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 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); /* dbn must be changed
134 } else if (ret
== -ENOENT
) {
135 /* not found is not error (e.g. hole); must return without
136 the mapped state flag. */
147 * nilfs_readpage() - implement readpage() method of nilfs_aops {}
148 * address_space_operations.
149 * @file - file struct of the file to be read
150 * @page - the page to be read
152 static int nilfs_readpage(struct file
*file
, struct page
*page
)
154 return mpage_readpage(page
, nilfs_get_block
);
158 * nilfs_readpages() - implement readpages() method of nilfs_aops {}
159 * address_space_operations.
160 * @file - file struct of the file to be read
161 * @mapping - address_space struct used for reading multiple pages
162 * @pages - the pages to be read
163 * @nr_pages - number of pages to be read
165 static int nilfs_readpages(struct file
*file
, struct address_space
*mapping
,
166 struct list_head
*pages
, unsigned nr_pages
)
168 return mpage_readpages(mapping
, pages
, nr_pages
, nilfs_get_block
);
171 static int nilfs_writepages(struct address_space
*mapping
,
172 struct writeback_control
*wbc
)
174 struct inode
*inode
= mapping
->host
;
177 if (inode
->i_sb
->s_flags
& MS_RDONLY
) {
178 nilfs_clear_dirty_pages(mapping
, false);
182 if (wbc
->sync_mode
== WB_SYNC_ALL
)
183 err
= nilfs_construct_dsync_segment(inode
->i_sb
, inode
,
189 static int nilfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
191 struct inode
*inode
= page
->mapping
->host
;
194 if (inode
->i_sb
->s_flags
& MS_RDONLY
) {
196 * It means that filesystem was remounted in read-only
197 * mode because of error or metadata corruption. But we
198 * have dirty pages that try to be flushed in background.
199 * So, here we simply discard this dirty page.
201 nilfs_clear_dirty_page(page
, false);
206 redirty_page_for_writepage(wbc
, page
);
209 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
210 err
= nilfs_construct_segment(inode
->i_sb
);
213 } else if (wbc
->for_reclaim
)
214 nilfs_flush_segment(inode
->i_sb
, inode
->i_ino
);
219 static int nilfs_set_page_dirty(struct page
*page
)
221 struct inode
*inode
= page
->mapping
->host
;
222 int ret
= __set_page_dirty_nobuffers(page
);
224 if (page_has_buffers(page
)) {
225 unsigned nr_dirty
= 0;
226 struct buffer_head
*bh
, *head
;
229 * This page is locked by callers, and no other thread
230 * concurrently marks its buffers dirty since they are
231 * only dirtied through routines in fs/buffer.c in
232 * which call sites of mark_buffer_dirty are protected
235 bh
= head
= page_buffers(page
);
237 /* Do not mark hole blocks dirty */
238 if (buffer_dirty(bh
) || !buffer_mapped(bh
))
241 set_buffer_dirty(bh
);
243 } while (bh
= bh
->b_this_page
, bh
!= head
);
246 nilfs_set_file_dirty(inode
, nr_dirty
);
248 unsigned nr_dirty
= 1 << (PAGE_SHIFT
- inode
->i_blkbits
