GFS2: Wait for unlock completion on umount
[linux-2.6.git] / fs / gfs2 / super.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
4  *
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  */
9
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
26 #include "gfs2.h"
27 #include "incore.h"
28 #include "bmap.h"
29 #include "dir.h"
30 #include "glock.h"
31 #include "glops.h"
32 #include "inode.h"
33 #include "log.h"
34 #include "meta_io.h"
35 #include "quota.h"
36 #include "recovery.h"
37 #include "rgrp.h"
38 #include "super.h"
39 #include "trans.h"
40 #include "util.h"
41 #include "sys.h"
42 #include "xattr.h"
43
44 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
45
46 enum {
47         Opt_lockproto,
48         Opt_locktable,
49         Opt_hostdata,
50         Opt_spectator,
51         Opt_ignore_local_fs,
52         Opt_localflocks,
53         Opt_localcaching,
54         Opt_debug,
55         Opt_nodebug,
56         Opt_upgrade,
57         Opt_acl,
58         Opt_noacl,
59         Opt_quota_off,
60         Opt_quota_account,
61         Opt_quota_on,
62         Opt_quota,
63         Opt_noquota,
64         Opt_suiddir,
65         Opt_nosuiddir,
66         Opt_data_writeback,
67         Opt_data_ordered,
68         Opt_meta,
69         Opt_discard,
70         Opt_nodiscard,
71         Opt_commit,
72         Opt_err_withdraw,
73         Opt_err_panic,
74         Opt_statfs_quantum,
75         Opt_statfs_percent,
76         Opt_quota_quantum,
77         Opt_barrier,
78         Opt_nobarrier,
79         Opt_error,
80 };
81
82 static const match_table_t tokens = {
83         {Opt_lockproto, "lockproto=%s"},
84         {Opt_locktable, "locktable=%s"},
85         {Opt_hostdata, "hostdata=%s"},
86         {Opt_spectator, "spectator"},
87         {Opt_ignore_local_fs, "ignore_local_fs"},
88         {Opt_localflocks, "localflocks"},
89         {Opt_localcaching, "localcaching"},
90         {Opt_debug, "debug"},
91         {Opt_nodebug, "nodebug"},
92         {Opt_upgrade, "upgrade"},
93         {Opt_acl, "acl"},
94         {Opt_noacl, "noacl"},
95         {Opt_quota_off, "quota=off"},
96         {Opt_quota_account, "quota=account"},
97         {Opt_quota_on, "quota=on"},
98         {Opt_quota, "quota"},
99         {Opt_noquota, "noquota"},
100         {Opt_suiddir, "suiddir"},
101         {Opt_nosuiddir, "nosuiddir"},
102         {Opt_data_writeback, "data=writeback"},
103         {Opt_data_ordered, "data=ordered"},
104         {Opt_meta, "meta"},
105         {Opt_discard, "discard"},
106         {Opt_nodiscard, "nodiscard"},
107         {Opt_commit, "commit=%d"},
108         {Opt_err_withdraw, "errors=withdraw"},
109         {Opt_err_panic, "errors=panic"},
110         {Opt_statfs_quantum, "statfs_quantum=%d"},
111         {Opt_statfs_percent, "statfs_percent=%d"},
112         {Opt_quota_quantum, "quota_quantum=%d"},
113         {Opt_barrier, "barrier"},
114         {Opt_nobarrier, "nobarrier"},
115         {Opt_error, NULL}
116 };
117
118 /**
119  * gfs2_mount_args - Parse mount options
120  * @args: The structure into which the parsed options will be written
121  * @options: The options to parse
122  *
123  * Return: errno
124  */
125
126 int gfs2_mount_args(struct gfs2_args *args, char *options)
127 {
128         char *o;
129         int token;
130         substring_t tmp[MAX_OPT_ARGS];
131         int rv;
132
133         /* Split the options into tokens with the "," character and
134            process them */
135
136         while (1) {
137                 o = strsep(&options, ",");
138                 if (o == NULL)
139                         break;
140                 if (*o == '\0')
141                         continue;
142
143                 token = match_token(o, tokens, tmp);
144                 switch (token) {
145                 case Opt_lockproto:
146                         match_strlcpy(args->ar_lockproto, &tmp[0],
147                                       GFS2_LOCKNAME_LEN);
148                         break;
149                 case Opt_locktable:
150                         match_strlcpy(args->ar_locktable, &tmp[0],
151                                       GFS2_LOCKNAME_LEN);
152                         break;
153                 case Opt_hostdata:
154                         match_strlcpy(args->ar_hostdata, &tmp[0],
155                                       GFS2_LOCKNAME_LEN);
156                         break;
157                 case Opt_spectator:
158                         args->ar_spectator = 1;
159                         break;
160                 case Opt_ignore_local_fs:
161                         args->ar_ignore_local_fs = 1;
162                         break;
163                 case Opt_localflocks:
164                         args->ar_localflocks = 1;
165                         break;
166                 case Opt_localcaching:
167                         args->ar_localcaching = 1;
168                         break;
169                 case Opt_debug:
170                         if (args->ar_errors == GFS2_ERRORS_PANIC) {
171                                 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
172                                        "are mutually exclusive.\n");
173                                 return -EINVAL;
174                         }
175                         args->ar_debug = 1;
176                         break;
177                 case Opt_nodebug:
178                         args->ar_debug = 0;
179                         break;
180                 case Opt_upgrade:
181                         args->ar_upgrade = 1;
182                         break;
183                 case Opt_acl:
184                         args->ar_posix_acl = 1;
185                         break;
186                 case Opt_noacl:
187                         args->ar_posix_acl = 0;
188                         break;
189                 case Opt_quota_off:
190                 case Opt_noquota:
191                         args->ar_quota = GFS2_QUOTA_OFF;
192                         break;
193                 case Opt_quota_account:
194                         args->ar_quota = GFS2_QUOTA_ACCOUNT;
195                         break;
196                 case Opt_quota_on:
197                 case Opt_quota:
198                         args->ar_quota = GFS2_QUOTA_ON;
199                         break;
200                 case Opt_suiddir:
201                         args->ar_suiddir = 1;
202                         break;
203                 case Opt_nosuiddir:
204                         args->ar_suiddir = 0;
205                         break;
206                 case Opt_data_writeback:
207                         args->ar_data = GFS2_DATA_WRITEBACK;
208                         break;
209                 case Opt_data_ordered:
210                         args->ar_data = GFS2_DATA_ORDERED;
211                         break;
212                 case Opt_meta:
213                         args->ar_meta = 1;
214                         break;
215                 case Opt_discard:
216                         args->ar_discard = 1;
217                         break;
218                 case Opt_nodiscard:
219                         args->ar_discard = 0;
220                         break;
221                 case Opt_commit:
222                         rv = match_int(&tmp[0], &args->ar_commit);
223                         if (rv || args->ar_commit <= 0) {
224                                 printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n");
225                                 return rv ? rv : -EINVAL;
226                         }
227                         break;
228                 case Opt_statfs_quantum:
229                         rv = match_int(&tmp[0], &args->ar_statfs_quantum);
230                         if (rv || args->ar_statfs_quantum < 0) {
231                                 printk(KERN_WARNING "GFS2: statfs_quantum mount option requires a non-negative numeric argument\n");
232                                 return rv ? rv : -EINVAL;
233                         }
234                         break;
235                 case Opt_quota_quantum:
236                         rv = match_int(&tmp[0], &args->ar_quota_quantum);
237                         if (rv || args->ar_quota_quantum <= 0) {
238                                 printk(KERN_WARNING "GFS2: quota_quantum mount option requires a positive numeric argument\n");
239                                 return rv ? rv : -EINVAL;
240                         }
241                         break;
242                 case Opt_statfs_percent:
243                         rv = match_int(&tmp[0], &args->ar_statfs_percent);
244                         if (rv || args->ar_statfs_percent < 0 ||
245                             args->ar_statfs_percent > 100) {
246                                 printk(KERN_WARNING "statfs_percent mount option requires a numeric argument between 0 and 100\n");
247                                 return rv ? rv : -EINVAL;
248                         }
249                         break;
250                 case Opt_err_withdraw:
251                         args->ar_errors = GFS2_ERRORS_WITHDRAW;
252                         break;
253                 case Opt_err_panic:
254                         if (args->ar_debug) {
255                                 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
256                                         "are mutually exclusive.\n");
257                                 return -EINVAL;
258                         }
259                         args->ar_errors = GFS2_ERRORS_PANIC;
260                         break;
261                 case Opt_barrier:
262                         args->ar_nobarrier = 0;
263                         break;
264                 case Opt_nobarrier:
265                         args->ar_nobarrier = 1;
266                         break;
267                 case Opt_error:
268                 default:
269                         printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o);
270                         return -EINVAL;
271                 }
272         }
273
274         return 0;
275 }
276
277 /**
278  * gfs2_jindex_free - Clear all the journal index information
279  * @sdp: The GFS2 superblock
280  *
281  */
282
283 void gfs2_jindex_free(struct gfs2_sbd *sdp)
284 {
285         struct list_head list, *head;
286         struct gfs2_jdesc *jd;
287         struct gfs2_journal_extent *jext;
288
289         spin_lock(&sdp->sd_jindex_spin);
290         list_add(&list, &sdp->sd_jindex_list);
291         list_del_init(&sdp->sd_jindex_list);
292         sdp->sd_journals = 0;
293         spin_unlock(&sdp->sd_jindex_spin);
294
295         while (!list_empty(&list)) {
296                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
297                 head = &jd->extent_list;
298                 while (!list_empty(head)) {
299                         jext = list_entry(head->next,
300                                           struct gfs2_journal_extent,
301                                           extent_list);
302                         list_del(&jext->extent_list);
303                         kfree(jext);
304                 }
305                 list_del(&jd->jd_list);
306                 iput(jd->jd_inode);
307                 kfree(jd);
308         }
309 }
310
311 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
312 {
313         struct gfs2_jdesc *jd;
314         int found = 0;
315
316         list_for_each_entry(jd, head, jd_list) {
317                 if (jd->jd_jid == jid) {
318                         found = 1;
319                         break;
320                 }
321         }
322
323         if (!found)
324                 jd = NULL;
325
326         return jd;
327 }
328
329 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
330 {
331         struct gfs2_jdesc *jd;
332
333         spin_lock(&sdp->sd_jindex_spin);
334         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
335         spin_unlock(&sdp->sd_jindex_spin);
336
337         return jd;
338 }
339
340 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
341 {
342         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
343         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
344         int ar;
345         int error;
346
347         if (ip->i_disksize < (8 << 20) || ip->i_disksize > (1 << 30) ||
348             (ip->i_disksize & (sdp->sd_sb.sb_bsize - 1))) {
349                 gfs2_consist_inode(ip);
350                 return -EIO;
351         }
352         jd->jd_blocks = ip->i_disksize >> sdp->sd_sb.sb_bsize_shift;
353
354         error = gfs2_write_alloc_required(ip, 0, ip->i_disksize, &ar);
355         if (!error && ar) {
356                 gfs2_consist_inode(ip);
357                 error = -EIO;
358         }
359
360         return error;
361 }
362
363 /**
364  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
365  * @sdp: the filesystem
366  *
367  * Returns: errno
368  */
369
