block: separate priority boosting from REQ_META
[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 #include <linux/writeback.h>
26 #include <linux/backing-dev.h>
27
28 #include "gfs2.h"
29 #include "incore.h"
30 #include "bmap.h"
31 #include "dir.h"
32 #include "glock.h"
33 #include "glops.h"
34 #include "inode.h"
35 #include "log.h"
36 #include "meta_io.h"
37 #include "quota.h"
38 #include "recovery.h"
39 #include "rgrp.h"
40 #include "super.h"
41 #include "trans.h"
42 #include "util.h"
43 #include "sys.h"
44 #include "xattr.h"
45
46 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
47
48 enum {
49         Opt_lockproto,
50         Opt_locktable,
51         Opt_hostdata,
52         Opt_spectator,
53         Opt_ignore_local_fs,
54         Opt_localflocks,
55         Opt_localcaching,
56         Opt_debug,
57         Opt_nodebug,
58         Opt_upgrade,
59         Opt_acl,
60         Opt_noacl,
61         Opt_quota_off,
62         Opt_quota_account,
63         Opt_quota_on,
64         Opt_quota,
65         Opt_noquota,
66         Opt_suiddir,
67         Opt_nosuiddir,
68         Opt_data_writeback,
69         Opt_data_ordered,
70         Opt_meta,
71         Opt_discard,
72         Opt_nodiscard,
73         Opt_commit,
74         Opt_err_withdraw,
75         Opt_err_panic,
76         Opt_statfs_quantum,
77         Opt_statfs_percent,
78         Opt_quota_quantum,
79         Opt_barrier,
80         Opt_nobarrier,
81         Opt_error,
82 };
83
84 static const match_table_t tokens = {
85         {Opt_lockproto, "lockproto=%s"},
86         {Opt_locktable, "locktable=%s"},
87         {Opt_hostdata, "hostdata=%s"},
88         {Opt_spectator, "spectator"},
89         {Opt_spectator, "norecovery"},
90         {Opt_ignore_local_fs, "ignore_local_fs"},
91         {Opt_localflocks, "localflocks"},
92         {Opt_localcaching, "localcaching"},
93         {Opt_debug, "debug"},
94         {Opt_nodebug, "nodebug"},
95         {Opt_upgrade, "upgrade"},
96         {Opt_acl, "acl"},
97         {Opt_noacl, "noacl"},
98         {Opt_quota_off, "quota=off"},
99         {Opt_quota_account, "quota=account"},
100         {Opt_quota_on, "quota=on"},
101         {Opt_quota, "quota"},
102         {Opt_noquota, "noquota"},
103         {Opt_suiddir, "suiddir"},
104         {Opt_nosuiddir, "nosuiddir"},
105         {Opt_data_writeback, "data=writeback"},
106         {Opt_data_ordered, "data=ordered"},
107         {Opt_meta, "meta"},
108         {Opt_discard, "discard"},
109         {Opt_nodiscard, "nodiscard"},
110         {Opt_commit, "commit=%d"},
111         {Opt_err_withdraw, "errors=withdraw"},
112         {Opt_err_panic, "errors=panic"},
113         {Opt_statfs_quantum, "statfs_quantum=%d"},
114         {Opt_statfs_percent, "statfs_percent=%d"},
115         {Opt_quota_quantum, "quota_quantum=%d"},
116         {Opt_barrier, "barrier"},
117         {Opt_nobarrier, "nobarrier"},
118         {Opt_error, NULL}
119 };
120
121 /**
122  * gfs2_mount_args - Parse mount options
123  * @args: The structure into which the parsed options will be written
124  * @options: The options to parse
125  *
126  * Return: errno
127  */
128
129 int gfs2_mount_args(struct gfs2_args *args, char *options)
130 {
131         char *o;
132         int token;
133         substring_t tmp[MAX_OPT_ARGS];
134         int rv;
135
136         /* Split the options into tokens with the "," character and
137            process them */
138
139         while (1) {
140                 o = strsep(&options, ",");
141                 if (o == NULL)
142                         break;
143                 if (*o == '\0')
144                         continue;
145
146                 token = match_token(o, tokens, tmp);
147                 switch (token) {
148                 case Opt_lockproto:
149                         match_strlcpy(args->ar_lockproto, &tmp[0],
150                                       GFS2_LOCKNAME_LEN);
151                         break;
152                 case Opt_locktable:
153                         match_strlcpy(args->ar_locktable, &tmp[0],
154                                       GFS2_LOCKNAME_LEN);
155                         break;
156                 case Opt_hostdata:
157                         match_strlcpy(args->ar_hostdata, &tmp[0],
158                                       GFS2_LOCKNAME_LEN);
159                         break;
160                 case Opt_spectator:
161                         args->ar_spectator = 1;
162                         break;
163                 case Opt_ignore_local_fs:
164                         /* Retained for backwards compat only */
165                         break;
166                 case Opt_localflocks:
167                         args->ar_localflocks = 1;
168                         break;
169                 case Opt_localcaching:
170                         /* Retained for backwards compat only */
171                         break;
172                 case Opt_debug:
173                         if (args->ar_errors == GFS2_ERRORS_PANIC) {
174                                 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
175                                        "are mutually exclusive.\n");
176                                 return -EINVAL;
177                         }
178                         args->ar_debug = 1;
179                         break;
180                 case Opt_nodebug:
181                         args->ar_debug = 0;
182                         break;
183                 case Opt_upgrade:
184                         /* Retained for backwards compat only */
185                         break;
186                 case Opt_acl:
187                         args->ar_posix_acl = 1;
188                         break;
189                 case Opt_noacl:
