]> Pileus Git - ~andy/linux/blob - fs/gfs2/super.c
GFS2: Improve statfs and quota usability
[~andy/linux] / 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
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"
42
43 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
44
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_error,
77 };
78
79 static const match_table_t tokens = {
80         {Opt_lockproto, "lockproto=%s"},
81         {Opt_locktable, "locktable=%s"},
82         {Opt_hostdata, "hostdata=%s"},
83         {Opt_spectator, "spectator"},
84         {Opt_ignore_local_fs, "ignore_local_fs"},
85         {Opt_localflocks, "localflocks"},
86         {Opt_localcaching, "localcaching"},
87         {Opt_debug, "debug"},
88         {Opt_nodebug, "nodebug"},
89         {Opt_upgrade, "upgrade"},
90         {Opt_acl, "acl"},
91         {Opt_noacl, "noacl"},
92         {Opt_quota_off, "quota=off"},
93         {Opt_quota_account, "quota=account"},
94         {Opt_quota_on, "quota=on"},
95         {Opt_quota, "quota"},
96         {Opt_noquota, "noquota"},
97         {Opt_suiddir, "suiddir"},
98         {Opt_nosuiddir, "nosuiddir"},
99         {Opt_data_writeback, "data=writeback"},
100         {Opt_data_ordered, "data=ordered"},
101         {Opt_meta, "meta"},
102         {Opt_discard, "discard"},
103         {Opt_nodiscard, "nodiscard"},
104         {Opt_commit, "commit=%d"},
105         {Opt_err_withdraw, "errors=withdraw"},
106         {Opt_err_panic, "errors=panic"},
107         {Opt_statfs_quantum, "statfs_quantum=%d"},
108         {Opt_statfs_percent, "statfs_percent=%d"},
109         {Opt_quota_quantum, "quota_quantum=%d"},
110         {Opt_error, NULL}
111 };
112
113 /**
114  * gfs2_mount_args - Parse mount options
115  * @args: The structure into which the parsed options will be written
116  * @options: The options to parse
117  *
118  * Return: errno
119  */
120
121 int gfs2_mount_args(struct gfs2_args *args, char *options)
122 {
123         char *o;
124         int token;
125         substring_t tmp[MAX_OPT_ARGS];
126         int rv;
127
128         /* Split the options into tokens with the "," character and
129            process them */
130
131         while (1) {
132                 o = strsep(&options, ",");
133                 if (o == NULL)
134                         break;
135                 if (*o == '\0')
136                         continue;
137
138                 token = match_token(o, tokens, tmp);
139                 switch (token) {
140                 case Opt_lockproto:
141                         match_strlcpy(args->ar_lockproto, &tmp[0],
142                                       GFS2_LOCKNAME_LEN);
143                         break;
144                 case Opt_locktable:
145                         match_strlcpy(args->ar_locktable, &tmp[0],
146                                       GFS2_LOCKNAME_LEN);
147                         break;
148                 case Opt_hostdata:
149                         match_strlcpy(args->ar_hostdata, &tmp[0],
150                                       GFS2_LOCKNAME_LEN);
151                         break;
152                 case Opt_spectator:
153                         args->ar_spectator = 1;
154                         break;
155                 case Opt_ignore_local_fs:
156                         args->ar_ignore_local_fs = 1;
157                         break;
158                 case Opt_localflocks:
159                         args->ar_localflocks = 1;
160                         break;
161                 case Opt_localcaching:
162                         args->ar_localcaching = 1;
163                         break;
164                 case Opt_debug:
165                         if (args->ar_errors == GFS2_ERRORS_PANIC) {
166                                 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
167                                        "are mutually exclusive.\n");
168                                 return -EINVAL;
169                         }
170                         args->ar_debug = 1;
171                         break;
172                 case Opt_nodebug:
173                         args->ar_debug = 0;
174                         break;
175                 case Opt_upgrade:
176                         args->ar_upgrade = 1;
177                         break;
178                 case Opt_acl:
179                         args->ar_posix_acl = 1;
180                         break;
181                 case Opt_noacl:
182                         args->ar_posix_acl = 0;
183                         break;
184                 case Opt_quota_off:
185                 case Opt_noquota:
186                         args->ar_quota = GFS2_QUOTA_OFF;
187                         break;
188                 case Opt_quota_account:
189                         args->ar_quota = GFS2_QUOTA_ACCOUNT;
190                         break;
191                 case Opt_quota_on:
192                 case Opt_quota:
193                         args->ar_quota = GFS2_QUOTA_ON;
194                         break;
195                 case Opt_suiddir:
196                         args->ar_suiddir = 1;
197                         break;
198                 case Opt_nosuiddir:
199                         args->ar_suiddir = 0;
200                         break;
201                 case Opt_data_writeback:
202                         args->ar_data = GFS2_DATA_WRITEBACK;
203                         break;
204                 case Opt_data_ordered:
205                         args->ar_data = GFS2_DATA_ORDERED;
206                         break;
207                 case Opt_meta:
208                         args->ar_meta = 1;
209                         break;
210                 case Opt_discard:
211                         args->ar_discard = 1;
212                         break;
213                 case Opt_nodiscard:
214                         args->ar_discard = 0;
215                         break;
216                 case Opt_commit:
217                         rv = match_int(&tmp[0], &args->ar_commit);
218                         if (rv || args->ar_commit <= 0) {
219                                 printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n");
