1 // SPDX-License-Identifier: GPL-2.0+
3 * inode.c - NILFS inode operations.
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
7 * Written by Ryusuke Konishi.
11 #include <linux/buffer_head.h>
12 #include <linux/gfp.h>
13 #include <linux/mpage.h>
14 #include <linux/pagemap.h>
15 #include <linux/writeback.h>
16 #include <linux/uio.h>
17 #include <linux/fiemap.h>
27 * struct nilfs_iget_args - arguments used during comparison between inodes
29 * @cno: checkpoint number
30 * @root: pointer on NILFS root object (mounted checkpoint)
31 * @for_gc: inode for GC flag
33 struct nilfs_iget_args
{
36 struct nilfs_root
*root
;
40 static int nilfs_iget_test(struct inode
*inode
, void *opaque
);
42 void nilfs_inode_add_blocks(struct inode
*inode
, int n
)
44 struct nilfs_root
*root
= NILFS_I(inode
)->i_root
;
46 inode_add_bytes(inode
, i_blocksize(inode
) * n
);
48 atomic64_add(n
, &root
->blocks_count
);
51 void nilfs_inode_sub_blocks(struct inode
*inode
, int n
)
53 struct nilfs_root
*root
= NILFS_I(inode
)->i_root
;
55 inode_sub_bytes(inode
, i_blocksize(inode
) * n
);
57 atomic64_sub(n
, &root
->blocks_count
);
61 * nilfs_get_block() - get a file block on the filesystem (callback function)
62 * @inode - inode struct of the target file
63 * @blkoff - file block number
64 * @bh_result - buffer head to be mapped on
65 * @create - indicate whether allocating the block or not when it has not
68 * This function does not issue actual read request of the specified data
69 * block. It is done by VFS.
71 int nilfs_get_block(struct inode
*inode
, sector_t blkoff
,
72 struct buffer_head
*bh_result
, int create
)
74 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
75 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
78 unsigned int maxblocks
= bh_result
->b_size
>> inode
->i_blkbits
;
80 down_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
81 ret
= nilfs_bmap_lookup_contig(ii
->i_bmap
, blkoff
, &blknum
, maxblocks
);
82 up_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
83 if (ret
>= 0) { /* found */
84 map_bh(bh_result
, inode
->i_sb
, blknum
);
86 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
89 /* data block was not found */
90 if (ret
== -ENOENT
&& create
) {
91 struct nilfs_transaction_info ti
;
93 bh_result
->b_blocknr
= 0;
94 err
= nilfs_transaction_begin(inode
->i_sb
, &ti
, 1);
97 err
= nilfs_bmap_insert(ii
->i_bmap
, blkoff
,
98 (unsigned long)bh_result
);
99 if (unlikely(err
!= 0)) {
100 if (err
== -EEXIST
) {
102 * The get_block() function could be called
103 * from multiple callers for an inode.
104 * However, the page having this block must
105 * be locked in this case.
107 nilfs_msg(inode
->i_sb
, KERN_WARNING
,
108 "%s (ino=%lu): a race condition while inserting a data block at offset=%llu",
109 __func__
, inode
->i_ino
,
110 (unsigned long long)blkoff
);
113 nilfs_transaction_abort(inode
->i_sb
);
116 nilfs_mark_inode_dirty_sync(inode
);
117 nilfs_transaction_commit(inode
->i_sb
); /* never fails */
118 /* Error handling should be detailed */
119 set_buffer_new(bh_result
);
120 set_buffer_delay(bh_result
);
121 map_bh(bh_result
, inode
->i_sb
, 0);
122 /* Disk block number must be changed to proper value */
124 } else if (ret
== -ENOENT
) {
126 * not found is not error (e.g. hole); must return without
127 * the mapped state flag.
139 * nilfs_readpage() - implement readpage() method of nilfs_aops {}
140 * address_space_operations.
