligned regions
3403 * will have the corresponding buffer head mapped if needed so that
3404 * that region of the page can be updated with the partial zero out.
3406 * This function assumes that the page has already been locked. The
3407 * The range to be discarded must be contained with in the given page.
3408 * If the specified range exceeds the end of the page it will be shortened
3409 * to the end of the page that corresponds to 'from'. This function is
3410 * appropriate for updating a page and it buffer heads to be unmapped and
3411 * zeroed for blocks that have been either released, or are going to be
3414 * handle: The journal handle
3415 * inode: The files inode
3416 * page: A locked page that contains the offset "from"
3417 * from: The starting byte offset (from the beginning of the file)
3418 * to begin discarding
3419 * len: The length of bytes to discard
3420 * flags: Optional flags that may be used:
3422 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3423 * Only zero the regions of the page whose buffer heads
3424 * have already been unmapped. This flag is appropriate
3425 * for updating the contents of a page whose blocks may
3426 * have already been released, and we only want to zero
3427 * out the regions that correspond to those released blocks.
3429 * Returns zero on success or negative on failure.
3431 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3432 struct inode *inode, struct page *page, loff_t from,
3433 loff_t length, int flags)
3435 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3436 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3437 unsigned int blocksize, max, pos;
3439 struct buffer_head *bh;
3442 blocksize = inode->i_sb->s_blocksize;
3443 max = PAGE_CACHE_SIZE - offset;
3445 if (index != page->index)
3449 * correct length if it does not fall between
3450 * 'from' and the end of the page
3452 if (length > max || length < 0)
3455 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3457 if (!page_has_buffers(page))
3458 create_empty_buffers(page, blocksize, 0);
3460 /* Find the buffer that contains "offset" */
3461 bh = page_buffers(page);
3463 while (offset >= pos) {
3464 bh = bh->b_this_page;
3470 while (pos < offset + length) {
3471 unsigned int end_of_block, range_to_discard;
3475 /* The length of space left to zero and unmap */
3476 range_to_discard = offset + length - pos;
3478 /* The length of space until the end of the block */
3479 end_of_block = blocksize - (pos & (blocksize-1));
3482 * Do not unmap or zero past end of block
3483 * for this buffer head
3485 if (range_to_discard > end_of_block)
3486 range_to_discard = end_of_block;
3490 * Skip this buffer head if we are only zeroing unampped
3491 * regions of the page
3493 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3497 /* If the range is block aligned, unmap */
3498 if (range_to_discard == blocksize) {
3499 clear_buffer_dirty(bh);
3501 clear_buffer_mapped(bh);
3502 clear_buffer_req(bh);
3503 clear_buffer_new(bh);
3504 clear_buffer_delay(bh);
3505 clear_buffer_unwritten(bh);
3506 clear_buffer_uptodate(bh);
3507 zero_user(page, pos, range_to_discard);
3508 BUFFER_TRACE(bh, "Buffer discarded");
3513 * If this block is not completely contained in the range
3514 * to be discarded, then it is not going to be released. Because
3515 * we need to keep this block, we need to make sure this part
3516 * of the page is uptodate before we modify it by writeing
3517 * partial zeros on it.
3519 if (!buffer_mapped(bh)) {
3521 * Buffer head must be mapped before we can read
3524 BUFFER_TRACE(bh, "unmapped");
3525 ext4_get_block(inode, iblock, bh, 0);
3526 /* unmapped? It's a hole - nothing to do */
3527 if (!buffer_mapped(bh)) {
3528 BUFFER_TRACE(bh, "still unmapped");
3533 /* Ok, it's mapped. Make sure it's up-to-date */
3534 if (PageUptodate(page))
3535 set_buffer_uptodate(bh);
3537 if (!buffer_uptodate(bh)) {
3539 ll_rw_block(READ, 1, &bh);
3541 /* Uhhuh. Read error. Complain and punt.*/
3542 if (!buffer_uptodate(bh))
3546 if (ext4_should_journal_data(inode)) {
3547 BUFFER_TRACE(bh, "get write access");
3548 err = ext4_journal_get_write_access(handle, bh);
3553 zero_user(page, pos, range_to_discard);
3556 if (ext4_should_journal_data(inode)) {
3557 err = ext4_handle_dirty_metadata(handle, inode, bh);
3559 mark_buffer_dirty(bh);
3561 BUFFER_TRACE(bh, "Partial buffer zeroed");
3563 bh = bh->b_this_page;
3565 pos += range_to_discard;
3571 int ext4_can_truncate(struct inode *inode)
3573 if (S_ISREG(inode->i_mode))
3575 if (S_ISDIR(inode->i_mode))
3577 if (S_ISLNK(inode->i_mode))
3578 return !ext4_inode_is_fast_symlink(inode);
3583 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3584 * associated with the given offset and length
3586 * @inode: File inode
3587 * @offset: The offset where the hole will begin
3588 * @len: The length of the hole
3590 * Returns: 0 on success or negative on failure
3593 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3595 struct inode *inode = file_inode(file);
3596 struct super_block *sb = inode->i_sb;
3597 ext4_lblk_t first_block, stop_block;
3598 struct address_space *mapping = inode->i_mapping;
3599 loff_t first_page, last_page, page_len;
3600 loff_t first_page_offset, last_page_offset;
3602 unsigned int credits;
3605 if (!S_ISREG(inode->i_mode))
3608 if (EXT4_SB(sb)->s_cluster_ratio > 1) {
3609 /* TODO: Add support for bigalloc file systems */
3613 trace_ext4_punch_hole(inode, offset, length);
3616 * Write out all dirty pages to avoid race conditions
3617 * Then release them.
