nilfs2: get rid of sget use for acquiring nilfs object
[linux-block.git] / fs / nilfs2 / super.c
CommitLineData
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
63MODULE_AUTHOR("NTT Corp.");
64MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
65 "(NILFS)");
783f6184
RK
66MODULE_LICENSE("GPL");
67
8c85e125 68static void nilfs_write_super(struct super_block *sb);
783f6184
RK
69static int nilfs_remount(struct super_block *sb, int *flags, char *data);
70static 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 */
85void 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;
e339ad31
RK
106 nilfs->ns_sbp[0]->s_state |=
107 cpu_to_le16(NILFS_ERROR_FS);
108 nilfs_commit_super(sbi, 1);
783f6184
RK
109 }
110 up_write(&nilfs->ns_sem);
111
112 if (nilfs_test_opt(sbi, ERRORS_RO)) {
113 printk(KERN_CRIT "Remounting filesystem read-only\n");
114 sb->s_flags |= MS_RDONLY;
115 }
116 }
117
118 if (nilfs_test_opt(sbi, ERRORS_PANIC))
119 panic("NILFS (device %s): panic forced after error\n",
120 sb->s_id);
121}
122
123void nilfs_warning(struct super_block *sb, const char *function,
124 const char *fmt, ...)
125{
126 va_list args;
127
128 va_start(args, fmt);
129 printk(KERN_WARNING "NILFS warning (device %s): %s: ",
130 sb->s_id, function);
131 vprintk(fmt, args);
132 printk("\n");
133 va_end(args);
134}
135
136static struct kmem_cache *nilfs_inode_cachep;
137
138struct inode *nilfs_alloc_inode(struct super_block *sb)
139{
140 struct nilfs_inode_info *ii;
141
142 ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
143 if (!ii)
144 return NULL;
145 ii->i_bh = NULL;
146 ii->i_state = 0;
147 ii->vfs_inode.i_version = 1;
148 nilfs_btnode_cache_init(&ii->i_btnode_cache);
149 return &ii->vfs_inode;
150}
151
152void nilfs_destroy_inode(struct inode *inode)
153{
154 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
155}
156
157static void init_once(void *obj)
158{
159 struct nilfs_inode_info *ii = obj;
160
161 INIT_LIST_HEAD(&ii->i_dirty);
162#ifdef CONFIG_NILFS_XATTR
163 init_rwsem(&ii->xattr_sem);
164#endif
165 nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
166 ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union;
167 inode_init_once(&ii->vfs_inode);
168}
169
170static int nilfs_init_inode_cache(void)
171{
172 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
173 sizeof(struct nilfs_inode_info),
174 0, SLAB_RECLAIM_ACCOUNT,
175 init_once);
176
177 return (nilfs_inode_cachep == NULL) ? -ENOMEM : 0;
178}
179
180static inline void nilfs_destroy_inode_cache(void)
181{
182 kmem_cache_destroy(nilfs_inode_cachep);
183}
184
185static void nilfs_clear_inode(struct inode *inode)
186{
187 struct nilfs_inode_info *ii = NILFS_I(inode);
783f6184
RK
188
189#ifdef CONFIG_NILFS_POSIX_ACL
190 if (ii->i_acl && ii->i_acl != NILFS_ACL_NOT_CACHED) {
191 posix_acl_release(ii->i_acl);
192 ii->i_acl = NILFS_ACL_NOT_CACHED;
193 }
194 if (ii->i_default_acl && ii->i_default_acl != NILFS_ACL_NOT_CACHED) {
195 posix_acl_release(ii->i_default_acl);
196 ii->i_default_acl = NILFS_ACL_NOT_CACHED;
197 }
198#endif
199 /*
200 * Free resources allocated in nilfs_read_inode(), here.
201 */
a2e7d2df 202 BUG_ON(!list_empty(&ii->i_dirty));
783f6184
RK
203 brelse(ii->i_bh);
204 ii->i_bh = NULL;
783f6184
RK
205
206 if (test_bit(NILFS_I_BMAP, &ii->i_state))
207 nilfs_bmap_clear(ii->i_bmap);
208
209 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
783f6184
RK
210}
211
e339ad31 212static int nilfs_sync_super(struct nilfs_sb_info *sbi, int dupsb)
783f6184
RK
213{
214 struct the_nilfs *nilfs = sbi->s_nilfs;
215 int err;
216 int barrier_done = 0;
217
218 if (nilfs_test_opt(sbi, BARRIER)) {
e339ad31 219 set_buffer_ordered(nilfs->ns_sbh[0]);
783f6184
RK
220 barrier_done = 1;
221 }
222 retry:
e339ad31
RK
223 set_buffer_dirty(nilfs->ns_sbh[0]);
224 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
783f6184
RK
225 if (err == -EOPNOTSUPP && barrier_done) {
226 nilfs_warning(sbi->s_super, __func__,
227 "barrier-based sync failed. "
228 "disabling barriers\n");
229 nilfs_clear_opt(sbi, BARRIER);
230 barrier_done = 0;
e339ad31 231 clear_buffer_ordered(nilfs->ns_sbh[0]);
783f6184
RK
232 goto retry;
233 }
e339ad31 234 if (unlikely(err)) {
783f6184
RK
235 printk(KERN_ERR
236 "NILFS: unable to write superblock (err=%d)\n", err);
e339ad31
RK
237 if (err == -EIO && nilfs->ns_sbh[1]) {
238 nilfs_fall_back_super_block(nilfs);
239 goto retry;
240 }
241 } else {
242 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
243
244 /*
245 * The latest segment becomes trailable from the position
246 * written in superblock.
