Commit | Line | Data |
---|---|---|
783f6184 RK |
1 | /* |
2 | * super.c - NILFS module and super block management. | |
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 | * linux/fs/ext2/super.c | |
24 | * | |
25 | * Copyright (C) 1992, 1993, 1994, 1995 | |
26 | * Remy Card (card@masi.ibp.fr) | |
27 | * Laboratoire MASI - Institut Blaise Pascal | |
28 | * Universite Pierre et Marie Curie (Paris VI) | |
29 | * | |
30 | * from | |
31 | * | |
32 | * linux/fs/minix/inode.c | |
33 | * | |
34 | * Copyright (C) 1991, 1992 Linus Torvalds | |
35 | * | |
36 | * Big-endian to little-endian byte-swapping/bitmaps by | |
37 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
38 | */ | |
39 | ||
40 | #include <linux/module.h> | |
41 | #include <linux/string.h> | |
42 | #include <linux/slab.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/blkdev.h> | |
45 | #include <linux/parser.h> | |
46 | #include <linux/random.h> | |
47 | #include <linux/crc32.h> | |
48 | #include <linux/smp_lock.h> | |
49 | #include <linux/vfs.h> | |
50 | #include <linux/writeback.h> | |
51 | #include <linux/kobject.h> | |
52 | #include <linux/exportfs.h> | |
53 | #include "nilfs.h" | |
54 | #include "mdt.h" | |
55 | #include "alloc.h" | |
56 | #include "page.h" | |
57 | #include "cpfile.h" | |
58 | #include "ifile.h" | |
59 | #include "dat.h" | |
60 | #include "segment.h" | |
61 | #include "segbuf.h" | |
62 | ||
63 | MODULE_AUTHOR("NTT Corp."); | |
64 | MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem " | |
65 | "(NILFS)"); | |
66 | MODULE_VERSION(NILFS_VERSION); | |
67 | MODULE_LICENSE("GPL"); | |
68 | ||
69 | static int nilfs_remount(struct super_block *sb, int *flags, char *data); | |
70 | static int test_exclusive_mount(struct file_system_type *fs_type, | |
71 | struct block_device *bdev, int flags); | |
72 | ||
73 | /** | |
74 | * nilfs_error() - report failure condition on a filesystem | |
75 | * | |
76 | * nilfs_error() sets an ERROR_FS flag on the superblock as well as | |
77 | * reporting an error message. It should be called when NILFS detects | |
78 | * incoherences or defects of meta data on disk. As for sustainable | |
79 | * errors such as a single-shot I/O error, nilfs_warning() or the printk() | |
80 | * function should be used instead. | |
81 | * | |
82 | * The segment constructor must not call this function because it can | |
83 | * kill itself. | |
84 | */ | |
85 | void nilfs_error(struct super_block *sb, const char *function, | |
86 | const char *fmt, ...) | |
87 | { | |
88 | struct nilfs_sb_info *sbi = NILFS_SB(sb); | |
89 | va_list args; | |
90 | ||
91 | va_start(args, fmt); | |
92 | printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function); | |
93 | vprintk(fmt, args); | |
94 | printk("\n"); | |
95 | va_end(args); | |
96 | ||
97 | if (!(sb->s_flags & MS_RDONLY)) { | |
98 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
99 | ||
100 | if (!nilfs_test_opt(sbi, ERRORS_CONT)) | |
101 | nilfs_detach_segment_constructor(sbi); | |
102 | ||
103 | down_write(&nilfs->ns_sem); | |
104 | if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) { | |
105 | nilfs->ns_mount_state |= NILFS_ERROR_FS; | |
106 | nilfs->ns_sbp->s_state |= cpu_to_le16(NILFS_ERROR_FS); | |
107 | nilfs_commit_super(sbi); | |
108 | } | |
109 | up_write(&nilfs->ns_sem); | |
110 | ||
111 | if (nilfs_test_opt(sbi, ERRORS_RO)) { | |
112 | printk(KERN_CRIT "Remounting filesystem read-only\n"); | |
113 | sb->s_flags |= MS_RDONLY; | |
114 | } | |
115 | } | |
116 | ||
117 | if (nilfs_test_opt(sbi, ERRORS_PANIC)) | |
118 | panic("NILFS (device %s): panic forced after error\n", | |
119 | sb->s_id); | |
120 | } | |
121 | ||
122 | void nilfs_warning(struct super_block *sb, const char *function, | |
123 | const char *fmt, ...) | |
124 | { | |
125 | va_list args; | |
126 | ||
127 | va_start(args, fmt); | |
128 | printk(KERN_WARNING "NILFS warning (device %s): %s: ", | |
129 | sb->s_id, function); | |
130 | vprintk(fmt, args); | |
131 | printk("\n"); | |
132 | va_end(args); | |
133 | } | |
134 | ||
135 | static struct kmem_cache *nilfs_inode_cachep; | |
136 | ||
137 | struct inode *nilfs_alloc_inode(struct super_block *sb) | |
138 | { | |
139 | struct nilfs_inode_info *ii; | |
140 | ||
141 | ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS); | |
142 | if (!ii) | |
143 | return NULL; | |
144 | ii->i_bh = NULL; | |
145 | ii->i_state = 0; | |
146 | ii->vfs_inode.i_version = 1; | |
147 | nilfs_btnode_cache_init(&ii->i_btnode_cache); | |
148 | return &ii->vfs_inode; | |
149 | } | |
150 | ||
151 | void nilfs_destroy_inode(struct inode *inode) | |
152 | { | |
153 | kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode)); | |
154 | } | |
155 | ||
156 | static void init_once(void *obj) | |
157 | { | |
158 | struct nilfs_inode_info *ii = obj; | |
159 | ||
160 | INIT_LIST_HEAD(&ii->i_dirty); | |
161 | #ifdef CONFIG_NILFS_XATTR | |
162 | init_rwsem(&ii->xattr_sem); | |
163 | #endif | |
164 | nilfs_btnode_cache_init_once(&ii->i_btnode_cache); | |
165 | ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union; | |
166 | inode_init_once(&ii->vfs_inode); | |
167 | } | |
168 | ||
169 | static int nilfs_init_inode_cache(void) | |
170 | { | |
171 | nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache", | |
172 | sizeof(struct nilfs_inode_info), | |
173 | 0, SLAB_RECLAIM_ACCOUNT, | |
174 | init_once); | |
175 | ||
176 | return (nilfs_inode_cachep == NULL) ? -ENOMEM : 0; | |
177 | } | |
178 | ||
179 | static inline void nilfs_destroy_inode_cache(void) | |
180 | { | |
181 | kmem_cache_destroy(nilfs_inode_cachep); | |
182 | } | |
183 | ||
184 | static void nilfs_clear_inode(struct inode *inode) | |
185 | { | |
