1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/inode.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
23 #include <linux/time.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/dax.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/bitops.h>
41 #include <linux/iomap.h>
42 #include <linux/iversion.h>
44 #include "ext4_jbd2.h"
49 #include <trace/events/ext4.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
54 struct ext4_inode_info *ei)
56 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
59 int offset = offsetof(struct ext4_inode, i_checksum_lo);
60 unsigned int csum_size = sizeof(dummy_csum);
62 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
63 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
65 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
66 EXT4_GOOD_OLD_INODE_SIZE - offset);
68 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
69 offset = offsetof(struct ext4_inode, i_checksum_hi);
70 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
71 EXT4_GOOD_OLD_INODE_SIZE,
72 offset - EXT4_GOOD_OLD_INODE_SIZE);
73 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
74 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
78 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
79 EXT4_INODE_SIZE(inode->i_sb) - offset);
85 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
86 struct ext4_inode_info *ei)
88 __u32 provided, calculated;
90 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
91 cpu_to_le32(EXT4_OS_LINUX) ||
92 !ext4_has_metadata_csum(inode->i_sb))
95 provided = le16_to_cpu(raw->i_checksum_lo);
96 calculated = ext4_inode_csum(inode, raw, ei);
97 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
98 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
99 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
101 calculated &= 0xFFFF;
103 return provided == calculated;
106 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
107 struct ext4_inode_info *ei)
111 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
112 cpu_to_le32(EXT4_OS_LINUX) ||
113 !ext4_has_metadata_csum(inode->i_sb))
116 csum = ext4_inode_csum(inode, raw, ei);
117 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
118 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
119 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
120 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
123 static inline int ext4_begin_ordered_truncate(struct inode *inode,
126 trace_ext4_begin_ordered_truncate(inode, new_size);
128 * If jinode is zero, then we never opened the file for
129 * writing, so there's no need to call
130 * jbd2_journal_begin_ordered_truncate() since there's no
131 * outstanding writes we need to flush.
133 if (!EXT4_I(inode)->jinode)
135 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
136 EXT4_I(inode)->jinode,
140 static void ext4_invalidatepage(struct page *page, unsigned int offset,
141 unsigned int length);
142 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
143 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
144 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
148 * Test whether an inode is a fast symlink.
149 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
151 int ext4_inode_is_fast_symlink(struct inode *inode)
153 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
154 int ea_blocks = EXT4_I(inode)->i_file_acl ?
155 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
157 if (ext4_has_inline_data(inode))
160 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
162 return S_ISLNK(inode->i_mode) && inode->i_size &&
163 (inode->i_size < EXT4_N_BLOCKS * 4);
167 * Called at the last iput() if i_nlink is zero.
169 void ext4_evict_inode(struct inode *inode)
174 * Credits for final inode cleanup and freeing:
175 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
176 * (xattr block freeing), bitmap, group descriptor (inode freeing)
178 int extra_credits = 6;
179 struct ext4_xattr_inode_array *ea_inode_array = NULL;
181 trace_ext4_evict_inode(inode);
183 if (inode->i_nlink) {
185 * When journalling data dirty buffers are tracked only in the
186 * journal. So although mm thinks everything is clean and
187 * ready for reaping the inode might still have some pages to
188 * write in the running transaction or waiting to be
189 * checkpointed. Thus calling jbd2_journal_invalidatepage()
190 * (via truncate_inode_pages()) to discard these buffers can
191 * cause data loss. Also even if we did not discard these
192 * buffers, we would have no way to find them after the inode
193 * is reaped and thus user could see stale data if he tries to
194 * read them before the transaction is checkpointed. So be
195 * careful and force everything to disk here... We use
196 * ei->i_datasync_tid to store the newest transaction
197 * containing inode's data.
199 * Note that directories do not have this problem because they
200 * don't use page cache.
202 if (inode->i_ino != EXT4_JOURNAL_INO &&
203 ext4_should_journal_data(inode) &&
204 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
205 inode->i_data.nrpages) {
206 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
207 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
209 jbd2_complete_transaction(journal, commit_tid);
210 filemap_write_and_wait(&inode->i_data);
212 truncate_inode_pages_final(&inode->i_data);
217 if (is_bad_inode(inode))
219 dquot_initialize(inode);
221 if (ext4_should_order_data(inode))
222 ext4_begin_ordered_truncate(inode, 0);
223 truncate_inode_pages_final(&inode->i_data);
226 * Protect us against freezing - iput() caller didn't have to have any
227 * protection against it
229 sb_start_intwrite(inode->i_sb);
231 if (!IS_NOQUOTA(inode))
232 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
235 * Block bitmap, group descriptor, and inode are accounted in both
236 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
238 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
239 ext4_blocks_for_truncate(inode) + extra_credits - 3);
240 if (IS_ERR(handle)) {
241 ext4_std_error(inode->i_sb, PTR_ERR(handle));
243 * If we're going to skip the normal cleanup, we still need to
244 * make sure that the in-core orphan linked list is properly
247 ext4_orphan_del(NULL, inode);
248 sb_end_intwrite(inode->i_sb);
253 ext4_handle_sync(handle);
256 * Set inode->i_size to 0 before calling ext4_truncate(). We need
257 * special handling of symlinks here because i_size is used to
258 * determine whether ext4_inode_info->i_data contains symlink data or
259 * block mappings. Setting i_size to 0 will remove its fast symlink
260 * status. Erase i_data so that it becomes a valid empty block map.
262 if (ext4_inode_is_fast_symlink(inode))
263 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
265 err = ext4_mark_inode_dirty(handle, inode);
267 ext4_warning(inode->i_sb,
268 "couldn't mark inode dirty (err %d)", err);
271 if (inode->i_blocks) {
272 err = ext4_truncate(inode);
274 ext4_error(inode->i_sb,
275 "couldn't truncate inode %lu (err %d)",
281 /* Remove xattr references. */
282 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
285 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
287 ext4_journal_stop(handle);
288 ext4_orphan_del(NULL, inode);
289 sb_end_intwrite(inode->i_sb);
290 ext4_xattr_inode_array_free(ea_inode_array);
295 * Kill off the orphan record which ext4_truncate created.
296 * AKPM: I think this can be inside the above `if'.
297 * Note that ext4_orphan_del() has to be able to cope with the
298 * deletion of a non-existent orphan - this is because we don't
299 * know if ext4_truncate() actually created an orphan record.
300 * (Well, we could do this if we need to, but heck - it works)
302 ext4_orphan_del(handle, inode);
303 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
306 * One subtle ordering requirement: if anything has gone wrong
307 * (transaction abort, IO errors, whatever), then we can still
308 * do these next steps (the fs will already have been marked as
309 * having errors), but we can't free the inode if the mark_dirty
312 if (ext4_mark_inode_dirty(handle, inode))
313 /* If that failed, just do the required in-core inode clear. */
314 ext4_clear_inode(inode);
316 ext4_free_inode(handle, inode);
317 ext4_journal_stop(handle);
318 sb_end_intwrite(inode->i_sb);
319 ext4_xattr_inode_array_free(ea_inode_array);
322 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
326 qsize_t *ext4_get_reserved_space(struct inode *inode)
328 return &EXT4_I(inode)->i_reserved_quota;
333 * Called with i_data_sem down, which is important since we can call
334 * ext4_discard_preallocations() from here.
336 void ext4_da_update_reserve_space(struct inode *inode,
337 int used, int quota_claim)
339 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
340 struct ext4_inode_info *ei = EXT4_I(inode);
342 spin_lock(&ei->i_block_reservation_lock);
343 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
344 if (unlikely(used > ei->i_reserved_data_blocks)) {
345 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
346 "with only %d reserved data blocks",
347 __func__, inode->i_ino, used,
348 ei->i_reserved_data_blocks);
350 used = ei->i_reserved_data_blocks;
353 /* Update per-inode reservations */
354 ei->i_reserved_data_blocks -= used;
355 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
357 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
359 /* Update quota subsystem for data blocks */
361 dquot_claim_block(inode, EXT4_C2B(sbi, used));
364 * We did fallocate with an offset that is already delayed
365 * allocated. So on delayed allocated writeback we should
366 * not re-claim the quota for fallocated blocks.
368 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
372 * If we have done all the pending block allocations and if
373 * there aren't any writers on the inode, we can discard the
374 * inode's preallocations.
376 if ((ei->i_reserved_data_blocks == 0) &&
377 !inode_is_open_for_write(inode))
378 ext4_discard_preallocations(inode);
381 static int __check_block_validity(struct inode *inode, const char *func,
383 struct ext4_map_blocks *map)
385 if (ext4_has_feature_journal(inode->i_sb) &&
387 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
389 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
391 ext4_error_inode(inode, func, line, map->m_pblk,
392 "lblock %lu mapped to illegal pblock %llu "
393 "(length %d)", (unsigned long) map->m_lblk,
394 map->m_pblk, map->m_len);
395 return -EFSCORRUPTED;
400 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
405 if (IS_ENCRYPTED(inode))
406 return fscrypt_zeroout_range(inode, lblk, pblk, len);
408 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
415 #define check_block_validity(inode, map) \
416 __check_block_validity((inode), __func__, __LINE__, (map))
418 #ifdef ES_AGGRESSIVE_TEST
419 static void ext4_map_blocks_es_recheck(handle_t *handle,
421 struct ext4_map_blocks *es_map,
422 struct ext4_map_blocks *map,
429 * There is a race window that the result is not the same.
430 * e.g. xfstests #223 when dioread_nolock enables. The reason
431 * is that we lookup a block mapping in extent status tree with
432 * out taking i_data_sem. So at the time the unwritten extent
433 * could be converted.
435 down_read(&EXT4_I(inode)->i_data_sem);
436 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
437 retval = ext4_ext_map_blocks(handle, inode, map, flags &
438 EXT4_GET_BLOCKS_KEEP_SIZE);
440 retval = ext4_ind_map_blocks(handle, inode, map, flags &
441 EXT4_GET_BLOCKS_KEEP_SIZE);
443 up_read((&EXT4_I(inode)->i_data_sem));
446 * We don't check m_len because extent will be collpased in status
447 * tree. So the m_len might not equal.
449 if (es_map->m_lblk != map->m_lblk ||
450 es_map->m_flags != map->m_flags ||
451 es_map->m_pblk != map->m_pblk) {
452 printk("ES cache assertion failed for inode: %lu "
453 "es_cached ex [%d/%d/%llu/%x] != "
454 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
455 inode->i_ino, es_map->m_lblk, es_map->m_len,
456 es_map->m_pblk, es_map->m_flags, map->m_lblk,
457 map->m_len, map->m_pblk, map->m_flags,
461 #endif /* ES_AGGRESSIVE_TEST */
464 * The ext4_map_blocks() function tries to look up the requested blocks,
465 * and returns if the blocks are already mapped.
467 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
468 * and store the allocated blocks in the result buffer head and mark it
471 * If file type is extents based, it will call ext4_ext_map_blocks(),
472 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
475 * On success, it returns the number of blocks being mapped or allocated. if
476 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
477 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
479 * It returns 0 if plain look up failed (blocks have not been allocated), in
480 * that case, @map is returned as unmapped but we still do fill map->m_len to
481 * indicate the length of a hole starting at map->m_lblk.
483 * It returns the error in case of allocation failure.
485 int ext4_map_blocks(handle_t *handle, struct inode *inode,
486 struct ext4_map_blocks *map, int flags)
488 struct extent_status es;
491 #ifdef ES_AGGRESSIVE_TEST
492 struct ext4_map_blocks orig_map;
494 memcpy(&orig_map, map, sizeof(*map));
498 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
499 "logical block %lu\n", inode->i_ino, flags, map->m_len,
500 (unsigned long) map->m_lblk);
503 * ext4_map_blocks returns an int, and m_len is an unsigned int
505 if (unlikely(map->m_len > INT_MAX))
506 map->m_len = INT_MAX;
508 /* We can handle the block number less than EXT_MAX_BLOCKS */
509 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
510 return -EFSCORRUPTED;
512 /* Lookup extent status tree firstly */
513 if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
514 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
515 map->m_pblk = ext4_es_pblock(&es) +
516 map->m_lblk - es.es_lblk;
517 map->m_flags |= ext4_es_is_written(&es) ?
518 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
519 retval = es.es_len - (map->m_lblk - es.es_lblk);
520 if (retval > map->m_len)
523 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
525 retval = es.es_len - (map->m_lblk - es.es_lblk);
526 if (retval > map->m_len)
533 #ifdef ES_AGGRESSIVE_TEST
534 ext4_map_blocks_es_recheck(handle, inode, map,
541 * Try to see if we can get the block without requesting a new
544 down_read(&EXT4_I(inode)->i_data_sem);
545 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
546 retval = ext4_ext_map_blocks(handle, inode, map, flags &
547 EXT4_GET_BLOCKS_KEEP_SIZE);
549 retval = ext4_ind_map_blocks(handle, inode, map, flags &
550 EXT4_GET_BLOCKS_KEEP_SIZE);
555 if (unlikely(retval != map->m_len)) {
556 ext4_warning(inode->i_sb,
557 "ES len assertion failed for inode "
558 "%lu: retval %d != map->m_len %d",
559 inode->i_ino, retval, map->m_len);
563 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
564 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
565 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
566 !(status & EXTENT_STATUS_WRITTEN) &&
567 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
568 map->m_lblk + map->m_len - 1))
569 status |= EXTENT_STATUS_DELAYED;
570 ret = ext4_es_insert_extent(inode, map->m_lblk,
571 map->m_len, map->m_pblk, status);
575 up_read((&EXT4_I(inode)->i_data_sem));
578 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
579 ret = check_block_validity(inode, map);
584 /* If it is only a block(s) look up */
585 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
589 * Returns if the blocks have already allocated
591 * Note that if blocks have been preallocated
592 * ext4_ext_get_block() returns the create = 0
593 * with buffer head unmapped.
595 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
597 * If we need to convert extent to unwritten
598 * we continue and do the actual work in
599 * ext4_ext_map_blocks()
601 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
605 * Here we clear m_flags because after allocating an new extent,
606 * it will be set again.
608 map->m_flags &= ~EXT4_MAP_FLAGS;
611 * New blocks allocate and/or writing to unwritten extent
612 * will possibly result in updating i_data, so we take
613 * the write lock of i_data_sem, and call get_block()
614 * with create == 1 flag.
616 down_write(&EXT4_I(inode)->i_data_sem);
619 * We need to check for EXT4 here because migrate
620 * could have changed the inode type in between
622 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
623 retval = ext4_ext_map_blocks(handle, inode, map, flags);
625 retval = ext4_ind_map_blocks(handle, inode, map, flags);
627 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
629 * We allocated new blocks which will result in
630 * i_data's format changing. Force the migrate
631 * to fail by clearing migrate flags
633 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
637 * Update reserved blocks/metadata blocks after successful
638 * block allocation which had been deferred till now. We don't
639 * support fallocate for non extent files. So we can update
640 * reserve space here.
643 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
644 ext4_da_update_reserve_space(inode, retval, 1);
650 if (unlikely(retval != map->m_len)) {
651 ext4_warning(inode->i_sb,
652 "ES len assertion failed for inode "
653 "%lu: retval %d != map->m_len %d",
654 inode->i_ino, retval, map->m_len);
659 * We have to zeroout blocks before inserting them into extent
660 * status tree. Otherwise someone could look them up there and
661 * use them before they are really zeroed. We also have to
662 * unmap metadata before zeroing as otherwise writeback can
663 * overwrite zeros with stale data from block device.
665 if (flags & EXT4_GET_BLOCKS_ZERO &&
666 map->m_flags & EXT4_MAP_MAPPED &&
667 map->m_flags & EXT4_MAP_NEW) {
668 ret = ext4_issue_zeroout(inode, map->m_lblk,
669 map->m_pblk, map->m_len);
677 * If the extent has been zeroed out, we don't need to update
678 * extent status tree.
680 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
681 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
682 if (ext4_es_is_written(&es))
685 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
686 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
687 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
688 !(status & EXTENT_STATUS_WRITTEN) &&
689 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
690 map->m_lblk + map->m_len - 1))
691 status |= EXTENT_STATUS_DELAYED;
692 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
693 map->m_pblk, status);
701 up_write((&EXT4_I(inode)->i_data_sem));
702 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
703 ret = check_block_validity(inode, map);
708 * Inodes with freshly allocated blocks where contents will be
709 * visible after transaction commit must be on transaction's
712 if (map->m_flags & EXT4_MAP_NEW &&
713 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
714 !(flags & EXT4_GET_BLOCKS_ZERO) &&
715 !ext4_is_quota_file(inode) &&
716 ext4_should_order_data(inode)) {
718 (loff_t)map->m_lblk << inode->i_blkbits;
719 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
721 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
722 ret = ext4_jbd2_inode_add_wait(handle, inode,
725 ret = ext4_jbd2_inode_add_write(handle, inode,
735 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
736 * we have to be careful as someone else may be manipulating b_state as well.
738 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
740 unsigned long old_state;
741 unsigned long new_state;
743 flags &= EXT4_MAP_FLAGS;
745 /* Dummy buffer_head? Set non-atomically. */
747 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
751 * Someone else may be modifying b_state. Be careful! This is ugly but
752 * once we get rid of using bh as a container for mapping information
753 * to pass to / from get_block functions, this can go away.
756 old_state = READ_ONCE(bh->b_state);
757 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
759 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
762 static int _ext4_get_block(struct inode *inode, sector_t iblock,
763 struct buffer_head *bh, int flags)
765 struct ext4_map_blocks map;
768 if (ext4_has_inline_data(inode))
772 map.m_len = bh->b_size >> inode->i_blkbits;
774 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
777 map_bh(bh, inode->i_sb, map.m_pblk);
778 ext4_update_bh_state(bh, map.m_flags);
779 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
781 } else if (ret == 0) {
782 /* hole case, need to fill in bh->b_size */
783 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
788 int ext4_get_block(struct inode *inode, sector_t iblock,
789 struct buffer_head *bh, int create)
791 return _ext4_get_block(inode, iblock, bh,
792 create ? EXT4_GET_BLOCKS_CREATE : 0);
796 * Get block function used when preparing for buffered write if we require
797 * creating an unwritten extent if blocks haven't been allocated. The extent
798 * will be converted to written after the IO is complete.