141 * @file - file struct of the file to be read
142 * @page - the page to be read
144 static int nilfs_readpage(struct file
*file
, struct page
*page
)
146 return mpage_readpage(page
, nilfs_get_block
);
149 static void nilfs_readahead(struct readahead_control
*rac
)
151 mpage_readahead(rac
, nilfs_get_block
);
154 static int nilfs_writepages(struct address_space
*mapping
,
155 struct writeback_control
*wbc
)
157 struct inode
*inode
= mapping
->host
;
160 if (sb_rdonly(inode
->i_sb
)) {
161 nilfs_clear_dirty_pages(mapping
, false);
165 if (wbc
->sync_mode
== WB_SYNC_ALL
)
166 err
= nilfs_construct_dsync_segment(inode
->i_sb
, inode
,
172 static int nilfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
174 struct inode
*inode
= page
->mapping
->host
;
177 if (sb_rdonly(inode
->i_sb
)) {
179 * It means that filesystem was remounted in read-only
180 * mode because of error or metadata corruption. But we
181 * have dirty pages that try to be flushed in background.
182 * So, here we simply discard this dirty page.
184 nilfs_clear_dirty_page(page
, false);
189 redirty_page_for_writepage(wbc
, page
);
192 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
193 err
= nilfs_construct_segment(inode
->i_sb
);
196 } else if (wbc
->for_reclaim
)
197 nilfs_flush_segment(inode
->i_sb
, inode
->i_ino
);
202 static int nilfs_set_page_dirty(struct page
*page
)
204 struct inode
*inode
= page
->mapping
->host
;
205 int ret
= __set_page_dirty_nobuffers(page
);
207 if (page_has_buffers(page
)) {
208 unsigned int nr_dirty
= 0;
209 struct buffer_head
*bh
, *head
;
212 * This page is locked by callers, and no other thread
213 * concurrently marks its buffers dirty since they are
214 * only dirtied through routines in fs/buffer.c in
215 * which call sites of mark_buffer_dirty are protected
218 bh
= head
= page_buffers(page
);
220 /* Do not mark hole blocks dirty */
221 if (buffer_dirty(bh
) || !buffer_mapped(bh
))
224 set_buffer_dirty(bh
);
226 } while (bh
= bh
->b_this_page
, bh
!= head
);
229 nilfs_set_file_dirty(inode
, nr_dirty
);
231 unsigned int nr_dirty
= 1 << (PAGE_SHIFT
- inode
->i_blkbits
);
233 nilfs_set_file_dirty(inode
, nr_dirty
);
238 void nilfs_write_failed(struct address_space
*mapping
, loff_t to
)
240 struct inode
*inode
= mapping
->host
;
242 if (to
> inode
->i_size
) {
243 truncate_pagecache(inode
, inode
->i_size
);
244 nilfs_truncate(inode
);
248 static int nilfs_write_begin(struct file
*file
, struct address_space
*mapping
,
249 loff_t pos
, unsigned len
, unsigned flags
,
250 struct page
**pagep
, void **fsdata
)
253 struct inode
*inode
= mapping
->host
;
254 int err
= nilfs_transaction_begin(inode
->i_sb
, NULL
, 1);
259 err
= block_write_begin(mapping
, pos
, len
, flags
, pagep
,
262 nilfs_write_failed(mapping
, pos
+ len
);
263 nilfs_transaction_abort(inode
->i_sb
);
268 static int nilfs_write_end(struct file
*file
, struct address_space
*mapping
,
269 loff_t pos
, unsigned len
, unsigned copied
,
270 struct page
*page
, void *fsdata
)
272 struct inode
*inode
= mapping
->host
;
273 unsigned int start
= pos
& (PAGE_SIZE
- 1);
274 unsigned int nr_dirty
;
277 nr_dirty
= nilfs_page_count_clean_buffers(page
, start
,
279 copied
= generic_write_end(file
, mapping
, pos
, len
, copied
, page
,
281 nilfs_set_file_dirty(inode
, nr_dirty
);
282 err
= nilfs_transaction_commit(inode
->i_sb
);
283 return err
? : copied
;
287 nilfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
289 struct inode
*inode
= file_inode(iocb
->ki_filp
);
291 if (iov_iter_rw(iter
) == WRITE
)
294 /* Needs synchronization with the cleaner */
295 return blockdev_direct_IO(iocb
, inode
, iter
, nilfs_get_block
);
298 const struct address_space_operations nilfs_aops
= {
299 .writepage
= nilfs_writepage
,
300 .readpage
= nilfs_readpage
,
301 .writepages
= nilfs_writepages
,
302 .set_page_dirty
= nilfs_set_page_dirty
,
303 .readahead
= nilfs_readahead
,
304 .write_begin
= nilfs_write_begin
,
