]> git.karo-electronics.de Git - linux-beck.git/blob - fs/nilfs2/inode.c
bbb47e8bde3e01ecf8b0e6d535861a50a36c9ee1
[linux-beck.git] / 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  * Written by Ryusuke Konishi.
17  *
18  */
19
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>
26 #include "nilfs.h"
27 #include "btnode.h"
28 #include "segment.h"
29 #include "page.h"
30 #include "mdt.h"
31 #include "cpfile.h"
32 #include "ifile.h"
33
34 /**
35  * struct nilfs_iget_args - arguments used during comparison between inodes
36  * @ino: inode number
37  * @cno: checkpoint number
38  * @root: pointer on NILFS root object (mounted checkpoint)
39  * @for_gc: inode for GC flag
40  */
41 struct nilfs_iget_args {
42         u64 ino;
43         __u64 cno;
44         struct nilfs_root *root;
45         int for_gc;
46 };
47
48 static int nilfs_iget_test(struct inode *inode, void *opaque);
49
50 void nilfs_inode_add_blocks(struct inode *inode, int n)
51 {
52         struct nilfs_root *root = NILFS_I(inode)->i_root;
53
54         inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
55         if (root)
56                 atomic64_add(n, &root->blocks_count);
57 }
58
59 void nilfs_inode_sub_blocks(struct inode *inode, int n)
60 {
61         struct nilfs_root *root = NILFS_I(inode)->i_root;
62
63         inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
64         if (root)
65                 atomic64_sub(n, &root->blocks_count);
66 }
67
68 /**
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
74  *      been allocated yet.
75  *
76  * This function does not issue actual read request of the specified data
77  * block. It is done by VFS.
78  */
79 int nilfs_get_block(struct inode *inode, sector_t blkoff,
80                     struct buffer_head *bh_result, int create)
81 {
82         struct nilfs_inode_info *ii = NILFS_I(inode);
83         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
84         __u64 blknum = 0;
85         int err = 0, ret;
86         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
87
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);
93                 if (ret > 0)
94                         bh_result->b_size = (ret << inode->i_blkbits);
95                 goto out;
96         }
97         /* data block was not found */
98         if (ret == -ENOENT && create) {
99                 struct nilfs_transaction_info ti;
100
101                 bh_result->b_blocknr = 0;
102                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
103                 if (unlikely(err))
104                         goto out;
105                 err = nilfs_bmap_insert(ii->i_bmap, blkoff,
106                                         (unsigned long)bh_result);
107                 if (unlikely(err != 0)) {
108                         if (err == -EEXIST) {
109                                 /*
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.
114                                  */
115                                 printk(KERN_WARNING
116                                        "nilfs_get_block: a race condition "
117                                        "while inserting a data block. "
118                                        "(inode number=%lu, file block "
119                                        "offset=%llu)\n",
120                                        inode->i_ino,
121                                        (unsigned long long)blkoff);
122                                 err = 0;
123                         }
124                         nilfs_transaction_abort(inode->i_sb);
125                         goto out;
126                 }
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
133                                                       to proper value */
134         } else if (ret == -ENOENT) {
135                 /* not found is not error (e.g. hole); must return without
136                    the mapped state flag. */
137                 ;
138         } else {
139                 err = ret;
140         }
141
142  out:
143         return err;
144 }
145
146 /**
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
151  */
152 static int nilfs_readpage(struct file *file, struct page *page)
153 {
154         return mpage_readpage(page, nilfs_get_block);
155 }
156
157 /**
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
164  */
165 static int nilfs_readpages(struct file *file, struct address_space *mapping,
166                            struct list_head *pages, unsigned nr_pages)
167 {
168         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
169 }
170
171 static int nilfs_writepages(struct address_space *mapping,
172                             struct writeback_control *wbc)
173 {
174         struct inode *inode = mapping->host;
175         int err = 0;
176
177         if (inode->i_sb->s_flags & MS_RDONLY) {
178                 nilfs_clear_dirty_pages(mapping, false);
179                 return -EROFS;
180         }
181
182         if (wbc->sync_mode == WB_SYNC_ALL)
183                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
184                                                     wbc->range_start,
185                                                     wbc->range_end);
186         return err;
187 }
188
189 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
190 {
191         struct inode *inode = page->mapping->host;
192         int err;
193
194         if (inode->i_sb->s_flags & MS_RDONLY) {
195                 /*
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.