186 | struct nilfs_inode_info *ii = NILFS_I(inode); | |
187 | struct nilfs_transaction_info ti; | |
188 | struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb); | |
189 | ||
190 | #ifdef CONFIG_NILFS_POSIX_ACL | |
191 | if (ii->i_acl && ii->i_acl != NILFS_ACL_NOT_CACHED) { | |
192 | posix_acl_release(ii->i_acl); | |
193 | ii->i_acl = NILFS_ACL_NOT_CACHED; | |
194 | } | |
195 | if (ii->i_default_acl && ii->i_default_acl != NILFS_ACL_NOT_CACHED) { | |
196 | posix_acl_release(ii->i_default_acl); | |
197 | ii->i_default_acl = NILFS_ACL_NOT_CACHED; | |
198 | } | |
199 | #endif | |
200 | /* | |
201 | * Free resources allocated in nilfs_read_inode(), here. | |
202 | */ | |
203 | nilfs_transaction_begin(inode->i_sb, &ti, 0); | |
204 | ||
205 | spin_lock(&sbi->s_inode_lock); | |
206 | if (!list_empty(&ii->i_dirty)) | |
207 | list_del_init(&ii->i_dirty); | |
208 | brelse(ii->i_bh); | |
209 | ii->i_bh = NULL; | |
210 | spin_unlock(&sbi->s_inode_lock); | |
211 | ||
212 | if (test_bit(NILFS_I_BMAP, &ii->i_state)) | |
213 | nilfs_bmap_clear(ii->i_bmap); | |
214 | ||
215 | nilfs_btnode_cache_clear(&ii->i_btnode_cache); | |
216 | ||
217 | nilfs_transaction_end(inode->i_sb, 0); | |
218 | } | |
219 | ||
220 | /** | |
221 | * nilfs_update_last_segment - change pointer to the latest segment | |
222 | * @sbi: nilfs_sb_info | |
223 | * @update_cno: flag whether to update checkpoint number. | |
224 | * | |
225 | * nilfs_update_last_segment() changes information in the super block | |
226 | * after a partial segment is written out successfully. The super | |
227 | * block is marked dirty. It will be written out at the next VFS sync | |
228 | * operations such as sync_supers() and generic_shutdown_super(). | |
229 | */ | |
230 | void nilfs_update_last_segment(struct nilfs_sb_info *sbi, int update_cno) | |
231 | { | |
232 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
233 | struct nilfs_super_block *sbp = nilfs->ns_sbp; | |
234 | ||
235 | /* nilfs->sem must be locked by the caller. */ | |
236 | spin_lock(&nilfs->ns_last_segment_lock); | |
237 | if (update_cno) | |
238 | nilfs->ns_last_cno = nilfs->ns_cno++; | |
239 | sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq); | |
240 | sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg); | |
241 | sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno); | |
242 | spin_unlock(&nilfs->ns_last_segment_lock); | |
243 | ||
244 | sbi->s_super->s_dirt = 1; /* must be set if delaying the call of | |
245 | nilfs_commit_super() */ | |
246 | } | |
247 | ||
248 | static int nilfs_sync_super(struct nilfs_sb_info *sbi) | |
249 | { | |
250 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
251 | int err; | |
252 | int barrier_done = 0; | |
253 | ||
254 | if (nilfs_test_opt(sbi, BARRIER)) { | |
255 | set_buffer_ordered(nilfs->ns_sbh); | |
256 | barrier_done = 1; | |
257 | } | |
258 | retry: | |
259 | set_buffer_dirty(nilfs->ns_sbh); | |
260 | err = sync_dirty_buffer(nilfs->ns_sbh); | |
261 | if (err == -EOPNOTSUPP && barrier_done) { | |
262 | nilfs_warning(sbi->s_super, __func__, | |
263 | "barrier-based sync failed. " | |
264 | "disabling barriers\n"); | |
265 | nilfs_clear_opt(sbi, BARRIER); | |
266 | barrier_done = 0; | |
267 | clear_buffer_ordered(nilfs->ns_sbh); | |
268 | goto retry; | |
269 | } | |
270 | if (unlikely(err)) | |
271 | printk(KERN_ERR | |
272 | "NILFS: unable to write superblock (err=%d)\n", err); | |
273 | else { | |
274 | nilfs_dispose_used_segments(nilfs); | |
275 | clear_nilfs_discontinued(nilfs); | |
276 | } | |
277 | ||
278 | return err; | |
279 | } | |
280 | ||
281 | int nilfs_commit_super(struct nilfs_sb_info *sbi) | |
282 | { | |
283 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
284 | struct nilfs_super_block *sbp = nilfs->ns_sbp; | |
285 | sector_t nfreeblocks; | |
286 | int err; | |
287 | ||
288 | /* nilfs->sem must be locked by the caller. */ | |
289 | err = nilfs_count_free_blocks(nilfs, &nfreeblocks); | |
290 | if (unlikely(err)) { | |
291 | printk(KERN_ERR "NILFS: failed to count free blocks\n"); | |
292 | return err; | |
293 | } | |
294 | sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks); | |
295 | sbp->s_wtime = cpu_to_le64(get_seconds()); | |
296 | sbp->s_sum = 0; | |
297 | sbp->s_sum = crc32_le(nilfs->ns_crc_seed, (unsigned char *)sbp, | |
298 | le16_to_cpu(sbp->s_bytes)); | |
299 | ||
300 | sbi->s_super->s_dirt = 0; | |
301 | return nilfs_sync_super(sbi); | |
302 | } | |
303 | ||
304 | static void nilfs_put_super(struct super_block *sb) | |
305 | { | |
306 | struct nilfs_sb_info *sbi = NILFS_SB(sb); | |
307 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
308 | ||
309 | nilfs_detach_segment_constructor(sbi); | |
310 | ||
311 | if (!(sb->s_flags & MS_RDONLY)) { | |
312 | down_write(&nilfs->ns_sem); | |
313 | nilfs->ns_sbp->s_state = cpu_to_le16(nilfs->ns_mount_state); | |
314 | nilfs_commit_super(sbi); | |
315 | up_write(&nilfs->ns_sem); | |
316 | } | |
317 | ||
318 | nilfs_detach_checkpoint(sbi); | |
319 | put_nilfs(sbi->s_nilfs); | |
320 | sbi->s_super = NULL; | |
321 | sb->s_fs_info = NULL; | |
322 | kfree(sbi); | |
323 | } | |
324 | ||
325 | /** | |
326 | * nilfs_write_super - write super block(s) of NILFS | |
327 | * @sb: super_block | |
328 | * | |
329 | * nilfs_write_super() gets a fs-dependent lock, writes super block(s), and | |
330 | * clears s_dirt. This function is called in the section protected by | |
331 | * lock_super(). | |
332 | * | |
333 | * The s_dirt flag is managed by each filesystem and we protect it by ns_sem | |
334 | * of the struct the_nilfs. Lock order must be as follows: | |
335 | * | |
336 | * 1. lock_super() | |
337 | * 2. down_write(&nilfs->ns_sem) | |
338 | * | |