800 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
801 struct buffer_head *bh_result, int create)
803 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
804 inode->i_ino, create);
805 return _ext4_get_block(inode, iblock, bh_result,
806 EXT4_GET_BLOCKS_IO_CREATE_EXT);
809 /* Maximum number of blocks we map for direct IO at once. */
810 #define DIO_MAX_BLOCKS 4096
813 * `handle' can be NULL if create is zero
815 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
816 ext4_lblk_t block, int map_flags)
818 struct ext4_map_blocks map;
819 struct buffer_head *bh;
820 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
823 J_ASSERT(handle != NULL || create == 0);
827 err = ext4_map_blocks(handle, inode, &map, map_flags);
830 return create ? ERR_PTR(-ENOSPC) : NULL;
834 bh = sb_getblk(inode->i_sb, map.m_pblk);
836 return ERR_PTR(-ENOMEM);
837 if (map.m_flags & EXT4_MAP_NEW) {
838 J_ASSERT(create != 0);
839 J_ASSERT(handle != NULL);
842 * Now that we do not always journal data, we should
843 * keep in mind whether this should always journal the
844 * new buffer as metadata. For now, regular file
845 * writes use ext4_get_block instead, so it's not a
849 BUFFER_TRACE(bh, "call get_create_access");
850 err = ext4_journal_get_create_access(handle, bh);
855 if (!buffer_uptodate(bh)) {
856 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
857 set_buffer_uptodate(bh);
860 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
861 err = ext4_handle_dirty_metadata(handle, inode, bh);
865 BUFFER_TRACE(bh, "not a new buffer");
872 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
873 ext4_lblk_t block, int map_flags)
875 struct buffer_head *bh;
877 bh = ext4_getblk(handle, inode, block, map_flags);
880 if (!bh || ext4_buffer_uptodate(bh))
882 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
884 if (buffer_uptodate(bh))
887 return ERR_PTR(-EIO);
890 /* Read a contiguous batch of blocks. */
891 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
892 bool wait, struct buffer_head **bhs)
896 for (i = 0; i < bh_count; i++) {
897 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
898 if (IS_ERR(bhs[i])) {
899 err = PTR_ERR(bhs[i]);
905 for (i = 0; i < bh_count; i++)
906 /* Note that NULL bhs[i] is valid because of holes. */
907 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
908 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
914 for (i = 0; i < bh_count; i++)
916 wait_on_buffer(bhs[i]);
918 for (i = 0; i < bh_count; i++) {
919 if (bhs[i] && !buffer_uptodate(bhs[i])) {
927 for (i = 0; i < bh_count; i++) {
934 int ext4_walk_page_buffers(handle_t *handle,
935 struct buffer_head *head,
939 int (*fn)(handle_t *handle,
940 struct buffer_head *bh))
942 struct buffer_head *bh;
943 unsigned block_start, block_end;
944 unsigned blocksize = head->b_size;
946 struct buffer_head *next;
948 for (bh = head, block_start = 0;
949 ret == 0 && (bh != head || !block_start);
950 block_start = block_end, bh = next) {
951 next = bh->b_this_page;
952 block_end = block_start + blocksize;
953 if (block_end <= from || block_start >= to) {
954 if (partial && !buffer_uptodate(bh))
958 err = (*fn)(handle, bh);
966 * To preserve ordering, it is essential that the hole instantiation and
967 * the data write be encapsulated in a single transaction. We cannot
968 * close off a transaction and start a new one between the ext4_get_block()
969 * and the commit_write(). So doing the jbd2_journal_start at the start of
970 * prepare_write() is the right place.
972 * Also, this function can nest inside ext4_writepage(). In that case, we
973 * *know* that ext4_writepage() has generated enough buffer credits to do the
974 * whole page. So we won't block on the journal in that case, which is good,
975 * because the caller may be PF_MEMALLOC.
977 * By accident, ext4 can be reentered when a transaction is open via
978 * quota file writes. If we were to commit the transaction while thus
979 * reentered, there can be a deadlock - we would be holding a quota
980 * lock, and the commit would never complete if another thread had a
981 * transaction open and was blocking on the quota lock - a ranking
984 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
985 * will _not_ run commit under these circumstances because handle->h_ref
986 * is elevated. We'll still have enough credits for the tiny quotafile
989 int do_journal_get_write_access(handle_t *handle,
990 struct buffer_head *bh)
992 int dirty = buffer_dirty(bh);
995 if (!buffer_mapped(bh) || buffer_freed(bh))
998 * __block_write_begin() could have dirtied some buffers. Clean
999 * the dirty bit as jbd2_journal_get_write_access() could complain
1000 * otherwise about fs integrity issues. Setting of the dirty bit
1001 * by __block_write_begin() isn't a real problem here as we clear
1002 * the bit before releasing a page lock and thus writeback cannot
1003 * ever write the buffer.
1006 clear_buffer_dirty(bh);
1007 BUFFER_TRACE(bh, "get write access");
1008 ret = ext4_journal_get_write_access(handle, bh);
1010 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1014 #ifdef CONFIG_FS_ENCRYPTION
1015 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1016 get_block_t *get_block)
1018 unsigned from = pos & (PAGE_SIZE - 1);
1019 unsigned to = from + len;
1020 struct inode *inode = page->mapping->host;
1021 unsigned block_start, block_end;
1024 unsigned blocksize = inode->i_sb->s_blocksize;
1026 struct buffer_head *bh, *head, *wait[2];
1030 BUG_ON(!PageLocked(page));
1031 BUG_ON(from > PAGE_SIZE);
1032 BUG_ON(to > PAGE_SIZE);
1035 if (!page_has_buffers(page))
1036 create_empty_buffers(page, blocksize, 0);
1037 head = page_buffers(page);
1038 bbits = ilog2(blocksize);
1039 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1041 for (bh = head, block_start = 0; bh != head || !block_start;
1042 block++, block_start = block_end, bh = bh->b_this_page) {
1043 block_end = block_start + blocksize;
1044 if (block_end <= from || block_start >= to) {
1045 if (PageUptodate(page)) {
1046 if (!buffer_uptodate(bh))
1047 set_buffer_uptodate(bh);
1052 clear_buffer_new(bh);
1053 if (!buffer_mapped(bh)) {
1054 WARN_ON(bh->b_size != blocksize);
1055 err = get_block(inode, block, bh, 1);
1058 if (buffer_new(bh)) {
1059 if (PageUptodate(page)) {
1060 clear_buffer_new(bh);
1061 set_buffer_uptodate(bh);
1062 mark_buffer_dirty(bh);
1065 if (block_end > to || block_start < from)
1066 zero_user_segments(page, to, block_end,
1071 if (PageUptodate(page)) {
1072 if (!buffer_uptodate(bh))
1073 set_buffer_uptodate(bh);
1076 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1077 !buffer_unwritten(bh) &&
1078 (block_start < from || block_end > to)) {
1079 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1080 wait[nr_wait++] = bh;
1084 * If we issued read requests, let them complete.
1086 for (i = 0; i < nr_wait; i++) {
1087 wait_on_buffer(wait[i]);
1088 if (!buffer_uptodate(wait[i]))
1091 if (unlikely(err)) {
1092 page_zero_new_buffers(page, from, to);
1093 } else if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
1094 for (i = 0; i < nr_wait; i++) {
1097 err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1098 bh_offset(wait[i]));
1100 clear_buffer_uptodate(wait[i]);
1110 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1111 loff_t pos, unsigned len, unsigned flags,
1112 struct page **pagep, void **fsdata)
1114 struct inode *inode = mapping->host;
1115 int ret, needed_blocks;
1122 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1125 trace_ext4_write_begin(inode, pos, len, flags);
1127 * Reserve one block more for addition to orphan list in case
1128 * we allocate blocks but write fails for some reason
1130 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1131 index = pos >> PAGE_SHIFT;
1132 from = pos & (PAGE_SIZE - 1);
1135 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1136 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1145 * grab_cache_page_write_begin() can take a long time if the
1146 * system is thrashing due to memory pressure, or if the page
1147 * is being written back. So grab it first before we start
1148 * the transaction handle. This also allows us to allocate
1149 * the page (if needed) without using GFP_NOFS.
1152 page = grab_cache_page_write_begin(mapping, index, flags);
1158 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1159 if (IS_ERR(handle)) {
1161 return PTR_ERR(handle);
1165 if (page->mapping != mapping) {
1166 /* The page got truncated from under us */
1169 ext4_journal_stop(handle);
1172 /* In case writeback began while the page was unlocked */
1173 wait_for_stable_page(page);
1175 #ifdef CONFIG_FS_ENCRYPTION
1176 if (ext4_should_dioread_nolock(inode))
1177 ret = ext4_block_write_begin(page, pos, len,
1178 ext4_get_block_unwritten);
1180 ret = ext4_block_write_begin(page, pos, len,
1183 if (ext4_should_dioread_nolock(inode))
1184 ret = __block_write_begin(page, pos, len,
1185 ext4_get_block_unwritten);
1187 ret = __block_write_begin(page, pos, len, ext4_get_block);
1189 if (!ret && ext4_should_journal_data(inode)) {
1190 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1192 do_journal_get_write_access);
1196 bool extended = (pos + len > inode->i_size) &&
1197 !ext4_verity_in_progress(inode);
1201 * __block_write_begin may have instantiated a few blocks
1202 * outside i_size. Trim these off again. Don't need
1203 * i_size_read because we hold i_mutex.
1205 * Add inode to orphan list in case we crash before
1208 if (extended && ext4_can_truncate(inode))
1209 ext4_orphan_add(handle, inode);
1211 ext4_journal_stop(handle);
1213 ext4_truncate_failed_write(inode);
1215 * If truncate failed early the inode might
1216 * still be on the orphan list; we need to
1217 * make sure the inode is removed from the
1218 * orphan list in that case.
1221 ext4_orphan_del(NULL, inode);
1224 if (ret == -ENOSPC &&
1225 ext4_should_retry_alloc(inode->i_sb, &retries))
1234 /* For write_end() in data=journal mode */
1235 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1238 if (!buffer_mapped(bh) || buffer_freed(bh))
1240 set_buffer_uptodate(bh);
1241 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1242 clear_buffer_meta(bh);
1243 clear_buffer_prio(bh);
1248 * We need to pick up the new inode size which generic_commit_write gave us
1249 * `file' can be NULL - eg, when called from page_symlink().
1251 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1252 * buffers are managed internally.
1254 static int ext4_write_end(struct file *file,
1255 struct address_space *mapping,
1256 loff_t pos, unsigned len, unsigned copied,
1257 struct page *page, void *fsdata)
1259 handle_t *handle = ext4_journal_current_handle();
1260 struct inode *inode = mapping->host;
1261 loff_t old_size = inode->i_size;
1263 int i_size_changed = 0;
1264 int inline_data = ext4_has_inline_data(inode);
1265 bool verity = ext4_verity_in_progress(inode);
1267 trace_ext4_write_end(inode, pos, len, copied);
1269 ret = ext4_write_inline_data_end(inode, pos, len,
1278 copied = block_write_end(file, mapping, pos,
1279 len, copied, page, fsdata);
1281 * it's important to update i_size while still holding page lock:
1282 * page writeout could otherwise come in and zero beyond i_size.
1284 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1285 * blocks are being written past EOF, so skip the i_size update.
1288 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1292 if (old_size < pos && !verity)
1293 pagecache_isize_extended(inode, old_size, pos);
1295 * Don't mark the inode dirty under page lock. First, it unnecessarily
1296 * makes the holding time of page lock longer. Second, it forces lock
1297 * ordering of page lock and transaction start for journaling
1300 if (i_size_changed || inline_data)
1301 ext4_mark_inode_dirty(handle, inode);
1303 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1304 /* if we have allocated more blocks and copied
1305 * less. We will have blocks allocated outside
1306 * inode->i_size. So truncate them
1308 ext4_orphan_add(handle, inode);
1310 ret2 = ext4_journal_stop(handle);
1314 if (pos + len > inode->i_size && !verity) {
1315 ext4_truncate_failed_write(inode);
1317 * If truncate failed early the inode might still be
1318 * on the orphan list; we need to make sure the inode
1319 * is removed from the orphan list in that case.
1322 ext4_orphan_del(NULL, inode);
1325 return ret ? ret : copied;
1329 * This is a private version of page_zero_new_buffers() which doesn't
1330 * set the buffer to be dirty, since in data=journalled mode we need
1331 * to call ext4_handle_dirty_metadata() instead.
1333 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1335 unsigned from, unsigned to)
1337 unsigned int block_start = 0, block_end;
1338 struct buffer_head *head, *bh;
1340 bh = head = page_buffers(page);
1342 block_end = block_start + bh->b_size;
1343 if (buffer_new(bh)) {
1344 if (block_end > from && block_start < to) {
1345 if (!PageUptodate(page)) {
1346 unsigned start, size;
1348 start = max(from, block_start);
1349 size = min(to, block_end) - start;
1351 zero_user(page, start, size);
1352 write_end_fn(handle, bh);
1354 clear_buffer_new(bh);
1357 block_start = block_end;
1358 bh = bh->b_this_page;
1359 } while (bh != head);
1362 static int ext4_journalled_write_end(struct file *file,
1363 struct address_space *mapping,
1364 loff_t pos, unsigned len, unsigned copied,
1365 struct page *page, void *fsdata)
1367 handle_t *handle = ext4_journal_current_handle();
1368 struct inode *inode = mapping->host;
1369 loff_t old_size = inode->i_size;
1373 int size_changed = 0;
1374 int inline_data = ext4_has_inline_data(inode);
1375 bool verity = ext4_verity_in_progress(inode);
1377 trace_ext4_journalled_write_end(inode, pos, len, copied);
1378 from = pos & (PAGE_SIZE - 1);
1381 BUG_ON(!ext4_handle_valid(handle));
1384 ret = ext4_write_inline_data_end(inode, pos, len,
1392 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1394 ext4_journalled_zero_new_buffers(handle, page, from, to);
1396 if (unlikely(copied < len))
1397 ext4_journalled_zero_new_buffers(handle, page,
1399 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1400 from + copied, &partial,
1403 SetPageUptodate(page);
1406 size_changed = ext4_update_inode_size(inode, pos + copied);
1407 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1408 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1412 if (old_size < pos && !verity)
1413 pagecache_isize_extended(inode, old_size, pos);
1415 if (size_changed || inline_data) {
1416 ret2 = ext4_mark_inode_dirty(handle, inode);
1421 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1422 /* if we have allocated more blocks and copied
1423 * less. We will have blocks allocated outside
1424 * inode->i_size. So truncate them
1426 ext4_orphan_add(handle, inode);
1429 ret2 = ext4_journal_stop(handle);
1432 if (pos + len > inode->i_size && !verity) {
1433 ext4_truncate_failed_write(inode);
1435 * If truncate failed early the inode might still be
1436 * on the orphan list; we need to make sure the inode
1437 * is removed from the orphan list in that case.
1440 ext4_orphan_del(NULL, inode);
1443 return ret ? ret : copied;
1447 * Reserve space for a single cluster
1449 static int ext4_da_reserve_space(struct inode *inode)
1451 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1452 struct ext4_inode_info *ei = EXT4_I(inode);
1456 * We will charge metadata quota at writeout time; this saves
1457 * us from metadata over-estimation, though we may go over by
1458 * a small amount in the end. Here we just reserve for data.
1460 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1464 spin_lock(&ei->i_block_reservation_lock);
1465 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1466 spin_unlock(&ei->i_block_reservation_lock);
1467 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1470 ei->i_reserved_data_blocks++;
1471 trace_ext4_da_reserve_space(inode);
1472 spin_unlock(&ei->i_block_reservation_lock);
1474 return 0; /* success */
1477 void ext4_da_release_space(struct inode *inode, int to_free)
1479 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1480 struct ext4_inode_info *ei = EXT4_I(inode);
1483 return; /* Nothing to release, exit */
1485 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1487 trace_ext4_da_release_space(inode, to_free);
1488 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1490 * if there aren't enough reserved blocks, then the
1491 * counter is messed up somewhere. Since this
1492 * function is called from invalidate page, it's
1493 * harmless to return without any action.
1495 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1496 "ino %lu, to_free %d with only %d reserved "
1497 "data blocks", inode->i_ino, to_free,
1498 ei->i_reserved_data_blocks);
1500 to_free = ei->i_reserved_data_blocks;
1502 ei->i_reserved_data_blocks -= to_free;
1504 /* update fs dirty data blocks counter */
1505 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1507 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1509 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1513 * Delayed allocation stuff
1516 struct mpage_da_data {
1517 struct inode *inode;
1518 struct writeback_control *wbc;
1520 pgoff_t first_page; /* The first page to write */
1521 pgoff_t next_page; /* Current page to examine */
1522 pgoff_t last_page; /* Last page to examine */
1524 * Extent to map - this can be after first_page because that can be
1525 * fully mapped. We somewhat abuse m_flags to store whether the extent
1526 * is delalloc or unwritten.