220                                 return rv ? rv : -EINVAL;
221                         }
222                         break;
223                 case Opt_statfs_quantum:
224                         rv = match_int(&tmp[0], &args->ar_statfs_quantum);
225                         if (rv || args->ar_statfs_quantum < 0) {
226                                 printk(KERN_WARNING "GFS2: statfs_quantum mount option requires a non-negative numeric argument\n");
227                                 return rv ? rv : -EINVAL;
228                         }
229                         break;
230                 case Opt_quota_quantum:
231                         rv = match_int(&tmp[0], &args->ar_quota_quantum);
232                         if (rv || args->ar_quota_quantum <= 0) {
233                                 printk(KERN_WARNING "GFS2: quota_quantum mount option requires a positive numeric argument\n");
234                                 return rv ? rv : -EINVAL;
235                         }
236                         break;
237                 case Opt_statfs_percent:
238                         rv = match_int(&tmp[0], &args->ar_statfs_percent);
239                         if (rv || args->ar_statfs_percent < 0 ||
240                             args->ar_statfs_percent > 100) {
241                                 printk(KERN_WARNING "statfs_percent mount option requires a numeric argument between 0 and 100\n");
242                                 return rv ? rv : -EINVAL;
243                         }
244                         break;
245                 case Opt_err_withdraw:
246                         args->ar_errors = GFS2_ERRORS_WITHDRAW;
247                         break;
248                 case Opt_err_panic:
249                         if (args->ar_debug) {
250                                 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
251                                         "are mutually exclusive.\n");
252                                 return -EINVAL;
253                         }
254                         args->ar_errors = GFS2_ERRORS_PANIC;
255                         break;
256                 case Opt_error:
257                 default:
258                         printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o);
259                         return -EINVAL;
260                 }
261         }
262
263         return 0;
264 }
265
266 /**
267  * gfs2_jindex_free - Clear all the journal index information
268  * @sdp: The GFS2 superblock
269  *
270  */
271
272 void gfs2_jindex_free(struct gfs2_sbd *sdp)
273 {
274         struct list_head list, *head;
275         struct gfs2_jdesc *jd;
276         struct gfs2_journal_extent *jext;
277
278         spin_lock(&sdp->sd_jindex_spin);
279         list_add(&list, &sdp->sd_jindex_list);
280         list_del_init(&sdp->sd_jindex_list);
281         sdp->sd_journals = 0;
282         spin_unlock(&sdp->sd_jindex_spin);
283
284         while (!list_empty(&list)) {
285                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
286                 head = &jd->extent_list;
287                 while (!list_empty(head)) {
288                         jext = list_entry(head->next,
289                                           struct gfs2_journal_extent,
290                                           extent_list);
291                         list_del(&jext->extent_list);
292                         kfree(jext);
293                 }
294                 list_del(&jd->jd_list);
295                 iput(jd->jd_inode);
296                 kfree(jd);
297         }
298 }
299
300 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
301 {
302         struct gfs2_jdesc *jd;
303         int found = 0;
304
305         list_for_each_entry(jd, head, jd_list) {
306                 if (jd->jd_jid == jid) {
307                         found = 1;
308                         break;
309                 }
310         }
311
312         if (!found)
313                 jd = NULL;
314
315         return jd;
316 }
317
318 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
319 {
320         struct gfs2_jdesc *jd;
321
322         spin_lock(&sdp->sd_jindex_spin);
323         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
324         spin_unlock(&sdp->sd_jindex_spin);
325
326         return jd;
327 }
328
329 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
330 {
331         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
332         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
333         int ar;
334         int error;
335
336         if (ip->i_disksize < (8 << 20) || ip->i_disksize > (1 << 30) ||
337             (ip->i_disksize & (sdp->sd_sb.sb_bsize - 1))) {
338                 gfs2_consist_inode(ip);
339                 return -EIO;
340         }
341         jd->jd_blocks = ip->i_disksize >> sdp->sd_sb.sb_bsize_shift;
342
343         error = gfs2_write_alloc_required(ip, 0, ip->i_disksize, &ar);
344         if (!error && ar) {
345                 gfs2_consist_inode(ip);
346                 error = -EIO;
347         }
348
349         return error;
350 }
351
352 /**
353  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
354  * @sdp: the filesystem
355  *
356  * Returns: errno
357  */
358
359 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
360 {
361         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
362         struct gfs2_glock *j_gl = ip->i_gl;
363         struct gfs2_holder t_gh;
364         struct gfs2_log_header_host head;
365         int error;
366
367         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
368         if (error)
369                 return error;
370
371         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
372
373         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
374         if (error)
375                 goto fail;
376
377         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
378                 gfs2_consist(sdp);
379                 error = -EIO;
380                 goto fail;
381         }
382
383         /*  Initialize some head of the log stuff  */
384         sdp->sd_log_sequence = head.lh_sequence + 1;
