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); | |
783f6184 RK |
187 | |
188 | #ifdef CONFIG_NILFS_POSIX_ACL | |
189 | if (ii->i_acl && ii->i_acl != NILFS_ACL_NOT_CACHED) { | |
190 | posix_acl_release(ii->i_acl); | |
191 | ii->i_acl = NILFS_ACL_NOT_CACHED; | |
192 | } | |
193 | if (ii->i_default_acl && ii->i_default_acl != NILFS_ACL_NOT_CACHED) { | |
194 | posix_acl_release(ii->i_default_acl); | |
195 | ii->i_default_acl = NILFS_ACL_NOT_CACHED; | |
196 | } | |
197 | #endif | |
198 | /* | |
199 | * Free resources allocated in nilfs_read_inode(), here. | |
200 | */ | |
a2e7d2df | 201 | BUG_ON(!list_empty(&ii->i_dirty)); |
783f6184 RK |
202 | brelse(ii->i_bh); |
203 | ii->i_bh = NULL; | |
783f6184 RK |
204 | |
205 | if (test_bit(NILFS_I_BMAP, &ii->i_state)) | |
206 | nilfs_bmap_clear(ii->i_bmap); | |
207 | ||
208 | nilfs_btnode_cache_clear(&ii->i_btnode_cache); | |
783f6184 RK |
209 | } |
210 | ||
211 | /** | |
212 | * nilfs_update_last_segment - change pointer to the latest segment | |
213 | * @sbi: nilfs_sb_info | |
214 | * @update_cno: flag whether to update checkpoint number. | |
215 | * | |
216 | * nilfs_update_last_segment() changes information in the super block | |
217 | * after a partial segment is written out successfully. The super | |
218 | * block is marked dirty. It will be written out at the next VFS sync | |
219 | * operations such as sync_supers() and generic_shutdown_super(). | |
220 | */ | |
221 | void nilfs_update_last_segment(struct nilfs_sb_info *sbi, int update_cno) | |
222 | { | |
223 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
224 | struct nilfs_super_block *sbp = nilfs->ns_sbp; | |
225 | ||
226 | /* nilfs->sem must be locked by the caller. */ | |
227 | spin_lock(&nilfs->ns_last_segment_lock); | |
228 | if (update_cno) | |
229 | nilfs->ns_last_cno = nilfs->ns_cno++; | |
230 | sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq); | |
231 | sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg); | |
232 | sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno); | |
233 | spin_unlock(&nilfs->ns_last_segment_lock); | |
234 | ||
235 | sbi->s_super->s_dirt = 1; /* must be set if delaying the call of | |
236 | nilfs_commit_super() */ | |
237 | } | |
238 | ||
239 | static int nilfs_sync_super(struct nilfs_sb_info *sbi) | |
240 | { | |
241 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
242 | int err; | |
243 | int barrier_done = 0; | |
244 | ||
245 | if (nilfs_test_opt(sbi, BARRIER)) { | |
246 | set_buffer_ordered(nilfs->ns_sbh); | |
247 | barrier_done = 1; | |
248 | } | |
249 | retry: | |
250 | set_buffer_dirty(nilfs->ns_sbh); | |
251 | err = sync_dirty_buffer(nilfs->ns_sbh); | |
252 | if (err == -EOPNOTSUPP && barrier_done) { | |
253 | nilfs_warning(sbi->s_super, __func__, | |
254 | "barrier-based sync failed. " | |
255 | "disabling barriers\n"); | |
256 | nilfs_clear_opt(sbi, BARRIER); | |
257 | barrier_done = 0; | |
258 | clear_buffer_ordered(nilfs->ns_sbh); | |
259 | goto retry; | |
260 | } | |
261 | if (unlikely(err)) | |
262 | printk(KERN_ERR | |
263 | "NILFS: unable to write superblock (err=%d)\n", err); | |
264 | else { | |
783f6184 | 265 | clear_nilfs_discontinued(nilfs); |
2c2e52fc RK |
266 | spin_lock(&nilfs->ns_last_segment_lock); |
267 | nilfs->ns_prot_seq = le64_to_cpu(nilfs->ns_sbp->s_last_seq); | |
268 | spin_unlock(&nilfs->ns_last_segment_lock); | |
783f6184 RK |
269 | } |
270 | ||
271 | return err; | |
272 | } | |
273 | ||
274 | int nilfs_commit_super(struct nilfs_sb_info *sbi) | |
275 | { | |
276 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
277 | struct nilfs_super_block *sbp = nilfs->ns_sbp; | |
278 | sector_t nfreeblocks; | |
279 | int err; | |
280 | ||
281 | /* nilfs->sem must be locked by the caller. */ | |
282 | err = nilfs_count_free_blocks(nilfs, &nfreeblocks); | |
283 | if (unlikely(err)) { | |
284 | printk(KERN_ERR "NILFS: failed to count free blocks\n"); | |
285 | return err; | |
286 | } | |
287 | sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks); | |
288 | sbp->s_wtime = cpu_to_le64(get_seconds()); | |
289 | sbp->s_sum = 0; | |
e6268746 RK |
290 | sbp->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed, |
291 | (unsigned char *)sbp, | |
292 | le16_to_cpu(sbp->s_bytes))); | |
783f6184 RK |
293 | sbi->s_super->s_dirt = 0; |
294 | return nilfs_sync_super(sbi); | |
295 | } | |
296 | ||
297 | static void nilfs_put_super(struct super_block *sb) | |
298 | { | |
299 | struct nilfs_sb_info *sbi = NILFS_SB(sb); | |
300 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
301 | ||
302 | nilfs_detach_segment_constructor(sbi); | |
303 | ||
304 | if (!(sb->s_flags & MS_RDONLY)) { | |
305 | down_write(&nilfs->ns_sem); | |
306 | nilfs->ns_sbp->s_state = cpu_to_le16(nilfs->ns_mount_state); | |
307 | nilfs_commit_super(sbi); | |
308 | up_write(&nilfs->ns_sem); | |
309 | } | |
310 | ||
311 | nilfs_detach_checkpoint(sbi); | |
312 | put_nilfs(sbi->s_nilfs); | |
313 | sbi->s_super = NULL; | |