1528 struct ext4_map_blocks map;
1529 struct ext4_io_submit io_submit; /* IO submission data */
1530 unsigned int do_map:1;
1533 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1538 struct pagevec pvec;
1539 struct inode *inode = mpd->inode;
1540 struct address_space *mapping = inode->i_mapping;
1542 /* This is necessary when next_page == 0. */
1543 if (mpd->first_page >= mpd->next_page)
1546 index = mpd->first_page;
1547 end = mpd->next_page - 1;
1549 ext4_lblk_t start, last;
1550 start = index << (PAGE_SHIFT - inode->i_blkbits);
1551 last = end << (PAGE_SHIFT - inode->i_blkbits);
1552 ext4_es_remove_extent(inode, start, last - start + 1);
1555 pagevec_init(&pvec);
1556 while (index <= end) {
1557 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1560 for (i = 0; i < nr_pages; i++) {
1561 struct page *page = pvec.pages[i];
1563 BUG_ON(!PageLocked(page));
1564 BUG_ON(PageWriteback(page));
1566 if (page_mapped(page))
1567 clear_page_dirty_for_io(page);
1568 block_invalidatepage(page, 0, PAGE_SIZE);
1569 ClearPageUptodate(page);
1573 pagevec_release(&pvec);
1577 static void ext4_print_free_blocks(struct inode *inode)
1579 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1580 struct super_block *sb = inode->i_sb;
1581 struct ext4_inode_info *ei = EXT4_I(inode);
1583 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1584 EXT4_C2B(EXT4_SB(inode->i_sb),
1585 ext4_count_free_clusters(sb)));
1586 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1587 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1588 (long long) EXT4_C2B(EXT4_SB(sb),
1589 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1590 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1591 (long long) EXT4_C2B(EXT4_SB(sb),
1592 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1593 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1594 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1595 ei->i_reserved_data_blocks);
1599 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1601 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1605 * ext4_insert_delayed_block - adds a delayed block to the extents status
1606 * tree, incrementing the reserved cluster/block
1607 * count or making a pending reservation
1610 * @inode - file containing the newly added block
1611 * @lblk - logical block to be added
1613 * Returns 0 on success, negative error code on failure.
1615 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1617 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1619 bool allocated = false;
1622 * If the cluster containing lblk is shared with a delayed,
1623 * written, or unwritten extent in a bigalloc file system, it's
1624 * already been accounted for and does not need to be reserved.
1625 * A pending reservation must be made for the cluster if it's
1626 * shared with a written or unwritten extent and doesn't already
1627 * have one. Written and unwritten extents can be purged from the
1628 * extents status tree if the system is under memory pressure, so
1629 * it's necessary to examine the extent tree if a search of the
1630 * extents status tree doesn't get a match.
1632 if (sbi->s_cluster_ratio == 1) {
1633 ret = ext4_da_reserve_space(inode);
1634 if (ret != 0) /* ENOSPC */
1636 } else { /* bigalloc */
1637 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1638 if (!ext4_es_scan_clu(inode,
1639 &ext4_es_is_mapped, lblk)) {
1640 ret = ext4_clu_mapped(inode,
1641 EXT4_B2C(sbi, lblk));
1645 ret = ext4_da_reserve_space(inode);
1646 if (ret != 0) /* ENOSPC */
1657 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1664 * This function is grabs code from the very beginning of
1665 * ext4_map_blocks, but assumes that the caller is from delayed write
1666 * time. This function looks up the requested blocks and sets the
1667 * buffer delay bit under the protection of i_data_sem.
1669 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1670 struct ext4_map_blocks *map,
1671 struct buffer_head *bh)
1673 struct extent_status es;
1675 sector_t invalid_block = ~((sector_t) 0xffff);
1676 #ifdef ES_AGGRESSIVE_TEST
1677 struct ext4_map_blocks orig_map;
1679 memcpy(&orig_map, map, sizeof(*map));
1682 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1686 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1687 "logical block %lu\n", inode->i_ino, map->m_len,
1688 (unsigned long) map->m_lblk);
1690 /* Lookup extent status tree firstly */
1691 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1692 if (ext4_es_is_hole(&es)) {
1694 down_read(&EXT4_I(inode)->i_data_sem);
1699 * Delayed extent could be allocated by fallocate.
1700 * So we need to check it.
1702 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1703 map_bh(bh, inode->i_sb, invalid_block);
1705 set_buffer_delay(bh);
1709 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1710 retval = es.es_len - (iblock - es.es_lblk);
1711 if (retval > map->m_len)
1712 retval = map->m_len;
1713 map->m_len = retval;
1714 if (ext4_es_is_written(&es))
1715 map->m_flags |= EXT4_MAP_MAPPED;
1716 else if (ext4_es_is_unwritten(&es))
1717 map->m_flags |= EXT4_MAP_UNWRITTEN;
1721 #ifdef ES_AGGRESSIVE_TEST
1722 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1728 * Try to see if we can get the block without requesting a new
1729 * file system block.
1731 down_read(&EXT4_I(inode)->i_data_sem);
1732 if (ext4_has_inline_data(inode))
1734 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1735 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1737 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1744 * XXX: __block_prepare_write() unmaps passed block,
1748 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1754 map_bh(bh, inode->i_sb, invalid_block);
1756 set_buffer_delay(bh);
1757 } else if (retval > 0) {
1759 unsigned int status;
1761 if (unlikely(retval != map->m_len)) {
1762 ext4_warning(inode->i_sb,
1763 "ES len assertion failed for inode "
1764 "%lu: retval %d != map->m_len %d",
1765 inode->i_ino, retval, map->m_len);
1769 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1770 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1771 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1772 map->m_pblk, status);
1778 up_read((&EXT4_I(inode)->i_data_sem));
1784 * This is a special get_block_t callback which is used by
1785 * ext4_da_write_begin(). It will either return mapped block or
1786 * reserve space for a single block.
1788 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1789 * We also have b_blocknr = -1 and b_bdev initialized properly
1791 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1792 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1793 * initialized properly.
1795 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1796 struct buffer_head *bh, int create)
1798 struct ext4_map_blocks map;
1801 BUG_ON(create == 0);
1802 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1804 map.m_lblk = iblock;
1808 * first, we need to know whether the block is allocated already
1809 * preallocated blocks are unmapped but should treated
1810 * the same as allocated blocks.
1812 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1816 map_bh(bh, inode->i_sb, map.m_pblk);
1817 ext4_update_bh_state(bh, map.m_flags);
1819 if (buffer_unwritten(bh)) {
1820 /* A delayed write to unwritten bh should be marked
1821 * new and mapped. Mapped ensures that we don't do
1822 * get_block multiple times when we write to the same
1823 * offset and new ensures that we do proper zero out
1824 * for partial write.
1827 set_buffer_mapped(bh);
1832 static int bget_one(handle_t *handle, struct buffer_head *bh)
1838 static int bput_one(handle_t *handle, struct buffer_head *bh)
1844 static int __ext4_journalled_writepage(struct page *page,
1847 struct address_space *mapping = page->mapping;
1848 struct inode *inode = mapping->host;
1849 struct buffer_head *page_bufs = NULL;
1850 handle_t *handle = NULL;
1851 int ret = 0, err = 0;
1852 int inline_data = ext4_has_inline_data(inode);
1853 struct buffer_head *inode_bh = NULL;
1855 ClearPageChecked(page);
1858 BUG_ON(page->index != 0);
1859 BUG_ON(len > ext4_get_max_inline_size(inode));
1860 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1861 if (inode_bh == NULL)
1864 page_bufs = page_buffers(page);
1869 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1873 * We need to release the page lock before we start the
1874 * journal, so grab a reference so the page won't disappear
1875 * out from under us.
1880 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1881 ext4_writepage_trans_blocks(inode));
1882 if (IS_ERR(handle)) {
1883 ret = PTR_ERR(handle);
1885 goto out_no_pagelock;
1887 BUG_ON(!ext4_handle_valid(handle));
1891 if (page->mapping != mapping) {
1892 /* The page got truncated from under us */
1893 ext4_journal_stop(handle);
1899 ret = ext4_mark_inode_dirty(handle, inode);
1901 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1902 do_journal_get_write_access);
1904 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1909 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1910 err = ext4_journal_stop(handle);
1914 if (!ext4_has_inline_data(inode))
1915 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1917 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1926 * Note that we don't need to start a transaction unless we're journaling data
1927 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1928 * need to file the inode to the transaction's list in ordered mode because if
1929 * we are writing back data added by write(), the inode is already there and if
1930 * we are writing back data modified via mmap(), no one guarantees in which
1931 * transaction the data will hit the disk. In case we are journaling data, we
1932 * cannot start transaction directly because transaction start ranks above page
1933 * lock so we have to do some magic.
1935 * This function can get called via...
1936 * - ext4_writepages after taking page lock (have journal handle)
1937 * - journal_submit_inode_data_buffers (no journal handle)
1938 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1939 * - grab_page_cache when doing write_begin (have journal handle)
1941 * We don't do any block allocation in this function. If we have page with
1942 * multiple blocks we need to write those buffer_heads that are mapped. This
1943 * is important for mmaped based write. So if we do with blocksize 1K
1944 * truncate(f, 1024);
1945 * a = mmap(f, 0, 4096);
1947 * truncate(f, 4096);
1948 * we have in the page first buffer_head mapped via page_mkwrite call back
1949 * but other buffer_heads would be unmapped but dirty (dirty done via the
1950 * do_wp_page). So writepage should write the first block. If we modify
1951 * the mmap area beyond 1024 we will again get a page_fault and the
1952 * page_mkwrite callback will do the block allocation and mark the
1953 * buffer_heads mapped.
1955 * We redirty the page if we have any buffer_heads that is either delay or
1956 * unwritten in the page.
1958 * We can get recursively called as show below.
1960 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1963 * But since we don't do any block allocation we should not deadlock.
1964 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1966 static int ext4_writepage(struct page *page,
1967 struct writeback_control *wbc)
1972 struct buffer_head *page_bufs = NULL;
1973 struct inode *inode = page->mapping->host;
1974 struct ext4_io_submit io_submit;
1975 bool keep_towrite = false;
1977 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
1978 ext4_invalidatepage(page, 0, PAGE_SIZE);
1983 trace_ext4_writepage(page);
1984 size = i_size_read(inode);
1985 if (page->index == size >> PAGE_SHIFT &&
1986 !ext4_verity_in_progress(inode))
1987 len = size & ~PAGE_MASK;
1991 page_bufs = page_buffers(page);
1993 * We cannot do block allocation or other extent handling in this
1994 * function. If there are buffers needing that, we have to redirty
1995 * the page. But we may reach here when we do a journal commit via
1996 * journal_submit_inode_data_buffers() and in that case we must write
1997 * allocated buffers to achieve data=ordered mode guarantees.
1999 * Also, if there is only one buffer per page (the fs block
2000 * size == the page size), if one buffer needs block
2001 * allocation or needs to modify the extent tree to clear the
2002 * unwritten flag, we know that the page can't be written at
2003 * all, so we might as well refuse the write immediately.
2004 * Unfortunately if the block size != page size, we can't as
2005 * easily detect this case using ext4_walk_page_buffers(), but
2006 * for the extremely common case, this is an optimization that
2007 * skips a useless round trip through ext4_bio_write_page().
2009 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2010 ext4_bh_delay_or_unwritten)) {
2011 redirty_page_for_writepage(wbc, page);
2012 if ((current->flags & PF_MEMALLOC) ||
2013 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2015 * For memory cleaning there's no point in writing only
2016 * some buffers. So just bail out. Warn if we came here
2017 * from direct reclaim.
2019 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2024 keep_towrite = true;
2027 if (PageChecked(page) && ext4_should_journal_data(inode))
2029 * It's mmapped pagecache. Add buffers and journal it. There
2030 * doesn't seem much point in redirtying the page here.
2032 return __ext4_journalled_writepage(page, len);
2034 ext4_io_submit_init(&io_submit, wbc);
2035 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2036 if (!io_submit.io_end) {
2037 redirty_page_for_writepage(wbc, page);
2041 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2042 ext4_io_submit(&io_submit);
2043 /* Drop io_end reference we got from init */
2044 ext4_put_io_end_defer(io_submit.io_end);
2048 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2054 BUG_ON(page->index != mpd->first_page);
2055 clear_page_dirty_for_io(page);
2057 * We have to be very careful here! Nothing protects writeback path
2058 * against i_size changes and the page can be writeably mapped into
2059 * page tables. So an application can be growing i_size and writing
2060 * data through mmap while writeback runs. clear_page_dirty_for_io()
2061 * write-protects our page in page tables and the page cannot get
2062 * written to again until we release page lock. So only after
2063 * clear_page_dirty_for_io() we are safe to sample i_size for
2064 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2065 * on the barrier provided by TestClearPageDirty in
2066 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2067 * after page tables are updated.
2069 size = i_size_read(mpd->inode);
2070 if (page->index == size >> PAGE_SHIFT &&
2071 !ext4_verity_in_progress(mpd->inode))
2072 len = size & ~PAGE_MASK;
2075 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2077 mpd->wbc->nr_to_write--;
2083 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2086 * mballoc gives us at most this number of blocks...
2087 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2088 * The rest of mballoc seems to handle chunks up to full group size.
2090 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2093 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2095 * @mpd - extent of blocks
2096 * @lblk - logical number of the block in the file
2097 * @bh - buffer head we want to add to the extent
2099 * The function is used to collect contig. blocks in the same state. If the
2100 * buffer doesn't require mapping for writeback and we haven't started the
2101 * extent of buffers to map yet, the function returns 'true' immediately - the
2102 * caller can write the buffer right away. Otherwise the function returns true
2103 * if the block has been added to the extent, false if the block couldn't be
2106 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2107 struct buffer_head *bh)
2109 struct ext4_map_blocks *map = &mpd->map;
2111 /* Buffer that doesn't need mapping for writeback? */
2112 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2113 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2114 /* So far no extent to map => we write the buffer right away */
2115 if (map->m_len == 0)
2120 /* First block in the extent? */
2121 if (map->m_len == 0) {
2122 /* We cannot map unless handle is started... */
2127 map->m_flags = bh->b_state & BH_FLAGS;
2131 /* Don't go larger than mballoc is willing to allocate */
2132 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2135 /* Can we merge the block to our big extent? */
2136 if (lblk == map->m_lblk + map->m_len &&
2137 (bh->b_state & BH_FLAGS) == map->m_flags) {
2145 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2147 * @mpd - extent of blocks for mapping
2148 * @head - the first buffer in the page
2149 * @bh - buffer we should start processing from
2150 * @lblk - logical number of the block in the file corresponding to @bh
2152 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2153 * the page for IO if all buffers in this page were mapped and there's no
2154 * accumulated extent of buffers to map or add buffers in the page to the
2155 * extent of buffers to map. The function returns 1 if the caller can continue
2156 * by processing the next page, 0 if it should stop adding buffers to the
2157 * extent to map because we cannot extend it anymore. It can also return value
2158 * < 0 in case of error during IO submission.
2160 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2161 struct buffer_head *head,
2162 struct buffer_head *bh,
2165 struct inode *inode = mpd->inode;
2167 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2168 >> inode->i_blkbits;
2170 if (ext4_verity_in_progress(inode))
2171 blocks = EXT_MAX_BLOCKS;
2174 BUG_ON(buffer_locked(bh));
2176 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2177 /* Found extent to map? */
2180 /* Buffer needs mapping and handle is not started? */
2183 /* Everything mapped so far and we hit EOF */
2186 } while (lblk++, (bh = bh->b_this_page) != head);
2187 /* So far everything mapped? Submit the page for IO. */
2188 if (mpd->map.m_len == 0) {
2189 err = mpage_submit_page(mpd, head->b_page);
2193 return lblk < blocks;
2197 * mpage_process_page - update page buffers corresponding to changed extent and
2198 * may submit fully mapped page for IO
2200 * @mpd - description of extent to map, on return next extent to map
2201 * @m_lblk - logical block mapping.
2202 * @m_pblk - corresponding physical mapping.
2203 * @map_bh - determines on return whether this page requires any further
2205 * Scan given page buffers corresponding to changed extent and update buffer
2206 * state according to new extent state.
2207 * We map delalloc buffers to their physical location, clear unwritten bits.
2208 * If the given page is not fully mapped, we update @map to the next extent in
2209 * the given page that needs mapping & return @map_bh as true.
2211 static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2212 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2215 struct buffer_head *head, *bh;
2216 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2217 ext4_lblk_t lblk = *m_lblk;
2218 ext4_fsblk_t pblock = *m_pblk;
2220 int blkbits = mpd->inode->i_blkbits;
2221 ssize_t io_end_size = 0;
2222 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2224 bh = head = page_buffers(page);
2226 if (lblk < mpd->map.m_lblk)
2228 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2230 * Buffer after end of mapped extent.
2231 * Find next buffer in the page to map.
2234 mpd->map.m_flags = 0;
2235 io_end_vec->size += io_end_size;
2238 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2241 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2242 io_end_vec = ext4_alloc_io_end_vec(io_end);
2243 io_end_vec->offset = mpd->map.m_lblk << blkbits;
2248 if (buffer_delay(bh)) {
2249 clear_buffer_delay(bh);
2250 bh->b_blocknr = pblock++;
2252 clear_buffer_unwritten(bh);
2253 io_end_size += (1 << blkbits);
2254 } while (lblk++, (bh = bh->b_this_page) != head);
2256 io_end_vec->size += io_end_size;
2266 * mpage_map_buffers - update buffers corresponding to changed extent and
2267 * submit fully mapped pages for IO
2269 * @mpd - description of extent to map, on return next extent to map
2271 * Scan buffers corresponding to changed extent (we expect corresponding pages
2272 * to be already locked) and update buffer state according to new extent state.
2273 * We map delalloc buffers to their physical location, clear unwritten bits,
2274 * and mark buffers as uninit when we perform writes to unwritten extents
2275 * and do extent conversion after IO is finished. If the last page is not fully
2276 * mapped, we update @map to the next extent in the last page that needs
2277 * mapping. Otherwise we submit the page for IO.
2279 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2281 struct pagevec pvec;
2283 struct inode *inode = mpd->inode;
2284 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2287 ext4_fsblk_t pblock;
2289 bool map_bh = false;
2291 start = mpd->map.m_lblk >> bpp_bits;
2292 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2293 lblk = start << bpp_bits;
2294 pblock = mpd->map.m_pblk;
2296 pagevec_init(&pvec);
2297 while (start <= end) {
2298 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2302 for (i = 0; i < nr_pages; i++) {
2303 struct page *page = pvec.pages[i];
2305 err = mpage_process_page(mpd, page, &lblk, &pblock,
2308 * If map_bh is true, means page may require further bh
2309 * mapping, or maybe the page was submitted for IO.
2310 * So we return to call further extent mapping.
