Merge branches 'acerhdf', 'acpi-pci-bind', 'bjorn-pci-root', 'bugzilla-12904', 'bugzi...
[linux-2.6] / fs / nilfs2 / inode.c
1 /*
2  * inode.c - NILFS inode operations.
3  *
4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5  *
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.
10  *
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.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
19  *
20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
21  *
22  */
23
24 #include <linux/buffer_head.h>
25 #include <linux/mpage.h>
26 #include <linux/writeback.h>
27 #include <linux/uio.h>
28 #include "nilfs.h"
29 #include "segment.h"
30 #include "page.h"
31 #include "mdt.h"
32 #include "cpfile.h"
33 #include "ifile.h"
34
35
36 /**
37  * nilfs_get_block() - get a file block on the filesystem (callback function)
38  * @inode - inode struct of the target file
39  * @blkoff - file block number
40  * @bh_result - buffer head to be mapped on
41  * @create - indicate whether allocating the block or not when it has not
42  *      been allocated yet.
43  *
44  * This function does not issue actual read request of the specified data
45  * block. It is done by VFS.
46  */
47 int nilfs_get_block(struct inode *inode, sector_t blkoff,
48                     struct buffer_head *bh_result, int create)
49 {
50         struct nilfs_inode_info *ii = NILFS_I(inode);
51         __u64 blknum = 0;
52         int err = 0, ret;
53         struct inode *dat = nilfs_dat_inode(NILFS_I_NILFS(inode));
54         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
55
56         down_read(&NILFS_MDT(dat)->mi_sem);
57         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
58         up_read(&NILFS_MDT(dat)->mi_sem);
59         if (ret >= 0) { /* found */
60                 map_bh(bh_result, inode->i_sb, blknum);
61                 if (ret > 0)
62                         bh_result->b_size = (ret << inode->i_blkbits);
63                 goto out;
64         }
65         /* data block was not found */
66         if (ret == -ENOENT && create) {
67                 struct nilfs_transaction_info ti;
68
69                 bh_result->b_blocknr = 0;
70                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
71                 if (unlikely(err))
72                         goto out;
73                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
74                                         (unsigned long)bh_result);
75                 if (unlikely(err != 0)) {
76                         if (err == -EEXIST) {
77                                 /*
78                                  * The get_block() function could be called
79                                  * from multiple callers for an inode.
80                                  * However, the page having this block must
81                                  * be locked in this case.
82                                  */
83                                 printk(KERN_WARNING
84                                        "nilfs_get_block: a race condition "
85                                        "while inserting a data block. "
86                                        "(inode number=%lu, file block "
87                                        "offset=%llu)\n",
88                                        inode->i_ino,
89                                        (unsigned long long)blkoff);
90                                 err = 0;
91                         } else if (err == -EINVAL) {
92                                 nilfs_error(inode->i_sb, __func__,
93                                             "broken bmap (inode=%lu)\n",
94                                             inode->i_ino);
95                                 err = -EIO;
96                         }
97                         nilfs_transaction_abort(inode->i_sb);
98                         goto out;
99                 }
100                 nilfs_transaction_commit(inode->i_sb); /* never fails */
101                 /* Error handling should be detailed */
102                 set_buffer_new(bh_result);
103                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
104                                                       to proper value */
105         } else if (ret == -ENOENT) {
106                 /* not found is not error (e.g. hole); must return without
107                    the mapped state flag. */
108                 ;
109         } else {
110                 err = ret;
111         }
112
113  out:
114         return err;
115 }
116
117 /**
118  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
119  * address_space_operations.
120  * @file - file struct of the file to be read
121  * @page - the page to be read
122  */
123 static int nilfs_readpage(struct file *file, struct page *page)
124 {
125         return mpage_readpage(page, nilfs_get_block);
126 }
127
128 /**
129  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
130  * address_space_operations.