370 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
371 {
372         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
373         struct gfs2_glock *j_gl = ip->i_gl;
374         struct gfs2_holder t_gh;
375         struct gfs2_log_header_host head;
376         int error;
377
378         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
379         if (error)
380                 return error;
381
382         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
383
384         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
385         if (error)
386                 goto fail;
387
388         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
389                 gfs2_consist(sdp);
390                 error = -EIO;
391                 goto fail;
392         }
393
394         /*  Initialize some head of the log stuff  */
395         sdp->sd_log_sequence = head.lh_sequence + 1;
396         gfs2_log_pointers_init(sdp, head.lh_blkno);
397
398         error = gfs2_quota_init(sdp);
399         if (error)
400                 goto fail;
401
402         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
403
404         gfs2_glock_dq_uninit(&t_gh);
405
406         return 0;
407
408 fail:
409         t_gh.gh_flags |= GL_NOCACHE;
410         gfs2_glock_dq_uninit(&t_gh);
411
412         return error;
413 }
414
415 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
416 {
417         const struct gfs2_statfs_change *str = buf;
418
419         sc->sc_total = be64_to_cpu(str->sc_total);
420         sc->sc_free = be64_to_cpu(str->sc_free);
421         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
422 }
423
424 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
425 {
426         struct gfs2_statfs_change *str = buf;
427
428         str->sc_total = cpu_to_be64(sc->sc_total);
429         str->sc_free = cpu_to_be64(sc->sc_free);
430         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
431 }
432
433 int gfs2_statfs_init(struct gfs2_sbd *sdp)
434 {
435         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
436         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
437         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
438         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
439         struct buffer_head *m_bh, *l_bh;
440         struct gfs2_holder gh;
441         int error;
442
443         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
444                                    &gh);
445         if (error)
446                 return error;
447
448         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
449         if (error)
450                 goto out;
451
452         if (sdp->sd_args.ar_spectator) {
453                 spin_lock(&sdp->sd_statfs_spin);
454                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
455                                       sizeof(struct gfs2_dinode));
456                 spin_unlock(&sdp->sd_statfs_spin);
457         } else {
458                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
459                 if (error)
460                         goto out_m_bh;
461
462                 spin_lock(&sdp->sd_statfs_spin);
463                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
464                                       sizeof(struct gfs2_dinode));
465                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
466                                       sizeof(struct gfs2_dinode));
467                 spin_unlock(&sdp->sd_statfs_spin);
468
469                 brelse(l_bh);
470         }
471
472 out_m_bh:
473         brelse(m_bh);
474 out:
475         gfs2_glock_dq_uninit(&gh);
476         return 0;
477 }
478
479 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
480                         s64 dinodes)
481 {
482         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
483         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
484         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
485         struct buffer_head *l_bh;
486         s64 x, y;
487         int need_sync = 0;
488         int error;
489
490         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
491         if (error)
492                 return;
493
494         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
495
496         spin_lock(&sdp->sd_statfs_spin);
497         l_sc->sc_total += total;
498         l_sc->sc_free += free;
499         l_sc->sc_dinodes += dinodes;
500         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
501         if (sdp->sd_args.ar_statfs_percent) {
502                 x = 100 * l_sc->sc_free;
503                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
504                 if (x >= y || x <= -y)
505                         need_sync = 1;
506         }
507         spin_unlock(&sdp->sd_statfs_spin);
508
509         brelse(l_bh);
510         if (need_sync)
511                 gfs2_wake_up_statfs(sdp);
512 }
513
514 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
515                    struct buffer_head *l_bh)
516 {
517         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
518         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
519         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
520         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
521
522         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
523
524         spin_lock(&sdp->sd_statfs_spin);
525         m_sc->sc_total += l_sc->sc_total;
526         m_sc->sc_free += l_sc->sc_free;
527         m_sc->sc_dinodes += l_sc->sc_dinodes;
528         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
529         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
530                0, sizeof(struct gfs2_statfs_change));
531         spin_unlock(&sdp->sd_statfs_spin);
532
533         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
534         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
535 }
536
537 int gfs2_statfs_sync(struct super_block *sb, int type)
538 {
539         struct gfs2_sbd *sdp = sb->s_fs_info;
540         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
541         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
542         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
543         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
544         struct gfs2_holder gh;
545         struct buffer_head *m_bh, *l_bh;
546         int error;
547
548         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