247 */
783f6184 248 clear_nilfs_discontinued(nilfs);
e339ad31
RK
249
250 /* update GC protection for recent segments */
251 if (nilfs->ns_sbh[1]) {
252 sbp = NULL;
253 if (dupsb) {
254 set_buffer_dirty(nilfs->ns_sbh[1]);
255 if (!sync_dirty_buffer(nilfs->ns_sbh[1]))
256 sbp = nilfs->ns_sbp[1];
257 }
258 }
259 if (sbp) {
260 spin_lock(&nilfs->ns_last_segment_lock);
261 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
262 spin_unlock(&nilfs->ns_last_segment_lock);
263 }
783f6184
RK
264 }
265
266 return err;
267}
268
e339ad31 269int nilfs_commit_super(struct nilfs_sb_info *sbi, int dupsb)
783f6184
RK
270{
271 struct the_nilfs *nilfs = sbi->s_nilfs;
e339ad31 272 struct nilfs_super_block **sbp = nilfs->ns_sbp;
783f6184 273 sector_t nfreeblocks;
e339ad31 274 time_t t;
783f6184
RK
275 int err;
276
277 /* nilfs->sem must be locked by the caller. */
e339ad31
RK
278 if (sbp[0]->s_magic != NILFS_SUPER_MAGIC) {
279 if (sbp[1] && sbp[1]->s_magic == NILFS_SUPER_MAGIC)
280 nilfs_swap_super_block(nilfs);
281 else {
282 printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
283 sbi->s_super->s_id);
284 return -EIO;
285 }
286 }
783f6184
RK
287 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
288 if (unlikely(err)) {
289 printk(KERN_ERR "NILFS: failed to count free blocks\n");
290 return err;
291 }
e339ad31
RK
292 spin_lock(&nilfs->ns_last_segment_lock);
293 sbp[0]->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
294 sbp[0]->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
295 sbp[0]->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
296 spin_unlock(&nilfs->ns_last_segment_lock);
297
298 t = get_seconds();
299 nilfs->ns_sbwtime[0] = t;
300 sbp[0]->s_free_blocks_count = cpu_to_le64(nfreeblocks);
301 sbp[0]->s_wtime = cpu_to_le64(t);
302 sbp[0]->s_sum = 0;
303 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
304 (unsigned char *)sbp[0],
305 nilfs->ns_sbsize));
306 if (dupsb && sbp[1]) {
307 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
308 nilfs->ns_sbwtime[1] = t;
309 }
783f6184 310 sbi->s_super->s_dirt = 0;
e339ad31 311 return nilfs_sync_super(sbi, dupsb);
783f6184
RK
312}
313
314static void nilfs_put_super(struct super_block *sb)
315{
316 struct nilfs_sb_info *sbi = NILFS_SB(sb);
317 struct the_nilfs *nilfs = sbi->s_nilfs;
318
6cfd0148
CH
319 lock_kernel();
320
8c85e125
CH
321 if (sb->s_dirt)
322 nilfs_write_super(sb);
323
783f6184
RK
324 nilfs_detach_segment_constructor(sbi);
325
326 if (!(sb->s_flags & MS_RDONLY)) {
327 down_write(&nilfs->ns_sem);
e339ad31
RK
328 nilfs->ns_sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
329 nilfs_commit_super(sbi, 1);
783f6184
RK
330 up_write(&nilfs->ns_sem);
331 }
332
333 nilfs_detach_checkpoint(sbi);
334 put_nilfs(sbi->s_nilfs);
335 sbi->s_super = NULL;
336 sb->s_fs_info = NULL;
337 kfree(sbi);
6cfd0148
CH
338
339 unlock_kernel();
783f6184
RK
340}
341
342/**
343 * nilfs_write_super - write super block(s) of NILFS
344 * @sb: super_block
345 *
346 * nilfs_write_super() gets a fs-dependent lock, writes super block(s), and
347 * clears s_dirt. This function is called in the section protected by
348 * lock_super().
349 *
350 * The s_dirt flag is managed by each filesystem and we protect it by ns_sem
351 * of the struct the_nilfs. Lock order must be as follows:
352 *
353 * 1. lock_super()
354 * 2. down_write(&nilfs->ns_sem)
355 *
356 * Inside NILFS, locking ns_sem is enough to protect s_dirt and the buffer
e339ad31 357 * of the super block (nilfs->ns_sbp[]).
783f6184
RK
358 *
359 * In most cases, VFS functions call lock_super() before calling these
360 * methods. So we must be careful not to bring on deadlocks when using
361 * lock_super(); see generic_shutdown_super(), write_super(), and so on.
362 *
363 * Note that order of lock_kernel() and lock_super() depends on contexts
364 * of VFS. We should also note that lock_kernel() can be used in its
365 * protective section and only the outermost one has an effect.
