Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/bp/bp
[linux-2.6/linux-2.6-openrd.git] / fs / gfs2 / super.c
blobc282ad41f3d1860bba4301495a7303de40b57d06
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>
25 #include "gfs2.h"
26 #include "incore.h"
27 #include "bmap.h"
28 #include "dir.h"
29 #include "glock.h"
30 #include "glops.h"
31 #include "inode.h"
32 #include "log.h"
33 #include "meta_io.h"
34 #include "quota.h"
35 #include "recovery.h"
36 #include "rgrp.h"
37 #include "super.h"
38 #include "trans.h"
39 #include "util.h"
40 #include "sys.h"
41 #include "xattr.h"
43 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
45 enum {
46 Opt_lockproto,
47 Opt_locktable,
48 Opt_hostdata,
49 Opt_spectator,
50 Opt_ignore_local_fs,
51 Opt_localflocks,
52 Opt_localcaching,
53 Opt_debug,
54 Opt_nodebug,
55 Opt_upgrade,
56 Opt_acl,
57 Opt_noacl,
58 Opt_quota_off,
59 Opt_quota_account,
60 Opt_quota_on,
61 Opt_quota,
62 Opt_noquota,
63 Opt_suiddir,
64 Opt_nosuiddir,
65 Opt_data_writeback,
66 Opt_data_ordered,
67 Opt_meta,
68 Opt_discard,
69 Opt_nodiscard,
70 Opt_commit,
71 Opt_err_withdraw,
72 Opt_err_panic,
73 Opt_statfs_quantum,
74 Opt_statfs_percent,
75 Opt_quota_quantum,
76 Opt_barrier,
77 Opt_nobarrier,
78 Opt_error,
81 static const match_table_t tokens = {
82 {Opt_lockproto, "lockproto=%s"},
83 {Opt_locktable, "locktable=%s"},
84 {Opt_hostdata, "hostdata=%s"},
85 {Opt_spectator, "spectator"},
86 {Opt_ignore_local_fs, "ignore_local_fs"},
87 {Opt_localflocks, "localflocks"},
88 {Opt_localcaching, "localcaching"},
89 {Opt_debug, "debug"},
90 {Opt_nodebug, "nodebug"},
91 {Opt_upgrade, "upgrade"},
92 {Opt_acl, "acl"},
93 {Opt_noacl, "noacl"},
94 {Opt_quota_off, "quota=off"},
95 {Opt_quota_account, "quota=account"},
96 {Opt_quota_on, "quota=on"},
97 {Opt_quota, "quota"},
98 {Opt_noquota, "noquota"},
99 {Opt_suiddir, "suiddir"},
100 {Opt_nosuiddir, "nosuiddir"},
101 {Opt_data_writeback, "data=writeback"},
102 {Opt_data_ordered, "data=ordered"},
103 {Opt_meta, "meta"},
104 {Opt_discard, "discard"},
105 {Opt_nodiscard, "nodiscard"},
106 {Opt_commit, "commit=%d"},
107 {Opt_err_withdraw, "errors=withdraw"},
108 {Opt_err_panic, "errors=panic"},
109 {Opt_statfs_quantum, "statfs_quantum=%d"},
110 {Opt_statfs_percent, "statfs_percent=%d"},
111 {Opt_quota_quantum, "quota_quantum=%d"},
112 {Opt_barrier, "barrier"},
113 {Opt_nobarrier, "nobarrier"},
114 {Opt_error, NULL}
118 * gfs2_mount_args - Parse mount options
119 * @args: The structure into which the parsed options will be written
120 * @options: The options to parse
122 * Return: errno
125 int gfs2_mount_args(struct gfs2_args *args, char *options)
127 char *o;
128 int token;
129 substring_t tmp[MAX_OPT_ARGS];
130 int rv;
132 /* Split the options into tokens with the "," character and
133 process them */
135 while (1) {
136 o = strsep(&options, ",");
137 if (o == NULL)
138 break;
139 if (*o == '\0')
140 continue;
142 token = match_token(o, tokens, tmp);
143 switch (token) {
144 case Opt_lockproto:
145 match_strlcpy(args->ar_lockproto, &tmp[0],
146 GFS2_LOCKNAME_LEN);
147 break;
148 case Opt_locktable:
149 match_strlcpy(args->ar_locktable, &tmp[0],
150 GFS2_LOCKNAME_LEN);
151 break;
152 case Opt_hostdata:
153 match_strlcpy(args->ar_hostdata, &tmp[0],
154 GFS2_LOCKNAME_LEN);
155 break;
156 case Opt_spectator:
157 args->ar_spectator = 1;
158 break;
159 case Opt_ignore_local_fs:
160 args->ar_ignore_local_fs = 1;
161 break;
162 case Opt_localflocks:
163 args->ar_localflocks = 1;
164 break;
165 case Opt_localcaching:
166 args->ar_localcaching = 1;
167 break;
168 case Opt_debug:
169 if (args->ar_errors == GFS2_ERRORS_PANIC) {
170 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
171 "are mutually exclusive.\n");
172 return -EINVAL;
174 args->ar_debug = 1;
175 break;
176 case Opt_nodebug:
177 args->ar_debug = 0;
178 break;
179 case Opt_upgrade:
180 args->ar_upgrade = 1;
181 break;
182 case Opt_acl:
183 args->ar_posix_acl = 1;
184 break;
185 case Opt_noacl:
186 args->ar_posix_acl = 0;
187 break;
188 case Opt_quota_off:
189 case Opt_noquota:
190 args->ar_quota = GFS2_QUOTA_OFF;
191 break;
192 case Opt_quota_account:
193 args->ar_quota = GFS2_QUOTA_ACCOUNT;
194 break;
195 case Opt_quota_on:
196 case Opt_quota:
197 args->ar_quota = GFS2_QUOTA_ON;
198 break;
199 case Opt_suiddir:
200 args->ar_suiddir = 1;
201 break;
202 case Opt_nosuiddir:
203 args->ar_suiddir = 0;
204 break;
205 case Opt_data_writeback:
206 args->ar_data = GFS2_DATA_WRITEBACK;