314 | sb->s_fs_info = NULL; | |
315 | kfree(sbi); | |
316 | } | |
317 | ||
318 | /** | |
319 | * nilfs_write_super - write super block(s) of NILFS | |
320 | * @sb: super_block | |
321 | * | |
322 | * nilfs_write_super() gets a fs-dependent lock, writes super block(s), and | |
323 | * clears s_dirt. This function is called in the section protected by | |
324 | * lock_super(). | |
325 | * | |
326 | * The s_dirt flag is managed by each filesystem and we protect it by ns_sem | |
327 | * of the struct the_nilfs. Lock order must be as follows: | |
328 | * | |
329 | * 1. lock_super() | |
330 | * 2. down_write(&nilfs->ns_sem) | |
331 | * | |
332 | * Inside NILFS, locking ns_sem is enough to protect s_dirt and the buffer | |
333 | * of the super block (nilfs->ns_sbp). | |
334 | * | |
335 | * In most cases, VFS functions call lock_super() before calling these | |
336 | * methods. So we must be careful not to bring on deadlocks when using | |
337 | * lock_super(); see generic_shutdown_super(), write_super(), and so on. | |
338 | * | |
339 | * Note that order of lock_kernel() and lock_super() depends on contexts | |
340 | * of VFS. We should also note that lock_kernel() can be used in its | |
341 | * protective section and only the outermost one has an effect. | |
342 | */ | |
343 | static void nilfs_write_super(struct super_block *sb) | |
344 | { | |
345 | struct nilfs_sb_info *sbi = NILFS_SB(sb); | |
346 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
347 | ||
348 | down_write(&nilfs->ns_sem); | |
349 | if (!(sb->s_flags & MS_RDONLY)) | |
350 | nilfs_commit_super(sbi); | |
351 | sb->s_dirt = 0; | |
352 | up_write(&nilfs->ns_sem); | |
353 | } | |
354 | ||
355 | static int nilfs_sync_fs(struct super_block *sb, int wait) | |
356 | { | |
357 | int err = 0; | |
358 | ||
359 | /* This function is called when super block should be written back */ | |
360 | if (wait) | |
361 | err = nilfs_construct_segment(sb); | |
362 | return err; | |
363 | } | |
364 | ||
365 | int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno) | |
366 | { | |
367 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
368 | struct nilfs_checkpoint *raw_cp; | |
369 | struct buffer_head *bh_cp; | |
370 | int err; | |
371 | ||
372 | down_write(&nilfs->ns_sem); | |
373 | list_add(&sbi->s_list, &nilfs->ns_supers); | |
374 | up_write(&nilfs->ns_sem); | |
375 | ||
376 | sbi->s_ifile = nilfs_mdt_new( | |
377 | nilfs, sbi->s_super, NILFS_IFILE_INO, NILFS_IFILE_GFP); | |
378 | if (!sbi->s_ifile) | |
379 | return -ENOMEM; | |
380 | ||
381 | err = nilfs_palloc_init_blockgroup(sbi->s_ifile, nilfs->ns_inode_size); | |
382 | if (unlikely(err)) | |
383 | goto failed; | |
384 | ||
385 | err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp, | |
386 | &bh_cp); | |
387 | if (unlikely(err)) { | |
388 | if (err == -ENOENT || err == -EINVAL) { | |
389 | printk(KERN_ERR | |
390 | "NILFS: Invalid checkpoint " | |
391 | "(checkpoint number=%llu)\n", | |
392 | (unsigned long long)cno); | |
393 | err = -EINVAL; | |
394 | } | |
395 | goto failed; | |
396 | } | |
397 | err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode); | |
398 | if (unlikely(err)) | |
399 | goto failed_bh; | |
400 | atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count)); | |
401 | atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count)); | |
402 | ||
403 | nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp); | |
404 | return 0; | |
405 | ||
406 | failed_bh: | |
407 | nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp); | |
408 | failed: | |
409 | nilfs_mdt_destroy(sbi->s_ifile); | |
410 | sbi->s_ifile = NULL; | |
411 | ||
412 | down_write(&nilfs->ns_sem); | |
413 | list_del_init(&sbi->s_list); | |
414 | up_write(&nilfs->ns_sem); | |
415 | ||
416 | return err; | |
417 | } | |
418 | ||
419 | void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi) | |
420 | { | |
421 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
422 | ||
423 | nilfs_mdt_clear(sbi->s_ifile); | |
424 | nilfs_mdt_destroy(sbi->s_ifile); | |
425 | sbi->s_ifile = NULL; | |
426 | down_write(&nilfs->ns_sem); | |
427 | list_del_init(&sbi->s_list); | |
428 | up_write(&nilfs->ns_sem); | |
429 | } | |
430 | ||
431 | static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi) | |
432 | { | |
433 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
434 | int err = 0; | |
435 | ||
436 | down_write(&nilfs->ns_sem); | |
437 | if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) { | |
438 | nilfs->ns_mount_state |= NILFS_VALID_FS; | |
439 | err = nilfs_commit_super(sbi); | |
440 | if (likely(!err)) | |
441 | printk(KERN_INFO "NILFS: recovery complete.\n"); | |
442 | } | |
443 | up_write(&nilfs->ns_sem); | |
444 | return err; | |
445 | } | |
446 | ||
447 | static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf) | |
448 | { | |
449 | struct super_block *sb = dentry->d_sb; | |
450 | struct nilfs_sb_info *sbi = NILFS_SB(sb); | |
451 | unsigned long long blocks; | |
452 | unsigned long overhead; | |