339 | * Inside NILFS, locking ns_sem is enough to protect s_dirt and the buffer | |
340 | * of the super block (nilfs->ns_sbp). | |
341 | * | |
342 | * In most cases, VFS functions call lock_super() before calling these | |
343 | * methods. So we must be careful not to bring on deadlocks when using | |
344 | * lock_super(); see generic_shutdown_super(), write_super(), and so on. | |
345 | * | |
346 | * Note that order of lock_kernel() and lock_super() depends on contexts | |
347 | * of VFS. We should also note that lock_kernel() can be used in its | |
348 | * protective section and only the outermost one has an effect. | |
349 | */ | |
350 | static void nilfs_write_super(struct super_block *sb) | |
351 | { | |
352 | struct nilfs_sb_info *sbi = NILFS_SB(sb); | |
353 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
354 | ||
355 | down_write(&nilfs->ns_sem); | |
356 | if (!(sb->s_flags & MS_RDONLY)) | |
357 | nilfs_commit_super(sbi); | |
358 | sb->s_dirt = 0; | |
359 | up_write(&nilfs->ns_sem); | |
360 | } | |
361 | ||
362 | static int nilfs_sync_fs(struct super_block *sb, int wait) | |
363 | { | |
364 | int err = 0; | |
365 | ||
366 | /* This function is called when super block should be written back */ | |
367 | if (wait) | |
368 | err = nilfs_construct_segment(sb); | |
369 | return err; | |
370 | } | |
371 | ||
372 | int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno) | |
373 | { | |
374 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
375 | struct nilfs_checkpoint *raw_cp; | |
376 | struct buffer_head *bh_cp; | |
377 | int err; | |
378 | ||
379 | down_write(&nilfs->ns_sem); | |
380 | list_add(&sbi->s_list, &nilfs->ns_supers); | |
381 | up_write(&nilfs->ns_sem); | |
382 | ||
383 | sbi->s_ifile = nilfs_mdt_new( | |
384 | nilfs, sbi->s_super, NILFS_IFILE_INO, NILFS_IFILE_GFP); | |
385 | if (!sbi->s_ifile) | |
386 | return -ENOMEM; | |
387 | ||
388 | err = nilfs_palloc_init_blockgroup(sbi->s_ifile, nilfs->ns_inode_size); | |
389 | if (unlikely(err)) | |
390 | goto failed; | |
391 | ||
392 | err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp, | |
393 | &bh_cp); | |
394 | if (unlikely(err)) { | |
395 | if (err == -ENOENT || err == -EINVAL) { | |
396 | printk(KERN_ERR | |
397 | "NILFS: Invalid checkpoint " | |
398 | "(checkpoint number=%llu)\n", | |
399 | (unsigned long long)cno); | |
400 | err = -EINVAL; | |
401 | } | |
402 | goto failed; | |
403 | } | |
404 | err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode); | |
405 | if (unlikely(err)) | |
406 | goto failed_bh; | |
407 | atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count)); | |
408 | atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count)); | |
409 | ||
410 | nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp); | |
411 | return 0; | |
412 | ||
413 | failed_bh: | |
414 | nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp); | |
415 | failed: | |
416 | nilfs_mdt_destroy(sbi->s_ifile); | |
417 | sbi->s_ifile = NULL; | |
418 | ||
419 | down_write(&nilfs->ns_sem); | |
420 | list_del_init(&sbi->s_list); | |
421 | up_write(&nilfs->ns_sem); | |
422 | ||
423 | return err; | |
424 | } | |
425 | ||
426 | void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi) | |
427 | { | |
428 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
429 | ||
430 | nilfs_mdt_clear(sbi->s_ifile); | |
431 | nilfs_mdt_destroy(sbi->s_ifile); | |
432 | sbi->s_ifile = NULL; | |
433 | down_write(&nilfs->ns_sem); | |
434 | list_del_init(&sbi->s_list); | |
435 | up_write(&nilfs->ns_sem); | |
436 | } | |
437 | ||
438 | static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi) | |
439 | { | |
440 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
441 | int err = 0; | |
442 | ||
443 | down_write(&nilfs->ns_sem); | |
444 | if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) { | |
445 | nilfs->ns_mount_state |= NILFS_VALID_FS; | |
446 | err = nilfs_commit_super(sbi); | |
447 | if (likely(!err)) | |
448 | printk(KERN_INFO "NILFS: recovery complete.\n"); | |
449 | } | |
450 | up_write(&nilfs->ns_sem); | |
451 | return err; | |
452 | } | |
453 | ||
454 | static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf) | |
455 | { | |
456 | struct super_block *sb = dentry->d_sb; | |
457 | struct nilfs_sb_info *sbi = NILFS_SB(sb); | |
458 | unsigned long long blocks; | |
459 | unsigned long overhead; | |
460 | unsigned long nrsvblocks; | |
461 | sector_t nfreeblocks; | |
462 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
463 | int err; | |
464 | ||
465 | /* | |
466 | * Compute all of the segment blocks | |
467 | * | |
468 | * The blocks before first segment and after last segment | |
469 | * are excluded. | |
470 | */ | |
471 | blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments | |
472 | - nilfs->ns_first_data_block; | |
473 | nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment; | |
474 | ||
475 | /* | |
476 | * Compute the overhead | |
477 | * | |
478 | * When distributing meta data blocks outside semgent structure, | |
479 | * We must count them as the overhead. | |
480 | */ | |
481 | overhead = 0; | |
482 | ||
483 | err = nilfs_count_free_blocks(nilfs, &nfreeblocks); | |
484 | if (unlikely(err)) | |
485 | return err; | |
486 | ||
487 | buf->f_type = NILFS_SUPER_MAGIC; | |
488 | buf->f_bsize = sb->s_blocksize; | |
489 | buf->f_blocks = blocks - overhead; | |
490 | buf->f_bfree = nfreeblocks; | |
491 | buf->f_bavail = (buf->f_bfree >= nrsvblocks) ? | |
492 | (buf->f_bfree - nrsvblocks) : 0; | |
493 | buf->f_files = atomic_read(&sbi->s_inodes_count); | |
494 | buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */ | |
495 | buf->f_namelen = NILFS_NAME_LEN; | |
496 | return 0; | |
497 | } | |