305 .write_end
= nilfs_write_end
,
306 /* .releasepage = nilfs_releasepage, */
307 .invalidatepage
= block_invalidatepage
,
308 .direct_IO
= nilfs_direct_IO
,
309 .is_partially_uptodate
= block_is_partially_uptodate
,
312 static int nilfs_insert_inode_locked(struct inode
*inode
,
313 struct nilfs_root
*root
,
316 struct nilfs_iget_args args
= {
317 .ino
= ino
, .root
= root
, .cno
= 0, .for_gc
= 0
320 return insert_inode_locked4(inode
, ino
, nilfs_iget_test
, &args
);
323 struct inode
*nilfs_new_inode(struct inode
*dir
, umode_t mode
)
325 struct super_block
*sb
= dir
->i_sb
;
326 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
328 struct nilfs_inode_info
*ii
;
329 struct nilfs_root
*root
;
333 inode
= new_inode(sb
);
334 if (unlikely(!inode
))
337 mapping_set_gfp_mask(inode
->i_mapping
,
338 mapping_gfp_constraint(inode
->i_mapping
, ~__GFP_FS
));
340 root
= NILFS_I(dir
)->i_root
;
342 ii
->i_state
= BIT(NILFS_I_NEW
);
345 err
= nilfs_ifile_create_inode(root
->ifile
, &ino
, &ii
->i_bh
);
347 goto failed_ifile_create_inode
;
348 /* reference count of i_bh inherits from nilfs_mdt_read_block() */
350 atomic64_inc(&root
->inodes_count
);
351 inode_init_owner(inode
, dir
, mode
);
353 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= current_time(inode
);
355 if (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)) {
356 err
= nilfs_bmap_read(ii
->i_bmap
, NULL
);
358 goto failed_after_creation
;
360 set_bit(NILFS_I_BMAP
, &ii
->i_state
);
361 /* No lock is needed; iget() ensures it. */
364 ii
->i_flags
= nilfs_mask_flags(
365 mode
, NILFS_I(dir
)->i_flags
& NILFS_FL_INHERITED
);
367 /* ii->i_file_acl = 0; */
368 /* ii->i_dir_acl = 0; */
369 ii
->i_dir_start_lookup
= 0;
370 nilfs_set_inode_flags(inode
);
371 spin_lock(&nilfs
->ns_next_gen_lock
);
372 inode
->i_generation
= nilfs
->ns_next_generation
++;
373 spin_unlock(&nilfs
->ns_next_gen_lock
);
374 if (nilfs_insert_inode_locked(inode
, root
, ino
) < 0) {
376 goto failed_after_creation
;
379 err
= nilfs_init_acl(inode
, dir
);
382 * Never occur. When supporting nilfs_init_acl(),
383 * proper cancellation of above jobs should be considered.
385 goto failed_after_creation
;
389 failed_after_creation
:
391 unlock_new_inode(inode
);
393 * raw_inode will be deleted through
394 * nilfs_evict_inode().
398 failed_ifile_create_inode
:
399 make_bad_inode(inode
);
405 void nilfs_set_inode_flags(struct inode
*inode
)
407 unsigned int flags
= NILFS_I(inode
)->i_flags
;
408 unsigned int new_fl
= 0;
410 if (flags
& FS_SYNC_FL
)
412 if (flags
& FS_APPEND_FL
)
414 if (flags
& FS_IMMUTABLE_FL
)
415 new_fl
|= S_IMMUTABLE
;
416 if (flags
& FS_NOATIME_FL
)
418 if (flags
& FS_DIRSYNC_FL
)
420 inode_set_flags(inode
, new_fl
, S_SYNC
| S_APPEND
| S_IMMUTABLE
|
421 S_NOATIME
| S_DIRSYNC
);
424 int nilfs_read_inode_common(struct inode
*inode
,
425 struct nilfs_inode
*raw_inode
)
427 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
430 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
431 i_uid_write(inode
, le32_to_cpu(raw_inode
->i_uid
));
432 i_gid_write(inode
, le32_to_cpu(raw_inode
->i_gid
));
433 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
434 inode
->i_size
= le64_to_cpu(raw_inode
->i_size
);
435 inode
->i_atime
.tv_sec
= le64_to_cpu(raw_inode
->i_mtime
);
436 inode
->i_ctime
.tv_sec
= le64_to_cpu(raw_inode
->i_ctime
);
437 inode
->i_mtime
.tv_sec
= le64_to_cpu(raw_inode
->i_mtime
);
438 inode
->i_atime
.tv_nsec
= le32_to_cpu(raw_inode
->i_mtime_nsec
);
439 inode
->i_ctime
.tv_nsec
= le32_to_cpu(raw_inode
->i_ctime_nsec
);
440 inode
->i_mtime
.tv_nsec
= le32_to_cpu(raw_inode
->i_mtime_nsec
);
441 if (inode
->i_nlink
== 0)
442 return -ESTALE
; /* this inode is deleted */
444 inode
->i_blocks
= le64_to_cpu(raw_inode
->i_blocks
);
445 ii
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
447 ii
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl
);
448 ii
->i_dir_acl
= S_ISREG(inode
->i_mode
) ?
