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 wait_on_page_writeback(page);
1037 if (ext4_should_dioread_nolock(inode))
1038 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1040 ret = __block_write_begin(page, pos, len, ext4_get_block);
1042 if (!ret && ext4_should_journal_data(inode)) {
1043 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1045 do_journal_get_write_access);
1051 * __block_write_begin may have instantiated a few blocks
1052 * outside i_size. Trim these off again. Don't need
1053 * i_size_read because we hold i_mutex.
1055 * Add inode to orphan list in case we crash before
1058 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1059 ext4_orphan_add(handle, inode);
1061 ext4_journal_stop(handle);
1062 if (pos + len > inode->i_size) {
1063 ext4_truncate_failed_write(inode);
1065 * If truncate failed early the inode might
1066 * still be on the orphan list; we need to
1067 * make sure the inode is removed from the
1068 * orphan list in that case.
1071 ext4_orphan_del(NULL, inode);
1074 if (ret == -ENOSPC &&
1075 ext4_should_retry_alloc(inode->i_sb, &retries))
1077 page_cache_release(page);
1084 /* For write_end() in data=journal mode */
1085 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1088 if (!buffer_mapped(bh) || buffer_freed(bh))
1090 set_buffer_uptodate(bh);
1091 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1092 clear_buffer_meta(bh);
1093 clear_buffer_prio(bh);
1098 * We need to pick up the new inode size which generic_commit_write gave us
1099 * `file' can be NULL - eg, when called from page_symlink().
1101 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1102 * buffers are managed internally.
1104 static int ext4_write_end(struct file *file,
1105 struct address_space *mapping,
1106 loff_t pos, unsigned len, unsigned copied,
1107 struct page *page, void *fsdata)
1109 handle_t *handle = ext4_journal_current_handle();
1110 struct inode *inode = mapping->host;
1112 int i_size_changed = 0;
1114 trace_ext4_write_end(inode, pos, len, copied);
1115 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1116 ret = ext4_jbd2_file_inode(handle, inode);
1119 page_cache_release(page);
1124 if (ext4_has_inline_data(inode)) {
1125 ret = ext4_write_inline_data_end(inode, pos, len,
1131 copied = block_write_end(file, mapping, pos,
1132 len, copied, page, fsdata);
1135 * No need to use i_size_read() here, the i_size
1136 * cannot change under us because we hole i_mutex.
1138 * But it's important to update i_size while still holding page lock:
1139 * page writeout could otherwise come in and zero beyond i_size.
1141 if (pos + copied > inode->i_size) {
1142 i_size_write(inode, pos + copied);
1146 if (pos + copied > EXT4_I(inode)->i_disksize) {
1147 /* We need to mark inode dirty even if
1148 * new_i_size is less that inode->i_size
1149 * but greater than i_disksize. (hint delalloc)
1151 ext4_update_i_disksize(inode, (pos + copied));
1155 page_cache_release(page);
1158 * Don't mark the inode dirty under page lock. First, it unnecessarily
1159 * makes the holding time of page lock longer. Second, it forces lock
1160 * ordering of page lock and transaction start for journaling
1164 ext4_mark_inode_dirty(handle, inode);
1168 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1169 /* if we have allocated more blocks and copied
1170 * less. We will have blocks allocated outside
1171 * inode->i_size. So truncate them
1173 ext4_orphan_add(handle, inode);
1175 ret2 = ext4_journal_stop(handle);
1179 if (pos + len > inode->i_size) {
1180 ext4_truncate_failed_write(inode);
1182 * If truncate failed early the inode might still be
1183 * on the orphan list; we need to make sure the inode
1184 * is removed from the orphan list in that case.
1187 ext4_orphan_del(NULL, inode);
1190 return ret ? ret : copied;
1193 static int ext4_journalled_write_end(struct file *file,
1194 struct address_space *mapping,
1195 loff_t pos, unsigned len, unsigned copied,
1196 struct page *page, void *fsdata)
1198 handle_t *handle = ext4_journal_current_handle();
1199 struct inode *inode = mapping->host;
1205 trace_ext4_journalled_write_end(inode, pos, len, copied);
1206 from = pos & (PAGE_CACHE_SIZE - 1);
1209 BUG_ON(!ext4_handle_valid(handle));
1211 if (ext4_has_inline_data(inode))
1212 copied = ext4_write_inline_data_end(inode, pos, len,
1216 if (!PageUptodate(page))
1218 page_zero_new_buffers(page, from+copied, to);
1221 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1222 to, &partial, write_end_fn);
1224 SetPageUptodate(page);
1226 new_i_size = pos + copied;
1227 if (new_i_size > inode->i_size)
1228 i_size_write(inode, pos+copied);
1229 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1230 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1231 if (new_i_size > EXT4_I(inode)->i_disksize) {
1232 ext4_update_i_disksize(inode, new_i_size);
1233 ret2 = ext4_mark_inode_dirty(handle, inode);
1239 page_cache_release(page);
1240 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1241 /* if we have allocated more blocks and copied
1242 * less. We will have blocks allocated outside
1243 * inode->i_size. So truncate them
1245 ext4_orphan_add(handle, inode);
1247 ret2 = ext4_journal_stop(handle);
1250 if (pos + len > inode->i_size) {
1251 ext4_truncate_failed_write(inode);
1253 * If truncate failed early the inode might still be
1254 * on the orphan list; we need to make sure the inode
1255 * is removed from the orphan list in that case.
1258 ext4_orphan_del(NULL, inode);
1261 return ret ? ret : copied;
1265 * Reserve a metadata for a single block located at lblock
1267 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1269 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1270 struct ext4_inode_info *ei = EXT4_I(inode);
1271 unsigned int md_needed;
1272 ext4_lblk_t save_last_lblock;
1276 * recalculate the amount of metadata blocks to reserve
1277 * in order to allocate nrblocks
1278 * worse case is one extent per block
1280 spin_lock(&ei->i_block_reservation_lock);
1282 * ext4_calc_metadata_amount() has side effects, which we have
1283 * to be prepared undo if we fail to claim space.
1285 save_len = ei->i_da_metadata_calc_len;
1286 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1287 md_needed = EXT4_NUM_B2C(sbi,
1288 ext4_calc_metadata_amount(inode, lblock));
1289 trace_ext4_da_reserve_space(inode, md_needed);
1292 * We do still charge estimated metadata to the sb though;
1293 * we cannot afford to run out of free blocks.
1295 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1296 ei->i_da_metadata_calc_len = save_len;
1297 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1298 spin_unlock(&ei->i_block_reservation_lock);
1301 ei->i_reserved_meta_blocks += md_needed;
1302 spin_unlock(&ei->i_block_reservation_lock);
1304 return 0; /* success */
1308 * Reserve a single cluster located at lblock
1310 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1312 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1313 struct ext4_inode_info *ei = EXT4_I(inode);
1314 unsigned int md_needed;
1316 ext4_lblk_t save_last_lblock;
1320 * We will charge metadata quota at writeout time; this saves
1321 * us from metadata over-estimation, though we may go over by
1322 * a small amount in the end. Here we just reserve for data.
1324 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1329 * recalculate the amount of metadata blocks to reserve
1330 * in order to allocate nrblocks
1331 * worse case is one extent per block
1333 spin_lock(&ei->i_block_reservation_lock);
1335 * ext4_calc_metadata_amount() has side effects, which we have
1336 * to be prepared undo if we fail to claim space.
1338 save_len = ei->i_da_metadata_calc_len;
1339 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1340 md_needed = EXT4_NUM_B2C(sbi,
1341 ext4_calc_metadata_amount(inode, lblock));
1342 trace_ext4_da_reserve_space(inode, md_needed);
1345 * We do still charge estimated metadata to the sb though;
1346 * we cannot afford to run out of free blocks.
1348 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1349 ei->i_da_metadata_calc_len = save_len;
1350 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1351 spin_unlock(&ei->i_block_reservation_lock);
1352 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1355 ei->i_reserved_data_blocks++;
1356 ei->i_reserved_meta_blocks += md_needed;
1357 spin_unlock(&ei->i_block_reservation_lock);
1359 return 0; /* success */
1362 static void ext4_da_release_space(struct inode *inode, int to_free)
1364 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1365 struct ext4_inode_info *ei = EXT4_I(inode);
1368 return; /* Nothing to release, exit */
1370 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1372 trace_ext4_da_release_space(inode, to_free);
1373 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1375 * if there aren't enough reserved blocks, then the
1376 * counter is messed up somewhere. Since this
1377 * function is called from invalidate page, it's
1378 * harmless to return without any action.
1380 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1381 "ino %lu, to_free %d with only %d reserved "
1382 "data blocks", inode->i_ino, to_free,
1383 ei->i_reserved_data_blocks);
1385 to_free = ei->i_reserved_data_blocks;
1387 ei->i_reserved_data_blocks -= to_free;
1389 if (ei->i_reserved_data_blocks == 0) {
1391 * We can release all of the reserved metadata blocks
1392 * only when we have written all of the delayed
1393 * allocation blocks.
1394 * Note that in case of bigalloc, i_reserved_meta_blocks,
1395 * i_reserved_data_blocks, etc. refer to number of clusters.
1397 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1398 ei->i_reserved_meta_blocks);
1399 ei->i_reserved_meta_blocks = 0;
1400 ei->i_da_metadata_calc_len = 0;
1403 /* update fs dirty data blocks counter */
1404 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1406 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1408 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1411 static void ext4_da_page_release_reservation(struct page *page,
1412 unsigned long offset)
1415 struct buffer_head *head, *bh;
1416 unsigned int curr_off = 0;
1417 struct inode *inode = page->mapping->host;
1418 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1422 head = page_buffers(page);
1425 unsigned int next_off = curr_off + bh->b_size;
1427 if ((offset <= curr_off) && (buffer_delay(bh))) {
1429 clear_buffer_delay(bh);
1431 curr_off = next_off;
1432 } while ((bh = bh->b_this_page) != head);
1435 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1436 ext4_es_remove_extent(inode, lblk, to_release);
1439 /* If we have released all the blocks belonging to a cluster, then we
1440 * need to release the reserved space for that cluster. */
1441 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1442 while (num_clusters > 0) {
1443 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1444 ((num_clusters - 1) << sbi->s_cluster_bits);
1445 if (sbi->s_cluster_ratio == 1 ||
1446 !ext4_find_delalloc_cluster(inode, lblk))
1447 ext4_da_release_space(inode, 1);
1454 * Delayed allocation stuff
1458 * mpage_da_submit_io - walks through extent of pages and try to write
1459 * them with writepage() call back
1461 * @mpd->inode: inode
1462 * @mpd->first_page: first page of the extent
1463 * @mpd->next_page: page after the last page of the extent
1465 * By the time mpage_da_submit_io() is called we expect all blocks
1466 * to be allocated. this may be wrong if allocation failed.
1468 * As pages are already locked by write_cache_pages(), we can't use it
1470 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1471 struct ext4_map_blocks *map)
1473 struct pagevec pvec;
1474 unsigned long index, end;
1475 int ret = 0, err, nr_pages, i;
1476 struct inode *inode = mpd->inode;
1477 struct address_space *mapping = inode->i_mapping;
1478 loff_t size = i_size_read(inode);
1479 unsigned int len, block_start;
1480 struct buffer_head *bh, *page_bufs = NULL;
1481 sector_t pblock = 0, cur_logical = 0;
1482 struct ext4_io_submit io_submit;
1484 BUG_ON(mpd->next_page <= mpd->first_page);
1485 memset(&io_submit, 0, sizeof(io_submit));
1487 * We need to start from the first_page to the next_page - 1
1488 * to make sure we also write the mapped dirty buffer_heads.
1489 * If we look at mpd->b_blocknr we would only be looking
1490 * at the currently mapped buffer_heads.