498 | ||
499 | static struct super_operations nilfs_sops = { | |
500 | .alloc_inode = nilfs_alloc_inode, | |
501 | .destroy_inode = nilfs_destroy_inode, | |
502 | .dirty_inode = nilfs_dirty_inode, | |
503 | /* .write_inode = nilfs_write_inode, */ | |
504 | /* .put_inode = nilfs_put_inode, */ | |
505 | /* .drop_inode = nilfs_drop_inode, */ | |
506 | .delete_inode = nilfs_delete_inode, | |
507 | .put_super = nilfs_put_super, | |
508 | .write_super = nilfs_write_super, | |
509 | .sync_fs = nilfs_sync_fs, | |
510 | /* .write_super_lockfs */ | |
511 | /* .unlockfs */ | |
512 | .statfs = nilfs_statfs, | |
513 | .remount_fs = nilfs_remount, | |
514 | .clear_inode = nilfs_clear_inode, | |
515 | /* .umount_begin */ | |
516 | /* .show_options */ | |
517 | }; | |
518 | ||
519 | static struct inode * | |
520 | nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation) | |
521 | { | |
522 | struct inode *inode; | |
523 | ||
524 | if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO && | |
525 | ino != NILFS_SKETCH_INO) | |
526 | return ERR_PTR(-ESTALE); | |
527 | ||
528 | inode = nilfs_iget(sb, ino); | |
529 | if (IS_ERR(inode)) | |
530 | return ERR_CAST(inode); | |
531 | if (generation && inode->i_generation != generation) { | |
532 | iput(inode); | |
533 | return ERR_PTR(-ESTALE); | |
534 | } | |
535 | ||
536 | return inode; | |
537 | } | |
538 | ||
539 | static struct dentry * | |
540 | nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len, | |
541 | int fh_type) | |
542 | { | |
543 | return generic_fh_to_dentry(sb, fid, fh_len, fh_type, | |
544 | nilfs_nfs_get_inode); | |
545 | } | |
546 | ||
547 | static struct dentry * | |
548 | nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len, | |
549 | int fh_type) | |
550 | { | |
551 | return generic_fh_to_parent(sb, fid, fh_len, fh_type, | |
552 | nilfs_nfs_get_inode); | |
553 | } | |
554 | ||
555 | static struct export_operations nilfs_export_ops = { | |
556 | .fh_to_dentry = nilfs_fh_to_dentry, | |
557 | .fh_to_parent = nilfs_fh_to_parent, | |
558 | .get_parent = nilfs_get_parent, | |
559 | }; | |
560 | ||
561 | enum { | |
562 | Opt_err_cont, Opt_err_panic, Opt_err_ro, | |
563 | Opt_barrier, Opt_snapshot, Opt_order, | |
564 | Opt_err, | |
565 | }; | |
566 | ||
567 | static match_table_t tokens = { | |
568 | {Opt_err_cont, "errors=continue"}, | |
569 | {Opt_err_panic, "errors=panic"}, | |
570 | {Opt_err_ro, "errors=remount-ro"}, | |
571 | {Opt_barrier, "barrier=%s"}, | |
572 | {Opt_snapshot, "cp=%u"}, | |
573 | {Opt_order, "order=%s"}, | |
574 | {Opt_err, NULL} | |
575 | }; | |
576 | ||
577 | static int match_bool(substring_t *s, int *result) | |
578 | { | |
579 | int len = s->to - s->from; | |
580 | ||
581 | if (strncmp(s->from, "on", len) == 0) | |
582 | *result = 1; | |
583 | else if (strncmp(s->from, "off", len) == 0) | |
584 | *result = 0; | |
585 | else | |
586 | return 1; | |
587 | return 0; | |
588 | } | |
589 | ||
590 | static int parse_options(char *options, struct super_block *sb) | |
591 | { | |
592 | struct nilfs_sb_info *sbi = NILFS_SB(sb); | |
593 | char *p; | |
594 | substring_t args[MAX_OPT_ARGS]; | |
595 | int option; | |
596 | ||
597 | if (!options) | |
598 | return 1; | |
599 | ||
600 | while ((p = strsep(&options, ",")) != NULL) { | |
601 | int token; | |
602 | if (!*p) | |
603 | continue; | |
604 | ||
605 | token = match_token(p, tokens, args); | |
606 | switch (token) { | |
607 | case Opt_barrier: | |
608 | if (match_bool(&args[0], &option)) | |
609 | return 0; | |
610 | if (option) | |
611 | nilfs_set_opt(sbi, BARRIER); | |
612 | else | |
613 | nilfs_clear_opt(sbi, BARRIER); | |
614 | break; | |
615 | case Opt_order: | |
616 | if (strcmp(args[0].from, "relaxed") == 0) | |
617 | /* Ordered data semantics */ | |
618 | nilfs_clear_opt(sbi, STRICT_ORDER); | |
619 | else if (strcmp(args[0].from, "strict") == 0) | |
620 | /* Strict in-order semantics */ | |
621 | nilfs_set_opt(sbi, STRICT_ORDER); | |
622 | else | |
623 | return 0; | |
624 | break; | |
625 | case Opt_err_panic: | |
626 | nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC); | |
627 | break; | |
628 | case Opt_err_ro: | |
629 | nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO); | |
630 | break; | |
631 | case Opt_err_cont: | |
632 | nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT); | |
633 | break; | |
634 | case Opt_snapshot: | |
635 | if (match_int(&args[0], &option) || option <= 0) | |
636 | return 0; | |
637 | if (!(sb->s_flags & MS_RDONLY)) | |
638 | return 0; | |
639 | sbi->s_snapshot_cno = option; | |
640 | nilfs_set_opt(sbi, SNAPSHOT); | |
641 | break; | |
642 | default: | |
643 | printk(KERN_ERR | |
644 | "NILFS: Unrecognized mount option \"%s\"\n", p); | |
645 | return 0; | |
646 | } | |
647 | } | |
648 | return 1; | |
649 | } | |
650 | ||
651 | static inline void | |
652 | nilfs_set_default_options(struct nilfs_sb_info *sbi, | |
653 | struct nilfs_super_block *sbp) | |
654 | { | |
655 | sbi->s_mount_opt = | |
656 | NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER; | |
657 | } | |
658 | ||
659 | static int nilfs_setup_super(struct nilfs_sb_info *sbi) | |
660 | { | |
661 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
662 | struct nilfs_super_block *sbp = nilfs->ns_sbp; | |
663 | int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count); | |
664 | int mnt_count = le16_to_cpu(sbp->s_mnt_count); | |
665 | ||
666 | /* nilfs->sem must be locked by the caller. */ | |
667 | if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) { | |
668 | printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n"); | |
669 | } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) { | |
670 | printk(KERN_WARNING | |