190                         args->ar_posix_acl = 0;
191                         break;
192                 case Opt_quota_off:
193                 case Opt_noquota:
194                         args->ar_quota = GFS2_QUOTA_OFF;
195                         break;
196                 case Opt_quota_account:
197                         args->ar_quota = GFS2_QUOTA_ACCOUNT;
198                         break;
199                 case Opt_quota_on:
200                 case Opt_quota:
201                         args->ar_quota = GFS2_QUOTA_ON;
202                         break;
203                 case Opt_suiddir:
204                         args->ar_suiddir = 1;
205                         break;
206                 case Opt_nosuiddir:
207                         args->ar_suiddir = 0;
208                         break;
209                 case Opt_data_writeback:
210                         args->ar_data = GFS2_DATA_WRITEBACK;
211                         break;
212                 case Opt_data_ordered:
213                         args->ar_data = GFS2_DATA_ORDERED;
214                         break;
215                 case Opt_meta:
216                         args->ar_meta = 1;
217                         break;
218                 case Opt_discard:
219                         args->ar_discard = 1;
220                         break;
221                 case Opt_nodiscard:
222                         args->ar_discard = 0;
223                         break;
224                 case Opt_commit:
225                         rv = match_int(&tmp[0], &args->ar_commit);
226                         if (rv || args->ar_commit <= 0) {
227                                 printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n");
228                                 return rv ? rv : -EINVAL;
229                         }
230                         break;
231                 case Opt_statfs_quantum:
232                         rv = match_int(&tmp[0], &args->ar_statfs_quantum);
233                         if (rv || args->ar_statfs_quantum < 0) {
234                                 printk(KERN_WARNING "GFS2: statfs_quantum mount option requires a non-negative numeric argument\n");
235                                 return rv ? rv : -EINVAL;
236                         }
237                         break;
238                 case Opt_quota_quantum:
239                         rv = match_int(&tmp[0], &args->ar_quota_quantum);
240                         if (rv || args->ar_quota_quantum <= 0) {
241                                 printk(KERN_WARNING "GFS2: quota_quantum mount option requires a positive numeric argument\n");
242                                 return rv ? rv : -EINVAL;
243                         }
244                         break;
245                 case Opt_statfs_percent:
246                         rv = match_int(&tmp[0], &args->ar_statfs_percent);
247                         if (rv || args->ar_statfs_percent < 0 ||
248                             args->ar_statfs_percent > 100) {
249                                 printk(KERN_WARNING "statfs_percent mount option requires a numeric argument between 0 and 100\n");
250                                 return rv ? rv : -EINVAL;
251                         }
252                         break;
253                 case Opt_err_withdraw:
254                         args->ar_errors = GFS2_ERRORS_WITHDRAW;
255                         break;
256                 case Opt_err_panic:
257                         if (args->ar_debug) {
258                                 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
259                                         "are mutually exclusive.\n");
260                                 return -EINVAL;
261                         }
262                         args->ar_errors = GFS2_ERRORS_PANIC;
263                         break;
264                 case Opt_barrier:
265                         args->ar_nobarrier = 0;
266                         break;
267                 case Opt_nobarrier:
268                         args->ar_nobarrier = 1;
269                         break;
270                 case Opt_error:
271                 default:
272                         printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o);
273                         return -EINVAL;
274                 }
275         }
276
277         return 0;
278 }
279
280 /**
281  * gfs2_jindex_free - Clear all the journal index information
282  * @sdp: The GFS2 superblock
283  *
284  */
285
286 void gfs2_jindex_free(struct gfs2_sbd *sdp)
287 {
288         struct list_head list, *head;
289         struct gfs2_jdesc *jd;
290         struct gfs2_journal_extent *jext;
291
292         spin_lock(&sdp->sd_jindex_spin);
293         list_add(&list, &sdp->sd_jindex_list);
294         list_del_init(&sdp->sd_jindex_list);
295         sdp->sd_journals = 0;
296         spin_unlock(&sdp->sd_jindex_spin);
297
298         while (!list_empty(&list)) {
299                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
300                 head = &jd->extent_list;
301                 while (!list_empty(head)) {
302                         jext = list_entry(head->next,
303                                           struct gfs2_journal_extent,
304                                           extent_list);
305                         list_del(&jext->extent_list);
306                         kfree(jext);
307                 }
308                 list_del(&jd->jd_list);
309                 iput(jd->jd_inode);
310                 kfree(jd);
311         }
312 }
313
314 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
315 {
316         struct gfs2_jdesc *jd;
317         int found = 0;
318
319         list_for_each_entry(jd, head, jd_list) {
