Linux 2.6.36-rc5
[linux-2.6/next.git] / fs / gfs2 / super.c
blob77cb9f830ee47eb51520bd8581ebc700b426455e
1 /*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
10 #include <linux/bio.h>
11 #include <linux/sched.h>
12 #include <linux/slab.h>
13 #include <linux/spinlock.h>
14 #include <linux/completion.h>
15 #include <linux/buffer_head.h>
16 #include <linux/statfs.h>
17 #include <linux/seq_file.h>
18 #include <linux/mount.h>
19 #include <linux/kthread.h>
20 #include <linux/delay.h>
21 #include <linux/gfs2_ondisk.h>
22 #include <linux/crc32.h>
23 #include <linux/time.h>
24 #include <linux/wait.h>
25 #include <linux/writeback.h>
27 #include "gfs2.h"
28 #include "incore.h"
29 #include "bmap.h"
30 #include "dir.h"
31 #include "glock.h"
32 #include "glops.h"
33 #include "inode.h"
34 #include "log.h"
35 #include "meta_io.h"
36 #include "quota.h"
37 #include "recovery.h"
38 #include "rgrp.h"
39 #include "super.h"
40 #include "trans.h"
41 #include "util.h"
42 #include "sys.h"
43 #include "xattr.h"
45 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
47 enum {
48 Opt_lockproto,
49 Opt_locktable,
50 Opt_hostdata,
51 Opt_spectator,
52 Opt_ignore_local_fs,
53 Opt_localflocks,
54 Opt_localcaching,
55 Opt_debug,
56 Opt_nodebug,
57 Opt_upgrade,
58 Opt_acl,
59 Opt_noacl,
60 Opt_quota_off,
61 Opt_quota_account,
62 Opt_quota_on,
63 Opt_quota,
64 Opt_noquota,
65 Opt_suiddir,
66 Opt_nosuiddir,
67 Opt_data_writeback,
68 Opt_data_ordered,
69 Opt_meta,
70 Opt_discard,
71 Opt_nodiscard,
72 Opt_commit,
73 Opt_err_withdraw,
74 Opt_err_panic,
75 Opt_statfs_quantum,
76 Opt_statfs_percent,
77 Opt_quota_quantum,
78 Opt_barrier,
79 Opt_nobarrier,
80 Opt_error,
83 static const match_table_t tokens = {
84 {Opt_lockproto, "lockproto=%s"},
85 {Opt_locktable, "locktable=%s"},
86 {Opt_hostdata, "hostdata=%s"},
87 {Opt_spectator, "spectator"},
88 {Opt_ignore_local_fs, "ignore_local_fs"},
89 {Opt_localflocks, "localflocks"},
90 {Opt_localcaching, "localcaching"},
91 {Opt_debug, "debug"},
92 {Opt_nodebug, "nodebug"},
93 {Opt_upgrade, "upgrade"},
94 {Opt_acl, "acl"},
95 {Opt_noacl, "noacl"},
96 {Opt_quota_off, "quota=off"},
97 {Opt_quota_account, "quota=account"},
98 {Opt_quota_on, "quota=on"},
99 {Opt_quota, "quota"},
100 {Opt_noquota, "noquota"},
101 {Opt_suiddir, "suiddir"},
102 {Opt_nosuiddir, "nosuiddir"},
103 {Opt_data_writeback, "data=writeback"},
104 {Opt_data_ordered, "data=ordered"},
105 {Opt_meta, "meta"},
106 {Opt_discard, "discard"},
107 {Opt_nodiscard, "nodiscard"},
108 {Opt_commit, "commit=%d"},
109 {Opt_err_withdraw, "errors=withdraw"},
110 {Opt_err_panic, "errors=panic"},
111 {Opt_statfs_quantum, "statfs_quantum=%d"},
112 {Opt_statfs_percent, "statfs_percent=%d"},
113 {Opt_quota_quantum, "quota_quantum=%d"},
114 {Opt_barrier, "barrier"},
115 {Opt_nobarrier, "nobarrier"},
116 {Opt_error, NULL}
120 * gfs2_mount_args - Parse mount options
121 * @args: The structure into which the parsed options will be written
122 * @options: The options to parse
124 * Return: errno
127 int gfs2_mount_args(struct gfs2_args *args, char *options)
129 char *o;
130 int token;
131 substring_t tmp[MAX_OPT_ARGS];
132 int rv;
134 /* Split the options into tokens with the "," character and
135 process them */
137 while (1) {
138 o = strsep(&options, ",");
139 if (o == NULL)
140 break;
141 if (*o == '\0')
142 continue;
144 token = match_token(o, tokens, tmp);
145 switch (token) {
146 case Opt_lockproto:
147 match_strlcpy(args->ar_lockproto, &tmp[0],
148 GFS2_LOCKNAME_LEN);
149 break;
150 case Opt_locktable:
151 match_strlcpy(args->ar_locktable, &tmp[0],
152 GFS2_LOCKNAME_LEN);
153 break;
154 case Opt_hostdata:
155 match_strlcpy(args->ar_hostdata, &tmp[0],
156 GFS2_LOCKNAME_LEN);
157 break;
158 case Opt_spectator:
159 args->ar_spectator = 1;
160 break;
161 case Opt_ignore_local_fs:
162 args->ar_ignore_local_fs = 1;
163 break;
164 case Opt_localflocks:
165 args->ar_localflocks = 1;
166 break;
167 case Opt_localcaching:
168 args->ar_localcaching = 1;
169 break;
170 case Opt_debug:
171 if (args->ar_errors == GFS2_ERRORS_PANIC) {
172 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
173 "are mutually exclusive.\n");
174 return -EINVAL;
176 args->ar_debug = 1;