671 | "NILFS warning: mounting fs with errors\n"); | |
672 | #if 0 | |
673 | } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) { | |
674 | printk(KERN_WARNING | |
675 | "NILFS warning: maximal mount count reached\n"); | |
676 | #endif | |
677 | } | |
678 | if (!max_mnt_count) | |
679 | sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT); | |
680 | ||
681 | sbp->s_mnt_count = cpu_to_le16(mnt_count + 1); | |
682 | sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS); | |
683 | sbp->s_mtime = cpu_to_le64(get_seconds()); | |
684 | return nilfs_commit_super(sbi); | |
685 | } | |
686 | ||
687 | struct nilfs_super_block * | |
688 | nilfs_load_super_block(struct super_block *sb, struct buffer_head **pbh) | |
689 | { | |
690 | int blocksize; | |
691 | unsigned long offset, sb_index; | |
692 | ||
693 | /* | |
694 | * Adjusting block size | |
695 | * Blocksize will be enlarged when it is smaller than hardware | |
696 | * sector size. | |
697 | * Disk format of superblock does not change. | |
698 | */ | |
699 | blocksize = sb_min_blocksize(sb, BLOCK_SIZE); | |
700 | if (!blocksize) { | |
701 | printk(KERN_ERR | |
702 | "NILFS: unable to set blocksize of superblock\n"); | |
703 | return NULL; | |
704 | } | |
705 | sb_index = NILFS_SB_OFFSET_BYTES / blocksize; | |
706 | offset = NILFS_SB_OFFSET_BYTES % blocksize; | |
707 | ||
708 | *pbh = sb_bread(sb, sb_index); | |
709 | if (!*pbh) { | |
710 | printk(KERN_ERR "NILFS: unable to read superblock\n"); | |
711 | return NULL; | |
712 | } | |
713 | return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset); | |
714 | } | |
715 | ||
716 | struct nilfs_super_block * | |
717 | nilfs_reload_super_block(struct super_block *sb, struct buffer_head **pbh, | |
718 | int blocksize) | |
719 | { | |
720 | struct nilfs_super_block *sbp; | |
721 | unsigned long offset, sb_index; | |
722 | int hw_blocksize = bdev_hardsect_size(sb->s_bdev); | |
723 | ||
724 | if (blocksize < hw_blocksize) { | |
725 | printk(KERN_ERR | |
726 | "NILFS: blocksize %d too small for device " | |
727 | "(sector-size = %d).\n", | |
728 | blocksize, hw_blocksize); | |
729 | goto failed_sbh; | |
730 | } | |
731 | brelse(*pbh); | |
732 | sb_set_blocksize(sb, blocksize); | |
733 | ||
734 | sb_index = NILFS_SB_OFFSET_BYTES / blocksize; | |
735 | offset = NILFS_SB_OFFSET_BYTES % blocksize; | |
736 | ||
737 | *pbh = sb_bread(sb, sb_index); | |
738 | if (!*pbh) { | |
739 | printk(KERN_ERR | |
740 | "NILFS: cannot read superblock on 2nd try.\n"); | |
741 | goto failed; | |
742 | } | |
743 | ||
744 | sbp = (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset); | |
745 | if (sbp->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) { | |
746 | printk(KERN_ERR | |
747 | "NILFS: !? Magic mismatch on 2nd try.\n"); | |
748 | goto failed_sbh; | |
749 | } | |
750 | return sbp; | |
751 | ||
752 | failed_sbh: | |
753 | brelse(*pbh); | |
754 | ||
755 | failed: | |
756 | return NULL; | |
757 | } | |
758 | ||
759 | int nilfs_store_magic_and_option(struct super_block *sb, | |
760 | struct nilfs_super_block *sbp, | |
761 | char *data) | |
762 | { | |
763 | struct nilfs_sb_info *sbi = NILFS_SB(sb); | |
764 | ||
765 | /* trying to fill super (1st stage) */ | |
766 | sb->s_magic = le16_to_cpu(sbp->s_magic); | |
767 | ||
768 | /* FS independent flags */ | |
769 | #ifdef NILFS_ATIME_DISABLE | |
770 | sb->s_flags |= MS_NOATIME; | |
771 | #endif | |
772 | ||
773 | if (sb->s_magic != NILFS_SUPER_MAGIC) { | |
774 | printk("NILFS: Can't find nilfs on dev %s.\n", sb->s_id); | |
775 | return -EINVAL; | |
776 | } | |
777 | ||
778 | nilfs_set_default_options(sbi, sbp); | |
779 | ||
780 | sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid); | |
781 | sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid); | |
782 | sbi->s_interval = le32_to_cpu(sbp->s_c_interval); | |
783 | sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max); | |
784 | ||
785 | if (!parse_options(data, sb)) | |
786 | return -EINVAL; | |
787 | ||
788 | return 0; | |
789 | } | |
790 | ||
791 | /** | |
792 | * nilfs_fill_super() - initialize a super block instance | |
793 | * @sb: super_block | |
794 | * @data: mount options | |
795 | * @silent: silent mode flag | |
796 | * @nilfs: the_nilfs struct | |
797 | * | |
798 | * This function is called exclusively by bd_mount_mutex. | |
799 | * So, the recovery process is protected from other simultaneous mounts. | |
800 | */ | |
801 | static int | |
802 | nilfs_fill_super(struct super_block *sb, void *data, int silent, | |
803 | struct the_nilfs *nilfs) | |
804 | { | |
805 | struct nilfs_sb_info *sbi; | |
806 | struct inode *root; | |
807 | __u64 cno; | |
808 | int err; | |
809 | ||
810 | sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); | |
811 | if (!sbi) | |
812 | return -ENOMEM; | |
813 | ||
814 | sb->s_fs_info = sbi; | |
815 | ||
816 | get_nilfs(nilfs); | |
817 | sbi->s_nilfs = nilfs; | |
818 | sbi->s_super = sb; | |
819 | ||
820 | err = init_nilfs(nilfs, sbi, (char *)data); | |
821 | if (err) | |
822 | goto failed_sbi; | |
823 | ||
824 | spin_lock_init(&sbi->s_inode_lock); | |
825 | INIT_LIST_HEAD(&sbi->s_dirty_files); | |
826 | INIT_LIST_HEAD(&sbi->s_list); | |
827 | ||
828 | /* | |
829 | * Following initialization is overlapped because | |
830 | * nilfs_sb_info structure has been cleared at the beginning. | |
831 | * But we reserve them to keep our interest and make ready | |
832 | * for the future change. | |
833 | */ | |
834 | get_random_bytes(&sbi->s_next_generation, | |
835 | sizeof(sbi->s_next_generation)); | |
836 | spin_lock_init(&sbi->s_next_gen_lock); | |
837 | ||
838 | sb->s_op = &nilfs_sops; | |
839 | sb->s_export_op = &nilfs_export_ops; | |