366 */
367static void nilfs_write_super(struct super_block *sb)
368{
369 struct nilfs_sb_info *sbi = NILFS_SB(sb);
370 struct the_nilfs *nilfs = sbi->s_nilfs;
371
372 down_write(&nilfs->ns_sem);
e339ad31
RK
373 if (!(sb->s_flags & MS_RDONLY)) {
374 struct nilfs_super_block **sbp = nilfs->ns_sbp;
375 u64 t = get_seconds();
376 int dupsb;
377
378 if (!nilfs_discontinued(nilfs) && t >= nilfs->ns_sbwtime[0] &&
379 t < nilfs->ns_sbwtime[0] + NILFS_SB_FREQ) {
380 up_write(&nilfs->ns_sem);
381 return;
382 }
383 dupsb = sbp[1] && t > nilfs->ns_sbwtime[1] + NILFS_ALTSB_FREQ;
384 nilfs_commit_super(sbi, dupsb);
385 }
783f6184
RK
386 sb->s_dirt = 0;
387 up_write(&nilfs->ns_sem);
388}
389
390static int nilfs_sync_fs(struct super_block *sb, int wait)
391{
392 int err = 0;
393
d731e063
CH
394 nilfs_write_super(sb);
395
783f6184
RK
396 /* This function is called when super block should be written back */
397 if (wait)
398 err = nilfs_construct_segment(sb);
399 return err;
400}
401
402int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
403{
404 struct the_nilfs *nilfs = sbi->s_nilfs;
405 struct nilfs_checkpoint *raw_cp;
406 struct buffer_head *bh_cp;
407 int err;
408
409 down_write(&nilfs->ns_sem);
410 list_add(&sbi->s_list, &nilfs->ns_supers);
411 up_write(&nilfs->ns_sem);
412
413 sbi->s_ifile = nilfs_mdt_new(
414 nilfs, sbi->s_super, NILFS_IFILE_INO, NILFS_IFILE_GFP);
415 if (!sbi->s_ifile)
416 return -ENOMEM;
417
418 err = nilfs_palloc_init_blockgroup(sbi->s_ifile, nilfs->ns_inode_size);
419 if (unlikely(err))
420 goto failed;
421
422 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
423 &bh_cp);
424 if (unlikely(err)) {
425 if (err == -ENOENT || err == -EINVAL) {
426 printk(KERN_ERR
427 "NILFS: Invalid checkpoint "
428 "(checkpoint number=%llu)\n",
429 (unsigned long long)cno);
430 err = -EINVAL;
431 }
432 goto failed;
433 }
434 err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
435 if (unlikely(err))
436 goto failed_bh;
437 atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
438 atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
439
440 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
441 return 0;
442
443 failed_bh:
444 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
445 failed:
446 nilfs_mdt_destroy(sbi->s_ifile);
447 sbi->s_ifile = NULL;
448
449 down_write(&nilfs->ns_sem);
450 list_del_init(&sbi->s_list);
451 up_write(&nilfs->ns_sem);
452
453 return err;
454}
455
456void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
457{
458 struct the_nilfs *nilfs = sbi->s_nilfs;
459
460 nilfs_mdt_clear(sbi->s_ifile);
461 nilfs_mdt_destroy(sbi->s_ifile);
462 sbi->s_ifile = NULL;
463 down_write(&nilfs->ns_sem);
464 list_del_init(&sbi->s_list);
465 up_write(&nilfs->ns_sem);
466}
467
468static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi)
469{
470 struct the_nilfs *nilfs = sbi->s_nilfs;
471 int err = 0;
472
473 down_write(&nilfs->ns_sem);
474 if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
475 nilfs->ns_mount_state |= NILFS_VALID_FS;
e339ad31 476 err = nilfs_commit_super(sbi, 1);
783f6184
RK
477 if (likely(!err))
478 printk(KERN_INFO "NILFS: recovery complete.\n");
479 }
480 up_write(&nilfs->ns_sem);
481 return err;
482}
483
484static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
485{
486 struct super_block *sb = dentry->d_sb;
487 struct nilfs_sb_info *sbi = NILFS_SB(sb);
c306af23
RK
488 struct the_nilfs *nilfs = sbi->s_nilfs;
489 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
783f6184
RK
490 unsigned long long blocks;
491 unsigned long overhead;
492 unsigned long nrsvblocks;
493 sector_t nfreeblocks;
783f6184
RK
494 int err;
495
496 /*
497 * Compute all of the segment blocks
498 *
499 * The blocks before first segment and after last segment
500 * are excluded.
501 */
502 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
503 - nilfs->ns_first_data_block;
504 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
505
506 /*
507 * Compute the overhead
508 *
509 * When distributing meta data blocks outside semgent structure,
510 * We must count them as the overhead.