131  * @file - file struct of the file to be read
132  * @mapping - address_space struct used for reading multiple pages
133  * @pages - the pages to be read
134  * @nr_pages - number of pages to be read
135  */
136 static int nilfs_readpages(struct file *file, struct address_space *mapping,
137                            struct list_head *pages, unsigned nr_pages)
138 {
139         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
140 }
141
142 static int nilfs_writepages(struct address_space *mapping,
143                             struct writeback_control *wbc)
144 {
145         struct inode *inode = mapping->host;
146         int err = 0;
147
148         if (wbc->sync_mode == WB_SYNC_ALL)
149                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
150                                                     wbc->range_start,
151                                                     wbc->range_end);
152         return err;
153 }
154
155 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
156 {
157         struct inode *inode = page->mapping->host;
158         int err;
159
160         redirty_page_for_writepage(wbc, page);
161         unlock_page(page);
162
163         if (wbc->sync_mode == WB_SYNC_ALL) {
164                 err = nilfs_construct_segment(inode->i_sb);
165                 if (unlikely(err))
166                         return err;
167         } else if (wbc->for_reclaim)
168                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
169
170         return 0;
171 }
172
173 static int nilfs_set_page_dirty(struct page *page)
174 {
175         int ret = __set_page_dirty_buffers(page);
176
177         if (ret) {
178                 struct inode *inode = page->mapping->host;
179                 struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
180                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
181
182                 nilfs_set_file_dirty(sbi, inode, nr_dirty);
183         }
184         return ret;
185 }
186
187 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
188                              loff_t pos, unsigned len, unsigned flags,
189                              struct page **pagep, void **fsdata)
190
191 {
192         struct inode *inode = mapping->host;
193         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
194
195         if (unlikely(err))
196                 return err;
197
198         *pagep = NULL;
199         err = block_write_begin(file, mapping, pos, len, flags, pagep,
200                                 fsdata, nilfs_get_block);
201         if (unlikely(err))
202                 nilfs_transaction_abort(inode->i_sb);
203         return err;
204 }
205
206 static int nilfs_write_end(struct file *file, struct address_space *mapping,
207                            loff_t pos, unsigned len, unsigned copied,
208                            struct page *page, void *fsdata)
209 {
210         struct inode *inode = mapping->host;
211         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
212         unsigned nr_dirty;
213         int err;
214
215         nr_dirty = nilfs_page_count_clean_buffers(page, start,
216                                                   start + copied);
217         copied = generic_write_end(file, mapping, pos, len, copied, page,
218                                    fsdata);
219         nilfs_set_file_dirty(NILFS_SB(inode->i_sb), inode, nr_dirty);
220         err = nilfs_transaction_commit(inode->i_sb);
221         return err ? : copied;
222 }
223
224 static ssize_t
225 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
226                 loff_t offset, unsigned long nr_segs)
227 {
228         struct file *file = iocb->ki_filp;
229         struct inode *inode = file->f_mapping->host;
230         ssize_t size;
231
232         if (rw == WRITE)
233                 return 0;
234
235         /* Needs synchronization with the cleaner */
236         size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
237                                   offset, nr_segs, nilfs_get_block, NULL);
238         return size;
239 }
240
241 struct address_space_operations nilfs_aops = {
242         .writepage              = nilfs_writepage,
243         .readpage               = nilfs_readpage,
244         .sync_page              = block_sync_page,
245         .writepages             = nilfs_writepages,
246         .set_page_dirty         = nilfs_set_page_dirty,
247         .readpages              = nilfs_readpages,
248         .write_begin            = nilfs_write_begin,
249         .write_end              = nilfs_write_end,
250         /* .releasepage         = nilfs_releasepage, */
251         .invalidatepage         = block_invalidatepage,
252         .direct_IO              = nilfs_direct_IO,
253         .is_partially_uptodate  = block_is_partially_uptodate,
254 };
255
256 struct inode *nilfs_new_inode(struct inode *dir, int mode)
257 {
258         struct super_block *sb = dir->i_sb;
259         struct nilfs_sb_info *sbi = NILFS_SB(sb);
