2 * inode.c - NILFS inode operations.
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * Written by Ryusuke Konishi.
20 #include <linux/buffer_head.h>
21 #include <linux/gfp.h>
22 #include <linux/mpage.h>
23 #include <linux/pagemap.h>
24 #include <linux/writeback.h>
25 #include <linux/uio.h>
35 * struct nilfs_iget_args - arguments used during comparison between inodes
37 * @cno: checkpoint number
38 * @root: pointer on NILFS root object (mounted checkpoint)
39 * @for_gc: inode for GC flag
41 struct nilfs_iget_args
{
44 struct nilfs_root
*root
;
48 static int nilfs_iget_test(struct inode
*inode
, void *opaque
);
50 void nilfs_inode_add_blocks(struct inode
*inode
, int n
)
52 struct nilfs_root
*root
= NILFS_I(inode
)->i_root
;
54 inode_add_bytes(inode
, (1 << inode
->i_blkbits
) * n
);
56 atomic64_add(n
, &root
->blocks_count
);
59 void nilfs_inode_sub_blocks(struct inode
*inode
, int n
)
61 struct nilfs_root
*root
= NILFS_I(inode
)->i_root
;
63 inode_sub_bytes(inode
, (1 << inode
->i_blkbits
) * n
);
65 atomic64_sub(n
, &root
->blocks_count
);
69 * nilfs_get_block() - get a file block on the filesystem (callback function)
70 * @inode - inode struct of the target file
71 * @blkoff - file block number
72 * @bh_result - buffer head to be mapped on
73 * @create - indicate whether allocating the block or not when it has not
76 * This function does not issue actual read request of the specified data
77 * block. It is done by VFS.
79 int nilfs_get_block(struct inode
*inode
, sector_t blkoff
,
80 struct buffer_head
*bh_result
, int create
)
82 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
83 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
86 unsigned int maxblocks
= bh_result
->b_size
>> inode
->i_blkbits
;
88 down_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
89 ret
= nilfs_bmap_lookup_contig(ii
->i_bmap
, blkoff
, &blknum
, maxblocks
);
90 up_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
91 if (ret
>= 0) { /* found */
92 map_bh(bh_result
, inode
->i_sb
, blknum
);
94 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
97 /* data block was not found */
98 if (ret
== -ENOENT
&& create
) {
99 struct nilfs_transaction_info ti
;
101 bh_result
->b_blocknr
= 0;
102 err
= nilfs_transaction_begin(inode
->i_sb
, &ti
, 1);
105 err
= nilfs_bmap_insert(ii
->i_bmap
, blkoff
,
106 (unsigned long)bh_result
);
107 if (unlikely(err
!= 0)) {
108 if (err
== -EEXIST
) {
110 * The get_block() function could be called
111 * from multiple callers for an inode.
112 * However, the page having this block must
113 * be locked in this case.
116 "nilfs_get_block: a race condition "
117 "while inserting a data block. "
118 "(inode number=%lu, file block "
121 (unsigned long long)blkoff
);
124 nilfs_transaction_abort(inode
->i_sb
);
127 nilfs_mark_inode_dirty_sync(inode
);
128 nilfs_transaction_commit(inode
->i_sb
); /* never fails */
129 /* Error handling should be detailed */
130 set_buffer_new(bh_result
);
131 set_buffer_delay(bh_result
);
132 map_bh(bh_result
, inode
->i_sb
, 0);
133 /* Disk block number must be changed to proper value */
135 } else if (ret
== -ENOENT
) {
137 * not found is not error (e.g. hole); must return without
138 * the mapped state flag.
150 * nilfs_readpage() - implement readpage() method of nilfs_aops {}
151 * address_space_operations.
152 * @file - file struct of the file to be read
153 * @page - the page to be read
155 static int nilfs_readpage(struct file
*file
, struct page
*page
)
157 return mpage_readpage(page
, nilfs_get_block
);
161 * nilfs_readpages() - implement readpages() method of nilfs_aops {}
162 * address_space_operations.