511 */
512 overhead = 0;
513
514 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
515 if (unlikely(err))
516 return err;
517
518 buf->f_type = NILFS_SUPER_MAGIC;
519 buf->f_bsize = sb->s_blocksize;
520 buf->f_blocks = blocks - overhead;
521 buf->f_bfree = nfreeblocks;
522 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
523 (buf->f_bfree - nrsvblocks) : 0;
524 buf->f_files = atomic_read(&sbi->s_inodes_count);
525 buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
526 buf->f_namelen = NILFS_NAME_LEN;
c306af23
RK
527 buf->f_fsid.val[0] = (u32)id;
528 buf->f_fsid.val[1] = (u32)(id >> 32);
529
783f6184
RK
530 return 0;
531}
532
533static struct super_operations nilfs_sops = {
534 .alloc_inode = nilfs_alloc_inode,
535 .destroy_inode = nilfs_destroy_inode,
536 .dirty_inode = nilfs_dirty_inode,
537 /* .write_inode = nilfs_write_inode, */
538 /* .put_inode = nilfs_put_inode, */
539 /* .drop_inode = nilfs_drop_inode, */
540 .delete_inode = nilfs_delete_inode,
541 .put_super = nilfs_put_super,
542 .write_super = nilfs_write_super,
543 .sync_fs = nilfs_sync_fs,
544 /* .write_super_lockfs */
545 /* .unlockfs */
546 .statfs = nilfs_statfs,
547 .remount_fs = nilfs_remount,
548 .clear_inode = nilfs_clear_inode,
549 /* .umount_begin */
550 /* .show_options */
551};
552
553static struct inode *
554nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
555{
556 struct inode *inode;
557
558 if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
559 ino != NILFS_SKETCH_INO)
560 return ERR_PTR(-ESTALE);
561
562 inode = nilfs_iget(sb, ino);
563 if (IS_ERR(inode))
564 return ERR_CAST(inode);
565 if (generation && inode->i_generation != generation) {
566 iput(inode);
567 return ERR_PTR(-ESTALE);
568 }
569
570 return inode;
571}
572
573static struct dentry *
574nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
575 int fh_type)
576{
577 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
578 nilfs_nfs_get_inode);
579}
580
581static struct dentry *
582nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
583 int fh_type)
584{
585 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
586 nilfs_nfs_get_inode);
587}
588
589static struct export_operations nilfs_export_ops = {
590 .fh_to_dentry = nilfs_fh_to_dentry,
591 .fh_to_parent = nilfs_fh_to_parent,
592 .get_parent = nilfs_get_parent,
593};
594
595enum {
596 Opt_err_cont, Opt_err_panic, Opt_err_ro,
597 Opt_barrier, Opt_snapshot, Opt_order,
598 Opt_err,
599};
600
601static match_table_t tokens = {
602 {Opt_err_cont, "errors=continue"},
603 {Opt_err_panic, "errors=panic"},
604 {Opt_err_ro, "errors=remount-ro"},
605 {Opt_barrier, "barrier=%s"},
606 {Opt_snapshot, "cp=%u"},
607 {Opt_order, "order=%s"},
608 {Opt_err, NULL}
609};
610
611static int match_bool(substring_t *s, int *result)
612{
613 int len = s->to - s->from;
614
615 if (strncmp(s->from, "on", len) == 0)
616 *result = 1;
617 else if (strncmp(s->from, "off", len) == 0)
618 *result = 0;
619 else
620 return 1;
621 return 0;
622}
623
624static int parse_options(char *options, struct super_block *sb)
625{
626 struct nilfs_sb_info *sbi = NILFS_SB(sb);
627 char *p;
628 substring_t args[MAX_OPT_ARGS];
629 int option;
630
631 if (!options)
632 return 1;
633
634 while ((p = strsep(&options, ",")) != NULL) {
635 int token;
636 if (!*p)
637 continue;
638
639 token = match_token(p, tokens, args);
640 switch (token) {
641 case Opt_barrier:
642 if (match_bool(&args[0], &option))
643 return 0;
644 if (option)
645 nilfs_set_opt(sbi, BARRIER);
646 else
647 nilfs_clear_opt(sbi, BARRIER);
648 break;
649 case Opt_order:
650 if (strcmp(args[0].from, "relaxed") == 0)
651 /* Ordered data semantics */
652 nilfs_clear_opt(sbi, STRICT_ORDER);
653 else if (strcmp(args[0].from, "strict") == 0)
654 /* Strict in-order semantics */
655 nilfs_set_opt(sbi, STRICT_ORDER);
656 else
657 return 0;
658 break;
659 case Opt_err_panic:
660 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
661 break;
662 case Opt_err_ro:
663 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
664 break;
665 case Opt_err_cont:
666 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
667 break;
668 case Opt_snapshot:
669 if (match_int(&args[0], &option) || option <= 0)
670 return 0;
671 if (!(sb->s_flags & MS_RDONLY))
672 return 0;
673 sbi->s_snapshot_cno = option;
674 nilfs_set_opt(sbi, SNAPSHOT);
675 break;
676 default:
677 printk(KERN_ERR
678 "NILFS: Unrecognized mount option \"%s\"\n", p);
679 return 0;
680 }
681 }
682 return 1;
683}
684
685static inline void
686nilfs_set_default_options(struct nilfs_sb_info *sbi,
687 struct nilfs_super_block *sbp)
688{
689 sbi->s_mount_opt =
690 NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
691}
692
693static int nilfs_setup_super(struct nilfs_sb_info *sbi)
694{
695 struct the_nilfs *nilfs = sbi->s_nilfs;
e339ad31 696 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
783f6184
RK
697 int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
698 int mnt_count = le16_to_cpu(sbp->s_mnt_count);
699
700 /* nilfs->sem must be locked by the caller. */
701 if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
702 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
703 } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
704 printk(KERN_WARNING
705 "NILFS warning: mounting fs with errors\n");
706#if 0
707 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
708 printk(KERN_WARNING
709 "NILFS warning: maximal mount count reached\n");
710#endif
711 }
712 if (!max_mnt_count)
713 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
714
715 sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
716 sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
717 sbp->s_mtime = cpu_to_le64(get_seconds());
e339ad31 718 return nilfs_commit_super(sbi, 1);
783f6184
RK
719}
720
e339ad31
RK
721struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
722 u64 pos, int blocksize,
723 struct buffer_head **pbh)
783f6184 724{
e339ad31
RK
725 unsigned long long sb_index = pos;
726 unsigned long offset;
783f6184 727
e339ad31 728 offset = do_div(sb_index, blocksize);
783f6184 729 *pbh = sb_bread(sb, sb_index);
e339ad31 730 if (!*pbh)
783f6184 731 return NULL;
783f6184
RK
732 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
733}
734
783f6184
RK
735int nilfs_store_magic_and_option(struct super_block *sb,
736 struct nilfs_super_block *sbp,
737 char *data)
738{
739 struct nilfs_sb_info *sbi = NILFS_SB(sb);
740
783f6184
RK
741 sb->s_magic = le16_to_cpu(sbp->s_magic);
742
743 /* FS independent flags */
744#ifdef NILFS_ATIME_DISABLE
745 sb->s_flags |= MS_NOATIME;
746#endif
747
783f6184
RK
748 nilfs_set_default_options(sbi, sbp);
749
750 sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
751 sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
752 sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
753 sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
754
e339ad31 755 return !parse_options(data, sb) ? -EINVAL : 0 ;
783f6184
RK
756}
757
758/**
759 * nilfs_fill_super() - initialize a super block instance
760 * @sb: super_block
761 * @data: mount options
762 * @silent: silent mode flag
763 * @nilfs: the_nilfs struct
764 *
765 * This function is called exclusively by bd_mount_mutex.
766 * So, the recovery process is protected from other simultaneous mounts.
767 */
768static int
769nilfs_fill_super(struct super_block *sb, void *data, int silent,
770 struct the_nilfs *nilfs)
771{
772 struct nilfs_sb_info *sbi;
773 struct inode *root;
774 __u64 cno;
775 int err;
776
777 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
778 if (!sbi)
779 return -ENOMEM;
780
781 sb->s_fs_info = sbi;
782
783 get_nilfs(nilfs);
784 sbi->s_nilfs = nilfs;
785 sbi->s_super = sb;
786
787 err = init_nilfs(nilfs, sbi, (char *)data);
788 if (err)
789 goto failed_sbi;
790
791 spin_lock_init(&sbi->s_inode_lock);
792 INIT_LIST_HEAD(&sbi->s_dirty_files);
793 INIT_LIST_HEAD(&sbi->s_list);
794
795 /*
796 * Following initialization is overlapped because
797 * nilfs_sb_info structure has been cleared at the beginning.
798 * But we reserve them to keep our interest and make ready
799 * for the future change.
800 */
801 get_random_bytes(&sbi->s_next_generation,
802 sizeof(sbi->s_next_generation));
803 spin_lock_init(&sbi->s_next_gen_lock);
804
805 sb->s_op = &nilfs_sops;
806 sb->s_export_op = &nilfs_export_ops;
807 sb->s_root = NULL;