207 break;
208 case Opt_data_ordered:
209 args->ar_data = GFS2_DATA_ORDERED;
210 break;
211 case Opt_meta:
212 args->ar_meta = 1;
213 break;
214 case Opt_discard:
215 args->ar_discard = 1;
216 break;
217 case Opt_nodiscard:
218 args->ar_discard = 0;
219 break;
220 case Opt_commit:
221 rv = match_int(&tmp[0], &args->ar_commit);
222 if (rv || args->ar_commit <= 0) {
223 printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n");
224 return rv ? rv : -EINVAL;
226 break;
227 case Opt_statfs_quantum:
228 rv = match_int(&tmp[0], &args->ar_statfs_quantum);
229 if (rv || args->ar_statfs_quantum < 0) {
230 printk(KERN_WARNING "GFS2: statfs_quantum mount option requires a non-negative numeric argument\n");
231 return rv ? rv : -EINVAL;
233 break;
234 case Opt_quota_quantum:
235 rv = match_int(&tmp[0], &args->ar_quota_quantum);
236 if (rv || args->ar_quota_quantum <= 0) {
237 printk(KERN_WARNING "GFS2: quota_quantum mount option requires a positive numeric argument\n");
238 return rv ? rv : -EINVAL;
240 break;
241 case Opt_statfs_percent:
242 rv = match_int(&tmp[0], &args->ar_statfs_percent);
243 if (rv || args->ar_statfs_percent < 0 ||
244 args->ar_statfs_percent > 100) {
245 printk(KERN_WARNING "statfs_percent mount option requires a numeric argument between 0 and 100\n");
246 return rv ? rv : -EINVAL;
248 break;
249 case Opt_err_withdraw:
250 args->ar_errors = GFS2_ERRORS_WITHDRAW;
251 break;
252 case Opt_err_panic:
253 if (args->ar_debug) {
254 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
255 "are mutually exclusive.\n");
256 return -EINVAL;
258 args->ar_errors = GFS2_ERRORS_PANIC;
259 break;
260 case Opt_barrier:
261 args->ar_nobarrier = 0;
262 break;
263 case Opt_nobarrier:
264 args->ar_nobarrier = 1;
265 break;
266 case Opt_error:
267 default:
268 printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o);
269 return -EINVAL;
273 return 0;
277 * gfs2_jindex_free - Clear all the journal index information
278 * @sdp: The GFS2 superblock
282 void gfs2_jindex_free(struct gfs2_sbd *sdp)
284 struct list_head list, *head;
285 struct gfs2_jdesc *jd;
286 struct gfs2_journal_extent *jext;
288 spin_lock(&sdp->sd_jindex_spin);
289 list_add(&list, &sdp->sd_jindex_list);
290 list_del_init(&sdp->sd_jindex_list);
291 sdp->sd_journals = 0;
292 spin_unlock(&sdp->sd_jindex_spin);
294 while (!list_empty(&list)) {
295 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
296 head = &jd->extent_list;
297 while (!list_empty(head)) {
298 jext = list_entry(head->next,
299 struct gfs2_journal_extent,
300 extent_list);
301 list_del(&jext->extent_list);
302 kfree(jext);
304 list_del(&jd->jd_list);
305 iput(jd->jd_inode);
306 kfree(jd);
310 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
312 struct gfs2_jdesc *jd;
313 int found = 0;
315 list_for_each_entry(jd, head, jd_list) {
316 if (jd->jd_jid == jid) {
317 found = 1;
318 break;
322 if (!found)
323 jd = NULL;
325 return jd;
328 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
330 struct gfs2_jdesc *jd;
332 spin_lock(&sdp->sd_jindex_spin);
333 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
334 spin_unlock(&sdp->sd_jindex_spin);
336 return jd;
339 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
341 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
342 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
343 int ar;
344 int error;
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 error = gfs2_write_alloc_required(ip, 0, ip->i_disksize, &ar);
354 if (!error && ar) {
355 gfs2_consist_inode(ip);
356 error = -EIO;
359 return error;
363 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
364 * @sdp: the filesystem
366 * Returns: errno
369 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
371 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
372 struct gfs2_glock *j_gl = ip->i_gl;
373 struct gfs2_holder t_gh;
374 struct gfs2_log_header_host head;
375 int error;
377 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
378 if (error)
379 return error;
381 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
383 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
384 if (error)
385 goto fail;
387 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
388 gfs2_consist(sdp);
389 error = -EIO;
390 goto fail;
393 /* Initialize some head of the log stuff */
394 sdp->sd_log_sequence = head.lh_sequence + 1;
395 gfs2_log_pointers_init(sdp, head.lh_blkno);
397 error = gfs2_quota_init(sdp);
398 if (error)
399 goto fail;