453 | unsigned long nrsvblocks; | |
454 | sector_t nfreeblocks; | |
455 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
456 | int err; | |
457 | ||
458 | /* | |
459 | * Compute all of the segment blocks | |
460 | * | |
461 | * The blocks before first segment and after last segment | |
462 | * are excluded. | |
463 | */ | |
464 | blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments | |
465 | - nilfs->ns_first_data_block; | |
466 | nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment; | |
467 | ||
468 | /* | |
469 | * Compute the overhead | |
470 | * | |
471 | * When distributing meta data blocks outside semgent structure, | |
472 | * We must count them as the overhead. | |
473 | */ | |
474 | overhead = 0; | |
475 | ||
476 | err = nilfs_count_free_blocks(nilfs, &nfreeblocks); | |
477 | if (unlikely(err)) | |
478 | return err; | |
479 | ||
480 | buf->f_type = NILFS_SUPER_MAGIC; | |
481 | buf->f_bsize = sb->s_blocksize; | |
482 | buf->f_blocks = blocks - overhead; | |
483 | buf->f_bfree = nfreeblocks; | |
484 | buf->f_bavail = (buf->f_bfree >= nrsvblocks) ? | |
485 | (buf->f_bfree - nrsvblocks) : 0; | |
486 | buf->f_files = atomic_read(&sbi->s_inodes_count); | |
487 | buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */ | |
488 | buf->f_namelen = NILFS_NAME_LEN; | |
489 | return 0; | |
490 | } | |
491 | ||
492 | static struct super_operations nilfs_sops = { | |
493 | .alloc_inode = nilfs_alloc_inode, | |
494 | .destroy_inode = nilfs_destroy_inode, | |
495 | .dirty_inode = nilfs_dirty_inode, | |
496 | /* .write_inode = nilfs_write_inode, */ | |
497 | /* .put_inode = nilfs_put_inode, */ | |
498 | /* .drop_inode = nilfs_drop_inode, */ | |
499 | .delete_inode = nilfs_delete_inode, | |
500 | .put_super = nilfs_put_super, | |
501 | .write_super = nilfs_write_super, | |
502 | .sync_fs = nilfs_sync_fs, | |
503 | /* .write_super_lockfs */ | |
504 | /* .unlockfs */ | |
505 | .statfs = nilfs_statfs, | |
506 | .remount_fs = nilfs_remount, | |
507 | .clear_inode = nilfs_clear_inode, | |
508 | /* .umount_begin */ | |
509 | /* .show_options */ | |
510 | }; | |
511 | ||
512 | static struct inode * | |
513 | nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation) | |
514 | { | |
515 | struct inode *inode; | |
516 | ||
517 | if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO && | |
518 | ino != NILFS_SKETCH_INO) | |
519 | return ERR_PTR(-ESTALE); | |
520 | ||
521 | inode = nilfs_iget(sb, ino); | |
522 | if (IS_ERR(inode)) | |
523 | return ERR_CAST(inode); | |
524 | if (generation && inode->i_generation != generation) { | |
525 | iput(inode); | |
526 | return ERR_PTR(-ESTALE); | |
527 | } | |
528 | ||
529 | return inode; | |
530 | } | |
531 | ||
532 | static struct dentry * | |
533 | nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len, | |
534 | int fh_type) | |
535 | { | |
536 | return generic_fh_to_dentry(sb, fid, fh_len, fh_type, | |
537 | nilfs_nfs_get_inode); | |
538 | } | |
539 | ||
540 | static struct dentry * | |
541 | nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len, | |
542 | int fh_type) | |
543 | { | |
544 | return generic_fh_to_parent(sb, fid, fh_len, fh_type, | |
545 | nilfs_nfs_get_inode); | |
546 | } | |
547 | ||
548 | static struct export_operations nilfs_export_ops = { | |
549 | .fh_to_dentry = nilfs_fh_to_dentry, | |
550 | .fh_to_parent = nilfs_fh_to_parent, | |
551 | .get_parent = nilfs_get_parent, | |
552 | }; | |
553 | ||
554 | enum { | |
555 | Opt_err_cont, Opt_err_panic, Opt_err_ro, | |
556 | Opt_barrier, Opt_snapshot, Opt_order, | |
557 | Opt_err, | |
558 | }; | |
559 | ||
560 | static match_table_t tokens = { | |
561 | {Opt_err_cont, "errors=continue"}, | |
562 | {Opt_err_panic, "errors=panic"}, | |
563 | {Opt_err_ro, "errors=remount-ro"}, | |
564 | {Opt_barrier, "barrier=%s"}, | |
565 | {Opt_snapshot, "cp=%u"}, | |
566 | {Opt_order, "order=%s"}, | |
567 | {Opt_err, NULL} | |
568 | }; | |
569 | ||
570 | static int match_bool(substring_t *s, int *result) | |
571 | { | |
572 | int len = s->to - s->from; | |
573 | ||
574 | if (strncmp(s->from, "on", len) == 0) | |
575 | *result = 1; | |
576 | else if (strncmp(s->from, "off", len) == 0) | |
577 | *result = 0; | |
578 | else | |
579 | return 1; | |
580 | return 0; | |
581 | } | |
582 | ||
583 | static int parse_options(char *options, struct super_block *sb) | |
584 | { | |
585 | struct nilfs_sb_info *sbi = NILFS_SB(sb); | |
586 | char *p; | |
587 | substring_t args[MAX_OPT_ARGS]; | |
588 | int option; | |
589 | ||
590 | if (!options) | |
591 | return 1; | |
592 | ||
593 | while ((p = strsep(&options, ",")) != NULL) { | |
594 | int token; | |
595 | if (!*p) | |
596 | continue; | |
597 | ||
598 | token = match_token(p, tokens, args); | |
599 | switch (token) { | |
600 | case Opt_barrier: | |
601 | if (match_bool(&args[0], &option)) | |
602 | return 0; | |
603 | if (option) | |
604 | nilfs_set_opt(sbi, BARRIER); | |
605 | else | |
606 | nilfs_clear_opt(sbi, BARRIER); | |