549                                    &gh);
550         if (error)
551                 return error;
552
553         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
554         if (error)
555                 goto out;
556
557         spin_lock(&sdp->sd_statfs_spin);
558         gfs2_statfs_change_in(m_sc, m_bh->b_data +
559                               sizeof(struct gfs2_dinode));
560         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
561                 spin_unlock(&sdp->sd_statfs_spin);
562                 goto out_bh;
563         }
564         spin_unlock(&sdp->sd_statfs_spin);
565
566         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
567         if (error)
568                 goto out_bh;
569
570         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
571         if (error)
572                 goto out_bh2;
573
574         update_statfs(sdp, m_bh, l_bh);
575         sdp->sd_statfs_force_sync = 0;
576
577         gfs2_trans_end(sdp);
578
579 out_bh2:
580         brelse(l_bh);
581 out_bh:
582         brelse(m_bh);
583 out:
584         gfs2_glock_dq_uninit(&gh);
585         return error;
586 }
587
588 struct lfcc {
589         struct list_head list;
590         struct gfs2_holder gh;
591 };
592
593 /**
594  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
595  *                            journals are clean
596  * @sdp: the file system
597  * @state: the state to put the transaction lock into
598  * @t_gh: the hold on the transaction lock
599  *
600  * Returns: errno
601  */
602
603 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
604                                     struct gfs2_holder *t_gh)
605 {
606         struct gfs2_inode *ip;
607         struct gfs2_jdesc *jd;
608         struct lfcc *lfcc;
609         LIST_HEAD(list);
610         struct gfs2_log_header_host lh;
611         int error;
612
613         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
614                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
615                 if (!lfcc) {
616                         error = -ENOMEM;
617                         goto out;
618                 }
619                 ip = GFS2_I(jd->jd_inode);
620                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
621                 if (error) {
622                         kfree(lfcc);
623                         goto out;
624                 }
625                 list_add(&lfcc->list, &list);
626         }
627
628         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
629                                    GL_NOCACHE, t_gh);
630
631         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
632                 error = gfs2_jdesc_check(jd);
633                 if (error)
634                         break;
635                 error = gfs2_find_jhead(jd, &lh);
636                 if (error)
637                         break;
638                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
639                         error = -EBUSY;
640                         break;
641                 }
642         }
643
644         if (error)
645                 gfs2_glock_dq_uninit(t_gh);
646
647 out:
648         while (!list_empty(&list)) {
649                 lfcc = list_entry(list.next, struct lfcc, list);
650                 list_del(&lfcc->list);
651                 gfs2_glock_dq_uninit(&lfcc->gh);
652                 kfree(lfcc);
653         }
654         return error;
655 }
656
657 /**
658  * gfs2_freeze_fs - freezes the file system
659  * @sdp: the file system
660  *
661  * This function flushes data and meta data for all machines by
662  * aquiring the transaction log exclusively.  All journals are
663  * ensured to be in a clean state as well.
664  *
665  * Returns: errno
666  */
667
668 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
669 {
670         int error = 0;
671
672         mutex_lock(&sdp->sd_freeze_lock);
673
674         if (!sdp->sd_freeze_count++) {
675                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
676                 if (error)
677                         sdp->sd_freeze_count--;
678         }
679
680         mutex_unlock(&sdp->sd_freeze_lock);
681
682         return error;
683 }
684
685 /**
686  * gfs2_unfreeze_fs - unfreezes the file system
687  * @sdp: the file system
688  *
689  * This function allows the file system to proceed by unlocking
690  * the exclusively held transaction lock.  Other GFS2 nodes are
691  * now free to acquire the lock shared and go on with their lives.
692  *
693  */
694
695 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
696 {
697         mutex_lock(&sdp->sd_freeze_lock);
698
699         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
700                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
701
702         mutex_unlock(&sdp->sd_freeze_lock);
703 }
704
705
706 /**
707  * gfs2_write_inode - Make sure the inode is stable on the disk
708  * @inode: The inode
709  * @sync: synchronous write flag
710  *
711  * Returns: errno
712  */
713
714 static int gfs2_write_inode(struct inode *inode, int sync)
715 {
716         struct gfs2_inode *ip = GFS2_I(inode);
717         struct gfs2_sbd *sdp = GFS2_SB(inode);
718         struct gfs2_holder gh;
719         struct buffer_head *bh;
720         struct timespec atime;
721         struct gfs2_dinode *di;
722         int ret = 0;
723
724         /* Check this is a "normal" inode, etc */
725         if (!test_bit(GIF_USER, &ip->i_flags) ||
726             (current->flags & PF_MEMALLOC))
727                 return 0;
728         ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
729         if (ret)
730                 goto do_flush;
731         ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
732         if (ret)
733                 goto do_unlock;
734         ret = gfs2_meta_inode_buffer(ip, &bh);
735         if (ret == 0) {
736                 di = (struct gfs2_dinode *)bh->b_data;
737                 atime.tv_sec = be64_to_cpu(di->di_atime);
738                 atime.tv_nsec = be32_to_cpu(di->di_atime_nsec);
739                 if (timespec_compare(&inode->i_atime, &atime) > 0) {
740                         gfs2_trans_add_bh(ip->i_gl, bh, 1);
741                         gfs2_dinode_out(ip, bh->b_data);
742                 }
743                 brelse(bh);
744         }
745         gfs2_trans_end(sdp);
746 do_unlock:
747         gfs2_glock_dq_uninit(&gh);
748 do_flush:
749         if (sync != 0)