320                 if (jd->jd_jid == jid) {
321                         found = 1;
322                         break;
323                 }
324         }
325
326         if (!found)
327                 jd = NULL;
328
329         return jd;
330 }
331
332 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
333 {
334         struct gfs2_jdesc *jd;
335
336         spin_lock(&sdp->sd_jindex_spin);
337         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
338         spin_unlock(&sdp->sd_jindex_spin);
339
340         return jd;
341 }
342
343 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
344 {
345         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
346         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
347         u64 size = i_size_read(jd->jd_inode);
348
349         if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, 1 << 30))
350                 return -EIO;
351
352         jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
353
354         if (gfs2_write_alloc_required(ip, 0, size)) {
355                 gfs2_consist_inode(ip);
356                 return -EIO;
357         }
358
359         return 0;
360 }
361
362 /**
363  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
364  * @sdp: the filesystem
365  *
366  * Returns: errno
367  */
368
369 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
370 {
371         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
372         struct gfs2_glock *j_gl = ip->i_gl;
373         struct gfs2_holder t_gh;
374         struct gfs2_log_header_host head;
375         int error;
376
377         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
378         if (error)
379                 return error;
380
381         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
382
383         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
384         if (error)
385                 goto fail;
386
387         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
388                 gfs2_consist(sdp);
389                 error = -EIO;
390                 goto fail;
391         }
392
393         /*  Initialize some head of the log stuff  */
394         sdp->sd_log_sequence = head.lh_sequence + 1;
395         gfs2_log_pointers_init(sdp, head.lh_blkno);
396
397         error = gfs2_quota_init(sdp);
398         if (error)
399                 goto fail;
400
401         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
402
403         gfs2_glock_dq_uninit(&t_gh);
404
405         return 0;
406
407 fail:
408         t_gh.gh_flags |= GL_NOCACHE;
409         gfs2_glock_dq_uninit(&t_gh);
410
411         return error;
412 }
413
414 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
415 {
416         const struct gfs2_statfs_change *str = buf;
417
418         sc->sc_total = be64_to_cpu(str->sc_total);
419         sc->sc_free = be64_to_cpu(str->sc_free);
420         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
421 }
422
423 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
424 {
425         struct gfs2_statfs_change *str = buf;
426
427         str->sc_total = cpu_to_be64(sc->sc_total);
428         str->sc_free = cpu_to_be64(sc->sc_free);
429         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
430 }
431
432 int gfs2_statfs_init(struct gfs2_sbd *sdp)
433 {
434         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
435         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
436         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
437         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
438         struct buffer_head *m_bh, *l_bh;
439         struct gfs2_holder gh;
440         int error;
441
442         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
443                                    &gh);
444         if (error)
445                 return error;
446
447         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
448         if (error)
449                 goto out;
450
451         if (sdp->sd_args.ar_spectator) {
452                 spin_lock(&sdp->sd_statfs_spin);
453                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
454                                       sizeof(struct gfs2_dinode));
455                 spin_unlock(&sdp->sd_statfs_spin);
456         } else {
457                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
458                 if (error)
459                         goto out_m_bh;
460
461                 spin_lock(&sdp->sd_statfs_spin);
462                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
463                                       sizeof(struct gfs2_dinode));
464                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
465                                       sizeof(struct gfs2_dinode));
466                 spin_unlock(&sdp->sd_statfs_spin);
467
468                 brelse(l_bh);
469         }
470
471 out_m_bh:
472         brelse(m_bh);
473 out:
474         gfs2_glock_dq_uninit(&gh);
475         return 0;
476 }
477
478 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
479                         s64 dinodes)
480 {
481         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
482         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
483         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
484         struct buffer_head *l_bh;
485         s64 x, y;
486         int need_sync = 0;
487         int error;
488
489         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
490         if (error)
491                 return;
492
493         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
494
495         spin_lock(&sdp->sd_statfs_spin);
496         l_sc->sc_total += total;
497         l_sc->sc_free += free;
498         l_sc->sc_dinodes += dinodes;