607 | break; | |
608 | case Opt_order: | |
609 | if (strcmp(args[0].from, "relaxed") == 0) | |
610 | /* Ordered data semantics */ | |
611 | nilfs_clear_opt(sbi, STRICT_ORDER); | |
612 | else if (strcmp(args[0].from, "strict") == 0) | |
613 | /* Strict in-order semantics */ | |
614 | nilfs_set_opt(sbi, STRICT_ORDER); | |
615 | else | |
616 | return 0; | |
617 | break; | |
618 | case Opt_err_panic: | |
619 | nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC); | |
620 | break; | |
621 | case Opt_err_ro: | |
622 | nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO); | |
623 | break; | |
624 | case Opt_err_cont: | |
625 | nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT); | |
626 | break; | |
627 | case Opt_snapshot: | |
628 | if (match_int(&args[0], &option) || option <= 0) | |
629 | return 0; | |
630 | if (!(sb->s_flags & MS_RDONLY)) | |
631 | return 0; | |
632 | sbi->s_snapshot_cno = option; | |
633 | nilfs_set_opt(sbi, SNAPSHOT); | |
634 | break; | |
635 | default: | |
636 | printk(KERN_ERR | |
637 | "NILFS: Unrecognized mount option \"%s\"\n", p); | |
638 | return 0; | |
639 | } | |
640 | } | |
641 | return 1; | |
642 | } | |
643 | ||
644 | static inline void | |
645 | nilfs_set_default_options(struct nilfs_sb_info *sbi, | |
646 | struct nilfs_super_block *sbp) | |
647 | { | |
648 | sbi->s_mount_opt = | |
649 | NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER; | |
650 | } | |
651 | ||
652 | static int nilfs_setup_super(struct nilfs_sb_info *sbi) | |
653 | { | |
654 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
655 | struct nilfs_super_block *sbp = nilfs->ns_sbp; | |
656 | int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count); | |
657 | int mnt_count = le16_to_cpu(sbp->s_mnt_count); | |
658 | ||
659 | /* nilfs->sem must be locked by the caller. */ | |
660 | if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) { | |
661 | printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n"); | |
662 | } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) { | |
663 | printk(KERN_WARNING | |
664 | "NILFS warning: mounting fs with errors\n"); | |
665 | #if 0 | |
666 | } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) { | |
667 | printk(KERN_WARNING | |
668 | "NILFS warning: maximal mount count reached\n"); | |
669 | #endif | |
670 | } | |
671 | if (!max_mnt_count) | |
672 | sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT); | |
673 | ||
674 | sbp->s_mnt_count = cpu_to_le16(mnt_count + 1); | |
675 | sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS); | |
676 | sbp->s_mtime = cpu_to_le64(get_seconds()); | |
677 | return nilfs_commit_super(sbi); | |
678 | } | |
679 | ||
680 | struct nilfs_super_block * | |
681 | nilfs_load_super_block(struct super_block *sb, struct buffer_head **pbh) | |
682 | { | |
683 | int blocksize; | |
684 | unsigned long offset, sb_index; | |
685 | ||
686 | /* | |
687 | * Adjusting block size | |
688 | * Blocksize will be enlarged when it is smaller than hardware | |
689 | * sector size. | |
690 | * Disk format of superblock does not change. | |
691 | */ | |
692 | blocksize = sb_min_blocksize(sb, BLOCK_SIZE); | |
693 | if (!blocksize) { | |
694 | printk(KERN_ERR | |
695 | "NILFS: unable to set blocksize of superblock\n"); | |
696 | return NULL; | |
697 | } | |
698 | sb_index = NILFS_SB_OFFSET_BYTES / blocksize; | |
699 | offset = NILFS_SB_OFFSET_BYTES % blocksize; | |
700 | ||
701 | *pbh = sb_bread(sb, sb_index); | |
702 | if (!*pbh) { | |
703 | printk(KERN_ERR "NILFS: unable to read superblock\n"); | |
704 | return NULL; | |
705 | } | |
706 | return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset); | |
707 | } | |
708 | ||
709 | struct nilfs_super_block * | |
710 | nilfs_reload_super_block(struct super_block *sb, struct buffer_head **pbh, | |
711 | int blocksize) | |
712 | { | |
713 | struct nilfs_super_block *sbp; | |
714 | unsigned long offset, sb_index; | |
715 | int hw_blocksize = bdev_hardsect_size(sb->s_bdev); | |
716 | ||
717 | if (blocksize < hw_blocksize) { | |
718 | printk(KERN_ERR | |
719 | "NILFS: blocksize %d too small for device " | |
720 | "(sector-size = %d).\n", | |
721 | blocksize, hw_blocksize); | |
722 | goto failed_sbh; | |
723 | } | |
724 | brelse(*pbh); | |
725 | sb_set_blocksize(sb, blocksize); | |
726 | ||
727 | sb_index = NILFS_SB_OFFSET_BYTES / blocksize; | |
728 | offset = NILFS_SB_OFFSET_BYTES % blocksize; | |
729 | ||
730 | *pbh = sb_bread(sb, sb_index); | |
731 | if (!*pbh) { | |
732 | printk(KERN_ERR | |
733 | "NILFS: cannot read superblock on 2nd try.\n"); | |
734 | goto failed; | |
735 | } | |
736 | ||
737 | sbp = (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset); | |
738 | if (sbp->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) { | |
739 | printk(KERN_ERR | |
740 | "NILFS: !? Magic mismatch on 2nd try.\n"); | |
741 | goto failed_sbh; | |
742 | } | |
743 | return sbp; | |
744 | ||
745 | failed_sbh: | |
746 | brelse(*pbh); | |
747 | ||
748 | failed: | |