1492 index = mpd->first_page;
1493 end = mpd->next_page - 1;
1495 pagevec_init(&pvec, 0);
1496 while (index <= end) {
1497 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1500 for (i = 0; i < nr_pages; i++) {
1502 struct page *page = pvec.pages[i];
1504 index = page->index;
1508 if (index == size >> PAGE_CACHE_SHIFT)
1509 len = size & ~PAGE_CACHE_MASK;
1511 len = PAGE_CACHE_SIZE;
1513 cur_logical = index << (PAGE_CACHE_SHIFT -
1515 pblock = map->m_pblk + (cur_logical -
1520 BUG_ON(!PageLocked(page));
1521 BUG_ON(PageWriteback(page));
1523 bh = page_bufs = page_buffers(page);
1526 if (map && (cur_logical >= map->m_lblk) &&
1527 (cur_logical <= (map->m_lblk +
1528 (map->m_len - 1)))) {
1529 if (buffer_delay(bh)) {
1530 clear_buffer_delay(bh);
1531 bh->b_blocknr = pblock;
1533 if (buffer_unwritten(bh) ||
1535 BUG_ON(bh->b_blocknr != pblock);
1536 if (map->m_flags & EXT4_MAP_UNINIT)
1537 set_buffer_uninit(bh);
1538 clear_buffer_unwritten(bh);
1542 * skip page if block allocation undone and
1545 if (ext4_bh_delay_or_unwritten(NULL, bh))
1547 bh = bh->b_this_page;
1548 block_start += bh->b_size;
1551 } while (bh != page_bufs);
1558 clear_page_dirty_for_io(page);
1559 err = ext4_bio_write_page(&io_submit, page, len,
1562 mpd->pages_written++;
1564 * In error case, we have to continue because
1565 * remaining pages are still locked
1570 pagevec_release(&pvec);
1572 ext4_io_submit(&io_submit);
1576 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1580 struct pagevec pvec;
1581 struct inode *inode = mpd->inode;
1582 struct address_space *mapping = inode->i_mapping;
1583 ext4_lblk_t start, last;
1585 index = mpd->first_page;
1586 end = mpd->next_page - 1;
1588 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1589 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1590 ext4_es_remove_extent(inode, start, last - start + 1);
1592 pagevec_init(&pvec, 0);
1593 while (index <= end) {
1594 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1597 for (i = 0; i < nr_pages; i++) {
1598 struct page *page = pvec.pages[i];
1599 if (page->index > end)
1601 BUG_ON(!PageLocked(page));
1602 BUG_ON(PageWriteback(page));
1603 block_invalidatepage(page, 0);
1604 ClearPageUptodate(page);
1607 index = pvec.pages[nr_pages - 1]->index + 1;
1608 pagevec_release(&pvec);
1613 static void ext4_print_free_blocks(struct inode *inode)
1615 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1616 struct super_block *sb = inode->i_sb;
1617 struct ext4_inode_info *ei = EXT4_I(inode);
1619 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1620 EXT4_C2B(EXT4_SB(inode->i_sb),
1621 ext4_count_free_clusters(sb)));
1622 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1623 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1624 (long long) EXT4_C2B(EXT4_SB(sb),
1625 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1626 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1627 (long long) EXT4_C2B(EXT4_SB(sb),
1628 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1629 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1630 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1631 ei->i_reserved_data_blocks);
1632 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1633 ei->i_reserved_meta_blocks);
1634 ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u",
1635 ei->i_allocated_meta_blocks);
1640 * mpage_da_map_and_submit - go through given space, map them
1641 * if necessary, and then submit them for I/O
1643 * @mpd - bh describing space
1645 * The function skips space we know is already mapped to disk blocks.
1648 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1650 int err, blks, get_blocks_flags;
1651 struct ext4_map_blocks map, *mapp = NULL;
1652 sector_t next = mpd->b_blocknr;
1653 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1654 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1655 handle_t *handle = NULL;
1658 * If the blocks are mapped already, or we couldn't accumulate
1659 * any blocks, then proceed immediately to the submission stage.
1661 if ((mpd->b_size == 0) ||
1662 ((mpd->b_state & (1 << BH_Mapped)) &&
1663 !(mpd->b_state & (1 << BH_Delay)) &&
1664 !(mpd->b_state & (1 << BH_Unwritten))))
1667 handle = ext4_journal_current_handle();
1671 * Call ext4_map_blocks() to allocate any delayed allocation
1672 * blocks, or to convert an uninitialized extent to be
1673 * initialized (in the case where we have written into
1674 * one or more preallocated blocks).
1676 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1677 * indicate that we are on the delayed allocation path. This
1678 * affects functions in many different parts of the allocation
1679 * call path. This flag exists primarily because we don't
1680 * want to change *many* call functions, so ext4_map_blocks()
1681 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1682 * inode's allocation semaphore is taken.
1684 * If the blocks in questions were delalloc blocks, set
1685 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1686 * variables are updated after the blocks have been allocated.
1689 map.m_len = max_blocks;
1691 * We're in delalloc path and it is possible that we're going to
1692 * need more metadata blocks than previously reserved. However
1693 * we must not fail because we're in writeback and there is
1694 * nothing we can do about it so it might result in data loss.
1695 * So use reserved blocks to allocate metadata if possible.
1697 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
1698 EXT4_GET_BLOCKS_METADATA_NOFAIL;
1699 if (ext4_should_dioread_nolock(mpd->inode))
1700 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1701 if (mpd->b_state & (1 << BH_Delay))
1702 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1705 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1707 struct super_block *sb = mpd->inode->i_sb;
1711 * If get block returns EAGAIN or ENOSPC and there
1712 * appears to be free blocks we will just let
1713 * mpage_da_submit_io() unlock all of the pages.
1718 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1724 * get block failure will cause us to loop in
1725 * writepages, because a_ops->writepage won't be able
1726 * to make progress. The page will be redirtied by
1727 * writepage and writepages will again try to write
1730 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1731 ext4_msg(sb, KERN_CRIT,
1732 "delayed block allocation failed for inode %lu "
1733 "at logical offset %llu with max blocks %zd "
1734 "with error %d", mpd->inode->i_ino,
1735 (unsigned long long) next,
1736 mpd->b_size >> mpd->inode->i_blkbits, err);
1737 ext4_msg(sb, KERN_CRIT,
1738 "This should not happen!! Data will be lost");
1740 ext4_print_free_blocks(mpd->inode);
1742 /* invalidate all the pages */
1743 ext4_da_block_invalidatepages(mpd);
1745 /* Mark this page range as having been completed */
1752 if (map.m_flags & EXT4_MAP_NEW) {
1753 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1756 for (i = 0; i < map.m_len; i++)
1757 unmap_underlying_metadata(bdev, map.m_pblk + i);
1761 * Update on-disk size along with block allocation.
1763 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1764 if (disksize > i_size_read(mpd->inode))
1765 disksize = i_size_read(mpd->inode);
1766 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1767 ext4_update_i_disksize(mpd->inode, disksize);
1768 err = ext4_mark_inode_dirty(handle, mpd->inode);
1770 ext4_error(mpd->inode->i_sb,
1771 "Failed to mark inode %lu dirty",
1776 mpage_da_submit_io(mpd, mapp);
1780 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1781 (1 << BH_Delay) | (1 << BH_Unwritten))
1784 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1786 * @mpd->lbh - extent of blocks
1787 * @logical - logical number of the block in the file
1788 * @b_state - b_state of the buffer head added
1790 * the function is used to collect contig. blocks in same state
1792 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical,
1793 unsigned long b_state)
1796 int blkbits = mpd->inode->i_blkbits;
1797 int nrblocks = mpd->b_size >> blkbits;
1800 * XXX Don't go larger than mballoc is willing to allocate
1801 * This is a stopgap solution. We eventually need to fold
1802 * mpage_da_submit_io() into this function and then call
1803 * ext4_map_blocks() multiple times in a loop
1805 if (nrblocks >= (8*1024*1024 >> blkbits))
1808 /* check if the reserved journal credits might overflow */
1809 if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) {
1810 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1812 * With non-extent format we are limited by the journal
1813 * credit available. Total credit needed to insert
1814 * nrblocks contiguous blocks is dependent on the
1815 * nrblocks. So limit nrblocks.
1821 * First block in the extent
1823 if (mpd->b_size == 0) {
1824 mpd->b_blocknr = logical;
1825 mpd->b_size = 1 << blkbits;
1826 mpd->b_state = b_state & BH_FLAGS;
1830 next = mpd->b_blocknr + nrblocks;
1832 * Can we merge the block to our big extent?
1834 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1835 mpd->b_size += 1 << blkbits;
1841 * We couldn't merge the block to our extent, so we
1842 * need to flush current extent and start new one
1844 mpage_da_map_and_submit(mpd);
1848 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1850 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1854 * This function is grabs code from the very beginning of
1855 * ext4_map_blocks, but assumes that the caller is from delayed write
1856 * time. This function looks up the requested blocks and sets the
1857 * buffer delay bit under the protection of i_data_sem.
1859 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1860 struct ext4_map_blocks *map,
1861 struct buffer_head *bh)
1863 struct extent_status es;
1865 sector_t invalid_block = ~((sector_t) 0xffff);
1866 #ifdef ES_AGGRESSIVE_TEST
1867 struct ext4_map_blocks orig_map;
1869 memcpy(&orig_map, map, sizeof(*map));
1872 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1876 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1877 "logical block %lu\n", inode->i_ino, map->m_len,
1878 (unsigned long) map->m_lblk);
1880 /* Lookup extent status tree firstly */
1881 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1883 if (ext4_es_is_hole(&es)) {
1885 down_read((&EXT4_I(inode)->i_data_sem));
1890 * Delayed extent could be allocated by fallocate.
1891 * So we need to check it.
1893 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1894 map_bh(bh, inode->i_sb, invalid_block);
1896 set_buffer_delay(bh);
1900 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1901 retval = es.es_len - (iblock - es.es_lblk);
1902 if (retval > map->m_len)
1903 retval = map->m_len;
1904 map->m_len = retval;
1905 if (ext4_es_is_written(&es))
1906 map->m_flags |= EXT4_MAP_MAPPED;
1907 else if (ext4_es_is_unwritten(&es))
1908 map->m_flags |= EXT4_MAP_UNWRITTEN;
1912 #ifdef ES_AGGRESSIVE_TEST
1913 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1919 * Try to see if we can get the block without requesting a new
1920 * file system block.
1922 down_read((&EXT4_I(inode)->i_data_sem));
1923 if (ext4_has_inline_data(inode)) {
1925 * We will soon create blocks for this page, and let
1926 * us pretend as if the blocks aren't allocated yet.
1927 * In case of clusters, we have to handle the work
1928 * of mapping from cluster so that the reserved space
1929 * is calculated properly.
1931 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1932 ext4_find_delalloc_cluster(inode, map->m_lblk))
1933 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1935 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1936 retval = ext4_ext_map_blocks(NULL, inode, map,
1937 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1939 retval = ext4_ind_map_blocks(NULL, inode, map,
1940 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1946 * XXX: __block_prepare_write() unmaps passed block,
1950 * If the block was allocated from previously allocated cluster,
1951 * then we don't need to reserve it again. However we still need
1952 * to reserve metadata for every block we're going to write.
1954 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1955 ret = ext4_da_reserve_space(inode, iblock);
1957 /* not enough space to reserve */
1962 ret = ext4_da_reserve_metadata(inode, iblock);
1964 /* not enough space to reserve */
1970 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1971 ~0, EXTENT_STATUS_DELAYED);
1977 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1978 * and it should not appear on the bh->b_state.
1980 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1982 map_bh(bh, inode->i_sb, invalid_block);
1984 set_buffer_delay(bh);
1985 } else if (retval > 0) {
1987 unsigned long long status;
1989 #ifdef ES_AGGRESSIVE_TEST
1990 if (retval != map->m_len) {
1991 printk("ES len assertation failed for inode: %lu "
1992 "retval %d != map->m_len %d "
1993 "in %s (lookup)\n", inode->i_ino, retval,
1994 map->m_len, __func__);
1998 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1999 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
2000 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
2001 map->m_pblk, status);
2007 up_read((&EXT4_I(inode)->i_data_sem));
2013 * This is a special get_blocks_t callback which is used by
2014 * ext4_da_write_begin(). It will either return mapped block or
2015 * reserve space for a single block.
2017 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2018 * We also have b_blocknr = -1 and b_bdev initialized properly
2020 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2021 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2022 * initialized properly.
2024 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2025 struct buffer_head *bh, int create)
2027 struct ext4_map_blocks map;
2030 BUG_ON(create == 0);
2031 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2033 map.m_lblk = iblock;
2037 * first, we need to know whether the block is allocated already
2038 * preallocated blocks are unmapped but should treated
2039 * the same as allocated blocks.
2041 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
2045 map_bh(bh, inode->i_sb, map.m_pblk);
2046 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2048 if (buffer_unwritten(bh)) {
2049 /* A delayed write to unwritten bh should be marked
2050 * new and mapped. Mapped ensures that we don't do
2051 * get_block multiple times when we write to the same
2052 * offset and new ensures that we do proper zero out
2053 * for partial write.
2056 set_buffer_mapped(bh);
2061 static int bget_one(handle_t *handle, struct buffer_head *bh)
2067 static int bput_one(handle_t *handle, struct buffer_head *bh)
2073 static int __ext4_journalled_writepage(struct page *page,
2076 struct address_space *mapping = page->mapping;
2077 struct inode *inode = mapping->host;
2078 struct buffer_head *page_bufs = NULL;
2079 handle_t *handle = NULL;
2080 int ret = 0, err = 0;
2081 int inline_data = ext4_has_inline_data(inode);
2082 struct buffer_head *inode_bh = NULL;
2084 ClearPageChecked(page);
2087 BUG_ON(page->index != 0);
2088 BUG_ON(len > ext4_get_max_inline_size(inode));
2089 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2090 if (inode_bh == NULL)
2093 page_bufs = page_buffers(page);
2098 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2101 /* As soon as we unlock the page, it can go away, but we have
2102 * references to buffers so we are safe */
2105 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2106 ext4_writepage_trans_blocks(inode));
2107 if (IS_ERR(handle)) {
2108 ret = PTR_ERR(handle);
2112 BUG_ON(!ext4_handle_valid(handle));
2115 ret = ext4_journal_get_write_access(handle, inode_bh);
2117 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2120 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2121 do_journal_get_write_access);
2123 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2128 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2129 err = ext4_journal_stop(handle);
2133 if (!ext4_has_inline_data(inode))
2134 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2136 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2143 * Note that we don't need to start a transaction unless we're journaling data
2144 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2145 * need to file the inode to the transaction's list in ordered mode because if
2146 * we are writing back data added by write(), the inode is already there and if
2147 * we are writing back data modified via mmap(), no one guarantees in which
2148 * transaction the data will hit the disk. In case we are journaling data, we
2149 * cannot start transaction directly because transaction start ranks above page
2150 * lock so we have to do some magic.
