Merge branch 'v6v7' into devel
[linux/fpc-iii.git] / fs / gfs2 / super.c
blobec73ed70bae1afc669b69e1cdbc162d93acec7d8
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_spectator, "norecovery"},
89 {Opt_ignore_local_fs, "ignore_local_fs"},
90 {Opt_localflocks, "localflocks"},
91 {Opt_localcaching, "localcaching"},
92 {Opt_debug, "debug"},
93 {Opt_nodebug, "nodebug"},
94 {Opt_upgrade, "upgrade"},
95 {Opt_acl, "acl"},
96 {Opt_noacl, "noacl"},
97 {Opt_quota_off, "quota=off"},
98 {Opt_quota_account, "quota=account"},
99 {Opt_quota_on, "quota=on"},
100 {Opt_quota, "quota"},
101 {Opt_noquota, "noquota"},
102 {Opt_suiddir, "suiddir"},
103 {Opt_nosuiddir, "nosuiddir"},
104 {Opt_data_writeback, "data=writeback"},
105 {Opt_data_ordered, "data=ordered"},
106 {Opt_meta, "meta"},
107 {Opt_discard, "discard"},
108 {Opt_nodiscard, "nodiscard"},
109 {Opt_commit, "commit=%d"},
110 {Opt_err_withdraw, "errors=withdraw"},
111 {Opt_err_panic, "errors=panic"},
112 {Opt_statfs_quantum, "statfs_quantum=%d"},
113 {Opt_statfs_percent, "statfs_percent=%d"},
114 {Opt_quota_quantum, "quota_quantum=%d"},
115 {Opt_barrier, "barrier"},
116 {Opt_nobarrier, "nobarrier"},
117 {Opt_error, NULL}
121 * gfs2_mount_args - Parse mount options
122 * @args: The structure into which the parsed options will be written
123 * @options: The options to parse
125 * Return: errno
128 int gfs2_mount_args(struct gfs2_args *args, char *options)
130 char *o;
131 int token;
132 substring_t tmp[MAX_OPT_ARGS];
133 int rv;
135 /* Split the options into tokens with the "," character and
136 process them */
138 while (1) {
139 o = strsep(&options, ",");
140 if (o == NULL)
141 break;
142 if (*o == '\0')
143 continue;
145 token = match_token(o, tokens, tmp);
146 switch (token) {
147 case Opt_lockproto:
148 match_strlcpy(args->ar_lockproto, &tmp[0],
149 GFS2_LOCKNAME_LEN);
150 break;
151 case Opt_locktable:
152 match_strlcpy(args->ar_locktable, &tmp[0],
153 GFS2_LOCKNAME_LEN);
154 break;
155 case Opt_hostdata:
156 match_strlcpy(args->ar_hostdata, &tmp[0],
157 GFS2_LOCKNAME_LEN);
158 break;
159 case Opt_spectator:
160 args->ar_spectator = 1;
161 break;
162 case Opt_ignore_local_fs:
163 /* Retained for backwards compat only */
164 break;
165 case Opt_localflocks:
166 args->ar_localflocks = 1;
167 break;
168 case Opt_localcaching:
169 /* Retained for backwards compat only */
170 break;
171 case Opt_debug:
172 if (args->ar_errors == GFS2_ERRORS_PANIC) {
173 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
174 "are mutually exclusive.\n");
175 return -EINVAL;
177 args->ar_debug = 1;
178 break;
179 case Opt_nodebug:
180 args->ar_debug = 0;
181 break;
182 case Opt_upgrade:
183 /* Retained for backwards compat only */
184 break;
185 case Opt_acl:
186 args->ar_posix_acl = 1;
187 break;
188 case Opt_noacl:
189 args->ar_posix_acl = 0;
190 break;
191 case Opt_quota_off:
192 case Opt_noquota:
193 args->ar_quota = GFS2_QUOTA_OFF;
194 break;
195 case Opt_quota_account:
196 args->ar_quota = GFS2_QUOTA_ACCOUNT;
197 break;
198 case Opt_quota_on:
199 case Opt_quota:
200 args->ar_quota = GFS2_QUOTA_ON;
201 break;
202 case Opt_suiddir:
203 args->ar_suiddir = 1;
204 break;
205 case Opt_nosuiddir:
206 args->ar_suiddir = 0;
207 break;
208 case Opt_data_writeback:
209 args->ar_data = GFS2_DATA_WRITEBACK;
210 break;
211 case Opt_data_ordered:
212 args->ar_data = GFS2_DATA_ORDERED;
213 break;
214 case Opt_meta:
215 args->ar_meta = 1;
216 break;
217 case Opt_discard:
218 args->ar_discard = 1;
219 break;
220 case Opt_nodiscard:
221 args->ar_discard = 0;
222 break;
223 case Opt_commit:
224 rv = match_int(&tmp[0], &args->ar_commit);