499         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
500         if (sdp->sd_args.ar_statfs_percent) {
501                 x = 100 * l_sc->sc_free;
502                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
503                 if (x >= y || x <= -y)
504                         need_sync = 1;
505         }
506         spin_unlock(&sdp->sd_statfs_spin);
507
508         brelse(l_bh);
509         if (need_sync)
510                 gfs2_wake_up_statfs(sdp);
511 }
512
513 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
514                    struct buffer_head *l_bh)
515 {
516         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
517         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
518         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
519         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
520
521         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
522
523         spin_lock(&sdp->sd_statfs_spin);
524         m_sc->sc_total += l_sc->sc_total;
525         m_sc->sc_free += l_sc->sc_free;
526         m_sc->sc_dinodes += l_sc->sc_dinodes;
527         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
528         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
529                0, sizeof(struct gfs2_statfs_change));
530         spin_unlock(&sdp->sd_statfs_spin);
531
532         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
533         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
534 }
535
536 int gfs2_statfs_sync(struct super_block *sb, int type)
537 {
538         struct gfs2_sbd *sdp = sb->s_fs_info;
539         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
540         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
541         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
542         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
543         struct gfs2_holder gh;
544         struct buffer_head *m_bh, *l_bh;
545         int error;
546
547         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
548                                    &gh);
549         if (error)
550                 return error;
551
552         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
553         if (error)
554                 goto out;
555
556         spin_lock(&sdp->sd_statfs_spin);
557         gfs2_statfs_change_in(m_sc, m_bh->b_data +
558                               sizeof(struct gfs2_dinode));
559         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
560                 spin_unlock(&sdp->sd_statfs_spin);
561                 goto out_bh;
562         }
563         spin_unlock(&sdp->sd_statfs_spin);
564
565         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
566         if (error)
567                 goto out_bh;
568
569         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
570         if (error)
571                 goto out_bh2;
572
573         update_statfs(sdp, m_bh, l_bh);
574         sdp->sd_statfs_force_sync = 0;
575
576         gfs2_trans_end(sdp);
577
578 out_bh2:
579         brelse(l_bh);
580 out_bh:
581         brelse(m_bh);
582 out:
583         gfs2_glock_dq_uninit(&gh);
584         return error;
585 }
586
587 struct lfcc {
588         struct list_head list;
589         struct gfs2_holder gh;
590 };
591
592 /**
593  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
594  *                            journals are clean
595  * @sdp: the file system
596  * @state: the state to put the transaction lock into
597  * @t_gh: the hold on the transaction lock
598  *
599  * Returns: errno
600  */
601
602 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
603                                     struct gfs2_holder *t_gh)
604 {
605         struct gfs2_inode *ip;
606         struct gfs2_jdesc *jd;
607         struct lfcc *lfcc;
608         LIST_HEAD(list);
609         struct gfs2_log_header_host lh;
610         int error;
611
612         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
613                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
614                 if (!lfcc) {
615                         error = -ENOMEM;
616                         goto out;
617                 }
618                 ip = GFS2_I(jd->jd_inode);
619                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
620                 if (error) {
621                         kfree(lfcc);
622                         goto out;
623                 }
624                 list_add(&lfcc->list, &list);
625         }
626
627         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
628                                    GL_NOCACHE, t_gh);
629
630         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
631                 error = gfs2_jdesc_check(jd);
632                 if (error)
633                         break;
634                 error = gfs2_find_jhead(jd, &lh);
635                 if (error)
636                         break;
637                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
638                         error = -EBUSY;
639                         break;
640                 }
641         }
642
643         if (error)
644                 gfs2_glock_dq_uninit(t_gh);
645
646 out:
647         while (!list_empty(&list)) {
648                 lfcc = list_entry(list.next, struct lfcc, list);
649                 list_del(&lfcc->list);
650                 gfs2_glock_dq_uninit(&lfcc->gh);
651                 kfree(lfcc);
652         }
653         return error;
654 }
655
656 /**
657  * gfs2_freeze_fs - freezes the file system
658  * @sdp: the file system
659  *
660  * This function flushes data and meta data for all machines by
661  * acquiring the transaction log exclusively.  All journals are
662  * ensured to be in a clean state as well.