2152 * This function can get called via...
2153 * - ext4_da_writepages after taking page lock (have journal handle)
2154 * - journal_submit_inode_data_buffers (no journal handle)
2155 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2156 * - grab_page_cache when doing write_begin (have journal handle)
2158 * We don't do any block allocation in this function. If we have page with
2159 * multiple blocks we need to write those buffer_heads that are mapped. This
2160 * is important for mmaped based write. So if we do with blocksize 1K
2161 * truncate(f, 1024);
2162 * a = mmap(f, 0, 4096);
2164 * truncate(f, 4096);
2165 * we have in the page first buffer_head mapped via page_mkwrite call back
2166 * but other buffer_heads would be unmapped but dirty (dirty done via the
2167 * do_wp_page). So writepage should write the first block. If we modify
2168 * the mmap area beyond 1024 we will again get a page_fault and the
2169 * page_mkwrite callback will do the block allocation and mark the
2170 * buffer_heads mapped.
2172 * We redirty the page if we have any buffer_heads that is either delay or
2173 * unwritten in the page.
2175 * We can get recursively called as show below.
2177 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2180 * But since we don't do any block allocation we should not deadlock.
2181 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2183 static int ext4_writepage(struct page *page,
2184 struct writeback_control *wbc)
2189 struct buffer_head *page_bufs = NULL;
2190 struct inode *inode = page->mapping->host;
2191 struct ext4_io_submit io_submit;
2193 trace_ext4_writepage(page);
2194 size = i_size_read(inode);
2195 if (page->index == size >> PAGE_CACHE_SHIFT)
2196 len = size & ~PAGE_CACHE_MASK;
2198 len = PAGE_CACHE_SIZE;
2200 page_bufs = page_buffers(page);
2202 * We cannot do block allocation or other extent handling in this
2203 * function. If there are buffers needing that, we have to redirty
2204 * the page. But we may reach here when we do a journal commit via
2205 * journal_submit_inode_data_buffers() and in that case we must write
2206 * allocated buffers to achieve data=ordered mode guarantees.
2208 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2209 ext4_bh_delay_or_unwritten)) {
2210 redirty_page_for_writepage(wbc, page);
2211 if (current->flags & PF_MEMALLOC) {
2213 * For memory cleaning there's no point in writing only
2214 * some buffers. So just bail out. Warn if we came here
2215 * from direct reclaim.
2217 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2224 if (PageChecked(page) && ext4_should_journal_data(inode))
2226 * It's mmapped pagecache. Add buffers and journal it. There
2227 * doesn't seem much point in redirtying the page here.
2229 return __ext4_journalled_writepage(page, len);
2231 memset(&io_submit, 0, sizeof(io_submit));
2232 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2233 ext4_io_submit(&io_submit);
2238 * This is called via ext4_da_writepages() to
2239 * calculate the total number of credits to reserve to fit
2240 * a single extent allocation into a single transaction,
2241 * ext4_da_writpeages() will loop calling this before
2242 * the block allocation.
2245 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2247 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2250 * With non-extent format the journal credit needed to
2251 * insert nrblocks contiguous block is dependent on
2252 * number of contiguous block. So we will limit
2253 * number of contiguous block to a sane value
2255 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2256 (max_blocks > EXT4_MAX_TRANS_DATA))
2257 max_blocks = EXT4_MAX_TRANS_DATA;
2259 return ext4_chunk_trans_blocks(inode, max_blocks);
2263 * write_cache_pages_da - walk the list of dirty pages of the given
2264 * address space and accumulate pages that need writing, and call
2265 * mpage_da_map_and_submit to map a single contiguous memory region
2266 * and then write them.
2268 static int write_cache_pages_da(handle_t *handle,
2269 struct address_space *mapping,
2270 struct writeback_control *wbc,
2271 struct mpage_da_data *mpd,
2272 pgoff_t *done_index)
2274 struct buffer_head *bh, *head;
2275 struct inode *inode = mapping->host;
2276 struct pagevec pvec;
2277 unsigned int nr_pages;
2280 long nr_to_write = wbc->nr_to_write;
2281 int i, tag, ret = 0;
2283 memset(mpd, 0, sizeof(struct mpage_da_data));
2286 pagevec_init(&pvec, 0);
2287 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2288 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2290 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2291 tag = PAGECACHE_TAG_TOWRITE;
2293 tag = PAGECACHE_TAG_DIRTY;
2295 *done_index = index;
2296 while (index <= end) {
2297 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2298 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2302 for (i = 0; i < nr_pages; i++) {
2303 struct page *page = pvec.pages[i];
2306 * At this point, the page may be truncated or
2307 * invalidated (changing page->mapping to NULL), or
2308 * even swizzled back from swapper_space to tmpfs file
2309 * mapping. However, page->index will not change
2310 * because we have a reference on the page.
2312 if (page->index > end)
2315 *done_index = page->index + 1;
2318 * If we can't merge this page, and we have
2319 * accumulated an contiguous region, write it
2321 if ((mpd->next_page != page->index) &&
2322 (mpd->next_page != mpd->first_page)) {
2323 mpage_da_map_and_submit(mpd);
2324 goto ret_extent_tail;
2330 * If the page is no longer dirty, or its
2331 * mapping no longer corresponds to inode we
2332 * are writing (which means it has been
2333 * truncated or invalidated), or the page is
2334 * already under writeback and we are not
2335 * doing a data integrity writeback, skip the page
2337 if (!PageDirty(page) ||
2338 (PageWriteback(page) &&
2339 (wbc->sync_mode == WB_SYNC_NONE)) ||
2340 unlikely(page->mapping != mapping)) {
2345 wait_on_page_writeback(page);
2346 BUG_ON(PageWriteback(page));
2349 * If we have inline data and arrive here, it means that
2350 * we will soon create the block for the 1st page, so
2351 * we'd better clear the inline data here.
2353 if (ext4_has_inline_data(inode)) {
2354 BUG_ON(ext4_test_inode_state(inode,
2355 EXT4_STATE_MAY_INLINE_DATA));
2356 ext4_destroy_inline_data(handle, inode);
2359 if (mpd->next_page != page->index)
2360 mpd->first_page = page->index;
2361 mpd->next_page = page->index + 1;
2362 logical = (sector_t) page->index <<
2363 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2365 /* Add all dirty buffers to mpd */
2366 head = page_buffers(page);
2369 BUG_ON(buffer_locked(bh));
2371 * We need to try to allocate unmapped blocks
2372 * in the same page. Otherwise we won't make
2373 * progress with the page in ext4_writepage
2375 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2376 mpage_add_bh_to_extent(mpd, logical,
2379 goto ret_extent_tail;
2380 } else if (buffer_dirty(bh) &&
2381 buffer_mapped(bh)) {
2383 * mapped dirty buffer. We need to
2384 * update the b_state because we look
2385 * at b_state in mpage_da_map_blocks.
2386 * We don't update b_size because if we
2387 * find an unmapped buffer_head later
2388 * we need to use the b_state flag of
2391 if (mpd->b_size == 0)
2393 bh->b_state & BH_FLAGS;
2396 } while ((bh = bh->b_this_page) != head);
2398 if (nr_to_write > 0) {
2400 if (nr_to_write == 0 &&
2401 wbc->sync_mode == WB_SYNC_NONE)
2403 * We stop writing back only if we are
2404 * not doing integrity sync. In case of
2405 * integrity sync we have to keep going
2406 * because someone may be concurrently
2407 * dirtying pages, and we might have
2408 * synced a lot of newly appeared dirty
2409 * pages, but have not synced all of the
2415 pagevec_release(&pvec);
2420 ret = MPAGE_DA_EXTENT_TAIL;
2422 pagevec_release(&pvec);
2428 static int ext4_da_writepages(struct address_space *mapping,
2429 struct writeback_control *wbc)
2432 int range_whole = 0;
2433 handle_t *handle = NULL;
2434 struct mpage_da_data mpd;
2435 struct inode *inode = mapping->host;
2436 int pages_written = 0;
2437 unsigned int max_pages;
2438 int range_cyclic, cycled = 1, io_done = 0;
2439 int needed_blocks, ret = 0;
2440 long desired_nr_to_write, nr_to_writebump = 0;
2441 loff_t range_start = wbc->range_start;
2442 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2443 pgoff_t done_index = 0;
2445 struct blk_plug plug;
2447 trace_ext4_da_writepages(inode, wbc);
2450 * No pages to write? This is mainly a kludge to avoid starting
2451 * a transaction for special inodes like journal inode on last iput()
2452 * because that could violate lock ordering on umount
2454 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2458 * If the filesystem has aborted, it is read-only, so return
2459 * right away instead of dumping stack traces later on that
2460 * will obscure the real source of the problem. We test
2461 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2462 * the latter could be true if the filesystem is mounted
2463 * read-only, and in that case, ext4_da_writepages should
2464 * *never* be called, so if that ever happens, we would want
2467 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2470 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2473 range_cyclic = wbc->range_cyclic;
2474 if (wbc->range_cyclic) {
2475 index = mapping->writeback_index;
2478 wbc->range_start = index << PAGE_CACHE_SHIFT;
2479 wbc->range_end = LLONG_MAX;
2480 wbc->range_cyclic = 0;
2483 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2484 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2488 * This works around two forms of stupidity. The first is in
2489 * the writeback code, which caps the maximum number of pages
2490 * written to be 1024 pages. This is wrong on multiple
2491 * levels; different architectues have a different page size,
2492 * which changes the maximum amount of data which gets
2493 * written. Secondly, 4 megabytes is way too small. XFS
2494 * forces this value to be 16 megabytes by multiplying
2495 * nr_to_write parameter by four, and then relies on its
2496 * allocator to allocate larger extents to make them
2497 * contiguous. Unfortunately this brings us to the second
2498 * stupidity, which is that ext4's mballoc code only allocates
2499 * at most 2048 blocks. So we force contiguous writes up to
2500 * the number of dirty blocks in the inode, or
2501 * sbi->max_writeback_mb_bump whichever is smaller.
2503 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2504 if (!range_cyclic && range_whole) {
2505 if (wbc->nr_to_write == LONG_MAX)
2506 desired_nr_to_write = wbc->nr_to_write;
2508 desired_nr_to_write = wbc->nr_to_write * 8;
2510 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2512 if (desired_nr_to_write > max_pages)
2513 desired_nr_to_write = max_pages;
2515 if (wbc->nr_to_write < desired_nr_to_write) {
2516 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2517 wbc->nr_to_write = desired_nr_to_write;
2521 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2522 tag_pages_for_writeback(mapping, index, end);
2524 blk_start_plug(&plug);
2525 while (!ret && wbc->nr_to_write > 0) {
2528 * we insert one extent at a time. So we need
2529 * credit needed for single extent allocation.
2530 * journalled mode is currently not supported
2533 BUG_ON(ext4_should_journal_data(inode));
2534 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2536 /* start a new transaction*/
2537 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2539 if (IS_ERR(handle)) {
2540 ret = PTR_ERR(handle);
2541 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2542 "%ld pages, ino %lu; err %d", __func__,
2543 wbc->nr_to_write, inode->i_ino, ret);
2544 blk_finish_plug(&plug);
2545 goto out_writepages;
2549 * Now call write_cache_pages_da() to find the next
2550 * contiguous region of logical blocks that need
2551 * blocks to be allocated by ext4 and submit them.
2553 ret = write_cache_pages_da(handle, mapping,
2554 wbc, &mpd, &done_index);
2556 * If we have a contiguous extent of pages and we
2557 * haven't done the I/O yet, map the blocks and submit
2560 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2561 mpage_da_map_and_submit(&mpd);
2562 ret = MPAGE_DA_EXTENT_TAIL;
2564 trace_ext4_da_write_pages(inode, &mpd);
2565 wbc->nr_to_write -= mpd.pages_written;
2567 ext4_journal_stop(handle);
2569 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2570 /* commit the transaction which would
2571 * free blocks released in the transaction
2574 jbd2_journal_force_commit_nested(sbi->s_journal);
2576 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2578 * Got one extent now try with rest of the pages.
2579 * If mpd.retval is set -EIO, journal is aborted.
2580 * So we don't need to write any more.
2582 pages_written += mpd.pages_written;
2585 } else if (wbc->nr_to_write)
2587 * There is no more writeout needed
2588 * or we requested for a noblocking writeout
2589 * and we found the device congested
2593 blk_finish_plug(&plug);
2594 if (!io_done && !cycled) {
2597 wbc->range_start = index << PAGE_CACHE_SHIFT;
2598 wbc->range_end = mapping->writeback_index - 1;
2603 wbc->range_cyclic = range_cyclic;
2604 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2606 * set the writeback_index so that range_cyclic
2607 * mode will write it back later
2609 mapping->writeback_index = done_index;
2612 wbc->nr_to_write -= nr_to_writebump;
2613 wbc->range_start = range_start;
2614 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2618 static int ext4_nonda_switch(struct super_block *sb)
2620 s64 free_clusters, dirty_clusters;
2621 struct ext4_sb_info *sbi = EXT4_SB(sb);
2624 * switch to non delalloc mode if we are running low
2625 * on free block. The free block accounting via percpu
2626 * counters can get slightly wrong with percpu_counter_batch getting
2627 * accumulated on each CPU without updating global counters
2628 * Delalloc need an accurate free block accounting. So switch
2629 * to non delalloc when we are near to error range.