163 * @file - file struct of the file to be read
164 * @mapping - address_space struct used for reading multiple pages
165 * @pages - the pages to be read
166 * @nr_pages - number of pages to be read
168 static int nilfs_readpages(struct file
*file
, struct address_space
*mapping
,
169 struct list_head
*pages
, unsigned int nr_pages
)
171 return mpage_readpages(mapping
, pages
, nr_pages
, nilfs_get_block
);
174 static int nilfs_writepages(struct address_space
*mapping
,
175 struct writeback_control
*wbc
)
177 struct inode
*inode
= mapping
->host
;
180 if (inode
->i_sb
->s_flags
& MS_RDONLY
) {
181 nilfs_clear_dirty_pages(mapping
, false);
185 if (wbc
->sync_mode
== WB_SYNC_ALL
)
186 err
= nilfs_construct_dsync_segment(inode
->i_sb
, inode
,
192 static int nilfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
194 struct inode
*inode
= page
->mapping
->host
;
197 if (inode
->i_sb
->s_flags
& MS_RDONLY
) {
199 * It means that filesystem was remounted in read-only
200 * mode because of error or metadata corruption. But we
201 * have dirty pages that try to be flushed in background.
202 * So, here we simply discard this dirty page.
204 nilfs_clear_dirty_page(page
, false);
209 redirty_page_for_writepage(wbc
, page
);
212 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
213 err
= nilfs_construct_segment(inode
->i_sb
);
216 } else if (wbc
->for_reclaim
)
217 nilfs_flush_segment(inode
->i_sb
, inode
->i_ino
);
222 static int nilfs_set_page_dirty(struct page
*page
)
224 struct inode
*inode
= page
->mapping
->host
;
225 int ret
= __set_page_dirty_nobuffers(page
);
227 if (page_has_buffers(page
)) {
228 unsigned int nr_dirty
= 0;
229 struct buffer_head
*bh
, *head
;
232 * This page is locked by callers, and no other thread
233 * concurrently marks its buffers dirty since they are
234 * only dirtied through routines in fs/buffer.c in
235 * which call sites of mark_buffer_dirty are protected
238 bh
= head
= page_buffers(page
);
240 /* Do not mark hole blocks dirty */
241 if (buffer_dirty(bh
) || !buffer_mapped(bh
))
244 set_buffer_dirty(bh
);
246 } while (bh
= bh
->b_this_page
, bh
!= head
);
249 nilfs_set_file_dirty(inode
, nr_dirty
);
251 unsigned int nr_dirty
= 1 << (PAGE_SHIFT
- inode
->i_blkbits
);
253 nilfs_set_file_dirty(inode
, nr_dirty
);
258 void nilfs_write_failed(struct address_space
*mapping
, loff_t to
)
260 struct inode
*inode
= mapping
->host
;
262 if (to
> inode
->i_size
) {
263 truncate_pagecache(inode
, inode
->i_size
);
264 nilfs_truncate(inode
);
268 static int nilfs_write_begin(struct file
*file
, struct address_space
*mapping
,
269 loff_t pos
, unsigned len
, unsigned flags
,
270 struct page
**pagep
, void **fsdata
)
273 struct inode
*inode
= mapping
->host
;
274 int err
= nilfs_transaction_begin(inode
->i_sb
, NULL
, 1);
279 err
= block_write_begin(mapping
, pos
, len
, flags
, pagep
,
282 nilfs_write_failed(mapping
, pos
+ len
);
283 nilfs_transaction_abort(inode
->i_sb
);
288 static int nilfs_write_end(struct file
*file
, struct address_space
*mapping
,
289 loff_t pos
, unsigned len
, unsigned copied
,
290 struct page
*page
, void *fsdata
)
292 struct inode
*inode
= mapping
->host
;