61239230 808 sb->s_time_gran = 1;
783f6184
RK
809
810 if (!nilfs_loaded(nilfs)) {
811 err = load_nilfs(nilfs, sbi);
812 if (err)
813 goto failed_sbi;
814 }
815 cno = nilfs_last_cno(nilfs);
816
817 if (sb->s_flags & MS_RDONLY) {
818 if (nilfs_test_opt(sbi, SNAPSHOT)) {
1f5abe7e
RK
819 err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
820 sbi->s_snapshot_cno);
821 if (err < 0)
822 goto failed_sbi;
823 if (!err) {
783f6184
RK
824 printk(KERN_ERR
825 "NILFS: The specified checkpoint is "
826 "not a snapshot "
827 "(checkpoint number=%llu).\n",
828 (unsigned long long)sbi->s_snapshot_cno);
829 err = -EINVAL;
830 goto failed_sbi;
831 }
832 cno = sbi->s_snapshot_cno;
833 } else
834 /* Read-only mount */
835 sbi->s_snapshot_cno = cno;
836 }
837
838 err = nilfs_attach_checkpoint(sbi, cno);
839 if (err) {
840 printk(KERN_ERR "NILFS: error loading a checkpoint"
841 " (checkpoint number=%llu).\n", (unsigned long long)cno);
842 goto failed_sbi;
843 }
844
845 if (!(sb->s_flags & MS_RDONLY)) {
cece5520 846 err = nilfs_attach_segment_constructor(sbi);
783f6184
RK
847 if (err)
848 goto failed_checkpoint;
849 }
850
851 root = nilfs_iget(sb, NILFS_ROOT_INO);
852 if (IS_ERR(root)) {
853 printk(KERN_ERR "NILFS: get root inode failed\n");
854 err = PTR_ERR(root);
855 goto failed_segctor;
856 }
857 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
858 iput(root);
859 printk(KERN_ERR "NILFS: corrupt root inode.\n");
860 err = -EINVAL;
861 goto failed_segctor;
862 }
863 sb->s_root = d_alloc_root(root);
864 if (!sb->s_root) {
865 iput(root);
866 printk(KERN_ERR "NILFS: get root dentry failed\n");
867 err = -ENOMEM;
868 goto failed_segctor;
869 }
870
871 if (!(sb->s_flags & MS_RDONLY)) {
872 down_write(&nilfs->ns_sem);
873 nilfs_setup_super(sbi);
874 up_write(&nilfs->ns_sem);
875 }
876
877 err = nilfs_mark_recovery_complete(sbi);
878 if (unlikely(err)) {
879 printk(KERN_ERR "NILFS: recovery failed.\n");
880 goto failed_root;
881 }
882
883 return 0;
884
885 failed_root:
886 dput(sb->s_root);
887 sb->s_root = NULL;
888
889 failed_segctor:
890 nilfs_detach_segment_constructor(sbi);
891
892 failed_checkpoint:
893 nilfs_detach_checkpoint(sbi);
894
895 failed_sbi:
896 put_nilfs(nilfs);
897 sb->s_fs_info = NULL;
898 kfree(sbi);
899 return err;
900}
901
902static int nilfs_remount(struct super_block *sb, int *flags, char *data)
903{
904 struct nilfs_sb_info *sbi = NILFS_SB(sb);
905 struct nilfs_super_block *sbp;
906 struct the_nilfs *nilfs = sbi->s_nilfs;
907 unsigned long old_sb_flags;
908 struct nilfs_mount_options old_opts;
909 int err;
910
337eb00a
AIB
911 lock_kernel();
912
783f6184
RK
913 old_sb_flags = sb->s_flags;
914 old_opts.mount_opt = sbi->s_mount_opt;
915 old_opts.snapshot_cno = sbi->s_snapshot_cno;
916
917 if (!parse_options(data, sb)) {
918 err = -EINVAL;
919 goto restore_opts;
920 }
921 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
922
923 if ((*flags & MS_RDONLY) &&
924 sbi->s_snapshot_cno != old_opts.snapshot_cno) {
925 printk(KERN_WARNING "NILFS (device %s): couldn't "
926 "remount to a different snapshot. \n",
927 sb->s_id);
928 err = -EINVAL;
929 goto restore_opts;
930 }
931
932 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
933 goto out;
934 if (*flags & MS_RDONLY) {
935 /* Shutting down the segment constructor */
936 nilfs_detach_segment_constructor(sbi);
937 sb->s_flags |= MS_RDONLY;
938
939 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
940 /* nilfs_set_opt(sbi, SNAPSHOT); */
941
942 /*
943 * Remounting a valid RW partition RDONLY, so set
944 * the RDONLY flag and then mark the partition as valid again.
945 */
946 down_write(&nilfs->ns_sem);
e339ad31 947 sbp = nilfs->ns_sbp[0];
783f6184
RK
948 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
949 (nilfs->ns_mount_state & NILFS_VALID_FS))
950 sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
951 sbp->s_mtime = cpu_to_le64(get_seconds());
e339ad31 952 nilfs_commit_super(sbi, 1);
783f6184
RK
953 up_write(&nilfs->ns_sem);
954 } else {
955 /*
956 * Mounting a RDONLY partition read-write, so reread and
957 * store the current valid flag. (It may have been changed
958 * by fsck since we originally mounted the partition.)