3619 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3620 ret = filemap_write_and_wait_range(mapping, offset,
3621 offset + length - 1);
3626 mutex_lock(&inode->i_mutex);
3627 /* It's not possible punch hole on append only file */
3628 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
3632 if (IS_SWAPFILE(inode)) {
3637 /* No need to punch hole beyond i_size */
3638 if (offset >= inode->i_size)
3642 * If the hole extends beyond i_size, set the hole
3643 * to end after the page that contains i_size
3645 if (offset + length > inode->i_size) {
3646 length = inode->i_size +
3647 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3651 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
3652 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
3654 first_page_offset = first_page << PAGE_CACHE_SHIFT;
3655 last_page_offset = last_page << PAGE_CACHE_SHIFT;
3657 /* Now release the pages */
3658 if (last_page_offset > first_page_offset) {
3659 truncate_pagecache_range(inode, first_page_offset,
3660 last_page_offset - 1);
3663 /* Wait all existing dio workers, newcomers will block on i_mutex */
3664 ext4_inode_block_unlocked_dio(inode);
3665 ret = ext4_flush_unwritten_io(inode);
3668 inode_dio_wait(inode);
3670 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3671 credits = ext4_writepage_trans_blocks(inode);
3673 credits = ext4_blocks_for_truncate(inode);
3674 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3675 if (IS_ERR(handle)) {
3676 ret = PTR_ERR(handle);
3677 ext4_std_error(sb, ret);
3682 * Now we need to zero out the non-page-aligned data in the
3683 * pages at the start and tail of the hole, and unmap the
3684 * buffer heads for the block aligned regions of the page that
3685 * were completely zeroed.
3687 if (first_page > last_page) {
3689 * If the file space being truncated is contained
3690 * within a page just zero out and unmap the middle of
3693 ret = ext4_discard_partial_page_buffers(handle,
3694 mapping, offset, length, 0);
3700 * zero out and unmap the partial page that contains
3701 * the start of the hole
3703 page_len = first_page_offset - offset;
3705 ret = ext4_discard_partial_page_buffers(handle, mapping,
3706 offset, page_len, 0);
3712 * zero out and unmap the partial page that contains
3713 * the end of the hole
3715 page_len = offset + length - last_page_offset;
3717 ret = ext4_discard_partial_page_buffers(handle, mapping,
3718 last_page_offset, page_len, 0);
3725 * If i_size is contained in the last page, we need to
3726 * unmap and zero the partial page after i_size
3728 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
3729 inode->i_size % PAGE_CACHE_SIZE != 0) {
3730 page_len = PAGE_CACHE_SIZE -
3731 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3734 ret = ext4_discard_partial_page_buffers(handle,
3735 mapping, inode->i_size, page_len, 0);
3742 first_block = (offset + sb->s_blocksize - 1) >>
3743 EXT4_BLOCK_SIZE_BITS(sb);
3744 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3746 /* If there are no blocks to remove, return now */
3747 if (first_block >= stop_block)
3750 down_write(&EXT4_I(inode)->i_data_sem);
3751 ext4_discard_preallocations(inode);
3753 ret = ext4_es_remove_extent(inode, first_block,
3754 stop_block - first_block);
3756 up_write(&EXT4_I(inode)->i_data_sem);
3760 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3761 ret = ext4_ext_remove_space(inode, first_block,
3764 ret = ext4_free_hole_blocks(handle, inode, first_block,
3767 ext4_discard_preallocations(inode);
3768 up_write(&EXT4_I(inode)->i_data_sem);
3770 ext4_handle_sync(handle);
3771 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3772 ext4_mark_inode_dirty(handle, inode);
3774 ext4_journal_stop(handle);
3776 ext4_inode_resume_unlocked_dio(inode);
3778 mutex_unlock(&inode->i_mutex);
3785 * We block out ext4_get_block() block instantiations across the entire
3786 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3787 * simultaneously on behalf of the same inode.
3789 * As we work through the truncate and commit bits of it to the journal there
3790 * is one core, guiding principle: the file's tree must always be consistent on
3791 * disk. We must be able to restart the truncate after a crash.
3793 * The file's tree may be transiently inconsistent in memory (although it
3794 * probably isn't), but whenever we close off and commit a journal transaction,
3795 * the contents of (the filesystem + the journal) must be consistent and
3796 * restartable. It's pretty simple, really: bottom up, right to left (although
3797 * left-to-right works OK too).
3799 * Note that at recovery time, journal replay occurs *before* the restart of
3800 * truncate against the orphan inode list.
3802 * The committed inode has the new, desired i_size (which is the same as
3803 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3804 * that this inode's truncate did not complete and it will again call
3805 * ext4_truncate() to have another go. So there will be instantiated blocks
3806 * to the right of the truncation point in a crashed ext4 filesystem. But
3807 * that's fine - as long as they are linked from the inode, the post-crash
3808 * ext4_truncate() run will find them and release them.
3810 void ext4_truncate(struct inode *inode)
3812 struct ext4_inode_info *ei = EXT4_I(inode);
3813 unsigned int credits;
3815 struct address_space *mapping = inode->i_mapping;
3819 * There is a possibility that we're either freeing the inode
3820 * or it completely new indode. In those cases we might not
3821 * have i_mutex locked because it's not necessary.
3823 if (!(inode->i_state & (I_NEW|I_FREEING)))
3824 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3825 trace_ext4_truncate_enter(inode);
3827 if (!ext4_can_truncate(inode))
3830 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3832 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3833 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3835 if (ext4_has_inline_data(inode)) {
3838 ext4_inline_data_truncate(inode, &has_inline);
3844 * finish any pending end_io work so we won't run the risk of
3845 * converting any truncated blocks to initialized later
3847 ext4_flush_unwritten_io(inode);
3849 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3850 credits = ext4_writepage_trans_blocks(inode);
3852 credits = ext4_blocks_for_truncate(inode);
3854 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3855 if (IS_ERR(handle)) {
3856 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3860 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
3861 page_len = PAGE_CACHE_SIZE -
3862 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3864 if (ext4_discard_partial_page_buffers(handle,
3865 mapping, inode->i_size, page_len, 0))
3870 * We add the inode to the orphan list, so that if this
3871 * truncate spans multiple transactions, and we crash, we will
3872 * resume the truncate when the filesystem recovers. It also
3873 * marks the inode dirty, to catch the new size.