449 0 : le32_to_cpu(raw_inode
->i_dir_acl
);
451 ii
->i_dir_start_lookup
= 0;
452 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
454 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
455 S_ISLNK(inode
->i_mode
)) {
456 err
= nilfs_bmap_read(ii
->i_bmap
, raw_inode
);
459 set_bit(NILFS_I_BMAP
, &ii
->i_state
);
460 /* No lock is needed; iget() ensures it. */
465 static int __nilfs_read_inode(struct super_block
*sb
,
466 struct nilfs_root
*root
, unsigned long ino
,
469 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
470 struct buffer_head
*bh
;
471 struct nilfs_inode
*raw_inode
;
474 down_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
475 err
= nilfs_ifile_get_inode_block(root
->ifile
, ino
, &bh
);
479 raw_inode
= nilfs_ifile_map_inode(root
->ifile
, ino
, bh
);
481 err
= nilfs_read_inode_common(inode
, raw_inode
);
485 if (S_ISREG(inode
->i_mode
)) {
486 inode
->i_op
= &nilfs_file_inode_operations
;
487 inode
->i_fop
= &nilfs_file_operations
;
488 inode
->i_mapping
->a_ops
= &nilfs_aops
;
489 } else if (S_ISDIR(inode
->i_mode
)) {
490 inode
->i_op
= &nilfs_dir_inode_operations
;
491 inode
->i_fop
= &nilfs_dir_operations
;
492 inode
->i_mapping
->a_ops
= &nilfs_aops
;
493 } else if (S_ISLNK(inode
->i_mode
)) {
494 inode
->i_op
= &nilfs_symlink_inode_operations
;
495 inode_nohighmem(inode
);
496 inode
->i_mapping
->a_ops
= &nilfs_aops
;
498 inode
->i_op
= &nilfs_special_inode_operations
;
500 inode
, inode
->i_mode
,
501 huge_decode_dev(le64_to_cpu(raw_inode
->i_device_code
)));
503 nilfs_ifile_unmap_inode(root
->ifile
, ino
, bh
);
505 up_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
506 nilfs_set_inode_flags(inode
);
507 mapping_set_gfp_mask(inode
->i_mapping
,
508 mapping_gfp_constraint(inode
->i_mapping
, ~__GFP_FS
));
512 nilfs_ifile_unmap_inode(root
->ifile
, ino
, bh
);
516 up_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
520 static int nilfs_iget_test(struct inode
*inode
, void *opaque
)
522 struct nilfs_iget_args
*args
= opaque
;
523 struct nilfs_inode_info
*ii
;
525 if (args
->ino
!= inode
->i_ino
|| args
->root
!= NILFS_I(inode
)->i_root
)
529 if (!test_bit(NILFS_I_GCINODE
, &ii
->i_state
))
530 return !args
->for_gc
;
532 return args
->for_gc
&& args
->cno
== ii
->i_cno
;
535 static int nilfs_iget_set(struct inode
*inode
, void *opaque
)
537 struct nilfs_iget_args
*args
= opaque
;
539 inode
->i_ino
= args
->ino
;
541 NILFS_I(inode
)->i_state
= BIT(NILFS_I_GCINODE
);
542 NILFS_I(inode
)->i_cno
= args
->cno
;
543 NILFS_I(inode
)->i_root
= NULL
;
545 if (args
->root
&& args
->ino
== NILFS_ROOT_INO
)
546 nilfs_get_root(args
->root
);
547 NILFS_I(inode
)->i_root
= args
->root
;
552 struct inode
*nilfs_ilookup(struct super_block