2312 if (err < 0 || map_bh == true)
2314 /* Page fully mapped - let IO run! */
2315 err = mpage_submit_page(mpd, page);
2319 pagevec_release(&pvec);
2321 /* Extent fully mapped and matches with page boundary. We are done. */
2323 mpd->map.m_flags = 0;
2326 pagevec_release(&pvec);
2330 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2332 struct inode *inode = mpd->inode;
2333 struct ext4_map_blocks *map = &mpd->map;
2334 int get_blocks_flags;
2335 int err, dioread_nolock;
2337 trace_ext4_da_write_pages_extent(inode, map);
2339 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2340 * to convert an unwritten extent to be initialized (in the case
2341 * where we have written into one or more preallocated blocks). It is
2342 * possible that we're going to need more metadata blocks than
2343 * previously reserved. However we must not fail because we're in
2344 * writeback and there is nothing we can do about it so it might result
2345 * in data loss. So use reserved blocks to allocate metadata if
2348 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2349 * the blocks in question are delalloc blocks. This indicates
2350 * that the blocks and quotas has already been checked when
2351 * the data was copied into the page cache.
2353 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2354 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2355 EXT4_GET_BLOCKS_IO_SUBMIT;
2356 dioread_nolock = ext4_should_dioread_nolock(inode);
2358 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2359 if (map->m_flags & (1 << BH_Delay))
2360 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2362 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2365 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2366 if (!mpd->io_submit.io_end->handle &&
2367 ext4_handle_valid(handle)) {
2368 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2369 handle->h_rsv_handle = NULL;
2371 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2374 BUG_ON(map->m_len == 0);
2379 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2380 * mpd->len and submit pages underlying it for IO
2382 * @handle - handle for journal operations
2383 * @mpd - extent to map
2384 * @give_up_on_write - we set this to true iff there is a fatal error and there
2385 * is no hope of writing the data. The caller should discard
2386 * dirty pages to avoid infinite loops.
2388 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2389 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2390 * them to initialized or split the described range from larger unwritten
2391 * extent. Note that we need not map all the described range since allocation
2392 * can return less blocks or the range is covered by more unwritten extents. We
2393 * cannot map more because we are limited by reserved transaction credits. On
2394 * the other hand we always make sure that the last touched page is fully
2395 * mapped so that it can be written out (and thus forward progress is
2396 * guaranteed). After mapping we submit all mapped pages for IO.
2398 static int mpage_map_and_submit_extent(handle_t *handle,
2399 struct mpage_da_data *mpd,
2400 bool *give_up_on_write)
2402 struct inode *inode = mpd->inode;
2403 struct ext4_map_blocks *map = &mpd->map;
2407 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2408 struct ext4_io_end_vec *io_end_vec = ext4_alloc_io_end_vec(io_end);
2410 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2412 err = mpage_map_one_extent(handle, mpd);
2414 struct super_block *sb = inode->i_sb;
2416 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2417 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2418 goto invalidate_dirty_pages;
2420 * Let the uper layers retry transient errors.
2421 * In the case of ENOSPC, if ext4_count_free_blocks()
2422 * is non-zero, a commit should free up blocks.
2424 if ((err == -ENOMEM) ||
2425 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2427 goto update_disksize;
2430 ext4_msg(sb, KERN_CRIT,
2431 "Delayed block allocation failed for "
2432 "inode %lu at logical offset %llu with"
2433 " max blocks %u with error %d",
2435 (unsigned long long)map->m_lblk,
2436 (unsigned)map->m_len, -err);
2437 ext4_msg(sb, KERN_CRIT,
2438 "This should not happen!! Data will "
2441 ext4_print_free_blocks(inode);
2442 invalidate_dirty_pages:
2443 *give_up_on_write = true;
2448 * Update buffer state, submit mapped pages, and get us new
2451 err = mpage_map_and_submit_buffers(mpd);
2453 goto update_disksize;
2454 } while (map->m_len);
2458 * Update on-disk size after IO is submitted. Races with
2459 * truncate are avoided by checking i_size under i_data_sem.
2461 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2462 if (disksize > EXT4_I(inode)->i_disksize) {
2466 down_write(&EXT4_I(inode)->i_data_sem);
2467 i_size = i_size_read(inode);
2468 if (disksize > i_size)
2470 if (disksize > EXT4_I(inode)->i_disksize)
2471 EXT4_I(inode)->i_disksize = disksize;
2472 up_write(&EXT4_I(inode)->i_data_sem);
2473 err2 = ext4_mark_inode_dirty(handle, inode);
2475 ext4_error(inode->i_sb,
2476 "Failed to mark inode %lu dirty",
2485 * Calculate the total number of credits to reserve for one writepages
2486 * iteration. This is called from ext4_writepages(). We map an extent of
2487 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2488 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2489 * bpp - 1 blocks in bpp different extents.
2491 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2493 int bpp = ext4_journal_blocks_per_page(inode);
2495 return ext4_meta_trans_blocks(inode,
2496 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2500 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2501 * and underlying extent to map
2503 * @mpd - where to look for pages
2505 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2506 * IO immediately. When we find a page which isn't mapped we start accumulating
2507 * extent of buffers underlying these pages that needs mapping (formed by
2508 * either delayed or unwritten buffers). We also lock the pages containing
2509 * these buffers. The extent found is returned in @mpd structure (starting at
2510 * mpd->lblk with length mpd->len blocks).
2512 * Note that this function can attach bios to one io_end structure which are
2513 * neither logically nor physically contiguous. Although it may seem as an
2514 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2515 * case as we need to track IO to all buffers underlying a page in one io_end.
2517 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2519 struct address_space *mapping = mpd->inode->i_mapping;
2520 struct pagevec pvec;
2521 unsigned int nr_pages;
2522 long left = mpd->wbc->nr_to_write;
2523 pgoff_t index = mpd->first_page;
2524 pgoff_t end = mpd->last_page;
2527 int blkbits = mpd->inode->i_blkbits;
2529 struct buffer_head *head;
2531 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2532 tag = PAGECACHE_TAG_TOWRITE;
2534 tag = PAGECACHE_TAG_DIRTY;
2536 pagevec_init(&pvec);
2538 mpd->next_page = index;
2539 while (index <= end) {
2540 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2545 for (i = 0; i < nr_pages; i++) {
2546 struct page *page = pvec.pages[i];
2549 * Accumulated enough dirty pages? This doesn't apply
2550 * to WB_SYNC_ALL mode. For integrity sync we have to
2551 * keep going because someone may be concurrently
2552 * dirtying pages, and we might have synced a lot of
2553 * newly appeared dirty pages, but have not synced all
2554 * of the old dirty pages.
2556 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2559 /* If we can't merge this page, we are done. */
2560 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2565 * If the page is no longer dirty, or its mapping no
2566 * longer corresponds to inode we are writing (which
2567 * means it has been truncated or invalidated), or the
2568 * page is already under writeback and we are not doing
2569 * a data integrity writeback, skip the page
2571 if (!PageDirty(page) ||
2572 (PageWriteback(page) &&
2573 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2574 unlikely(page->mapping != mapping)) {
2579 wait_on_page_writeback(page);
2580 BUG_ON(PageWriteback(page));
2582 if (mpd->map.m_len == 0)
2583 mpd->first_page = page->index;
2584 mpd->next_page = page->index + 1;
2585 /* Add all dirty buffers to mpd */
2586 lblk = ((ext4_lblk_t)page->index) <<
2587 (PAGE_SHIFT - blkbits);
2588 head = page_buffers(page);
2589 err = mpage_process_page_bufs(mpd, head, head, lblk);
2595 pagevec_release(&pvec);
2600 pagevec_release(&pvec);
2604 static int ext4_writepages(struct address_space *mapping,
2605 struct writeback_control *wbc)
2607 pgoff_t writeback_index = 0;
2608 long nr_to_write = wbc->nr_to_write;
2609 int range_whole = 0;
2611 handle_t *handle = NULL;
2612 struct mpage_da_data mpd;
2613 struct inode *inode = mapping->host;
2614 int needed_blocks, rsv_blocks = 0, ret = 0;
2615 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2617 struct blk_plug plug;
2618 bool give_up_on_write = false;
2620 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2623 percpu_down_read(&sbi->s_journal_flag_rwsem);
2624 trace_ext4_writepages(inode, wbc);
2627 * No pages to write? This is mainly a kludge to avoid starting
2628 * a transaction for special inodes like journal inode on last iput()
2629 * because that could violate lock ordering on umount
2631 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2632 goto out_writepages;
2634 if (ext4_should_journal_data(inode)) {
2635 ret = generic_writepages(mapping, wbc);
2636 goto out_writepages;
2640 * If the filesystem has aborted, it is read-only, so return
2641 * right away instead of dumping stack traces later on that
2642 * will obscure the real source of the problem. We test
2643 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2644 * the latter could be true if the filesystem is mounted
2645 * read-only, and in that case, ext4_writepages should
2646 * *never* be called, so if that ever happens, we would want
2649 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2650 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2652 goto out_writepages;
2656 * If we have inline data and arrive here, it means that
2657 * we will soon create the block for the 1st page, so
2658 * we'd better clear the inline data here.
2660 if (ext4_has_inline_data(inode)) {
2661 /* Just inode will be modified... */
2662 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2663 if (IS_ERR(handle)) {
2664 ret = PTR_ERR(handle);
2665 goto out_writepages;
2667 BUG_ON(ext4_test_inode_state(inode,
2668 EXT4_STATE_MAY_INLINE_DATA));
2669 ext4_destroy_inline_data(handle, inode);
2670 ext4_journal_stop(handle);
2673 if (ext4_should_dioread_nolock(inode)) {
2675 * We may need to convert up to one extent per block in
2676 * the page and we may dirty the inode.
2678 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2679 PAGE_SIZE >> inode->i_blkbits);
2682 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2685 if (wbc->range_cyclic) {
2686 writeback_index = mapping->writeback_index;
2687 if (writeback_index)
2689 mpd.first_page = writeback_index;
2692 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2693 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2698 ext4_io_submit_init(&mpd.io_submit, wbc);
2700 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2701 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2703 blk_start_plug(&plug);
2706 * First writeback pages that don't need mapping - we can avoid
2707 * starting a transaction unnecessarily and also avoid being blocked
2708 * in the block layer on device congestion while having transaction
2712 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2713 if (!mpd.io_submit.io_end) {
2717 ret = mpage_prepare_extent_to_map(&mpd);
2718 /* Unlock pages we didn't use */
2719 mpage_release_unused_pages(&mpd, false);
2720 /* Submit prepared bio */
2721 ext4_io_submit(&mpd.io_submit);
2722 ext4_put_io_end_defer(mpd.io_submit.io_end);
2723 mpd.io_submit.io_end = NULL;
2727 while (!done && mpd.first_page <= mpd.last_page) {
2728 /* For each extent of pages we use new io_end */
2729 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2730 if (!mpd.io_submit.io_end) {
2736 * We have two constraints: We find one extent to map and we
2737 * must always write out whole page (makes a difference when
2738 * blocksize < pagesize) so that we don't block on IO when we
2739 * try to write out the rest of the page. Journalled mode is
2740 * not supported by delalloc.
2742 BUG_ON(ext4_should_journal_data(inode));
2743 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2745 /* start a new transaction */
2746 handle = ext4_journal_start_with_reserve(inode,
2747 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2748 if (IS_ERR(handle)) {
2749 ret = PTR_ERR(handle);
2750 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2751 "%ld pages, ino %lu; err %d", __func__,
2752 wbc->nr_to_write, inode->i_ino, ret);
2753 /* Release allocated io_end */
2754 ext4_put_io_end(mpd.io_submit.io_end);
2755 mpd.io_submit.io_end = NULL;
2760 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2761 ret = mpage_prepare_extent_to_map(&mpd);
2764 ret = mpage_map_and_submit_extent(handle, &mpd,
2768 * We scanned the whole range (or exhausted
2769 * nr_to_write), submitted what was mapped and
2770 * didn't find anything needing mapping. We are
2777 * Caution: If the handle is synchronous,
2778 * ext4_journal_stop() can wait for transaction commit
2779 * to finish which may depend on writeback of pages to
2780 * complete or on page lock to be released. In that
2781 * case, we have to wait until after after we have
2782 * submitted all the IO, released page locks we hold,
2783 * and dropped io_end reference (for extent conversion
2784 * to be able to complete) before stopping the handle.
2786 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2787 ext4_journal_stop(handle);
2791 /* Unlock pages we didn't use */
2792 mpage_release_unused_pages(&mpd, give_up_on_write);
2793 /* Submit prepared bio */
2794 ext4_io_submit(&mpd.io_submit);
2797 * Drop our io_end reference we got from init. We have
2798 * to be careful and use deferred io_end finishing if
2799 * we are still holding the transaction as we can
2800 * release the last reference to io_end which may end
2801 * up doing unwritten extent conversion.
2804 ext4_put_io_end_defer(mpd.io_submit.io_end);
2805 ext4_journal_stop(handle);
2807 ext4_put_io_end(mpd.io_submit.io_end);
2808 mpd.io_submit.io_end = NULL;
2810 if (ret == -ENOSPC && sbi->s_journal) {
2812 * Commit the transaction which would
2813 * free blocks released in the transaction
2816 jbd2_journal_force_commit_nested(sbi->s_journal);
2820 /* Fatal error - ENOMEM, EIO... */
2825 blk_finish_plug(&plug);
2826 if (!ret && !cycled && wbc->nr_to_write > 0) {
2828 mpd.last_page = writeback_index - 1;
2834 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2836 * Set the writeback_index so that range_cyclic
2837 * mode will write it back later
2839 mapping->writeback_index = mpd.first_page;
2842 trace_ext4_writepages_result(inode, wbc, ret,
2843 nr_to_write - wbc->nr_to_write);
2844 percpu_up_read(&sbi->s_journal_flag_rwsem);
2848 static int ext4_dax_writepages(struct address_space *mapping,
2849 struct writeback_control *wbc)
2852 long nr_to_write = wbc->nr_to_write;
2853 struct inode *inode = mapping->host;
2854 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2856 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2859 percpu_down_read(&sbi->s_journal_flag_rwsem);
2860 trace_ext4_writepages(inode, wbc);
2862 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, wbc);
2863 trace_ext4_writepages_result(inode, wbc, ret,
2864 nr_to_write - wbc->nr_to_write);
2865 percpu_up_read(&sbi->s_journal_flag_rwsem);
2869 static int ext4_nonda_switch(struct super_block *sb)
2871 s64 free_clusters, dirty_clusters;
2872 struct ext4_sb_info *sbi = EXT4_SB(sb);
2875 * switch to non delalloc mode if we are running low
2876 * on free block. The free block accounting via percpu
2877 * counters can get slightly wrong with percpu_counter_batch getting
2878 * accumulated on each CPU without updating global counters
2879 * Delalloc need an accurate free block accounting. So switch
2880 * to non delalloc when we are near to error range.
2883 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2885 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2887 * Start pushing delalloc when 1/2 of free blocks are dirty.
2889 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2890 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2892 if (2 * free_clusters < 3 * dirty_clusters ||
2893 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2895 * free block count is less than 150% of dirty blocks
2896 * or free blocks is less than watermark
2903 /* We always reserve for an inode update; the superblock could be there too */
2904 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2906 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2909 if (pos + len <= 0x7fffffffULL)
2912 /* We might need to update the superblock to set LARGE_FILE */
2916 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2917 loff_t pos, unsigned len, unsigned flags,
2918 struct page **pagep, void **fsdata)
2920 int ret, retries = 0;
2923 struct inode *inode = mapping->host;
2926 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2929 index = pos >> PAGE_SHIFT;
2931 if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) ||
2932 ext4_verity_in_progress(inode)) {
2933 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2934 return ext4_write_begin(file, mapping, pos,
2935 len, flags, pagep, fsdata);
2937 *fsdata = (void *)0;
2938 trace_ext4_da_write_begin(inode, pos, len, flags);
2940 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2941 ret = ext4_da_write_inline_data_begin(mapping, inode,
2951 * grab_cache_page_write_begin() can take a long time if the
2952 * system is thrashing due to memory pressure, or if the page
2953 * is being written back. So grab it first before we start
2954 * the transaction handle. This also allows us to allocate
2955 * the page (if needed) without using GFP_NOFS.
2958 page = grab_cache_page_write_begin(mapping, index, flags);
2964 * With delayed allocation, we don't log the i_disksize update
2965 * if there is delayed block allocation. But we still need
2966 * to journalling the i_disksize update if writes to the end
2967 * of file which has an already mapped buffer.
2970 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2971 ext4_da_write_credits(inode, pos, len));
2972 if (IS_ERR(handle)) {
2974 return PTR_ERR(handle);
2978 if (page->mapping != mapping) {
2979 /* The page got truncated from under us */
2982 ext4_journal_stop(handle);
2985 /* In case writeback began while the page was unlocked */
2986 wait_for_stable_page(page);
2988 #ifdef CONFIG_FS_ENCRYPTION
2989 ret = ext4_block_write_begin(page, pos, len,
2990 ext4_da_get_block_prep);
2992 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2996 ext4_journal_stop(handle);
2998 * block_write_begin may have instantiated a few blocks
2999 * outside i_size. Trim these off again. Don't need
3000 * i_size_read because we hold i_mutex.