200                  */
201                 nilfs_clear_dirty_page(page, false);
202                 unlock_page(page);
203                 return -EROFS;
204         }
205
206         redirty_page_for_writepage(wbc, page);
207         unlock_page(page);
208
209         if (wbc->sync_mode == WB_SYNC_ALL) {
210                 err = nilfs_construct_segment(inode->i_sb);
211                 if (unlikely(err))
212                         return err;
213         } else if (wbc->for_reclaim)
214                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
215
216         return 0;
217 }
218
219 static int nilfs_set_page_dirty(struct page *page)
220 {
221         struct inode *inode = page->mapping->host;
222         int ret = __set_page_dirty_nobuffers(page);
223
224         if (page_has_buffers(page)) {
225                 unsigned nr_dirty = 0;
226                 struct buffer_head *bh, *head;
227
228                 /*
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
233                  * by page lock.
234                  */
235                 bh = head = page_buffers(page);
236                 do {
237                         /* Do not mark hole blocks dirty */
238                         if (buffer_dirty(bh) || !buffer_mapped(bh))
239                                 continue;
240
241                         set_buffer_dirty(bh);
242                         nr_dirty++;
243                 } while (bh = bh->b_this_page, bh != head);
244
245                 if (nr_dirty)
246                         nilfs_set_file_dirty(inode, nr_dirty);
247         } else if (ret) {
248                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
249
250                 nilfs_set_file_dirty(inode, nr_dirty);
251         }
252         return ret;
253 }
254
255 void nilfs_write_failed(struct address_space *mapping, loff_t to)
256 {
257         struct inode *inode = mapping->host;
258
259         if (to > inode->i_size) {
260                 truncate_pagecache(inode, inode->i_size);
261                 nilfs_truncate(inode);
262         }
263 }
264
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)
268
269 {
270         struct inode *inode = mapping->host;
271         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
272
273         if (unlikely(err))
274                 return err;
275
276         err = block_write_begin(mapping, pos, len, flags, pagep,
277                                 nilfs_get_block);
278         if (unlikely(err)) {
279                 nilfs_write_failed(mapping, pos + len);
280                 nilfs_transaction_abort(inode->i_sb);
281         }
282         return err;
283 }
284
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)
288 {
289         struct inode *inode = mapping->host;
290         unsigned start = pos & (PAGE_SIZE - 1);
291         unsigned nr_dirty;
292         int err;
293
294         nr_dirty = nilfs_page_count_clean_buffers(page, start,
295                                                   start + copied);
296         copied = generic_write_end(file, mapping, pos, len, copied, page,
297                                    fsdata);
298         nilfs_set_file_dirty(inode, nr_dirty);
299         err = nilfs_transaction_commit(inode->i_sb);
300         return err ? : copied;
301 }
302
303 static ssize_t
304 nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
305 {
306         struct inode *inode = file_inode(iocb->ki_filp);
307
308         if (iov_iter_rw(iter) == WRITE)
309                 return 0;
310
311         /* Needs synchronization with the cleaner */
312         return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
313 }
314
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,
327 };
328
329 static int nilfs_insert_inode_locked(struct inode *inode,
330                                      struct nilfs_root *root,
331                                      unsigned long ino)
332 {
333         struct nilfs_iget_args args = {
334                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
335         };
336
337         return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
338 }
339
340 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
341 {
342         struct super_block *sb = dir->i_sb;
343         struct the_nilfs *nilfs = sb->s_fs_info;
344         struct inode *inode;
345         struct nilfs_inode_info *ii;
346         struct nilfs_root *root;
347         int err = -ENOMEM;
348         ino_t ino;
349
350         inode = new_inode(sb);
351         if (unlikely(!inode))
352                 goto failed;
353
354         mapping_set_gfp_mask(inode->i_mapping,
355                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
356
357         root = NILFS_I(dir)->i_root;
358         ii = NILFS_I(inode);
359         ii->i_state = 1 << NILFS_I_NEW;
360         ii->i_root = root;
361
362         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
363         if (unlikely(err))
364                 goto failed_ifile_create_inode;
365         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
366
367         atomic64_inc(&root->inodes_count);
368         inode_init_owner(inode, dir, mode);
369         inode->i_ino = ino;
370         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
371
372         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
373                 err = nilfs_bmap_read(ii->i_bmap, NULL);
374                 if (err < 0)
375                         goto failed_after_creation;
376
377                 set_bit(NILFS_I_BMAP, &ii->i_state);
378                 /* No lock is needed; iget() ensures it. */
379         }
380
381         ii->i_flags = nilfs_mask_flags(
382                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