385         gfs2_log_pointers_init(sdp, head.lh_blkno);
386
387         error = gfs2_quota_init(sdp);
388         if (error)
389                 goto fail;
390
391         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
392
393         gfs2_glock_dq_uninit(&t_gh);
394
395         return 0;
396
397 fail:
398         t_gh.gh_flags |= GL_NOCACHE;
399         gfs2_glock_dq_uninit(&t_gh);
400
401         return error;
402 }
403
404 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
405 {
406         const struct gfs2_statfs_change *str = buf;
407
408         sc->sc_total = be64_to_cpu(str->sc_total);
409         sc->sc_free = be64_to_cpu(str->sc_free);
410         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
411 }
412
413 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
414 {
415         struct gfs2_statfs_change *str = buf;
416
417         str->sc_total = cpu_to_be64(sc->sc_total);
418         str->sc_free = cpu_to_be64(sc->sc_free);
419         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
420 }
421
422 int gfs2_statfs_init(struct gfs2_sbd *sdp)
423 {
424         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
425         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
426         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
427         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
428         struct buffer_head *m_bh, *l_bh;
429         struct gfs2_holder gh;
430         int error;
431
432         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
433                                    &gh);
434         if (error)
435                 return error;
436
437         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
438         if (error)
439                 goto out;
440
441         if (sdp->sd_args.ar_spectator) {
442                 spin_lock(&sdp->sd_statfs_spin);
443                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
444                                       sizeof(struct gfs2_dinode));
445                 spin_unlock(&sdp->sd_statfs_spin);
446         } else {
447                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
448                 if (error)
449                         goto out_m_bh;
450
451                 spin_lock(&sdp->sd_statfs_spin);
452                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
453                                       sizeof(struct gfs2_dinode));
454                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
455                                       sizeof(struct gfs2_dinode));
456                 spin_unlock(&sdp->sd_statfs_spin);
457
458                 brelse(l_bh);
459         }
460
461 out_m_bh:
462         brelse(m_bh);
463 out:
464         gfs2_glock_dq_uninit(&gh);
465         return 0;
466 }
467
468 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
469                         s64 dinodes)
470 {
471         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
472         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
473         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
474         struct buffer_head *l_bh;
475         int percent, sync_percent;
476         int error;
477
478         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
479         if (error)
480                 return;
481
482         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
483
484         spin_lock(&sdp->sd_statfs_spin);
485         l_sc->sc_total += total;
486         l_sc->sc_free += free;
487         l_sc->sc_dinodes += dinodes;
488         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
489         if (m_sc->sc_free)
490                 percent = (100 * l_sc->sc_free) / m_sc->sc_free;
491         else
492                 percent = 100;
493         spin_unlock(&sdp->sd_statfs_spin);
494
495         brelse(l_bh);
496         sync_percent = sdp->sd_args.ar_statfs_percent;
497         if (sync_percent && (percent >= sync_percent ||
498                              percent <= -sync_percent))
499                 gfs2_wake_up_statfs(sdp);
500 }
501
502 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
503                    struct buffer_head *l_bh)
504 {
505         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
506         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
507         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
508         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
509
510         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
511
512         spin_lock(&sdp->sd_statfs_spin);
513         m_sc->sc_total += l_sc->sc_total;
514         m_sc->sc_free += l_sc->sc_free;
515         m_sc->sc_dinodes += l_sc->sc_dinodes;
516         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
517         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
518                0, sizeof(struct gfs2_statfs_change));
519         spin_unlock(&sdp->sd_statfs_spin);
520
521         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
522         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
523 }
524
525 int gfs2_statfs_sync(struct super_block *sb, int type)
526 {
527         struct gfs2_sbd *sdp = sb->s_fs_info;
528         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
529         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
530         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
531         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
532         struct gfs2_holder gh;
533         struct buffer_head *m_bh, *l_bh;
534         int error;
535
536         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
537                                    &gh);
538         if (error)
539                 return error;
540
541         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
542         if (error)
543                 goto out;
544
545         spin_lock(&sdp->sd_statfs_spin);
546         gfs2_statfs_change_in(m_sc, m_bh->b_data +
547                               sizeof(struct gfs2_dinode));
548         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
549                 spin_unlock(&sdp->sd_statfs_spin);