3002 if (pos + len > inode->i_size)
3003 ext4_truncate_failed_write(inode);
3005 if (ret == -ENOSPC &&
3006 ext4_should_retry_alloc(inode->i_sb, &retries))
3018 * Check if we should update i_disksize
3019 * when write to the end of file but not require block allocation
3021 static int ext4_da_should_update_i_disksize(struct page *page,
3022 unsigned long offset)
3024 struct buffer_head *bh;
3025 struct inode *inode = page->mapping->host;
3029 bh = page_buffers(page);
3030 idx = offset >> inode->i_blkbits;
3032 for (i = 0; i < idx; i++)
3033 bh = bh->b_this_page;
3035 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3040 static int ext4_da_write_end(struct file *file,
3041 struct address_space *mapping,
3042 loff_t pos, unsigned len, unsigned copied,
3043 struct page *page, void *fsdata)
3045 struct inode *inode = mapping->host;
3047 handle_t *handle = ext4_journal_current_handle();
3049 unsigned long start, end;
3050 int write_mode = (int)(unsigned long)fsdata;
3052 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3053 return ext4_write_end(file, mapping, pos,
3054 len, copied, page, fsdata);
3056 trace_ext4_da_write_end(inode, pos, len, copied);
3057 start = pos & (PAGE_SIZE - 1);
3058 end = start + copied - 1;
3061 * generic_write_end() will run mark_inode_dirty() if i_size
3062 * changes. So let's piggyback the i_disksize mark_inode_dirty
3065 new_i_size = pos + copied;
3066 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3067 if (ext4_has_inline_data(inode) ||
3068 ext4_da_should_update_i_disksize(page, end)) {
3069 ext4_update_i_disksize(inode, new_i_size);
3070 /* We need to mark inode dirty even if
3071 * new_i_size is less that inode->i_size
3072 * bu greater than i_disksize.(hint delalloc)
3074 ext4_mark_inode_dirty(handle, inode);
3078 if (write_mode != CONVERT_INLINE_DATA &&
3079 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3080 ext4_has_inline_data(inode))
3081 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3084 ret2 = generic_write_end(file, mapping, pos, len, copied,
3090 ret2 = ext4_journal_stop(handle);
3094 return ret ? ret : copied;
3098 * Force all delayed allocation blocks to be allocated for a given inode.
3100 int ext4_alloc_da_blocks(struct inode *inode)
3102 trace_ext4_alloc_da_blocks(inode);
3104 if (!EXT4_I(inode)->i_reserved_data_blocks)
3108 * We do something simple for now. The filemap_flush() will
3109 * also start triggering a write of the data blocks, which is
3110 * not strictly speaking necessary (and for users of
3111 * laptop_mode, not even desirable). However, to do otherwise
3112 * would require replicating code paths in:
3114 * ext4_writepages() ->
3115 * write_cache_pages() ---> (via passed in callback function)
3116 * __mpage_da_writepage() -->
3117 * mpage_add_bh_to_extent()
3118 * mpage_da_map_blocks()
3120 * The problem is that write_cache_pages(), located in
3121 * mm/page-writeback.c, marks pages clean in preparation for
3122 * doing I/O, which is not desirable if we're not planning on
3125 * We could call write_cache_pages(), and then redirty all of
3126 * the pages by calling redirty_page_for_writepage() but that
3127 * would be ugly in the extreme. So instead we would need to
3128 * replicate parts of the code in the above functions,
3129 * simplifying them because we wouldn't actually intend to
3130 * write out the pages, but rather only collect contiguous
3131 * logical block extents, call the multi-block allocator, and
3132 * then update the buffer heads with the block allocations.
3134 * For now, though, we'll cheat by calling filemap_flush(),
3135 * which will map the blocks, and start the I/O, but not
3136 * actually wait for the I/O to complete.
3138 return filemap_flush(inode->i_mapping);
3142 * bmap() is special. It gets used by applications such as lilo and by
3143 * the swapper to find the on-disk block of a specific piece of data.
3145 * Naturally, this is dangerous if the block concerned is still in the
3146 * journal. If somebody makes a swapfile on an ext4 data-journaling
3147 * filesystem and enables swap, then they may get a nasty shock when the
3148 * data getting swapped to that swapfile suddenly gets overwritten by
3149 * the original zero's written out previously to the journal and
3150 * awaiting writeback in the kernel's buffer cache.
3152 * So, if we see any bmap calls here on a modified, data-journaled file,
3153 * take extra steps to flush any blocks which might be in the cache.
3155 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3157 struct inode *inode = mapping->host;
3162 * We can get here for an inline file via the FIBMAP ioctl
3164 if (ext4_has_inline_data(inode))
3167 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3168 test_opt(inode->i_sb, DELALLOC)) {
3170 * With delalloc we want to sync the file
3171 * so that we can make sure we allocate
3174 filemap_write_and_wait(mapping);
3177 if (EXT4_JOURNAL(inode) &&
3178 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3180 * This is a REALLY heavyweight approach, but the use of
3181 * bmap on dirty files is expected to be extremely rare:
3182 * only if we run lilo or swapon on a freshly made file
3183 * do we expect this to happen.
3185 * (bmap requires CAP_SYS_RAWIO so this does not
3186 * represent an unprivileged user DOS attack --- we'd be
3187 * in trouble if mortal users could trigger this path at
3190 * NB. EXT4_STATE_JDATA is not set on files other than
3191 * regular files. If somebody wants to bmap a directory
3192 * or symlink and gets confused because the buffer
3193 * hasn't yet been flushed to disk, they deserve
3194 * everything they get.
3197 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3198 journal = EXT4_JOURNAL(inode);
3199 jbd2_journal_lock_updates(journal);
3200 err = jbd2_journal_flush(journal);
3201 jbd2_journal_unlock_updates(journal);
3207 return generic_block_bmap(mapping, block, ext4_get_block);
3210 static int ext4_readpage(struct file *file, struct page *page)
3213 struct inode *inode = page->mapping->host;
3215 trace_ext4_readpage(page);
3217 if (ext4_has_inline_data(inode))
3218 ret = ext4_readpage_inline(inode, page);
3221 return ext4_mpage_readpages(page->mapping, NULL, page, 1,
3228 ext4_readpages(struct file *file, struct address_space *mapping,
3229 struct list_head *pages, unsigned nr_pages)
3231 struct inode *inode = mapping->host;
3233 /* If the file has inline data, no need to do readpages. */
3234 if (ext4_has_inline_data(inode))
3237 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages, true);
3240 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3241 unsigned int length)
3243 trace_ext4_invalidatepage(page, offset, length);
3245 /* No journalling happens on data buffers when this function is used */
3246 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3248 block_invalidatepage(page, offset, length);
3251 static int __ext4_journalled_invalidatepage(struct page *page,
3252 unsigned int offset,
3253 unsigned int length)
3255 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3257 trace_ext4_journalled_invalidatepage(page, offset, length);
3260 * If it's a full truncate we just forget about the pending dirtying
3262 if (offset == 0 && length == PAGE_SIZE)
3263 ClearPageChecked(page);
3265 return jbd2_journal_invalidatepage(journal, page, offset, length);
3268 /* Wrapper for aops... */
3269 static void ext4_journalled_invalidatepage(struct page *page,
3270 unsigned int offset,
3271 unsigned int length)
3273 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3276 static int ext4_releasepage(struct page *page, gfp_t wait)
3278 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3280 trace_ext4_releasepage(page);
3282 /* Page has dirty journalled data -> cannot release */
3283 if (PageChecked(page))
3286 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3288 return try_to_free_buffers(page);
3291 static bool ext4_inode_datasync_dirty(struct inode *inode)
3293 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3296 return !jbd2_transaction_committed(journal,
3297 EXT4_I(inode)->i_datasync_tid);
3298 /* Any metadata buffers to write? */
3299 if (!list_empty(&inode->i_mapping->private_list))
3301 return inode->i_state & I_DIRTY_DATASYNC;
3304 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3305 struct ext4_map_blocks *map, loff_t offset,
3308 u8 blkbits = inode->i_blkbits;
3311 * Writes that span EOF might trigger an I/O size update on completion,
3312 * so consider them to be dirty for the purpose of O_DSYNC, even if
3313 * there is no other metadata changes being made or are pending.
3316 if (ext4_inode_datasync_dirty(inode) ||
3317 offset + length > i_size_read(inode))
3318 iomap->flags |= IOMAP_F_DIRTY;
3320 if (map->m_flags & EXT4_MAP_NEW)
3321 iomap->flags |= IOMAP_F_NEW;
3323 iomap->bdev = inode->i_sb->s_bdev;
3324 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3325 iomap->offset = (u64) map->m_lblk << blkbits;
3326 iomap->length = (u64) map->m_len << blkbits;
3329 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3330 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3331 * set. In order for any allocated unwritten extents to be converted
3332 * into written extents correctly within the ->end_io() handler, we
3333 * need to ensure that the iomap->type is set appropriately. Hence, the
3334 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3337 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3338 iomap->type = IOMAP_UNWRITTEN;
3339 iomap->addr = (u64) map->m_pblk << blkbits;
3340 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3341 iomap->type = IOMAP_MAPPED;
3342 iomap->addr = (u64) map->m_pblk << blkbits;
3344 iomap->type = IOMAP_HOLE;
3345 iomap->addr = IOMAP_NULL_ADDR;
3349 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3353 u8 blkbits = inode->i_blkbits;
3354 int ret, dio_credits, m_flags = 0, retries = 0;
3357 * Trim the mapping request to the maximum value that we can map at
3358 * once for direct I/O.
3360 if (map->m_len > DIO_MAX_BLOCKS)
3361 map->m_len = DIO_MAX_BLOCKS;
3362 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3366 * Either we allocate blocks and then don't get an unwritten extent, so
3367 * in that case we have reserved enough credits. Or, the blocks are
3368 * already allocated and unwritten. In that case, the extent conversion
3369 * fits into the credits as well.
3371 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3373 return PTR_ERR(handle);
3376 * DAX and direct I/O are the only two operations that are currently
3377 * supported with IOMAP_WRITE.
3379 WARN_ON(!IS_DAX(inode) && !(flags & IOMAP_DIRECT));
3381 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3383 * We use i_size instead of i_disksize here because delalloc writeback
3384 * can complete at any point during the I/O and subsequently push the
3385 * i_disksize out to i_size. This could be beyond where direct I/O is
3386 * happening and thus expose allocated blocks to direct I/O reads.
3388 else if ((map->m_lblk * (1 << blkbits)) >= i_size_read(inode))
3389 m_flags = EXT4_GET_BLOCKS_CREATE;
3390 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3391 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3393 ret = ext4_map_blocks(handle, inode, map, m_flags);
3396 * We cannot fill holes in indirect tree based inodes as that could
3397 * expose stale data in the case of a crash. Use the magic error code
3398 * to fallback to buffered I/O.
3400 if (!m_flags && !ret)
3403 ext4_journal_stop(handle);
3404 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3411 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3412 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3415 struct ext4_map_blocks map;
3416 u8 blkbits = inode->i_blkbits;
3418 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3421 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3425 * Calculate the first and last logical blocks respectively.
3427 map.m_lblk = offset >> blkbits;
3428 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3429 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3431 if (flags & IOMAP_WRITE)
3432 ret = ext4_iomap_alloc(inode, &map, flags);
3434 ret = ext4_map_blocks(NULL, inode, &map, 0);
3439 ext4_set_iomap(inode, iomap, &map, offset, length);
3444 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3445 ssize_t written, unsigned flags, struct iomap *iomap)
3448 * Check to see whether an error occurred while writing out the data to
3449 * the allocated blocks. If so, return the magic error code so that we
3450 * fallback to buffered I/O and attempt to complete the remainder of
3451 * the I/O. Any blocks that may have been allocated in preparation for
3452 * the direct I/O will be reused during buffered I/O.
3454 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3460 const struct iomap_ops ext4_iomap_ops = {
3461 .iomap_begin = ext4_iomap_begin,
3462 .iomap_end = ext4_iomap_end,
3465 static bool ext4_iomap_is_delalloc(struct inode *inode,
3466 struct ext4_map_blocks *map)
3468 struct extent_status es;
3469 ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3471 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3472 map->m_lblk, end, &es);
3474 if (!es.es_len || es.es_lblk > end)
3477 if (es.es_lblk > map->m_lblk) {
3478 map->m_len = es.es_lblk - map->m_lblk;
3482 offset = map->m_lblk - es.es_lblk;
3483 map->m_len = es.es_len - offset;
3488 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3489 loff_t length, unsigned int flags,
3490 struct iomap *iomap, struct iomap *srcmap)
3493 bool delalloc = false;
3494 struct ext4_map_blocks map;
3495 u8 blkbits = inode->i_blkbits;
3497 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3500 if (ext4_has_inline_data(inode)) {
3501 ret = ext4_inline_data_iomap(inode, iomap);
3502 if (ret != -EAGAIN) {
3503 if (ret == 0 && offset >= iomap->length)
3510 * Calculate the first and last logical block respectively.
3512 map.m_lblk = offset >> blkbits;
3513 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3514 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3516 ret = ext4_map_blocks(NULL, inode, &map, 0);
3520 delalloc = ext4_iomap_is_delalloc(inode, &map);
3522 ext4_set_iomap(inode, iomap, &map, offset, length);
3523 if (delalloc && iomap->type == IOMAP_HOLE)
3524 iomap->type = IOMAP_DELALLOC;
3529 const struct iomap_ops ext4_iomap_report_ops = {
3530 .iomap_begin = ext4_iomap_begin_report,
3534 * Pages can be marked dirty completely asynchronously from ext4's journalling
3535 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3536 * much here because ->set_page_dirty is called under VFS locks. The page is
3537 * not necessarily locked.
3539 * We cannot just dirty the page and leave attached buffers clean, because the
3540 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3541 * or jbddirty because all the journalling code will explode.
3543 * So what we do is to mark the page "pending dirty" and next time writepage
3544 * is called, propagate that into the buffers appropriately.
3546 static int ext4_journalled_set_page_dirty(struct page *page)
3548 SetPageChecked(page);
3549 return __set_page_dirty_nobuffers(page);
3552 static int ext4_set_page_dirty(struct page *page)
3554 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3555 WARN_ON_ONCE(!page_has_buffers(page));
3556 return __set_page_dirty_buffers(page);
3559 static const struct address_space_operations ext4_aops = {
3560 .readpage = ext4_readpage,
3561 .readpages = ext4_readpages,
3562 .writepage = ext4_writepage,
3563 .writepages = ext4_writepages,
3564 .write_begin = ext4_write_begin,
3565 .write_end = ext4_write_end,
3566 .set_page_dirty = ext4_set_page_dirty,
3568 .invalidatepage = ext4_invalidatepage,
3569 .releasepage = ext4_releasepage,
3570 .direct_IO = noop_direct_IO,
3571 .migratepage = buffer_migrate_page,
3572 .is_partially_uptodate = block_is_partially_uptodate,
3573 .error_remove_page = generic_error_remove_page,
3576 static const struct address_space_operations ext4_journalled_aops = {
3577 .readpage = ext4_readpage,
3578 .readpages = ext4_readpages,
3579 .writepage = ext4_writepage,
3580 .writepages = ext4_writepages,
3581 .write_begin = ext4_write_begin,
3582 .write_end = ext4_journalled_write_end,
3583 .set_page_dirty = ext4_journalled_set_page_dirty,
3585 .invalidatepage = ext4_journalled_invalidatepage,
3586 .releasepage = ext4_releasepage,
3587 .direct_IO = noop_direct_IO,
3588 .is_partially_uptodate = block_is_partially_uptodate,
3589 .error_remove_page = generic_error_remove_page,
3592 static const struct address_space_operations ext4_da_aops = {
3593 .readpage = ext4_readpage,
3594 .readpages = ext4_readpages,
3595 .writepage = ext4_writepage,
3596 .writepages = ext4_writepages,
3597 .write_begin = ext4_da_write_begin,
3598 .write_end = ext4_da_write_end,
3599 .set_page_dirty = ext4_set_page_dirty,
3601 .invalidatepage = ext4_invalidatepage,
3602 .releasepage = ext4_releasepage,
3603 .direct_IO = noop_direct_IO,
3604 .migratepage = buffer_migrate_page,
3605 .is_partially_uptodate = block_is_partially_uptodate,
3606 .error_remove_page = generic_error_remove_page,
3609 static const struct address_space_operations ext4_dax_aops = {
3610 .writepages = ext4_dax_writepages,
3611 .direct_IO = noop_direct_IO,
3612 .set_page_dirty = noop_set_page_dirty,
3614 .invalidatepage = noop_invalidatepage,
3617 void ext4_set_aops(struct inode *inode)
3619 switch (ext4_inode_journal_mode(inode)) {
3620 case EXT4_INODE_ORDERED_DATA_MODE:
3621 case EXT4_INODE_WRITEBACK_DATA_MODE:
3623 case EXT4_INODE_JOURNAL_DATA_MODE:
3624 inode->i_mapping->a_ops = &ext4_journalled_aops;
3630 inode->i_mapping->a_ops = &ext4_dax_aops;
3631 else if (test_opt(inode->i_sb, DELALLOC))
3632 inode->i_mapping->a_ops = &ext4_da_aops;
3634 inode->i_mapping->a_ops = &ext4_aops;
3637 static int __ext4_block_zero_page_range(handle_t *handle,
3638 struct address_space *mapping, loff_t from, loff_t length)
3640 ext4_fsblk_t index = from >> PAGE_SHIFT;
3641 unsigned offset = from & (PAGE_SIZE-1);
3642 unsigned blocksize, pos;
3644 struct inode *inode = mapping->host;
3645 struct buffer_head *bh;
3649 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3650 mapping_gfp_constraint(mapping, ~__GFP_FS));
3654 blocksize = inode->i_sb->s_blocksize;
3656 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3658 if (!page_has_buffers(page))
3659 create_empty_buffers(page, blocksize, 0);
3661 /* Find the buffer that contains "offset" */
3662 bh = page_buffers(page);
3664 while (offset >= pos) {
3665 bh = bh->b_this_page;
3669 if (buffer_freed(bh)) {
3670 BUFFER_TRACE(bh, "freed: skip");
3673 if (!buffer_mapped(bh)) {
3674 BUFFER_TRACE(bh, "unmapped");
3675 ext4_get_block(inode, iblock, bh, 0);
3676 /* unmapped? It's a hole - nothing to do */
3677 if (!buffer_mapped(bh)) {
3678 BUFFER_TRACE(bh, "still unmapped");
3683 /* Ok, it's mapped. Make sure it's up-to-date */
3684 if (PageUptodate(page))
3685 set_buffer_uptodate(bh);
3687 if (!buffer_uptodate(bh)) {
3689 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
3691 /* Uhhuh. Read error. Complain and punt. */
3692 if (!buffer_uptodate(bh))
3694 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
3695 /* We expect the key to be set. */
3696 BUG_ON(!fscrypt_has_encryption_key(inode));
3697 WARN_ON_ONCE(fscrypt_decrypt_pagecache_blocks(
3698 page, blocksize, bh_offset(bh)));
3701 if (ext4_should_journal_data(inode)) {
3702 BUFFER_TRACE(bh, "get write access");
3703 err = ext4_journal_get_write_access(handle, bh);
3707 zero_user(page, offset, length);
3708 BUFFER_TRACE(bh, "zeroed end of block");
3710 if (ext4_should_journal_data(inode)) {
3711 err = ext4_handle_dirty_metadata(handle, inode, bh);
3714 mark_buffer_dirty(bh);
3715 if (ext4_should_order_data(inode))
3716 err = ext4_jbd2_inode_add_write(handle, inode, from,
3727 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3728 * starting from file offset 'from'. The range to be zero'd must
3729 * be contained with in one block. If the specified range exceeds
3730 * the end of the block it will be shortened to end of the block
3731 * that cooresponds to 'from'
3733 static int ext4_block_zero_page_range(handle_t *handle,
3734 struct address_space *mapping, loff_t from, loff_t length)
3736 struct inode *inode = mapping->host;
3737 unsigned offset = from & (PAGE_SIZE-1);
3738 unsigned blocksize = inode->i_sb->s_blocksize;
3739 unsigned max = blocksize - (offset & (blocksize - 1));
3742 * correct length if it does not fall between
3743 * 'from' and the end of the block
3745 if (length > max || length < 0)
3748 if (IS_DAX(inode)) {
3749 return iomap_zero_range(inode, from, length, NULL,
3752 return __ext4_block_zero_page_range(handle, mapping, from, length);
3756 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3757 * up to the end of the block which corresponds to `from'.