840 | sb->s_root = NULL; | |
841 | ||
842 | if (!nilfs_loaded(nilfs)) { | |
843 | err = load_nilfs(nilfs, sbi); | |
844 | if (err) | |
845 | goto failed_sbi; | |
846 | } | |
847 | cno = nilfs_last_cno(nilfs); | |
848 | ||
849 | if (sb->s_flags & MS_RDONLY) { | |
850 | if (nilfs_test_opt(sbi, SNAPSHOT)) { | |
851 | if (!nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, | |
852 | sbi->s_snapshot_cno)) { | |
853 | printk(KERN_ERR | |
854 | "NILFS: The specified checkpoint is " | |
855 | "not a snapshot " | |
856 | "(checkpoint number=%llu).\n", | |
857 | (unsigned long long)sbi->s_snapshot_cno); | |
858 | err = -EINVAL; | |
859 | goto failed_sbi; | |
860 | } | |
861 | cno = sbi->s_snapshot_cno; | |
862 | } else | |
863 | /* Read-only mount */ | |
864 | sbi->s_snapshot_cno = cno; | |
865 | } | |
866 | ||
867 | err = nilfs_attach_checkpoint(sbi, cno); | |
868 | if (err) { | |
869 | printk(KERN_ERR "NILFS: error loading a checkpoint" | |
870 | " (checkpoint number=%llu).\n", (unsigned long long)cno); | |
871 | goto failed_sbi; | |
872 | } | |
873 | ||
874 | if (!(sb->s_flags & MS_RDONLY)) { | |
875 | err = nilfs_attach_segment_constructor(sbi, NULL); | |
876 | if (err) | |
877 | goto failed_checkpoint; | |
878 | } | |
879 | ||
880 | root = nilfs_iget(sb, NILFS_ROOT_INO); | |
881 | if (IS_ERR(root)) { | |
882 | printk(KERN_ERR "NILFS: get root inode failed\n"); | |
883 | err = PTR_ERR(root); | |
884 | goto failed_segctor; | |
885 | } | |
886 | if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { | |
887 | iput(root); | |
888 | printk(KERN_ERR "NILFS: corrupt root inode.\n"); | |
889 | err = -EINVAL; | |
890 | goto failed_segctor; | |
891 | } | |
892 | sb->s_root = d_alloc_root(root); | |
893 | if (!sb->s_root) { | |
894 | iput(root); | |
895 | printk(KERN_ERR "NILFS: get root dentry failed\n"); | |
896 | err = -ENOMEM; | |
897 | goto failed_segctor; | |
898 | } | |
899 | ||
900 | if (!(sb->s_flags & MS_RDONLY)) { | |
901 | down_write(&nilfs->ns_sem); | |
902 | nilfs_setup_super(sbi); | |
903 | up_write(&nilfs->ns_sem); | |
904 | } | |
905 | ||
906 | err = nilfs_mark_recovery_complete(sbi); | |
907 | if (unlikely(err)) { | |
908 | printk(KERN_ERR "NILFS: recovery failed.\n"); | |
909 | goto failed_root; | |
910 | } | |
911 | ||
912 | return 0; | |
913 | ||
914 | failed_root: | |
915 | dput(sb->s_root); | |
916 | sb->s_root = NULL; | |
917 | ||
918 | failed_segctor: | |
919 | nilfs_detach_segment_constructor(sbi); | |
920 | ||
921 | failed_checkpoint: | |
922 | nilfs_detach_checkpoint(sbi); | |
923 | ||
924 | failed_sbi: | |
925 | put_nilfs(nilfs); | |
926 | sb->s_fs_info = NULL; | |
927 | kfree(sbi); | |
928 | return err; | |
929 | } | |
930 | ||
931 | static int nilfs_remount(struct super_block *sb, int *flags, char *data) | |
932 | { | |
933 | struct nilfs_sb_info *sbi = NILFS_SB(sb); | |
934 | struct nilfs_super_block *sbp; | |
935 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
936 | unsigned long old_sb_flags; | |
937 | struct nilfs_mount_options old_opts; | |
938 | int err; | |
939 | ||
940 | old_sb_flags = sb->s_flags; | |
941 | old_opts.mount_opt = sbi->s_mount_opt; | |
942 | old_opts.snapshot_cno = sbi->s_snapshot_cno; | |
943 | ||
944 | if (!parse_options(data, sb)) { | |
945 | err = -EINVAL; | |
946 | goto restore_opts; | |
947 | } | |
948 | sb->s_flags = (sb->s_flags & ~MS_POSIXACL); | |
949 | ||
950 | if ((*flags & MS_RDONLY) && | |
951 | sbi->s_snapshot_cno != old_opts.snapshot_cno) { | |
952 | printk(KERN_WARNING "NILFS (device %s): couldn't " | |
953 | "remount to a different snapshot. \n", | |
954 | sb->s_id); | |
955 | err = -EINVAL; | |
956 | goto restore_opts; | |
957 | } | |
958 | ||
959 | if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY)) | |
960 | goto out; | |
961 | if (*flags & MS_RDONLY) { | |
962 | /* Shutting down the segment constructor */ | |
963 | nilfs_detach_segment_constructor(sbi); | |
964 | sb->s_flags |= MS_RDONLY; | |
965 | ||
966 | sbi->s_snapshot_cno = nilfs_last_cno(nilfs); | |
967 | /* nilfs_set_opt(sbi, SNAPSHOT); */ | |
968 | ||
969 | /* | |
970 | * Remounting a valid RW partition RDONLY, so set | |
971 | * the RDONLY flag and then mark the partition as valid again. | |
972 | */ | |
973 | down_write(&nilfs->ns_sem); | |
974 | sbp = nilfs->ns_sbp; | |
975 | if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) && | |
976 | (nilfs->ns_mount_state & NILFS_VALID_FS)) | |
977 | sbp->s_state = cpu_to_le16(nilfs->ns_mount_state); | |
978 | sbp->s_mtime = cpu_to_le64(get_seconds()); | |
979 | nilfs_commit_super(sbi); | |
980 | up_write(&nilfs->ns_sem); | |
981 | } else { | |
982 | /* | |
983 | * Mounting a RDONLY partition read-write, so reread and | |
984 | * store the current valid flag. (It may have been changed | |
985 | * by fsck since we originally mounted the partition.) | |
986 | */ | |
987 | down(&sb->s_bdev->bd_mount_sem); | |
988 | /* Check existing RW-mount */ | |
989 | if (test_exclusive_mount(sb->s_type, sb->s_bdev, 0)) { | |
990 | printk(KERN_WARNING "NILFS (device %s): couldn't " | |
991 | "remount because a RW-mount exists.\n", | |
992 | sb->s_id); | |
993 | err = -EBUSY; | |
994 | goto rw_remount_failed; | |
995 | } | |
996 | if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) { | |
997 | printk(KERN_WARNING "NILFS (device %s): couldn't " | |
998 | "remount because the current RO-mount is not " | |
999 | "the latest one.\n", | |
1000 | sb->s_id); | |
1001 | err = -EINVAL; | |
1002 | goto rw_remount_failed; | |
1003 | } | |
1004 | sb->s_flags &= ~MS_RDONLY; | |
1005 | nilfs_clear_opt(sbi, SNAPSHOT); | |
1006 | sbi->s_snapshot_cno = 0; | |
1007 | ||
1008 | err = nilfs_attach_segment_constructor(sbi, NULL); | |