401 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
403 gfs2_glock_dq_uninit(&t_gh);
405 return 0;
407 fail:
408 t_gh.gh_flags |= GL_NOCACHE;
409 gfs2_glock_dq_uninit(&t_gh);
411 return error;
414 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
416 const struct gfs2_statfs_change *str = buf;
418 sc->sc_total = be64_to_cpu(str->sc_total);
419 sc->sc_free = be64_to_cpu(str->sc_free);
420 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
423 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
425 struct gfs2_statfs_change *str = buf;
427 str->sc_total = cpu_to_be64(sc->sc_total);
428 str->sc_free = cpu_to_be64(sc->sc_free);
429 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
432 int gfs2_statfs_init(struct gfs2_sbd *sdp)
434 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
435 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
436 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
437 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
438 struct buffer_head *m_bh, *l_bh;
439 struct gfs2_holder gh;
440 int error;
442 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
443 &gh);
444 if (error)
445 return error;
447 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
448 if (error)
449 goto out;
451 if (sdp->sd_args.ar_spectator) {
452 spin_lock(&sdp->sd_statfs_spin);
453 gfs2_statfs_change_in(m_sc, m_bh->b_data +
454 sizeof(struct gfs2_dinode));
455 spin_unlock(&sdp->sd_statfs_spin);
456 } else {
457 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
458 if (error)
459 goto out_m_bh;
461 spin_lock(&sdp->sd_statfs_spin);
462 gfs2_statfs_change_in(m_sc, m_bh->b_data +
463 sizeof(struct gfs2_dinode));
464 gfs2_statfs_change_in(l_sc, l_bh->b_data +
465 sizeof(struct gfs2_dinode));
466 spin_unlock(&sdp->sd_statfs_spin);
468 brelse(l_bh);
471 out_m_bh:
472 brelse(m_bh);
473 out:
474 gfs2_glock_dq_uninit(&gh);
475 return 0;
478 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
479 s64 dinodes)
481 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
482 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
483 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
484 struct buffer_head *l_bh;
485 s64 x, y;
486 int need_sync = 0;
487 int error;
489 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
490 if (error)
491 return;
493 gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
495 spin_lock(&sdp->sd_statfs_spin);
496 l_sc->sc_total += total;
497 l_sc->sc_free += free;
498 l_sc->sc_dinodes += dinodes;
499 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
500 if (sdp->sd_args.ar_statfs_percent) {
501 x = 100 * l_sc->sc_free;
502 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
503 if (x >= y || x <= -y)
504 need_sync = 1;
506 spin_unlock(&sdp->sd_statfs_spin);
508 brelse(l_bh);
509 if (need_sync)
510 gfs2_wake_up_statfs(sdp);
513 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
514 struct buffer_head *l_bh)
516 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
517 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
518 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
519 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
521 gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
523 spin_lock(&sdp->sd_statfs_spin);
524 m_sc->sc_total += l_sc->sc_total;
525 m_sc->sc_free += l_sc->sc_free;
526 m_sc->sc_dinodes += l_sc->sc_dinodes;
527 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
528 memset(l_bh->b_data + sizeof(struct gfs2_dinode),
529 0, sizeof(struct gfs2_statfs_change));
530 spin_unlock(&sdp->sd_statfs_spin);
532 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
533 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
536 int gfs2_statfs_sync(struct super_block *sb, int type)
538 struct gfs2_sbd *sdp = sb->s_fs_info;
539 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
540 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
541 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
542 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
543 struct gfs2_holder gh;
544 struct buffer_head *m_bh, *l_bh;
545 int error;
547 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
548 &gh);
549 if (error)
550 return error;
552 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
553 if (error)
554 goto out;
556 spin_lock(&sdp->sd_statfs_spin);
557 gfs2_statfs_change_in(m_sc, m_bh->b_data +
558 sizeof(struct gfs2_dinode));
559 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
560 spin_unlock(&sdp->sd_statfs_spin);