);
250 nilfs_set_file_dirty(inode
, nr_dirty
);
255 void nilfs_write_failed(struct address_space
*mapping
, loff_t to
)
257 struct inode
*inode
= mapping
->host
;
259 if (to
> inode
->i_size
) {
260 truncate_pagecache(inode
, inode
->i_size
);
261 nilfs_truncate(inode
);
265 static int nilfs_write_begin(struct file
*file
, struct address_space
*mapping
,
266 loff_t pos
, unsigned len
, unsigned flags
,
267 struct page
**pagep
, void **fsdata
)
270 struct inode
*inode
= mapping
->host
;
271 int err
= nilfs_transaction_begin(inode
->i_sb
, NULL
, 1);
276 err
= block_write_begin(mapping
, pos
, len
, flags
, pagep
,
279 nilfs_write_failed(mapping
, pos
+ len
);
280 nilfs_transaction_abort(inode
->i_sb
);
285 static int nilfs_write_end(struct file
*file
, struct address_space
*mapping
,
286 loff_t pos
, unsigned len
, unsigned copied
,
287 struct page
*page
, void *fsdata
)
289 struct inode
*inode
= mapping
->host
;
290 unsigned start
= pos
& (PAGE_SIZE
- 1);
294 nr_dirty
= nilfs_page_count_clean_buffers(page
, start
,
296 copied
= generic_write_end(file
, mapping
, pos
, len
, copied
, page
,
298 nilfs_set_file_dirty(inode
, nr_dirty
);
299 err
= nilfs_transaction_commit(inode
->i_sb
);
300 return err
? : copied
;
304 nilfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
306 struct inode
*inode
= file_inode(iocb
->ki_filp
);
308 if (iov_iter_rw(iter
) == WRITE
)
311 /* Needs synchronization with the cleaner */
312 return blockdev_direct_IO(iocb
, inode
, iter
, nilfs_get_block
);
315 const struct address_space_operations nilfs_aops
= {
316 .writepage
= nilfs_writepage
,
317 .readpage
= nilfs_readpage
,
318 .writepages
= nilfs_writepages
,
319 .set_page_dirty
= nilfs_set_page_dirty
,
320 .readpages
= nilfs_readpages
,
321 .write_begin
= nilfs_write_begin
,
322 .write_end
= nilfs_write_end
,
323 /* .releasepage = nilfs_releasepage, */
324 .invalidatepage
= block_invalidatepage
,
325 .direct_IO
= nilfs_direct_IO
,
326 .is_partially_uptodate
= block_is_partially_uptodate
,
329 static int nilfs_insert_inode_locked(struct inode
*inode
,
330 struct nilfs_root
*root
,
333 struct nilfs_iget_args args
= {
334 .ino
= ino
, .root
= root
, .cno
= 0, .for_gc
= 0
337 return insert_inode_locked4(inode
, ino
, nilfs_iget_test
, &args
);
340 struct inode
*nilfs_new_inode(struct inode
*dir
, umode_t mode
)
342 struct super_block
*sb
= dir
->i_sb
;
343 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
345 struct nilfs_inode_info
*ii
;
346 struct nilfs_root
*root
;
350 inode
= new_inode(sb
);
351 if (unlikely(!inode
))
354 mapping_set_gfp_mask(inode
->i_mapping
,
355 mapping_gfp_constraint(inode
->i_mapping
, ~__GFP_FS
));
357 root
= NILFS_I(dir
)->i_root
;
359 ii
->i_state
= 1 << NILFS_I_NEW
;
362 err
= nilfs_ifile_create_inode(root
->ifile
, &ino
, &ii
->i_bh
);
364 goto failed_ifile_create_inode
;
365 /* reference count of i_bh inherits from nilfs_mdt_read_block() */