1009 | if (err) | |
1010 | goto rw_remount_failed; | |
1011 | ||
1012 | down_write(&nilfs->ns_sem); | |
1013 | nilfs_setup_super(sbi); | |
1014 | up_write(&nilfs->ns_sem); | |
1015 | ||
1016 | up(&sb->s_bdev->bd_mount_sem); | |
1017 | } | |
1018 | out: | |
1019 | return 0; | |
1020 | ||
1021 | rw_remount_failed: | |
1022 | up(&sb->s_bdev->bd_mount_sem); | |
1023 | restore_opts: | |
1024 | sb->s_flags = old_sb_flags; | |
1025 | sbi->s_mount_opt = old_opts.mount_opt; | |
1026 | sbi->s_snapshot_cno = old_opts.snapshot_cno; | |
1027 | return err; | |
1028 | } | |
1029 | ||
1030 | struct nilfs_super_data { | |
1031 | struct block_device *bdev; | |
1032 | __u64 cno; | |
1033 | int flags; | |
1034 | }; | |
1035 | ||
1036 | /** | |
1037 | * nilfs_identify - pre-read mount options needed to identify mount instance | |
1038 | * @data: mount options | |
1039 | * @sd: nilfs_super_data | |
1040 | */ | |
1041 | static int nilfs_identify(char *data, struct nilfs_super_data *sd) | |
1042 | { | |
1043 | char *p, *options = data; | |
1044 | substring_t args[MAX_OPT_ARGS]; | |
1045 | int option, token; | |
1046 | int ret = 0; | |
1047 | ||
1048 | do { | |
1049 | p = strsep(&options, ","); | |
1050 | if (p != NULL && *p) { | |
1051 | token = match_token(p, tokens, args); | |
1052 | if (token == Opt_snapshot) { | |
1053 | if (!(sd->flags & MS_RDONLY)) | |
1054 | ret++; | |
1055 | else { | |
1056 | ret = match_int(&args[0], &option); | |
1057 | if (!ret) { | |
1058 | if (option > 0) | |
1059 | sd->cno = option; | |
1060 | else | |
1061 | ret++; | |
1062 | } | |
1063 | } | |
1064 | } | |
1065 | if (ret) | |
1066 | printk(KERN_ERR | |
1067 | "NILFS: invalid mount option: %s\n", p); | |
1068 | } | |
1069 | if (!options) | |
1070 | break; | |
1071 | BUG_ON(options == data); | |
1072 | *(options - 1) = ','; | |
1073 | } while (!ret); | |
1074 | return ret; | |
1075 | } | |
1076 | ||
1077 | static int nilfs_set_bdev_super(struct super_block *s, void *data) | |
1078 | { | |
1079 | struct nilfs_super_data *sd = data; | |
1080 | ||
1081 | s->s_bdev = sd->bdev; | |
1082 | s->s_dev = s->s_bdev->bd_dev; | |
1083 | return 0; | |
1084 | } | |
1085 | ||
1086 | static int nilfs_test_bdev_super(struct super_block *s, void *data) | |
1087 | { | |
1088 | struct nilfs_super_data *sd = data; | |
1089 | ||
1090 | return s->s_bdev == sd->bdev; | |
1091 | } | |
1092 | ||
1093 | static int nilfs_test_bdev_super2(struct super_block *s, void *data) | |
1094 | { | |
1095 | struct nilfs_super_data *sd = data; | |
1096 | int ret; | |
1097 | ||
1098 | if (s->s_bdev != sd->bdev) | |
1099 | return 0; | |
1100 | ||
1101 | if (!((s->s_flags | sd->flags) & MS_RDONLY)) | |
1102 | return 1; /* Reuse an old R/W-mode super_block */ | |
1103 | ||
1104 | if (s->s_flags & sd->flags & MS_RDONLY) { | |
1105 | if (down_read_trylock(&s->s_umount)) { | |
1106 | ret = s->s_root && | |
1107 | (sd->cno == NILFS_SB(s)->s_snapshot_cno); | |
1108 | up_read(&s->s_umount); | |
1109 | /* | |
1110 | * This path is locked with sb_lock by sget(). | |
1111 | * So, drop_super() causes deadlock. | |
1112 | */ | |
1113 | return ret; | |
1114 | } | |
1115 | } | |
1116 | return 0; | |
1117 | } | |
1118 | ||
1119 | static int | |
1120 | nilfs_get_sb(struct file_system_type *fs_type, int flags, | |
1121 | const char *dev_name, void *data, struct vfsmount *mnt) | |
1122 | { | |
1123 | struct nilfs_super_data sd; | |
1124 | struct super_block *s, *s2; | |
1125 | struct the_nilfs *nilfs = NULL; | |
1126 | int err, need_to_close = 1; | |
1127 | ||
1128 | sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type); | |
1129 | if (IS_ERR(sd.bdev)) | |
1130 | return PTR_ERR(sd.bdev); | |
1131 | ||
1132 | /* | |
1133 | * To get mount instance using sget() vfs-routine, NILFS needs | |
1134 | * much more information than normal filesystems to identify mount | |
1135 | * instance. For snapshot mounts, not only a mount type (ro-mount | |
1136 | * or rw-mount) but also a checkpoint number is required. | |
1137 | * The results are passed in sget() using nilfs_super_data. | |
1138 | */ | |
1139 | sd.cno = 0; | |
1140 | sd.flags = flags; | |
1141 | if (nilfs_identify((char *)data, &sd)) { | |
1142 | err = -EINVAL; | |
1143 | goto failed; | |
1144 | } | |
1145 | ||
1146 | /* | |
1147 | * once the super is inserted into the list by sget, s_umount | |
1148 | * will protect the lockfs code from trying to start a snapshot | |
1149 | * while we are mounting | |
1150 | */ | |
1151 | down(&sd.bdev->bd_mount_sem); | |
1152 | if (!sd.cno && | |
1153 | (err = test_exclusive_mount(fs_type, sd.bdev, flags ^ MS_RDONLY))) { | |
1154 | err = (err < 0) ? : -EBUSY; | |
1155 | goto failed_unlock; | |
1156 | } | |
1157 | ||
1158 | /* | |
1159 | * Phase-1: search any existent instance and get the_nilfs | |
1160 | */ | |
1161 | s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd); | |
1162 | if (IS_ERR(s)) | |
1163 | goto error_s; | |
1164 | ||
1165 | if (!s->s_root) { | |
1166 | err = -ENOMEM; | |
1167 | nilfs = alloc_nilfs(sd.bdev); | |
1168 | if (!nilfs) | |
1169 | goto cancel_new; | |
1170 | } else { | |
1171 | struct nilfs_sb_info *sbi = NILFS_SB(s); | |
1172 | ||
1173 | BUG_ON(!sbi || !sbi->s_nilfs); | |
1174 | /* | |
1175 | * s_umount protects super_block from unmount process; | |
1176 | * It covers pointers of nilfs_sb_info and the_nilfs. | |
1177 | */ | |
1178 | nilfs = sbi->s_nilfs; | |
1179 | get_nilfs(nilfs); | |
1180 | up_write(&s->s_umount); | |
1181 | ||
1182 | /* | |
1183 | * Phase-2: search specified snapshot or R/W mode super_block | |
1184 | */ | |
1185 | if (!sd.cno) | |
1186 | /* trying to get the latest checkpoint. */ | |
1187 | sd.cno = nilfs_last_cno(nilfs); | |
1188 | ||