959 */
960 down(&sb->s_bdev->bd_mount_sem);
961 /* Check existing RW-mount */
962 if (test_exclusive_mount(sb->s_type, sb->s_bdev, 0)) {
963 printk(KERN_WARNING "NILFS (device %s): couldn't "
964 "remount because a RW-mount exists.\n",
965 sb->s_id);
966 err = -EBUSY;
967 goto rw_remount_failed;
968 }
969 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
970 printk(KERN_WARNING "NILFS (device %s): couldn't "
971 "remount because the current RO-mount is not "
972 "the latest one.\n",
973 sb->s_id);
974 err = -EINVAL;
975 goto rw_remount_failed;
976 }
977 sb->s_flags &= ~MS_RDONLY;
978 nilfs_clear_opt(sbi, SNAPSHOT);
979 sbi->s_snapshot_cno = 0;
980
cece5520 981 err = nilfs_attach_segment_constructor(sbi);
783f6184
RK
982 if (err)
983 goto rw_remount_failed;
984
985 down_write(&nilfs->ns_sem);
986 nilfs_setup_super(sbi);
987 up_write(&nilfs->ns_sem);
988
989 up(&sb->s_bdev->bd_mount_sem);
990 }
991 out:
337eb00a 992 unlock_kernel();
783f6184
RK
993 return 0;
994
995 rw_remount_failed:
996 up(&sb->s_bdev->bd_mount_sem);
997 restore_opts:
998 sb->s_flags = old_sb_flags;
999 sbi->s_mount_opt = old_opts.mount_opt;
1000 sbi->s_snapshot_cno = old_opts.snapshot_cno;
337eb00a 1001 unlock_kernel();
783f6184
RK
1002 return err;
1003}
1004
1005struct nilfs_super_data {
1006 struct block_device *bdev;
1007 __u64 cno;
1008 int flags;
1009};
1010
1011/**
1012 * nilfs_identify - pre-read mount options needed to identify mount instance
1013 * @data: mount options
1014 * @sd: nilfs_super_data
1015 */
1016static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1017{
1018 char *p, *options = data;
1019 substring_t args[MAX_OPT_ARGS];
1020 int option, token;
1021 int ret = 0;
1022
1023 do {
1024 p = strsep(&options, ",");
1025 if (p != NULL && *p) {
1026 token = match_token(p, tokens, args);
1027 if (token == Opt_snapshot) {
1028 if (!(sd->flags & MS_RDONLY))
1029 ret++;
1030 else {
1031 ret = match_int(&args[0], &option);
1032 if (!ret) {
1033 if (option > 0)
1034 sd->cno = option;
1035 else
1036 ret++;
1037 }
1038 }
1039 }
1040 if (ret)
1041 printk(KERN_ERR
1042 "NILFS: invalid mount option: %s\n", p);
1043 }
1044 if (!options)
1045 break;
1046 BUG_ON(options == data);
1047 *(options - 1) = ',';
1048 } while (!ret);
1049 return ret;
1050}
1051
1052static int nilfs_set_bdev_super(struct super_block *s, void *data)
1053{
1054 struct nilfs_super_data *sd = data;
1055
1056 s->s_bdev = sd->bdev;
1057 s->s_dev = s->s_bdev->bd_dev;
1058 return 0;
1059}
1060
1061static int nilfs_test_bdev_super(struct super_block *s, void *data)
783f6184
RK
1062{
1063 struct nilfs_super_data *sd = data;
1064 int ret;
1065
1066 if (s->s_bdev != sd->bdev)
1067 return 0;
1068
1069 if (!((s->s_flags | sd->flags) & MS_RDONLY))
1070 return 1; /* Reuse an old R/W-mode super_block */
1071
1072 if (s->s_flags & sd->flags & MS_RDONLY) {
1073 if (down_read_trylock(&s->s_umount)) {
1074 ret = s->s_root &&
1075 (sd->cno == NILFS_SB(s)->s_snapshot_cno);
1076 up_read(&s->s_umount);
1077 /*
1078 * This path is locked with sb_lock by sget().
1079 * So, drop_super() causes deadlock.
1080 */
1081 return ret;
1082 }
1083 }
1084 return 0;
1085}
1086
1087static int
1088nilfs_get_sb(struct file_system_type *fs_type, int flags,
1089 const char *dev_name, void *data, struct vfsmount *mnt)
1090{
1091 struct nilfs_super_data sd;
33c8e57c
RK
1092 struct super_block *s;
1093 struct the_nilfs *nilfs;
783f6184
RK
1094 int err, need_to_close = 1;
1095
1096 sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1097 if (IS_ERR(sd.bdev))
1098 return PTR_ERR(sd.bdev);
1099
1100 /*
1101 * To get mount instance using sget() vfs-routine, NILFS needs
1102 * much more information than normal filesystems to identify mount
1103 * instance. For snapshot mounts, not only a mount type (ro-mount
1104 * or rw-mount) but also a checkpoint number is required.
1105 * The results are passed in sget() using nilfs_super_data.
1106 */
1107 sd.cno = 0;
1108 sd.flags = flags;
1109 if (nilfs_identify((char *)data, &sd)) {
1110 err = -EINVAL;
1111 goto failed;
1112 }
1113
33c8e57c
RK
1114 nilfs = find_or_create_nilfs(sd.bdev);
1115 if (!nilfs) {
1116 err = -ENOMEM;
1117 goto failed;
1118 }
1119
783f6184
RK
1120 down(&sd.bdev->bd_mount_sem);
1121 if (!sd.cno &&
1122 (err = test_exclusive_mount(fs_type, sd.bdev, flags ^ MS_RDONLY))) {
1123 err = (err < 0) ? : -EBUSY;
1124 goto failed_unlock;
1125 }
1126
1127 /*
33c8e57c 1128 * Search specified snapshot or R/W mode super_block
783f6184 1129 */
33c8e57c
RK
1130 if (!sd.cno)
1131 /* trying to get the latest checkpoint. */
1132 sd.cno = nilfs_last_cno(nilfs);
783f6184 1133
33c8e57c
RK
1134 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1135 if (IS_ERR(s)) {
1136 err = PTR_ERR(s);
1137 goto failed_unlock;
783f6184
RK
1138 }
1139
1140 if (!s->s_root) {
1141 char b[BDEVNAME_SIZE];
1142
33c8e57c 1143 /* New superblock instance created */
783f6184
RK
1144 s->s_flags = flags;
1145 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1146 sb_set_blocksize(s, block_size(sd.bdev));
1147
1148 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1149 if (err)
1150 goto cancel_new;
1151
1152 s->s_flags |= MS_ACTIVE;
1153 need_to_close = 0;
783f6184
RK
1154 }
1155
1156 up(&sd.bdev->bd_mount_sem);
1157 put_nilfs(nilfs);
1158 if (need_to_close)
1159 close_bdev_exclusive(sd.bdev, flags);
1160 simple_set_mnt(mnt, s);
1161 return 0;
1162
783f6184
RK
1163 failed_unlock:
1164 up(&sd.bdev->bd_mount_sem);
33c8e57c 1165 put_nilfs(nilfs);
783f6184
RK
1166 failed:
1167 close_bdev_exclusive(sd.bdev, flags);
1168
1169 return err;
1170
1171 cancel_new:
1172 /* Abandoning the newly allocated superblock */
1173 up(&sd.bdev->bd_mount_sem);
33c8e57c 1174 put_nilfs(nilfs);
783f6184
RK
1175 up_write(&s->s_umount);
1176 deactivate_super(s);
1177 /*
1178 * deactivate_super() invokes close_bdev_exclusive().