383
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) {
392                 err = -EIO;
393                 goto failed_after_creation;
394         }
395
396         err = nilfs_init_acl(inode, dir);
397         if (unlikely(err))
398                 goto failed_after_creation; /* never occur. When supporting
399                                     nilfs_init_acl(), proper cancellation of
400                                     above jobs should be considered */
401
402         return inode;
403
404  failed_after_creation:
405         clear_nlink(inode);
406         unlock_new_inode(inode);
407         iput(inode);  /* raw_inode will be deleted through
408                          nilfs_evict_inode() */
409         goto failed;
410
411  failed_ifile_create_inode:
412         make_bad_inode(inode);
413         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
414                          called */
415  failed:
416         return ERR_PTR(err);
417 }
418
419 void nilfs_set_inode_flags(struct inode *inode)
420 {
421         unsigned int flags = NILFS_I(inode)->i_flags;
422         unsigned int new_fl = 0;
423
424         if (flags & FS_SYNC_FL)
425                 new_fl |= S_SYNC;
426         if (flags & FS_APPEND_FL)
427                 new_fl |= S_APPEND;
428         if (flags & FS_IMMUTABLE_FL)
429                 new_fl |= S_IMMUTABLE;
430         if (flags & FS_NOATIME_FL)
431                 new_fl |= S_NOATIME;
432         if (flags & FS_DIRSYNC_FL)
433                 new_fl |= S_DIRSYNC;
434         inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
435                         S_NOATIME | S_DIRSYNC);
436 }
437
438 int nilfs_read_inode_common(struct inode *inode,
439                             struct nilfs_inode *raw_inode)
440 {
441         struct nilfs_inode_info *ii = NILFS_I(inode);
442         int err;
443
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 */
457
458         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
459         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
460 #if 0
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);
464 #endif
465         ii->i_dir_start_lookup = 0;
466         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
467
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);
471                 if (err < 0)
472                         return err;
473                 set_bit(NILFS_I_BMAP, &ii->i_state);
474                 /* No lock is needed; iget() ensures it. */
475         }
476         return 0;
477 }
478
479 static int __nilfs_read_inode(struct super_block *sb,
480                               struct nilfs_root *root, unsigned long ino,
481                               struct inode *inode)
482 {
483         struct the_nilfs *nilfs = sb->s_fs_info;
484         struct buffer_head *bh;
485         struct nilfs_inode *raw_inode;
486         int err;
487
488         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
489         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
490         if (unlikely(err))
491                 goto bad_inode;
492
493         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
494
495         err = nilfs_read_inode_common(inode, raw_inode);
496         if (err)
497                 goto failed_unmap;
498
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;
511         } else {
512                 inode->i_op = &nilfs_special_inode_operations;
513                 init_special_inode(
514                         inode, inode->i_mode,
515                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
516         }
517         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
518         brelse(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));
523         return 0;
524
525  failed_unmap:
526         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
527         brelse(bh);
528
529  bad_inode:
530         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
531         return err;
532 }
533
534 static int nilfs_iget_test(struct inode *inode, void *opaque)
535 {
536         struct nilfs_iget_args *args = opaque;
537         struct nilfs_inode_info *ii;
538
539         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
540                 return 0;
541
542         ii = NILFS_I(inode);
543         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
544                 return !args->for_gc;
545
546         return args->for_gc && args->cno == ii->i_cno;
547 }
548
549 static int nilfs_iget_set(struct inode *inode, void *opaque)
550 {
551         struct nilfs_iget_args *args = opaque;
552
553         inode->i_ino = args->ino;
554         if (args->for_gc) {
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;
558         } else {
559                 if (args->root && args->ino == NILFS_ROOT_INO)
560                         nilfs_get_root(args->root);
561                 NILFS_I(inode)->i_root = args->root;
562         }
563         return 0;
564 }
565
566 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
567                             unsigned long ino)
568 {
569         struct nilfs_iget_args args = {
570                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
571         };
572
573         return ilookup5(sb, ino, nilfs_iget_test, &args);
574 }
575
576 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
577                                 unsigned long ino)
578 {
579         struct nilfs_iget_args args = {
580                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
581         };
582
583         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
584 }
585
586 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
587                          unsigned long ino)