177 break;
178 case Opt_nodebug:
179 args->ar_debug = 0;
180 break;
181 case Opt_upgrade:
182 args->ar_upgrade = 1;
183 break;
184 case Opt_acl:
185 args->ar_posix_acl = 1;
186 break;
187 case Opt_noacl:
188 args->ar_posix_acl = 0;
189 break;
190 case Opt_quota_off:
191 case Opt_noquota:
192 args->ar_quota = GFS2_QUOTA_OFF;
193 break;
194 case Opt_quota_account:
195 args->ar_quota = GFS2_QUOTA_ACCOUNT;
196 break;
197 case Opt_quota_on:
198 case Opt_quota:
199 args->ar_quota = GFS2_QUOTA_ON;
200 break;
201 case Opt_suiddir:
202 args->ar_suiddir = 1;
203 break;
204 case Opt_nosuiddir:
205 args->ar_suiddir = 0;
206 break;
207 case Opt_data_writeback:
208 args->ar_data = GFS2_DATA_WRITEBACK;
209 break;
210 case Opt_data_ordered:
211 args->ar_data = GFS2_DATA_ORDERED;
212 break;
213 case Opt_meta:
214 args->ar_meta = 1;
215 break;
216 case Opt_discard:
217 args->ar_discard = 1;
218 break;
219 case Opt_nodiscard:
220 args->ar_discard = 0;
221 break;
222 case Opt_commit:
223 rv = match_int(&tmp[0], &args->ar_commit);
224 if (rv || args->ar_commit <= 0) {
225 printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n");
226 return rv ? rv : -EINVAL;
228 break;
229 case Opt_statfs_quantum:
230 rv = match_int(&tmp[0], &args->ar_statfs_quantum);
231 if (rv || args->ar_statfs_quantum < 0) {
232 printk(KERN_WARNING "GFS2: statfs_quantum mount option requires a non-negative numeric argument\n");
233 return rv ? rv : -EINVAL;
235 break;
236 case Opt_quota_quantum:
237 rv = match_int(&tmp[0], &args->ar_quota_quantum);
238 if (rv || args->ar_quota_quantum <= 0) {
239 printk(KERN_WARNING "GFS2: quota_quantum mount option requires a positive numeric argument\n");
240 return rv ? rv : -EINVAL;
242 break;
243 case Opt_statfs_percent:
244 rv = match_int(&tmp[0], &args->ar_statfs_percent);
245 if (rv || args->ar_statfs_percent < 0 ||
246 args->ar_statfs_percent > 100) {
247 printk(KERN_WARNING "statfs_percent mount option requires a numeric argument between 0 and 100\n");
248 return rv ? rv : -EINVAL;
250 break;
251 case Opt_err_withdraw:
252 args->ar_errors = GFS2_ERRORS_WITHDRAW;
253 break;
254 case Opt_err_panic:
255 if (args->ar_debug) {
256 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
257 "are mutually exclusive.\n");
258 return -EINVAL;
260 args->ar_errors = GFS2_ERRORS_PANIC;
261 break;
262 case Opt_barrier:
263 args->ar_nobarrier = 0;
264 break;
265 case Opt_nobarrier:
266 args->ar_nobarrier = 1;
267 break;
268 case Opt_error:
269 default:
270 printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o);
271 return -EINVAL;
275 return 0;
279 * gfs2_jindex_free - Clear all the journal index information
280 * @sdp: The GFS2 superblock
284 void gfs2_jindex_free(struct gfs2_sbd *sdp)
286 struct list_head list, *head;
287 struct gfs2_jdesc *jd;
288 struct gfs2_journal_extent *jext;
290 spin_lock(&sdp->sd_jindex_spin);
291 list_add(&list, &sdp->sd_jindex_list);
292 list_del_init(&sdp->sd_jindex_list);
293 sdp->sd_journals = 0;
294 spin_unlock(&sdp->sd_jindex_spin);
296 while (!list_empty(&list)) {
297 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
298 head = &jd->extent_list;
299 while (!list_empty(head)) {
300 jext = list_entry(head->next,
301 struct gfs2_journal_extent,
302 extent_list);
303 list_del(&jext->extent_list);
304 kfree(jext);
306 list_del(&jd->jd_list);
307 iput(jd->jd_inode);
308 kfree(jd);
312 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
314 struct gfs2_jdesc *jd;
315 int found = 0;
317 list_for_each_entry(jd, head, jd_list) {
318 if (jd->jd_jid == jid) {
319 found = 1;
320 break;
324 if (!found)
325 jd = NULL;
327 return jd;
330 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
332 struct gfs2_jdesc *jd;
334 spin_lock(&sdp->sd_jindex_spin);
335 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
336 spin_unlock(&sdp->sd_jindex_spin);
338 return jd;
341 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
343 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
344 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
346 if (ip->i_disksize < (8 << 20) || ip->i_disksize > (1 << 30) ||
347 (ip->i_disksize & (sdp->sd_sb.sb_bsize - 1))) {
348 gfs2_consist_inode(ip);
349 return -EIO;