3875 * Implication: the file must always be in a sane, consistent
3876 * truncatable state while each transaction commits.
3878 if (ext4_orphan_add(handle, inode))
3881 down_write(&EXT4_I(inode)->i_data_sem);
3883 ext4_discard_preallocations(inode);
3885 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3886 ext4_ext_truncate(handle, inode);
3888 ext4_ind_truncate(handle, inode);
3890 up_write(&ei->i_data_sem);
3893 ext4_handle_sync(handle);
3897 * If this was a simple ftruncate() and the file will remain alive,
3898 * then we need to clear up the orphan record which we created above.
3899 * However, if this was a real unlink then we were called by
3900 * ext4_delete_inode(), and we allow that function to clean up the
3901 * orphan info for us.
3904 ext4_orphan_del(handle, inode);
3906 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3907 ext4_mark_inode_dirty(handle, inode);
3908 ext4_journal_stop(handle);
3910 trace_ext4_truncate_exit(inode);
3914 * ext4_get_inode_loc returns with an extra refcount against the inode's
3915 * underlying buffer_head on success. If 'in_mem' is true, we have all
3916 * data in memory that is needed to recreate the on-disk version of this
3919 static int __ext4_get_inode_loc(struct inode *inode,
3920 struct ext4_iloc *iloc, int in_mem)
3922 struct ext4_group_desc *gdp;
3923 struct buffer_head *bh;
3924 struct super_block *sb = inode->i_sb;
3926 int inodes_per_block, inode_offset;
3929 if (!ext4_valid_inum(sb, inode->i_ino))
3932 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3933 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3938 * Figure out the offset within the block group inode table
3940 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3941 inode_offset = ((inode->i_ino - 1) %
3942 EXT4_INODES_PER_GROUP(sb));
3943 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3944 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3946 bh = sb_getblk(sb, block);
3949 if (!buffer_uptodate(bh)) {
3953 * If the buffer has the write error flag, we have failed
3954 * to write out another inode in the same block. In this
3955 * case, we don't have to read the block because we may
3956 * read the old inode data successfully.
3958 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3959 set_buffer_uptodate(bh);
3961 if (buffer_uptodate(bh)) {
3962 /* someone brought it uptodate while we waited */
3968 * If we have all information of the inode in memory and this
3969 * is the only valid inode in the block, we need not read the
3973 struct buffer_head *bitmap_bh;
3976 start = inode_offset & ~(inodes_per_block - 1);
3978 /* Is the inode bitmap in cache? */
3979 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3980 if (unlikely(!bitmap_bh))
3984 * If the inode bitmap isn't in cache then the
3985 * optimisation may end up performing two reads instead
3986 * of one, so skip it.
3988 if (!buffer_uptodate(bitmap_bh)) {
3992 for (i = start; i < start + inodes_per_block; i++) {
3993 if (i == inode_offset)
3995 if (ext4_test_bit(i, bitmap_bh->b_data))
3999 if (i == start + inodes_per_block) {
4000 /* all other inodes are free, so skip I/O */
4001 memset(bh->b_data, 0, bh->b_size);
4002 set_buffer_uptodate(bh);
4010 * If we need to do any I/O, try to pre-readahead extra
4011 * blocks from the inode table.
4013 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4014 ext4_fsblk_t b, end, table;
4016 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4018 table = ext4_inode_table(sb, gdp);
4019 /* s_inode_readahead_blks is always a power of 2 */
4020 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4024 num = EXT4_INODES_PER_GROUP(sb);
4025 if (ext4_has_group_desc_csum(sb))
4026 num -= ext4_itable_unused_count(sb, gdp);
4027 table += num / inodes_per_block;
4031 sb_breadahead(sb, b++);
4035 * There are other valid inodes in the buffer, this inode
4036 * has in-inode xattrs, or we don't have this inode in memory.
4037 * Read the block from disk.