1179 * We must finish all post-cleaning before this call;
1180 * put_nilfs() and unlocking bd_mount_sem need the block device.
1181 */
1182 return err;
1183}
1184
1185static int nilfs_test_bdev_super3(struct super_block *s, void *data)
1186{
1187 struct nilfs_super_data *sd = data;
1188 int ret;
1189
1190 if (s->s_bdev != sd->bdev)
1191 return 0;
1192 if (down_read_trylock(&s->s_umount)) {
1193 ret = (s->s_flags & MS_RDONLY) && s->s_root &&
1194 nilfs_test_opt(NILFS_SB(s), SNAPSHOT);
1195 up_read(&s->s_umount);
1196 if (ret)
1197 return 0; /* ignore snapshot mounts */
1198 }
1199 return !((sd->flags ^ s->s_flags) & MS_RDONLY);
1200}
1201
1202static int __false_bdev_super(struct super_block *s, void *data)
1203{
1204#if 0 /* XXX: workaround for lock debug. This is not good idea */
1205 up_write(&s->s_umount);
1206#endif
1207 return -EFAULT;
1208}
1209
1210/**
1211 * test_exclusive_mount - check whether an exclusive RW/RO mount exists or not.
1212 * fs_type: filesystem type
1213 * bdev: block device
1214 * flag: 0 (check rw-mount) or MS_RDONLY (check ro-mount)
1215 * res: pointer to an integer to store result
1216 *
1217 * This function must be called within a section protected by bd_mount_mutex.
1218 */
1219static int test_exclusive_mount(struct file_system_type *fs_type,
1220 struct block_device *bdev, int flags)
1221{
1222 struct super_block *s;
1223 struct nilfs_super_data sd = { .flags = flags, .bdev = bdev };
1224
1225 s = sget(fs_type, nilfs_test_bdev_super3, __false_bdev_super, &sd);
1226 if (IS_ERR(s)) {
1227 if (PTR_ERR(s) != -EFAULT)
1228 return PTR_ERR(s);
1229 return 0; /* Not found */
1230 }
1231 up_write(&s->s_umount);
1232 deactivate_super(s);
1233 return 1; /* Found */
1234}
1235
1236struct file_system_type nilfs_fs_type = {
1237 .owner = THIS_MODULE,
1238 .name = "nilfs2",
1239 .get_sb = nilfs_get_sb,
1240 .kill_sb = kill_block_super,
1241 .fs_flags = FS_REQUIRES_DEV,
1242};
1243
1244static int __init init_nilfs_fs(void)
1245{
1246 int err;
1247
1248 err = nilfs_init_inode_cache();
1249 if (err)
1250 goto failed;
1251
1252 err = nilfs_init_transaction_cache();
1253 if (err)
1254 goto failed_inode_cache;
1255
1256 err = nilfs_init_segbuf_cache();
1257 if (err)
1258 goto failed_transaction_cache;
1259
1260 err = nilfs_btree_path_cache_init();
1261 if (err)
1262 goto failed_segbuf_cache;
1263
1264 err = register_filesystem(&nilfs_fs_type);
1265 if (err)
1266 goto failed_btree_path_cache;
1267
1268 return 0;
1269
1270 failed_btree_path_cache:
1271 nilfs_btree_path_cache_destroy();
1272
1273 failed_segbuf_cache:
1274 nilfs_destroy_segbuf_cache();
1275
1276 failed_transaction_cache:
1277 nilfs_destroy_transaction_cache();
1278
1279 failed_inode_cache:
1280 nilfs_destroy_inode_cache();
1281
1282 failed:
1283 return err;
1284}
1285
1286static void __exit exit_nilfs_fs(void)
1287{
1288 nilfs_destroy_segbuf_cache();
1289 nilfs_destroy_transaction_cache();
1290 nilfs_destroy_inode_cache();
1291 nilfs_btree_path_cache_destroy();
1292 unregister_filesystem(&nilfs_fs_type);
1293}
1294
1295module_init(init_nilfs_fs)
1296module_exit(exit_nilfs_fs)