293 unsigned int start
= pos
& (PAGE_SIZE
- 1);
294 unsigned int nr_dirty
;
297 nr_dirty
= nilfs_page_count_clean_buffers(page
, start
,
299 copied
= generic_write_end(file
, mapping
, pos
, len
, copied
, page
,
301 nilfs_set_file_dirty(inode
, nr_dirty
);
302 err
= nilfs_transaction_commit(inode
->i_sb
);
303 return err
? : copied
;
307 nilfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
309 struct inode
*inode
= file_inode(iocb
->ki_filp
);
311 if (iov_iter_rw(iter
) == WRITE
)
314 /* Needs synchronization with the cleaner */
315 return blockdev_direct_IO(iocb
, inode
, iter
, nilfs_get_block
);
318 const struct address_space_operations nilfs_aops
= {
319 .writepage
= nilfs_writepage
,
320 .readpage
= nilfs_readpage
,
321 .writepages
= nilfs_writepages
,
322 .set_page_dirty
= nilfs_set_page_dirty
,
323 .readpages
= nilfs_readpages
,
324 .write_begin
= nilfs_write_begin
,
325 .write_end
= nilfs_write_end
,
326 /* .releasepage = nilfs_releasepage, */
327 .invalidatepage
= block_invalidatepage
,
328 .direct_IO
= nilfs_direct_IO
,
329 .is_partially_uptodate
= block_is_partially_uptodate
,
332 static int nilfs_insert_inode_locked(struct inode
*inode
,
333 struct nilfs_root
*root
,
336 struct nilfs_iget_args args
= {
337 .ino
= ino
, .root
= root
, .cno
= 0, .for_gc
= 0
340 return insert_inode_locked4(inode
, ino
, nilfs_iget_test
, &args
);
343 struct inode
*nilfs_new_inode(struct inode
*dir
, umode_t mode
)
345 struct super_block
*sb
= dir
->i_sb
;
346 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
348 struct nilfs_inode_info
*ii
;
349 struct nilfs_root
*root
;
353 inode
= new_inode(sb
);
354 if (unlikely(!inode
))
357 mapping_set_gfp_mask(inode
->i_mapping
,
358 mapping_gfp_constraint(inode
->i_mapping
, ~__GFP_FS
));
360 root
= NILFS_I(dir
)->i_root
;
362 ii
->i_state
= 1 << NILFS_I_NEW
;
365 err
= nilfs_ifile_create_inode(root
->ifile
, &ino
, &ii
->i_bh
);
367 goto failed_ifile_create_inode
;
368 /* reference count of i_bh inherits from nilfs_mdt_read_block() */
370 atomic64_inc(&root
->inodes_count
);
371 inode_init_owner(inode
, dir
, mode
);
373 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
375 if (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)) {
376 err
= nilfs_bmap_read(ii
->i_bmap
, NULL
);
378 goto failed_after_creation
;
380 set_bit(NILFS_I_BMAP
, &ii
->i_state
);
381 /* No lock is needed; iget() ensures it. */
384 ii
->i_flags
= nilfs_mask_flags(
385 mode
, NILFS_I(dir
)->i_flags
& NILFS_FL_INHERITED
);
387 /* ii->i_file_acl = 0; */
388 /* ii->i_dir_acl = 0; */
389 ii
->i_dir_start_lookup
= 0;
390 nilfs_set_inode_flags(inode
);
391 spin_lock(&nilfs
->ns_next_gen_lock
);
392 inode
->i_generation
= nilfs
->ns_next_generation
++;
393 spin_unlock(&nilfs
->ns_next_gen_lock
);
394 if (nilfs_insert_inode_locked(inode
, root
, ino
) < 0) {
396 goto failed_after_creation
;
399 err
= nilfs_init_acl(inode
, dir
);
402 * Never occur. When supporting nilfs_init_acl(),
403 * proper cancellation of above jobs should be considered.
405 goto failed_after_creation
;
409 failed_after_creation
:
411 unlock_new_inode(inode
);
413 * raw_inode will be deleted through
414 * nilfs_evict_inode().