749 | return NULL; | |
750 | } | |
751 | ||
752 | int nilfs_store_magic_and_option(struct super_block *sb, | |
753 | struct nilfs_super_block *sbp, | |
754 | char *data) | |
755 | { | |
756 | struct nilfs_sb_info *sbi = NILFS_SB(sb); | |
757 | ||
758 | /* trying to fill super (1st stage) */ | |
759 | sb->s_magic = le16_to_cpu(sbp->s_magic); | |
760 | ||
761 | /* FS independent flags */ | |
762 | #ifdef NILFS_ATIME_DISABLE | |
763 | sb->s_flags |= MS_NOATIME; | |
764 | #endif | |
765 | ||
766 | if (sb->s_magic != NILFS_SUPER_MAGIC) { | |
767 | printk("NILFS: Can't find nilfs on dev %s.\n", sb->s_id); | |
768 | return -EINVAL; | |
769 | } | |
770 | ||
771 | nilfs_set_default_options(sbi, sbp); | |
772 | ||
773 | sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid); | |
774 | sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid); | |
775 | sbi->s_interval = le32_to_cpu(sbp->s_c_interval); | |
776 | sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max); | |
777 | ||
778 | if (!parse_options(data, sb)) | |
779 | return -EINVAL; | |
780 | ||
781 | return 0; | |
782 | } | |
783 | ||
784 | /** | |
785 | * nilfs_fill_super() - initialize a super block instance | |
786 | * @sb: super_block | |
787 | * @data: mount options | |
788 | * @silent: silent mode flag | |
789 | * @nilfs: the_nilfs struct | |
790 | * | |
791 | * This function is called exclusively by bd_mount_mutex. | |
792 | * So, the recovery process is protected from other simultaneous mounts. | |
793 | */ | |
794 | static int | |
795 | nilfs_fill_super(struct super_block *sb, void *data, int silent, | |
796 | struct the_nilfs *nilfs) | |
797 | { | |
798 | struct nilfs_sb_info *sbi; | |
799 | struct inode *root; | |
800 | __u64 cno; | |
801 | int err; | |
802 | ||
803 | sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); | |
804 | if (!sbi) | |
805 | return -ENOMEM; | |
806 | ||
807 | sb->s_fs_info = sbi; | |
808 | ||
809 | get_nilfs(nilfs); | |
810 | sbi->s_nilfs = nilfs; | |
811 | sbi->s_super = sb; | |
812 | ||
813 | err = init_nilfs(nilfs, sbi, (char *)data); | |
814 | if (err) | |
815 | goto failed_sbi; | |
816 | ||
817 | spin_lock_init(&sbi->s_inode_lock); | |
818 | INIT_LIST_HEAD(&sbi->s_dirty_files); | |
819 | INIT_LIST_HEAD(&sbi->s_list); | |
820 | ||
821 | /* | |
822 | * Following initialization is overlapped because | |
823 | * nilfs_sb_info structure has been cleared at the beginning. | |
824 | * But we reserve them to keep our interest and make ready | |
825 | * for the future change. | |
826 | */ | |
827 | get_random_bytes(&sbi->s_next_generation, | |
828 | sizeof(sbi->s_next_generation)); | |
829 | spin_lock_init(&sbi->s_next_gen_lock); | |
830 | ||
831 | sb->s_op = &nilfs_sops; | |
832 | sb->s_export_op = &nilfs_export_ops; | |
833 | sb->s_root = NULL; | |
834 | ||
835 | if (!nilfs_loaded(nilfs)) { | |
836 | err = load_nilfs(nilfs, sbi); | |
837 | if (err) | |
838 | goto failed_sbi; | |
839 | } | |
840 | cno = nilfs_last_cno(nilfs); | |
841 | ||
842 | if (sb->s_flags & MS_RDONLY) { | |
843 | if (nilfs_test_opt(sbi, SNAPSHOT)) { | |
1f5abe7e RK |
844 | err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, |
845 | sbi->s_snapshot_cno); | |
846 | if (err < 0) | |
847 | goto failed_sbi; | |
848 | if (!err) { | |
783f6184 RK |
849 | printk(KERN_ERR |
850 | "NILFS: The specified checkpoint is " | |
851 | "not a snapshot " | |
852 | "(checkpoint number=%llu).\n", | |
853 | (unsigned long long)sbi->s_snapshot_cno); | |
854 | err = -EINVAL; | |
855 | goto failed_sbi; | |
856 | } | |
857 | cno = sbi->s_snapshot_cno; | |
858 | } else | |
859 | /* Read-only mount */ | |
860 | sbi->s_snapshot_cno = cno; | |
861 | } | |
862 | ||
863 | err = nilfs_attach_checkpoint(sbi, cno); | |
864 | if (err) { | |
865 | printk(KERN_ERR "NILFS: error loading a checkpoint" | |
866 | " (checkpoint number=%llu).\n", (unsigned long long)cno); | |
867 | goto failed_sbi; | |
868 | } | |
869 | ||
870 | if (!(sb->s_flags & MS_RDONLY)) { | |
cece5520 | 871 | err = nilfs_attach_segment_constructor(sbi); |
783f6184 RK |
872 | if (err) |
873 | goto failed_checkpoint; | |
874 | } | |
875 | ||
876 | root = nilfs_iget(sb, NILFS_ROOT_INO); | |
877 | if (IS_ERR(root)) { | |
878 | printk(KERN_ERR "NILFS: get root inode failed\n"); | |
879 | err = PTR_ERR(root); | |
880 | goto failed_segctor; | |
881 | } | |
882 | if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { | |
883 | iput(root); | |
884 | printk(KERN_ERR "NILFS: corrupt root inode.\n"); | |
885 | err = -EINVAL; | |
886 | goto failed_segctor; | |
887 | } | |
888 | sb->s_root = d_alloc_root(root); | |
889 | if (!sb->s_root) { | |
890 | iput(root); | |
891 | printk(KERN_ERR "NILFS: get root dentry failed\n"); | |
892 | err = -ENOMEM; | |
893 | goto failed_segctor; | |
894 | } | |
895 | ||
896 | if (!(sb->s_flags & MS_RDONLY)) { | |
897 | down_write(&nilfs->ns_sem); | |
898 | nilfs_setup_super(sbi); | |
899 | up_write(&nilfs->ns_sem); | |
900 | } | |
901 | ||
902 | err = nilfs_mark_recovery_complete(sbi); | |