663  *
664  * Returns: errno
665  */
666
667 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
668 {
669         int error = 0;
670
671         mutex_lock(&sdp->sd_freeze_lock);
672
673         if (!sdp->sd_freeze_count++) {
674                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
675                 if (error)
676                         sdp->sd_freeze_count--;
677         }
678
679         mutex_unlock(&sdp->sd_freeze_lock);
680
681         return error;
682 }
683
684 /**
685  * gfs2_unfreeze_fs - unfreezes the file system
686  * @sdp: the file system
687  *
688  * This function allows the file system to proceed by unlocking
689  * the exclusively held transaction lock.  Other GFS2 nodes are
690  * now free to acquire the lock shared and go on with their lives.
691  *
692  */
693
694 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
695 {
696         mutex_lock(&sdp->sd_freeze_lock);
697
698         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
699                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
700
701         mutex_unlock(&sdp->sd_freeze_lock);
702 }
703
704 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
705 {
706         struct gfs2_dinode *str = buf;
707
708         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
709         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
710         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
711         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
712         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
713         str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
714         str->di_uid = cpu_to_be32(ip->i_inode.i_uid);
715         str->di_gid = cpu_to_be32(ip->i_inode.i_gid);
716         str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
717         str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
718         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
719         str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
720         str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
721         str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
722
723         str->di_goal_meta = cpu_to_be64(ip->i_goal);
724         str->di_goal_data = cpu_to_be64(ip->i_goal);
725         str->di_generation = cpu_to_be64(ip->i_generation);
726
727         str->di_flags = cpu_to_be32(ip->i_diskflags);
728         str->di_height = cpu_to_be16(ip->i_height);
729         str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
730                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
731                                              GFS2_FORMAT_DE : 0);
732         str->di_depth = cpu_to_be16(ip->i_depth);
733         str->di_entries = cpu_to_be32(ip->i_entries);
734
735         str->di_eattr = cpu_to_be64(ip->i_eattr);
736         str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
737         str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
738         str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
739 }
740
741 /**
742  * gfs2_write_inode - Make sure the inode is stable on the disk
743  * @inode: The inode
744  * @wbc: The writeback control structure
745  *
746  * Returns: errno
747  */
748
749 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
750 {
751         struct gfs2_inode *ip = GFS2_I(inode);
752         struct gfs2_sbd *sdp = GFS2_SB(inode);
753         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
754         struct backing_dev_info *bdi = metamapping->backing_dev_info;
755         struct gfs2_holder gh;
756         struct buffer_head *bh;
757         struct timespec atime;
758         struct gfs2_dinode *di;
759         int ret = -EAGAIN;
760         int unlock_required = 0;
761
762         /* Skip timestamp update, if this is from a memalloc */
763         if (current->flags & PF_MEMALLOC)
764                 goto do_flush;
765         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
766                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
767                 if (ret)
768                         goto do_flush;
769                 unlock_required = 1;
770         }
771         ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
772         if (ret)
773                 goto do_unlock;
774         ret = gfs2_meta_inode_buffer(ip, &bh);
775         if (ret == 0) {
776                 di = (struct gfs2_dinode *)bh->b_data;
777                 atime.tv_sec = be64_to_cpu(di->di_atime);
778                 atime.tv_nsec = be32_to_cpu(di->di_atime_nsec);
779                 if (timespec_compare(&inode->i_atime, &atime) > 0) {
780                         gfs2_trans_add_bh(ip->i_gl, bh, 1);
781                         gfs2_dinode_out(ip, bh->b_data);
782                 }