4039 trace_ext4_load_inode(inode);
4041 bh->b_end_io = end_buffer_read_sync;
4042 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4044 if (!buffer_uptodate(bh)) {
4045 EXT4_ERROR_INODE_BLOCK(inode, block,
4046 "unable to read itable block");
4056 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4058 /* We have all inode data except xattrs in memory here. */
4059 return __ext4_get_inode_loc(inode, iloc,
4060 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4063 void ext4_set_inode_flags(struct inode *inode)
4065 unsigned int flags = EXT4_I(inode)->i_flags;
4066 unsigned int new_fl = 0;
4068 if (flags & EXT4_SYNC_FL)
4070 if (flags & EXT4_APPEND_FL)
4072 if (flags & EXT4_IMMUTABLE_FL)
4073 new_fl |= S_IMMUTABLE;
4074 if (flags & EXT4_NOATIME_FL)
4075 new_fl |= S_NOATIME;
4076 if (flags & EXT4_DIRSYNC_FL)
4077 new_fl |= S_DIRSYNC;
4078 set_mask_bits(&inode->i_flags,
4079 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC, new_fl);
4082 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4083 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4085 unsigned int vfs_fl;
4086 unsigned long old_fl, new_fl;
4089 vfs_fl = ei->vfs_inode.i_flags;
4090 old_fl = ei->i_flags;
4091 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4092 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4094 if (vfs_fl & S_SYNC)
4095 new_fl |= EXT4_SYNC_FL;
4096 if (vfs_fl & S_APPEND)
4097 new_fl |= EXT4_APPEND_FL;
4098 if (vfs_fl & S_IMMUTABLE)
4099 new_fl |= EXT4_IMMUTABLE_FL;
4100 if (vfs_fl & S_NOATIME)
4101 new_fl |= EXT4_NOATIME_FL;
4102 if (vfs_fl & S_DIRSYNC)
4103 new_fl |= EXT4_DIRSYNC_FL;
4104 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4107 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4108 struct ext4_inode_info *ei)
4111 struct inode *inode = &(ei->vfs_inode);
4112 struct super_block *sb = inode->i_sb;
4114 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4115 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4116 /* we are using combined 48 bit field */
4117 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4118 le32_to_cpu(raw_inode->i_blocks_lo);
4119 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4120 /* i_blocks represent file system block size */
4121 return i_blocks << (inode->i_blkbits - 9);
4126 return le32_to_cpu(raw_inode->i_blocks_lo);
4130 static inline void ext4_iget_extra_inode(struct inode *inode,
4131 struct ext4_inode *raw_inode,
4132 struct ext4_inode_info *ei)
4134 __le32 *magic = (void *)raw_inode +
4135 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4136 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4137 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4138 ext4_find_inline_data_nolock(inode);
4140 EXT4_I(inode)->i_inline_off = 0;
4143 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4145 struct ext4_iloc iloc;
4146 struct ext4_inode *raw_inode;
4147 struct ext4_inode_info *ei;
4148 struct inode *inode;
4149 journal_t *journal = EXT4_SB(sb)->s_journal;
4155 inode = iget_locked(sb, ino);
4157 return ERR_PTR(-ENOMEM);
4158 if (!(inode->i_state & I_NEW))
4164 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4167 raw_inode = ext4_raw_inode(&iloc);
4169 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4170 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4171 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4172 EXT4_INODE_SIZE(inode->i_sb)) {
4173 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4174 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4175 EXT4_INODE_SIZE(inode->i_sb));
4180 ei->i_extra_isize = 0;
4182 /* Precompute checksum seed for inode metadata */
4183 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4184 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
4185 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4187 __le32 inum = cpu_to_le32(inode->i_ino);
4188 __le32 gen = raw_inode->i_generation;
4189 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4191 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4195 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4196 EXT4_ERROR_INODE(inode, "checksum invalid");
4201 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4202 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4203 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4204 if (!(test_opt(inode->i_sb, NO_UID32))) {
4205 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4206 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4208 i_uid_write(inode, i_uid);
4209 i_gid_write(inode, i_gid);
4210 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4212 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4213 ei->i_inline_off = 0;
4214 ei->i_dir_start_lookup = 0;
4215 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4216 /* We now have enough fields to check if the inode was active or not.
4217 * This is needed because nfsd might try to access dead inodes
4218 * the test is that same one that e2fsck uses
4219 * NeilBrown 1999oct15
4221 if (inode->i_nlink == 0) {
4222 if ((inode->i_mode == 0 ||
4223 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4224 ino != EXT4_BOOT_LOADER_INO) {
4225 /* this inode is deleted */
4229 /* The only unlinked inodes we let through here have
4230 * valid i_mode and are being read by the orphan
4231 * recovery code: that's fine, we're about to complete
4232 * the process of deleting those.
4233 * OR it is the EXT4_BOOT_LOADER_INO which is
4234 * not initialized on a new filesystem. */
4236 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4237 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4238 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4239 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4241 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4242 inode->i_size = ext4_isize(raw_inode);
4243 ei->i_disksize = inode->i_size;
4245 ei->i_reserved_quota = 0;
4247 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4248 ei->i_block_group = iloc.block_group;
4249 ei->i_last_alloc_group = ~0;
4251 * NOTE! The in-memory inode i_data array is in little-endian order
4252 * even on big-endian machines: we do NOT byteswap the block numbers!
4254 for (block = 0; block < EXT4_N_BLOCKS; block++)
4255 ei->i_data[block] = raw_inode->i_block[block];
4256 INIT_LIST_HEAD(&ei->i_orphan);
4259 * Set transaction id's of transactions that have to be committed
4260 * to finish f[data]sync. We set them to currently running transaction
4261 * as we cannot be sure that the inode or some of its metadata isn't
4262 * part of the transaction - the inode could have been reclaimed and
4263 * now it is reread from disk.