903 | if (unlikely(err)) { | |
904 | printk(KERN_ERR "NILFS: recovery failed.\n"); | |
905 | goto failed_root; | |
906 | } | |
907 | ||
908 | return 0; | |
909 | ||
910 | failed_root: | |
911 | dput(sb->s_root); | |
912 | sb->s_root = NULL; | |
913 | ||
914 | failed_segctor: | |
915 | nilfs_detach_segment_constructor(sbi); | |
916 | ||
917 | failed_checkpoint: | |
918 | nilfs_detach_checkpoint(sbi); | |
919 | ||
920 | failed_sbi: | |
921 | put_nilfs(nilfs); | |
922 | sb->s_fs_info = NULL; | |
923 | kfree(sbi); | |
924 | return err; | |
925 | } | |
926 | ||
927 | static int nilfs_remount(struct super_block *sb, int *flags, char *data) | |
928 | { | |
929 | struct nilfs_sb_info *sbi = NILFS_SB(sb); | |
930 | struct nilfs_super_block *sbp; | |
931 | struct the_nilfs *nilfs = sbi->s_nilfs; | |
932 | unsigned long old_sb_flags; | |
933 | struct nilfs_mount_options old_opts; | |
934 | int err; | |
935 | ||
936 | old_sb_flags = sb->s_flags; | |
937 | old_opts.mount_opt = sbi->s_mount_opt; | |
938 | old_opts.snapshot_cno = sbi->s_snapshot_cno; | |
939 | ||
940 | if (!parse_options(data, sb)) { | |
941 | err = -EINVAL; | |
942 | goto restore_opts; | |
943 | } | |
944 | sb->s_flags = (sb->s_flags & ~MS_POSIXACL); | |
945 | ||
946 | if ((*flags & MS_RDONLY) && | |
947 | sbi->s_snapshot_cno != old_opts.snapshot_cno) { | |
948 | printk(KERN_WARNING "NILFS (device %s): couldn't " | |
949 | "remount to a different snapshot. \n", | |
950 | sb->s_id); | |
951 | err = -EINVAL; | |
952 | goto restore_opts; | |
953 | } | |
954 | ||
955 | if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY)) | |
956 | goto out; | |
957 | if (*flags & MS_RDONLY) { | |
958 | /* Shutting down the segment constructor */ | |
959 | nilfs_detach_segment_constructor(sbi); | |
960 | sb->s_flags |= MS_RDONLY; | |
961 | ||
962 | sbi->s_snapshot_cno = nilfs_last_cno(nilfs); | |
963 | /* nilfs_set_opt(sbi, SNAPSHOT); */ | |
964 | ||
965 | /* | |
966 | * Remounting a valid RW partition RDONLY, so set | |
967 | * the RDONLY flag and then mark the partition as valid again. | |
968 | */ | |
969 | down_write(&nilfs->ns_sem); | |
970 | sbp = nilfs->ns_sbp; | |
971 | if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) && | |
972 | (nilfs->ns_mount_state & NILFS_VALID_FS)) | |
973 | sbp->s_state = cpu_to_le16(nilfs->ns_mount_state); | |
974 | sbp->s_mtime = cpu_to_le64(get_seconds()); | |
975 | nilfs_commit_super(sbi); | |
976 | up_write(&nilfs->ns_sem); | |
977 | } else { | |
978 | /* | |
979 | * Mounting a RDONLY partition read-write, so reread and | |
980 | * store the current valid flag. (It may have been changed | |
981 | * by fsck since we originally mounted the partition.) | |
982 | */ | |
983 | down(&sb->s_bdev->bd_mount_sem); | |
984 | /* Check existing RW-mount */ | |
985 | if (test_exclusive_mount(sb->s_type, sb->s_bdev, 0)) { | |
986 | printk(KERN_WARNING "NILFS (device %s): couldn't " | |
987 | "remount because a RW-mount exists.\n", | |
988 | sb->s_id); | |
989 | err = -EBUSY; | |
990 | goto rw_remount_failed; | |
991 | } | |
992 | if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) { | |
993 | printk(KERN_WARNING "NILFS (device %s): couldn't " | |
994 | "remount because the current RO-mount is not " | |
995 | "the latest one.\n", | |
996 | sb->s_id); | |
997 | err = -EINVAL; | |
998 | goto rw_remount_failed; | |
999 | } | |
1000 | sb->s_flags &= ~MS_RDONLY; | |
1001 | nilfs_clear_opt(sbi, SNAPSHOT); | |
1002 | sbi->s_snapshot_cno = 0; | |
1003 | ||
cece5520 | 1004 | err = nilfs_attach_segment_constructor(sbi); |
783f6184 RK |
1005 | if (err) |
1006 | goto rw_remount_failed; | |
1007 | ||
1008 | down_write(&nilfs->ns_sem); | |
1009 | nilfs_setup_super(sbi); | |
1010 | up_write(&nilfs->ns_sem); | |
1011 | ||
1012 | up(&sb->s_bdev->bd_mount_sem); | |
1013 | } | |
1014 | out: | |
1015 | return 0; | |
1016 | ||
1017 | rw_remount_failed: | |
1018 | up(&sb->s_bdev->bd_mount_sem); | |
1019 | restore_opts: | |
1020 | sb->s_flags = old_sb_flags; | |
1021 | sbi->s_mount_opt = old_opts.mount_opt; | |
1022 | sbi->s_snapshot_cno = old_opts.snapshot_cno; | |
1023 | return err; | |
1024 | } | |
1025 | ||
1026 | struct nilfs_super_data { | |
1027 | struct block_device *bdev; | |
1028 | __u64 cno; | |
1029 | int flags; | |
1030 | }; | |
1031 | ||
1032 | /** | |
1033 | * nilfs_identify - pre-read mount options needed to identify mount instance | |
1034 | * @data: mount options | |
1035 | * @sd: nilfs_super_data | |
1036 | */ | |
1037 | static int nilfs_identify(char *data, struct nilfs_super_data *sd) | |
1038 | { | |
1039 | char *p, *options = data; | |
1040 | substring_t args[MAX_OPT_ARGS]; | |
1041 | int option, token; | |
1042 | int ret = 0; | |
1043 | ||
1044 | do { | |
1045 | p = strsep(&options, ","); | |
1046 | if (p != NULL && *p) { | |
1047 | token = match_token(p, tokens, args); | |
1048 | if (token == Opt_snapshot) { | |
1049 | if (!(sd->flags & MS_RDONLY)) | |
1050 | ret++; | |
1051 | else { | |