367 atomic64_inc(&root
->inodes_count
);
368 inode_init_owner(inode
, dir
, mode
);
370 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
372 if (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)) {
373 err
= nilfs_bmap_read(ii
->i_bmap
, NULL
);
375 goto failed_after_creation
;
377 set_bit(NILFS_I_BMAP
, &ii
->i_state
);
378 /* No lock is needed; iget() ensures it. */
381 ii
->i_flags
= nilfs_mask_flags(
382 mode
, NILFS_I(dir
)->i_flags
& NILFS_FL_INHERITED
);
384 /* ii->i_file_acl = 0; */
385 /* ii->i_dir_acl = 0; */
386 ii
->i_dir_start_lookup
= 0;
387 nilfs_set_inode_flags(inode
);
388 spin_lock(&nilfs
->ns_next_gen_lock
);
389 inode
->i_generation
= nilfs
->ns_next_generation
++;
390 spin_unlock(&nilfs
->ns_next_gen_lock
);
391 if (nilfs_insert_inode_locked(inode
, root
, ino
) < 0) {
393 goto failed_after_creation
;
396 err
= nilfs_init_acl(inode
, dir
);
398 goto failed_after_creation
; /* never occur. When supporting
399 nilfs_init_acl(), proper cancellation of
400 above jobs should be considered */
404 failed_after_creation
:
406 unlock_new_inode(inode
);
407 iput(inode
); /* raw_inode will be deleted through
408 nilfs_evict_inode() */
411 failed_ifile_create_inode
:
412 make_bad_inode(inode
);
413 iput(inode
); /* if i_nlink == 1, generic_forget_inode() will be
419 void nilfs_set_inode_flags(struct inode
*inode
)
421 unsigned int flags
= NILFS_I(inode
)->i_flags
;
422 unsigned int new_fl
= 0;
424 if (flags
& FS_SYNC_FL
)
426 if (flags
& FS_APPEND_FL
)
428 if (flags
& FS_IMMUTABLE_FL
)
429 new_fl
|= S_IMMUTABLE
;
430 if (flags
& FS_NOATIME_FL
)
432 if (flags
& FS_DIRSYNC_FL
)
434 inode_set_flags(inode
, new_fl
, S_SYNC
| S_APPEND
| S_IMMUTABLE
|
435 S_NOATIME
| S_DIRSYNC
);
438 int nilfs_read_inode_common(struct inode
*inode
,
439 struct nilfs_inode
*raw_inode
)
441 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
444 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
445 i_uid_write(inode
, le32_to_cpu(raw_inode
->i_uid
));
446 i_gid_write(inode
, le32_to_cpu(raw_inode
->i_gid
));
447 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
448 inode
->i_size
= le64_to_cpu(raw_inode
->i_size
);
449 inode
->i_atime
.tv_sec
= le64_to_cpu(raw_inode
->i_mtime
);
450 inode
->i_ctime
.tv_sec
= le64_to_cpu(raw_inode
->i_ctime
);
451 inode
->i_mtime
.tv_sec
= le64_to_cpu(raw_inode
->i_mtime
);
452 inode
->i_atime
.tv_nsec
= le32_to_cpu(raw_inode
->i_mtime_nsec
);
453 inode
->i_ctime
.tv_nsec
= le32_to_cpu(raw_inode
->i_ctime_nsec
);
454 inode
->i_mtime
.tv_nsec
= le32_to_cpu(raw_inode
->i_mtime_nsec
);
455 if (inode
->i_nlink
== 0)
456 return -ESTALE
; /* this inode is deleted */
458 inode
->i_blocks
= le64_to_cpu(raw_inode
->i_blocks
);
459 ii
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
461 ii
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl
);
462 ii
->i_dir_acl
= S_ISREG(inode
->i_mode
) ?