2632 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2634 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2636 * Start pushing delalloc when 1/2 of free blocks are dirty.
2638 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2639 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2641 if (2 * free_clusters < 3 * dirty_clusters ||
2642 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2644 * free block count is less than 150% of dirty blocks
2645 * or free blocks is less than watermark
2652 /* We always reserve for an inode update; the superblock could be there too */
2653 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2655 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
2656 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2659 if (pos + len <= 0x7fffffffULL)
2662 /* We might need to update the superblock to set LARGE_FILE */
2666 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2667 loff_t pos, unsigned len, unsigned flags,
2668 struct page **pagep, void **fsdata)
2670 int ret, retries = 0;
2673 struct inode *inode = mapping->host;
2676 index = pos >> PAGE_CACHE_SHIFT;
2678 if (ext4_nonda_switch(inode->i_sb)) {
2679 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2680 return ext4_write_begin(file, mapping, pos,
2681 len, flags, pagep, fsdata);
2683 *fsdata = (void *)0;
2684 trace_ext4_da_write_begin(inode, pos, len, flags);
2686 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2687 ret = ext4_da_write_inline_data_begin(mapping, inode,
2697 * grab_cache_page_write_begin() can take a long time if the
2698 * system is thrashing due to memory pressure, or if the page
2699 * is being written back. So grab it first before we start
2700 * the transaction handle. This also allows us to allocate
2701 * the page (if needed) without using GFP_NOFS.
2704 page = grab_cache_page_write_begin(mapping, index, flags);
2710 * With delayed allocation, we don't log the i_disksize update
2711 * if there is delayed block allocation. But we still need
2712 * to journalling the i_disksize update if writes to the end
2713 * of file which has an already mapped buffer.
2716 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2717 ext4_da_write_credits(inode, pos, len));
2718 if (IS_ERR(handle)) {
2719 page_cache_release(page);
2720 return PTR_ERR(handle);
2724 if (page->mapping != mapping) {
2725 /* The page got truncated from under us */
2727 page_cache_release(page);
2728 ext4_journal_stop(handle);
2731 /* In case writeback began while the page was unlocked */
2732 wait_on_page_writeback(page);
2734 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2737 ext4_journal_stop(handle);
2739 * block_write_begin may have instantiated a few blocks
2740 * outside i_size. Trim these off again. Don't need
2741 * i_size_read because we hold i_mutex.
2743 if (pos + len > inode->i_size)
2744 ext4_truncate_failed_write(inode);
2746 if (ret == -ENOSPC &&
2747 ext4_should_retry_alloc(inode->i_sb, &retries))
2750 page_cache_release(page);
2759 * Check if we should update i_disksize
2760 * when write to the end of file but not require block allocation
2762 static int ext4_da_should_update_i_disksize(struct page *page,
2763 unsigned long offset)
2765 struct buffer_head *bh;
2766 struct inode *inode = page->mapping->host;
2770 bh = page_buffers(page);
2771 idx = offset >> inode->i_blkbits;
2773 for (i = 0; i < idx; i++)
2774 bh = bh->b_this_page;
2776 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2781 static int ext4_da_write_end(struct file *file,
2782 struct address_space *mapping,
2783 loff_t pos, unsigned len, unsigned copied,
2784 struct page *page, void *fsdata)
2786 struct inode *inode = mapping->host;
2788 handle_t *handle = ext4_journal_current_handle();
2790 unsigned long start, end;
2791 int write_mode = (int)(unsigned long)fsdata;
2793 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2794 return ext4_write_end(file, mapping, pos,
2795 len, copied, page, fsdata);
2797 trace_ext4_da_write_end(inode, pos, len, copied);
2798 start = pos & (PAGE_CACHE_SIZE - 1);
2799 end = start + copied - 1;
2802 * generic_write_end() will run mark_inode_dirty() if i_size
2803 * changes. So let's piggyback the i_disksize mark_inode_dirty
2806 new_i_size = pos + copied;
2807 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2808 if (ext4_has_inline_data(inode) ||
2809 ext4_da_should_update_i_disksize(page, end)) {
2810 down_write(&EXT4_I(inode)->i_data_sem);
2811 if (new_i_size > EXT4_I(inode)->i_disksize)
2812 EXT4_I(inode)->i_disksize = new_i_size;
2813 up_write(&EXT4_I(inode)->i_data_sem);
2814 /* We need to mark inode dirty even if
2815 * new_i_size is less that inode->i_size
2816 * bu greater than i_disksize.(hint delalloc)
2818 ext4_mark_inode_dirty(handle, inode);
2822 if (write_mode != CONVERT_INLINE_DATA &&
2823 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2824 ext4_has_inline_data(inode))
2825 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2828 ret2 = generic_write_end(file, mapping, pos, len, copied,
2834 ret2 = ext4_journal_stop(handle);
2838 return ret ? ret : copied;
2841 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2844 * Drop reserved blocks
2846 BUG_ON(!PageLocked(page));
2847 if (!page_has_buffers(page))
2850 ext4_da_page_release_reservation(page, offset);
2853 ext4_invalidatepage(page, offset);
2859 * Force all delayed allocation blocks to be allocated for a given inode.
2861 int ext4_alloc_da_blocks(struct inode *inode)
2863 trace_ext4_alloc_da_blocks(inode);
2865 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2866 !EXT4_I(inode)->i_reserved_meta_blocks)
2870 * We do something simple for now. The filemap_flush() will
2871 * also start triggering a write of the data blocks, which is
2872 * not strictly speaking necessary (and for users of
2873 * laptop_mode, not even desirable). However, to do otherwise
2874 * would require replicating code paths in:
2876 * ext4_da_writepages() ->
2877 * write_cache_pages() ---> (via passed in callback function)
2878 * __mpage_da_writepage() -->
2879 * mpage_add_bh_to_extent()
2880 * mpage_da_map_blocks()
2882 * The problem is that write_cache_pages(), located in
2883 * mm/page-writeback.c, marks pages clean in preparation for
2884 * doing I/O, which is not desirable if we're not planning on
2887 * We could call write_cache_pages(), and then redirty all of
2888 * the pages by calling redirty_page_for_writepage() but that
2889 * would be ugly in the extreme. So instead we would need to
2890 * replicate parts of the code in the above functions,
2891 * simplifying them because we wouldn't actually intend to
2892 * write out the pages, but rather only collect contiguous
2893 * logical block extents, call the multi-block allocator, and
2894 * then update the buffer heads with the block allocations.
2896 * For now, though, we'll cheat by calling filemap_flush(),
2897 * which will map the blocks, and start the I/O, but not
2898 * actually wait for the I/O to complete.
2900 return filemap_flush(inode->i_mapping);
2904 * bmap() is special. It gets used by applications such as lilo and by
2905 * the swapper to find the on-disk block of a specific piece of data.
2907 * Naturally, this is dangerous if the block concerned is still in the
2908 * journal. If somebody makes a swapfile on an ext4 data-journaling
2909 * filesystem and enables swap, then they may get a nasty shock when the
2910 * data getting swapped to that swapfile suddenly gets overwritten by
2911 * the original zero's written out previously to the journal and
2912 * awaiting writeback in the kernel's buffer cache.
2914 * So, if we see any bmap calls here on a modified, data-journaled file,
2915 * take extra steps to flush any blocks which might be in the cache.
2917 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2919 struct inode *inode = mapping->host;
2924 * We can get here for an inline file via the FIBMAP ioctl
2926 if (ext4_has_inline_data(inode))
2929 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2930 test_opt(inode->i_sb, DELALLOC)) {
2932 * With delalloc we want to sync the file
2933 * so that we can make sure we allocate
2936 filemap_write_and_wait(mapping);
2939 if (EXT4_JOURNAL(inode) &&
2940 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2942 * This is a REALLY heavyweight approach, but the use of
2943 * bmap on dirty files is expected to be extremely rare:
2944 * only if we run lilo or swapon on a freshly made file
2945 * do we expect this to happen.
2947 * (bmap requires CAP_SYS_RAWIO so this does not
2948 * represent an unprivileged user DOS attack --- we'd be
2949 * in trouble if mortal users could trigger this path at
2952 * NB. EXT4_STATE_JDATA is not set on files other than
2953 * regular files. If somebody wants to bmap a directory
2954 * or symlink and gets confused because the buffer
2955 * hasn't yet been flushed to disk, they deserve
2956 * everything they get.
2959 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2960 journal = EXT4_JOURNAL(inode);
2961 jbd2_journal_lock_updates(journal);
2962 err = jbd2_journal_flush(journal);
2963 jbd2_journal_unlock_updates(journal);
2969 return generic_block_bmap(mapping, block, ext4_get_block);
2972 static int ext4_readpage(struct file *file, struct page *page)
2975 struct inode *inode = page->mapping->host;
2977 trace_ext4_readpage(page);
2979 if (ext4_has_inline_data(inode))
2980 ret = ext4_readpage_inline(inode, page);
2983 return mpage_readpage(page, ext4_get_block);
2989 ext4_readpages(struct file *file, struct address_space *mapping,
2990 struct list_head *pages, unsigned nr_pages)
2992 struct inode *inode = mapping->host;
2994 /* If the file has inline data, no need to do readpages. */
2995 if (ext4_has_inline_data(inode))
2998 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3001 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3003 trace_ext4_invalidatepage(page, offset);
3005 /* No journalling happens on data buffers when this function is used */
3006 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3008 block_invalidatepage(page, offset);
3011 static int __ext4_journalled_invalidatepage(struct page *page,
3012 unsigned long offset)
3014 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3016 trace_ext4_journalled_invalidatepage(page, offset);
3019 * If it's a full truncate we just forget about the pending dirtying
3022 ClearPageChecked(page);
3024 return jbd2_journal_invalidatepage(journal, page, offset);
3027 /* Wrapper for aops... */
3028 static void ext4_journalled_invalidatepage(struct page *page,
3029 unsigned long offset)
3031 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
3034 static int ext4_releasepage(struct page *page, gfp_t wait)
3036 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3038 trace_ext4_releasepage(page);
3040 /* Page has dirty journalled data -> cannot release */
3041 if (PageChecked(page))
3044 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3046 return try_to_free_buffers(page);
3050 * ext4_get_block used when preparing for a DIO write or buffer write.
3051 * We allocate an uinitialized extent if blocks haven't been allocated.
3052 * The extent will be converted to initialized after the IO is complete.
3054 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3055 struct buffer_head *bh_result, int create)
3057 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3058 inode->i_ino, create);
3059 return _ext4_get_block(inode, iblock, bh_result,
3060 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3063 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3064 struct buffer_head *bh_result, int create)
3066 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3067 inode->i_ino, create);
3068 return _ext4_get_block(inode, iblock, bh_result,
3069 EXT4_GET_BLOCKS_NO_LOCK);
3072 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3073 ssize_t size, void *private, int ret,
3076 struct inode *inode = file_inode(iocb->ki_filp);
3077 ext4_io_end_t *io_end = iocb->private;
3079 /* if not async direct IO or dio with 0 bytes write, just return */
3080 if (!io_end || !size)
3083 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3084 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3085 iocb->private, io_end->inode->i_ino, iocb, offset,
3088 iocb->private = NULL;
3090 /* if not aio dio with unwritten extents, just free io and return */
3091 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3092 ext4_free_io_end(io_end);
3094 inode_dio_done(inode);
3096 aio_complete(iocb, ret, 0);
3100 io_end->offset = offset;
3101 io_end->size = size;
3103 io_end->iocb = iocb;
3104 io_end->result = ret;
3107 ext4_add_complete_io(io_end);
3111 * For ext4 extent files, ext4 will do direct-io write to holes,
3112 * preallocated extents, and those write extend the file, no need to
3113 * fall back to buffered IO.
3115 * For holes, we fallocate those blocks, mark them as uninitialized
3116 * If those blocks were preallocated, we mark sure they are split, but
3117 * still keep the range to write as uninitialized.
3119 * The unwritten extents will be converted to written when DIO is completed.
3120 * For async direct IO, since the IO may still pending when return, we
3121 * set up an end_io call back function, which will do the conversion
3122 * when async direct IO completed.
3124 * If the O_DIRECT write will extend the file then add this inode to the
3125 * orphan list. So recovery will truncate it back to the original size
3126 * if the machine crashes during the write.
3129 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3130 const struct iovec *iov, loff_t offset,
3131 unsigned long nr_segs)
3133 struct file *file = iocb->ki_filp;
3134 struct inode *inode = file->f_mapping->host;
3136 size_t count = iov_length(iov, nr_segs);
3138 get_block_t *get_block_func = NULL;
3140 loff_t final_size = offset + count;
3142 /* Use the old path for reads and writes beyond i_size. */
3143 if (rw != WRITE || final_size > inode->i_size)
3144 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3146 BUG_ON(iocb->private == NULL);
3148 /* If we do a overwrite dio, i_mutex locking can be released */
3149 overwrite = *((int *)iocb->private);
3152 atomic_inc(&inode->i_dio_count);
3153 down_read(&EXT4_I(inode)->i_data_sem);
3154 mutex_unlock(&inode->i_mutex);
3158 * We could direct write to holes and fallocate.