3758 * This required during truncate. We need to physically zero the tail end
3759 * of that block so it doesn't yield old data if the file is later grown.
3761 static int ext4_block_truncate_page(handle_t *handle,
3762 struct address_space *mapping, loff_t from)
3764 unsigned offset = from & (PAGE_SIZE-1);
3767 struct inode *inode = mapping->host;
3769 /* If we are processing an encrypted inode during orphan list handling */
3770 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3773 blocksize = inode->i_sb->s_blocksize;
3774 length = blocksize - (offset & (blocksize - 1));
3776 return ext4_block_zero_page_range(handle, mapping, from, length);
3779 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3780 loff_t lstart, loff_t length)
3782 struct super_block *sb = inode->i_sb;
3783 struct address_space *mapping = inode->i_mapping;
3784 unsigned partial_start, partial_end;
3785 ext4_fsblk_t start, end;
3786 loff_t byte_end = (lstart + length - 1);
3789 partial_start = lstart & (sb->s_blocksize - 1);
3790 partial_end = byte_end & (sb->s_blocksize - 1);
3792 start = lstart >> sb->s_blocksize_bits;
3793 end = byte_end >> sb->s_blocksize_bits;
3795 /* Handle partial zero within the single block */
3797 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3798 err = ext4_block_zero_page_range(handle, mapping,
3802 /* Handle partial zero out on the start of the range */
3803 if (partial_start) {
3804 err = ext4_block_zero_page_range(handle, mapping,
3805 lstart, sb->s_blocksize);
3809 /* Handle partial zero out on the end of the range */
3810 if (partial_end != sb->s_blocksize - 1)
3811 err = ext4_block_zero_page_range(handle, mapping,
3812 byte_end - partial_end,
3817 int ext4_can_truncate(struct inode *inode)
3819 if (S_ISREG(inode->i_mode))
3821 if (S_ISDIR(inode->i_mode))
3823 if (S_ISLNK(inode->i_mode))
3824 return !ext4_inode_is_fast_symlink(inode);
3829 * We have to make sure i_disksize gets properly updated before we truncate
3830 * page cache due to hole punching or zero range. Otherwise i_disksize update
3831 * can get lost as it may have been postponed to submission of writeback but
3832 * that will never happen after we truncate page cache.
3834 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3838 loff_t size = i_size_read(inode);
3840 WARN_ON(!inode_is_locked(inode));
3841 if (offset > size || offset + len < size)
3844 if (EXT4_I(inode)->i_disksize >= size)
3847 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3849 return PTR_ERR(handle);
3850 ext4_update_i_disksize(inode, size);
3851 ext4_mark_inode_dirty(handle, inode);
3852 ext4_journal_stop(handle);
3857 static void ext4_wait_dax_page(struct ext4_inode_info *ei)
3859 up_write(&ei->i_mmap_sem);
3861 down_write(&ei->i_mmap_sem);
3864 int ext4_break_layouts(struct inode *inode)
3866 struct ext4_inode_info *ei = EXT4_I(inode);
3870 if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
3874 page = dax_layout_busy_page(inode->i_mapping);
3878 error = ___wait_var_event(&page->_refcount,
3879 atomic_read(&page->_refcount) == 1,
3880 TASK_INTERRUPTIBLE, 0, 0,
3881 ext4_wait_dax_page(ei));
3882 } while (error == 0);
3888 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3889 * associated with the given offset and length
3891 * @inode: File inode
3892 * @offset: The offset where the hole will begin
3893 * @len: The length of the hole
3895 * Returns: 0 on success or negative on failure
3898 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3900 struct super_block *sb = inode->i_sb;
3901 ext4_lblk_t first_block, stop_block;
3902 struct address_space *mapping = inode->i_mapping;
3903 loff_t first_block_offset, last_block_offset;
3905 unsigned int credits;
3908 if (!S_ISREG(inode->i_mode))
3911 trace_ext4_punch_hole(inode, offset, length, 0);
3913 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
3914 if (ext4_has_inline_data(inode)) {
3915 down_write(&EXT4_I(inode)->i_mmap_sem);
3916 ret = ext4_convert_inline_data(inode);
3917 up_write(&EXT4_I(inode)->i_mmap_sem);
3923 * Write out all dirty pages to avoid race conditions
3924 * Then release them.
3926 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3927 ret = filemap_write_and_wait_range(mapping, offset,
3928 offset + length - 1);
3935 /* No need to punch hole beyond i_size */
3936 if (offset >= inode->i_size)
3940 * If the hole extends beyond i_size, set the hole
3941 * to end after the page that contains i_size
3943 if (offset + length > inode->i_size) {
3944 length = inode->i_size +
3945 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3949 if (offset & (sb->s_blocksize - 1) ||
3950 (offset + length) & (sb->s_blocksize - 1)) {
3952 * Attach jinode to inode for jbd2 if we do any zeroing of
3955 ret = ext4_inode_attach_jinode(inode);
3961 /* Wait all existing dio workers, newcomers will block on i_mutex */
3962 inode_dio_wait(inode);
3965 * Prevent page faults from reinstantiating pages we have released from
3968 down_write(&EXT4_I(inode)->i_mmap_sem);
3970 ret = ext4_break_layouts(inode);
3974 first_block_offset = round_up(offset, sb->s_blocksize);
3975 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3977 /* Now release the pages and zero block aligned part of pages*/
3978 if (last_block_offset > first_block_offset) {
3979 ret = ext4_update_disksize_before_punch(inode, offset, length);
3982 truncate_pagecache_range(inode, first_block_offset,
3986 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3987 credits = ext4_writepage_trans_blocks(inode);
3989 credits = ext4_blocks_for_truncate(inode);
3990 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3991 if (IS_ERR(handle)) {
3992 ret = PTR_ERR(handle);
3993 ext4_std_error(sb, ret);
3997 ret = ext4_zero_partial_blocks(handle, inode, offset,
4002 first_block = (offset + sb->s_blocksize - 1) >>
4003 EXT4_BLOCK_SIZE_BITS(sb);
4004 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4006 /* If there are blocks to remove, do it */
4007 if (stop_block > first_block) {
4009 down_write(&EXT4_I(inode)->i_data_sem);
4010 ext4_discard_preallocations(inode);
4012 ret = ext4_es_remove_extent(inode, first_block,
4013 stop_block - first_block);
4015 up_write(&EXT4_I(inode)->i_data_sem);
4019 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4020 ret = ext4_ext_remove_space(inode, first_block,
4023 ret = ext4_ind_remove_space(handle, inode, first_block,
4026 up_write(&EXT4_I(inode)->i_data_sem);
4029 ext4_handle_sync(handle);
4031 inode->i_mtime = inode->i_ctime = current_time(inode);
4032 ext4_mark_inode_dirty(handle, inode);
4034 ext4_update_inode_fsync_trans(handle, inode, 1);
4036 ext4_journal_stop(handle);
4038 up_write(&EXT4_I(inode)->i_mmap_sem);
4040 inode_unlock(inode);
4044 int ext4_inode_attach_jinode(struct inode *inode)
4046 struct ext4_inode_info *ei = EXT4_I(inode);
4047 struct jbd2_inode *jinode;
4049 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4052 jinode = jbd2_alloc_inode(GFP_KERNEL);
4053 spin_lock(&inode->i_lock);
4056 spin_unlock(&inode->i_lock);
4059 ei->jinode = jinode;
4060 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4063 spin_unlock(&inode->i_lock);
4064 if (unlikely(jinode != NULL))
4065 jbd2_free_inode(jinode);
4072 * We block out ext4_get_block() block instantiations across the entire
4073 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4074 * simultaneously on behalf of the same inode.
4076 * As we work through the truncate and commit bits of it to the journal there
4077 * is one core, guiding principle: the file's tree must always be consistent on
4078 * disk. We must be able to restart the truncate after a crash.
4080 * The file's tree may be transiently inconsistent in memory (although it
4081 * probably isn't), but whenever we close off and commit a journal transaction,
4082 * the contents of (the filesystem + the journal) must be consistent and
4083 * restartable. It's pretty simple, really: bottom up, right to left (although
4084 * left-to-right works OK too).
4086 * Note that at recovery time, journal replay occurs *before* the restart of
4087 * truncate against the orphan inode list.
4089 * The committed inode has the new, desired i_size (which is the same as
4090 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4091 * that this inode's truncate did not complete and it will again call
4092 * ext4_truncate() to have another go. So there will be instantiated blocks
4093 * to the right of the truncation point in a crashed ext4 filesystem. But
4094 * that's fine - as long as they are linked from the inode, the post-crash
4095 * ext4_truncate() run will find them and release them.
4097 int ext4_truncate(struct inode *inode)
4099 struct ext4_inode_info *ei = EXT4_I(inode);
4100 unsigned int credits;
4103 struct address_space *mapping = inode->i_mapping;
4106 * There is a possibility that we're either freeing the inode
4107 * or it's a completely new inode. In those cases we might not
4108 * have i_mutex locked because it's not necessary.
4110 if (!(inode->i_state & (I_NEW|I_FREEING)))
4111 WARN_ON(!inode_is_locked(inode));
4112 trace_ext4_truncate_enter(inode);
4114 if (!ext4_can_truncate(inode))
4117 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4119 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4120 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4122 if (ext4_has_inline_data(inode)) {
4125 err = ext4_inline_data_truncate(inode, &has_inline);
4132 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4133 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4134 if (ext4_inode_attach_jinode(inode) < 0)
4138 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4139 credits = ext4_writepage_trans_blocks(inode);
4141 credits = ext4_blocks_for_truncate(inode);
4143 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4145 return PTR_ERR(handle);
4147 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4148 ext4_block_truncate_page(handle, mapping, inode->i_size);
4151 * We add the inode to the orphan list, so that if this
4152 * truncate spans multiple transactions, and we crash, we will
4153 * resume the truncate when the filesystem recovers. It also
4154 * marks the inode dirty, to catch the new size.
4156 * Implication: the file must always be in a sane, consistent
4157 * truncatable state while each transaction commits.
4159 err = ext4_orphan_add(handle, inode);
4163 down_write(&EXT4_I(inode)->i_data_sem);
4165 ext4_discard_preallocations(inode);
4167 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4168 err = ext4_ext_truncate(handle, inode);
4170 ext4_ind_truncate(handle, inode);
4172 up_write(&ei->i_data_sem);
4177 ext4_handle_sync(handle);
4181 * If this was a simple ftruncate() and the file will remain alive,
4182 * then we need to clear up the orphan record which we created above.
4183 * However, if this was a real unlink then we were called by
4184 * ext4_evict_inode(), and we allow that function to clean up the
4185 * orphan info for us.
4188 ext4_orphan_del(handle, inode);
4190 inode->i_mtime = inode->i_ctime = current_time(inode);
4191 ext4_mark_inode_dirty(handle, inode);
4192 ext4_journal_stop(handle);
4194 trace_ext4_truncate_exit(inode);
4199 * ext4_get_inode_loc returns with an extra refcount against the inode's
4200 * underlying buffer_head on success. If 'in_mem' is true, we have all
4201 * data in memory that is needed to recreate the on-disk version of this
4204 static int __ext4_get_inode_loc(struct inode *inode,
4205 struct ext4_iloc *iloc, int in_mem)
4207 struct ext4_group_desc *gdp;
4208 struct buffer_head *bh;
4209 struct super_block *sb = inode->i_sb;
4211 struct blk_plug plug;
4212 int inodes_per_block, inode_offset;
4215 if (inode->i_ino < EXT4_ROOT_INO ||
4216 inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4217 return -EFSCORRUPTED;
4219 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4220 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4225 * Figure out the offset within the block group inode table
4227 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4228 inode_offset = ((inode->i_ino - 1) %
4229 EXT4_INODES_PER_GROUP(sb));
4230 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4231 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4233 bh = sb_getblk(sb, block);
4236 if (!buffer_uptodate(bh)) {
4240 * If the buffer has the write error flag, we have failed
4241 * to write out another inode in the same block. In this
4242 * case, we don't have to read the block because we may
4243 * read the old inode data successfully.
4245 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4246 set_buffer_uptodate(bh);
4248 if (buffer_uptodate(bh)) {
4249 /* someone brought it uptodate while we waited */
4255 * If we have all information of the inode in memory and this
4256 * is the only valid inode in the block, we need not read the
4260 struct buffer_head *bitmap_bh;
4263 start = inode_offset & ~(inodes_per_block - 1);
4265 /* Is the inode bitmap in cache? */
4266 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4267 if (unlikely(!bitmap_bh))
4271 * If the inode bitmap isn't in cache then the
4272 * optimisation may end up performing two reads instead
4273 * of one, so skip it.
4275 if (!buffer_uptodate(bitmap_bh)) {
4279 for (i = start; i < start + inodes_per_block; i++) {
4280 if (i == inode_offset)
4282 if (ext4_test_bit(i, bitmap_bh->b_data))
4286 if (i == start + inodes_per_block) {
4287 /* all other inodes are free, so skip I/O */
4288 memset(bh->b_data, 0, bh->b_size);
4289 set_buffer_uptodate(bh);
4297 * If we need to do any I/O, try to pre-readahead extra
4298 * blocks from the inode table.
4300 blk_start_plug(&plug);
4301 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4302 ext4_fsblk_t b, end, table;
4304 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4306 table = ext4_inode_table(sb, gdp);
4307 /* s_inode_readahead_blks is always a power of 2 */
4308 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4312 num = EXT4_INODES_PER_GROUP(sb);
4313 if (ext4_has_group_desc_csum(sb))
4314 num -= ext4_itable_unused_count(sb, gdp);
4315 table += num / inodes_per_block;
4319 sb_breadahead(sb, b++);
4323 * There are other valid inodes in the buffer, this inode
4324 * has in-inode xattrs, or we don't have this inode in memory.
4325 * Read the block from disk.
4327 trace_ext4_load_inode(inode);
4329 bh->b_end_io = end_buffer_read_sync;
4330 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4331 blk_finish_plug(&plug);
4333 if (!buffer_uptodate(bh)) {
4334 EXT4_ERROR_INODE_BLOCK(inode, block,
4335 "unable to read itable block");
4345 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4347 /* We have all inode data except xattrs in memory here. */
4348 return __ext4_get_inode_loc(inode, iloc,
4349 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4352 static bool ext4_should_use_dax(struct inode *inode)
4354 if (!test_opt(inode->i_sb, DAX))
4356 if (!S_ISREG(inode->i_mode))
4358 if (ext4_should_journal_data(inode))
4360 if (ext4_has_inline_data(inode))
4362 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4364 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4369 void ext4_set_inode_flags(struct inode *inode)
4371 unsigned int flags = EXT4_I(inode)->i_flags;
4372 unsigned int new_fl = 0;
4374 if (flags & EXT4_SYNC_FL)
4376 if (flags & EXT4_APPEND_FL)
4378 if (flags & EXT4_IMMUTABLE_FL)
4379 new_fl |= S_IMMUTABLE;
4380 if (flags & EXT4_NOATIME_FL)
4381 new_fl |= S_NOATIME;
4382 if (flags & EXT4_DIRSYNC_FL)
4383 new_fl |= S_DIRSYNC;
4384 if (ext4_should_use_dax(inode))
4386 if (flags & EXT4_ENCRYPT_FL)
4387 new_fl |= S_ENCRYPTED;
4388 if (flags & EXT4_CASEFOLD_FL)
4389 new_fl |= S_CASEFOLD;
4390 if (flags & EXT4_VERITY_FL)
4392 inode_set_flags(inode, new_fl,
4393 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4394 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4397 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4398 struct ext4_inode_info *ei)
4401 struct inode *inode = &(ei->vfs_inode);
4402 struct super_block *sb = inode->i_sb;
4404 if (ext4_has_feature_huge_file(sb)) {
4405 /* we are using combined 48 bit field */
4406 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4407 le32_to_cpu(raw_inode->i_blocks_lo);
4408 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4409 /* i_blocks represent file system block size */
4410 return i_blocks << (inode->i_blkbits - 9);
4415 return le32_to_cpu(raw_inode->i_blocks_lo);
4419 static inline int ext4_iget_extra_inode(struct inode *inode,
4420 struct ext4_inode *raw_inode,
4421 struct ext4_inode_info *ei)
4423 __le32 *magic = (void *)raw_inode +
4424 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4426 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4427 EXT4_INODE_SIZE(inode->i_sb) &&
4428 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4429 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4430 return ext4_find_inline_data_nolock(inode);
4432 EXT4_I(inode)->i_inline_off = 0;
4436 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4438 if (!ext4_has_feature_project(inode->i_sb))
4440 *projid = EXT4_I(inode)->i_projid;
4445 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4446 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4449 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4451 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4452 inode_set_iversion_raw(inode, val);
4454 inode_set_iversion_queried(inode, val);
4456 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4458 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4459 return inode_peek_iversion_raw(inode);
4461 return inode_peek_iversion(inode);
4464 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4465 ext4_iget_flags flags, const char *function,
4468 struct ext4_iloc iloc;
4469 struct ext4_inode *raw_inode;
4470 struct ext4_inode_info *ei;
4471 struct inode *inode;
4472 journal_t *journal = EXT4_SB(sb)->s_journal;
4480 if ((!(flags & EXT4_IGET_SPECIAL) &&
4481 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4482 (ino < EXT4_ROOT_INO) ||
4483 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4484 if (flags & EXT4_IGET_HANDLE)
4485 return ERR_PTR(-ESTALE);
4486 __ext4_error(sb, function, line,
4487 "inode #%lu: comm %s: iget: illegal inode #",
4488 ino, current->comm);
4489 return ERR_PTR(-EFSCORRUPTED);
4492 inode = iget_locked(sb, ino);
4494 return ERR_PTR(-ENOMEM);
4495 if (!(inode->i_state & I_NEW))
4501 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4504 raw_inode = ext4_raw_inode(&iloc);
4506 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4507 ext4_error_inode(inode, function, line, 0,
4508 "iget: root inode unallocated");
4509 ret = -EFSCORRUPTED;
4513 if ((flags & EXT4_IGET_HANDLE) &&
4514 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4519 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4520 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4521 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4522 EXT4_INODE_SIZE(inode->i_sb) ||
4523 (ei->i_extra_isize & 3)) {
4524 ext4_error_inode(inode, function, line, 0,
4525 "iget: bad extra_isize %u "
4528 EXT4_INODE_SIZE(inode->i_sb));
4529 ret = -EFSCORRUPTED;
4533 ei->i_extra_isize = 0;
4535 /* Precompute checksum seed for inode metadata */
4536 if (ext4_has_metadata_csum(sb)) {
4537 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4539 __le32 inum = cpu_to_le32(inode->i_ino);
4540 __le32 gen = raw_inode->i_generation;
4541 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4543 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4547 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4548 ext4_error_inode(inode, function, line, 0,
4549 "iget: checksum invalid");
4554 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4555 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4556 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4557 if (ext4_has_feature_project(sb) &&
4558 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4559 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4560 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4562 i_projid = EXT4_DEF_PROJID;
4564 if (!(test_opt(inode->i_sb, NO_UID32))) {
4565 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4566 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4568 i_uid_write(inode, i_uid);
4569 i_gid_write(inode, i_gid);
4570 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4571 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4573 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4574 ei->i_inline_off = 0;
4575 ei->i_dir_start_lookup = 0;
4576 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4577 /* We now have enough fields to check if the inode was active or not.