418 failed_ifile_create_inode
:
419 make_bad_inode(inode
);
425 void nilfs_set_inode_flags(struct inode
*inode
)
427 unsigned int flags
= NILFS_I(inode
)->i_flags
;
428 unsigned int new_fl
= 0;
430 if (flags
& FS_SYNC_FL
)
432 if (flags
& FS_APPEND_FL
)
434 if (flags
& FS_IMMUTABLE_FL
)
435 new_fl
|= S_IMMUTABLE
;
436 if (flags
& FS_NOATIME_FL
)
438 if (flags
& FS_DIRSYNC_FL
)
440 inode_set_flags(inode
, new_fl
, S_SYNC
| S_APPEND
| S_IMMUTABLE
|
441 S_NOATIME
| S_DIRSYNC
);
444 int nilfs_read_inode_common(struct inode
*inode
,
445 struct nilfs_inode
*raw_inode
)
447 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
450 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
451 i_uid_write(inode
, le32_to_cpu(raw_inode
->i_uid
));
452 i_gid_write(inode
, le32_to_cpu(raw_inode
->i_gid
));
453 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
454 inode
->i_size
= le64_to_cpu(raw_inode
->i_size
);
455 inode
->i_atime
.tv_sec
= le64_to_cpu(raw_inode
->i_mtime
);
456 inode
->i_ctime
.tv_sec
= le64_to_cpu(raw_inode
->i_ctime
);
457 inode
->i_mtime
.tv_sec
= le64_to_cpu(raw_inode
->i_mtime
);
458 inode
->i_atime
.tv_nsec
= le32_to_cpu(raw_inode
->i_mtime_nsec
);
459 inode
->i_ctime
.tv_nsec
= le32_to_cpu(raw_inode
->i_ctime_nsec
);
460 inode
->i_mtime
.tv_nsec
= le32_to_cpu(raw_inode
->i_mtime_nsec
);
461 if (inode
->i_nlink
== 0)
462 return -ESTALE
; /* this inode is deleted */
464 inode
->i_blocks
= le64_to_cpu(raw_inode
->i_blocks
);
465 ii
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
467 ii
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl
);
468 ii
->i_dir_acl
= S_ISREG(inode
->i_mode
) ?
469 0 : le32_to_cpu(raw_inode
->i_dir_acl
);
471 ii
->i_dir_start_lookup
= 0;
472 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
474 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
475 S_ISLNK(inode
->i_mode
)) {
476 err
= nilfs_bmap_read(ii
->i_bmap
, raw_inode
);
479 set_bit(NILFS_I_BMAP
, &ii
->i_state
);
480 /* No lock is needed; iget() ensures it. */
485 static int __nilfs_read_inode(struct super_block
*sb
,
486 struct nilfs_root
*root
, unsigned long ino
,
489 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
490 struct buffer_head
*bh
;
491 struct nilfs_inode
*raw_inode
;
494 down_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
495 err
= nilfs_ifile_get_inode_block(root
->ifile
, ino
, &bh
);
499 raw_inode
= nilfs_ifile_map_inode(root
->ifile
, ino
, bh
);
501 err
= nilfs_read_inode_common(inode
, raw_inode
);
505 if (S_ISREG(inode
->i_mode
)) {
506 inode
->i_op
= &nilfs_file_inode_operations
;
507 inode
->i_fop
= &nilfs_file_operations
;
508 inode
->i_mapping
->a_ops
= &nilfs_aops
;
509 } else if (S_ISDIR(inode
->i_mode
)) {
510 inode
->i_op
= &nilfs_dir_inode_operations
;
511 inode
->i_fop
= &nilfs_dir_operations
;
512 inode
->i_mapping
->a_ops
= &nilfs_aops
;
513 } else if (S_ISLNK(inode
->i_mode
)) {
514 inode
->i_op
= &nilfs_symlink_inode_operations
;
515 inode_nohighmem(inode
);
516 inode
->i_mapping
->a_ops
= &nilfs_aops
;
518 inode
->i_op
= &nilfs_special_inode_operations
;
520 inode
, inode
->i_mode
,
521 huge_decode_dev(le64_to_cpu(raw_inode
->i_device_code
)));
523 nilfs_ifile_unmap_inode(root
->ifile
, ino
, bh
);
525 up_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
526 nilfs_set_inode_flags(inode
);
527 mapping_set_gfp_mask(inode
->i_mapping
,
528 mapping_gfp_constraint(inode
->i_mapping
, ~__GFP_FS
));
532 nilfs_ifile_unmap_inode(root
->ifile
, ino
, bh
);
536 up_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