750                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
751         return ret;
752 }
753
754 /**
755  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
756  * @sdp: the filesystem
757  *
758  * Returns: errno
759  */
760
761 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
762 {
763         struct gfs2_holder t_gh;
764         int error;
765
766         flush_workqueue(gfs2_delete_workqueue);
767         gfs2_quota_sync(sdp->sd_vfs, 0);
768         gfs2_statfs_sync(sdp->sd_vfs, 0);
769
770         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
771                                    &t_gh);
772         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
773                 return error;
774
775         gfs2_meta_syncfs(sdp);
776         gfs2_log_shutdown(sdp);
777
778         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
779
780         if (t_gh.gh_gl)
781                 gfs2_glock_dq_uninit(&t_gh);
782
783         gfs2_quota_cleanup(sdp);
784
785         return error;
786 }
787
788 static int gfs2_umount_recovery_wait(void *word)
789 {
790         schedule();
791         return 0;
792 }
793
794 /**
795  * gfs2_put_super - Unmount the filesystem
796  * @sb: The VFS superblock
797  *
798  */
799
800 static void gfs2_put_super(struct super_block *sb)
801 {
802         struct gfs2_sbd *sdp = sb->s_fs_info;
803         int error;
804         struct gfs2_jdesc *jd;
805
806         /*  Unfreeze the filesystem, if we need to  */
807
808         mutex_lock(&sdp->sd_freeze_lock);
809         if (sdp->sd_freeze_count)
810                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
811         mutex_unlock(&sdp->sd_freeze_lock);
812
813         /* No more recovery requests */
814         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
815         smp_mb();
816
817         /* Wait on outstanding recovery */
818 restart:
819         spin_lock(&sdp->sd_jindex_spin);
820         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
821                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
822                         continue;
823                 spin_unlock(&sdp->sd_jindex_spin);
824                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
825                             gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
826                 goto restart;
827         }
828         spin_unlock(&sdp->sd_jindex_spin);
829
830         kthread_stop(sdp->sd_quotad_process);
831         kthread_stop(sdp->sd_logd_process);
832
833         if (!(sb->s_flags & MS_RDONLY)) {
834                 error = gfs2_make_fs_ro(sdp);
835                 if (error)
836                         gfs2_io_error(sdp);
837         }
838         /*  At this point, we're through modifying the disk  */
839
840         /*  Release stuff  */
841
842         iput(sdp->sd_jindex);
843         iput(sdp->sd_statfs_inode);
844         iput(sdp->sd_rindex);
845         iput(sdp->sd_quota_inode);
846
847         gfs2_glock_put(sdp->sd_rename_gl);
848         gfs2_glock_put(sdp->sd_trans_gl);
849
850         if (!sdp->sd_args.ar_spectator) {
851                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
852                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
853                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
854                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
855                 iput(sdp->sd_sc_inode);
856                 iput(sdp->sd_qc_inode);
857         }
858
859         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
860         gfs2_clear_rgrpd(sdp);
861         gfs2_jindex_free(sdp);
862         /*  Take apart glock structures and buffer lists  */
863         gfs2_gl_hash_clear(sdp);
864         /* Wait for dlm to reply to all our unlock requests */
865         wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
866         /*  Unmount the locking protocol  */
867         gfs2_lm_unmount(sdp);
868
869         /*  At this point, we're through participating in the lockspace  */
870         gfs2_sys_fs_del(sdp);
871 }
872
873 /**
874  * gfs2_sync_fs - sync the filesystem
875  * @sb: the superblock
876  *
877  * Flushes the log to disk.
878  */
879
880 static int gfs2_sync_fs(struct super_block *sb, int wait)
881 {
882         if (wait && sb->s_fs_info)
883                 gfs2_log_flush(sb->s_fs_info, NULL);
884         return 0;
885 }
886
887 /**
888  * gfs2_freeze - prevent further writes to the filesystem
889  * @sb: the VFS structure for the filesystem
890  *
891  */
892
893 static int gfs2_freeze(struct super_block *sb)
894 {
895         struct gfs2_sbd *sdp = sb->s_fs_info;
896         int error;
897
898         if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
899                 return -EINVAL;
900
901         for (;;) {
902                 error = gfs2_freeze_fs(sdp);
903                 if (!error)
904                         break;
905
906                 switch (error) {
907                 case -EBUSY:
908                         fs_err(sdp, "waiting for recovery before freeze\n");
909                         break;
910
911                 default:
912                         fs_err(sdp, "error freezing FS: %d\n", error);
913                         break;
914                 }
915
916                 fs_err(sdp, "retrying...\n");
917                 msleep(1000);
918         }
919         return 0;
920 }
921
922 /**
923  * gfs2_unfreeze - reallow writes to the filesystem
924  * @sb: the VFS structure for the filesystem
925  *
926  */
927
928 static int gfs2_unfreeze(struct super_block *sb)
929 {
930         gfs2_unfreeze_fs(sb->s_fs_info);
931         return 0;
932 }
933
934 /**
935  * statfs_fill - fill in the sg for a given RG
936  * @rgd: the RG
937  * @sc: the sc structure
938  *
939  * Returns: 0 on success, -ESTALE if the LVB is invalid
940  */
941
942 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
943                             struct gfs2_statfs_change_host *sc)
944 {
945         gfs2_rgrp_verify(rgd);
946         sc->sc_total += rgd->rd_data;
947         sc->sc_free += rgd->rd_free;
948         sc->sc_dinodes += rgd->rd_dinodes;
949         return 0;
950 }
951
952 /**
953  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
954  * @sdp: the filesystem
955  * @sc: the sc info that will be returned
956  *
957  * Any error (other than a signal) will cause this routine to fall back
958  * to the synchronous version.
959  *
960  * FIXME: This really shouldn't busy wait like this.