*sb
, struct nilfs_root
*root
,
555 struct nilfs_iget_args args
= {
556 .ino
= ino
, .root
= root
, .cno
= 0, .for_gc
= 0
559 return ilookup5(sb
, ino
, nilfs_iget_test
, &args
);
562 struct inode
*nilfs_iget_locked(struct super_block
*sb
, struct nilfs_root
*root
,
565 struct nilfs_iget_args args
= {
566 .ino
= ino
, .root
= root
, .cno
= 0, .for_gc
= 0
569 return iget5_locked(sb
, ino
, nilfs_iget_test
, nilfs_iget_set
, &args
);
572 struct inode
*nilfs_iget(struct super_block
*sb
, struct nilfs_root
*root
,
578 inode
= nilfs_iget_locked(sb
, root
, ino
);
579 if (unlikely(!inode
))
580 return ERR_PTR(-ENOMEM
);
581 if (!(inode
->i_state
& I_NEW
))
584 err
= __nilfs_read_inode(sb
, root
, ino
, inode
);
589 unlock_new_inode(inode
);
593 struct inode
*nilfs_iget_for_gc(struct super_block
*sb
, unsigned long ino
,
596 struct nilfs_iget_args args
= {
597 .ino
= ino
, .root
= NULL
, .cno
= cno
, .for_gc
= 1
602 inode
= iget5_locked(sb
, ino
, nilfs_iget_test
, nilfs_iget_set
, &args
);
603 if (unlikely(!inode
))
604 return ERR_PTR(-ENOMEM
);
605 if (!(inode
->i_state
& I_NEW
))
608 err
= nilfs_init_gcinode(inode
);
613 unlock_new_inode(inode
);
617 void nilfs_write_inode_common(struct inode
*inode
,
618 struct nilfs_inode
*raw_inode
, int has_bmap
)
620 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
622 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
623 raw_inode
->i_uid
= cpu_to_le32(i_uid_read(inode
));
624 raw_inode
->i_gid
= cpu_to_le32(i_gid_read(inode
));
625 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
626 raw_inode
->i_size
= cpu_to_le64(inode
->i_size
);
627 raw_inode
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
628 raw_inode
->i_mtime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
629 raw_inode
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
630 raw_inode
->i_mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
631 raw_inode
->i_blocks
= cpu_to_le64(inode
->i_blocks
);
633 raw_inode
->i_flags
= cpu_to_le32(ii
->i_flags
);
634 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
636 if (NILFS_ROOT_METADATA_FILE(inode
->i_ino
)) {
637 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
639 /* zero-fill unused portion in the case of super root block */
640 raw_inode
->i_xattr
= 0;
641 raw_inode
->i_pad
= 0;
642 memset((void *)raw_inode
+ sizeof(*raw_inode
), 0,
643 nilfs
->ns_inode_size
- sizeof(*raw_inode
));
647 nilfs_bmap_write(ii
->i_bmap
, raw_inode
);
648 else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
))
649 raw_inode
->i_device_code
=
650 cpu_to_le64(huge_encode_dev(inode
->i_rdev
));
652 * When extending inode, nilfs->ns_inode_size should be checked
653 * for substitutions of appended fields.