588 {
589         struct inode *inode;
590         int err;
591
592         inode = nilfs_iget_locked(sb, root, ino);
593         if (unlikely(!inode))
594                 return ERR_PTR(-ENOMEM);
595         if (!(inode->i_state & I_NEW))
596                 return inode;
597
598         err = __nilfs_read_inode(sb, root, ino, inode);
599         if (unlikely(err)) {
600                 iget_failed(inode);
601                 return ERR_PTR(err);
602         }
603         unlock_new_inode(inode);
604         return inode;
605 }
606
607 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
608                                 __u64 cno)
609 {
610         struct nilfs_iget_args args = {
611                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
612         };
613         struct inode *inode;
614         int err;
615
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))
620                 return inode;
621
622         err = nilfs_init_gcinode(inode);
623         if (unlikely(err)) {
624                 iget_failed(inode);
625                 return ERR_PTR(err);
626         }
627         unlock_new_inode(inode);
628         return inode;
629 }
630
631 void nilfs_write_inode_common(struct inode *inode,
632                               struct nilfs_inode *raw_inode, int has_bmap)
633 {
634         struct nilfs_inode_info *ii = NILFS_I(inode);
635
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);
646
647         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
648         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
649
650         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
651                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
652
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));
658         }
659
660         if (has_bmap)
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 */
667 }
668
669 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
670 {
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;
675
676         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
677
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);
682
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
686                    before writing */
687         nilfs_ifile_unmap_inode(ifile, ino, ibh);
688 }
689
690 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
691
692 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
693                                 unsigned long from)
694 {
695         __u64 b;
696         int ret;
697
698         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
699                 return;
700 repeat:
701         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
702         if (ret == -ENOENT)
703                 return;
704         else if (ret < 0)
705                 goto failed;
706
707         if (b < from)
708                 return;
709
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))
715                 goto repeat;
716
717 failed:
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);
721 }
722
723 void nilfs_truncate(struct inode *inode)
724 {
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);
730
731         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
732                 return;
733         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
734                 return;
735
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 */
739
740         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
741
742         nilfs_truncate_bmap(ii, blkoff);
743
744         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
745         if (IS_SYNC(inode))
746                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
747
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. */
753 }
754
755 static void nilfs_clear_inode(struct inode *inode)
756 {
757         struct nilfs_inode_info *ii = NILFS_I(inode);
758
759         /*
760          * Free resources allocated in nilfs_read_inode(), here.
761          */
762         BUG_ON(!list_empty(&ii->i_dirty));
763         brelse(ii->i_bh);
764         ii->i_bh = NULL;
765
766         if (nilfs_is_metadata_file_inode(inode))
767                 nilfs_mdt_clear(inode);
768
769         if (test_bit(NILFS_I_BMAP, &ii->i_state))
770                 nilfs_bmap_clear(ii->i_bmap);
771
772         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
773
774         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
775                 nilfs_put_root(ii->i_root);
776 }
777
778 void nilfs_evict_inode(struct inode *inode)
779 {
780         struct nilfs_transaction_info ti;
781         struct super_block *sb = inode->i_sb;
782         struct nilfs_inode_info *ii = NILFS_I(inode);
783         int ret;
784
785         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
786                 truncate_inode_pages_final(&inode->i_data);
787                 clear_inode(inode);
788                 nilfs_clear_inode(inode);
789                 return;
790         }
791         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
792
793         truncate_inode_pages_final(&inode->i_data);
794
795         /* TODO: some of the following operations may fail.  */
796         nilfs_truncate_bmap(ii, 0);
797         nilfs_mark_inode_dirty(inode);
798         clear_inode(inode);
799
800         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
801         if (!ret)
802                 atomic64_dec(&ii->i_root->inodes_count);
803
804         nilfs_clear_inode(inode);
805
806         if (IS_SYNC(inode))
807                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
808         nilfs_transaction_commit(sb);
809         /* May construct a logical segment and may fail in sync mode.