961  *
962  * Returns: errno
963  */
964
965 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
966 {
967         struct gfs2_holder ri_gh;
968         struct gfs2_rgrpd *rgd_next;
969         struct gfs2_holder *gha, *gh;
970         unsigned int slots = 64;
971         unsigned int x;
972         int done;
973         int error = 0, err;
974
975         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
976         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
977         if (!gha)
978                 return -ENOMEM;
979
980         error = gfs2_rindex_hold(sdp, &ri_gh);
981         if (error)
982                 goto out;
983
984         rgd_next = gfs2_rgrpd_get_first(sdp);
985
986         for (;;) {
987                 done = 1;
988
989                 for (x = 0; x < slots; x++) {
990                         gh = gha + x;
991
992                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
993                                 err = gfs2_glock_wait(gh);
994                                 if (err) {
995                                         gfs2_holder_uninit(gh);
996                                         error = err;
997                                 } else {
998                                         if (!error)
999                                                 error = statfs_slow_fill(
1000                                                         gh->gh_gl->gl_object, sc);
1001                                         gfs2_glock_dq_uninit(gh);
1002                                 }
1003                         }
1004
1005                         if (gh->gh_gl)
1006                                 done = 0;
1007                         else if (rgd_next && !error) {
1008                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1009                                                            LM_ST_SHARED,
1010                                                            GL_ASYNC,
1011                                                            gh);
1012                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1013                                 done = 0;
1014                         }
1015
1016                         if (signal_pending(current))
1017                                 error = -ERESTARTSYS;
1018                 }
1019
1020                 if (done)
1021                         break;
1022
1023                 yield();
1024         }
1025
1026         gfs2_glock_dq_uninit(&ri_gh);
1027
1028 out:
1029         kfree(gha);
1030         return error;
1031 }
1032
1033 /**
1034  * gfs2_statfs_i - Do a statfs
1035  * @sdp: the filesystem
1036  * @sg: the sg structure
1037  *
1038  * Returns: errno
1039  */
1040
1041 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1042 {
1043         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1044         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1045
1046         spin_lock(&sdp->sd_statfs_spin);
1047
1048         *sc = *m_sc;
1049         sc->sc_total += l_sc->sc_total;
1050         sc->sc_free += l_sc->sc_free;
1051         sc->sc_dinodes += l_sc->sc_dinodes;
1052
1053         spin_unlock(&sdp->sd_statfs_spin);
1054
1055         if (sc->sc_free < 0)
1056                 sc->sc_free = 0;
1057         if (sc->sc_free > sc->sc_total)
1058                 sc->sc_free = sc->sc_total;
1059         if (sc->sc_dinodes < 0)
1060                 sc->sc_dinodes = 0;
1061
1062         return 0;
1063 }
1064
1065 /**
1066  * gfs2_statfs - Gather and return stats about the filesystem
1067  * @sb: The superblock
1068  * @statfsbuf: The buffer
1069  *
1070  * Returns: 0 on success or error code
1071  */
1072
1073 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1074 {
1075         struct super_block *sb = dentry->d_inode->i_sb;
1076         struct gfs2_sbd *sdp = sb->s_fs_info;
1077         struct gfs2_statfs_change_host sc;
1078         int error;
1079
1080         if (gfs2_tune_get(sdp, gt_statfs_slow))
1081                 error = gfs2_statfs_slow(sdp, &sc);
1082         else
1083                 error = gfs2_statfs_i(sdp, &sc);
1084
1085         if (error)
1086                 return error;
1087
1088         buf->f_type = GFS2_MAGIC;
1089         buf->f_bsize = sdp->sd_sb.sb_bsize;
1090         buf->f_blocks = sc.sc_total;
1091         buf->f_bfree = sc.sc_free;
1092         buf->f_bavail = sc.sc_free;
1093         buf->f_files = sc.sc_dinodes + sc.sc_free;
1094         buf->f_ffree = sc.sc_free;
1095         buf->f_namelen = GFS2_FNAMESIZE;
1096
1097         return 0;
1098 }
1099
1100 /**
1101  * gfs2_remount_fs - called when the FS is remounted
1102  * @sb:  the filesystem
1103  * @flags:  the remount flags
1104  * @data:  extra data passed in (not used right now)
1105  *
1106  * Returns: errno
1107  */
1108
1109 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1110 {
1111         struct gfs2_sbd *sdp = sb->s_fs_info;
1112         struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1113         struct gfs2_tune *gt = &sdp->sd_tune;
1114         int error;
1115
1116         spin_lock(&gt->gt_spin);
1117         args.ar_commit = gt->gt_log_flush_secs;
1118         args.ar_quota_quantum = gt->gt_quota_quantum;
1119         if (gt->gt_statfs_slow)
1120                 args.ar_statfs_quantum = 0;
1121         else
1122                 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1123         spin_unlock(&gt->gt_spin);
1124         error = gfs2_mount_args(&args, data);
1125         if (error)
1126                 return error;
1127
1128         /* Not allowed to change locking details */
1129         if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1130             strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1131             strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1132                 return -EINVAL;
1133
1134         /* Some flags must not be changed */
1135         if (args_neq(&args, &sdp->sd_args, spectator) ||
1136             args_neq(&args, &sdp->sd_args, ignore_local_fs) ||
1137             args_neq(&args, &sdp->sd_args, localflocks) ||
1138             args_neq(&args, &sdp->sd_args, localcaching) ||