657 void nilfs_update_inode(struct inode
*inode
, struct buffer_head
*ibh
, int flags
)
659 ino_t ino
= inode
->i_ino
;
660 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
661 struct inode
*ifile
= ii
->i_root
->ifile
;
662 struct nilfs_inode
*raw_inode
;
664 raw_inode
= nilfs_ifile_map_inode(ifile
, ino
, ibh
);
666 if (test_and_clear_bit(NILFS_I_NEW
, &ii
->i_state
))
667 memset(raw_inode
, 0, NILFS_MDT(ifile
)->mi_entry_size
);
668 if (flags
& I_DIRTY_DATASYNC
)
669 set_bit(NILFS_I_INODE_SYNC
, &ii
->i_state
);
671 nilfs_write_inode_common(inode
, raw_inode
, 0);
673 * XXX: call with has_bmap = 0 is a workaround to avoid
674 * deadlock of bmap. This delays update of i_bmap to just
678 nilfs_ifile_unmap_inode(ifile
, ino
, ibh
);
681 #define NILFS_MAX_TRUNCATE_BLOCKS 16384 /* 64MB for 4KB block */
683 static void nilfs_truncate_bmap(struct nilfs_inode_info
*ii
,
689 if (!test_bit(NILFS_I_BMAP
, &ii
->i_state
))
692 ret
= nilfs_bmap_last_key(ii
->i_bmap
, &b
);
701 b
-= min_t(__u64
, NILFS_MAX_TRUNCATE_BLOCKS
, b
- from
);
702 ret
= nilfs_bmap_truncate(ii
->i_bmap
, b
);
703 nilfs_relax_pressure_in_lock(ii
->vfs_inode
.i_sb
);
704 if (!ret
|| (ret
== -ENOMEM
&&
705 nilfs_bmap_truncate(ii
->i_bmap
, b
) == 0))
709 nilfs_msg(ii
->vfs_inode
.i_sb
, KERN_WARNING
,
710 "error %d truncating bmap (ino=%lu)", ret
,
711 ii
->vfs_inode
.i_ino
);
714 void nilfs_truncate(struct inode
*inode
)
716 unsigned long blkoff
;
717 unsigned int blocksize
;
718 struct nilfs_transaction_info ti
;
719 struct super_block
*sb
= inode
->i_sb
;
720 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
722 if (!test_bit(NILFS_I_BMAP
, &ii
->i_state
))
724 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
727 blocksize
= sb
->s_blocksize
;
728 blkoff
= (inode
->i_size
+ blocksize
- 1) >> sb
->s_blocksize_bits
;
729 nilfs_transaction_begin(sb
, &ti
, 0); /* never fails */
731 block_truncate_page(inode
->i_mapping
, inode
->i_size
, nilfs_get_block
);
733 nilfs_truncate_bmap(ii
, blkoff
);
735 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
737 nilfs_set_transaction_flag(NILFS_TI_SYNC
);
739 nilfs_mark_inode_dirty(inode
);
740 nilfs_set_file_dirty(inode
, 0);
741 nilfs_transaction_commit(sb
);
743 * May construct a logical segment and may fail in sync mode.
744 * But truncate has no return value.
748 static void nilfs_clear_inode(struct inode
*inode
)
750 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
753 * Free resources allocated in nilfs_read_inode(), here.
755 BUG_ON(!list_empty(&ii
->i_dirty
));
759 if (nilfs_is_metadata_file_inode(inode
))
760 nilfs_mdt_clear(inode
);
762 if (test_bit(NILFS_I_BMAP
, &ii
->i_state
))
763 nilfs_bmap_clear(ii
->i_bmap
);
765 nilfs_btnode_cache_clear(&ii
->i_btnode_cache
);
767 if (ii
->i_root
&& inode
->i_ino
== NILFS_ROOT_INO
)
768 nilfs_put_root(ii
->i_root
);
771 void nilfs_evict_inode(struct inode
*inode
)
773 struct nilfs_transaction_info ti
;
774 struct super_block
*sb
= inode
->i_sb
;
775 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
778 if (inode
->i_nlink
|| !ii
->i_root
|| unlikely(is_bad_inode(inode
))) {
779 truncate_inode_pages_final(&inode
->i_data
);
781 nilfs_clear_inode(inode
);
784 nilfs_transaction_begin(sb
, &ti
, 0); /* never fails */
786 truncate_inode_pages_final(&inode
->i_data
);
788 /* TODO: some of the following operations may fail. */
789 nilfs_truncate_bmap(ii
, 0);
790 nilfs_mark_inode_dirty(inode
);
793 ret
= nilfs_ifile_delete_inode(ii
->i_root
->ifile
, inode
->i_ino
);
795 atomic64_dec(&ii
->i_root
->inodes_count
);
797 nilfs_clear_inode(inode
);
800 nilfs_set_transaction_flag(NILFS_TI_SYNC
);
801 nilfs_transaction_commit(sb
);
803 * May construct a logical segment and may fail in sync mode.
804 * But delete_inode has no return value.