1052 | ret = match_int(&args[0], &option); | |
1053 | if (!ret) { | |
1054 | if (option > 0) | |
1055 | sd->cno = option; | |
1056 | else | |
1057 | ret++; | |
1058 | } | |
1059 | } | |
1060 | } | |
1061 | if (ret) | |
1062 | printk(KERN_ERR | |
1063 | "NILFS: invalid mount option: %s\n", p); | |
1064 | } | |
1065 | if (!options) | |
1066 | break; | |
1067 | BUG_ON(options == data); | |
1068 | *(options - 1) = ','; | |
1069 | } while (!ret); | |
1070 | return ret; | |
1071 | } | |
1072 | ||
1073 | static int nilfs_set_bdev_super(struct super_block *s, void *data) | |
1074 | { | |
1075 | struct nilfs_super_data *sd = data; | |
1076 | ||
1077 | s->s_bdev = sd->bdev; | |
1078 | s->s_dev = s->s_bdev->bd_dev; | |
1079 | return 0; | |
1080 | } | |
1081 | ||
1082 | static int nilfs_test_bdev_super(struct super_block *s, void *data) | |
1083 | { | |
1084 | struct nilfs_super_data *sd = data; | |
1085 | ||
1086 | return s->s_bdev == sd->bdev; | |
1087 | } | |
1088 | ||
1089 | static int nilfs_test_bdev_super2(struct super_block *s, void *data) | |
1090 | { | |
1091 | struct nilfs_super_data *sd = data; | |
1092 | int ret; | |
1093 | ||
1094 | if (s->s_bdev != sd->bdev) | |
1095 | return 0; | |
1096 | ||
1097 | if (!((s->s_flags | sd->flags) & MS_RDONLY)) | |
1098 | return 1; /* Reuse an old R/W-mode super_block */ | |
1099 | ||
1100 | if (s->s_flags & sd->flags & MS_RDONLY) { | |
1101 | if (down_read_trylock(&s->s_umount)) { | |
1102 | ret = s->s_root && | |
1103 | (sd->cno == NILFS_SB(s)->s_snapshot_cno); | |
1104 | up_read(&s->s_umount); | |
1105 | /* | |
1106 | * This path is locked with sb_lock by sget(). | |
1107 | * So, drop_super() causes deadlock. | |
1108 | */ | |
1109 | return ret; | |
1110 | } | |
1111 | } | |
1112 | return 0; | |
1113 | } | |
1114 | ||
1115 | static int | |
1116 | nilfs_get_sb(struct file_system_type *fs_type, int flags, | |
1117 | const char *dev_name, void *data, struct vfsmount *mnt) | |
1118 | { | |
1119 | struct nilfs_super_data sd; | |
1120 | struct super_block *s, *s2; | |
1121 | struct the_nilfs *nilfs = NULL; | |
1122 | int err, need_to_close = 1; | |
1123 | ||
1124 | sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type); | |
1125 | if (IS_ERR(sd.bdev)) | |
1126 | return PTR_ERR(sd.bdev); | |
1127 | ||
1128 | /* | |
1129 | * To get mount instance using sget() vfs-routine, NILFS needs | |
1130 | * much more information than normal filesystems to identify mount | |
1131 | * instance. For snapshot mounts, not only a mount type (ro-mount | |
1132 | * or rw-mount) but also a checkpoint number is required. | |
1133 | * The results are passed in sget() using nilfs_super_data. | |
1134 | */ | |
1135 | sd.cno = 0; | |
1136 | sd.flags = flags; | |
1137 | if (nilfs_identify((char *)data, &sd)) { | |
1138 | err = -EINVAL; | |
1139 | goto failed; | |
1140 | } | |
1141 | ||
1142 | /* | |
1143 | * once the super is inserted into the list by sget, s_umount | |
1144 | * will protect the lockfs code from trying to start a snapshot | |
1145 | * while we are mounting | |
1146 | */ | |
1147 | down(&sd.bdev->bd_mount_sem); | |
1148 | if (!sd.cno && | |
1149 | (err = test_exclusive_mount(fs_type, sd.bdev, flags ^ MS_RDONLY))) { | |
1150 | err = (err < 0) ? : -EBUSY; | |
1151 | goto failed_unlock; | |
1152 | } | |
1153 | ||
1154 | /* | |
1155 | * Phase-1: search any existent instance and get the_nilfs | |
1156 | */ | |
1157 | s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd); | |
1158 | if (IS_ERR(s)) | |
1159 | goto error_s; | |
1160 | ||
1161 | if (!s->s_root) { | |
1162 | err = -ENOMEM; | |
1163 | nilfs = alloc_nilfs(sd.bdev); | |
1164 | if (!nilfs) | |
1165 | goto cancel_new; | |
1166 | } else { | |
1167 | struct nilfs_sb_info *sbi = NILFS_SB(s); | |
1168 | ||
783f6184 RK |
1169 | /* |
1170 | * s_umount protects super_block from unmount process; | |
1171 | * It covers pointers of nilfs_sb_info and the_nilfs. | |
1172 | */ | |
1173 | nilfs = sbi->s_nilfs; | |
1174 | get_nilfs(nilfs); | |
1175 | up_write(&s->s_umount); | |
1176 | ||
1177 | /* | |
1178 | * Phase-2: search specified snapshot or R/W mode super_block | |
1179 | */ | |
1180 | if (!sd.cno) | |
1181 | /* trying to get the latest checkpoint. */ | |
1182 | sd.cno = nilfs_last_cno(nilfs); | |
1183 | ||
1184 | s2 = sget(fs_type, nilfs_test_bdev_super2, | |
1185 | nilfs_set_bdev_super, &sd); | |
1186 | deactivate_super(s); | |
1187 | /* | |
1188 | * Although deactivate_super() invokes close_bdev_exclusive() at | |
1189 | * kill_block_super(). Here, s is an existent mount; we need | |
1190 | * one more close_bdev_exclusive() call. | |
1191 | */ | |
1192 | s = s2; | |
1193 | if (IS_ERR(s)) | |
1194 | goto error_s; | |
1195 | } | |
1196 | ||
1197 | if (!s->s_root) { | |
1198 | char b[BDEVNAME_SIZE]; | |
1199 | ||
1200 | s->s_flags = flags; | |
1201 | strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id)); | |
1202 | sb_set_blocksize(s, block_size(sd.bdev)); | |
1203 | ||