225 if (rv || args->ar_commit <= 0) {
226 printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n");
227 return rv ? rv : -EINVAL;
229 break;
230 case Opt_statfs_quantum:
231 rv = match_int(&tmp[0], &args->ar_statfs_quantum);
232 if (rv || args->ar_statfs_quantum < 0) {
233 printk(KERN_WARNING "GFS2: statfs_quantum mount option requires a non-negative numeric argument\n");
234 return rv ? rv : -EINVAL;
236 break;
237 case Opt_quota_quantum:
238 rv = match_int(&tmp[0], &args->ar_quota_quantum);
239 if (rv || args->ar_quota_quantum <= 0) {
240 printk(KERN_WARNING "GFS2: quota_quantum mount option requires a positive numeric argument\n");
241 return rv ? rv : -EINVAL;
243 break;
244 case Opt_statfs_percent:
245 rv = match_int(&tmp[0], &args->ar_statfs_percent);
246 if (rv || args->ar_statfs_percent < 0 ||
247 args->ar_statfs_percent > 100) {
248 printk(KERN_WARNING "statfs_percent mount option requires a numeric argument between 0 and 100\n");
249 return rv ? rv : -EINVAL;
251 break;
252 case Opt_err_withdraw:
253 args->ar_errors = GFS2_ERRORS_WITHDRAW;
254 break;
255 case Opt_err_panic:
256 if (args->ar_debug) {
257 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
258 "are mutually exclusive.\n");
259 return -EINVAL;
261 args->ar_errors = GFS2_ERRORS_PANIC;
262 break;
263 case Opt_barrier:
264 args->ar_nobarrier = 0;
265 break;
266 case Opt_nobarrier:
267 args->ar_nobarrier = 1;
268 break;
269 case Opt_error:
270 default:
271 printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o);
272 return -EINVAL;
276 return 0;
280 * gfs2_jindex_free - Clear all the journal index information
281 * @sdp: The GFS2 superblock
285 void gfs2_jindex_free(struct gfs2_sbd *sdp)
287 struct list_head list, *head;
288 struct gfs2_jdesc *jd;
289 struct gfs2_journal_extent *jext;
291 spin_lock(&sdp->sd_jindex_spin);
292 list_add(&list, &sdp->sd_jindex_list);
293 list_del_init(&sdp->sd_jindex_list);
294 sdp->sd_journals = 0;
295 spin_unlock(&sdp->sd_jindex_spin);
297 while (!list_empty(&list)) {
298 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
299 head = &jd->extent_list;
300 while (!list_empty(head)) {
301 jext = list_entry(head->next,
302 struct gfs2_journal_extent,
303 extent_list);
304 list_del(&jext->extent_list);
305 kfree(jext);
307 list_del(&jd->jd_list);
308 iput(jd->jd_inode);
309 kfree(jd);
313 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
315 struct gfs2_jdesc *jd;
316 int found = 0;
318 list_for_each_entry(jd, head, jd_list) {
319 if (jd->jd_jid == jid) {
320 found = 1;
321 break;
325 if (!found)
326 jd = NULL;
328 return jd;
331 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
333 struct gfs2_jdesc *jd;
335 spin_lock(&sdp->sd_jindex_spin);
336 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
337 spin_unlock(&sdp->sd_jindex_spin);
339 return jd;
342 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
344 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
345 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
346 u64 size = i_size_read(jd->jd_inode);
348 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, 1 << 30))
349 return -EIO;
351 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
353 if (gfs2_write_alloc_required(ip, 0, size)) {
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 gfs2_gl_hash_clear(sdp);
861 /* Unmount the locking protocol */
862 gfs2_lm_unmount(sdp);
864 /* At this point, we're through participating in the lockspace */
865 gfs2_sys_fs_del(sdp);
869 * gfs2_sync_fs - sync the filesystem
870 * @sb: the superblock
872 * Flushes the log to disk.