3160 * Allocated blocks to fill the hole are marked as
3161 * uninitialized to prevent parallel buffered read to expose
3162 * the stale data before DIO complete the data IO.
3164 * As to previously fallocated extents, ext4 get_block will
3165 * just simply mark the buffer mapped but still keep the
3166 * extents uninitialized.
3168 * For non AIO case, we will convert those unwritten extents
3169 * to written after return back from blockdev_direct_IO.
3171 * For async DIO, the conversion needs to be deferred when the
3172 * IO is completed. The ext4 end_io callback function will be
3173 * called to take care of the conversion work. Here for async
3174 * case, we allocate an io_end structure to hook to the iocb.
3176 iocb->private = NULL;
3177 ext4_inode_aio_set(inode, NULL);
3178 if (!is_sync_kiocb(iocb)) {
3179 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
3184 io_end->flag |= EXT4_IO_END_DIRECT;
3185 iocb->private = io_end;
3187 * we save the io structure for current async direct
3188 * IO, so that later ext4_map_blocks() could flag the
3189 * io structure whether there is a unwritten extents
3190 * needs to be converted when IO is completed.
3192 ext4_inode_aio_set(inode, io_end);
3196 get_block_func = ext4_get_block_write_nolock;
3198 get_block_func = ext4_get_block_write;
3199 dio_flags = DIO_LOCKING;
3201 ret = __blockdev_direct_IO(rw, iocb, inode,
3202 inode->i_sb->s_bdev, iov,
3210 ext4_inode_aio_set(inode, NULL);
3212 * The io_end structure takes a reference to the inode, that
3213 * structure needs to be destroyed and the reference to the
3214 * inode need to be dropped, when IO is complete, even with 0
3215 * byte write, or failed.
3217 * In the successful AIO DIO case, the io_end structure will
3218 * be destroyed and the reference to the inode will be dropped
3219 * after the end_io call back function is called.
3221 * In the case there is 0 byte write, or error case, since VFS
3222 * direct IO won't invoke the end_io call back function, we
3223 * need to free the end_io structure here.
3225 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3226 ext4_free_io_end(iocb->private);
3227 iocb->private = NULL;
3228 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3229 EXT4_STATE_DIO_UNWRITTEN)) {
3232 * for non AIO case, since the IO is already
3233 * completed, we could do the conversion right here
3235 err = ext4_convert_unwritten_extents(inode,
3239 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3243 /* take i_mutex locking again if we do a ovewrite dio */
3245 inode_dio_done(inode);
3246 up_read(&EXT4_I(inode)->i_data_sem);
3247 mutex_lock(&inode->i_mutex);
3253 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3254 const struct iovec *iov, loff_t offset,
3255 unsigned long nr_segs)
3257 struct file *file = iocb->ki_filp;
3258 struct inode *inode = file->f_mapping->host;
3262 * If we are doing data journalling we don't support O_DIRECT
3264 if (ext4_should_journal_data(inode))
3267 /* Let buffer I/O handle the inline data case. */
3268 if (ext4_has_inline_data(inode))
3271 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3272 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3273 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3275 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3276 trace_ext4_direct_IO_exit(inode, offset,
3277 iov_length(iov, nr_segs), rw, ret);
3282 * Pages can be marked dirty completely asynchronously from ext4's journalling
3283 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3284 * much here because ->set_page_dirty is called under VFS locks. The page is
3285 * not necessarily locked.
3287 * We cannot just dirty the page and leave attached buffers clean, because the
3288 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3289 * or jbddirty because all the journalling code will explode.
3291 * So what we do is to mark the page "pending dirty" and next time writepage
3292 * is called, propagate that into the buffers appropriately.
3294 static int ext4_journalled_set_page_dirty(struct page *page)
3296 SetPageChecked(page);
3297 return __set_page_dirty_nobuffers(page);
3300 static const struct address_space_operations ext4_aops = {
3301 .readpage = ext4_readpage,
3302 .readpages = ext4_readpages,
3303 .writepage = ext4_writepage,
3304 .write_begin = ext4_write_begin,
3305 .write_end = ext4_write_end,
3307 .invalidatepage = ext4_invalidatepage,
3308 .releasepage = ext4_releasepage,
3309 .direct_IO = ext4_direct_IO,
3310 .migratepage = buffer_migrate_page,
3311 .is_partially_uptodate = block_is_partially_uptodate,
3312 .error_remove_page = generic_error_remove_page,
3315 static const struct address_space_operations ext4_journalled_aops = {
3316 .readpage = ext4_readpage,
3317 .readpages = ext4_readpages,
3318 .writepage = ext4_writepage,
3319 .write_begin = ext4_write_begin,
3320 .write_end = ext4_journalled_write_end,
3321 .set_page_dirty = ext4_journalled_set_page_dirty,
3323 .invalidatepage = ext4_journalled_invalidatepage,
3324 .releasepage = ext4_releasepage,
3325 .direct_IO = ext4_direct_IO,
3326 .is_partially_uptodate = block_is_partially_uptodate,
3327 .error_remove_page = generic_error_remove_page,
3330 static const struct address_space_operations ext4_da_aops = {
3331 .readpage = ext4_readpage,
3332 .readpages = ext4_readpages,
3333 .writepage = ext4_writepage,
3334 .writepages = ext4_da_writepages,
3335 .write_begin = ext4_da_write_begin,
3336 .write_end = ext4_da_write_end,
3338 .invalidatepage = ext4_da_invalidatepage,
3339 .releasepage = ext4_releasepage,
3340 .direct_IO = ext4_direct_IO,
3341 .migratepage = buffer_migrate_page,
3342 .is_partially_uptodate = block_is_partially_uptodate,
3343 .error_remove_page = generic_error_remove_page,
3346 void ext4_set_aops(struct inode *inode)
3348 switch (ext4_inode_journal_mode(inode)) {
3349 case EXT4_INODE_ORDERED_DATA_MODE:
3350 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3352 case EXT4_INODE_WRITEBACK_DATA_MODE:
3353 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3355 case EXT4_INODE_JOURNAL_DATA_MODE:
3356 inode->i_mapping->a_ops = &ext4_journalled_aops;
3361 if (test_opt(inode->i_sb, DELALLOC))
3362 inode->i_mapping->a_ops = &ext4_da_aops;
3364 inode->i_mapping->a_ops = &ext4_aops;
3369 * ext4_discard_partial_page_buffers()
3370 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3371 * This function finds and locks the page containing the offset
3372 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3373 * Calling functions that already have the page locked should call
3374 * ext4_discard_partial_page_buffers_no_lock directly.
3376 int ext4_discard_partial_page_buffers(handle_t *handle,
3377 struct address_space *mapping, loff_t from,
3378 loff_t length, int flags)
3380 struct inode *inode = mapping->host;
3384 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3385 mapping_gfp_mask(mapping) & ~__GFP_FS);
3389 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3390 from, length, flags);
3393 page_cache_release(page);
3398 * ext4_discard_partial_page_buffers_no_lock()
3399 * Zeros a page range of length 'length' starting from offset 'from'.
3400 * Buffer heads that correspond to the block aligned regions of the
3401 * zeroed range will be unmapped. Unblock aligned regions
3402 * will have the corresponding buffer head mapped if needed so that
3403 * that region of the page can be updated with the partial zero out.
3405 * This function assumes that the page has already been locked. The
3406 * The range to be discarded must be contained with in the given page.
3407 * If the specified range exceeds the end of the page it will be shortened
3408 * to the end of the page that corresponds to 'from'. This function is
3409 * appropriate for updating a page and it buffer heads to be unmapped and
3410 * zeroed for blocks that have been either released, or are going to be
3413 * handle: The journal handle
3414 * inode: The files inode
3415 * page: A locked page that contains the offset "from"
3416 * from: The starting byte offset (from the beginning of the file)
3417 * to begin discarding
3418 * len: The length of bytes to discard
3419 * flags: Optional flags that may be used:
3421 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3422 * Only zero the regions of the page whose buffer heads
3423 * have already been unmapped. This flag is appropriate
3424 * for updating the contents of a page whose blocks may
3425 * have already been released, and we only want to zero
3426 * out the regions that correspond to those released blocks.
3428 * Returns zero on success or negative on failure.
3430 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3431 struct inode *inode, struct page *page, loff_t from,
3432 loff_t length, int flags)
3434 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3435 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3436 unsigned int blocksize, max, pos;
3438 struct buffer_head *bh;
3441 blocksize = inode->i_sb->s_blocksize;
3442 max = PAGE_CACHE_SIZE - offset;
3444 if (index != page->index)
3448 * correct length if it does not fall between
3449 * 'from' and the end of the page
3451 if (length > max || length < 0)
3454 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3456 if (!page_has_buffers(page))
3457 create_empty_buffers(page, blocksize, 0);
3459 /* Find the buffer that contains "offset" */
3460 bh = page_buffers(page);
3462 while (offset >= pos) {
3463 bh = bh->b_this_page;
3469 while (pos < offset + length) {
3470 unsigned int end_of_block, range_to_discard;
3474 /* The length of space left to zero and unmap */
3475 range_to_discard = offset + length - pos;
3477 /* The length of space until the end of the block */
3478 end_of_block = blocksize - (pos & (blocksize-1));
3481 * Do not unmap or zero past end of block
3482 * for this buffer head
3484 if (range_to_discard > end_of_block)
3485 range_to_discard = end_of_block;
3489 * Skip this buffer head if we are only zeroing unampped
3490 * regions of the page
3492 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3496 /* If the range is block aligned, unmap */
3497 if (range_to_discard == blocksize) {
3498 clear_buffer_dirty(bh);
3500 clear_buffer_mapped(bh);
3501 clear_buffer_req(bh);
3502 clear_buffer_new(bh);
3503 clear_buffer_delay(bh);
3504 clear_buffer_unwritten(bh);
3505 clear_buffer_uptodate(bh);
3506 zero_user(page, pos, range_to_discard);
3507 BUFFER_TRACE(bh, "Buffer discarded");
3512 * If this block is not completely contained in the range
3513 * to be discarded, then it is not going to be released. Because
3514 * we need to keep this block, we need to make sure this part
3515 * of the page is uptodate before we modify it by writeing
3516 * partial zeros on it.
3518 if (!buffer_mapped(bh)) {
3520 * Buffer head must be mapped before we can read
3523 BUFFER_TRACE(bh, "unmapped");
3524 ext4_get_block(inode, iblock, bh, 0);
3525 /* unmapped? It's a hole - nothing to do */
3526 if (!buffer_mapped(bh)) {
3527 BUFFER_TRACE(bh, "still unmapped");
3532 /* Ok, it's mapped. Make sure it's up-to-date */
3533 if (PageUptodate(page))
3534 set_buffer_uptodate(bh);
3536 if (!buffer_uptodate(bh)) {
3538 ll_rw_block(READ, 1, &bh);
3540 /* Uhhuh. Read error. Complain and punt.*/
3541 if (!buffer_uptodate(bh))
3545 if (ext4_should_journal_data(inode)) {
3546 BUFFER_TRACE(bh, "get write access");
3547 err = ext4_journal_get_write_access(handle, bh);
3552 zero_user(page, pos, range_to_discard);
3555 if (ext4_should_journal_data(inode)) {
3556 err = ext4_handle_dirty_metadata(handle, inode, bh);
3558 mark_buffer_dirty(bh);
3560 BUFFER_TRACE(bh, "Partial buffer zeroed");
3562 bh = bh->b_this_page;
3564 pos += range_to_discard;
3570 int ext4_can_truncate(struct inode *inode)
3572 if (S_ISREG(inode->i_mode))
3574 if (S_ISDIR(inode->i_mode))
3576 if (S_ISLNK(inode->i_mode))
3577 return !ext4_inode_is_fast_symlink(inode);
3582 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3583 * associated with the given offset and length
3585 * @inode: File inode
3586 * @offset: The offset where the hole will begin
3587 * @len: The length of the hole
3589 * Returns: 0 on success or negative on failure
3592 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3594 struct inode *inode = file_inode(file);
3595 struct super_block *sb = inode->i_sb;
3596 ext4_lblk_t first_block, stop_block;
3597 struct address_space *mapping = inode->i_mapping;
3598 loff_t first_page, last_page, page_len;
3599 loff_t first_page_offset, last_page_offset;
3601 unsigned int credits;
3604 if (!S_ISREG(inode->i_mode))
3607 if (EXT4_SB(sb)->s_cluster_ratio > 1) {
3608 /* TODO: Add support for bigalloc file systems */
3612 trace_ext4_punch_hole(inode, offset, length);
3615 * Write out all dirty pages to avoid race conditions
3616 * Then release them.