810            But delete_inode has no return value. */
811 }
812
813 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
814 {
815         struct nilfs_transaction_info ti;
816         struct inode *inode = d_inode(dentry);
817         struct super_block *sb = inode->i_sb;
818         int err;
819
820         err = inode_change_ok(inode, iattr);
821         if (err)
822                 return err;
823
824         err = nilfs_transaction_begin(sb, &ti, 0);
825         if (unlikely(err))
826                 return err;
827
828         if ((iattr->ia_valid & ATTR_SIZE) &&
829             iattr->ia_size != i_size_read(inode)) {
830                 inode_dio_wait(inode);
831                 truncate_setsize(inode, iattr->ia_size);
832                 nilfs_truncate(inode);
833         }
834
835         setattr_copy(inode, iattr);
836         mark_inode_dirty(inode);
837
838         if (iattr->ia_valid & ATTR_MODE) {
839                 err = nilfs_acl_chmod(inode);
840                 if (unlikely(err))
841                         goto out_err;
842         }
843
844         return nilfs_transaction_commit(sb);
845
846 out_err:
847         nilfs_transaction_abort(sb);
848         return err;
849 }
850
851 int nilfs_permission(struct inode *inode, int mask)
852 {
853         struct nilfs_root *root = NILFS_I(inode)->i_root;
854
855         if ((mask & MAY_WRITE) && root &&
856             root->cno != NILFS_CPTREE_CURRENT_CNO)
857                 return -EROFS; /* snapshot is not writable */
858
859         return generic_permission(inode, mask);
860 }
861
862 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
863 {
864         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
865         struct nilfs_inode_info *ii = NILFS_I(inode);
866         int err;
867
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,
872                                                   inode->i_ino, pbh);
873                 if (unlikely(err))
874                         return err;
875                 spin_lock(&nilfs->ns_inode_lock);
876                 if (ii->i_bh == NULL)
877                         ii->i_bh = *pbh;
878                 else {
879                         brelse(*pbh);
880                         *pbh = ii->i_bh;
881                 }
882         } else
883                 *pbh = ii->i_bh;
884
885         get_bh(*pbh);
886         spin_unlock(&nilfs->ns_inode_lock);
887         return 0;
888 }
889
890 int nilfs_inode_dirty(struct inode *inode)
891 {
892         struct nilfs_inode_info *ii = NILFS_I(inode);
893         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
894         int ret = 0;
895
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);
901         }
902         return ret;
903 }
904
905 int nilfs_set_file_dirty(struct inode *inode, unsigned nr_dirty)
906 {
907         struct nilfs_inode_info *ii = NILFS_I(inode);
908         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
909
910         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
911
912         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
913                 return 0;
914
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
922                            this inode. */
923                         nilfs_warning(inode->i_sb, __func__,
924                                       "cannot get inode (ino=%lu)",
925                                       inode->i_ino);
926                         spin_unlock(&nilfs->ns_inode_lock);
927                         return -EINVAL; /* NILFS_I_DIRTY may remain for
928                                            freeing inode */
929                 }
930                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
931                 set_bit(NILFS_I_QUEUED, &ii->i_state);
932         }
933         spin_unlock(&nilfs->ns_inode_lock);
934         return 0;
935 }
936
937 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
938 {
939         struct buffer_head *ibh;
940         int err;
941
942         err = nilfs_load_inode_block(inode, &ibh);
943         if (unlikely(err)) {
944                 nilfs_warning(inode->i_sb, __func__,
945                               "failed to reget inode block.");
946                 return err;
947         }
948         nilfs_update_inode(inode, ibh, flags);
949         mark_buffer_dirty(ibh);
950         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
951         brelse(ibh);
952         return 0;
953 }
954
955 /**
956  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
957  * @inode: inode of the file to be registered.
958  *
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.