875 static int gfs2_sync_fs(struct super_block *sb, int wait)
877 if (wait && sb->s_fs_info)
878 gfs2_log_flush(sb->s_fs_info, NULL);
879 return 0;
883 * gfs2_freeze - prevent further writes to the filesystem
884 * @sb: the VFS structure for the filesystem
888 static int gfs2_freeze(struct super_block *sb)
890 struct gfs2_sbd *sdp = sb->s_fs_info;
891 int error;
893 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
894 return -EINVAL;
896 for (;;) {
897 error = gfs2_freeze_fs(sdp);
898 if (!error)
899 break;
901 switch (error) {
902 case -EBUSY:
903 fs_err(sdp, "waiting for recovery before freeze\n");
904 break;
906 default:
907 fs_err(sdp, "error freezing FS: %d\n", error);
908 break;
911 fs_err(sdp, "retrying...\n");
912 msleep(1000);
914 return 0;
918 * gfs2_unfreeze - reallow writes to the filesystem
919 * @sb: the VFS structure for the filesystem
923 static int gfs2_unfreeze(struct super_block *sb)
925 gfs2_unfreeze_fs(sb->s_fs_info);
926 return 0;
930 * statfs_fill - fill in the sg for a given RG
931 * @rgd: the RG
932 * @sc: the sc structure
934 * Returns: 0 on success, -ESTALE if the LVB is invalid
937 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
938 struct gfs2_statfs_change_host *sc)
940 gfs2_rgrp_verify(rgd);
941 sc->sc_total += rgd->rd_data;
942 sc->sc_free += rgd->rd_free;
943 sc->sc_dinodes += rgd->rd_dinodes;
944 return 0;
948 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
949 * @sdp: the filesystem
950 * @sc: the sc info that will be returned
952 * Any error (other than a signal) will cause this routine to fall back
953 * to the synchronous version.
955 * FIXME: This really shouldn't busy wait like this.
957 * Returns: errno
960 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
962 struct gfs2_holder ri_gh;
963 struct gfs2_rgrpd *rgd_next;
964 struct gfs2_holder *gha, *gh;
965 unsigned int slots = 64;
966 unsigned int x;
967 int done;
968 int error = 0, err;
970 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
971 gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
972 if (!gha)
973 return -ENOMEM;
975 error = gfs2_rindex_hold(sdp, &ri_gh);
976 if (error)
977 goto out;
979 rgd_next = gfs2_rgrpd_get_first(sdp);
981 for (;;) {
982 done = 1;
984 for (x = 0; x < slots; x++) {
985 gh = gha + x;
987 if (gh->gh_gl && gfs2_glock_poll(gh)) {
988 err = gfs2_glock_wait(gh);
989 if (err) {
990 gfs2_holder_uninit(gh);
991 error = err;
992 } else {
993 if (!error)
994 error = statfs_slow_fill(
995 gh->gh_gl->gl_object, sc);
996 gfs2_glock_dq_uninit(gh);
1000 if (gh->gh_gl)
1001 done = 0;
1002 else if (rgd_next && !error) {
1003 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1004 LM_ST_SHARED,
1005 GL_ASYNC,
1006 gh);
1007 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1008 done = 0;
1011 if (signal_pending(current))
1012 error = -ERESTARTSYS;
1015 if (done)
1016 break;
1018 yield();
1021 gfs2_glock_dq_uninit(&ri_gh);
1023 out:
1024 kfree(gha);
1025 return error;
1029 * gfs2_statfs_i - Do a statfs
1030 * @sdp: the filesystem
1031 * @sg: the sg structure
1033 * Returns: errno
1036 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1038 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1039 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1041 spin_lock(&sdp->sd_statfs_spin);
1043 *sc = *m_sc;
1044 sc->sc_total += l_sc->sc_total;
1045 sc->sc_free += l_sc->sc_free;
1046 sc->sc_dinodes += l_sc->sc_dinodes;
1048 spin_unlock(&sdp->sd_statfs_spin);
1050 if (sc->sc_free < 0)
1051 sc->sc_free = 0;
1052 if (sc->sc_free > sc->sc_total)
1053 sc->sc_free = sc->sc_total;
1054 if (sc->sc_dinodes < 0)
1055 sc->sc_dinodes = 0;
1057 return 0;
1061 * gfs2_statfs - Gather and return stats about the filesystem