808 int nilfs_setattr(struct dentry
*dentry
, struct iattr
*iattr
)
810 struct nilfs_transaction_info ti
;
811 struct inode
*inode
= d_inode(dentry
);
812 struct super_block
*sb
= inode
->i_sb
;
815 err
= setattr_prepare(dentry
, iattr
);
819 err
= nilfs_transaction_begin(sb
, &ti
, 0);
823 if ((iattr
->ia_valid
& ATTR_SIZE
) &&
824 iattr
->ia_size
!= i_size_read(inode
)) {
825 inode_dio_wait(inode
);
826 truncate_setsize(inode
, iattr
->ia_size
);
827 nilfs_truncate(inode
);
830 setattr_copy(inode
, iattr
);
831 mark_inode_dirty(inode
);
833 if (iattr
->ia_valid
& ATTR_MODE
) {
834 err
= nilfs_acl_chmod(inode
);
839 return nilfs_transaction_commit(sb
);
842 nilfs_transaction_abort(sb
);
846 int nilfs_permission(struct inode
*inode
, int mask
)
848 struct nilfs_root
*root
= NILFS_I(inode
)->i_root
;
850 if ((mask
& MAY_WRITE
) && root
&&
851 root
->cno
!= NILFS_CPTREE_CURRENT_CNO
)
852 return -EROFS
; /* snapshot is not writable */
854 return generic_permission(inode
, mask
);
857 int nilfs_load_inode_block(struct inode
*inode
, struct buffer_head
**pbh
)
859 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
860 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
863 spin_lock(&nilfs
->ns_inode_lock
);
864 if (ii
->i_bh
== NULL
) {
865 spin_unlock(&nilfs
->ns_inode_lock
);
866 err
= nilfs_ifile_get_inode_block(ii
->i_root
->ifile
,
870 spin_lock(&nilfs
->ns_inode_lock
);
871 if (ii
->i_bh
== NULL
)
881 spin_unlock(&nilfs
->ns_inode_lock
);
885 int nilfs_inode_dirty(struct inode
*inode
)
887 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
888 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
891 if (!list_empty(&ii
->i_dirty
)) {
892 spin_lock(&nilfs
->ns_inode_lock
);
893 ret
= test_bit(NILFS_I_DIRTY
, &ii
->i_state
) ||
894 test_bit(NILFS_I_BUSY
, &ii
->i_state
);
895 spin_unlock(&nilfs
->ns_inode_lock
);
900 int nilfs_set_file_dirty(struct inode
*inode
, unsigned int nr_dirty
)
902 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
903 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
905 atomic_add(nr_dirty
, &nilfs
->ns_ndirtyblks
);
907 if (test_and_set_bit(NILFS_I_DIRTY
, &ii
->i_state
))
910 spin_lock(&nilfs
->ns_inode_lock
);
911 if (!test_bit(NILFS_I_QUEUED
, &ii
->i_state
) &&
912 !test_bit(NILFS_I_BUSY
, &ii
->i_state
)) {
914 * Because this routine may race with nilfs_dispose_list(),
915 * we have to check NILFS_I_QUEUED here, too.
917 if (list_empty(&ii
->i_dirty
) && igrab(inode
) == NULL
) {
919 * This will happen when somebody is freeing
922 nilfs_msg(inode
->i_sb
, KERN_WARNING
,
923 "cannot set file dirty (ino=%lu): the file is being freed",
925 spin_unlock(&nilfs
->ns_inode_lock
);
927 * NILFS_I_DIRTY may remain for
931 list_move_tail(&ii
->i_dirty
, &nilfs
->ns_dirty_files
);
932 set_bit(NILFS_I_QUEUED
, &ii
->i_state
);
934 spin_unlock(&nilfs
->ns_inode_lock
);
938 int __nilfs_mark_inode_dirty(struct inode
*inode
, int flags
)
940 struct buffer_head
*ibh
;
943 err
= nilfs_load_inode_block(inode
, &ibh
);
945 nilfs_msg(inode
->i_sb
, KERN_WARNING
,
946 "cannot mark inode dirty (ino=%lu): error %d loading inode block",
950 nilfs_update_inode(inode
, ibh
, flags
);
951 mark_buffer_dirty(ibh
);
952 nilfs_mdt_mark_dirty(NILFS_I(inode
)->i_root
->ifile
);
958 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
959 * @inode: inode of the file to be registered.