260         struct inode *inode;
261         struct nilfs_inode_info *ii;
262         int err = -ENOMEM;
263         ino_t ino;
264
265         inode = new_inode(sb);
266         if (unlikely(!inode))
267                 goto failed;
268
269         mapping_set_gfp_mask(inode->i_mapping,
270                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
271
272         ii = NILFS_I(inode);
273         ii->i_state = 1 << NILFS_I_NEW;
274
275         err = nilfs_ifile_create_inode(sbi->s_ifile, &ino, &ii->i_bh);
276         if (unlikely(err))
277                 goto failed_ifile_create_inode;
278         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
279
280         atomic_inc(&sbi->s_inodes_count);
281
282         inode->i_uid = current_fsuid();
283         if (dir->i_mode & S_ISGID) {
284                 inode->i_gid = dir->i_gid;
285                 if (S_ISDIR(mode))
286                         mode |= S_ISGID;
287         } else
288                 inode->i_gid = current_fsgid();
289
290         inode->i_mode = mode;
291         inode->i_ino = ino;
292         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
293
294         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
295                 err = nilfs_bmap_read(ii->i_bmap, NULL);
296                 if (err < 0)
297                         goto failed_bmap;
298
299                 set_bit(NILFS_I_BMAP, &ii->i_state);
300                 /* No lock is needed; iget() ensures it. */
301         }
302
303         ii->i_flags = NILFS_I(dir)->i_flags;
304         if (S_ISLNK(mode))
305                 ii->i_flags &= ~(NILFS_IMMUTABLE_FL | NILFS_APPEND_FL);
306         if (!S_ISDIR(mode))
307                 ii->i_flags &= ~NILFS_DIRSYNC_FL;
308
309         /* ii->i_file_acl = 0; */
310         /* ii->i_dir_acl = 0; */
311         ii->i_dir_start_lookup = 0;
312 #ifdef CONFIG_NILFS_FS_POSIX_ACL
313         ii->i_acl = NULL;
314         ii->i_default_acl = NULL;
315 #endif
316         ii->i_cno = 0;
317         nilfs_set_inode_flags(inode);
318         spin_lock(&sbi->s_next_gen_lock);
319         inode->i_generation = sbi->s_next_generation++;
320         spin_unlock(&sbi->s_next_gen_lock);
321         insert_inode_hash(inode);
322
323         err = nilfs_init_acl(inode, dir);
324         if (unlikely(err))
325                 goto failed_acl; /* never occur. When supporting
326                                     nilfs_init_acl(), proper cancellation of
327                                     above jobs should be considered */
328
329         mark_inode_dirty(inode);
330         return inode;
331
332  failed_acl:
333  failed_bmap:
334         inode->i_nlink = 0;
335         iput(inode);  /* raw_inode will be deleted through
336                          generic_delete_inode() */
337         goto failed;
338
339  failed_ifile_create_inode:
340         make_bad_inode(inode);
341         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
342                          called */
343  failed:
344         return ERR_PTR(err);
345 }
346
347 void nilfs_free_inode(struct inode *inode)
348 {
349         struct super_block *sb = inode->i_sb;
350         struct nilfs_sb_info *sbi = NILFS_SB(sb);
351
352         clear_inode(inode);
353         /* XXX: check error code? Is there any thing I can do? */
354         (void) nilfs_ifile_delete_inode(sbi->s_ifile, inode->i_ino);
355         atomic_dec(&sbi->s_inodes_count);
356 }
357
358 void nilfs_set_inode_flags(struct inode *inode)
359 {
360         unsigned int flags = NILFS_I(inode)->i_flags;
361
362         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
363                             S_DIRSYNC);
364         if (flags & NILFS_SYNC_FL)
365                 inode->i_flags |= S_SYNC;
366         if (flags & NILFS_APPEND_FL)
367                 inode->i_flags |= S_APPEND;
368         if (flags & NILFS_IMMUTABLE_FL)
369                 inode->i_flags |= S_IMMUTABLE;
370 #ifndef NILFS_ATIME_DISABLE
371         if (flags & NILFS_NOATIME_FL)
372 #endif
373                 inode->i_flags |= S_NOATIME;
374         if (flags & NILFS_DIRSYNC_FL)
375                 inode->i_flags |= S_DIRSYNC;
376         mapping_set_gfp_mask(inode->i_mapping,
377                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
378 }
379
380 int nilfs_read_inode_common(struct inode *inode,
381                             struct nilfs_inode *raw_inode)
382 {
383         struct nilfs_inode_info *ii = NILFS_I(inode);
384         int err;
385
386         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
387         inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
388         inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
389         inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
390         inode->i_size = le64_to_cpu(raw_inode->i_size);
391         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
392         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