463 0 : le32_to_cpu(raw_inode
->i_dir_acl
);
465 ii
->i_dir_start_lookup
= 0;
466 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
468 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
469 S_ISLNK(inode
->i_mode
)) {
470 err
= nilfs_bmap_read(ii
->i_bmap
, raw_inode
);
473 set_bit(NILFS_I_BMAP
, &ii
->i_state
);
474 /* No lock is needed; iget() ensures it. */
479 static int __nilfs_read_inode(struct super_block
*sb
,
480 struct nilfs_root
*root
, unsigned long ino
,
483 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
484 struct buffer_head
*bh
;
485 struct nilfs_inode
*raw_inode
;
488 down_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
489 err
= nilfs_ifile_get_inode_block(root
->ifile
, ino
, &bh
);
493 raw_inode
= nilfs_ifile_map_inode(root
->ifile
, ino
, bh
);
495 err
= nilfs_read_inode_common(inode
, raw_inode
);
499 if (S_ISREG(inode
->i_mode
)) {
500 inode
->i_op
= &nilfs_file_inode_operations
;
501 inode
->i_fop
= &nilfs_file_operations
;
502 inode
->i_mapping
->a_ops
= &nilfs_aops
;
503 } else if (S_ISDIR(inode
->i_mode
)) {
504 inode
->i_op
= &nilfs_dir_inode_operations
;
505 inode
->i_fop
= &nilfs_dir_operations
;
506 inode
->i_mapping
->a_ops
= &nilfs_aops
;
507 } else if (S_ISLNK(inode
->i_mode
)) {
508 inode
->i_op
= &nilfs_symlink_inode_operations
;
509 inode_nohighmem(inode
);
510 inode
->i_mapping
->a_ops
= &nilfs_aops
;
512 inode
->i_op
= &nilfs_special_inode_operations
;
514 inode
, inode
->i_mode
,
515 huge_decode_dev(le64_to_cpu(raw_inode
->i_device_code
)));
517 nilfs_ifile_unmap_inode(root
->ifile
, ino
, bh
);
519 up_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
520 nilfs_set_inode_flags(inode
);
521 mapping_set_gfp_mask(inode
->i_mapping
,
522 mapping_gfp_constraint(inode
->i_mapping
, ~__GFP_FS
));
526 nilfs_ifile_unmap_inode(root
->ifile
, ino
, bh
);
530 up_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
534 static int nilfs_iget_test(struct inode
*inode
, void *opaque
)
536 struct nilfs_iget_args
*args
= opaque
;
537 struct nilfs_inode_info
*ii
;
539 if (args
->ino
!= inode
->i_ino
|| args
->root
!= NILFS_I(inode
)->i_root
)
543 if (!test_bit(NILFS_I_GCINODE
, &ii
->i_state
))
544 return !args
->for_gc
;
546 return args
->for_gc
&& args
->cno
== ii
->i_cno
;
549 static int nilfs_iget_set(struct inode
*inode
, void *opaque
)
551 struct nilfs_iget_args
*args
= opaque
;
553 inode
->i_ino
= args
->ino
;
555 NILFS_I(inode
)->i_state
= 1 << NILFS_I_GCINODE
;
556 NILFS_I(inode
)->i_cno
= args
->cno
;
557 NILFS_I(inode
)->i_root
= NULL
;
559 if (args
->root
&& args
->ino
== NILFS_ROOT_INO
)
560 nilfs_get_root(args
->root
);
561 NILFS_I(inode
)->i_root
= args
->root
;
566 struct inode
*nilfs_ilookup(struct super_block
*sb
, struct nilfs_root
*root
,
569 struct nilfs_iget_args args
= {
570 .ino
= ino
, .root
= root
, .cno
= 0, .for_gc
= 0
573 return ilookup5(sb
, ino
, nilfs_iget_test
, &args
);
576 struct inode
*nilfs_iget_locked(struct super_block
*sb
, struct nilfs_root
*root
,
579 struct nilfs_iget_args args
= {
580 .ino
= ino
, .root
= root
, .cno
= 0, .for_gc
= 0
583 return iget5_locked(sb
, ino
, nilfs_iget_test
, nilfs_iget_set
, &args
);
586 struct inode
*nilfs_iget(struct super_block
*sb
, struct nilfs_root
*root
,
592 inode
= nilfs_iget_locked(sb
, root
, ino
);
593 if (unlikely(!inode
))
594 return ERR_PTR(-ENOMEM
);
595 if (!(inode
->i_state
& I_NEW
))
598 err
= __nilfs_read_inode(sb
, root
, ino
, inode
);
603 unlock_new_inode(inode
);
607 struct inode
*nilfs_iget_for_gc(struct super_block
*sb
, unsigned long ino
,
610 struct nilfs_iget_args args
= {
611 .ino
= ino
, .root
= NULL
, .cno
= cno
, .for_gc
= 1
616 inode
= iget5_locked(sb
, ino
, nilfs_iget_test
, nilfs_iget_set
, &args
);
617 if (unlikely(!inode
))
618 return ERR_PTR(-ENOMEM
);
619 if (!(inode
->i_state
& I_NEW
))
622 err
= nilfs_init_gcinode(inode
);