961 * nilfs_dirty_inode() loads a inode block containing the specified
962 * @inode and copies data from a nilfs_inode to a corresponding inode
963 * entry in the inode block. This operation is excluded from the segment
964 * construction. This function can be called both as a single operation
965 * and as a part of indivisible file operations.
967 void nilfs_dirty_inode(struct inode
*inode
, int flags
)
969 struct nilfs_transaction_info ti
;
970 struct nilfs_mdt_info
*mdi
= NILFS_MDT(inode
);
972 if (is_bad_inode(inode
)) {
973 nilfs_msg(inode
->i_sb
, KERN_WARNING
,
974 "tried to mark bad_inode dirty. ignored.");
979 nilfs_mdt_mark_dirty(inode
);
982 nilfs_transaction_begin(inode
->i_sb
, &ti
, 0);
983 __nilfs_mark_inode_dirty(inode
, flags
);
984 nilfs_transaction_commit(inode
->i_sb
); /* never fails */
987 int nilfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
988 __u64 start
, __u64 len
)
990 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
991 __u64 logical
= 0, phys
= 0, size
= 0;
994 sector_t blkoff
, end_blkoff
;
995 sector_t delalloc_blkoff
;
996 unsigned long delalloc_blklen
;
997 unsigned int blkbits
= inode
->i_blkbits
;
1000 ret
= fiemap_prep(inode
, fieinfo
, start
, &len
, 0);
1006 isize
= i_size_read(inode
);
1008 blkoff
= start
>> blkbits
;
1009 end_blkoff
= (start
+ len
- 1) >> blkbits
;
1011 delalloc_blklen
= nilfs_find_uncommitted_extent(inode
, blkoff
,
1016 unsigned int maxblocks
;
1018 if (delalloc_blklen
&& blkoff
== delalloc_blkoff
) {
1020 /* End of the current extent */
1021 ret
= fiemap_fill_next_extent(
1022 fieinfo
, logical
, phys
, size
, flags
);
1026 if (blkoff
> end_blkoff
)
1029 flags
= FIEMAP_EXTENT_MERGED
| FIEMAP_EXTENT_DELALLOC
;
1030 logical
= blkoff
<< blkbits
;
1032 size
= delalloc_blklen
<< blkbits
;
1034 blkoff
= delalloc_blkoff
+ delalloc_blklen
;
1035 delalloc_blklen
= nilfs_find_uncommitted_extent(
1036 inode
, blkoff
, &delalloc_blkoff
);
1041 * Limit the number of blocks that we look up so as
1042 * not to get into the next delayed allocation extent.
1044 maxblocks
= INT_MAX
;
1045 if (delalloc_blklen
)
1046 maxblocks
= min_t(sector_t
, delalloc_blkoff
- blkoff
,
1050 down_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
1051 n
= nilfs_bmap_lookup_contig(
1052 NILFS_I(inode
)->i_bmap
, blkoff
, &blkphy
, maxblocks
);
1053 up_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
1058 if (unlikely(n
!= -ENOENT
))
1063 past_eof
= ((blkoff
<< blkbits
) >= isize
);
1066 /* End of the current extent */
1069 flags
|= FIEMAP_EXTENT_LAST
;
1071 ret
= fiemap_fill_next_extent(
1072 fieinfo
, logical
, phys
, size
, flags
);
1077 if (blkoff
> end_blkoff
|| past_eof
)
1081 if (phys
&& blkphy
<< blkbits
== phys
+ size
) {
1082 /* The current extent goes on */
1083 size
+= n
<< blkbits
;
1085 /* Terminate the current extent */
1086 ret
= fiemap_fill_next_extent(
1087 fieinfo
, logical
, phys
, size
,
1089 if (ret
|| blkoff
> end_blkoff
)
1092 /* Start another extent */
1093 flags
= FIEMAP_EXTENT_MERGED
;
1094 logical
= blkoff
<< blkbits
;
1095 phys
= blkphy
<< blkbits
;
1096 size
= n
<< blkbits
;
1099 /* Start a new extent */
1100 flags
= FIEMAP_EXTENT_MERGED
;
1101 logical
= blkoff
<< blkbits
;
1102 phys
= blkphy
<< blkbits
;
1103 size
= n
<< blkbits
;
1110 /* If ret is 1 then we just hit the end of the extent array */
1114 inode_unlock(inode
);