393         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
394         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
395         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
396         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
397         if (inode->i_nlink == 0 && inode->i_mode == 0)
398                 return -EINVAL; /* this inode is deleted */
399
400         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
401         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
402 #if 0
403         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
404         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
405                 0 : le32_to_cpu(raw_inode->i_dir_acl);
406 #endif
407         ii->i_cno = 0;
408         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
409
410         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
411             S_ISLNK(inode->i_mode)) {
412                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
413                 if (err < 0)
414                         return err;
415                 set_bit(NILFS_I_BMAP, &ii->i_state);
416                 /* No lock is needed; iget() ensures it. */
417         }
418         return 0;
419 }
420
421 static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
422                               struct inode *inode)
423 {
424         struct nilfs_sb_info *sbi = NILFS_SB(sb);
425         struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
426         struct buffer_head *bh;
427         struct nilfs_inode *raw_inode;
428         int err;
429
430         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
431         err = nilfs_ifile_get_inode_block(sbi->s_ifile, ino, &bh);
432         if (unlikely(err))
433                 goto bad_inode;
434
435         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, bh);
436
437 #ifdef CONFIG_NILFS_FS_POSIX_ACL
438         ii->i_acl = NILFS_ACL_NOT_CACHED;
439         ii->i_default_acl = NILFS_ACL_NOT_CACHED;
440 #endif
441         if (nilfs_read_inode_common(inode, raw_inode))
442                 goto failed_unmap;
443
444         if (S_ISREG(inode->i_mode)) {
445                 inode->i_op = &nilfs_file_inode_operations;
446                 inode->i_fop = &nilfs_file_operations;
447                 inode->i_mapping->a_ops = &nilfs_aops;
448         } else if (S_ISDIR(inode->i_mode)) {
449                 inode->i_op = &nilfs_dir_inode_operations;
450                 inode->i_fop = &nilfs_dir_operations;
451                 inode->i_mapping->a_ops = &nilfs_aops;
452         } else if (S_ISLNK(inode->i_mode)) {
453                 inode->i_op = &nilfs_symlink_inode_operations;
454                 inode->i_mapping->a_ops = &nilfs_aops;
455         } else {
456                 inode->i_op = &nilfs_special_inode_operations;
457                 init_special_inode(
458                         inode, inode->i_mode,
459                         new_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
460         }
461         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
462         brelse(bh);
463         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
464         nilfs_set_inode_flags(inode);
465         return 0;
466
467  failed_unmap:
468         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
469         brelse(bh);
470
471  bad_inode:
472         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
473         return err;
474 }
475
476 struct inode *nilfs_iget(struct super_block *sb, unsigned long ino)
477 {
478         struct inode *inode;
479         int err;
480
481         inode = iget_locked(sb, ino);
482         if (unlikely(!inode))
483                 return ERR_PTR(-ENOMEM);
484         if (!(inode->i_state & I_NEW))
485                 return inode;
486
487         err = __nilfs_read_inode(sb, ino, inode);
488         if (unlikely(err)) {
489                 iget_failed(inode);
490                 return ERR_PTR(err);
491         }
492         unlock_new_inode(inode);
493         return inode;
494 }
495
496 void nilfs_write_inode_common(struct inode *inode,
497                               struct nilfs_inode *raw_inode, int has_bmap)
498 {
499         struct nilfs_inode_info *ii = NILFS_I(inode);
500
501         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
502         raw_inode->i_uid = cpu_to_le32(inode->i_uid);
503         raw_inode->i_gid = cpu_to_le32(inode->i_gid);
504         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
505         raw_inode->i_size = cpu_to_le64(inode->i_size);
506         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
507         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
508         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
509         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
510         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
511
512         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