540 static int nilfs_iget_test(struct inode
*inode
, void *opaque
)
542 struct nilfs_iget_args
*args
= opaque
;
543 struct nilfs_inode_info
*ii
;
545 if (args
->ino
!= inode
->i_ino
|| args
->root
!= NILFS_I(inode
)->i_root
)
549 if (!test_bit(NILFS_I_GCINODE
, &ii
->i_state
))
550 return !args
->for_gc
;
552 return args
->for_gc
&& args
->cno
== ii
->i_cno
;
555 static int nilfs_iget_set(struct inode
*inode
, void *opaque
)
557 struct nilfs_iget_args
*args
= opaque
;
559 inode
->i_ino
= args
->ino
;
561 NILFS_I(inode
)->i_state
= 1 << NILFS_I_GCINODE
;
562 NILFS_I(inode
)->i_cno
= args
->cno
;
563 NILFS_I(inode
)->i_root
= NULL
;
565 if (args
->root
&& args
->ino
== NILFS_ROOT_INO
)
566 nilfs_get_root(args
->root
);
567 NILFS_I(inode
)->i_root
= args
->root
;
572 struct inode
*nilfs_ilookup(struct super_block
*sb
, struct nilfs_root
*root
,
575 struct nilfs_iget_args args
= {
576 .ino
= ino
, .root
= root
, .cno
= 0, .for_gc
= 0
579 return ilookup5(sb
, ino
, nilfs_iget_test
, &args
);
582 struct inode
*nilfs_iget_locked(struct super_block
*sb
, struct nilfs_root
*root
,
585 struct nilfs_iget_args args
= {
586 .ino
= ino
, .root
= root
, .cno
= 0, .for_gc
= 0
589 return iget5_locked(sb
, ino
, nilfs_iget_test
, nilfs_iget_set
, &args
);
592 struct inode
*nilfs_iget(struct super_block
*sb
, struct nilfs_root
*root
,
598 inode
= nilfs_iget_locked(sb
, root
, ino
);
599 if (unlikely(!inode
))
600 return ERR_PTR(-ENOMEM
);
601 if (!(inode
->i_state
& I_NEW
))
604 err
= __nilfs_read_inode(sb
, root
, ino
, inode
);
609 unlock_new_inode(inode
);
613 struct inode
*nilfs_iget_for_gc(struct super_block
*sb
, unsigned long ino
,
616 struct nilfs_iget_args args
= {
617 .ino
= ino
, .root
= NULL
, .cno
= cno
, .for_gc
= 1
622 inode
= iget5_locked(sb
, ino
, nilfs_iget_test
, nilfs_iget_set
, &args
);
623 if (unlikely(!inode
))
624 return ERR_PTR(-ENOMEM
);
625 if (!(inode
->i_state
& I_NEW
))
628 err
= nilfs_init_gcinode(inode
);
633 unlock_new_inode(inode
);
637 void nilfs_write_inode_common(struct inode
*inode
,
638 struct nilfs_inode
*raw_inode
, int has_bmap
)
640 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
642 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
643 raw_inode
->i_uid
= cpu_to_le32(i_uid_read(inode
));
644 raw_inode
->i_gid
= cpu_to_le32(i_gid_read(inode
));
645 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
646 raw_inode
->i_size
= cpu_to_le64(inode
->i_size
);
647 raw_inode
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
648 raw_inode
->i_mtime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
649 raw_inode
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
650 raw_inode
->i_mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
651 raw_inode
->i_blocks
= cpu_to_le64(inode
->i_blocks
);
653 raw_inode
->i_flags
= cpu_to_le32(ii
->i_flags
);
654 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
656 if (NILFS_ROOT_METADATA_FILE(inode
->i_ino
)) {
657 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
659 /* zero-fill unused portion in the case of super root block */
660 raw_inode
->i_xattr
= 0;
661 raw_inode
->i_pad
= 0;
662 memset((void *)raw_inode
+ sizeof(*raw_inode
), 0,
663 nilfs
->ns_inode_size
- sizeof(*raw_inode
));
667 nilfs_bmap_write(ii
->i_bmap
, raw_inode
);
668 else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
))
669 raw_inode
->i_device_code
=
670 cpu_to_le64(huge_encode_dev(inode
->i_rdev
));
672 * When extending inode, nilfs->ns_inode_size should be checked
673 * for substitutions of appended fields.