550                 goto out_bh;
551         }
552         spin_unlock(&sdp->sd_statfs_spin);
553
554         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
555         if (error)
556                 goto out_bh;
557
558         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
559         if (error)
560                 goto out_bh2;
561
562         update_statfs(sdp, m_bh, l_bh);
563         sdp->sd_statfs_force_sync = 0;
564
565         gfs2_trans_end(sdp);
566
567 out_bh2:
568         brelse(l_bh);
569 out_bh:
570         brelse(m_bh);
571 out:
572         gfs2_glock_dq_uninit(&gh);
573         return error;
574 }
575
576 struct lfcc {
577         struct list_head list;
578         struct gfs2_holder gh;
579 };
580
581 /**
582  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
583  *                            journals are clean
584  * @sdp: the file system
585  * @state: the state to put the transaction lock into
586  * @t_gh: the hold on the transaction lock
587  *
588  * Returns: errno
589  */
590
591 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
592                                     struct gfs2_holder *t_gh)
593 {
594         struct gfs2_inode *ip;
595         struct gfs2_jdesc *jd;
596         struct lfcc *lfcc;
597         LIST_HEAD(list);
598         struct gfs2_log_header_host lh;
599         int error;
600
601         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
602                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
603                 if (!lfcc) {
604                         error = -ENOMEM;
605                         goto out;
606                 }
607                 ip = GFS2_I(jd->jd_inode);
608                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
609                 if (error) {
610                         kfree(lfcc);
611                         goto out;
612                 }
613                 list_add(&lfcc->list, &list);
614         }
615
616         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
617                                    GL_NOCACHE, t_gh);
618
619         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
620                 error = gfs2_jdesc_check(jd);
621                 if (error)
622                         break;
623                 error = gfs2_find_jhead(jd, &lh);
624                 if (error)
625                         break;
626                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
627                         error = -EBUSY;
628                         break;
629                 }
630         }
631
632         if (error)
633                 gfs2_glock_dq_uninit(t_gh);
634
635 out:
636         while (!list_empty(&list)) {
637                 lfcc = list_entry(list.next, struct lfcc, list);
638                 list_del(&lfcc->list);
639                 gfs2_glock_dq_uninit(&lfcc->gh);
640                 kfree(lfcc);
641         }
642         return error;
643 }
644
645 /**
646  * gfs2_freeze_fs - freezes the file system
647  * @sdp: the file system
648  *
649  * This function flushes data and meta data for all machines by
650  * aquiring the transaction log exclusively.  All journals are
651  * ensured to be in a clean state as well.
652  *
653  * Returns: errno
654  */
655
656 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
657 {
658         int error = 0;
659
660         mutex_lock(&sdp->sd_freeze_lock);
661
662         if (!sdp->sd_freeze_count++) {
663                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
664                 if (error)
665                         sdp->sd_freeze_count--;
666         }
667
668         mutex_unlock(&sdp->sd_freeze_lock);
669
670         return error;
671 }
672
673 /**
674  * gfs2_unfreeze_fs - unfreezes the file system
675  * @sdp: the file system
676  *
677  * This function allows the file system to proceed by unlocking
678  * the exclusively held transaction lock.  Other GFS2 nodes are
679  * now free to acquire the lock shared and go on with their lives.
680  *
681  */
682
683 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
684 {
685         mutex_lock(&sdp->sd_freeze_lock);
686
687         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
688                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
689
690         mutex_unlock(&sdp->sd_freeze_lock);
691 }
692
693
694 /**
695  * gfs2_write_inode - Make sure the inode is stable on the disk
696  * @inode: The inode
697  * @sync: synchronous write flag
698  *
699  * Returns: errno
700  */
701
702 static int gfs2_write_inode(struct inode *inode, int sync)
703 {
704         struct gfs2_inode *ip = GFS2_I(inode);
705         struct gfs2_sbd *sdp = GFS2_SB(inode);
706         struct gfs2_holder gh;
707         struct buffer_head *bh;
708         struct timespec atime;
709         struct gfs2_dinode *di;
710         int ret = 0;
711
712         /* Check this is a "normal" inode, etc */
713         if (!test_bit(GIF_USER, &ip->i_flags) ||
714             (current->flags & PF_MEMALLOC))
715                 return 0;
716         ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
717         if (ret)
718                 goto do_flush;
719         ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
720         if (ret)
721                 goto do_unlock;
722         ret = gfs2_meta_inode_buffer(ip, &bh);
723         if (ret == 0) {
724                 di = (struct gfs2_dinode *)bh->b_data;
725                 atime.tv_sec = be64_to_cpu(di->di_atime);
726                 atime.tv_nsec = be32_to_cpu(di->di_atime_nsec);
727                 if (timespec_compare(&inode->i_atime, &atime) > 0) {
728                         gfs2_trans_add_bh(ip->i_gl, bh, 1);
729                         gfs2_dinode_out(ip, bh->b_data);
730                 }
731                 brelse(bh);
732         }
733         gfs2_trans_end(sdp);
734 do_unlock:
735         gfs2_glock_dq_uninit(&gh);
736 do_flush:
737         if (sync != 0)
738                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
739         return ret;
740 }
741
742 /**
743  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
744  * @sdp: the filesystem
745  *