627 unlock_new_inode(inode
);
631 void nilfs_write_inode_common(struct inode
*inode
,
632 struct nilfs_inode
*raw_inode
, int has_bmap
)
634 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
636 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
637 raw_inode
->i_uid
= cpu_to_le32(i_uid_read(inode
));
638 raw_inode
->i_gid
= cpu_to_le32(i_gid_read(inode
));
639 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
640 raw_inode
->i_size
= cpu_to_le64(inode
->i_size
);
641 raw_inode
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
642 raw_inode
->i_mtime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
643 raw_inode
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
644 raw_inode
->i_mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
645 raw_inode
->i_blocks
= cpu_to_le64(inode
->i_blocks
);
647 raw_inode
->i_flags
= cpu_to_le32(ii
->i_flags
);
648 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
650 if (NILFS_ROOT_METADATA_FILE(inode
->i_ino
)) {
651 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
653 /* zero-fill unused portion in the case of super root block */
654 raw_inode
->i_xattr
= 0;
655 raw_inode
->i_pad
= 0;
656 memset((void *)raw_inode
+ sizeof(*raw_inode
), 0,
657 nilfs
->ns_inode_size
- sizeof(*raw_inode
));
661 nilfs_bmap_write(ii
->i_bmap
, raw_inode
);
662 else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
))
663 raw_inode
->i_device_code
=
664 cpu_to_le64(huge_encode_dev(inode
->i_rdev
));
665 /* When extending inode, nilfs->ns_inode_size should be checked
666 for substitutions of appended fields */
669 void nilfs_update_inode(struct inode
*inode
, struct buffer_head
*ibh
, int flags
)
671 ino_t ino
= inode
->i_ino
;
672 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
673 struct inode
*ifile
= ii
->i_root
->ifile
;
674 struct nilfs_inode
*raw_inode
;
676 raw_inode
= nilfs_ifile_map_inode(ifile
, ino
, ibh
);
678 if (test_and_clear_bit(NILFS_I_NEW
, &ii
->i_state
))
679 memset(raw_inode
, 0, NILFS_MDT(ifile
)->mi_entry_size
);
680 if (flags
& I_DIRTY_DATASYNC
)
681 set_bit(NILFS_I_INODE_SYNC
, &ii
->i_state
);
683 nilfs_write_inode_common(inode
, raw_inode
, 0);
684 /* XXX: call with has_bmap = 0 is a workaround to avoid
685 deadlock of bmap. This delays update of i_bmap to just
687 nilfs_ifile_unmap_inode(ifile
, ino
, ibh
);
690 #define NILFS_MAX_TRUNCATE_BLOCKS 16384 /* 64MB for 4KB block */
692 static void nilfs_truncate_bmap(struct nilfs_inode_info
*ii
,
698 if (!test_bit(NILFS_I_BMAP
, &ii
->i_state
))
701 ret
= nilfs_bmap_last_key(ii
->i_bmap
, &b
);
710 b
-= min_t(__u64
, NILFS_MAX_TRUNCATE_BLOCKS
, b
- from
);
711 ret
= nilfs_bmap_truncate(ii
->i_bmap
, b
);
712 nilfs_relax_pressure_in_lock(ii
->vfs_inode
.i_sb
);
713 if (!ret
|| (ret
== -ENOMEM
&&
714 nilfs_bmap_truncate(ii
->i_bmap
, b
) == 0))
718 nilfs_warning(ii
->vfs_inode
.i_sb
, __func__
,
719 "failed to truncate bmap (ino=%lu, err=%d)",
720 ii
->vfs_inode
.i_ino
, ret
);
723 void nilfs_truncate(struct inode
*inode
)
725 unsigned long blkoff
;
726 unsigned int blocksize
;
727 struct nilfs_transaction_info ti
;
728 struct super_block
*sb
= inode
->i_sb
;
729 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
731 if (!test_bit(NILFS_I_BMAP
, &ii
->i_state
))
733 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
736 blocksize
= sb
->s_blocksize
;
737 blkoff
= (inode
->i_size
+ blocksize
- 1) >> sb
->s_blocksize_bits
;
738 nilfs_transaction_begin(sb
, &ti
, 0); /* never fails */
740 block_truncate_page(inode
->i_mapping
, inode
->i_size
, nilfs_get_block
);
742 nilfs_truncate_bmap(ii
, blkoff
);
744 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
746 nilfs_set_transaction_flag(NILFS_TI_SYNC
);
748 nilfs_mark_inode_dirty(inode
);
749 nilfs_set_file_dirty(inode
, 0);
750 nilfs_transaction_commit(sb
);
751 /* May construct a logical segment and may fail in sync mode.