351 jd->jd_blocks = ip->i_disksize >> sdp->sd_sb.sb_bsize_shift;
353 if (gfs2_write_alloc_required(ip, 0, ip->i_disksize)) {
354 gfs2_consist_inode(ip);
355 return -EIO;
358 return 0;
362 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
363 * @sdp: the filesystem
365 * Returns: errno
368 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
370 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
371 struct gfs2_glock *j_gl = ip->i_gl;
372 struct gfs2_holder t_gh;
373 struct gfs2_log_header_host head;
374 int error;
376 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
377 if (error)
378 return error;
380 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
382 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
383 if (error)
384 goto fail;
386 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
387 gfs2_consist(sdp);
388 error = -EIO;
389 goto fail;
392 /* Initialize some head of the log stuff */
393 sdp->sd_log_sequence = head.lh_sequence + 1;
394 gfs2_log_pointers_init(sdp, head.lh_blkno);
396 error = gfs2_quota_init(sdp);
397 if (error)
398 goto fail;
400 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
402 gfs2_glock_dq_uninit(&t_gh);
404 return 0;
406 fail:
407 t_gh.gh_flags |= GL_NOCACHE;
408 gfs2_glock_dq_uninit(&t_gh);
410 return error;
413 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
415 const struct gfs2_statfs_change *str = buf;
417 sc->sc_total = be64_to_cpu(str->sc_total);
418 sc->sc_free = be64_to_cpu(str->sc_free);
419 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
422 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
424 struct gfs2_statfs_change *str = buf;
426 str->sc_total = cpu_to_be64(sc->sc_total);
427 str->sc_free = cpu_to_be64(sc->sc_free);
428 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
431 int gfs2_statfs_init(struct gfs2_sbd *sdp)
433 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
434 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
435 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
436 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
437 struct buffer_head *m_bh, *l_bh;
438 struct gfs2_holder gh;
439 int error;
441 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
442 &gh);
443 if (error)
444 return error;
446 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
447 if (error)
448 goto out;
450 if (sdp->sd_args.ar_spectator) {
451 spin_lock(&sdp->sd_statfs_spin);
452 gfs2_statfs_change_in(m_sc, m_bh->b_data +
453 sizeof(struct gfs2_dinode));
454 spin_unlock(&sdp->sd_statfs_spin);
455 } else {
456 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
457 if (error)
458 goto out_m_bh;
460 spin_lock(&sdp->sd_statfs_spin);
461 gfs2_statfs_change_in(m_sc, m_bh->b_data +
462 sizeof(struct gfs2_dinode));
463 gfs2_statfs_change_in(l_sc, l_bh->b_data +
464 sizeof(struct gfs2_dinode));
465 spin_unlock(&sdp->sd_statfs_spin);
467 brelse(l_bh);
470 out_m_bh:
471 brelse(m_bh);
472 out:
473 gfs2_glock_dq_uninit(&gh);
474 return 0;
477 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
478 s64 dinodes)
480 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
481 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
482 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
483 struct buffer_head *l_bh;
484 s64 x, y;
485 int need_sync = 0;
486 int error;
488 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
489 if (error)
490 return;
492 gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
494 spin_lock(&sdp->sd_statfs_spin);
495 l_sc->sc_total += total;
496 l_sc->sc_free += free;
497 l_sc->sc_dinodes += dinodes;
498 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
499 if (sdp->sd_args.ar_statfs_percent) {
500 x = 100 * l_sc->sc_free;
501 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
502 if (x >= y || x <= -y)
503 need_sync = 1;
505 spin_unlock(&sdp->sd_statfs_spin);
507 brelse(l_bh);
508 if (need_sync)
509 gfs2_wake_up_statfs(sdp);
512 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
513 struct buffer_head *l_bh)
515 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
516 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
517 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