746  * Returns: errno
747  */
748
749 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
750 {
751         struct gfs2_holder t_gh;
752         int error;
753
754         flush_workqueue(gfs2_delete_workqueue);
755         gfs2_quota_sync(sdp->sd_vfs, 0);
756         gfs2_statfs_sync(sdp->sd_vfs, 0);
757
758         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
759                                    &t_gh);
760         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
761                 return error;
762
763         gfs2_meta_syncfs(sdp);
764         gfs2_log_shutdown(sdp);
765
766         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
767
768         if (t_gh.gh_gl)
769                 gfs2_glock_dq_uninit(&t_gh);
770
771         gfs2_quota_cleanup(sdp);
772
773         return error;
774 }
775
776 static int gfs2_umount_recovery_wait(void *word)
777 {
778         schedule();
779         return 0;
780 }
781
782 /**
783  * gfs2_put_super - Unmount the filesystem
784  * @sb: The VFS superblock
785  *
786  */
787
788 static void gfs2_put_super(struct super_block *sb)
789 {
790         struct gfs2_sbd *sdp = sb->s_fs_info;
791         int error;
792         struct gfs2_jdesc *jd;
793
794         /*  Unfreeze the filesystem, if we need to  */
795
796         mutex_lock(&sdp->sd_freeze_lock);
797         if (sdp->sd_freeze_count)
798                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
799         mutex_unlock(&sdp->sd_freeze_lock);
800
801         /* No more recovery requests */
802         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
803         smp_mb();
804
805         /* Wait on outstanding recovery */
806 restart:
807         spin_lock(&sdp->sd_jindex_spin);
808         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
809                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
810                         continue;
811                 spin_unlock(&sdp->sd_jindex_spin);
812                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
813                             gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
814                 goto restart;
815         }
816         spin_unlock(&sdp->sd_jindex_spin);
817
818         kthread_stop(sdp->sd_quotad_process);
819         kthread_stop(sdp->sd_logd_process);
820
821         if (!(sb->s_flags & MS_RDONLY)) {
822                 error = gfs2_make_fs_ro(sdp);
823                 if (error)
824                         gfs2_io_error(sdp);
825         }
826         /*  At this point, we're through modifying the disk  */
827
828         /*  Release stuff  */
829
830         iput(sdp->sd_jindex);
831         iput(sdp->sd_statfs_inode);
832         iput(sdp->sd_rindex);
833         iput(sdp->sd_quota_inode);
834
835         gfs2_glock_put(sdp->sd_rename_gl);
836         gfs2_glock_put(sdp->sd_trans_gl);
837
838         if (!sdp->sd_args.ar_spectator) {
839                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
840                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
841                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
842                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
843                 iput(sdp->sd_sc_inode);
844                 iput(sdp->sd_qc_inode);
845         }
846
847         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
848         gfs2_clear_rgrpd(sdp);
849         gfs2_jindex_free(sdp);
850         /*  Take apart glock structures and buffer lists  */
851         gfs2_gl_hash_clear(sdp);
852         /*  Unmount the locking protocol  */
853         gfs2_lm_unmount(sdp);
854
855         /*  At this point, we're through participating in the lockspace  */
856         gfs2_sys_fs_del(sdp);
857 }
858
859 /**
860  * gfs2_sync_fs - sync the filesystem
861  * @sb: the superblock
862  *
863  * Flushes the log to disk.
864  */
865
866 static int gfs2_sync_fs(struct super_block *sb, int wait)
867 {
868         if (wait && sb->s_fs_info)
869                 gfs2_log_flush(sb->s_fs_info, NULL);
870         return 0;
871 }
872
873 /**
874  * gfs2_freeze - prevent further writes to the filesystem
875  * @sb: the VFS structure for the filesystem
876  *
877  */
878
879 static int gfs2_freeze(struct super_block *sb)
880 {
881         struct gfs2_sbd *sdp = sb->s_fs_info;
882         int error;
883
884         if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
885                 return -EINVAL;
886
887         for (;;) {
888                 error = gfs2_freeze_fs(sdp);
889                 if (!error)
890                         break;
891
892                 switch (error) {
893                 case -EBUSY:
894                         fs_err(sdp, "waiting for recovery before freeze\n");
895                         break;
896
897                 default:
898                         fs_err(sdp, "error freezing FS: %d\n", error);
899                         break;
900                 }
901
902                 fs_err(sdp, "retrying...\n");
903                 msleep(1000);
904         }
905         return 0;
906 }
907
908 /**
909  * gfs2_unfreeze - reallow writes to the filesystem
910  * @sb: the VFS structure for the filesystem
911  *
912  */
913
914 static int gfs2_unfreeze(struct super_block *sb)
915 {
916         gfs2_unfreeze_fs(sb->s_fs_info);
917         return 0;
918 }
919
920 /**
921  * statfs_fill - fill in the sg for a given RG
922  * @rgd: the RG
923  * @sc: the sc structure
924  *
925  * Returns: 0 on success, -ESTALE if the LVB is invalid
926  */
927
928 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
929                             struct gfs2_statfs_change_host *sc)
930 {
931         gfs2_rgrp_verify(rgd);
932         sc->sc_total += rgd->rd_data;
933         sc->sc_free += rgd->rd_free;
934         sc->sc_dinodes += rgd->rd_dinodes;
935         return 0;
936 }
937
938 /**
939  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
940  * @sdp: the filesystem
941  * @sc: the sc info that will be returned
942  *
943  * Any error (other than a signal) will cause this routine to fall back
944  * to the synchronous version.
945  *
946  * FIXME: This really shouldn't busy wait like this.