677 void nilfs_update_inode(struct inode
*inode
, struct buffer_head
*ibh
, int flags
)
679 ino_t ino
= inode
->i_ino
;
680 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
681 struct inode
*ifile
= ii
->i_root
->ifile
;
682 struct nilfs_inode
*raw_inode
;
684 raw_inode
= nilfs_ifile_map_inode(ifile
, ino
, ibh
);
686 if (test_and_clear_bit(NILFS_I_NEW
, &ii
->i_state
))
687 memset(raw_inode
, 0, NILFS_MDT(ifile
)->mi_entry_size
);
688 if (flags
& I_DIRTY_DATASYNC
)
689 set_bit(NILFS_I_INODE_SYNC
, &ii
->i_state
);
691 nilfs_write_inode_common(inode
, raw_inode
, 0);
693 * XXX: call with has_bmap = 0 is a workaround to avoid
694 * deadlock of bmap. This delays update of i_bmap to just
698 nilfs_ifile_unmap_inode(ifile
, ino
, ibh
);
701 #define NILFS_MAX_TRUNCATE_BLOCKS 16384 /* 64MB for 4KB block */
703 static void nilfs_truncate_bmap(struct nilfs_inode_info
*ii
,
709 if (!test_bit(NILFS_I_BMAP
, &ii
->i_state
))
712 ret
= nilfs_bmap_last_key(ii
->i_bmap
, &b
);
721 b
-= min_t(__u64
, NILFS_MAX_TRUNCATE_BLOCKS
, b
- from
);
722 ret
= nilfs_bmap_truncate(ii
->i_bmap
, b
);
723 nilfs_relax_pressure_in_lock(ii
->vfs_inode
.i_sb
);
724 if (!ret
|| (ret
== -ENOMEM
&&
725 nilfs_bmap_truncate(ii
->i_bmap
, b
) == 0))
729 nilfs_warning(ii
->vfs_inode
.i_sb
, __func__
,
730 "failed to truncate bmap (ino=%lu, err=%d)",
731 ii
->vfs_inode
.i_ino
, ret
);
734 void nilfs_truncate(struct inode
*inode
)
736 unsigned long blkoff
;
737 unsigned int blocksize
;
738 struct nilfs_transaction_info ti
;
739 struct super_block
*sb
= inode
->i_sb
;
740 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
742 if (!test_bit(NILFS_I_BMAP
, &ii
->i_state
))
744 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
747 blocksize
= sb
->s_blocksize
;
748 blkoff
= (inode
->i_size
+ blocksize
- 1) >> sb
->s_blocksize_bits
;
749 nilfs_transaction_begin(sb
, &ti
, 0); /* never fails */
751 block_truncate_page(inode
->i_mapping
, inode
->i_size
, nilfs_get_block
);
753 nilfs_truncate_bmap(ii
, blkoff
);
755 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
757 nilfs_set_transaction_flag(NILFS_TI_SYNC
);
759 nilfs_mark_inode_dirty(inode
);
760 nilfs_set_file_dirty(inode
, 0);
761 nilfs_transaction_commit(sb
);
763 * May construct a logical segment and may fail in sync mode.
764 * But truncate has no return value.
768 static void nilfs_clear_inode(struct inode
*inode
)
770 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
773 * Free resources allocated in nilfs_read_inode(), here.
775 BUG_ON(!list_empty(&ii
->i_dirty
));
779 if (nilfs_is_metadata_file_inode(inode
))
780 nilfs_mdt_clear(inode
);
782 if (test_bit(NILFS_I_BMAP
, &ii
->i_state
))
783 nilfs_bmap_clear(ii
->i_bmap
);
785 nilfs_btnode_cache_clear(&ii
->i_btnode_cache
);
787 if (ii
->i_root
&& inode
->i_ino
== NILFS_ROOT_INO
)
788 nilfs_put_root(ii
->i_root
);
791 void nilfs_evict_inode(struct inode
*inode
)
793 struct nilfs_transaction_info ti
;
794 struct super_block
*sb
= inode
->i_sb
;
795 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
798 if (inode
->i_nlink
|| !ii
->i_root
|| unlikely(is_bad_inode(inode
))) {
799 truncate_inode_pages_final(&inode
->i_data
);
801 nilfs_clear_inode(inode
);
804 nilfs_transaction_begin(sb
, &ti
, 0); /* never fails */
806 truncate_inode_pages_final(&inode
->i_data
);
808 /* TODO: some of the following operations may fail. */
809 nilfs_truncate_bmap(ii
, 0);
810 nilfs_mark_inode_dirty(inode
);
813 ret
= nilfs_ifile_delete_inode(ii
->i_root
->ifile
, inode
->i_ino
);
815 atomic64_dec(&ii
->i_root
->inodes_count
);
817 nilfs_clear_inode(inode
);
820 nilfs_set_transaction_flag(NILFS_TI_SYNC
);
821 nilfs_transaction_commit(sb
);
823 * May construct a logical segment and may fail in sync mode.
824 * But delete_inode has no return value.