518 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
520 gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
522 spin_lock(&sdp->sd_statfs_spin);
523 m_sc->sc_total += l_sc->sc_total;
524 m_sc->sc_free += l_sc->sc_free;
525 m_sc->sc_dinodes += l_sc->sc_dinodes;
526 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
527 memset(l_bh->b_data + sizeof(struct gfs2_dinode),
528 0, sizeof(struct gfs2_statfs_change));
529 spin_unlock(&sdp->sd_statfs_spin);
531 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
532 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
535 int gfs2_statfs_sync(struct super_block *sb, int type)
537 struct gfs2_sbd *sdp = sb->s_fs_info;
538 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
539 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
540 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
541 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
542 struct gfs2_holder gh;
543 struct buffer_head *m_bh, *l_bh;
544 int error;
546 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
547 &gh);
548 if (error)
549 return error;
551 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
552 if (error)
553 goto out;
555 spin_lock(&sdp->sd_statfs_spin);
556 gfs2_statfs_change_in(m_sc, m_bh->b_data +
557 sizeof(struct gfs2_dinode));
558 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
559 spin_unlock(&sdp->sd_statfs_spin);
560 goto out_bh;
562 spin_unlock(&sdp->sd_statfs_spin);
564 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
565 if (error)
566 goto out_bh;
568 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
569 if (error)
570 goto out_bh2;
572 update_statfs(sdp, m_bh, l_bh);
573 sdp->sd_statfs_force_sync = 0;
575 gfs2_trans_end(sdp);
577 out_bh2:
578 brelse(l_bh);
579 out_bh:
580 brelse(m_bh);
581 out:
582 gfs2_glock_dq_uninit(&gh);
583 return error;
586 struct lfcc {
587 struct list_head list;
588 struct gfs2_holder gh;
592 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
593 * journals are clean
594 * @sdp: the file system
595 * @state: the state to put the transaction lock into
596 * @t_gh: the hold on the transaction lock
598 * Returns: errno
601 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
602 struct gfs2_holder *t_gh)
604 struct gfs2_inode *ip;
605 struct gfs2_jdesc *jd;
606 struct lfcc *lfcc;
607 LIST_HEAD(list);
608 struct gfs2_log_header_host lh;
609 int error;
611 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
612 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
613 if (!lfcc) {
614 error = -ENOMEM;
615 goto out;
617 ip = GFS2_I(jd->jd_inode);
618 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
619 if (error) {
620 kfree(lfcc);
621 goto out;
623 list_add(&lfcc->list, &list);
626 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
627 GL_NOCACHE, t_gh);
629 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
630 error = gfs2_jdesc_check(jd);
631 if (error)
632 break;
633 error = gfs2_find_jhead(jd, &lh);
634 if (error)
635 break;
636 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
637 error = -EBUSY;
638 break;
642 if (error)
643 gfs2_glock_dq_uninit(t_gh);
645 out:
646 while (!list_empty(&list)) {
647 lfcc = list_entry(list.next, struct lfcc, list);
648 list_del(&lfcc->list);
649 gfs2_glock_dq_uninit(&lfcc->gh);
650 kfree(lfcc);
652 return error;
656 * gfs2_freeze_fs - freezes the file system
657 * @sdp: the file system
659 * This function flushes data and meta data for all machines by
660 * aquiring the transaction log exclusively. All journals are
661 * ensured to be in a clean state as well.
663 * Returns: errno
666 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
668 int error = 0;
670 mutex_lock(&sdp->sd_freeze_lock);
672 if (!sdp->sd_freeze_count++) {
673 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
674 if (error)
675 sdp->sd_freeze_count--;
678 mutex_unlock(&sdp->sd_freeze_lock);
680 return error;
684 * gfs2_unfreeze_fs - unfreezes the file system
685 * @sdp: the file system
687 * This function allows the file system to proceed by unlocking
688 * the exclusively held transaction lock. Other GFS2 nodes are
689 * now free to acquire the lock shared and go on with their lives.