4266 transaction_t *transaction;
4269 read_lock(&journal->j_state_lock);
4270 if (journal->j_running_transaction)
4271 transaction = journal->j_running_transaction;
4273 transaction = journal->j_committing_transaction;
4275 tid = transaction->t_tid;
4277 tid = journal->j_commit_sequence;
4278 read_unlock(&journal->j_state_lock);
4279 ei->i_sync_tid = tid;
4280 ei->i_datasync_tid = tid;
4283 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4284 if (ei->i_extra_isize == 0) {
4285 /* The extra space is currently unused. Use it. */
4286 ei->i_extra_isize = sizeof(struct ext4_inode) -
4287 EXT4_GOOD_OLD_INODE_SIZE;
4289 ext4_iget_extra_inode(inode, raw_inode, ei);
4293 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4294 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4295 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4296 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4298 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4299 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4300 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4302 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4306 if (ei->i_file_acl &&
4307 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4308 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4312 } else if (!ext4_has_inline_data(inode)) {
4313 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4314 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4315 (S_ISLNK(inode->i_mode) &&
4316 !ext4_inode_is_fast_symlink(inode))))
4317 /* Validate extent which is part of inode */
4318 ret = ext4_ext_check_inode(inode);
4319 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4320 (S_ISLNK(inode->i_mode) &&
4321 !ext4_inode_is_fast_symlink(inode))) {
4322 /* Validate block references which are part of inode */
4323 ret = ext4_ind_check_inode(inode);
4329 if (S_ISREG(inode->i_mode)) {
4330 inode->i_op = &ext4_file_inode_operations;
4331 inode->i_fop = &ext4_file_operations;
4332 ext4_set_aops(inode);
4333 } else if (S_ISDIR(inode->i_mode)) {
4334 inode->i_op = &ext4_dir_inode_operations;
4335 inode->i_fop = &ext4_dir_operations;
4336 } else if (S_ISLNK(inode->i_mode)) {
4337 if (ext4_inode_is_fast_symlink(inode)) {
4338 inode->i_op = &ext4_fast_symlink_inode_operations;
4339 nd_terminate_link(ei->i_data, inode->i_size,
4340 sizeof(ei->i_data) - 1);
4342 inode->i_op = &ext4_symlink_inode_operations;
4343 ext4_set_aops(inode);
4345 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4346 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4347 inode->i_op = &ext4_special_inode_operations;
4348 if (raw_inode->i_block[0])
4349 init_special_inode(inode, inode->i_mode,
4350 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4352 init_special_inode(inode, inode->i_mode,
4353 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4354 } else if (ino == EXT4_BOOT_LOADER_INO) {
4355 make_bad_inode(inode);
4358 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4362 ext4_set_inode_flags(inode);
4363 unlock_new_inode(inode);
4369 return ERR_PTR(ret);
4372 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4374 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4375 return ERR_PTR(-EIO);
4376 return ext4_iget(sb, ino);
4379 static int ext4_inode_blocks_set(handle_t *handle,
4380 struct ext4_inode *raw_inode,
4381 struct ext4_inode_info *ei)
4383 struct inode *inode = &(ei->vfs_inode);
4384 u64 i_blocks = inode->i_blocks;
4385 struct super_block *sb = inode->i_sb;
4387 if (i_blocks <= ~0U) {
4389 * i_blocks can be represented in a 32 bit variable
4390 * as multiple of 512 bytes
4392 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4393 raw_inode->i_blocks_high = 0;
4394 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4397 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4400 if (i_blocks <= 0xffffffffffffULL) {
4402 * i_blocks can be represented in a 48 bit variable
4403 * as multiple of 512 bytes
4405 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4406 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4407 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4409 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4410 /* i_block is stored in file system block size */
4411 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4412 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4413 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4419 * Post the struct inode info into an on-disk inode location in the
4420 * buffer-cache. This gobbles the caller's reference to the
4421 * buffer_head in the inode location struct.
4423 * The caller must have write access to iloc->bh.
4425 static int ext4_do_update_inode(handle_t *handle,
4426 struct inode *inode,
4427 struct ext4_iloc *iloc)
4429 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4430 struct ext4_inode_info *ei = EXT4_I(inode);
4431 struct buffer_head *bh = iloc->bh;
4432 int err = 0, rc, block;
4433 int need_datasync = 0;
4437 /* For fields not not tracking in the in-memory inode,
4438 * initialise them to zero for new inodes. */
4439 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4440 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4442 ext4_get_inode_flags(ei);
4443 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4444 i_uid = i_uid_read(inode);
4445 i_gid = i_gid_read(inode);
4446 if (!(test_opt(inode->i_sb, NO_UID32))) {
4447 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4448 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4450 * Fix up interoperability with old kernels. Otherwise, old inodes get
4451 * re-used with the upper 16 bits of the uid/gid intact
4454 raw_inode->i_uid_high =
4455 cpu_to_le16(high_16_bits(i_uid));
4456 raw_inode->i_gid_high =
4457 cpu_to_le16(high_16_bits(i_gid));
4459 raw_inode->i_uid_high = 0;
4460 raw_inode->i_gid_high = 0;
4463 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4464 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4465 raw_inode->i_uid_high = 0;
4466 raw_inode->i_gid_high = 0;
4468 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4470 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4471 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4472 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4473 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4475 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4477 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4478 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4479 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4480 cpu_to_le32(EXT4_OS_HURD))
4481 raw_inode->i_file_acl_high =
4482 cpu_to_le16(ei->i_file_acl >> 32);
4483 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4484 if (ei->i_disksize != ext4_isize(raw_inode)) {
4485 ext4_isize_set(raw_inode, ei->i_disksize);
4488 if (ei->i_disksize > 0x7fffffffULL) {
4489 struct super_block *sb = inode->i_sb;
4490 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4491 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4492 EXT4_SB(sb)->s_es->s_rev_level ==
4493 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4494 /* If this is the first large file
4495 * created, add a flag to the superblock.
4497 err = ext4_journal_get_write_access(handle,
4498 EXT4_SB(sb)->s_sbh);
4501 ext4_update_dynamic_rev(sb);
4502 EXT4_SET_RO_COMPAT_FEATURE(sb,
4503 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4504 ext4_handle_sync(handle);
4505 err = ext4_handle_dirty_super(handle, sb);
4508 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4509 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4510 if (old_valid_dev(inode->i_rdev)) {
4511 raw_inode->i_block[0] =
4512 cpu_to_le32(old_encode_dev(inode->i_rdev));
4513 raw_inode->i_block[1] = 0;
4515 raw_inode->i_block[0] = 0;
4516 raw_inode->i_block[1] =
4517 cpu_to_le32(new_encode_dev(inode->i_rdev));
4518 raw_inode->i_block[2] = 0;
4520 } else if (!ext4_has_inline_data(inode)) {
4521 for (block = 0; block < EXT4_N_BLOCKS; block++)
4522 raw_inode->i_block[block] = ei->i_data[block];
4525 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4526 if (ei->i_extra_isize) {
4527 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4528 raw_inode->i_version_hi =
4529 cpu_to_le32(inode->i_version >> 32);
4530 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4533 ext4_inode_csum_set(inode, raw_inode, ei);
4535 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4536 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4539 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4541 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4544 ext4_std_error(inode->i_sb, err);
4549 * ext4_write_inode()
4551 * We are called from a few places:
4553 * - Within generic_file_write() for O_SYNC files.
4554 * Here, there will be no transaction running. We wait for any running
4555 * transaction to commit.
4557 * - Within sys_sync(), kupdate and such.