752 But truncate has no return value. */
755 static void nilfs_clear_inode(struct inode
*inode
)
757 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
758 struct nilfs_mdt_info
*mdi
= NILFS_MDT(inode
);
761 * Free resources allocated in nilfs_read_inode(), here.
763 BUG_ON(!list_empty(&ii
->i_dirty
));
767 if (mdi
&& mdi
->mi_palloc_cache
)
768 nilfs_palloc_destroy_cache(inode
);
770 if (test_bit(NILFS_I_BMAP
, &ii
->i_state
))
771 nilfs_bmap_clear(ii
->i_bmap
);
773 nilfs_btnode_cache_clear(&ii
->i_btnode_cache
);
775 if (ii
->i_root
&& inode
->i_ino
== NILFS_ROOT_INO
)
776 nilfs_put_root(ii
->i_root
);
779 void nilfs_evict_inode(struct inode
*inode
)
781 struct nilfs_transaction_info ti
;
782 struct super_block
*sb
= inode
->i_sb
;
783 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
786 if (inode
->i_nlink
|| !ii
->i_root
|| unlikely(is_bad_inode(inode
))) {
787 truncate_inode_pages_final(&inode
->i_data
);
789 nilfs_clear_inode(inode
);
792 nilfs_transaction_begin(sb
, &ti
, 0); /* never fails */
794 truncate_inode_pages_final(&inode
->i_data
);
796 /* TODO: some of the following operations may fail. */
797 nilfs_truncate_bmap(ii
, 0);
798 nilfs_mark_inode_dirty(inode
);
801 ret
= nilfs_ifile_delete_inode(ii
->i_root
->ifile
, inode
->i_ino
);
803 atomic64_dec(&ii
->i_root
->inodes_count
);
805 nilfs_clear_inode(inode
);
808 nilfs_set_transaction_flag(NILFS_TI_SYNC
);
809 nilfs_transaction_commit(sb
);
810 /* May construct a logical segment and may fail in sync mode.