693 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
695 mutex_lock(&sdp->sd_freeze_lock);
697 if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
698 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
700 mutex_unlock(&sdp->sd_freeze_lock);
705 * gfs2_write_inode - Make sure the inode is stable on the disk
706 * @inode: The inode
707 * @sync: synchronous write flag
709 * Returns: errno
712 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
714 struct gfs2_inode *ip = GFS2_I(inode);
715 struct gfs2_sbd *sdp = GFS2_SB(inode);
716 struct gfs2_holder gh;
717 struct buffer_head *bh;
718 struct timespec atime;
719 struct gfs2_dinode *di;
720 int ret = 0;
722 /* Check this is a "normal" inode, etc */
723 if (current->flags & PF_MEMALLOC)
724 return 0;
725 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
726 if (ret)
727 goto do_flush;
728 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
729 if (ret)
730 goto do_unlock;
731 ret = gfs2_meta_inode_buffer(ip, &bh);
732 if (ret == 0) {
733 di = (struct gfs2_dinode *)bh->b_data;
734 atime.tv_sec = be64_to_cpu(di->di_atime);
735 atime.tv_nsec = be32_to_cpu(di->di_atime_nsec);
736 if (timespec_compare(&inode->i_atime, &atime) > 0) {
737 gfs2_trans_add_bh(ip->i_gl, bh, 1);
738 gfs2_dinode_out(ip, bh->b_data);
740 brelse(bh);
742 gfs2_trans_end(sdp);
743 do_unlock:
744 gfs2_glock_dq_uninit(&gh);
745 do_flush:
746 if (wbc->sync_mode == WB_SYNC_ALL)
747 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
748 return ret;
752 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
753 * @sdp: the filesystem
755 * Returns: errno
758 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
760 struct gfs2_holder t_gh;
761 int error;
763 flush_workqueue(gfs2_delete_workqueue);
764 gfs2_quota_sync(sdp->sd_vfs, 0, 1);
765 gfs2_statfs_sync(sdp->sd_vfs, 0);
767 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
768 &t_gh);
769 if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
770 return error;
772 gfs2_meta_syncfs(sdp);
773 gfs2_log_shutdown(sdp);
775 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
777 if (t_gh.gh_gl)
778 gfs2_glock_dq_uninit(&t_gh);
780 gfs2_quota_cleanup(sdp);
782 return error;
785 static int gfs2_umount_recovery_wait(void *word)
787 schedule();
788 return 0;
792 * gfs2_put_super - Unmount the filesystem
793 * @sb: The VFS superblock
797 static void gfs2_put_super(struct super_block *sb)
799 struct gfs2_sbd *sdp = sb->s_fs_info;
800 int error;
801 struct gfs2_jdesc *jd;
803 /* Unfreeze the filesystem, if we need to */
805 mutex_lock(&sdp->sd_freeze_lock);
806 if (sdp->sd_freeze_count)
807 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
808 mutex_unlock(&sdp->sd_freeze_lock);
810 /* No more recovery requests */
811 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
812 smp_mb();
814 /* Wait on outstanding recovery */
815 restart:
816 spin_lock(&sdp->sd_jindex_spin);
817 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
818 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
819 continue;
820 spin_unlock(&sdp->sd_jindex_spin);
821 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
822 gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
823 goto restart;
825 spin_unlock(&sdp->sd_jindex_spin);
827 kthread_stop(sdp->sd_quotad_process);
828 kthread_stop(sdp->sd_logd_process);
830 if (!(sb->s_flags & MS_RDONLY)) {
831 error = gfs2_make_fs_ro(sdp);
832 if (error)
833 gfs2_io_error(sdp);
835 /* At this point, we're through modifying the disk */
837 /* Release stuff */
839 iput(sdp->sd_jindex);
840 iput(sdp->sd_statfs_inode);
841 iput(sdp->sd_rindex);
842 iput(sdp->sd_quota_inode);
844 gfs2_glock_put(sdp->sd_rename_gl);
845 gfs2_glock_put(sdp->sd_trans_gl);
847 if (!sdp->sd_args.ar_spectator) {
848 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
849 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
850 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
851 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
852 iput(sdp->sd_sc_inode);
853 iput(sdp->sd_qc_inode);
856 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
857 gfs2_clear_rgrpd(sdp);
858 gfs2_jindex_free(sdp);
859 /* Take apart glock structures and buffer lists */
860 invalidate_inodes(sdp->sd_vfs);
861 gfs2_gl_hash_clear(sdp);
862 /* Unmount the locking protocol */
863 gfs2_lm_unmount(sdp);
865 /* At this point, we're through participating in the lockspace */
866 gfs2_sys_fs_del(sdp);
870 * gfs2_sync_fs - sync the filesystem
871 * @sb: the superblock
873 * Flushes the log to disk.
876 static int gfs2_sync_fs(struct super_block *sb, int wait)
878 if (wait && sb->s_fs_info)
879 gfs2_log_flush(sb->s_fs_info, NULL);
880 return 0;
884 * gfs2_freeze - prevent further writes to the filesystem
885 * @sb: the VFS structure for the filesystem
889 static int gfs2_freeze(struct super_block *sb)
891 struct gfs2_sbd *sdp = sb->s_fs_info;
892 int error;
894 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
895 return -EINVAL;
897 for (;;) {
898 error = gfs2_freeze_fs(sdp);
899 if (!error)
900 break;
902 switch (error) {
903 case -EBUSY:
904 fs_err(sdp, "waiting for recovery before freeze\n");
905 break;
907 default:
908 fs_err(sdp, "error freezing FS: %d\n", error);
909 break;
912 fs_err(sdp, "retrying...\n");
913 msleep(1000);
915 return 0;
919 * gfs2_unfreeze - reallow writes to the filesystem
920 * @sb: the VFS structure for the filesystem
924 static int gfs2_unfreeze(struct super_block *sb)
926 gfs2_unfreeze_fs(sb->s_fs_info);
927 return 0;
931 * statfs_fill - fill in the sg for a given RG
932 * @rgd: the RG
933 * @sc: the sc structure
935 * Returns: 0 on success, -ESTALE if the LVB is invalid
938 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
939 struct gfs2_statfs_change_host *sc)
941 gfs2_rgrp_verify(rgd);
942 sc->sc_total += rgd->rd_data;
943 sc->sc_free += rgd->rd_free;
944 sc->sc_dinodes += rgd->rd_dinodes;
945 return 0;
949 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
950 * @sdp: the filesystem
951 * @sc: the sc info that will be returned
953 * Any error (other than a signal) will cause this routine to fall back
954 * to the synchronous version.