1139             args_neq(&args, &sdp->sd_args, meta))
1140                 return -EINVAL;
1141
1142         if (sdp->sd_args.ar_spectator)
1143                 *flags |= MS_RDONLY;
1144
1145         if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1146                 if (*flags & MS_RDONLY)
1147                         error = gfs2_make_fs_ro(sdp);
1148                 else
1149                         error = gfs2_make_fs_rw(sdp);
1150                 if (error)
1151                         return error;
1152         }
1153
1154         sdp->sd_args = args;
1155         if (sdp->sd_args.ar_posix_acl)
1156                 sb->s_flags |= MS_POSIXACL;
1157         else
1158                 sb->s_flags &= ~MS_POSIXACL;
1159         if (sdp->sd_args.ar_nobarrier)
1160                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1161         else
1162                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1163         spin_lock(&gt->gt_spin);
1164         gt->gt_log_flush_secs = args.ar_commit;
1165         gt->gt_quota_quantum = args.ar_quota_quantum;
1166         if (args.ar_statfs_quantum) {
1167                 gt->gt_statfs_slow = 0;
1168                 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1169         }
1170         else {
1171                 gt->gt_statfs_slow = 1;
1172                 gt->gt_statfs_quantum = 30;
1173         }
1174         spin_unlock(&gt->gt_spin);
1175
1176         gfs2_online_uevent(sdp);
1177         return 0;
1178 }
1179
1180 /**
1181  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1182  * @inode: The inode to drop
1183  *
1184  * If we've received a callback on an iopen lock then its because a
1185  * remote node tried to deallocate the inode but failed due to this node
1186  * still having the inode open. Here we mark the link count zero
1187  * since we know that it must have reached zero if the GLF_DEMOTE flag
1188  * is set on the iopen glock. If we didn't do a disk read since the
1189  * remote node removed the final link then we might otherwise miss
1190  * this event. This check ensures that this node will deallocate the
1191  * inode's blocks, or alternatively pass the baton on to another
1192  * node for later deallocation.
1193  */
1194
1195 static void gfs2_drop_inode(struct inode *inode)
1196 {
1197         struct gfs2_inode *ip = GFS2_I(inode);
1198
1199         if (test_bit(GIF_USER, &ip->i_flags) && inode->i_nlink) {
1200                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1201                 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1202                         clear_nlink(inode);
1203         }
1204         generic_drop_inode(inode);
1205 }
1206
1207 /**
1208  * gfs2_clear_inode - Deallocate an inode when VFS is done with it
1209  * @inode: The VFS inode
1210  *
1211  */
1212
1213 static void gfs2_clear_inode(struct inode *inode)
1214 {
1215         struct gfs2_inode *ip = GFS2_I(inode);
1216
1217         /* This tells us its a "real" inode and not one which only
1218          * serves to contain an address space (see rgrp.c, meta_io.c)
1219          * which therefore doesn't have its own glocks.
1220          */
1221         if (test_bit(GIF_USER, &ip->i_flags)) {
1222                 ip->i_gl->gl_object = NULL;
1223                 gfs2_glock_put(ip->i_gl);
1224                 ip->i_gl = NULL;
1225                 if (ip->i_iopen_gh.gh_gl) {
1226                         ip->i_iopen_gh.gh_gl->gl_object = NULL;
1227                         gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1228                 }
1229         }
1230 }
1231
1232 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1233 {
1234         do {
1235                 if (d1 == d2)
1236                         return 1;
1237                 d1 = d1->d_parent;
1238         } while (!IS_ROOT(d1));
1239         return 0;
1240 }
1241
1242 /**
1243  * gfs2_show_options - Show mount options for /proc/mounts
1244  * @s: seq_file structure
1245  * @mnt: vfsmount
1246  *
1247  * Returns: 0 on success or error code
1248  */
1249
1250 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1251 {
1252         struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
1253         struct gfs2_args *args = &sdp->sd_args;
1254         int val;
1255
1256         if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir))
1257                 seq_printf(s, ",meta");
1258         if (args->ar_lockproto[0])
1259                 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1260         if (args->ar_locktable[0])
1261                 seq_printf(s, ",locktable=%s", args->ar_locktable);
1262         if (args->ar_hostdata[0])
1263                 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1264         if (args->ar_spectator)
1265                 seq_printf(s, ",spectator");
1266         if (args->ar_ignore_local_fs)
1267                 seq_printf(s, ",ignore_local_fs");
1268         if (args->ar_localflocks)
1269                 seq_printf(s, ",localflocks");
1270         if (args->ar_localcaching)
1271                 seq_printf(s, ",localcaching");
1272         if (args->ar_debug)
1273                 seq_printf(s, ",debug");
1274         if (args->ar_upgrade)
1275                 seq_printf(s, ",upgrade");
1276         if (args->ar_posix_acl)
1277                 seq_printf(s, ",acl");
1278         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1279                 char *state;
1280                 switch (args->ar_quota) {
1281                 case GFS2_QUOTA_OFF:
1282                         state = "off";
1283                         break;
1284                 case GFS2_QUOTA_ACCOUNT:
1285                         state = "account";
1286                         break;
1287                 case GFS2_QUOTA_ON:
1288                         state = "on";
1289                         break;
1290                 default:
1291                         state = "unknown";
1292                         break;
1293                 }
1294                 seq_printf(s, ",quota=%s", state);
1295         }
1296         if (args->ar_suiddir)