561 goto out_bh;
563 spin_unlock(&sdp->sd_statfs_spin);
565 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
566 if (error)
567 goto out_bh;
569 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
570 if (error)
571 goto out_bh2;
573 update_statfs(sdp, m_bh, l_bh);
574 sdp->sd_statfs_force_sync = 0;
576 gfs2_trans_end(sdp);
578 out_bh2:
579 brelse(l_bh);
580 out_bh:
581 brelse(m_bh);
582 out:
583 gfs2_glock_dq_uninit(&gh);
584 return error;
587 struct lfcc {
588 struct list_head list;
589 struct gfs2_holder gh;
593 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
594 * journals are clean
595 * @sdp: the file system
596 * @state: the state to put the transaction lock into
597 * @t_gh: the hold on the transaction lock
599 * Returns: errno
602 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
603 struct gfs2_holder *t_gh)
605 struct gfs2_inode *ip;
606 struct gfs2_jdesc *jd;
607 struct lfcc *lfcc;
608 LIST_HEAD(list);
609 struct gfs2_log_header_host lh;
610 int error;
612 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
613 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
614 if (!lfcc) {
615 error = -ENOMEM;
616 goto out;
618 ip = GFS2_I(jd->jd_inode);
619 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
620 if (error) {
621 kfree(lfcc);
622 goto out;
624 list_add(&lfcc->list, &list);
627 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
628 GL_NOCACHE, t_gh);
630 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
631 error = gfs2_jdesc_check(jd);
632 if (error)
633 break;
634 error = gfs2_find_jhead(jd, &lh);
635 if (error)
636 break;
637 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
638 error = -EBUSY;
639 break;
643 if (error)
644 gfs2_glock_dq_uninit(t_gh);
646 out:
647 while (!list_empty(&list)) {
648 lfcc = list_entry(list.next, struct lfcc, list);
649 list_del(&lfcc->list);
650 gfs2_glock_dq_uninit(&lfcc->gh);
651 kfree(lfcc);
653 return error;
657 * gfs2_freeze_fs - freezes the file system
658 * @sdp: the file system
660 * This function flushes data and meta data for all machines by
661 * aquiring the transaction log exclusively. All journals are
662 * ensured to be in a clean state as well.
664 * Returns: errno
667 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
669 int error = 0;
671 mutex_lock(&sdp->sd_freeze_lock);
673 if (!sdp->sd_freeze_count++) {
674 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
675 if (error)
676 sdp->sd_freeze_count--;
679 mutex_unlock(&sdp->sd_freeze_lock);
681 return error;
685 * gfs2_unfreeze_fs - unfreezes the file system
686 * @sdp: the file system
688 * This function allows the file system to proceed by unlocking
689 * the exclusively held transaction lock. Other GFS2 nodes are
690 * now free to acquire the lock shared and go on with their lives.
694 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
696 mutex_lock(&sdp->sd_freeze_lock);
698 if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
699 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
701 mutex_unlock(&sdp->sd_freeze_lock);
706 * gfs2_write_inode - Make sure the inode is stable on the disk
707 * @inode: The inode
708 * @sync: synchronous write flag
710 * Returns: errno
713 static int gfs2_write_inode(struct inode *inode, int sync)
715 struct gfs2_inode *ip = GFS2_I(inode);
716 struct gfs2_sbd *sdp = GFS2_SB(inode);
717 struct gfs2_holder gh;
718 struct buffer_head *bh;
719 struct timespec atime;
720 struct gfs2_dinode *di;
721 int ret = 0;
723 /* Check this is a "normal" inode, etc */
724 if (!test_bit(GIF_USER, &ip->i_flags) ||
725 (current->flags & PF_MEMALLOC))
726 return 0;
727 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
728 if (ret)
729 goto do_flush;
730 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
731 if (ret)
732 goto do_unlock;
733 ret = gfs2_meta_inode_buffer(ip, &bh);
734 if (ret == 0) {
735 di = (struct gfs2_dinode *)bh->b_data;
736 atime.tv_sec = be64_to_cpu(di->di_atime);
737 atime.tv_nsec = be32_to_cpu(di->di_atime_nsec);
738 if (timespec_compare(&inode->i_atime, &atime) > 0) {
739 gfs2_trans_add_bh(ip->i_gl, bh, 1);
740 gfs2_dinode_out(ip, bh->b_data);
742 brelse(bh);
744 gfs2_trans_end(sdp);
745 do_unlock:
746 gfs2_glock_dq_uninit(&gh);
747 do_flush:
748 if (sync != 0)
749 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
750 return ret;
754 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
755 * @sdp: the filesystem
757 * Returns: errno