1062 * @sb: The superblock
1063 * @statfsbuf: The buffer
1065 * Returns: 0 on success or error code
1068 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1070 struct super_block *sb = dentry->d_inode->i_sb;
1071 struct gfs2_sbd *sdp = sb->s_fs_info;
1072 struct gfs2_statfs_change_host sc;
1073 int error;
1075 if (gfs2_tune_get(sdp, gt_statfs_slow))
1076 error = gfs2_statfs_slow(sdp, &sc);
1077 else
1078 error = gfs2_statfs_i(sdp, &sc);
1080 if (error)
1081 return error;
1083 buf->f_type = GFS2_MAGIC;
1084 buf->f_bsize = sdp->sd_sb.sb_bsize;
1085 buf->f_blocks = sc.sc_total;
1086 buf->f_bfree = sc.sc_free;
1087 buf->f_bavail = sc.sc_free;
1088 buf->f_files = sc.sc_dinodes + sc.sc_free;
1089 buf->f_ffree = sc.sc_free;
1090 buf->f_namelen = GFS2_FNAMESIZE;
1092 return 0;
1096 * gfs2_remount_fs - called when the FS is remounted
1097 * @sb: the filesystem
1098 * @flags: the remount flags
1099 * @data: extra data passed in (not used right now)
1101 * Returns: errno
1104 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1106 struct gfs2_sbd *sdp = sb->s_fs_info;
1107 struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1108 struct gfs2_tune *gt = &sdp->sd_tune;
1109 int error;
1111 spin_lock(&gt->gt_spin);
1112 args.ar_commit = gt->gt_logd_secs;
1113 args.ar_quota_quantum = gt->gt_quota_quantum;
1114 if (gt->gt_statfs_slow)
1115 args.ar_statfs_quantum = 0;
1116 else
1117 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1118 spin_unlock(&gt->gt_spin);
1119 error = gfs2_mount_args(&args, data);
1120 if (error)
1121 return error;
1123 /* Not allowed to change locking details */
1124 if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1125 strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1126 strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1127 return -EINVAL;
1129 /* Some flags must not be changed */
1130 if (args_neq(&args, &sdp->sd_args, spectator) ||
1131 args_neq(&args, &sdp->sd_args, localflocks) ||
1132 args_neq(&args, &sdp->sd_args, meta))
1133 return -EINVAL;
1135 if (sdp->sd_args.ar_spectator)
1136 *flags |= MS_RDONLY;
1138 if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1139 if (*flags & MS_RDONLY)
1140 error = gfs2_make_fs_ro(sdp);
1141 else
1142 error = gfs2_make_fs_rw(sdp);
1143 if (error)
1144 return error;
1147 sdp->sd_args = args;
1148 if (sdp->sd_args.ar_posix_acl)
1149 sb->s_flags |= MS_POSIXACL;
1150 else
1151 sb->s_flags &= ~MS_POSIXACL;
1152 if (sdp->sd_args.ar_nobarrier)
1153 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1154 else
1155 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1156 spin_lock(&gt->gt_spin);
1157 gt->gt_logd_secs = args.ar_commit;
1158 gt->gt_quota_quantum = args.ar_quota_quantum;
1159 if (args.ar_statfs_quantum) {
1160 gt->gt_statfs_slow = 0;
1161 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1163 else {
1164 gt->gt_statfs_slow = 1;
1165 gt->gt_statfs_quantum = 30;
1167 spin_unlock(&gt->gt_spin);
1169 gfs2_online_uevent(sdp);
1170 return 0;
1174 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1175 * @inode: The inode to drop
1177 * If we've received a callback on an iopen lock then its because a
1178 * remote node tried to deallocate the inode but failed due to this node
1179 * still having the inode open. Here we mark the link count zero
1180 * since we know that it must have reached zero if the GLF_DEMOTE flag
1181 * is set on the iopen glock. If we didn't do a disk read since the
1182 * remote node removed the final link then we might otherwise miss
1183 * this event. This check ensures that this node will deallocate the
1184 * inode's blocks, or alternatively pass the baton on to another
1185 * node for later deallocation.