4558 * We wait on commit, if tol to.
4560 * - Within prune_icache() (PF_MEMALLOC == true)
4561 * Here we simply return. We can't afford to block kswapd on the
4564 * In all cases it is actually safe for us to return without doing anything,
4565 * because the inode has been copied into a raw inode buffer in
4566 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4569 * Note that we are absolutely dependent upon all inode dirtiers doing the
4570 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4571 * which we are interested.
4573 * It would be a bug for them to not do this. The code:
4575 * mark_inode_dirty(inode)
4577 * inode->i_size = expr;
4579 * is in error because a kswapd-driven write_inode() could occur while
4580 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4581 * will no longer be on the superblock's dirty inode list.
4583 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4587 if (current->flags & PF_MEMALLOC)
4590 if (EXT4_SB(inode->i_sb)->s_journal) {
4591 if (ext4_journal_current_handle()) {
4592 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4597 if (wbc->sync_mode != WB_SYNC_ALL)
4600 err = ext4_force_commit(inode->i_sb);
4602 struct ext4_iloc iloc;
4604 err = __ext4_get_inode_loc(inode, &iloc, 0);
4607 if (wbc->sync_mode == WB_SYNC_ALL)
4608 sync_dirty_buffer(iloc.bh);
4609 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4610 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4611 "IO error syncing inode");
4620 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4621 * buffers that are attached to a page stradding i_size and are undergoing
4622 * commit. In that case we have to wait for commit to finish and try again.
4624 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4628 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4629 tid_t commit_tid = 0;
4632 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4634 * All buffers in the last page remain valid? Then there's nothing to
4635 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4638 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4641 page = find_lock_page(inode->i_mapping,
4642 inode->i_size >> PAGE_CACHE_SHIFT);
4645 ret = __ext4_journalled_invalidatepage(page, offset);
4647 page_cache_release(page);
4651 read_lock(&journal->j_state_lock);
4652 if (journal->j_committing_transaction)
4653 commit_tid = journal->j_committing_transaction->t_tid;
4654 read_unlock(&journal->j_state_lock);
4656 jbd2_log_wait_commit(journal, commit_tid);
4663 * Called from notify_change.
4665 * We want to trap VFS attempts to truncate the file as soon as
4666 * possible. In particular, we want to make sure that when the VFS
4667 * shrinks i_size, we put the inode on the orphan list and modify
4668 * i_disksize immediately, so that during the subsequent flushing of
4669 * dirty pages and freeing of disk blocks, we can guarantee that any
4670 * commit will leave the blocks being flushed in an unused state on
4671 * disk. (On recovery, the inode will get truncated and the blocks will
4672 * be freed, so we have a strong guarantee that no future commit will
4673 * leave these blocks visible to the user.)
4675 * Another thing we have to assure is that if we are in ordered mode
4676 * and inode is still attached to the committing transaction, we must
4677 * we start writeout of all the dirty pages which are being truncated.
4678 * This way we are sure that all the data written in the previous
4679 * transaction are already on disk (truncate waits for pages under
4682 * Called with inode->i_mutex down.
4684 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4686 struct inode *inode = dentry->d_inode;
4689 const unsigned int ia_valid = attr->ia_valid;
4691 error = inode_change_ok(inode, attr);
4695 if (is_quota_modification(inode, attr))
4696 dquot_initialize(inode);
4697 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4698 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4701 /* (user+group)*(old+new) structure, inode write (sb,
4702 * inode block, ? - but truncate inode update has it) */
4703 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4704 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4705 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4706 if (IS_ERR(handle)) {
4707 error = PTR_ERR(handle);
4710 error = dquot_transfer(inode, attr);
4712 ext4_journal_stop(handle);
4715 /* Update corresponding info in inode so that everything is in
4716 * one transaction */
4717 if (attr->ia_valid & ATTR_UID)
4718 inode->i_uid = attr->ia_uid;
4719 if (attr->ia_valid & ATTR_GID)
4720 inode->i_gid = attr->ia_gid;
4721 error = ext4_mark_inode_dirty(handle, inode);
4722 ext4_journal_stop(handle);
4725 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4727 loff_t oldsize = inode->i_size;
4729 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4730 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4732 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4736 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4737 inode_inc_iversion(inode);
4739 if (S_ISREG(inode->i_mode) &&
4740 (attr->ia_size < inode->i_size)) {
4741 if (ext4_should_order_data(inode)) {
4742 error = ext4_begin_ordered_truncate(inode,
4747 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4748 if (IS_ERR(handle)) {
4749 error = PTR_ERR(handle);
4752 if (ext4_handle_valid(handle)) {
4753 error = ext4_orphan_add(handle, inode);
4756 EXT4_I(inode)->i_disksize = attr->ia_size;
4757 rc = ext4_mark_inode_dirty(handle, inode);
4760 ext4_journal_stop(handle);
4762 ext4_orphan_del(NULL, inode);
4767 i_size_write(inode, attr->ia_size);
4769 * Blocks are going to be removed from the inode. Wait
4770 * for dio in flight. Temporarily disable
4771 * dioread_nolock to prevent livelock.
4774 if (!ext4_should_journal_data(inode)) {
4775 ext4_inode_block_unlocked_dio(inode);
4776 inode_dio_wait(inode);
4777 ext4_inode_resume_unlocked_dio(inode);
4779 ext4_wait_for_tail_page_commit(inode);
4782 * Truncate pagecache after we've waited for commit
4783 * in data=journal mode to make pages freeable.
4785 truncate_pagecache(inode, oldsize, inode->i_size);
4788 * We want to call ext4_truncate() even if attr->ia_size ==
4789 * inode->i_size for cases like truncation of fallocated space
4791 if (attr->ia_valid & ATTR_SIZE)
4792 ext4_truncate(inode);
4795 setattr_copy(inode, attr);
4796 mark_inode_dirty(inode);
4800 * If the call to ext4_truncate failed to get a transaction handle at
4801 * all, we need to clean up the in-core orphan list manually.