956 * FIXME: This really shouldn't busy wait like this.
958 * Returns: errno
961 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
963 struct gfs2_holder ri_gh;
964 struct gfs2_rgrpd *rgd_next;
965 struct gfs2_holder *gha, *gh;
966 unsigned int slots = 64;
967 unsigned int x;
968 int done;
969 int error = 0, err;
971 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
972 gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
973 if (!gha)
974 return -ENOMEM;
976 error = gfs2_rindex_hold(sdp, &ri_gh);
977 if (error)
978 goto out;
980 rgd_next = gfs2_rgrpd_get_first(sdp);
982 for (;;) {
983 done = 1;
985 for (x = 0; x < slots; x++) {
986 gh = gha + x;
988 if (gh->gh_gl && gfs2_glock_poll(gh)) {
989 err = gfs2_glock_wait(gh);
990 if (err) {
991 gfs2_holder_uninit(gh);
992 error = err;
993 } else {
994 if (!error)
995 error = statfs_slow_fill(
996 gh->gh_gl->gl_object, sc);
997 gfs2_glock_dq_uninit(gh);
1001 if (gh->gh_gl)
1002 done = 0;
1003 else if (rgd_next && !error) {
1004 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1005 LM_ST_SHARED,
1006 GL_ASYNC,
1007 gh);
1008 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1009 done = 0;
1012 if (signal_pending(current))
1013 error = -ERESTARTSYS;
1016 if (done)
1017 break;
1019 yield();
1022 gfs2_glock_dq_uninit(&ri_gh);
1024 out:
1025 kfree(gha);
1026 return error;
1030 * gfs2_statfs_i - Do a statfs
1031 * @sdp: the filesystem
1032 * @sg: the sg structure
1034 * Returns: errno
1037 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1039 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1040 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1042 spin_lock(&sdp->sd_statfs_spin);
1044 *sc = *m_sc;
1045 sc->sc_total += l_sc->sc_total;
1046 sc->sc_free += l_sc->sc_free;
1047 sc->sc_dinodes += l_sc->sc_dinodes;
1049 spin_unlock(&sdp->sd_statfs_spin);
1051 if (sc->sc_free < 0)
1052 sc->sc_free = 0;
1053 if (sc->sc_free > sc->sc_total)
1054 sc->sc_free = sc->sc_total;
1055 if (sc->sc_dinodes < 0)
1056 sc->sc_dinodes = 0;
1058 return 0;
1062 * gfs2_statfs - Gather and return stats about the filesystem
1063 * @sb: The superblock
1064 * @statfsbuf: The buffer
1066 * Returns: 0 on success or error code
1069 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1071 struct super_block *sb = dentry->d_inode->i_sb;
1072 struct gfs2_sbd *sdp = sb->s_fs_info;
1073 struct gfs2_statfs_change_host sc;
1074 int error;
1076 if (gfs2_tune_get(sdp, gt_statfs_slow))
1077 error = gfs2_statfs_slow(sdp, &sc);
1078 else
1079 error = gfs2_statfs_i(sdp, &sc);
1081 if (error)
1082 return error;
1084 buf->f_type = GFS2_MAGIC;
1085 buf->f_bsize = sdp->sd_sb.sb_bsize;
1086 buf->f_blocks = sc.sc_total;
1087 buf->f_bfree = sc.sc_free;
1088 buf->f_bavail = sc.sc_free;
1089 buf->f_files = sc.sc_dinodes + sc.sc_free;
1090 buf->f_ffree = sc.sc_free;
1091 buf->f_namelen = GFS2_FNAMESIZE;
1093 return 0;
1097 * gfs2_remount_fs - called when the FS is remounted
1098 * @sb: the filesystem
1099 * @flags: the remount flags
1100 * @data: extra data passed in (not used right now)
1102 * Returns: errno
1105 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1107 struct gfs2_sbd *sdp = sb->s_fs_info;
1108 struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1109 struct gfs2_tune *gt = &sdp->sd_tune;
1110 int error;
1112 spin_lock(&gt->gt_spin);
1113 args.ar_commit = gt->gt_logd_secs;
1114 args.ar_quota_quantum = gt->gt_quota_quantum;
1115 if (gt->gt_statfs_slow)
1116 args.ar_statfs_quantum = 0;
1117 else
1118 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1119 spin_unlock(&gt->gt_spin);
1120 error = gfs2_mount_args(&args, data);
1121 if (error)
1122 return error;
1124 /* Not allowed to change locking details */
1125 if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1126 strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1127 strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1128 return -EINVAL;
1130 /* Some flags must not be changed */
1131 if (args_neq(&args, &sdp->sd_args, spectator) ||
1132 args_neq(&args, &sdp->sd_args, ignore_local_fs) ||
1133 args_neq(&args, &sdp->sd_args, localflocks) ||
1134 args_neq(&args, &sdp->sd_args, localcaching) ||