1297                 seq_printf(s, ",suiddir");
1298         if (args->ar_data != GFS2_DATA_DEFAULT) {
1299                 char *state;
1300                 switch (args->ar_data) {
1301                 case GFS2_DATA_WRITEBACK:
1302                         state = "writeback";
1303                         break;
1304                 case GFS2_DATA_ORDERED:
1305                         state = "ordered";
1306                         break;
1307                 default:
1308                         state = "unknown";
1309                         break;
1310                 }
1311                 seq_printf(s, ",data=%s", state);
1312         }
1313         if (args->ar_discard)
1314                 seq_printf(s, ",discard");
1315         val = sdp->sd_tune.gt_log_flush_secs;
1316         if (val != 60)
1317                 seq_printf(s, ",commit=%d", val);
1318         val = sdp->sd_tune.gt_statfs_quantum;
1319         if (val != 30)
1320                 seq_printf(s, ",statfs_quantum=%d", val);
1321         val = sdp->sd_tune.gt_quota_quantum;
1322         if (val != 60)
1323                 seq_printf(s, ",quota_quantum=%d", val);
1324         if (args->ar_statfs_percent)
1325                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1326         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1327                 const char *state;
1328
1329                 switch (args->ar_errors) {
1330                 case GFS2_ERRORS_WITHDRAW:
1331                         state = "withdraw";
1332                         break;
1333                 case GFS2_ERRORS_PANIC:
1334                         state = "panic";
1335                         break;
1336                 default:
1337                         state = "unknown";
1338                         break;
1339                 }
1340                 seq_printf(s, ",errors=%s", state);
1341         }
1342         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1343                 seq_printf(s, ",nobarrier");
1344
1345         return 0;
1346 }
1347
1348 /*
1349  * We have to (at the moment) hold the inodes main lock to cover
1350  * the gap between unlocking the shared lock on the iopen lock and
1351  * taking the exclusive lock. I'd rather do a shared -> exclusive
1352  * conversion on the iopen lock, but we can change that later. This
1353  * is safe, just less efficient.
1354  */
1355
1356 static void gfs2_delete_inode(struct inode *inode)
1357 {
1358         struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
1359         struct gfs2_inode *ip = GFS2_I(inode);
1360         struct gfs2_holder gh;
1361         int error;
1362
1363         if (!test_bit(GIF_USER, &ip->i_flags))
1364                 goto out;
1365
1366         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1367         if (unlikely(error)) {
1368                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1369                 goto out;
1370         }
1371
1372         error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1373         if (error)
1374                 goto out_truncate;
1375
1376         gfs2_glock_dq_wait(&ip->i_iopen_gh);
1377         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1378         error = gfs2_glock_nq(&ip->i_iopen_gh);
1379         if (error)
1380                 goto out_truncate;
1381
1382         if (S_ISDIR(inode->i_mode) &&
1383             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1384                 error = gfs2_dir_exhash_dealloc(ip);
1385                 if (error)
1386                         goto out_unlock;
1387         }
1388
1389         if (ip->i_eattr) {
1390                 error = gfs2_ea_dealloc(ip);
1391                 if (error)
1392                         goto out_unlock;
1393         }
1394
1395         if (!gfs2_is_stuffed(ip)) {
1396                 error = gfs2_file_dealloc(ip);
1397                 if (error)
1398                         goto out_unlock;
1399         }
1400
1401         error = gfs2_dinode_dealloc(ip);
1402         if (error)
1403                 goto out_unlock;
1404
1405 out_truncate:
1406         error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1407         if (error)
1408                 goto out_unlock;
1409         /* Needs to be done before glock release & also in a transaction */
1410         truncate_inode_pages(&inode->i_data, 0);
1411         gfs2_trans_end(sdp);
1412
1413 out_unlock:
1414         if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
1415                 gfs2_glock_dq(&ip->i_iopen_gh);
1416         gfs2_holder_uninit(&ip->i_iopen_gh);
1417         gfs2_glock_dq_uninit(&gh);
1418         if (error && error != GLR_TRYFAILED && error != -EROFS)
1419                 fs_warn(sdp, "gfs2_delete_inode: %d\n", error);
1420 out:
1421         truncate_inode_pages(&inode->i_data, 0);
1422         clear_inode(inode);
1423 }
1424
1425 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1426 {
1427         struct gfs2_inode *ip;
1428
1429         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1430         if (ip) {
1431                 ip->i_flags = 0;
1432                 ip->i_gl = NULL;
1433         }
1434         return &ip->i_inode;
1435 }
1436
1437 static void gfs2_destroy_inode(struct inode *inode)
1438 {
1439         kmem_cache_free(gfs2_inode_cachep, inode);
1440 }
1441
1442 const struct super_operations gfs2_super_ops = {
1443         .alloc_inode            = gfs2_alloc_inode,
1444         .destroy_inode          = gfs2_destroy_inode,
1445         .write_inode            = gfs2_write_inode,
1446         .delete_inode           = gfs2_delete_inode,
1447         .put_super              = gfs2_put_super,
1448         .sync_fs                = gfs2_sync_fs,
1449         .freeze_fs              = gfs2_freeze,
1450         .unfreeze_fs            = gfs2_unfreeze,
1451         .statfs                 = gfs2_statfs,
1452         .remount_fs             = gfs2_remount_fs,
1453         .clear_inode            = gfs2_clear_inode,
1454         .drop_inode             = gfs2_drop_inode,
1455         .show_options           = gfs2_show_options,
1456 };
1457