811 But delete_inode has no return value. */
814 int nilfs_setattr(struct dentry
*dentry
, struct iattr
*iattr
)
816 struct nilfs_transaction_info ti
;
817 struct inode
*inode
= d_inode(dentry
);
818 struct super_block
*sb
= inode
->i_sb
;
821 err
= inode_change_ok(inode
, iattr
);
825 err
= nilfs_transaction_begin(sb
, &ti
, 0);
829 if ((iattr
->ia_valid
& ATTR_SIZE
) &&
830 iattr
->ia_size
!= i_size_read(inode
)) {
831 inode_dio_wait(inode
);
832 truncate_setsize(inode
, iattr
->ia_size
);
833 nilfs_truncate(inode
);
836 setattr_copy(inode
, iattr
);
837 mark_inode_dirty(inode
);
839 if (iattr
->ia_valid
& ATTR_MODE
) {
840 err
= nilfs_acl_chmod(inode
);
845 return nilfs_transaction_commit(sb
);
848 nilfs_transaction_abort(sb
);
852 int nilfs_permission(struct inode
*inode
, int mask
)
854 struct nilfs_root
*root
= NILFS_I(inode
)->i_root
;
855 if ((mask
& MAY_WRITE
) && root
&&
856 root
->cno
!= NILFS_CPTREE_CURRENT_CNO
)
857 return -EROFS
; /* snapshot is not writable */
859 return generic_permission(inode
, mask
);
862 int nilfs_load_inode_block(struct inode
*inode
, struct buffer_head
**pbh
)
864 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
865 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
868 spin_lock(&nilfs
->ns_inode_lock
);
869 if (ii
->i_bh
== NULL
) {
870 spin_unlock(&nilfs
->ns_inode_lock
);
871 err
= nilfs_ifile_get_inode_block(ii
->i_root
->ifile
,
875 spin_lock(&nilfs
->ns_inode_lock
);
876 if (ii
->i_bh
== NULL
)
886 spin_unlock(&nilfs
->ns_inode_lock
);
890 int nilfs_inode_dirty(struct inode
*inode
)
892 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
893 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
896 if (!list_empty(&ii
->i_dirty
)) {
897 spin_lock(&nilfs
->ns_inode_lock
);
898 ret
= test_bit(NILFS_I_DIRTY
, &ii
->i_state
) ||
899 test_bit(NILFS_I_BUSY
, &ii
->i_state
);
900 spin_unlock(&nilfs
->ns_inode_lock
);
905 int nilfs_set_file_dirty(struct inode
*inode
, unsigned nr_dirty
)
907 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
908 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
910 atomic_add(nr_dirty
, &nilfs
->ns_ndirtyblks
);
912 if (test_and_set_bit(NILFS_I_DIRTY
, &ii
->i_state
))
915 spin_lock(&nilfs
->ns_inode_lock
);
916 if (!test_bit(NILFS_I_QUEUED
, &ii
->i_state
) &&
917 !test_bit(NILFS_I_BUSY
, &ii
->i_state
)) {
918 /* Because this routine may race with nilfs_dispose_list(),
919 we have to check NILFS_I_QUEUED here, too. */
920 if (list_empty(&ii
->i_dirty
) && igrab(inode
) == NULL
) {
921 /* This will happen when somebody is freeing
923 nilfs_warning(inode
->i_sb
, __func__
,
924 "cannot get inode (ino=%lu)\n",
926 spin_unlock(&nilfs
->ns_inode_lock
);
927 return -EINVAL
; /* NILFS_I_DIRTY may remain for
930 list_move_tail(&ii
->i_dirty
, &nilfs
->ns_dirty_files
);
931 set_bit(NILFS_I_QUEUED
, &ii
->i_state
);
933 spin_unlock(&nilfs
->ns_inode_lock
);
937 int __nilfs_mark_inode_dirty(struct inode
*inode
, int flags
)
939 struct buffer_head
*ibh
;
942 err
= nilfs_load_inode_block(inode
, &ibh
);
944 nilfs_warning(inode
->i_sb
, __func__
,
945 "failed to reget inode block.\n");
948 nilfs_update_inode(inode
, ibh
, flags
);
949 mark_buffer_dirty(ibh
);
950 nilfs_mdt_mark_dirty(NILFS_I(inode
)->i_root
->ifile
);
956 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
957 * @inode: inode of the file to be registered.
959 * nilfs_dirty_inode() loads a inode block containing the specified
960 * @inode and copies data from a nilfs_inode to a corresponding inode
961 * entry in the inode block. This operation is excluded from the segment
962 * construction. This function can be called both as a single operation
963 * and as a part of indivisible file operations.