1135 args_neq(&args, &sdp->sd_args, meta))
1136 return -EINVAL;
1138 if (sdp->sd_args.ar_spectator)
1139 *flags |= MS_RDONLY;
1141 if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1142 if (*flags & MS_RDONLY)
1143 error = gfs2_make_fs_ro(sdp);
1144 else
1145 error = gfs2_make_fs_rw(sdp);
1146 if (error)
1147 return error;
1150 sdp->sd_args = args;
1151 if (sdp->sd_args.ar_posix_acl)
1152 sb->s_flags |= MS_POSIXACL;
1153 else
1154 sb->s_flags &= ~MS_POSIXACL;
1155 if (sdp->sd_args.ar_nobarrier)
1156 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1157 else
1158 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1159 spin_lock(&gt->gt_spin);
1160 gt->gt_logd_secs = args.ar_commit;
1161 gt->gt_quota_quantum = args.ar_quota_quantum;
1162 if (args.ar_statfs_quantum) {
1163 gt->gt_statfs_slow = 0;
1164 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1166 else {
1167 gt->gt_statfs_slow = 1;
1168 gt->gt_statfs_quantum = 30;
1170 spin_unlock(&gt->gt_spin);
1172 gfs2_online_uevent(sdp);
1173 return 0;
1177 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1178 * @inode: The inode to drop
1180 * If we've received a callback on an iopen lock then its because a
1181 * remote node tried to deallocate the inode but failed due to this node
1182 * still having the inode open. Here we mark the link count zero
1183 * since we know that it must have reached zero if the GLF_DEMOTE flag
1184 * is set on the iopen glock. If we didn't do a disk read since the
1185 * remote node removed the final link then we might otherwise miss
1186 * this event. This check ensures that this node will deallocate the
1187 * inode's blocks, or alternatively pass the baton on to another
1188 * node for later deallocation.
1191 static int gfs2_drop_inode(struct inode *inode)
1193 struct gfs2_inode *ip = GFS2_I(inode);
1195 if (inode->i_nlink) {
1196 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1197 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1198 clear_nlink(inode);
1200 return generic_drop_inode(inode);
1203 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1205 do {
1206 if (d1 == d2)
1207 return 1;
1208 d1 = d1->d_parent;
1209 } while (!IS_ROOT(d1));
1210 return 0;
1214 * gfs2_show_options - Show mount options for /proc/mounts
1215 * @s: seq_file structure
1216 * @mnt: vfsmount
1218 * Returns: 0 on success or error code
1221 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1223 struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
1224 struct gfs2_args *args = &sdp->sd_args;
1225 int val;
1227 if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir))
1228 seq_printf(s, ",meta");
1229 if (args->ar_lockproto[0])
1230 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1231 if (args->ar_locktable[0])
1232 seq_printf(s, ",locktable=%s", args->ar_locktable);
1233 if (args->ar_hostdata[0])
1234 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1235 if (args->ar_spectator)
1236 seq_printf(s, ",spectator");
1237 if (args->ar_ignore_local_fs)
1238 seq_printf(s, ",ignore_local_fs");
1239 if (args->ar_localflocks)
1240 seq_printf(s, ",localflocks");
1241 if (args->ar_localcaching)
1242 seq_printf(s, ",localcaching");
1243 if (args->ar_debug)
1244 seq_printf(s, ",debug");
1245 if (args->ar_upgrade)
1246 seq_printf(s, ",upgrade");
1247 if (args->ar_posix_acl)
1248 seq_printf(s, ",acl");
1249 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1250 char *state;
1251 switch (args->ar_quota) {
1252 case GFS2_QUOTA_OFF:
1253 state = "off";
1254 break;
1255 case GFS2_QUOTA_ACCOUNT:
1256 state = "account";
1257 break;
1258 case GFS2_QUOTA_ON:
1259 state = "on";
1260 break;
1261 default:
1262 state = "unknown";
1263 break;
1265 seq_printf(s, ",quota=%s", state);
1267 if (args->ar_suiddir)
1268 seq_printf(s, ",suiddir");
1269 if (args->ar_data != GFS2_DATA_DEFAULT) {
1270 char *state;
1271 switch (args->ar_data) {
1272 case GFS2_DATA_WRITEBACK:
1273 state = "writeback";
1274 break;
1275 case GFS2_DATA_ORDERED:
1276 state = "ordered";
1277 break;
1278 default:
1279 state = "unknown";
1280 break;
1282 seq_printf(s, ",data=%s", state);
1284 if (args->ar_discard)