760 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
762 struct gfs2_holder t_gh;
763 int error;
765 flush_workqueue(gfs2_delete_workqueue);
766 gfs2_quota_sync(sdp->sd_vfs, 0);
767 gfs2_statfs_sync(sdp->sd_vfs, 0);
769 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
770 &t_gh);
771 if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
772 return error;
774 gfs2_meta_syncfs(sdp);
775 gfs2_log_shutdown(sdp);
777 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
779 if (t_gh.gh_gl)
780 gfs2_glock_dq_uninit(&t_gh);
782 gfs2_quota_cleanup(sdp);
784 return error;
787 static int gfs2_umount_recovery_wait(void *word)
789 schedule();
790 return 0;
794 * gfs2_put_super - Unmount the filesystem
795 * @sb: The VFS superblock
799 static void gfs2_put_super(struct super_block *sb)
801 struct gfs2_sbd *sdp = sb->s_fs_info;
802 int error;
803 struct gfs2_jdesc *jd;
805 /* Unfreeze the filesystem, if we need to */
807 mutex_lock(&sdp->sd_freeze_lock);
808 if (sdp->sd_freeze_count)
809 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
810 mutex_unlock(&sdp->sd_freeze_lock);
812 /* No more recovery requests */
813 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
814 smp_mb();
816 /* Wait on outstanding recovery */
817 restart:
818 spin_lock(&sdp->sd_jindex_spin);
819 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
820 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
821 continue;
822 spin_unlock(&sdp->sd_jindex_spin);
823 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
824 gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
825 goto restart;
827 spin_unlock(&sdp->sd_jindex_spin);
829 kthread_stop(sdp->sd_quotad_process);
830 kthread_stop(sdp->sd_logd_process);
832 if (!(sb->s_flags & MS_RDONLY)) {
833 error = gfs2_make_fs_ro(sdp);
834 if (error)
835 gfs2_io_error(sdp);
837 /* At this point, we're through modifying the disk */
839 /* Release stuff */
841 iput(sdp->sd_jindex);
842 iput(sdp->sd_statfs_inode);
843 iput(sdp->sd_rindex);
844 iput(sdp->sd_quota_inode);
846 gfs2_glock_put(sdp->sd_rename_gl);
847 gfs2_glock_put(sdp->sd_trans_gl);
849 if (!sdp->sd_args.ar_spectator) {
850 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
851 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
852 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
853 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
854 iput(sdp->sd_sc_inode);
855 iput(sdp->sd_qc_inode);
858 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
859 gfs2_clear_rgrpd(sdp);
860 gfs2_jindex_free(sdp);
861 /* Take apart glock structures and buffer lists */
862 gfs2_gl_hash_clear(sdp);
863 /* Unmount the locking protocol */
864 gfs2_lm_unmount(sdp);
866 /* At this point, we're through participating in the lockspace */
867 gfs2_sys_fs_del(sdp);
871 * gfs2_sync_fs - sync the filesystem
872 * @sb: the superblock
874 * Flushes the log to disk.
877 static int gfs2_sync_fs(struct super_block *sb, int wait)
879 if (wait && sb->s_fs_info)
880 gfs2_log_flush(sb->s_fs_info, NULL);
881 return 0;
885 * gfs2_freeze - prevent further writes to the filesystem
886 * @sb: the VFS structure for the filesystem
890 static int gfs2_freeze(struct super_block *sb)
892 struct gfs2_sbd *sdp = sb->s_fs_info;
893 int error;
895 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
896 return -EINVAL;
898 for (;;) {
899 error = gfs2_freeze_fs(sdp);
900 if (!error)
901 break;
903 switch (error) {
904 case -EBUSY:
905 fs_err(sdp, "waiting for recovery before freeze\n");
906 break;
908 default:
909 fs_err(sdp, "error freezing FS: %d\n", error);
910 break;
913 fs_err(sdp, "retrying...\n");
914 msleep(1000);
916 return 0;
920 * gfs2_unfreeze - reallow writes to the filesystem
921 * @sb: the VFS structure for the filesystem
925 static int gfs2_unfreeze(struct super_block *sb)
927 gfs2_unfreeze_fs(sb->s_fs_info);
928 return 0;
932 * statfs_fill - fill in the sg for a given RG
933 * @rgd: the RG
934 * @sc: the sc structure
936 * Returns: 0 on success, -ESTALE if the LVB is invalid
939 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
940 struct gfs2_statfs_change_host *sc)
942 gfs2_rgrp_verify(rgd);
943 sc->sc_total += rgd->rd_data;
944 sc->sc_free += rgd->rd_free;
945 sc->sc_dinodes += rgd->rd_dinodes;
946 return 0;
950 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
951 * @sdp: the filesystem
952 * @sc: the sc info that will be returned
954 * Any error (other than a signal) will cause this routine to fall back
955 * to the synchronous version.