1188 static int gfs2_drop_inode(struct inode *inode)
1190 struct gfs2_inode *ip = GFS2_I(inode);
1192 if (inode->i_nlink) {
1193 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1194 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1195 clear_nlink(inode);
1197 return generic_drop_inode(inode);
1200 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1202 do {
1203 if (d1 == d2)
1204 return 1;
1205 d1 = d1->d_parent;
1206 } while (!IS_ROOT(d1));
1207 return 0;
1211 * gfs2_show_options - Show mount options for /proc/mounts
1212 * @s: seq_file structure
1213 * @mnt: vfsmount
1215 * Returns: 0 on success or error code
1218 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1220 struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
1221 struct gfs2_args *args = &sdp->sd_args;
1222 int val;
1224 if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir))
1225 seq_printf(s, ",meta");
1226 if (args->ar_lockproto[0])
1227 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1228 if (args->ar_locktable[0])
1229 seq_printf(s, ",locktable=%s", args->ar_locktable);
1230 if (args->ar_hostdata[0])
1231 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1232 if (args->ar_spectator)
1233 seq_printf(s, ",spectator");
1234 if (args->ar_localflocks)
1235 seq_printf(s, ",localflocks");
1236 if (args->ar_debug)
1237 seq_printf(s, ",debug");
1238 if (args->ar_posix_acl)
1239 seq_printf(s, ",acl");
1240 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1241 char *state;
1242 switch (args->ar_quota) {
1243 case GFS2_QUOTA_OFF:
1244 state = "off";
1245 break;
1246 case GFS2_QUOTA_ACCOUNT:
1247 state = "account";
1248 break;
1249 case GFS2_QUOTA_ON:
1250 state = "on";
1251 break;
1252 default:
1253 state = "unknown";
1254 break;
1256 seq_printf(s, ",quota=%s", state);
1258 if (args->ar_suiddir)
1259 seq_printf(s, ",suiddir");
1260 if (args->ar_data != GFS2_DATA_DEFAULT) {
1261 char *state;
1262 switch (args->ar_data) {
1263 case GFS2_DATA_WRITEBACK:
1264 state = "writeback";
1265 break;
1266 case GFS2_DATA_ORDERED:
1267 state = "ordered";
1268 break;
1269 default:
1270 state = "unknown";
1271 break;
1273 seq_printf(s, ",data=%s", state);
1275 if (args->ar_discard)
1276 seq_printf(s, ",discard");
1277 val = sdp->sd_tune.gt_logd_secs;
1278 if (val != 30)
1279 seq_printf(s, ",commit=%d", val);
1280 val = sdp->sd_tune.gt_statfs_quantum;
1281 if (val != 30)
1282 seq_printf(s, ",statfs_quantum=%d", val);
1283 val = sdp->sd_tune.gt_quota_quantum;
1284 if (val != 60)
1285 seq_printf(s, ",quota_quantum=%d", val);
1286 if (args->ar_statfs_percent)
1287 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1288 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1289 const char *state;
1291 switch (args->ar_errors) {
1292 case GFS2_ERRORS_WITHDRAW:
1293 state = "withdraw";
1294 break;
1295 case GFS2_ERRORS_PANIC:
1296 state = "panic";
1297 break;
1298 default:
1299 state = "unknown";
1300 break;
1302 seq_printf(s, ",errors=%s", state);
1304 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1305 seq_printf(s, ",nobarrier");
1306 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1307 seq_printf(s, ",demote_interface_used");
1308 return 0;
1312 * We have to (at the moment) hold the inodes main lock to cover
1313 * the gap between unlocking the shared lock on the iopen lock and
1314 * taking the exclusive lock. I'd rather do a shared -> exclusive
1315 * conversion on the iopen lock, but we can change that later. This
1316 * is safe, just less efficient.