1204 | err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs); | |
1205 | if (err) | |
1206 | goto cancel_new; | |
1207 | ||
1208 | s->s_flags |= MS_ACTIVE; | |
1209 | need_to_close = 0; | |
1210 | } else if (!(s->s_flags & MS_RDONLY)) { | |
1211 | err = -EBUSY; | |
1212 | } | |
1213 | ||
1214 | up(&sd.bdev->bd_mount_sem); | |
1215 | put_nilfs(nilfs); | |
1216 | if (need_to_close) | |
1217 | close_bdev_exclusive(sd.bdev, flags); | |
1218 | simple_set_mnt(mnt, s); | |
1219 | return 0; | |
1220 | ||
1221 | error_s: | |
1222 | up(&sd.bdev->bd_mount_sem); | |
1223 | if (nilfs) | |
1224 | put_nilfs(nilfs); | |
1225 | close_bdev_exclusive(sd.bdev, flags); | |
1226 | return PTR_ERR(s); | |
1227 | ||
1228 | failed_unlock: | |
1229 | up(&sd.bdev->bd_mount_sem); | |
1230 | failed: | |
1231 | close_bdev_exclusive(sd.bdev, flags); | |
1232 | ||
1233 | return err; | |
1234 | ||
1235 | cancel_new: | |
1236 | /* Abandoning the newly allocated superblock */ | |
1237 | up(&sd.bdev->bd_mount_sem); | |
1238 | if (nilfs) | |
1239 | put_nilfs(nilfs); | |
1240 | up_write(&s->s_umount); | |
1241 | deactivate_super(s); | |
1242 | /* | |
1243 | * deactivate_super() invokes close_bdev_exclusive(). | |
1244 | * We must finish all post-cleaning before this call; | |
1245 | * put_nilfs() and unlocking bd_mount_sem need the block device. | |
1246 | */ | |
1247 | return err; | |
1248 | } | |
1249 | ||
1250 | static int nilfs_test_bdev_super3(struct super_block *s, void *data) | |
1251 | { | |
1252 | struct nilfs_super_data *sd = data; | |
1253 | int ret; | |
1254 | ||
1255 | if (s->s_bdev != sd->bdev) | |
1256 | return 0; | |
1257 | if (down_read_trylock(&s->s_umount)) { | |
1258 | ret = (s->s_flags & MS_RDONLY) && s->s_root && | |
1259 | nilfs_test_opt(NILFS_SB(s), SNAPSHOT); | |
1260 | up_read(&s->s_umount); | |
1261 | if (ret) | |
1262 | return 0; /* ignore snapshot mounts */ | |
1263 | } | |
1264 | return !((sd->flags ^ s->s_flags) & MS_RDONLY); | |
1265 | } | |
1266 | ||
1267 | static int __false_bdev_super(struct super_block *s, void *data) | |
1268 | { | |
1269 | #if 0 /* XXX: workaround for lock debug. This is not good idea */ | |
1270 | up_write(&s->s_umount); | |
1271 | #endif | |
1272 | return -EFAULT; | |
1273 | } | |
1274 | ||
1275 | /** | |
1276 | * test_exclusive_mount - check whether an exclusive RW/RO mount exists or not. | |
1277 | * fs_type: filesystem type | |
1278 | * bdev: block device | |
1279 | * flag: 0 (check rw-mount) or MS_RDONLY (check ro-mount) | |
1280 | * res: pointer to an integer to store result | |
1281 | * | |
1282 | * This function must be called within a section protected by bd_mount_mutex. | |
1283 | */ | |
1284 | static int test_exclusive_mount(struct file_system_type *fs_type, | |
1285 | struct block_device *bdev, int flags) | |
1286 | { | |
1287 | struct super_block *s; | |
1288 | struct nilfs_super_data sd = { .flags = flags, .bdev = bdev }; | |
1289 | ||
1290 | s = sget(fs_type, nilfs_test_bdev_super3, __false_bdev_super, &sd); | |
1291 | if (IS_ERR(s)) { | |
1292 | if (PTR_ERR(s) != -EFAULT) | |
1293 | return PTR_ERR(s); | |
1294 | return 0; /* Not found */ | |
1295 | } | |
1296 | up_write(&s->s_umount); | |
1297 | deactivate_super(s); | |
1298 | return 1; /* Found */ | |
1299 | } | |
1300 | ||
1301 | struct file_system_type nilfs_fs_type = { | |
1302 | .owner = THIS_MODULE, | |
1303 | .name = "nilfs2", | |
1304 | .get_sb = nilfs_get_sb, | |
1305 | .kill_sb = kill_block_super, | |
1306 | .fs_flags = FS_REQUIRES_DEV, | |
1307 | }; | |
1308 | ||
1309 | static int __init init_nilfs_fs(void) | |
1310 | { | |
1311 | int err; | |
1312 | ||
1313 | err = nilfs_init_inode_cache(); | |
1314 | if (err) | |
1315 | goto failed; | |
1316 | ||
1317 | err = nilfs_init_transaction_cache(); | |
1318 | if (err) | |
1319 | goto failed_inode_cache; | |
1320 | ||
1321 | err = nilfs_init_segbuf_cache(); | |
1322 | if (err) | |
1323 | goto failed_transaction_cache; | |
1324 | ||
1325 | err = nilfs_btree_path_cache_init(); | |
1326 | if (err) | |
1327 | goto failed_segbuf_cache; | |
1328 | ||
1329 | err = register_filesystem(&nilfs_fs_type); | |
1330 | if (err) | |
1331 | goto failed_btree_path_cache; | |
1332 | ||
1333 | return 0; | |
1334 | ||
1335 | failed_btree_path_cache: | |
1336 | nilfs_btree_path_cache_destroy(); | |
1337 | ||
1338 | failed_segbuf_cache: | |
1339 | nilfs_destroy_segbuf_cache(); | |
1340 | ||
1341 | failed_transaction_cache: | |
1342 | nilfs_destroy_transaction_cache(); | |
1343 | ||
1344 | failed_inode_cache: | |
1345 | nilfs_destroy_inode_cache(); | |
1346 | ||
1347 | failed: | |
1348 | return err; | |
1349 | } | |
1350 | ||
1351 | static void __exit exit_nilfs_fs(void) | |
1352 | { | |
1353 | nilfs_destroy_segbuf_cache(); | |
1354 | nilfs_destroy_transaction_cache(); | |
1355 | nilfs_destroy_inode_cache(); | |
1356 | nilfs_btree_path_cache_destroy(); | |
1357 | unregister_filesystem(&nilfs_fs_type); | |
1358 | } | |
1359 | ||
1360 | module_init(init_nilfs_fs) | |
1361 | module_exit(exit_nilfs_fs) |