3618 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3619 ret = filemap_write_and_wait_range(mapping, offset,
3620 offset + length - 1);
3625 mutex_lock(&inode->i_mutex);
3626 /* It's not possible punch hole on append only file */
3627 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
3631 if (IS_SWAPFILE(inode)) {
3636 /* No need to punch hole beyond i_size */
3637 if (offset >= inode->i_size)
3641 * If the hole extends beyond i_size, set the hole
3642 * to end after the page that contains i_size
3644 if (offset + length > inode->i_size) {
3645 length = inode->i_size +
3646 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3650 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
3651 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
3653 first_page_offset = first_page << PAGE_CACHE_SHIFT;
3654 last_page_offset = last_page << PAGE_CACHE_SHIFT;
3656 /* Now release the pages */
3657 if (last_page_offset > first_page_offset) {
3658 truncate_pagecache_range(inode, first_page_offset,
3659 last_page_offset - 1);
3662 /* Wait all existing dio workers, newcomers will block on i_mutex */
3663 ext4_inode_block_unlocked_dio(inode);
3664 ret = ext4_flush_unwritten_io(inode);
3667 inode_dio_wait(inode);
3669 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3670 credits = ext4_writepage_trans_blocks(inode);
3672 credits = ext4_blocks_for_truncate(inode);
3673 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3674 if (IS_ERR(handle)) {
3675 ret = PTR_ERR(handle);
3676 ext4_std_error(sb, ret);
3681 * Now we need to zero out the non-page-aligned data in the
3682 * pages at the start and tail of the hole, and unmap the
3683 * buffer heads for the block aligned regions of the page that
3684 * were completely zeroed.
3686 if (first_page > last_page) {
3688 * If the file space being truncated is contained
3689 * within a page just zero out and unmap the middle of
3692 ret = ext4_discard_partial_page_buffers(handle,
3693 mapping, offset, length, 0);
3699 * zero out and unmap the partial page that contains
3700 * the start of the hole
3702 page_len = first_page_offset - offset;
3704 ret = ext4_discard_partial_page_buffers(handle, mapping,
3705 offset, page_len, 0);
3711 * zero out and unmap the partial page that contains
3712 * the end of the hole
3714 page_len = offset + length - last_page_offset;
3716 ret = ext4_discard_partial_page_buffers(handle, mapping,
3717 last_page_offset, page_len, 0);
3724 * If i_size is contained in the last page, we need to
3725 * unmap and zero the partial page after i_size
3727 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
3728 inode->i_size % PAGE_CACHE_SIZE != 0) {
3729 page_len = PAGE_CACHE_SIZE -
3730 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3733 ret = ext4_discard_partial_page_buffers(handle,
3734 mapping, inode->i_size, page_len, 0);
3741 first_block = (offset + sb->s_blocksize - 1) >>
3742 EXT4_BLOCK_SIZE_BITS(sb);
3743 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3745 /* If there are no blocks to remove, return now */
3746 if (first_block >= stop_block)
3749 down_write(&EXT4_I(inode)->i_data_sem);
3750 ext4_discard_preallocations(inode);
3752 ret = ext4_es_remove_extent(inode, first_block,
3753 stop_block - first_block);
3755 up_write(&EXT4_I(inode)->i_data_sem);
3759 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3760 ret = ext4_ext_remove_space(inode, first_block,
3763 ret = ext4_free_hole_blocks(handle, inode, first_block,
3766 ext4_discard_preallocations(inode);
3767 up_write(&EXT4_I(inode)->i_data_sem);
3769 ext4_handle_sync(handle);
3770 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3771 ext4_mark_inode_dirty(handle, inode);
3773 ext4_journal_stop(handle);
3775 ext4_inode_resume_unlocked_dio(inode);
3777 mutex_unlock(&inode->i_mutex);
3784 * We block out ext4_get_block() block instantiations across the entire
3785 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3786 * simultaneously on behalf of the same inode.
3788 * As we work through the truncate and commit bits of it to the journal there
3789 * is one core, guiding principle: the file's tree must always be consistent on
3790 * disk. We must be able to restart the truncate after a crash.
3792 * The file's tree may be transiently inconsistent in memory (although it
3793 * probably isn't), but whenever we close off and commit a journal transaction,
3794 * the contents of (the filesystem + the journal) must be consistent and
3795 * restartable. It's pretty simple, really: bottom up, right to left (although
3796 * left-to-right works OK too).
3798 * Note that at recovery time, journal replay occurs *before* the restart of
3799 * truncate against the orphan inode list.
3801 * The committed inode has the new, desired i_size (which is the same as
3802 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3803 * that this inode's truncate did not complete and it will again call
3804 * ext4_truncate() to have another go. So there will be instantiated blocks
3805 * to the right of the truncation point in a crashed ext4 filesystem. But
3806 * that's fine - as long as they are linked from the inode, the post-crash
3807 * ext4_truncate() run will find them and release them.
3809 void ext4_truncate(struct inode *inode)
3811 struct ext4_inode_info *ei = EXT4_I(inode);
3812 unsigned int credits;
3814 struct address_space *mapping = inode->i_mapping;
3818 * There is a possibility that we're either freeing the inode
3819 * or it completely new indode. In those cases we might not
3820 * have i_mutex locked because it's not necessary.
3822 if (!(inode->i_state & (I_NEW|I_FREEING)))
3823 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3824 trace_ext4_truncate_enter(inode);
3826 if (!ext4_can_truncate(inode))
3829 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3831 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3832 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3834 if (ext4_has_inline_data(inode)) {
3837 ext4_inline_data_truncate(inode, &has_inline);
3843 * finish any pending end_io work so we won't run the risk of
3844 * converting any truncated blocks to initialized later
3846 ext4_flush_unwritten_io(inode);
3848 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3849 credits = ext4_writepage_trans_blocks(inode);
3851 credits = ext4_blocks_for_truncate(inode);
3853 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3854 if (IS_ERR(handle)) {
3855 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3859 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
3860 page_len = PAGE_CACHE_SIZE -
3861 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3863 if (ext4_discard_partial_page_buffers(handle,
3864 mapping, inode->i_size, page_len, 0))
3869 * We add the inode to the orphan list, so that if this
3870 * truncate spans multiple transactions, and we crash, we will
3871 * resume the truncate when the filesystem recovers. It also
3872 * marks the inode dirty, to catch the new size.
3874 * Implication: the file must always be in a sane, consistent
3875 * truncatable state while each transaction commits.
3877 if (ext4_orphan_add(handle, inode))
3880 down_write(&EXT4_I(inode)->i_data_sem);
3882 ext4_discard_preallocations(inode);
3884 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3885 ext4_ext_truncate(handle, inode);
3887 ext4_ind_truncate(handle, inode);
3889 up_write(&ei->i_data_sem);
3892 ext4_handle_sync(handle);
3896 * If this was a simple ftruncate() and the file will remain alive,
3897 * then we need to clear up the orphan record which we created above.
3898 * However, if this was a real unlink then we were called by
3899 * ext4_delete_inode(), and we allow that function to clean up the
3900 * orphan info for us.
3903 ext4_orphan_del(handle, inode);
3905 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3906 ext4_mark_inode_dirty(handle, inode);
3907 ext4_journal_stop(handle);
3909 trace_ext4_truncate_exit(inode);
3913 * ext4_get_inode_loc returns with an extra refcount against the inode's
3914 * underlying buffer_head on success. If 'in_mem' is true, we have all
3915 * data in memory that is needed to recreate the on-disk version of this
3918 static int __ext4_get_inode_loc(struct inode *inode,
3919 struct ext4_iloc *iloc, int in_mem)
3921 struct ext4_group_desc *gdp;
3922 struct buffer_head *bh;
3923 struct super_block *sb = inode->i_sb;
3925 int inodes_per_block, inode_offset;
3928 if (!ext4_valid_inum(sb, inode->i_ino))
3931 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3932 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3937 * Figure out the offset within the block group inode table
3939 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3940 inode_offset = ((inode->i_ino - 1) %
3941 EXT4_INODES_PER_GROUP(sb));
3942 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3943 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3945 bh = sb_getblk(sb, block);
3948 if (!buffer_uptodate(bh)) {
3952 * If the buffer has the write error flag, we have failed
3953 * to write out another inode in the same block. In this
3954 * case, we don't have to read the block because we may
3955 * read the old inode data successfully.
3957 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3958 set_buffer_uptodate(bh);
3960 if (buffer_uptodate(bh)) {
3961 /* someone brought it uptodate while we waited */
3967 * If we have all information of the inode in memory and this
3968 * is the only valid inode in the block, we need not read the
3972 struct buffer_head *bitmap_bh;
3975 start = inode_offset & ~(inodes_per_block - 1);
3977 /* Is the inode bitmap in cache? */
3978 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3979 if (unlikely(!bitmap_bh))
3983 * If the inode bitmap isn't in cache then the
3984 * optimisation may end up performing two reads instead
3985 * of one, so skip it.
3987 if (!buffer_uptodate(bitmap_bh)) {
3991 for (i = start; i < start + inodes_per_block; i++) {
3992 if (i == inode_offset)
3994 if (ext4_test_bit(i, bitmap_bh->b_data))
3998 if (i == start + inodes_per_block) {
3999 /* all other inodes are free, so skip I/O */
4000 memset(bh->b_data, 0, bh->b_size);
4001 set_buffer_uptodate(bh);
4009 * If we need to do any I/O, try to pre-readahead extra
4010 * blocks from the inode table.
4012 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4013 ext4_fsblk_t b, end, table;
4015 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4017 table = ext4_inode_table(sb, gdp);
4018 /* s_inode_readahead_blks is always a power of 2 */
4019 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4023 num = EXT4_INODES_PER_GROUP(sb);
4024 if (ext4_has_group_desc_csum(sb))
4025 num -= ext4_itable_unused_count(sb, gdp);
4026 table += num / inodes_per_block;
4030 sb_breadahead(sb, b++);
4034 * There are other valid inodes in the buffer, this inode
4035 * has in-inode xattrs, or we don't have this inode in memory.
4036 * Read the block from disk.
4038 trace_ext4_load_inode(inode);
4040 bh->b_end_io = end_buffer_read_sync;
4041 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4043 if (!buffer_uptodate(bh)) {
4044 EXT4_ERROR_INODE_BLOCK(inode, block,
4045 "unable to read itable block");
4055 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4057 /* We have all inode data except xattrs in memory here. */
4058 return __ext4_get_inode_loc(inode, iloc,
4059 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4062 void ext4_set_inode_flags(struct inode *inode)
4064 unsigned int flags = EXT4_I(inode)->i_flags;
4065 unsigned int new_fl = 0;
4067 if (flags & EXT4_SYNC_FL)
4069 if (flags & EXT4_APPEND_FL)
4071 if (flags & EXT4_IMMUTABLE_FL)
4072 new_fl |= S_IMMUTABLE;
4073 if (flags & EXT4_NOATIME_FL)
4074 new_fl |= S_NOATIME;
4075 if (flags & EXT4_DIRSYNC_FL)
4076 new_fl |= S_DIRSYNC;
4077 set_mask_bits(&inode->i_flags,
4078 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC, new_fl);
4081 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4082 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4084 unsigned int vfs_fl;
4085 unsigned long old_fl, new_fl;
4088 vfs_fl = ei->vfs_inode.i_flags;
4089 old_fl = ei->i_flags;
4090 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4091 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4093 if (vfs_fl & S_SYNC)
4094 new_fl |= EXT4_SYNC_FL;
4095 if (vfs_fl & S_APPEND)
4096 new_fl |= EXT4_APPEND_FL;
4097 if (vfs_fl & S_IMMUTABLE)
4098 new_fl |= EXT4_IMMUTABLE_FL;
4099 if (vfs_fl & S_NOATIME)
4100 new_fl |= EXT4_NOATIME_FL;
4101 if (vfs_fl & S_DIRSYNC)
4102 new_fl |= EXT4_DIRSYNC_FL;
4103 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4106 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4107 struct ext4_inode_info *ei)
4110 struct inode *inode = &(ei->vfs_inode);
4111 struct super_block *sb = inode->i_sb;
4113 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4114 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4115 /* we are using combined 48 bit field */
4116 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4117 le32_to_cpu(raw_inode->i_blocks_lo);
4118 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4119 /* i_blocks represent file system block size */
4120 return i_blocks << (inode->i_blkbits - 9);
4125 return le32_to_cpu(raw_inode->i_blocks_lo);
4129 static inline void ext4_iget_extra_inode(struct inode *inode,
4130 struct ext4_inode *raw_inode,
4131 struct ext4_inode_info *ei)
4133 __le32 *magic = (void *)raw_inode +
4134 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4135 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4136 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4137 ext4_find_inline_data_nolock(inode);
4139 EXT4_I(inode)->i_inline_off = 0;
4142 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4144 struct ext4_iloc iloc;
4145 struct ext4_inode *raw_inode;
4146 struct ext4_inode_info *ei;
4147 struct inode *inode;
4148 journal_t *journal = EXT4_SB(sb)->s_journal;
4154 inode = iget_locked(sb, ino);
4156 return ERR_PTR(-ENOMEM);
4157 if (!(inode->i_state & I_NEW))
4163 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4166 raw_inode = ext4_raw_inode(&iloc);
4168 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4169 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4170 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4171 EXT4_INODE_SIZE(inode->i_sb)) {
4172 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4173 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4174 EXT4_INODE_SIZE(inode->i_sb));
4179 ei->i_extra_isize = 0;
4181 /* Precompute checksum seed for inode metadata */
4182 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4183 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
4184 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4186 __le32 inum = cpu_to_le32(inode->i_ino);
4187 __le32 gen = raw_inode->i_generation;
4188 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4190 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4194 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4195 EXT4_ERROR_INODE(inode, "checksum invalid");
4200 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4201 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4202 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4203 if (!(test_opt(inode->i_sb, NO_UID32))) {
4204 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4205 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4207 i_uid_write(inode, i_uid);
4208 i_gid_write(inode, i_gid);
4209 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4211 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4212 ei->i_inline_off = 0;
4213 ei->i_dir_start_lookup = 0;
4214 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4215 /* We now have enough fields to check if the inode was active or not.