513         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
514
515         if (has_bmap)
516                 nilfs_bmap_write(ii->i_bmap, raw_inode);
517         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
518                 raw_inode->i_device_code =
519                         cpu_to_le64(new_encode_dev(inode->i_rdev));
520         /* When extending inode, nilfs->ns_inode_size should be checked
521            for substitutions of appended fields */
522 }
523
524 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
525 {
526         ino_t ino = inode->i_ino;
527         struct nilfs_inode_info *ii = NILFS_I(inode);
528         struct super_block *sb = inode->i_sb;
529         struct nilfs_sb_info *sbi = NILFS_SB(sb);
530         struct nilfs_inode *raw_inode;
531
532         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, ibh);
533
534         /* The buffer is guarded with lock_buffer() by the caller */
535         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
536                 memset(raw_inode, 0, NILFS_MDT(sbi->s_ifile)->mi_entry_size);
537         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
538
539         nilfs_write_inode_common(inode, raw_inode, 0);
540                 /* XXX: call with has_bmap = 0 is a workaround to avoid
541                    deadlock of bmap. This delays update of i_bmap to just
542                    before writing */
543         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, ibh);
544 }
545
546 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
547
548 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
549                                 unsigned long from)
550 {
551         unsigned long b;
552         int ret;
553
554         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
555                 return;
556  repeat:
557         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
558         if (ret == -ENOENT)
559                 return;
560         else if (ret < 0)
561                 goto failed;
562
563         if (b < from)
564                 return;
565
566         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
567         ret = nilfs_bmap_truncate(ii->i_bmap, b);
568         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
569         if (!ret || (ret == -ENOMEM &&
570                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
571                 goto repeat;
572
573  failed:
574         if (ret == -EINVAL)
575                 nilfs_error(ii->vfs_inode.i_sb, __func__,
576                             "bmap is broken (ino=%lu)", ii->vfs_inode.i_ino);
577         else
578                 nilfs_warning(ii->vfs_inode.i_sb, __func__,
579                               "failed to truncate bmap (ino=%lu, err=%d)",
580                               ii->vfs_inode.i_ino, ret);
581 }
582
583 void nilfs_truncate(struct inode *inode)
584 {
585         unsigned long blkoff;
586         unsigned int blocksize;
587         struct nilfs_transaction_info ti;
588         struct super_block *sb = inode->i_sb;
589         struct nilfs_inode_info *ii = NILFS_I(inode);
590
591         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
592                 return;
593         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
594                 return;
595
596         blocksize = sb->s_blocksize;
597         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
598         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
599
600         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
601
602         nilfs_truncate_bmap(ii, blkoff);
603
604         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
605         if (IS_SYNC(inode))
606                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
607
608         nilfs_set_file_dirty(NILFS_SB(sb), inode, 0);
609         nilfs_transaction_commit(sb);
610         /* May construct a logical segment and may fail in sync mode.
611            But truncate has no return value. */
612 }
613
614 void nilfs_delete_inode(struct inode *inode)
615 {
616         struct nilfs_transaction_info ti;
617         struct super_block *sb = inode->i_sb;
618         struct nilfs_inode_info *ii = NILFS_I(inode);
619
620         if (unlikely(is_bad_inode(inode))) {
621                 if (inode->i_data.nrpages)
622                         truncate_inode_pages(&inode->i_data, 0);
623                 clear_inode(inode);
624                 return;
625         }
626         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
627
628         if (inode->i_data.nrpages)
629                 truncate_inode_pages(&inode->i_data, 0);
630
631         nilfs_truncate_bmap(ii, 0);
632         nilfs_free_inode(inode);
633         /* nilfs_free_inode() marks inode buffer dirty */
634         if (IS_SYNC(inode))
635                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
636         nilfs_transaction_commit(sb);
637         /* May construct a logical segment and may fail in sync mode.