4578 * This is needed because nfsd might try to access dead inodes
4579 * the test is that same one that e2fsck uses
4580 * NeilBrown 1999oct15
4582 if (inode->i_nlink == 0) {
4583 if ((inode->i_mode == 0 ||
4584 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4585 ino != EXT4_BOOT_LOADER_INO) {
4586 /* this inode is deleted */
4590 /* The only unlinked inodes we let through here have
4591 * valid i_mode and are being read by the orphan
4592 * recovery code: that's fine, we're about to complete
4593 * the process of deleting those.
4594 * OR it is the EXT4_BOOT_LOADER_INO which is
4595 * not initialized on a new filesystem. */
4597 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4598 ext4_set_inode_flags(inode);
4599 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4600 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4601 if (ext4_has_feature_64bit(sb))
4603 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4604 inode->i_size = ext4_isize(sb, raw_inode);
4605 if ((size = i_size_read(inode)) < 0) {
4606 ext4_error_inode(inode, function, line, 0,
4607 "iget: bad i_size value: %lld", size);
4608 ret = -EFSCORRUPTED;
4611 ei->i_disksize = inode->i_size;
4613 ei->i_reserved_quota = 0;
4615 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4616 ei->i_block_group = iloc.block_group;
4617 ei->i_last_alloc_group = ~0;
4619 * NOTE! The in-memory inode i_data array is in little-endian order
4620 * even on big-endian machines: we do NOT byteswap the block numbers!
4622 for (block = 0; block < EXT4_N_BLOCKS; block++)
4623 ei->i_data[block] = raw_inode->i_block[block];
4624 INIT_LIST_HEAD(&ei->i_orphan);
4627 * Set transaction id's of transactions that have to be committed
4628 * to finish f[data]sync. We set them to currently running transaction
4629 * as we cannot be sure that the inode or some of its metadata isn't
4630 * part of the transaction - the inode could have been reclaimed and
4631 * now it is reread from disk.
4634 transaction_t *transaction;
4637 read_lock(&journal->j_state_lock);
4638 if (journal->j_running_transaction)
4639 transaction = journal->j_running_transaction;
4641 transaction = journal->j_committing_transaction;
4643 tid = transaction->t_tid;
4645 tid = journal->j_commit_sequence;
4646 read_unlock(&journal->j_state_lock);
4647 ei->i_sync_tid = tid;
4648 ei->i_datasync_tid = tid;
4651 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4652 if (ei->i_extra_isize == 0) {
4653 /* The extra space is currently unused. Use it. */
4654 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4655 ei->i_extra_isize = sizeof(struct ext4_inode) -
4656 EXT4_GOOD_OLD_INODE_SIZE;
4658 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4664 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4665 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4666 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4667 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4669 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4670 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4672 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4673 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4675 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4677 ext4_inode_set_iversion_queried(inode, ivers);
4681 if (ei->i_file_acl &&
4682 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4683 ext4_error_inode(inode, function, line, 0,
4684 "iget: bad extended attribute block %llu",
4686 ret = -EFSCORRUPTED;
4688 } else if (!ext4_has_inline_data(inode)) {
4689 /* validate the block references in the inode */
4690 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4691 (S_ISLNK(inode->i_mode) &&
4692 !ext4_inode_is_fast_symlink(inode))) {
4693 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4694 ret = ext4_ext_check_inode(inode);
4696 ret = ext4_ind_check_inode(inode);
4702 if (S_ISREG(inode->i_mode)) {
4703 inode->i_op = &ext4_file_inode_operations;
4704 inode->i_fop = &ext4_file_operations;
4705 ext4_set_aops(inode);
4706 } else if (S_ISDIR(inode->i_mode)) {
4707 inode->i_op = &ext4_dir_inode_operations;
4708 inode->i_fop = &ext4_dir_operations;
4709 } else if (S_ISLNK(inode->i_mode)) {
4710 /* VFS does not allow setting these so must be corruption */
4711 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4712 ext4_error_inode(inode, function, line, 0,
4713 "iget: immutable or append flags "
4714 "not allowed on symlinks");
4715 ret = -EFSCORRUPTED;
4718 if (IS_ENCRYPTED(inode)) {
4719 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4720 ext4_set_aops(inode);
4721 } else if (ext4_inode_is_fast_symlink(inode)) {
4722 inode->i_link = (char *)ei->i_data;
4723 inode->i_op = &ext4_fast_symlink_inode_operations;
4724 nd_terminate_link(ei->i_data, inode->i_size,
4725 sizeof(ei->i_data) - 1);
4727 inode->i_op = &ext4_symlink_inode_operations;
4728 ext4_set_aops(inode);
4730 inode_nohighmem(inode);
4731 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4732 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4733 inode->i_op = &ext4_special_inode_operations;
4734 if (raw_inode->i_block[0])
4735 init_special_inode(inode, inode->i_mode,
4736 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4738 init_special_inode(inode, inode->i_mode,
4739 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4740 } else if (ino == EXT4_BOOT_LOADER_INO) {
4741 make_bad_inode(inode);
4743 ret = -EFSCORRUPTED;
4744 ext4_error_inode(inode, function, line, 0,
4745 "iget: bogus i_mode (%o)", inode->i_mode);
4748 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
4749 ext4_error_inode(inode, function, line, 0,
4750 "casefold flag without casefold feature");
4753 unlock_new_inode(inode);
4759 return ERR_PTR(ret);
4762 static int ext4_inode_blocks_set(handle_t *handle,
4763 struct ext4_inode *raw_inode,
4764 struct ext4_inode_info *ei)
4766 struct inode *inode = &(ei->vfs_inode);
4767 u64 i_blocks = inode->i_blocks;
4768 struct super_block *sb = inode->i_sb;
4770 if (i_blocks <= ~0U) {
4772 * i_blocks can be represented in a 32 bit variable
4773 * as multiple of 512 bytes
4775 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4776 raw_inode->i_blocks_high = 0;
4777 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4780 if (!ext4_has_feature_huge_file(sb))
4783 if (i_blocks <= 0xffffffffffffULL) {
4785 * i_blocks can be represented in a 48 bit variable
4786 * as multiple of 512 bytes
4788 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4789 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4790 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4792 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4793 /* i_block is stored in file system block size */
4794 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4795 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4796 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4801 struct other_inode {
4802 unsigned long orig_ino;
4803 struct ext4_inode *raw_inode;
4806 static int other_inode_match(struct inode * inode, unsigned long ino,
4809 struct other_inode *oi = (struct other_inode *) data;
4811 if ((inode->i_ino != ino) ||
4812 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4814 ((inode->i_state & I_DIRTY_TIME) == 0))
4816 spin_lock(&inode->i_lock);
4817 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4818 I_DIRTY_INODE)) == 0) &&
4819 (inode->i_state & I_DIRTY_TIME)) {
4820 struct ext4_inode_info *ei = EXT4_I(inode);
4822 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4823 spin_unlock(&inode->i_lock);
4825 spin_lock(&ei->i_raw_lock);
4826 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4827 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4828 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4829 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4830 spin_unlock(&ei->i_raw_lock);
4831 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4834 spin_unlock(&inode->i_lock);
4839 * Opportunistically update the other time fields for other inodes in
4840 * the same inode table block.
4842 static void ext4_update_other_inodes_time(struct super_block *sb,
4843 unsigned long orig_ino, char *buf)
4845 struct other_inode oi;
4847 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4848 int inode_size = EXT4_INODE_SIZE(sb);
4850 oi.orig_ino = orig_ino;
4852 * Calculate the first inode in the inode table block. Inode
4853 * numbers are one-based. That is, the first inode in a block
4854 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4856 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4857 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4858 if (ino == orig_ino)
4860 oi.raw_inode = (struct ext4_inode *) buf;
4861 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4866 * Post the struct inode info into an on-disk inode location in the
4867 * buffer-cache. This gobbles the caller's reference to the
4868 * buffer_head in the inode location struct.
4870 * The caller must have write access to iloc->bh.
4872 static int ext4_do_update_inode(handle_t *handle,
4873 struct inode *inode,
4874 struct ext4_iloc *iloc)
4876 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4877 struct ext4_inode_info *ei = EXT4_I(inode);
4878 struct buffer_head *bh = iloc->bh;
4879 struct super_block *sb = inode->i_sb;
4880 int err = 0, rc, block;
4881 int need_datasync = 0, set_large_file = 0;
4886 spin_lock(&ei->i_raw_lock);
4888 /* For fields not tracked in the in-memory inode,
4889 * initialise them to zero for new inodes. */
4890 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4891 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4893 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4894 i_uid = i_uid_read(inode);
4895 i_gid = i_gid_read(inode);
4896 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4897 if (!(test_opt(inode->i_sb, NO_UID32))) {
4898 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4899 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4901 * Fix up interoperability with old kernels. Otherwise, old inodes get
4902 * re-used with the upper 16 bits of the uid/gid intact
4904 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4905 raw_inode->i_uid_high = 0;
4906 raw_inode->i_gid_high = 0;
4908 raw_inode->i_uid_high =
4909 cpu_to_le16(high_16_bits(i_uid));
4910 raw_inode->i_gid_high =
4911 cpu_to_le16(high_16_bits(i_gid));
4914 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4915 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4916 raw_inode->i_uid_high = 0;
4917 raw_inode->i_gid_high = 0;
4919 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4921 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4922 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4923 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4924 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4926 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4928 spin_unlock(&ei->i_raw_lock);
4931 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4932 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4933 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4934 raw_inode->i_file_acl_high =
4935 cpu_to_le16(ei->i_file_acl >> 32);
4936 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4937 if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
4938 ext4_isize_set(raw_inode, ei->i_disksize);
4941 if (ei->i_disksize > 0x7fffffffULL) {
4942 if (!ext4_has_feature_large_file(sb) ||
4943 EXT4_SB(sb)->s_es->s_rev_level ==
4944 cpu_to_le32(EXT4_GOOD_OLD_REV))
4947 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4948 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4949 if (old_valid_dev(inode->i_rdev)) {
4950 raw_inode->i_block[0] =
4951 cpu_to_le32(old_encode_dev(inode->i_rdev));
4952 raw_inode->i_block[1] = 0;
4954 raw_inode->i_block[0] = 0;
4955 raw_inode->i_block[1] =
4956 cpu_to_le32(new_encode_dev(inode->i_rdev));
4957 raw_inode->i_block[2] = 0;
4959 } else if (!ext4_has_inline_data(inode)) {
4960 for (block = 0; block < EXT4_N_BLOCKS; block++)
4961 raw_inode->i_block[block] = ei->i_data[block];
4964 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4965 u64 ivers = ext4_inode_peek_iversion(inode);
4967 raw_inode->i_disk_version = cpu_to_le32(ivers);
4968 if (ei->i_extra_isize) {
4969 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4970 raw_inode->i_version_hi =
4971 cpu_to_le32(ivers >> 32);
4972 raw_inode->i_extra_isize =
4973 cpu_to_le16(ei->i_extra_isize);
4977 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
4978 i_projid != EXT4_DEF_PROJID);
4980 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4981 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4982 raw_inode->i_projid = cpu_to_le32(i_projid);
4984 ext4_inode_csum_set(inode, raw_inode, ei);
4985 spin_unlock(&ei->i_raw_lock);
4986 if (inode->i_sb->s_flags & SB_LAZYTIME)
4987 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4990 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4991 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4994 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4995 if (set_large_file) {
4996 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4997 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5000 ext4_set_feature_large_file(sb);
5001 ext4_handle_sync(handle);
5002 err = ext4_handle_dirty_super(handle, sb);
5004 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5007 ext4_std_error(inode->i_sb, err);
5012 * ext4_write_inode()
5014 * We are called from a few places:
5016 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5017 * Here, there will be no transaction running. We wait for any running
5018 * transaction to commit.
5020 * - Within flush work (sys_sync(), kupdate and such).
5021 * We wait on commit, if told to.
5023 * - Within iput_final() -> write_inode_now()
5024 * We wait on commit, if told to.
5026 * In all cases it is actually safe for us to return without doing anything,
5027 * because the inode has been copied into a raw inode buffer in
5028 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5031 * Note that we are absolutely dependent upon all inode dirtiers doing the
5032 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5033 * which we are interested.
5035 * It would be a bug for them to not do this. The code:
5037 * mark_inode_dirty(inode)
5039 * inode->i_size = expr;
5041 * is in error because write_inode() could occur while `stuff()' is running,
5042 * and the new i_size will be lost. Plus the inode will no longer be on the
5043 * superblock's dirty inode list.
5045 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5049 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5050 sb_rdonly(inode->i_sb))
5053 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5056 if (EXT4_SB(inode->i_sb)->s_journal) {
5057 if (ext4_journal_current_handle()) {
5058 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5064 * No need to force transaction in WB_SYNC_NONE mode. Also
5065 * ext4_sync_fs() will force the commit after everything is
5068 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5071 err = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5072 EXT4_I(inode)->i_sync_tid);
5074 struct ext4_iloc iloc;
5076 err = __ext4_get_inode_loc(inode, &iloc, 0);
5080 * sync(2) will flush the whole buffer cache. No need to do
5081 * it here separately for each inode.
5083 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5084 sync_dirty_buffer(iloc.bh);
5085 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5086 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5087 "IO error syncing inode");
5096 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5097 * buffers that are attached to a page stradding i_size and are undergoing
5098 * commit. In that case we have to wait for commit to finish and try again.
5100 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5104 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5105 tid_t commit_tid = 0;
5108 offset = inode->i_size & (PAGE_SIZE - 1);
5110 * All buffers in the last page remain valid? Then there's nothing to
5111 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5114 if (offset > PAGE_SIZE - i_blocksize(inode))
5117 page = find_lock_page(inode->i_mapping,
5118 inode->i_size >> PAGE_SHIFT);
5121 ret = __ext4_journalled_invalidatepage(page, offset,
5122 PAGE_SIZE - offset);
5128 read_lock(&journal->j_state_lock);
5129 if (journal->j_committing_transaction)
5130 commit_tid = journal->j_committing_transaction->t_tid;
5131 read_unlock(&journal->j_state_lock);
5133 jbd2_log_wait_commit(journal, commit_tid);
5140 * Called from notify_change.
5142 * We want to trap VFS attempts to truncate the file as soon as
5143 * possible. In particular, we want to make sure that when the VFS
5144 * shrinks i_size, we put the inode on the orphan list and modify
5145 * i_disksize immediately, so that during the subsequent flushing of
5146 * dirty pages and freeing of disk blocks, we can guarantee that any
5147 * commit will leave the blocks being flushed in an unused state on
5148 * disk. (On recovery, the inode will get truncated and the blocks will
5149 * be freed, so we have a strong guarantee that no future commit will
5150 * leave these blocks visible to the user.)
5152 * Another thing we have to assure is that if we are in ordered mode
5153 * and inode is still attached to the committing transaction, we must
5154 * we start writeout of all the dirty pages which are being truncated.
5155 * This way we are sure that all the data written in the previous
5156 * transaction are already on disk (truncate waits for pages under
5159 * Called with inode->i_mutex down.