965 void nilfs_dirty_inode(struct inode
*inode
, int flags
)
967 struct nilfs_transaction_info ti
;
968 struct nilfs_mdt_info
*mdi
= NILFS_MDT(inode
);
970 if (is_bad_inode(inode
)) {
971 nilfs_warning(inode
->i_sb
, __func__
,
972 "tried to mark bad_inode dirty. ignored.\n");
977 nilfs_mdt_mark_dirty(inode
);
980 nilfs_transaction_begin(inode
->i_sb
, &ti
, 0);
981 __nilfs_mark_inode_dirty(inode
, flags
);
982 nilfs_transaction_commit(inode
->i_sb
); /* never fails */
985 int nilfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
986 __u64 start
, __u64 len
)
988 struct the_nilfs
*nilfs
= inode
->i_sb
->s_fs_info
;
989 __u64 logical
= 0, phys
= 0, size
= 0;
992 sector_t blkoff
, end_blkoff
;
993 sector_t delalloc_blkoff
;
994 unsigned long delalloc_blklen
;
995 unsigned int blkbits
= inode
->i_blkbits
;
998 ret
= fiemap_check_flags(fieinfo
, FIEMAP_FLAG_SYNC
);
1004 isize
= i_size_read(inode
);
1006 blkoff
= start
>> blkbits
;
1007 end_blkoff
= (start
+ len
- 1) >> blkbits
;
1009 delalloc_blklen
= nilfs_find_uncommitted_extent(inode
, blkoff
,
1014 unsigned int maxblocks
;
1016 if (delalloc_blklen
&& blkoff
== delalloc_blkoff
) {
1018 /* End of the current extent */
1019 ret
= fiemap_fill_next_extent(
1020 fieinfo
, logical
, phys
, size
, flags
);
1024 if (blkoff
> end_blkoff
)
1027 flags
= FIEMAP_EXTENT_MERGED
| FIEMAP_EXTENT_DELALLOC
;
1028 logical
= blkoff
<< blkbits
;
1030 size
= delalloc_blklen
<< blkbits
;
1032 blkoff
= delalloc_blkoff
+ delalloc_blklen
;
1033 delalloc_blklen
= nilfs_find_uncommitted_extent(
1034 inode
, blkoff
, &delalloc_blkoff
);
1039 * Limit the number of blocks that we look up so as
1040 * not to get into the next delayed allocation extent.
1042 maxblocks
= INT_MAX
;
1043 if (delalloc_blklen
)
1044 maxblocks
= min_t(sector_t
, delalloc_blkoff
- blkoff
,
1048 down_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
1049 n
= nilfs_bmap_lookup_contig(
1050 NILFS_I(inode
)->i_bmap
, blkoff
, &blkphy
, maxblocks
);
1051 up_read(&NILFS_MDT(nilfs
->ns_dat
)->mi_sem
);
1056 if (unlikely(n
!= -ENOENT
))
1061 past_eof
= ((blkoff
<< blkbits
) >= isize
);
1064 /* End of the current extent */
1067 flags
|= FIEMAP_EXTENT_LAST
;
1069 ret
= fiemap_fill_next_extent(
1070 fieinfo
, logical
, phys
, size
, flags
);
1075 if (blkoff
> end_blkoff
|| past_eof
)
1079 if (phys
&& blkphy
<< blkbits
== phys
+ size
) {
1080 /* The current extent goes on */
1081 size
+= n
<< blkbits
;
1083 /* Terminate the current extent */
1084 ret
= fiemap_fill_next_extent(
1085 fieinfo
, logical
, phys
, size
,
1087 if (ret
|| blkoff
> end_blkoff
)
1090 /* Start another extent */
1091 flags
= FIEMAP_EXTENT_MERGED
;
1092 logical
= blkoff
<< blkbits
;
1093 phys
= blkphy
<< blkbits
;
1094 size
= n
<< blkbits
;
1097 /* Start a new extent */
1098 flags
= FIEMAP_EXTENT_MERGED
;
1099 logical
= blkoff
<< blkbits
;
1100 phys
= blkphy
<< blkbits
;
1101 size
= n
<< blkbits
;
1108 /* If ret is 1 then we just hit the end of the extent array */
1112 inode_unlock(inode
);