957 * FIXME: This really shouldn't busy wait like this.
959 * Returns: errno
962 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
964 struct gfs2_holder ri_gh;
965 struct gfs2_rgrpd *rgd_next;
966 struct gfs2_holder *gha, *gh;
967 unsigned int slots = 64;
968 unsigned int x;
969 int done;
970 int error = 0, err;
972 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
973 gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
974 if (!gha)
975 return -ENOMEM;
977 error = gfs2_rindex_hold(sdp, &ri_gh);
978 if (error)
979 goto out;
981 rgd_next = gfs2_rgrpd_get_first(sdp);
983 for (;;) {
984 done = 1;
986 for (x = 0; x < slots; x++) {
987 gh = gha + x;
989 if (gh->gh_gl && gfs2_glock_poll(gh)) {
990 err = gfs2_glock_wait(gh);
991 if (err) {
992 gfs2_holder_uninit(gh);
993 error = err;
994 } else {
995 if (!error)
996 error = statfs_slow_fill(
997 gh->gh_gl->gl_object, sc);
998 gfs2_glock_dq_uninit(gh);
1002 if (gh->gh_gl)
1003 done = 0;
1004 else if (rgd_next && !error) {
1005 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1006 LM_ST_SHARED,
1007 GL_ASYNC,
1008 gh);
1009 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1010 done = 0;
1013 if (signal_pending(current))
1014 error = -ERESTARTSYS;
1017 if (done)
1018 break;
1020 yield();
1023 gfs2_glock_dq_uninit(&ri_gh);
1025 out:
1026 kfree(gha);
1027 return error;
1031 * gfs2_statfs_i - Do a statfs
1032 * @sdp: the filesystem
1033 * @sg: the sg structure
1035 * Returns: errno
1038 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1040 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1041 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1043 spin_lock(&sdp->sd_statfs_spin);
1045 *sc = *m_sc;
1046 sc->sc_total += l_sc->sc_total;
1047 sc->sc_free += l_sc->sc_free;
1048 sc->sc_dinodes += l_sc->sc_dinodes;
1050 spin_unlock(&sdp->sd_statfs_spin);
1052 if (sc->sc_free < 0)
1053 sc->sc_free = 0;
1054 if (sc->sc_free > sc->sc_total)
1055 sc->sc_free = sc->sc_total;
1056 if (sc->sc_dinodes < 0)
1057 sc->sc_dinodes = 0;
1059 return 0;
1063 * gfs2_statfs - Gather and return stats about the filesystem
1064 * @sb: The superblock
1065 * @statfsbuf: The buffer
1067 * Returns: 0 on success or error code
1070 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1072 struct super_block *sb = dentry->d_inode->i_sb;
1073 struct gfs2_sbd *sdp = sb->s_fs_info;
1074 struct gfs2_statfs_change_host sc;
1075 int error;
1077 if (gfs2_tune_get(sdp, gt_statfs_slow))
1078 error = gfs2_statfs_slow(sdp, &sc);
1079 else
1080 error = gfs2_statfs_i(sdp, &sc);
1082 if (error)
1083 return error;
1085 buf->f_type = GFS2_MAGIC;
1086 buf->f_bsize = sdp->sd_sb.sb_bsize;
1087 buf->f_blocks = sc.sc_total;
1088 buf->f_bfree = sc.sc_free;
1089 buf->f_bavail = sc.sc_free;
1090 buf->f_files = sc.sc_dinodes + sc.sc_free;
1091 buf->f_ffree = sc.sc_free;
1092 buf->f_namelen = GFS2_FNAMESIZE;
1094 return 0;
1098 * gfs2_remount_fs - called when the FS is remounted
1099 * @sb: the filesystem
1100 * @flags: the remount flags
1101 * @data: extra data passed in (not used right now)
1103 * Returns: errno
1106 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1108 struct gfs2_sbd *sdp = sb->s_fs_info;
1109 struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1110 struct gfs2_tune *gt = &sdp->sd_tune;
1111 int error;
1113 spin_lock(&gt->gt_spin);
1114 args.ar_commit = gt->gt_log_flush_secs;
1115 args.ar_quota_quantum = gt->gt_quota_quantum;
1116 if (gt->gt_statfs_slow)
1117 args.ar_statfs_quantum = 0;
1118 else
1119 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1120 spin_unlock(&gt->gt_spin);
1121 error = gfs2_mount_args(&args, data);
1122 if (error)
1123 return error;
1125 /* Not allowed to change locking details */
1126 if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1127 strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1128 strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1129 return -EINVAL;
1131 /* Some flags must not be changed */
1132 if (args_neq(&args, &sdp->sd_args, spectator) ||
1133 args_neq(&args, &sdp->sd_args, ignore_local_fs) ||
1134 args_neq(&args, &sdp->sd_args, localflocks) ||
1135 args_neq(&args, &sdp->sd_args, localcaching) ||
1136 args_neq(&args, &sdp->sd_args, meta))
1137 return -EINVAL;
1139 if (sdp->sd_args.ar_spectator)
1140 *flags |= MS_RDONLY;