828 int nilfs_setattr(struct dentry
*dentry
, struct iattr
*iattr
)
830 struct nilfs_transaction_info ti
;
831 struct inode
*inode
= d_inode(dentry
);
832 struct super_block
*sb
= inode
->i_sb
;
835 err
= inode_change_ok(inode
, iattr
);
839 err
= nilfs_transaction_begin(sb
, &ti
, 0);
843 if ((iattr
->ia_valid
& ATTR_SIZE
) &&
844 iattr
->ia_size
!= i_size_read(inode
)) {
845 inode_dio_wait(inode
);
846 truncate_setsize(inode
, iattr
->ia_size
);
847 nilfs_truncate(inode
);
850 setattr_copy(inode
, iattr
);
851 mark_inode_dirty(inode
);
853 if (iattr
->ia_valid
& ATTR_MODE
) {
854 err
= nilfs_acl_chmod(inode
);
859 return nilfs_transaction_commit(sb
);
862 nilfs_transaction_abort(sb
);
866 int nilfs_permission(struct inode
*inode
, int mask
)
868 struct nilfs_root
*root
= NILFS_I(inode
)->i_root
;
870 if ((mask
& MAY_WRITE
) && root
&&
871 root
->cno
!= NILFS_CPTREE_CURRENT_CNO
)
872 return -EROFS
; /* snapshot is not writable */
874 return generic_permission(inode
, mask
);
877 int nilfs_load_inode_block(struct inode
*inode
, struct buffer_head
**pbh
)
879 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
880 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
883 spin_lock(&nilfs
->ns_inode_lock
);
884 if (ii
->i_bh
== NULL
) {
885 spin_unlock(&nilfs
->ns_inode_lock
);
886 err
= nilfs_ifile_get_inode_block(ii
->i_root
->ifile
,
890 spin_lock(&nilfs
->ns_inode_lock
);
891 if (ii
->i_bh
== NULL
)
901 spin_unlock(&nilfs
->ns_inode_lock
);
905 int nilfs_inode_dirty(struct inode
*inode
)
907 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
908 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
911 if (!list_empty(&ii
->i_dirty
)) {
912 spin_lock(&nilfs
->ns_inode_lock
);
913 ret
= test_bit(NILFS_I_DIRTY
, &ii
->i_state
) ||
914 test_bit(NILFS_I_BUSY
, &ii
->i_state
);
915 spin_unlock(&nilfs
->ns_inode_lock
);
920 int nilfs_set_file_dirty(struct inode
*inode
, unsigned int nr_dirty
)
922 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
923 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
925 atomic_add(nr_dirty
, &nilfs
->ns_ndirtyblks
);
927 if (test_and_set_bit(NILFS_I_DIRTY
, &ii
->i_state
))
930 spin_lock(&nilfs
->ns_inode_lock
);
931 if (!test_bit(NILFS_I_QUEUED
, &ii
->i_state
) &&
932 !test_bit(NILFS_I_BUSY
, &ii
->i_state
)) {
934 * Because this routine may race with nilfs_dispose_list(),
935 * we have to check NILFS_I_QUEUED here, too.
937 if (list_empty(&ii
->i_dirty
) && igrab(inode
) == NULL
) {
939 * This will happen when somebody is freeing
942 nilfs_warning(inode
->i_sb
, __func__
,
943 "cannot get inode (ino=%lu)",
945 spin_unlock(&nilfs
->ns_inode_lock
);
947 * NILFS_I_DIRTY may remain for
951 list_move_tail(&ii
->i_dirty
, &nilfs
->ns_dirty_files
);
952 set_bit(NILFS_I_QUEUED
, &ii
->i_state
);
954 spin_unlock(&nilfs
->ns_inode_lock
);
958 int __nilfs_mark_inode_dirty(struct inode
*inode
, int flags
)
960 struct buffer_head
*ibh
;
963 err
= nilfs_load_inode_block(inode
, &ibh
);
965 nilfs_warning(inode
->i_sb
, __func__
,
966 "failed to reget inode block.");
969 nilfs_update_inode(inode
, ibh
, flags
);
970 mark_buffer_dirty(ibh
);
971 nilfs_mdt_mark_dirty(NILFS_I(inode
)->i_root
->ifile
);
977 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
978 * @inode: inode of the file to be registered.