638            But delete_inode has no return value. */
639 }
640
641 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
642 {
643         struct nilfs_transaction_info ti;
644         struct inode *inode = dentry->d_inode;
645         struct super_block *sb = inode->i_sb;
646         int err;
647
648         err = inode_change_ok(inode, iattr);
649         if (err)
650                 return err;
651
652         err = nilfs_transaction_begin(sb, &ti, 0);
653         if (unlikely(err))
654                 return err;
655         err = inode_setattr(inode, iattr);
656         if (!err && (iattr->ia_valid & ATTR_MODE))
657                 err = nilfs_acl_chmod(inode);
658         if (likely(!err))
659                 err = nilfs_transaction_commit(sb);
660         else
661                 nilfs_transaction_abort(sb);
662
663         return err;
664 }
665
666 int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
667                            struct buffer_head **pbh)
668 {
669         struct nilfs_inode_info *ii = NILFS_I(inode);
670         int err;
671
672         spin_lock(&sbi->s_inode_lock);
673         /* Caller of this function MUST lock s_inode_lock */
674         if (ii->i_bh == NULL) {
675                 spin_unlock(&sbi->s_inode_lock);
676                 err = nilfs_ifile_get_inode_block(sbi->s_ifile, inode->i_ino,
677                                                   pbh);
678                 if (unlikely(err))
679                         return err;
680                 spin_lock(&sbi->s_inode_lock);
681                 if (ii->i_bh == NULL)
682                         ii->i_bh = *pbh;
683                 else {
684                         brelse(*pbh);
685                         *pbh = ii->i_bh;
686                 }
687         } else
688                 *pbh = ii->i_bh;
689
690         get_bh(*pbh);
691         spin_unlock(&sbi->s_inode_lock);
692         return 0;
693 }
694
695 int nilfs_inode_dirty(struct inode *inode)
696 {
697         struct nilfs_inode_info *ii = NILFS_I(inode);
698         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
699         int ret = 0;
700
701         if (!list_empty(&ii->i_dirty)) {
702                 spin_lock(&sbi->s_inode_lock);
703                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
704                         test_bit(NILFS_I_BUSY, &ii->i_state);
705                 spin_unlock(&sbi->s_inode_lock);
706         }
707         return ret;
708 }
709
710 int nilfs_set_file_dirty(struct nilfs_sb_info *sbi, struct inode *inode,
711                          unsigned nr_dirty)
712 {
713         struct nilfs_inode_info *ii = NILFS_I(inode);
714
715         atomic_add(nr_dirty, &sbi->s_nilfs->ns_ndirtyblks);
716
717         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
718                 return 0;
719
720         spin_lock(&sbi->s_inode_lock);
721         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
722             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
723                 /* Because this routine may race with nilfs_dispose_list(),
724                    we have to check NILFS_I_QUEUED here, too. */
725                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
726                         /* This will happen when somebody is freeing
727                            this inode. */
728                         nilfs_warning(sbi->s_super, __func__,
729                                       "cannot get inode (ino=%lu)\n",
730                                       inode->i_ino);
731                         spin_unlock(&sbi->s_inode_lock);
732                         return -EINVAL; /* NILFS_I_DIRTY may remain for
733                                            freeing inode */
734                 }
735                 list_del(&ii->i_dirty);
736                 list_add_tail(&ii->i_dirty, &sbi->s_dirty_files);
737                 set_bit(NILFS_I_QUEUED, &ii->i_state);
738         }
739         spin_unlock(&sbi->s_inode_lock);
740         return 0;
741 }
742
743 int nilfs_mark_inode_dirty(struct inode *inode)
744 {
745         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
746         struct buffer_head *ibh;
747         int err;
748
749         err = nilfs_load_inode_block(sbi, inode, &ibh);
750         if (unlikely(err)) {
751                 nilfs_warning(inode->i_sb, __func__,
752                               "failed to reget inode block.\n");
753                 return err;
754         }
755         lock_buffer(ibh);
756         nilfs_update_inode(inode, ibh);
757         unlock_buffer(ibh);
758         nilfs_mdt_mark_buffer_dirty(ibh);
759         nilfs_mdt_mark_dirty(sbi->s_ifile);
760         brelse(ibh);
761         return 0;
762 }
763
764 /**
765  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
766  * @inode: inode of the file to be registered.
767  *
768  * nilfs_dirty_inode() loads a inode block containing the specified
769  * @inode and copies data from a nilfs_inode to a corresponding inode
770  * entry in the inode block. This operation is excluded from the segment
771  * construction. This function can be called both as a single operation
772  * and as a part of indivisible file operations.
773  */
774 void nilfs_dirty_inode(struct inode *inode)
775 {
776         struct nilfs_transaction_info ti;
777
778         if (is_bad_inode(inode)) {
779                 nilfs_warning(inode->i_sb, __func__,
780                               "tried to mark bad_inode dirty. ignored.\n");
781                 dump_stack();
782                 return;
783         }
784         nilfs_transaction_begin(inode->i_sb, &ti, 0);
785         nilfs_mark_inode_dirty(inode);
786         nilfs_transaction_commit(inode->i_sb); /* never fails */
787 }