947  *
948  * Returns: errno
949  */
950
951 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
952 {
953         struct gfs2_holder ri_gh;
954         struct gfs2_rgrpd *rgd_next;
955         struct gfs2_holder *gha, *gh;
956         unsigned int slots = 64;
957         unsigned int x;
958         int done;
959         int error = 0, err;
960
961         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
962         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
963         if (!gha)
964                 return -ENOMEM;
965
966         error = gfs2_rindex_hold(sdp, &ri_gh);
967         if (error)
968                 goto out;
969
970         rgd_next = gfs2_rgrpd_get_first(sdp);
971
972         for (;;) {
973                 done = 1;
974
975                 for (x = 0; x < slots; x++) {
976                         gh = gha + x;
977
978                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
979                                 err = gfs2_glock_wait(gh);
980                                 if (err) {
981                                         gfs2_holder_uninit(gh);
982                                         error = err;
983                                 } else {
984                                         if (!error)
985                                                 error = statfs_slow_fill(
986                                                         gh->gh_gl->gl_object, sc);
987                                         gfs2_glock_dq_uninit(gh);
988                                 }
989                         }
990
991                         if (gh->gh_gl)
992                                 done = 0;
993                         else if (rgd_next && !error) {
994                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
995                                                            LM_ST_SHARED,
996                                                            GL_ASYNC,
997                                                            gh);
998                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
999                                 done = 0;
1000                         }
1001
1002                         if (signal_pending(current))
1003                                 error = -ERESTARTSYS;
1004                 }
1005
1006                 if (done)
1007                         break;
1008
1009                 yield();
1010         }
1011
1012         gfs2_glock_dq_uninit(&ri_gh);
1013
1014 out:
1015         kfree(gha);
1016         return error;
1017 }
1018
1019 /**
1020  * gfs2_statfs_i - Do a statfs
1021  * @sdp: the filesystem
1022  * @sg: the sg structure
1023  *
1024  * Returns: errno
1025  */
1026
1027 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1028 {
1029         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1030         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1031
1032         spin_lock(&sdp->sd_statfs_spin);
1033
1034         *sc = *m_sc;
1035         sc->sc_total += l_sc->sc_total;
1036         sc->sc_free += l_sc->sc_free;
1037         sc->sc_dinodes += l_sc->sc_dinodes;
1038
1039         spin_unlock(&sdp->sd_statfs_spin);
1040
1041         if (sc->sc_free < 0)
1042                 sc->sc_free = 0;
1043         if (sc->sc_free > sc->sc_total)
1044                 sc->sc_free = sc->sc_total;
1045         if (sc->sc_dinodes < 0)
1046                 sc->sc_dinodes = 0;
1047
1048         return 0;
1049 }
1050
1051 /**
1052  * gfs2_statfs - Gather and return stats about the filesystem
1053  * @sb: The superblock
1054  * @statfsbuf: The buffer
1055  *
1056  * Returns: 0 on success or error code
1057  */
1058
1059 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1060 {
1061         struct super_block *sb = dentry->d_inode->i_sb;
1062         struct gfs2_sbd *sdp = sb->s_fs_info;
1063         struct gfs2_statfs_change_host sc;
1064         int error;
1065
1066         if (gfs2_tune_get(sdp, gt_statfs_slow))
1067                 error = gfs2_statfs_slow(sdp, &sc);
1068         else
1069                 error = gfs2_statfs_i(sdp, &sc);
1070
1071         if (error)
1072                 return error;
1073
1074         buf->f_type = GFS2_MAGIC;
1075         buf->f_bsize = sdp->sd_sb.sb_bsize;
1076         buf->f_blocks = sc.sc_total;
1077         buf->f_bfree = sc.sc_free;
1078         buf->f_bavail = sc.sc_free;
1079         buf->f_files = sc.sc_dinodes + sc.sc_free;
1080         buf->f_ffree = sc.sc_free;
1081         buf->f_namelen = GFS2_FNAMESIZE;
1082
1083         return 0;
1084 }
1085
1086 /**
1087  * gfs2_remount_fs - called when the FS is remounted
1088  * @sb:  the filesystem
1089  * @flags:  the remount flags
1090  * @data:  extra data passed in (not used right now)
1091  *
1092  * Returns: errno
1093  */
1094
1095 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1096 {
1097         struct gfs2_sbd *sdp = sb->s_fs_info;
1098         struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1099         struct gfs2_tune *gt = &sdp->sd_tune;
1100         int error;
1101
1102         spin_lock(&gt->gt_spin);
1103         args.ar_commit = gt->gt_log_flush_secs;
1104         args.ar_quota_quantum = gt->gt_quota_quantum;
1105         if (gt->gt_statfs_slow)
1106                 args.ar_statfs_quantum = 0;
1107         else
1108                 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1109         spin_unlock(&gt->gt_spin);
1110         error = gfs2_mount_args(&args, data);
1111         if (error)
1112                 return error;
1113
1114         /* Not allowed to change locking details */
1115         if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1116             strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1117             strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1118                 return -EINVAL;
1119
1120         /* Some flags must not be changed */