4216 * This is needed because nfsd might try to access dead inodes
4217 * the test is that same one that e2fsck uses
4218 * NeilBrown 1999oct15
4220 if (inode->i_nlink == 0) {
4221 if ((inode->i_mode == 0 ||
4222 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4223 ino != EXT4_BOOT_LOADER_INO) {
4224 /* this inode is deleted */
4228 /* The only unlinked inodes we let through here have
4229 * valid i_mode and are being read by the orphan
4230 * recovery code: that's fine, we're about to complete
4231 * the process of deleting those.
4232 * OR it is the EXT4_BOOT_LOADER_INO which is
4233 * not initialized on a new filesystem. */
4235 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4236 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4237 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4238 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4240 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4241 inode->i_size = ext4_isize(raw_inode);
4242 ei->i_disksize = inode->i_size;
4244 ei->i_reserved_quota = 0;
4246 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4247 ei->i_block_group = iloc.block_group;
4248 ei->i_last_alloc_group = ~0;
4250 * NOTE! The in-memory inode i_data array is in little-endian order
4251 * even on big-endian machines: we do NOT byteswap the block numbers!
4253 for (block = 0; block < EXT4_N_BLOCKS; block++)
4254 ei->i_data[block] = raw_inode->i_block[block];
4255 INIT_LIST_HEAD(&ei->i_orphan);
4258 * Set transaction id's of transactions that have to be committed
4259 * to finish f[data]sync. We set them to currently running transaction
4260 * as we cannot be sure that the inode or some of its metadata isn't
4261 * part of the transaction - the inode could have been reclaimed and
4262 * now it is reread from disk.
4265 transaction_t *transaction;
4268 read_lock(&journal->j_state_lock);
4269 if (journal->j_running_transaction)
4270 transaction = journal->j_running_transaction;
4272 transaction = journal->j_committing_transaction;
4274 tid = transaction->t_tid;
4276 tid = journal->j_commit_sequence;
4277 read_unlock(&journal->j_state_lock);
4278 ei->i_sync_tid = tid;
4279 ei->i_datasync_tid = tid;
4282 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4283 if (ei->i_extra_isize == 0) {
4284 /* The extra space is currently unused. Use it. */
4285 ei->i_extra_isize = sizeof(struct ext4_inode) -
4286 EXT4_GOOD_OLD_INODE_SIZE;
4288 ext4_iget_extra_inode(inode, raw_inode, ei);
4292 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4293 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4294 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4295 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4297 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4298 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4299 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4301 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4305 if (ei->i_file_acl &&
4306 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4307 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4311 } else if (!ext4_has_inline_data(inode)) {
4312 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4313 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4314 (S_ISLNK(inode->i_mode) &&
4315 !ext4_inode_is_fast_symlink(inode))))
4316 /* Validate extent which is part of inode */
4317 ret = ext4_ext_check_inode(inode);
4318 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4319 (S_ISLNK(inode->i_mode) &&
4320 !ext4_inode_is_fast_symlink(inode))) {
4321 /* Validate block references which are part of inode */
4322 ret = ext4_ind_check_inode(inode);
4328 if (S_ISREG(inode->i_mode)) {
4329 inode->i_op = &ext4_file_inode_operations;
4330 inode->i_fop = &ext4_file_operations;
4331 ext4_set_aops(inode);
4332 } else if (S_ISDIR(inode->i_mode)) {
4333 inode->i_op = &ext4_dir_inode_operations;
4334 inode->i_fop = &ext4_dir_operations;
4335 } else if (S_ISLNK(inode->i_mode)) {
4336 if (ext4_inode_is_fast_symlink(inode)) {
4337 inode->i_op = &ext4_fast_symlink_inode_operations;
4338 nd_terminate_link(ei->i_data, inode->i_size,
4339 sizeof(ei->i_data) - 1);
4341 inode->i_op = &ext4_symlink_inode_operations;
4342 ext4_set_aops(inode);
4344 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4345 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4346 inode->i_op = &ext4_special_inode_operations;
4347 if (raw_inode->i_block[0])
4348 init_special_inode(inode, inode->i_mode,
4349 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4351 init_special_inode(inode, inode->i_mode,
4352 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4353 } else if (ino == EXT4_BOOT_LOADER_INO) {
4354 make_bad_inode(inode);
4357 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4361 ext4_set_inode_flags(inode);
4362 unlock_new_inode(inode);
4368 return ERR_PTR(ret);
4371 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4373 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4374 return ERR_PTR(-EIO);
4375 return ext4_iget(sb, ino);
4378 static int ext4_inode_blocks_set(handle_t *handle,
4379 struct ext4_inode *raw_inode,
4380 struct ext4_inode_info *ei)
4382 struct inode *inode = &(ei->vfs_inode);
4383 u64 i_blocks = inode->i_blocks;
4384 struct super_block *sb = inode->i_sb;
4386 if (i_blocks <= ~0U) {
4388 * i_blocks can be represented in a 32 bit variable
4389 * as multiple of 512 bytes
4391 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4392 raw_inode->i_blocks_high = 0;
4393 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4396 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4399 if (i_blocks <= 0xffffffffffffULL) {
4401 * i_blocks can be represented in a 48 bit variable
4402 * as multiple of 512 bytes
4404 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4405 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4406 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4408 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4409 /* i_block is stored in file system block size */
4410 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4411 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4412 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4418 * Post the struct inode info into an on-disk inode location in the
4419 * buffer-cache. This gobbles the caller's reference to the
4420 * buffer_head in the inode location struct.
4422 * The caller must have write access to iloc->bh.
4424 static int ext4_do_update_inode(handle_t *handle,
4425 struct inode *inode,
4426 struct ext4_iloc *iloc)
4428 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4429 struct ext4_inode_info *ei = EXT4_I(inode);
4430 struct buffer_head *bh = iloc->bh;
4431 int err = 0, rc, block;
4432 int need_datasync = 0;
4436 /* For fields not not tracking in the in-memory inode,
4437 * initialise them to zero for new inodes. */
4438 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4439 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4441 ext4_get_inode_flags(ei);
4442 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4443 i_uid = i_uid_read(inode);
4444 i_gid = i_gid_read(inode);
4445 if (!(test_opt(inode->i_sb, NO_UID32))) {
4446 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4447 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4449 * Fix up interoperability with old kernels. Otherwise, old inodes get
4450 * re-used with the upper 16 bits of the uid/gid intact
4453 raw_inode->i_uid_high =
4454 cpu_to_le16(high_16_bits(i_uid));
4455 raw_inode->i_gid_high =
4456 cpu_to_le16(high_16_bits(i_gid));
4458 raw_inode->i_uid_high = 0;
4459 raw_inode->i_gid_high = 0;
4462 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4463 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4464 raw_inode->i_uid_high = 0;
4465 raw_inode->i_gid_high = 0;
4467 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4469 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4470 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4471 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4472 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4474 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4476 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4477 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4478 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4479 cpu_to_le32(EXT4_OS_HURD))
4480 raw_inode->i_file_acl_high =
4481 cpu_to_le16(ei->i_file_acl >> 32);
4482 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4483 if (ei->i_disksize != ext4_isize(raw_inode)) {
4484 ext4_isize_set(raw_inode, ei->i_disksize);
4487 if (ei->i_disksize > 0x7fffffffULL) {
4488 struct super_block *sb = inode->i_sb;
4489 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4490 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4491 EXT4_SB(sb)->s_es->s_rev_level ==
4492 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4493 /* If this is the first large file
4494 * created, add a flag to the superblock.
4496 err = ext4_journal_get_write_access(handle,
4497 EXT4_SB(sb)->s_sbh);
4500 ext4_update_dynamic_rev(sb);
4501 EXT4_SET_RO_COMPAT_FEATURE(sb,
4502 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4503 ext4_handle_sync(handle);
4504 err = ext4_handle_dirty_super(handle, sb);
4507 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4508 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4509 if (old_valid_dev(inode->i_rdev)) {
4510 raw_inode->i_block[0] =
4511 cpu_to_le32(old_encode_dev(inode->i_rdev));
4512 raw_inode->i_block[1] = 0;
4514 raw_inode->i_block[0] = 0;
4515 raw_inode->i_block[1] =
4516 cpu_to_le32(new_encode_dev(inode->i_rdev));
4517 raw_inode->i_block[2] = 0;
4519 } else if (!ext4_has_inline_data(inode)) {
4520 for (block = 0; block < EXT4_N_BLOCKS; block++)
4521 raw_inode->i_block[block] = ei->i_data[block];
4524 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4525 if (ei->i_extra_isize) {
4526 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4527 raw_inode->i_version_hi =
4528 cpu_to_le32(inode->i_version >> 32);
4529 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4532 ext4_inode_csum_set(inode, raw_inode, ei);
4534 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4535 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4538 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4540 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4543 ext4_std_error(inode->i_sb, err);
4548 * ext4_write_inode()
4550 * We are called from a few places:
4552 * - Within generic_file_write() for O_SYNC files.
4553 * Here, there will be no transaction running. We wait for any running
4554 * transaction to commit.
4556 * - Within sys_sync(), kupdate and such.
4557 * We wait on commit, if tol to.
4559 * - Within prune_icache() (PF_MEMALLOC == true)
4560 * Here we simply return. We can't afford to block kswapd on the
4563 * In all cases it is actually safe for us to return without doing anything,
4564 * because the inode has been copied into a raw inode buffer in
4565 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4568 * Note that we are absolutely dependent upon all inode dirtiers doing the
4569 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4570 * which we are interested.
4572 * It would be a bug for them to not do this. The code:
4574 * mark_inode_dirty(inode)
4576 * inode->i_size = expr;
4578 * is in error because a kswapd-driven write_inode() could occur while
4579 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4580 * will no longer be on the superblock's dirty inode list.
4582 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4586 if (current->flags & PF_MEMALLOC)
4589 if (EXT4_SB(inode->i_sb)->s_journal) {
4590 if (ext4_journal_current_handle()) {
4591 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4596 if (wbc->sync_mode != WB_SYNC_ALL)
4599 err = ext4_force_commit(inode->i_sb);
4601 struct ext4_iloc iloc;
4603 err = __ext4_get_inode_loc(inode, &iloc, 0);
4606 if (wbc->sync_mode == WB_SYNC_ALL)
4607 sync_dirty_buffer(iloc.bh);
4608 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4609 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4610 "IO error syncing inode");
4619 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4620 * buffers that are attached to a page stradding i_size and are undergoing
4621 * commit. In that case we have to wait for commit to finish and try again.
4623 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4627 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4628 tid_t commit_tid = 0;
4631 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4633 * All buffers in the last page remain valid? Then there's nothing to
4634 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4637 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4640 page = find_lock_page(inode->i_mapping,
4641 inode->i_size >> PAGE_CACHE_SHIFT);
4644 ret = __ext4_journalled_invalidatepage(page, offset);
4646 page_cache_release(page);
4650 read_lock(&journal->j_state_lock);
4651 if (journal->j_committing_transaction)
4652 commit_tid = journal->j_committing_transaction->t_tid;
4653 read_unlock(&journal->j_state_lock);
4655 jbd2_log_wait_commit(journal, commit_tid);
4662 * Called from notify_change.
4664 * We want to trap VFS attempts to truncate the file as soon as
4665 * possible. In particular, we want to make sure that when the VFS
4666 * shrinks i_size, we put the inode on the orphan list and modify
4667 * i_disksize immediately, so that during the subsequent flushing of
4668 * dirty pages and freeing of disk blocks, we can guarantee that any
4669 * commit will leave the blocks being flushed in an unused state on
4670 * disk. (On recovery, the inode will get truncated and the blocks will
4671 * be freed, so we have a strong guarantee that no future commit will
4672 * leave these blocks visible to the user.)
4674 * Another thing we have to assure is that if we are in ordered mode
4675 * and inode is still attached to the committing transaction, we must
4676 * we start writeout of all the dirty pages which are being truncated.
4677 * This way we are sure that all the data written in the previous
4678 * transaction are already on disk (truncate waits for pages under
4681 * Called with inode->i_mutex down.