980 * nilfs_dirty_inode() loads a inode block containing the specified
981 * @inode and copies data from a nilfs_inode to a corresponding inode
982 * entry in the inode block. This operation is excluded from the segment
983 * construction. This function can be called both as a single operation
984 * and as a part of indivisible file operations.
986 void nilfs_dirty_inode(struct inode
*inode
, int flags
)
988 struct nilfs_transaction_info ti
;
989 struct nilfs_mdt_info
*mdi
= NILFS_MDT(inode
);
991 if (is_bad_inode(inode
)) {
992 nilfs_warning(inode
->i_sb
, __func__
,
993 "tried to mark bad_inode dirty. ignored.");
998 nilfs_mdt_mark_dirty(inode
);
1001 nilfs_transaction_begin(inode
->i_sb
, &ti
, 0);
1002 __nilfs_mark_inode_dirty(inode
, flags
);
1003 nilfs_transaction_commit(inode
->i_sb
); /* never fails */
1006 int nilfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
1007 __u64 start
, __u64 len
)
1009 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
1010 __u64 logical
= 0, phys
= 0, size
= 0;
1013 sector_t blkoff
, end_blkoff
;
1014 sector_t delalloc_blkoff
;
1015 unsigned long delalloc_blklen
;
1016 unsigned int blkbits
= inode
->i_blkbits
;
1019 ret
= fiemap_check_flags(fieinfo
, FIEMAP_FLAG_SYNC
);
1025 isize
= i_size_read(inode
);
1027 blkoff
= start
>> blkbits
;
1028 end_blkoff
= (start
+ len
- 1) >> blkbits
;
1030 delalloc_blklen
= nilfs_find_uncommitted_extent(inode
, blkoff
,
1035 unsigned int maxblocks
;
1037 if (delalloc_blklen
&& blkoff
== delalloc_blkoff
) {
1039 /* End of the current extent */
1040 ret
= fiemap_fill_next_extent(
1041 fieinfo
, logical
, phys
, size
, flags
);
1045 if (blkoff
> end_blkoff
)
1048 flags
= FIEMAP_EXTENT_MERGED
| FIEMAP_EXTENT_DELALLOC
;
1049 logical
= blkoff
<< blkbits
;
1051 size
= delalloc_blklen
<< blkbits
;
1053 blkoff
= delalloc_blkoff
+ delalloc_blklen
;
1054 delalloc_blklen
= nilfs_find_uncommitted_extent(
1055 inode
, blkoff
, &delalloc_blkoff
);
1060 * Limit the number of blocks that we look up so as
1061 * not to get into the next delayed allocation extent.
1063 maxblocks
= INT_MAX
;
1064 if (delalloc_blklen
)
1065 maxblocks
= min_t(sector_t
, delalloc_blkoff
- blkoff
,
1069 down_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
1070 n
= nilfs_bmap_lookup_contig(
1071 NILFS_I(inode
)->i_bmap
, blkoff
, &blkphy
, maxblocks
);
1072 up_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
1077 if (unlikely(n
!= -ENOENT
))
1082 past_eof
= ((blkoff
<< blkbits
) >= isize
);
1085 /* End of the current extent */
1088 flags
|= FIEMAP_EXTENT_LAST
;
1090 ret
= fiemap_fill_next_extent(
1091 fieinfo
, logical
, phys
, size
, flags
);
1096 if (blkoff
> end_blkoff
|| past_eof
)
1100 if (phys
&& blkphy
<< blkbits
== phys
+ size
) {
1101 /* The current extent goes on */
1102 size
+= n
<< blkbits
;
1104 /* Terminate the current extent */
1105 ret
= fiemap_fill_next_extent(
1106 fieinfo
, logical
, phys
, size
,
1108 if (ret
|| blkoff
> end_blkoff
)
1111 /* Start another extent */
1112 flags
= FIEMAP_EXTENT_MERGED
;
1113 logical
= blkoff
<< blkbits
;
1114 phys
= blkphy
<< blkbits
;
1115 size
= n
<< blkbits
;
1118 /* Start a new extent */
1119 flags
= FIEMAP_EXTENT_MERGED
;
1120 logical
= blkoff
<< blkbits
;
1121 phys
= blkphy
<< blkbits
;
1122 size
= n
<< blkbits
;
1129 /* If ret is 1 then we just hit the end of the extent array */
1133 inode_unlock(inode
);