1319 static void gfs2_evict_inode(struct inode *inode)
1321 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
1322 struct gfs2_inode *ip = GFS2_I(inode);
1323 struct gfs2_holder gh;
1324 int error;
1326 if (inode->i_nlink)
1327 goto out;
1329 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1330 if (unlikely(error)) {
1331 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1332 goto out;
1335 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1336 if (error)
1337 goto out_truncate;
1339 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1340 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1341 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1342 error = gfs2_glock_nq(&ip->i_iopen_gh);
1343 if (error)
1344 goto out_truncate;
1346 if (S_ISDIR(inode->i_mode) &&
1347 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1348 error = gfs2_dir_exhash_dealloc(ip);
1349 if (error)
1350 goto out_unlock;
1353 if (ip->i_eattr) {
1354 error = gfs2_ea_dealloc(ip);
1355 if (error)
1356 goto out_unlock;
1359 if (!gfs2_is_stuffed(ip)) {
1360 error = gfs2_file_dealloc(ip);
1361 if (error)
1362 goto out_unlock;
1365 error = gfs2_dinode_dealloc(ip);
1366 if (error)
1367 goto out_unlock;
1369 out_truncate:
1370 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1371 if (error)
1372 goto out_unlock;
1373 /* Needs to be done before glock release & also in a transaction */
1374 truncate_inode_pages(&inode->i_data, 0);
1375 gfs2_trans_end(sdp);
1377 out_unlock:
1378 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
1379 gfs2_glock_dq(&ip->i_iopen_gh);
1380 gfs2_holder_uninit(&ip->i_iopen_gh);
1381 gfs2_glock_dq_uninit(&gh);
1382 if (error && error != GLR_TRYFAILED && error != -EROFS)
1383 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1384 out:
1385 truncate_inode_pages(&inode->i_data, 0);
1386 end_writeback(inode);
1388 ip->i_gl->gl_object = NULL;
1389 gfs2_glock_put(ip->i_gl);
1390 ip->i_gl = NULL;
1391 if (ip->i_iopen_gh.gh_gl) {
1392 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1393 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1397 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1399 struct gfs2_inode *ip;
1401 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1402 if (ip) {
1403 ip->i_flags = 0;
1404 ip->i_gl = NULL;
1406 return &ip->i_inode;
1409 static void gfs2_i_callback(struct rcu_head *head)
1411 struct inode *inode = container_of(head, struct inode, i_rcu);
1412 INIT_LIST_HEAD(&inode->i_dentry);
1413 kmem_cache_free(gfs2_inode_cachep, inode);
1416 static void gfs2_destroy_inode(struct inode *inode)
1418 call_rcu(&inode->i_rcu, gfs2_i_callback);
1421 const struct super_operations gfs2_super_ops = {
1422 .alloc_inode = gfs2_alloc_inode,
1423 .destroy_inode = gfs2_destroy_inode,
1424 .write_inode = gfs2_write_inode,
1425 .evict_inode = gfs2_evict_inode,
1426 .put_super = gfs2_put_super,
1427 .sync_fs = gfs2_sync_fs,
1428 .freeze_fs = gfs2_freeze,
1429 .unfreeze_fs = gfs2_unfreeze,
1430 .statfs = gfs2_statfs,
1431 .remount_fs = gfs2_remount_fs,
1432 .drop_inode = gfs2_drop_inode,
1433 .show_options = gfs2_show_options,