5161 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5163 struct inode *inode = d_inode(dentry);
5166 const unsigned int ia_valid = attr->ia_valid;
5168 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5171 if (unlikely(IS_IMMUTABLE(inode)))
5174 if (unlikely(IS_APPEND(inode) &&
5175 (ia_valid & (ATTR_MODE | ATTR_UID |
5176 ATTR_GID | ATTR_TIMES_SET))))
5179 error = setattr_prepare(dentry, attr);
5183 error = fscrypt_prepare_setattr(dentry, attr);
5187 error = fsverity_prepare_setattr(dentry, attr);
5191 if (is_quota_modification(inode, attr)) {
5192 error = dquot_initialize(inode);
5196 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5197 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5200 /* (user+group)*(old+new) structure, inode write (sb,
5201 * inode block, ? - but truncate inode update has it) */
5202 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5203 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5204 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5205 if (IS_ERR(handle)) {
5206 error = PTR_ERR(handle);
5210 /* dquot_transfer() calls back ext4_get_inode_usage() which
5211 * counts xattr inode references.
5213 down_read(&EXT4_I(inode)->xattr_sem);
5214 error = dquot_transfer(inode, attr);
5215 up_read(&EXT4_I(inode)->xattr_sem);
5218 ext4_journal_stop(handle);
5221 /* Update corresponding info in inode so that everything is in
5222 * one transaction */
5223 if (attr->ia_valid & ATTR_UID)
5224 inode->i_uid = attr->ia_uid;
5225 if (attr->ia_valid & ATTR_GID)
5226 inode->i_gid = attr->ia_gid;
5227 error = ext4_mark_inode_dirty(handle, inode);
5228 ext4_journal_stop(handle);
5231 if (attr->ia_valid & ATTR_SIZE) {
5233 loff_t oldsize = inode->i_size;
5234 int shrink = (attr->ia_size < inode->i_size);
5236 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5237 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5239 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5242 if (!S_ISREG(inode->i_mode))
5245 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5246 inode_inc_iversion(inode);
5249 if (ext4_should_order_data(inode)) {
5250 error = ext4_begin_ordered_truncate(inode,
5256 * Blocks are going to be removed from the inode. Wait
5257 * for dio in flight.
5259 inode_dio_wait(inode);
5262 down_write(&EXT4_I(inode)->i_mmap_sem);
5264 rc = ext4_break_layouts(inode);
5266 up_write(&EXT4_I(inode)->i_mmap_sem);
5270 if (attr->ia_size != inode->i_size) {
5271 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5272 if (IS_ERR(handle)) {
5273 error = PTR_ERR(handle);
5276 if (ext4_handle_valid(handle) && shrink) {
5277 error = ext4_orphan_add(handle, inode);
5281 * Update c/mtime on truncate up, ext4_truncate() will
5282 * update c/mtime in shrink case below
5285 inode->i_mtime = current_time(inode);
5286 inode->i_ctime = inode->i_mtime;
5288 down_write(&EXT4_I(inode)->i_data_sem);
5289 EXT4_I(inode)->i_disksize = attr->ia_size;
5290 rc = ext4_mark_inode_dirty(handle, inode);
5294 * We have to update i_size under i_data_sem together
5295 * with i_disksize to avoid races with writeback code
5296 * running ext4_wb_update_i_disksize().
5299 i_size_write(inode, attr->ia_size);
5300 up_write(&EXT4_I(inode)->i_data_sem);
5301 ext4_journal_stop(handle);
5305 pagecache_isize_extended(inode, oldsize,
5307 } else if (ext4_should_journal_data(inode)) {
5308 ext4_wait_for_tail_page_commit(inode);
5313 * Truncate pagecache after we've waited for commit
5314 * in data=journal mode to make pages freeable.
5316 truncate_pagecache(inode, inode->i_size);
5318 * Call ext4_truncate() even if i_size didn't change to
5319 * truncate possible preallocated blocks.
5321 if (attr->ia_size <= oldsize) {
5322 rc = ext4_truncate(inode);
5327 up_write(&EXT4_I(inode)->i_mmap_sem);
5331 setattr_copy(inode, attr);
5332 mark_inode_dirty(inode);
5336 * If the call to ext4_truncate failed to get a transaction handle at
5337 * all, we need to clean up the in-core orphan list manually.
5339 if (orphan && inode->i_nlink)
5340 ext4_orphan_del(NULL, inode);
5342 if (!error && (ia_valid & ATTR_MODE))
5343 rc = posix_acl_chmod(inode, inode->i_mode);
5346 ext4_std_error(inode->i_sb, error);
5352 int ext4_getattr(const struct path *path, struct kstat *stat,
5353 u32 request_mask, unsigned int query_flags)
5355 struct inode *inode = d_inode(path->dentry);
5356 struct ext4_inode *raw_inode;
5357 struct ext4_inode_info *ei = EXT4_I(inode);
5360 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5361 stat->result_mask |= STATX_BTIME;
5362 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5363 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5366 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5367 if (flags & EXT4_APPEND_FL)
5368 stat->attributes |= STATX_ATTR_APPEND;
5369 if (flags & EXT4_COMPR_FL)
5370 stat->attributes |= STATX_ATTR_COMPRESSED;
5371 if (flags & EXT4_ENCRYPT_FL)
5372 stat->attributes |= STATX_ATTR_ENCRYPTED;
5373 if (flags & EXT4_IMMUTABLE_FL)
5374 stat->attributes |= STATX_ATTR_IMMUTABLE;
5375 if (flags & EXT4_NODUMP_FL)
5376 stat->attributes |= STATX_ATTR_NODUMP;
5378 stat->attributes_mask |= (STATX_ATTR_APPEND |
5379 STATX_ATTR_COMPRESSED |
5380 STATX_ATTR_ENCRYPTED |
5381 STATX_ATTR_IMMUTABLE |
5384 generic_fillattr(inode, stat);
5388 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5389 u32 request_mask, unsigned int query_flags)
5391 struct inode *inode = d_inode(path->dentry);
5392 u64 delalloc_blocks;
5394 ext4_getattr(path, stat, request_mask, query_flags);
5397 * If there is inline data in the inode, the inode will normally not
5398 * have data blocks allocated (it may have an external xattr block).
5399 * Report at least one sector for such files, so tools like tar, rsync,
5400 * others don't incorrectly think the file is completely sparse.
5402 if (unlikely(ext4_has_inline_data(inode)))
5403 stat->blocks += (stat->size + 511) >> 9;
5406 * We can't update i_blocks if the block allocation is delayed
5407 * otherwise in the case of system crash before the real block
5408 * allocation is done, we will have i_blocks inconsistent with
5409 * on-disk file blocks.
5410 * We always keep i_blocks updated together with real
5411 * allocation. But to not confuse with user, stat
5412 * will return the blocks that include the delayed allocation
5413 * blocks for this file.
5415 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5416 EXT4_I(inode)->i_reserved_data_blocks);
5417 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5421 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5424 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5425 return ext4_ind_trans_blocks(inode, lblocks);
5426 return ext4_ext_index_trans_blocks(inode, pextents);
5430 * Account for index blocks, block groups bitmaps and block group
5431 * descriptor blocks if modify datablocks and index blocks
5432 * worse case, the indexs blocks spread over different block groups
5434 * If datablocks are discontiguous, they are possible to spread over
5435 * different block groups too. If they are contiguous, with flexbg,
5436 * they could still across block group boundary.
5438 * Also account for superblock, inode, quota and xattr blocks
5440 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5443 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5449 * How many index blocks need to touch to map @lblocks logical blocks
5450 * to @pextents physical extents?
5452 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5457 * Now let's see how many group bitmaps and group descriptors need
5460 groups = idxblocks + pextents;
5462 if (groups > ngroups)
5464 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5465 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5467 /* bitmaps and block group descriptor blocks */
5468 ret += groups + gdpblocks;
5470 /* Blocks for super block, inode, quota and xattr blocks */
5471 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5477 * Calculate the total number of credits to reserve to fit
5478 * the modification of a single pages into a single transaction,
5479 * which may include multiple chunks of block allocations.
5481 * This could be called via ext4_write_begin()
5483 * We need to consider the worse case, when
5484 * one new block per extent.
5486 int ext4_writepage_trans_blocks(struct inode *inode)
5488 int bpp = ext4_journal_blocks_per_page(inode);
5491 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5493 /* Account for data blocks for journalled mode */
5494 if (ext4_should_journal_data(inode))
5500 * Calculate the journal credits for a chunk of data modification.
5502 * This is called from DIO, fallocate or whoever calling
5503 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5505 * journal buffers for data blocks are not included here, as DIO
5506 * and fallocate do no need to journal data buffers.
5508 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5510 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5514 * The caller must have previously called ext4_reserve_inode_write().
5515 * Give this, we know that the caller already has write access to iloc->bh.
5517 int ext4_mark_iloc_dirty(handle_t *handle,
5518 struct inode *inode, struct ext4_iloc *iloc)
5522 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5526 if (IS_I_VERSION(inode))
5527 inode_inc_iversion(inode);
5529 /* the do_update_inode consumes one bh->b_count */
5532 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5533 err = ext4_do_update_inode(handle, inode, iloc);
5539 * On success, We end up with an outstanding reference count against
5540 * iloc->bh. This _must_ be cleaned up later.
5544 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5545 struct ext4_iloc *iloc)
5549 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5552 err = ext4_get_inode_loc(inode, iloc);
5554 BUFFER_TRACE(iloc->bh, "get_write_access");
5555 err = ext4_journal_get_write_access(handle, iloc->bh);
5561 ext4_std_error(inode->i_sb, err);
5565 static int __ext4_expand_extra_isize(struct inode *inode,
5566 unsigned int new_extra_isize,
5567 struct ext4_iloc *iloc,
5568 handle_t *handle, int *no_expand)
5570 struct ext4_inode *raw_inode;
5571 struct ext4_xattr_ibody_header *header;
5574 raw_inode = ext4_raw_inode(iloc);
5576 header = IHDR(inode, raw_inode);
5578 /* No extended attributes present */
5579 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5580 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5581 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5582 EXT4_I(inode)->i_extra_isize, 0,
5583 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5584 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5588 /* try to expand with EAs present */
5589 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5593 * Inode size expansion failed; don't try again
5602 * Expand an inode by new_extra_isize bytes.
5603 * Returns 0 on success or negative error number on failure.
5605 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5606 unsigned int new_extra_isize,
5607 struct ext4_iloc iloc,
5613 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5617 * In nojournal mode, we can immediately attempt to expand
5618 * the inode. When journaled, we first need to obtain extra
5619 * buffer credits since we may write into the EA block
5620 * with this same handle. If journal_extend fails, then it will
5621 * only result in a minor loss of functionality for that inode.
5622 * If this is felt to be critical, then e2fsck should be run to
5623 * force a large enough s_min_extra_isize.
5625 if (ext4_journal_extend(handle,
5626 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5629 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5632 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5633 handle, &no_expand);
5634 ext4_write_unlock_xattr(inode, &no_expand);
5639 int ext4_expand_extra_isize(struct inode *inode,
5640 unsigned int new_extra_isize,
5641 struct ext4_iloc *iloc)
5647 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5652 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5653 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5654 if (IS_ERR(handle)) {
5655 error = PTR_ERR(handle);
5660 ext4_write_lock_xattr(inode, &no_expand);
5662 BUFFER_TRACE(iloc->bh, "get_write_access");
5663 error = ext4_journal_get_write_access(handle, iloc->bh);
5669 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5670 handle, &no_expand);
5672 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5676 ext4_write_unlock_xattr(inode, &no_expand);
5678 ext4_journal_stop(handle);
5683 * What we do here is to mark the in-core inode as clean with respect to inode
5684 * dirtiness (it may still be data-dirty).
5685 * This means that the in-core inode may be reaped by prune_icache
5686 * without having to perform any I/O. This is a very good thing,
5687 * because *any* task may call prune_icache - even ones which
5688 * have a transaction open against a different journal.
5690 * Is this cheating? Not really. Sure, we haven't written the
5691 * inode out, but prune_icache isn't a user-visible syncing function.
5692 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5693 * we start and wait on commits.
5695 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5697 struct ext4_iloc iloc;
5698 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5702 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5703 err = ext4_reserve_inode_write(handle, inode, &iloc);
5707 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5708 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5711 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5715 * ext4_dirty_inode() is called from __mark_inode_dirty()
5717 * We're really interested in the case where a file is being extended.
5718 * i_size has been changed by generic_commit_write() and we thus need
5719 * to include the updated inode in the current transaction.
5721 * Also, dquot_alloc_block() will always dirty the inode when blocks
5722 * are allocated to the file.
5724 * If the inode is marked synchronous, we don't honour that here - doing
5725 * so would cause a commit on atime updates, which we don't bother doing.
5726 * We handle synchronous inodes at the highest possible level.
5728 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5729 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5730 * to copy into the on-disk inode structure are the timestamp files.
5732 void ext4_dirty_inode(struct inode *inode, int flags)
5736 if (flags == I_DIRTY_TIME)
5738 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5742 ext4_mark_inode_dirty(handle, inode);
5744 ext4_journal_stop(handle);
5749 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5754 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5757 * We have to be very careful here: changing a data block's
5758 * journaling status dynamically is dangerous. If we write a
5759 * data block to the journal, change the status and then delete
5760 * that block, we risk forgetting to revoke the old log record
5761 * from the journal and so a subsequent replay can corrupt data.
5762 * So, first we make sure that the journal is empty and that
5763 * nobody is changing anything.
5766 journal = EXT4_JOURNAL(inode);
5769 if (is_journal_aborted(journal))
5772 /* Wait for all existing dio workers */
5773 inode_dio_wait(inode);
5776 * Before flushing the journal and switching inode's aops, we have
5777 * to flush all dirty data the inode has. There can be outstanding
5778 * delayed allocations, there can be unwritten extents created by
5779 * fallocate or buffered writes in dioread_nolock mode covered by
5780 * dirty data which can be converted only after flushing the dirty
5781 * data (and journalled aops don't know how to handle these cases).
5784 down_write(&EXT4_I(inode)->i_mmap_sem);
5785 err = filemap_write_and_wait(inode->i_mapping);
5787 up_write(&EXT4_I(inode)->i_mmap_sem);
5792 percpu_down_write(&sbi->s_journal_flag_rwsem);
5793 jbd2_journal_lock_updates(journal);
5796 * OK, there are no updates running now, and all cached data is
5797 * synced to disk. We are now in a completely consistent state
5798 * which doesn't have anything in the journal, and we know that
5799 * no filesystem updates are running, so it is safe to modify
5800 * the inode's in-core data-journaling state flag now.
5804 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5806 err = jbd2_journal_flush(journal);
5808 jbd2_journal_unlock_updates(journal);
5809 percpu_up_write(&sbi->s_journal_flag_rwsem);
5812 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5814 ext4_set_aops(inode);
5816 jbd2_journal_unlock_updates(journal);
5817 percpu_up_write(&sbi->s_journal_flag_rwsem);
5820 up_write(&EXT4_I(inode)->i_mmap_sem);
5822 /* Finally we can mark the inode as dirty. */
5824 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5826 return PTR_ERR(handle);
5828 err = ext4_mark_inode_dirty(handle, inode);
5829 ext4_handle_sync(handle);
5830 ext4_journal_stop(handle);
5831 ext4_std_error(inode->i_sb, err);
5836 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5838 return !buffer_mapped(bh);
5841 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
5843 struct vm_area_struct *vma = vmf->vma;
5844 struct page *page = vmf->page;
5849 struct file *file = vma->vm_file;
5850 struct inode *inode = file_inode(file);
5851 struct address_space *mapping = inode->i_mapping;
5853 get_block_t *get_block;
5856 if (unlikely(IS_IMMUTABLE(inode)))
5857 return VM_FAULT_SIGBUS;
5859 sb_start_pagefault(inode->i_sb);
5860 file_update_time(vma->vm_file);
5862 down_read(&EXT4_I(inode)->i_mmap_sem);
5864 err = ext4_convert_inline_data(inode);
5868 /* Delalloc case is easy... */
5869 if (test_opt(inode->i_sb, DELALLOC) &&
5870 !ext4_should_journal_data(inode) &&
5871 !ext4_nonda_switch(inode->i_sb)) {
5873 err = block_page_mkwrite(vma, vmf,
5874 ext4_da_get_block_prep);
5875 } while (err == -ENOSPC &&
5876 ext4_should_retry_alloc(inode->i_sb, &retries));
5881 size = i_size_read(inode);
5882 /* Page got truncated from under us? */
5883 if (page->mapping != mapping || page_offset(page) > size) {
5885 ret = VM_FAULT_NOPAGE;
5889 if (page->index == size >> PAGE_SHIFT)
5890 len = size & ~PAGE_MASK;
5894 * Return if we have all the buffers mapped. This avoids the need to do
5895 * journal_start/journal_stop which can block and take a long time
5897 if (page_has_buffers(page)) {
5898 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5900 ext4_bh_unmapped)) {
5901 /* Wait so that we don't change page under IO */
5902 wait_for_stable_page(page);
5903 ret = VM_FAULT_LOCKED;
5908 /* OK, we need to fill the hole... */
5909 if (ext4_should_dioread_nolock(inode))
5910 get_block = ext4_get_block_unwritten;
5912 get_block = ext4_get_block;
5914 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5915 ext4_writepage_trans_blocks(inode));
5916 if (IS_ERR(handle)) {
5917 ret = VM_FAULT_SIGBUS;
5920 err = block_page_mkwrite(vma, vmf, get_block);
5921 if (!err && ext4_should_journal_data(inode)) {
5922 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5923 PAGE_SIZE, NULL, do_journal_get_write_access)) {
5925 ret = VM_FAULT_SIGBUS;
5926 ext4_journal_stop(handle);
5929 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5931 ext4_journal_stop(handle);
5932 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5935 ret = block_page_mkwrite_return(err);
5937 up_read(&EXT4_I(inode)->i_mmap_sem);
5938 sb_end_pagefault(inode->i_sb);
5942 vm_fault_t ext4_filemap_fault(struct vm_fault *vmf)
5944 struct inode *inode = file_inode(vmf->vma->vm_file);
5947 down_read(&EXT4_I(inode)->i_mmap_sem);
5948 ret = filemap_fault(vmf);
5949 up_read(&EXT4_I(inode)->i_mmap_sem);