1142 if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1143 if (*flags & MS_RDONLY)
1144 error = gfs2_make_fs_ro(sdp);
1145 else
1146 error = gfs2_make_fs_rw(sdp);
1147 if (error)
1148 return error;
1151 sdp->sd_args = args;
1152 if (sdp->sd_args.ar_posix_acl)
1153 sb->s_flags |= MS_POSIXACL;
1154 else
1155 sb->s_flags &= ~MS_POSIXACL;
1156 if (sdp->sd_args.ar_nobarrier)
1157 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1158 else
1159 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1160 spin_lock(&gt->gt_spin);
1161 gt->gt_log_flush_secs = args.ar_commit;
1162 gt->gt_quota_quantum = args.ar_quota_quantum;
1163 if (args.ar_statfs_quantum) {
1164 gt->gt_statfs_slow = 0;
1165 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1167 else {
1168 gt->gt_statfs_slow = 1;
1169 gt->gt_statfs_quantum = 30;
1171 spin_unlock(&gt->gt_spin);
1173 gfs2_online_uevent(sdp);
1174 return 0;
1178 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1179 * @inode: The inode to drop
1181 * If we've received a callback on an iopen lock then its because a
1182 * remote node tried to deallocate the inode but failed due to this node
1183 * still having the inode open. Here we mark the link count zero
1184 * since we know that it must have reached zero if the GLF_DEMOTE flag
1185 * is set on the iopen glock. If we didn't do a disk read since the
1186 * remote node removed the final link then we might otherwise miss
1187 * this event. This check ensures that this node will deallocate the
1188 * inode's blocks, or alternatively pass the baton on to another
1189 * node for later deallocation.
1192 static void gfs2_drop_inode(struct inode *inode)
1194 struct gfs2_inode *ip = GFS2_I(inode);
1196 if (test_bit(GIF_USER, &ip->i_flags) && inode->i_nlink) {
1197 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1198 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1199 clear_nlink(inode);
1201 generic_drop_inode(inode);
1205 * gfs2_clear_inode - Deallocate an inode when VFS is done with it
1206 * @inode: The VFS inode
1210 static void gfs2_clear_inode(struct inode *inode)
1212 struct gfs2_inode *ip = GFS2_I(inode);
1214 /* This tells us its a "real" inode and not one which only
1215 * serves to contain an address space (see rgrp.c, meta_io.c)
1216 * which therefore doesn't have its own glocks.
1218 if (test_bit(GIF_USER, &ip->i_flags)) {
1219 ip->i_gl->gl_object = NULL;
1220 gfs2_glock_put(ip->i_gl);
1221 ip->i_gl = NULL;
1222 if (ip->i_iopen_gh.gh_gl) {
1223 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1224 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1229 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1231 do {
1232 if (d1 == d2)
1233 return 1;
1234 d1 = d1->d_parent;
1235 } while (!IS_ROOT(d1));
1236 return 0;
1240 * gfs2_show_options - Show mount options for /proc/mounts
1241 * @s: seq_file structure
1242 * @mnt: vfsmount
1244 * Returns: 0 on success or error code
1247 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1249 struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
1250 struct gfs2_args *args = &sdp->sd_args;
1251 int val;
1253 if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir))
1254 seq_printf(s, ",meta");
1255 if (args->ar_lockproto[0])
1256 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1257 if (args->ar_locktable[0])
1258 seq_printf(s, ",locktable=%s", args->ar_locktable);
1259 if (args->ar_hostdata[0])
1260 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1261 if (args->ar_spectator)
1262 seq_printf(s, ",spectator");
1263 if (args->ar_ignore_local_fs)
1264 seq_printf(s, ",ignore_local_fs");
1265 if (args->ar_localflocks)
1266 seq_printf(s, ",localflocks");
1267 if (args->ar_localcaching)
1268 seq_printf(s, ",localcaching");
1269 if (args->ar_debug)
1270 seq_printf(s, ",debug");
1271 if (args->ar_upgrade)
1272 seq_printf(s, ",upgrade");
1273 if (args->ar_posix_acl)
1274 seq_printf(s, ",acl");
1275 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1276 char *state;
1277 switch (args->ar_quota) {
1278 case GFS2_QUOTA_OFF:
1279 state = "off";
1280 break;
1281 case GFS2_QUOTA_ACCOUNT:
1282 state = "account";
1283 break;
1284 case GFS2_QUOTA_ON:
1285 state = "on";
1286 break;
1287 default:
1288 state = "unknown";
1289 break;
1291 seq_printf(s, ",quota=%s", state);