1121         if (args_neq(&args, &sdp->sd_args, spectator) ||
1122             args_neq(&args, &sdp->sd_args, ignore_local_fs) ||
1123             args_neq(&args, &sdp->sd_args, localflocks) ||
1124             args_neq(&args, &sdp->sd_args, localcaching) ||
1125             args_neq(&args, &sdp->sd_args, meta))
1126                 return -EINVAL;
1127
1128         if (sdp->sd_args.ar_spectator)
1129                 *flags |= MS_RDONLY;
1130
1131         if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1132                 if (*flags & MS_RDONLY)
1133                         error = gfs2_make_fs_ro(sdp);
1134                 else
1135                         error = gfs2_make_fs_rw(sdp);
1136                 if (error)
1137                         return error;
1138         }
1139
1140         sdp->sd_args = args;
1141         if (sdp->sd_args.ar_posix_acl)
1142                 sb->s_flags |= MS_POSIXACL;
1143         else
1144                 sb->s_flags &= ~MS_POSIXACL;
1145         spin_lock(&gt->gt_spin);
1146         gt->gt_log_flush_secs = args.ar_commit;
1147         gt->gt_quota_quantum = args.ar_quota_quantum;
1148         if (args.ar_statfs_quantum) {
1149                 gt->gt_statfs_slow = 0;
1150                 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1151         }
1152         else {
1153                 gt->gt_statfs_slow = 1;
1154                 gt->gt_statfs_quantum = 30;
1155         }
1156         spin_unlock(&gt->gt_spin);
1157
1158         gfs2_online_uevent(sdp);
1159         return 0;
1160 }
1161
1162 /**
1163  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1164  * @inode: The inode to drop
1165  *
1166  * If we've received a callback on an iopen lock then its because a
1167  * remote node tried to deallocate the inode but failed due to this node
1168  * still having the inode open. Here we mark the link count zero
1169  * since we know that it must have reached zero if the GLF_DEMOTE flag
1170  * is set on the iopen glock. If we didn't do a disk read since the
1171  * remote node removed the final link then we might otherwise miss
1172  * this event. This check ensures that this node will deallocate the
1173  * inode's blocks, or alternatively pass the baton on to another
1174  * node for later deallocation.
1175  */
1176
1177 static void gfs2_drop_inode(struct inode *inode)
1178 {
1179         struct gfs2_inode *ip = GFS2_I(inode);
1180
1181         if (test_bit(GIF_USER, &ip->i_flags) && inode->i_nlink) {
1182                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1183                 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1184                         clear_nlink(inode);
1185         }
1186         generic_drop_inode(inode);
1187 }
1188
1189 /**
1190  * gfs2_clear_inode - Deallocate an inode when VFS is done with it
1191  * @inode: The VFS inode
1192  *
1193  */
1194
1195 static void gfs2_clear_inode(struct inode *inode)
1196 {
1197         struct gfs2_inode *ip = GFS2_I(inode);
1198
1199         /* This tells us its a "real" inode and not one which only
1200          * serves to contain an address space (see rgrp.c, meta_io.c)
1201          * which therefore doesn't have its own glocks.
1202          */
1203         if (test_bit(GIF_USER, &ip->i_flags)) {
1204                 ip->i_gl->gl_object = NULL;
1205                 gfs2_glock_put(ip->i_gl);
1206                 ip->i_gl = NULL;
1207                 if (ip->i_iopen_gh.gh_gl) {
1208                         ip->i_iopen_gh.gh_gl->gl_object = NULL;
1209                         gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1210                 }
1211         }
1212 }
1213
1214 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1215 {
1216         do {
1217                 if (d1 == d2)
1218                         return 1;
1219                 d1 = d1->d_parent;
1220         } while (!IS_ROOT(d1));
1221         return 0;
1222 }
1223
1224 /**
1225  * gfs2_show_options - Show mount options for /proc/mounts
1226  * @s: seq_file structure
1227  * @mnt: vfsmount
1228  *
1229  * Returns: 0 on success or error code
1230  */
1231
1232 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1233 {
1234         struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
1235         struct gfs2_args *args = &sdp->sd_args;
1236         int val;
1237
1238         if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir))
1239                 seq_printf(s, ",meta");
1240         if (args->ar_lockproto[0])
1241                 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1242         if (args->ar_locktable[0])
1243                 seq_printf(s, ",locktable=%s", args->ar_locktable);
1244         if (args->ar_hostdata[0])
1245                 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1246         if (args->ar_spectator)
1247                 seq_printf(s, ",spectator");
1248         if (args->ar_ignore_local_fs)
1249                 seq_printf(s, ",ignore_local_fs");
1250         if (args->ar_localflocks)
1251                 seq_printf(s, ",localflocks");
1252         if (args->ar_localcaching)
1253                 seq_printf(s, ",localcaching");
1254         if (args->ar_debug)
1255                 seq_printf(s, ",debug");
1256         if (args->ar_upgrade)
1257                 seq_printf(s, ",upgrade");
1258         if (args->ar_posix_acl)
1259                 seq_printf(s, ",acl");
1260         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1261                 char *state;
1262                 switch (args->ar_quota) {
1263                 case GFS2_QUOTA_OFF:
1264                         state = "off";
1265                         break;
1266                 case GFS2_QUOTA_ACCOUNT:
1267                         state = "account";
1268                         break;
1269                 case GFS2_QUOTA_ON:
1270                         state = "on";
1271                         break;
1272                 default:
1273                         state = "unknown";
1274                         break;
1275                 }
1276                 seq_printf(s, ",quota=%s", state);
1277         }
1278         if (args->ar_suiddir)
1279                 seq_printf(s, ",suiddir");
1280         if (args->ar_data != GFS2_DATA_DEFAULT) {
1281                 char *state;
1282                 switch (args->ar_data) {
1283                 case GFS2_DATA_WRITEBACK:
1284                         state = "writeback";
1285                         break;
1286                 case GFS2_DATA_ORDERED:
1287                         state = "ordered";
1288                         break;
1289                 default:
1290                         state = "unknown";
1291                         break;
1292                 }
1293                 seq_printf(s, ",data=%s", state);
1294         }
1295         if (args->ar_discard)
1296                 seq_printf(s, ",discard");
1297         val = sdp->sd_tune.gt_log_flush_secs;
1298         if (val != 60)
1299                 seq_printf(s, ",commit=%d", val);
1300         val = sdp->sd_tune.gt_statfs_quantum;
1301         if (val != 30)
1302                 seq_printf(s, ",statfs_quantum=%d", val);
1303         val = sdp->sd_tune.gt_quota_quantum;
1304         if (val != 60)
1305                 seq_printf(s, ",quota_quantum=%d", val);
1306         if (args->ar_statfs_percent)
1307                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1308         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1309                 const char *state;
1310
1311                 switch (args->ar_errors) {
1312                 case GFS2_ERRORS_WITHDRAW:
1313                         state = "withdraw";
1314                         break;
1315                 case GFS2_ERRORS_PANIC:
1316                         state = "panic";
1317                         break;
1318                 default:
1319                         state = "unknown";
1320                         break;
1321                 }
1322                 seq_printf(s, ",errors=%s", state);
1323         }
1324         return 0;
1325 }
1326
1327 /*
1328  * We have to (at the moment) hold the inodes main lock to cover
1329  * the gap between unlocking the shared lock on the iopen lock and
1330  * taking the exclusive lock. I'd rather do a shared -> exclusive
1331  * conversion on the iopen lock, but we can change that later. This
1332  * is safe, just less efficient.
1333  */
1334
1335 static void gfs2_delete_inode(struct inode *inode)
1336 {
1337         struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
1338         struct gfs2_inode *ip = GFS2_I(inode);
1339         struct gfs2_holder gh;
1340         int error;
1341
1342         if (!test_bit(GIF_USER, &ip->i_flags))
1343                 goto out;
1344
1345         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1346         if (unlikely(error)) {
1347                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1348                 goto out;
1349         }
1350
1351         error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1352         if (error)
1353                 goto out_truncate;
1354
1355         gfs2_glock_dq_wait(&ip->i_iopen_gh);
1356         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1357         error = gfs2_glock_nq(&ip->i_iopen_gh);
1358         if (error)
1359                 goto out_truncate;
1360
1361         if (S_ISDIR(inode->i_mode) &&
1362             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1363                 error = gfs2_dir_exhash_dealloc(ip);
1364                 if (error)
1365                         goto out_unlock;
1366         }
1367
1368         if (ip->i_eattr) {
1369                 error = gfs2_ea_dealloc(ip);
1370                 if (error)
1371                         goto out_unlock;
1372         }
1373
1374         if (!gfs2_is_stuffed(ip)) {
1375                 error = gfs2_file_dealloc(ip);
1376                 if (error)
1377                         goto out_unlock;
1378         }
1379
1380         error = gfs2_dinode_dealloc(ip);
1381         if (error)
1382                 goto out_unlock;
1383
1384 out_truncate:
1385         error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1386         if (error)
1387                 goto out_unlock;
1388         /* Needs to be done before glock release & also in a transaction */
1389         truncate_inode_pages(&inode->i_data, 0);
1390         gfs2_trans_end(sdp);
1391
1392 out_unlock:
1393         if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
1394                 gfs2_glock_dq(&ip->i_iopen_gh);
1395         gfs2_holder_uninit(&ip->i_iopen_gh);
1396         gfs2_glock_dq_uninit(&gh);
1397         if (error && error != GLR_TRYFAILED && error != -EROFS)
1398                 fs_warn(sdp, "gfs2_delete_inode: %d\n", error);
1399 out:
1400         truncate_inode_pages(&inode->i_data, 0);
1401         clear_inode(inode);
1402 }
1403
1404 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1405 {
1406         struct gfs2_inode *ip;
1407
1408         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1409         if (ip) {
1410                 ip->i_flags = 0;
1411                 ip->i_gl = NULL;
1412         }
1413         return &ip->i_inode;
1414 }
1415
1416 static void gfs2_destroy_inode(struct inode *inode)
1417 {
1418         kmem_cache_free(gfs2_inode_cachep, inode);
1419 }
1420
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         .delete_inode           = gfs2_delete_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         .clear_inode            = gfs2_clear_inode,
1433         .drop_inode             = gfs2_drop_inode,
1434         .show_options           = gfs2_show_options,
1435 };
1436