4803 if (orphan && inode->i_nlink)
4804 ext4_orphan_del(NULL, inode);
4806 if (!rc && (ia_valid & ATTR_MODE))
4807 rc = ext4_acl_chmod(inode);
4810 ext4_std_error(inode->i_sb, error);
4816 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4819 struct inode *inode;
4820 unsigned long long delalloc_blocks;
4822 inode = dentry->d_inode;
4823 generic_fillattr(inode, stat);
4826 * We can't update i_blocks if the block allocation is delayed
4827 * otherwise in the case of system crash before the real block
4828 * allocation is done, we will have i_blocks inconsistent with
4829 * on-disk file blocks.
4830 * We always keep i_blocks updated together with real
4831 * allocation. But to not confuse with user, stat
4832 * will return the blocks that include the delayed allocation
4833 * blocks for this file.
4835 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4836 EXT4_I(inode)->i_reserved_data_blocks);
4838 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits-9);
4842 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4844 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4845 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4846 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4850 * Account for index blocks, block groups bitmaps and block group
4851 * descriptor blocks if modify datablocks and index blocks
4852 * worse case, the indexs blocks spread over different block groups
4854 * If datablocks are discontiguous, they are possible to spread over
4855 * different block groups too. If they are contiguous, with flexbg,
4856 * they could still across block group boundary.
4858 * Also account for superblock, inode, quota and xattr blocks
4860 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4862 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4868 * How many index blocks need to touch to modify nrblocks?
4869 * The "Chunk" flag indicating whether the nrblocks is
4870 * physically contiguous on disk
4872 * For Direct IO and fallocate, they calls get_block to allocate
4873 * one single extent at a time, so they could set the "Chunk" flag
4875 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4880 * Now let's see how many group bitmaps and group descriptors need
4890 if (groups > ngroups)
4892 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4893 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4895 /* bitmaps and block group descriptor blocks */
4896 ret += groups + gdpblocks;
4898 /* Blocks for super block, inode, quota and xattr blocks */
4899 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4905 * Calculate the total number of credits to reserve to fit
4906 * the modification of a single pages into a single transaction,
4907 * which may include multiple chunks of block allocations.
4909 * This could be called via ext4_write_begin()
4911 * We need to consider the worse case, when
4912 * one new block per extent.
4914 int ext4_writepage_trans_blocks(struct inode *inode)
4916 int bpp = ext4_journal_blocks_per_page(inode);
4919 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4921 /* Account for data blocks for journalled mode */
4922 if (ext4_should_journal_data(inode))
4928 * Calculate the journal credits for a chunk of data modification.
4930 * This is called from DIO, fallocate or whoever calling
4931 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4933 * journal buffers for data blocks are not included here, as DIO
4934 * and fallocate do no need to journal data buffers.
4936 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4938 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4942 * The caller must have previously called ext4_reserve_inode_write().
4943 * Give this, we know that the caller already has write access to iloc->bh.
4945 int ext4_mark_iloc_dirty(handle_t *handle,
4946 struct inode *inode, struct ext4_iloc *iloc)
4950 if (IS_I_VERSION(inode))
4951 inode_inc_iversion(inode);
4953 /* the do_update_inode consumes one bh->b_count */
4956 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4957 err = ext4_do_update_inode(handle, inode, iloc);
4963 * On success, We end up with an outstanding reference count against
4964 * iloc->bh. This _must_ be cleaned up later.
4968 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4969 struct ext4_iloc *iloc)
4973 err = ext4_get_inode_loc(inode, iloc);
4975 BUFFER_TRACE(iloc->bh, "get_write_access");
4976 err = ext4_journal_get_write_access(handle, iloc->bh);
4982 ext4_std_error(inode->i_sb, err);
4987 * Expand an inode by new_extra_isize bytes.
4988 * Returns 0 on success or negative error number on failure.
4990 static int ext4_expand_extra_isize(struct inode *inode,
4991 unsigned int new_extra_isize,
4992 struct ext4_iloc iloc,
4995 struct ext4_inode *raw_inode;
4996 struct ext4_xattr_ibody_header *header;
4998 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5001 raw_inode = ext4_raw_inode(&iloc);
5003 header = IHDR(inode, raw_inode);
5005 /* No extended attributes present */
5006 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5007 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5008 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5010 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5014 /* try to expand with EAs present */
5015 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5020 * What we do here is to mark the in-core inode as clean with respect to inode
5021 * dirtiness (it may still be data-dirty).
5022 * This means that the in-core inode may be reaped by prune_icache
5023 * without having to perform any I/O. This is a very good thing,
5024 * because *any* task may call prune_icache - even ones which
5025 * have a transaction open against a different journal.
5027 * Is this cheating? Not really. Sure, we haven't written the
5028 * inode out, but prune_icache isn't a user-visible syncing function.
5029 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5030 * we start and wait on commits.
5032 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5034 struct ext4_iloc iloc;
5035 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5036 static unsigned int mnt_count;
5040 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5041 err = ext4_reserve_inode_write(handle, inode, &iloc);
5042 if (ext4_handle_valid(handle) &&
5043 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5044 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5046 * We need extra buffer credits since we may write into EA block
5047 * with this same handle. If journal_extend fails, then it will
5048 * only result in a minor loss of functionality for that inode.
5049 * If this is felt to be critical, then e2fsck should be run to
5050 * force a large enough s_min_extra_isize.
5052 if ((jbd2_journal_extend(handle,
5053 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5054 ret = ext4_expand_extra_isize(inode,
5055 sbi->s_want_extra_isize,
5058 ext4_set_inode_state(inode,
5059 EXT4_STATE_NO_EXPAND);
5061 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5062 ext4_warning(inode->i_sb,
5063 "Unable to expand inode %lu. Delete"
5064 " some EAs or run e2fsck.",
5067 le16_to_cpu(sbi->s_es->s_mnt_count);
5073 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5078 * ext4_dirty_inode() is called from __mark_inode_dirty()
5080 * We're really interested in the case where a file is being extended.