1293 if (args->ar_suiddir)
1294 seq_printf(s, ",suiddir");
1295 if (args->ar_data != GFS2_DATA_DEFAULT) {
1296 char *state;
1297 switch (args->ar_data) {
1298 case GFS2_DATA_WRITEBACK:
1299 state = "writeback";
1300 break;
1301 case GFS2_DATA_ORDERED:
1302 state = "ordered";
1303 break;
1304 default:
1305 state = "unknown";
1306 break;
1308 seq_printf(s, ",data=%s", state);
1310 if (args->ar_discard)
1311 seq_printf(s, ",discard");
1312 val = sdp->sd_tune.gt_log_flush_secs;
1313 if (val != 60)
1314 seq_printf(s, ",commit=%d", val);
1315 val = sdp->sd_tune.gt_statfs_quantum;
1316 if (val != 30)
1317 seq_printf(s, ",statfs_quantum=%d", val);
1318 val = sdp->sd_tune.gt_quota_quantum;
1319 if (val != 60)
1320 seq_printf(s, ",quota_quantum=%d", val);
1321 if (args->ar_statfs_percent)
1322 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1323 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1324 const char *state;
1326 switch (args->ar_errors) {
1327 case GFS2_ERRORS_WITHDRAW:
1328 state = "withdraw";
1329 break;
1330 case GFS2_ERRORS_PANIC:
1331 state = "panic";
1332 break;
1333 default:
1334 state = "unknown";
1335 break;
1337 seq_printf(s, ",errors=%s", state);
1339 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1340 seq_printf(s, ",nobarrier");
1342 return 0;
1346 * We have to (at the moment) hold the inodes main lock to cover
1347 * the gap between unlocking the shared lock on the iopen lock and
1348 * taking the exclusive lock. I'd rather do a shared -> exclusive
1349 * conversion on the iopen lock, but we can change that later. This
1350 * is safe, just less efficient.
1353 static void gfs2_delete_inode(struct inode *inode)
1355 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
1356 struct gfs2_inode *ip = GFS2_I(inode);
1357 struct gfs2_holder gh;
1358 int error;
1360 if (!test_bit(GIF_USER, &ip->i_flags))
1361 goto out;
1363 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1364 if (unlikely(error)) {
1365 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1366 goto out;
1369 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1370 if (error)
1371 goto out_truncate;
1373 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1374 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1375 error = gfs2_glock_nq(&ip->i_iopen_gh);
1376 if (error)
1377 goto out_truncate;
1379 if (S_ISDIR(inode->i_mode) &&
1380 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1381 error = gfs2_dir_exhash_dealloc(ip);
1382 if (error)
1383 goto out_unlock;
1386 if (ip->i_eattr) {
1387 error = gfs2_ea_dealloc(ip);
1388 if (error)
1389 goto out_unlock;
1392 if (!gfs2_is_stuffed(ip)) {
1393 error = gfs2_file_dealloc(ip);
1394 if (error)
1395 goto out_unlock;
1398 error = gfs2_dinode_dealloc(ip);
1399 if (error)
1400 goto out_unlock;
1402 out_truncate:
1403 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1404 if (error)
1405 goto out_unlock;
1406 /* Needs to be done before glock release & also in a transaction */
1407 truncate_inode_pages(&inode->i_data, 0);
1408 gfs2_trans_end(sdp);
1410 out_unlock:
1411 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
1412 gfs2_glock_dq(&ip->i_iopen_gh);
1413 gfs2_holder_uninit(&ip->i_iopen_gh);
1414 gfs2_glock_dq_uninit(&gh);
1415 if (error && error != GLR_TRYFAILED && error != -EROFS)
1416 fs_warn(sdp, "gfs2_delete_inode: %d\n", error);
1417 out:
1418 truncate_inode_pages(&inode->i_data, 0);
1419 clear_inode(inode);
1422 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1424 struct gfs2_inode *ip;
1426 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1427 if (ip) {
1428 ip->i_flags = 0;
1429 ip->i_gl = NULL;
1431 return &ip->i_inode;
1434 static void gfs2_destroy_inode(struct inode *inode)
1436 kmem_cache_free(gfs2_inode_cachep, inode);
1439 const struct super_operations gfs2_super_ops = {
1440 .alloc_inode = gfs2_alloc_inode,
1441 .destroy_inode = gfs2_destroy_inode,
1442 .write_inode = gfs2_write_inode,
1443 .delete_inode = gfs2_delete_inode,
1444 .put_super = gfs2_put_super,
1445 .sync_fs = gfs2_sync_fs,
1446 .freeze_fs = gfs2_freeze,
1447 .unfreeze_fs = gfs2_unfreeze,
1448 .statfs = gfs2_statfs,
1449 .remount_fs = gfs2_remount_fs,
1450 .clear_inode = gfs2_clear_inode,
1451 .drop_inode = gfs2_drop_inode,
1452 .show_options = gfs2_show_options,