4683 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4685 struct inode *inode = dentry->d_inode;
4688 const unsigned int ia_valid = attr->ia_valid;
4690 error = inode_change_ok(inode, attr);
4694 if (is_quota_modification(inode, attr))
4695 dquot_initialize(inode);
4696 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4697 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4700 /* (user+group)*(old+new) structure, inode write (sb,
4701 * inode block, ? - but truncate inode update has it) */
4702 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4703 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4704 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4705 if (IS_ERR(handle)) {
4706 error = PTR_ERR(handle);
4709 error = dquot_transfer(inode, attr);
4711 ext4_journal_stop(handle);
4714 /* Update corresponding info in inode so that everything is in
4715 * one transaction */
4716 if (attr->ia_valid & ATTR_UID)
4717 inode->i_uid = attr->ia_uid;
4718 if (attr->ia_valid & ATTR_GID)
4719 inode->i_gid = attr->ia_gid;
4720 error = ext4_mark_inode_dirty(handle, inode);
4721 ext4_journal_stop(handle);
4724 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4726 loff_t oldsize = inode->i_size;
4728 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4729 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4731 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4735 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4736 inode_inc_iversion(inode);
4738 if (S_ISREG(inode->i_mode) &&
4739 (attr->ia_size < inode->i_size)) {
4740 if (ext4_should_order_data(inode)) {
4741 error = ext4_begin_ordered_truncate(inode,
4746 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4747 if (IS_ERR(handle)) {
4748 error = PTR_ERR(handle);
4751 if (ext4_handle_valid(handle)) {
4752 error = ext4_orphan_add(handle, inode);
4755 EXT4_I(inode)->i_disksize = attr->ia_size;
4756 rc = ext4_mark_inode_dirty(handle, inode);
4759 ext4_journal_stop(handle);
4761 ext4_orphan_del(NULL, inode);
4766 i_size_write(inode, attr->ia_size);
4768 * Blocks are going to be removed from the inode. Wait
4769 * for dio in flight. Temporarily disable
4770 * dioread_nolock to prevent livelock.
4773 if (!ext4_should_journal_data(inode)) {
4774 ext4_inode_block_unlocked_dio(inode);
4775 inode_dio_wait(inode);
4776 ext4_inode_resume_unlocked_dio(inode);
4778 ext4_wait_for_tail_page_commit(inode);
4781 * Truncate pagecache after we've waited for commit
4782 * in data=journal mode to make pages freeable.
4784 truncate_pagecache(inode, oldsize, inode->i_size);
4787 * We want to call ext4_truncate() even if attr->ia_size ==
4788 * inode->i_size for cases like truncation of fallocated space
4790 if (attr->ia_valid & ATTR_SIZE)
4791 ext4_truncate(inode);
4794 setattr_copy(inode, attr);
4795 mark_inode_dirty(inode);
4799 * If the call to ext4_truncate failed to get a transaction handle at
4800 * all, we need to clean up the in-core orphan list manually.
4802 if (orphan && inode->i_nlink)
4803 ext4_orphan_del(NULL, inode);
4805 if (!rc && (ia_valid & ATTR_MODE))
4806 rc = ext4_acl_chmod(inode);
4809 ext4_std_error(inode->i_sb, error);
4815 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4818 struct inode *inode;
4819 unsigned long long delalloc_blocks;
4821 inode = dentry->d_inode;
4822 generic_fillattr(inode, stat);
4825 * We can't update i_blocks if the block allocation is delayed
4826 * otherwise in the case of system crash before the real block
4827 * allocation is done, we will have i_blocks inconsistent with
4828 * on-disk file blocks.
4829 * We always keep i_blocks updated together with real
4830 * allocation. But to not confuse with user, stat
4831 * will return the blocks that include the delayed allocation
4832 * blocks for this file.
4834 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4835 EXT4_I(inode)->i_reserved_data_blocks);
4837 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits-9);
4841 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4843 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4844 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4845 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4849 * Account for index blocks, block groups bitmaps and block group
4850 * descriptor blocks if modify datablocks and index blocks
4851 * worse case, the indexs blocks spread over different block groups
4853 * If datablocks are discontiguous, they are possible to spread over
4854 * different block groups too. If they are contiguous, with flexbg,
4855 * they could still across block group boundary.
4857 * Also account for superblock, inode, quota and xattr blocks
4859 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4861 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4867 * How many index blocks need to touch to modify nrblocks?
4868 * The "Chunk" flag indicating whether the nrblocks is
4869 * physically contiguous on disk
4871 * For Direct IO and fallocate, they calls get_block to allocate
4872 * one single extent at a time, so they could set the "Chunk" flag
4874 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4879 * Now let's see how many group bitmaps and group descriptors need
4889 if (groups > ngroups)
4891 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4892 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4894 /* bitmaps and block group descriptor blocks */
4895 ret += groups + gdpblocks;
4897 /* Blocks for super block, inode, quota and xattr blocks */
4898 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4904 * Calculate the total number of credits to reserve to fit
4905 * the modification of a single pages into a single transaction,
4906 * which may include multiple chunks of block allocations.
4908 * This could be called via ext4_write_begin()
4910 * We need to consider the worse case, when
4911 * one new block per extent.
4913 int ext4_writepage_trans_blocks(struct inode *inode)
4915 int bpp = ext4_journal_blocks_per_page(inode);
4918 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4920 /* Account for data blocks for journalled mode */
4921 if (ext4_should_journal_data(inode))
4927 * Calculate the journal credits for a chunk of data modification.
4929 * This is called from DIO, fallocate or whoever calling
4930 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4932 * journal buffers for data blocks are not included here, as DIO
4933 * and fallocate do no need to journal data buffers.
4935 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4937 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4941 * The caller must have previously called ext4_reserve_inode_write().
4942 * Give this, we know that the caller already has write access to iloc->bh.
4944 int ext4_mark_iloc_dirty(handle_t *handle,
4945 struct inode *inode, struct ext4_iloc *iloc)
4949 if (IS_I_VERSION(inode))
4950 inode_inc_iversion(inode);
4952 /* the do_update_inode consumes one bh->b_count */
4955 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4956 err = ext4_do_update_inode(handle, inode, iloc);
4962 * On success, We end up with an outstanding reference count against
4963 * iloc->bh. This _must_ be cleaned up later.
4967 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4968 struct ext4_iloc *iloc)
4972 err = ext4_get_inode_loc(inode, iloc);
4974 BUFFER_TRACE(iloc->bh, "get_write_access");
4975 err = ext4_journal_get_write_access(handle, iloc->bh);
4981 ext4_std_error(inode->i_sb, err);
4986 * Expand an inode by new_extra_isize bytes.
4987 * Returns 0 on success or negative error number on failure.
4989 static int ext4_expand_extra_isize(struct inode *inode,
4990 unsigned int new_extra_isize,
4991 struct ext4_iloc iloc,
4994 struct ext4_inode *raw_inode;
4995 struct ext4_xattr_ibody_header *header;
4997 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5000 raw_inode = ext4_raw_inode(&iloc);
5002 header = IHDR(inode, raw_inode);
5004 /* No extended attributes present */
5005 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5006 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5007 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5009 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5013 /* try to expand with EAs present */
5014 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5019 * What we do here is to mark the in-core inode as clean with respect to inode
5020 * dirtiness (it may still be data-dirty).
5021 * This means that the in-core inode may be reaped by prune_icache
5022 * without having to perform any I/O. This is a very good thing,
5023 * because *any* task may call prune_icache - even ones which
5024 * have a transaction open against a different journal.
5026 * Is this cheating? Not really. Sure, we haven't written the
5027 * inode out, but prune_icache isn't a user-visible syncing function.
5028 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5029 * we start and wait on commits.
5031 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5033 struct ext4_iloc iloc;
5034 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5035 static unsigned int mnt_count;
5039 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5040 err = ext4_reserve_inode_write(handle, inode, &iloc);
5041 if (ext4_handle_valid(handle) &&
5042 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5043 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5045 * We need extra buffer credits since we may write into EA block
5046 * with this same handle. If journal_extend fails, then it will
5047 * only result in a minor loss of functionality for that inode.
5048 * If this is felt to be critical, then e2fsck should be run to
5049 * force a large enough s_min_extra_isize.
5051 if ((jbd2_journal_extend(handle,
5052 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5053 ret = ext4_expand_extra_isize(inode,
5054 sbi->s_want_extra_isize,
5057 ext4_set_inode_state(inode,
5058 EXT4_STATE_NO_EXPAND);
5060 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5061 ext4_warning(inode->i_sb,
5062 "Unable to expand inode %lu. Delete"
5063 " some EAs or run e2fsck.",
5066 le16_to_cpu(sbi->s_es->s_mnt_count);
5072 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5077 * ext4_dirty_inode() is called from __mark_inode_dirty()
5079 * We're really interested in the case where a file is being extended.
5080 * i_size has been changed by generic_commit_write() and we thus need
5081 * to include the updated inode in the current transaction.
5083 * Also, dquot_alloc_block() will always dirty the inode when blocks
5084 * are allocated to the file.
5086 * If the inode is marked synchronous, we don't honour that here - doing
5087 * so would cause a commit on atime updates, which we don't bother doing.
5088 * We handle synchronous inodes at the highest possible level.
5090 void ext4_dirty_inode(struct inode *inode, int flags)
5094 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5098 ext4_mark_inode_dirty(handle, inode);
5100 ext4_journal_stop(handle);
5107 * Bind an inode's backing buffer_head into this transaction, to prevent
5108 * it from being flushed to disk early. Unlike
5109 * ext4_reserve_inode_write, this leaves behind no bh reference and
5110 * returns no iloc structure, so the caller needs to repeat the iloc
5111 * lookup to mark the inode dirty later.
5113 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5115 struct ext4_iloc iloc;
5119 err = ext4_get_inode_loc(inode, &iloc);
5121 BUFFER_TRACE(iloc.bh, "get_write_access");
5122 err = jbd2_journal_get_write_access(handle, iloc.bh);
5124 err = ext4_handle_dirty_metadata(handle,
5130 ext4_std_error(inode->i_sb, err);
5135 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5142 * We have to be very careful here: changing a data block's
5143 * journaling status dynamically is dangerous. If we write a
5144 * data block to the journal, change the status and then delete
5145 * that block, we risk forgetting to revoke the old log record
5146 * from the journal and so a subsequent replay can corrupt data.
5147 * So, first we make sure that the journal is empty and that
5148 * nobody is changing anything.
5151 journal = EXT4_JOURNAL(inode);
5154 if (is_journal_aborted(journal))
5156 /* We have to allocate physical blocks for delalloc blocks
5157 * before flushing journal. otherwise delalloc blocks can not
5158 * be allocated any more. even more truncate on delalloc blocks
5159 * could trigger BUG by flushing delalloc blocks in journal.
5160 * There is no delalloc block in non-journal data mode.
5162 if (val && test_opt(inode->i_sb, DELALLOC)) {
5163 err = ext4_alloc_da_blocks(inode);
5168 /* Wait for all existing dio workers */
5169 ext4_inode_block_unlocked_dio(inode);
5170 inode_dio_wait(inode);
5172 jbd2_journal_lock_updates(journal);
5175 * OK, there are no updates running now, and all cached data is
5176 * synced to disk. We are now in a completely consistent state
5177 * which doesn't have anything in the journal, and we know that
5178 * no filesystem updates are running, so it is safe to modify
5179 * the inode's in-core data-journaling state flag now.
5183 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5185 jbd2_journal_flush(journal);
5186 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5188 ext4_set_aops(inode);
5190 jbd2_journal_unlock_updates(journal);
5191 ext4_inode_resume_unlocked_dio(inode);
5193 /* Finally we can mark the inode as dirty. */
5195 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5197 return PTR_ERR(handle);
5199 err = ext4_mark_inode_dirty(handle, inode);
5200 ext4_handle_sync(handle);
5201 ext4_journal_stop(handle);
5202 ext4_std_error(inode->i_sb, err);
5207 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5209 return !buffer_mapped(bh);
5212 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5214 struct page *page = vmf->page;
5218 struct file *file = vma->vm_file;
5219 struct inode *inode = file_inode(file);
5220 struct address_space *mapping = inode->i_mapping;
5222 get_block_t *get_block;
5225 sb_start_pagefault(inode->i_sb);
5226 file_update_time(vma->vm_file);
5227 /* Delalloc case is easy... */
5228 if (test_opt(inode->i_sb, DELALLOC) &&
5229 !ext4_should_journal_data(inode) &&
5230 !ext4_nonda_switch(inode->i_sb)) {
5232 ret = __block_page_mkwrite(vma, vmf,
5233 ext4_da_get_block_prep);
5234 } while (ret == -ENOSPC &&
5235 ext4_should_retry_alloc(inode->i_sb, &retries));
5240 size = i_size_read(inode);
5241 /* Page got truncated from under us? */
5242 if (page->mapping != mapping || page_offset(page) > size) {
5244 ret = VM_FAULT_NOPAGE;
5248 if (page->index == size >> PAGE_CACHE_SHIFT)
5249 len = size & ~PAGE_CACHE_MASK;
5251 len = PAGE_CACHE_SIZE;
5253 * Return if we have all the buffers mapped. This avoids the need to do
5254 * journal_start/journal_stop which can block and take a long time
5256 if (page_has_buffers(page)) {
5257 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5259 ext4_bh_unmapped)) {
5260 /* Wait so that we don't change page under IO */
5261 wait_for_stable_page(page);
5262 ret = VM_FAULT_LOCKED;
5267 /* OK, we need to fill the hole... */
5268 if (ext4_should_dioread_nolock(inode))
5269 get_block = ext4_get_block_write;
5271 get_block = ext4_get_block;
5273 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5274 ext4_writepage_trans_blocks(inode));
5275 if (IS_ERR(handle)) {
5276 ret = VM_FAULT_SIGBUS;
5279 ret = __block_page_mkwrite(vma, vmf, get_block);
5280 if (!ret && ext4_should_journal_data(inode)) {
5281 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5282 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5284 ret = VM_FAULT_SIGBUS;
5285 ext4_journal_stop(handle);
5288 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5290 ext4_journal_stop(handle);
5291 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5294 ret = block_page_mkwrite_return(ret);
5296 sb_end_pagefault(inode->i_sb);