1189 | s2 = sget(fs_type, nilfs_test_bdev_super2, | |
1190 | nilfs_set_bdev_super, &sd); | |
1191 | deactivate_super(s); | |
1192 | /* | |
1193 | * Although deactivate_super() invokes close_bdev_exclusive() at | |
1194 | * kill_block_super(). Here, s is an existent mount; we need | |
1195 | * one more close_bdev_exclusive() call. | |
1196 | */ | |
1197 | s = s2; | |
1198 | if (IS_ERR(s)) | |
1199 | goto error_s; | |
1200 | } | |
1201 | ||
1202 | if (!s->s_root) { | |
1203 | char b[BDEVNAME_SIZE]; | |
1204 | ||
1205 | s->s_flags = flags; | |
1206 | strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id)); | |
1207 | sb_set_blocksize(s, block_size(sd.bdev)); | |
1208 | ||
1209 | err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs); | |
1210 | if (err) | |
1211 | goto cancel_new; | |
1212 | ||
1213 | s->s_flags |= MS_ACTIVE; | |
1214 | need_to_close = 0; | |
1215 | } else if (!(s->s_flags & MS_RDONLY)) { | |
1216 | err = -EBUSY; | |
1217 | } | |
1218 | ||
1219 | up(&sd.bdev->bd_mount_sem); | |
1220 | put_nilfs(nilfs); | |
1221 | if (need_to_close) | |
1222 | close_bdev_exclusive(sd.bdev, flags); | |
1223 | simple_set_mnt(mnt, s); | |
1224 | return 0; | |
1225 | ||
1226 | error_s: | |
1227 | up(&sd.bdev->bd_mount_sem); | |
1228 | if (nilfs) | |
1229 | put_nilfs(nilfs); | |
1230 | close_bdev_exclusive(sd.bdev, flags); | |
1231 | return PTR_ERR(s); | |
1232 | ||
1233 | failed_unlock: | |
1234 | up(&sd.bdev->bd_mount_sem); | |
1235 | failed: | |
1236 | close_bdev_exclusive(sd.bdev, flags); | |
1237 | ||
1238 | return err; | |
1239 | ||
1240 | cancel_new: | |
1241 | /* Abandoning the newly allocated superblock */ | |
1242 | up(&sd.bdev->bd_mount_sem); | |
1243 | if (nilfs) | |
1244 | put_nilfs(nilfs); | |
1245 | up_write(&s->s_umount); | |
1246 | deactivate_super(s); | |
1247 | /* | |
1248 | * deactivate_super() invokes close_bdev_exclusive(). | |
1249 | * We must finish all post-cleaning before this call; | |
1250 | * put_nilfs() and unlocking bd_mount_sem need the block device. | |
1251 | */ | |
1252 | return err; | |
1253 | } | |
1254 | ||
1255 | static int nilfs_test_bdev_super3(struct super_block *s, void *data) | |
1256 | { | |
1257 | struct nilfs_super_data *sd = data; | |
1258 | int ret; | |
1259 | ||
1260 | if (s->s_bdev != sd->bdev) | |
1261 | return 0; | |
1262 | if (down_read_trylock(&s->s_umount)) { | |
1263 | ret = (s->s_flags & MS_RDONLY) && s->s_root && | |
1264 | nilfs_test_opt(NILFS_SB(s), SNAPSHOT); | |
1265 | up_read(&s->s_umount); | |
1266 | if (ret) | |
1267 | return 0; /* ignore snapshot mounts */ | |
1268 | } | |
1269 | return !((sd->flags ^ s->s_flags) & MS_RDONLY); | |
1270 | } | |
1271 | ||
1272 | static int __false_bdev_super(struct super_block *s, void *data) | |
1273 | { | |
1274 | #if 0 /* XXX: workaround for lock debug. This is not good idea */ | |
1275 | up_write(&s->s_umount); | |
1276 | #endif | |
1277 | return -EFAULT; | |
1278 | } | |
1279 | ||
1280 | /** | |
1281 | * test_exclusive_mount - check whether an exclusive RW/RO mount exists or not. | |
1282 | * fs_type: filesystem type | |
1283 | * bdev: block device | |
1284 | * flag: 0 (check rw-mount) or MS_RDONLY (check ro-mount) | |
1285 | * res: pointer to an integer to store result | |
1286 | * | |
1287 | * This function must be called within a section protected by bd_mount_mutex. | |
1288 | */ | |
1289 | static int test_exclusive_mount(struct file_system_type *fs_type, | |
1290 | struct block_device *bdev, int flags) | |
1291 | { | |
1292 | struct super_block *s; | |
1293 | struct nilfs_super_data sd = { .flags = flags, .bdev = bdev }; | |
1294 | ||
1295 | s = sget(fs_type, nilfs_test_bdev_super3, __false_bdev_super, &sd); | |
1296 | if (IS_ERR(s)) { | |
1297 | if (PTR_ERR(s) != -EFAULT) | |
1298 | return PTR_ERR(s); | |
1299 | return 0; /* Not found */ | |
1300 | } | |
1301 | up_write(&s->s_umount); | |
1302 | deactivate_super(s); | |
1303 | return 1; /* Found */ | |
1304 | } | |
1305 | ||
1306 | struct file_system_type nilfs_fs_type = { | |
1307 | .owner = THIS_MODULE, | |
1308 | .name = "nilfs2", | |
1309 | .get_sb = nilfs_get_sb, | |
1310 | .kill_sb = kill_block_super, | |
1311 | .fs_flags = FS_REQUIRES_DEV, | |
1312 | }; | |
1313 | ||
1314 | static int __init init_nilfs_fs(void) | |
1315 | { | |
1316 | int err; | |
1317 | ||
1318 | err = nilfs_init_inode_cache(); | |
1319 | if (err) | |
1320 | goto failed; | |
1321 | ||
1322 | err = nilfs_init_transaction_cache(); | |
1323 | if (err) | |
1324 | goto failed_inode_cache; | |
1325 | ||
1326 | err = nilfs_init_segbuf_cache(); | |
1327 | if (err) | |
1328 | goto failed_transaction_cache; | |
1329 | ||
1330 | err = nilfs_btree_path_cache_init(); | |
1331 | if (err) | |
1332 | goto failed_segbuf_cache; | |
1333 | ||
1334 | err = register_filesystem(&nilfs_fs_type); | |
1335 | if (err) | |
1336 | goto failed_btree_path_cache; | |
1337 | ||
1338 | return 0; | |
1339 | ||
1340 | failed_btree_path_cache: | |
1341 | nilfs_btree_path_cache_destroy(); | |
1342 | ||
1343 | failed_segbuf_cache: | |
1344 | nilfs_destroy_segbuf_cache(); | |
1345 | ||
1346 | failed_transaction_cache: | |
1347 | nilfs_destroy_transaction_cache(); | |
1348 | ||
1349 | failed_inode_cache: | |
1350 | nilfs_destroy_inode_cache(); | |
1351 | ||
1352 | failed: | |
1353 | return err; | |
1354 | } | |
1355 | ||
1356 | static void __exit exit_nilfs_fs(void) | |
1357 | { | |
1358 | nilfs_destroy_segbuf_cache(); | |
1359 | nilfs_destroy_transaction_cache(); | |
1360 | nilfs_destroy_inode_cache(); | |
1361 | nilfs_btree_path_cache_destroy(); | |
1362 | unregister_filesystem(&nilfs_fs_type); | |
1363 | } | |
1364 | ||
1365 | module_init(init_nilfs_fs) | |
1366 | module_exit(exit_nilfs_fs) |