1285 seq_printf(s, ",discard");
1286 val = sdp->sd_tune.gt_logd_secs;
1287 if (val != 30)
1288 seq_printf(s, ",commit=%d", val);
1289 val = sdp->sd_tune.gt_statfs_quantum;
1290 if (val != 30)
1291 seq_printf(s, ",statfs_quantum=%d", val);
1292 val = sdp->sd_tune.gt_quota_quantum;
1293 if (val != 60)
1294 seq_printf(s, ",quota_quantum=%d", val);
1295 if (args->ar_statfs_percent)
1296 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1297 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1298 const char *state;
1300 switch (args->ar_errors) {
1301 case GFS2_ERRORS_WITHDRAW:
1302 state = "withdraw";
1303 break;
1304 case GFS2_ERRORS_PANIC:
1305 state = "panic";
1306 break;
1307 default:
1308 state = "unknown";
1309 break;
1311 seq_printf(s, ",errors=%s", state);
1313 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1314 seq_printf(s, ",nobarrier");
1315 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1316 seq_printf(s, ",demote_interface_used");
1317 return 0;
1321 * We have to (at the moment) hold the inodes main lock to cover
1322 * the gap between unlocking the shared lock on the iopen lock and
1323 * taking the exclusive lock. I'd rather do a shared -> exclusive
1324 * conversion on the iopen lock, but we can change that later. This
1325 * is safe, just less efficient.
1328 static void gfs2_evict_inode(struct inode *inode)
1330 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
1331 struct gfs2_inode *ip = GFS2_I(inode);
1332 struct gfs2_holder gh;
1333 int error;
1335 if (inode->i_nlink)
1336 goto out;
1338 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1339 if (unlikely(error)) {
1340 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1341 goto out;
1344 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1345 if (error)
1346 goto out_truncate;
1348 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1349 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1350 error = gfs2_glock_nq(&ip->i_iopen_gh);
1351 if (error)
1352 goto out_truncate;
1354 if (S_ISDIR(inode->i_mode) &&
1355 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1356 error = gfs2_dir_exhash_dealloc(ip);
1357 if (error)
1358 goto out_unlock;
1361 if (ip->i_eattr) {
1362 error = gfs2_ea_dealloc(ip);
1363 if (error)
1364 goto out_unlock;
1367 if (!gfs2_is_stuffed(ip)) {
1368 error = gfs2_file_dealloc(ip);
1369 if (error)
1370 goto out_unlock;
1373 error = gfs2_dinode_dealloc(ip);
1374 if (error)
1375 goto out_unlock;
1377 out_truncate:
1378 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1379 if (error)
1380 goto out_unlock;
1381 /* Needs to be done before glock release & also in a transaction */
1382 truncate_inode_pages(&inode->i_data, 0);
1383 gfs2_trans_end(sdp);
1385 out_unlock:
1386 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
1387 gfs2_glock_dq(&ip->i_iopen_gh);
1388 gfs2_holder_uninit(&ip->i_iopen_gh);
1389 gfs2_glock_dq_uninit(&gh);
1390 if (error && error != GLR_TRYFAILED && error != -EROFS)
1391 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1392 out:
1393 truncate_inode_pages(&inode->i_data, 0);
1394 end_writeback(inode);
1396 ip->i_gl->gl_object = NULL;
1397 gfs2_glock_put(ip->i_gl);
1398 ip->i_gl = NULL;
1399 if (ip->i_iopen_gh.gh_gl) {
1400 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1401 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1405 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1407 struct gfs2_inode *ip;
1409 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1410 if (ip) {
1411 ip->i_flags = 0;
1412 ip->i_gl = NULL;
1414 return &ip->i_inode;
1417 static void gfs2_destroy_inode(struct inode *inode)
1419 kmem_cache_free(gfs2_inode_cachep, inode);
1422 const struct super_operations gfs2_super_ops = {
1423 .alloc_inode = gfs2_alloc_inode,
1424 .destroy_inode = gfs2_destroy_inode,
1425 .write_inode = gfs2_write_inode,
1426 .evict_inode = gfs2_evict_inode,
1427 .put_super = gfs2_put_super,
1428 .sync_fs = gfs2_sync_fs,
1429 .freeze_fs = gfs2_freeze,
1430 .unfreeze_fs = gfs2_unfreeze,
1431 .statfs = gfs2_statfs,
1432 .remount_fs = gfs2_remount_fs,
1433 .drop_inode = gfs2_drop_inode,
1434 .show_options = gfs2_show_options,