5081 * i_size has been changed by generic_commit_write() and we thus need
5082 * to include the updated inode in the current transaction.
5084 * Also, dquot_alloc_block() will always dirty the inode when blocks
5085 * are allocated to the file.
5087 * If the inode is marked synchronous, we don't honour that here - doing
5088 * so would cause a commit on atime updates, which we don't bother doing.
5089 * We handle synchronous inodes at the highest possible level.
5091 void ext4_dirty_inode(struct inode *inode, int flags)
5095 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5099 ext4_mark_inode_dirty(handle, inode);
5101 ext4_journal_stop(handle);
5108 * Bind an inode's backing buffer_head into this transaction, to prevent
5109 * it from being flushed to disk early. Unlike
5110 * ext4_reserve_inode_write, this leaves behind no bh reference and
5111 * returns no iloc structure, so the caller needs to repeat the iloc
5112 * lookup to mark the inode dirty later.
5114 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5116 struct ext4_iloc iloc;
5120 err = ext4_get_inode_loc(inode, &iloc);
5122 BUFFER_TRACE(iloc.bh, "get_write_access");
5123 err = jbd2_journal_get_write_access(handle, iloc.bh);
5125 err = ext4_handle_dirty_metadata(handle,
5131 ext4_std_error(inode->i_sb, err);
5136 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5143 * We have to be very careful here: changing a data block's
5144 * journaling status dynamically is dangerous. If we write a
5145 * data block to the journal, change the status and then delete
5146 * that block, we risk forgetting to revoke the old log record
5147 * from the journal and so a subsequent replay can corrupt data.
5148 * So, first we make sure that the journal is empty and that
5149 * nobody is changing anything.
5152 journal = EXT4_JOURNAL(inode);
5155 if (is_journal_aborted(journal))
5157 /* We have to allocate physical blocks for delalloc blocks
5158 * before flushing journal. otherwise delalloc blocks can not
5159 * be allocated any more. even more truncate on delalloc blocks
5160 * could trigger BUG by flushing delalloc blocks in journal.
5161 * There is no delalloc block in non-journal data mode.
5163 if (val && test_opt(inode->i_sb, DELALLOC)) {
5164 err = ext4_alloc_da_blocks(inode);
5169 /* Wait for all existing dio workers */
5170 ext4_inode_block_unlocked_dio(inode);
5171 inode_dio_wait(inode);
5173 jbd2_journal_lock_updates(journal);
5176 * OK, there are no updates running now, and all cached data is
5177 * synced to disk. We are now in a completely consistent state
5178 * which doesn't have anything in the journal, and we know that
5179 * no filesystem updates are running, so it is safe to modify
5180 * the inode's in-core data-journaling state flag now.
5184 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5186 jbd2_journal_flush(journal);
5187 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5189 ext4_set_aops(inode);
5191 jbd2_journal_unlock_updates(journal);
5192 ext4_inode_resume_unlocked_dio(inode);
5194 /* Finally we can mark the inode as dirty. */
5196 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5198 return PTR_ERR(handle);
5200 err = ext4_mark_inode_dirty(handle, inode);
5201 ext4_handle_sync(handle);
5202 ext4_journal_stop(handle);
5203 ext4_std_error(inode->i_sb, err);
5208 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5210 return !buffer_mapped(bh);
5213 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5215 struct page *page = vmf->page;
5219 struct file *file = vma->vm_file;
5220 struct inode *inode = file_inode(file);
5221 struct address_space *mapping = inode->i_mapping;
5223 get_block_t *get_block;
5226 sb_start_pagefault(inode->i_sb);
5227 file_update_time(vma->vm_file);
5228 /* Delalloc case is easy... */
5229 if (test_opt(inode->i_sb, DELALLOC) &&
5230 !ext4_should_journal_data(inode) &&
5231 !ext4_nonda_switch(inode->i_sb)) {
5233 ret = __block_page_mkwrite(vma, vmf,
5234 ext4_da_get_block_prep);
5235 } while (ret == -ENOSPC &&
5236 ext4_should_retry_alloc(inode->i_sb, &retries));
5241 size = i_size_read(inode);
5242 /* Page got truncated from under us? */
5243 if (page->mapping != mapping || page_offset(page) > size) {
5245 ret = VM_FAULT_NOPAGE;
5249 if (page->index == size >> PAGE_CACHE_SHIFT)
5250 len = size & ~PAGE_CACHE_MASK;
5252 len = PAGE_CACHE_SIZE;
5254 * Return if we have all the buffers mapped. This avoids the need to do
5255 * journal_start/journal_stop which can block and take a long time
5257 if (page_has_buffers(page)) {
5258 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5260 ext4_bh_unmapped)) {
5261 /* Wait so that we don't change page under IO */
5262 wait_for_stable_page(page);
5263 ret = VM_FAULT_LOCKED;
5268 /* OK, we need to fill the hole... */
5269 if (ext4_should_dioread_nolock(inode))
5270 get_block = ext4_get_block_write;
5272 get_block = ext4_get_block;
5274 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5275 ext4_writepage_trans_blocks(inode));
5276 if (IS_ERR(handle)) {
5277 ret = VM_FAULT_SIGBUS;
5280 ret = __block_page_mkwrite(vma, vmf, get_block);
5281 if (!ret && ext4_should_journal_data(inode)) {
5282 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5283 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5285 ret = VM_FAULT_SIGBUS;
5286 ext4_journal_stop(handle);
5289 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5291 ext4_journal_stop(handle);
5292 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5295 ret = block_page_mkwrite_return(ret);
5297 sb_end_pagefault(inode->i_sb);