964  */
965 void nilfs_dirty_inode(struct inode *inode, int flags)
966 {
967         struct nilfs_transaction_info ti;
968         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
969
970         if (is_bad_inode(inode)) {
971                 nilfs_warning(inode->i_sb, __func__,
972                               "tried to mark bad_inode dirty. ignored.");
973                 dump_stack();
974                 return;
975         }
976         if (mdi) {
977                 nilfs_mdt_mark_dirty(inode);
978                 return;
979         }
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 */
983 }
984
985 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
986                  __u64 start, __u64 len)
987 {
988         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
989         __u64 logical = 0, phys = 0, size = 0;
990         __u32 flags = 0;
991         loff_t isize;
992         sector_t blkoff, end_blkoff;
993         sector_t delalloc_blkoff;
994         unsigned long delalloc_blklen;
995         unsigned int blkbits = inode->i_blkbits;
996         int ret, n;
997
998         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
999         if (ret)
1000                 return ret;
1001
1002         inode_lock(inode);
1003
1004         isize = i_size_read(inode);
1005
1006         blkoff = start >> blkbits;
1007         end_blkoff = (start + len - 1) >> blkbits;
1008
1009         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1010                                                         &delalloc_blkoff);
1011
1012         do {
1013                 __u64 blkphy;
1014                 unsigned int maxblocks;
1015
1016                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1017                         if (size) {
1018                                 /* End of the current extent */
1019                                 ret = fiemap_fill_next_extent(
1020                                         fieinfo, logical, phys, size, flags);
1021                                 if (ret)
1022                                         break;
1023                         }
1024                         if (blkoff > end_blkoff)
1025                                 break;
1026
1027                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1028                         logical = blkoff << blkbits;
1029                         phys = 0;
1030                         size = delalloc_blklen << blkbits;
1031
1032                         blkoff = delalloc_blkoff + delalloc_blklen;
1033                         delalloc_blklen = nilfs_find_uncommitted_extent(
1034                                 inode, blkoff, &delalloc_blkoff);
1035                         continue;
1036                 }
1037
1038                 /*
1039                  * Limit the number of blocks that we look up so as
1040                  * not to get into the next delayed allocation extent.
1041                  */
1042                 maxblocks = INT_MAX;
1043                 if (delalloc_blklen)
1044                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1045                                           maxblocks);
1046                 blkphy = 0;
1047
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);
1052
1053                 if (n < 0) {
1054                         int past_eof;
1055
1056                         if (unlikely(n != -ENOENT))
1057                                 break; /* error */
1058
1059                         /* HOLE */
1060                         blkoff++;
1061                         past_eof = ((blkoff << blkbits) >= isize);
1062
1063                         if (size) {
1064                                 /* End of the current extent */
1065
1066                                 if (past_eof)
1067                                         flags |= FIEMAP_EXTENT_LAST;
1068
1069                                 ret = fiemap_fill_next_extent(
1070                                         fieinfo, logical, phys, size, flags);
1071                                 if (ret)
1072                                         break;
1073                                 size = 0;
1074                         }
1075                         if (blkoff > end_blkoff || past_eof)
1076                                 break;
1077                 } else {
1078                         if (size) {
1079                                 if (phys && blkphy << blkbits == phys + size) {
1080                                         /* The current extent goes on */
1081                                         size += n << blkbits;
1082                                 } else {
1083                                         /* Terminate the current extent */
1084                                         ret = fiemap_fill_next_extent(
1085                                                 fieinfo, logical, phys, size,
1086                                                 flags);
1087                                         if (ret || blkoff > end_blkoff)
1088                                                 break;
1089
1090                                         /* Start another extent */
1091                                         flags = FIEMAP_EXTENT_MERGED;
1092                                         logical = blkoff << blkbits;
1093                                         phys = blkphy << blkbits;
1094                                         size = n << blkbits;
1095                                 }
1096                         } else {
1097                                 /* Start a new extent */
1098                                 flags = FIEMAP_EXTENT_MERGED;
1099                                 logical = blkoff << blkbits;
1100                                 phys = blkphy << blkbits;
1101                                 size = n << blkbits;
1102                         }
1103                         blkoff += n;
1104                 }
1105                 cond_resched();
1106         } while (true);
1107
1108         /* If ret is 1 then we just hit the end of the extent array */
1109         if (ret == 1)
1110                 ret = 0;
1111
1112         inode_unlock(inode);
1113         return ret;
1114 }