783                 brelse(bh);
784         }
785         gfs2_trans_end(sdp);
786 do_unlock:
787         if (unlock_required)
788                 gfs2_glock_dq_uninit(&gh);
789 do_flush:
790         if (wbc->sync_mode == WB_SYNC_ALL)
791                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
792         filemap_fdatawrite(metamapping);
793         if (bdi->dirty_exceeded)
794                 gfs2_ail1_flush(sdp, wbc);
795         if (!ret && (wbc->sync_mode == WB_SYNC_ALL))
796                 ret = filemap_fdatawait(metamapping);
797         if (ret)
798                 mark_inode_dirty_sync(inode);
799         return ret;
800 }
801
802 /**
803  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
804  * @sdp: the filesystem
805  *
806  * Returns: errno
807  */
808
809 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
810 {
811         struct gfs2_holder t_gh;
812         int error;
813
814         flush_workqueue(gfs2_delete_workqueue);
815         gfs2_quota_sync(sdp->sd_vfs, 0, 1);
816         gfs2_statfs_sync(sdp->sd_vfs, 0);
817
818         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
819                                    &t_gh);
820         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
821                 return error;
822
823         gfs2_meta_syncfs(sdp);
824         gfs2_log_shutdown(sdp);
825
826         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
827
828         if (t_gh.gh_gl)
829                 gfs2_glock_dq_uninit(&t_gh);
830
831         gfs2_quota_cleanup(sdp);
832
833         return error;
834 }
835
836 static int gfs2_umount_recovery_wait(void *word)
837 {
838         schedule();
839         return 0;
840 }
841
842 /**
843  * gfs2_put_super - Unmount the filesystem
844  * @sb: The VFS superblock
845  *
846  */
847
848 static void gfs2_put_super(struct super_block *sb)
849 {
850         struct gfs2_sbd *sdp = sb->s_fs_info;
851         int error;
852         struct gfs2_jdesc *jd;
853
854         /*  Unfreeze the filesystem, if we need to  */
855
856         mutex_lock(&sdp->sd_freeze_lock);
857         if (sdp->sd_freeze_count)
858                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
859         mutex_unlock(&sdp->sd_freeze_lock);
860
861         /* No more recovery requests */
862         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
863         smp_mb();
864
865         /* Wait on outstanding recovery */
866 restart:
867         spin_lock(&sdp->sd_jindex_spin);
868         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
869                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
870                         continue;
871                 spin_unlock(&sdp->sd_jindex_spin);
872                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
873                             gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
874                 goto restart;
875         }
876         spin_unlock(&sdp->sd_jindex_spin);
877
878         kthread_stop(sdp->sd_quotad_process);
879         kthread_stop(sdp->sd_logd_process);
880
881         if (!(sb->s_flags & MS_RDONLY)) {
882                 error = gfs2_make_fs_ro(sdp);
883                 if (error)
884                         gfs2_io_error(sdp);
885         }
886         /*  At this point, we're through modifying the disk  */
887
888         /*  Release stuff  */
889
890         iput(sdp->sd_jindex);
891         iput(sdp->sd_statfs_inode);
892         iput(sdp->sd_rindex);
893         iput(sdp->sd_quota_inode);
894
895         gfs2_glock_put(sdp->sd_rename_gl);
896         gfs2_glock_put(sdp->sd_trans_gl);
897
898         if (!sdp->sd_args.ar_spectator) {
899                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
900                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
901                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
902                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
903                 iput(sdp->sd_sc_inode);
904                 iput(sdp->sd_qc_inode);
905         }
906
907         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
908         gfs2_clear_rgrpd(sdp);
909         gfs2_jindex_free(sdp);
910         /*  Take apart glock structures and buffer lists  */
911         gfs2_gl_hash_clear(sdp);
912         /*  Unmount the locking protocol  */
913         gfs2_lm_unmount(sdp);
914
915         /*  At this point, we're through participating in the lockspace  */
916         gfs2_sys_fs_del(sdp);
917 }
918
919 /**
920  * gfs2_sync_fs - sync the filesystem
921  * @sb: the superblock
922  *
923  * Flushes the log to disk.
924  */
925
926 static int gfs2_sync_fs(struct super_block *sb, int wait)
927 {
928         struct gfs2_sbd *sdp = sb->s_fs_info;
929         if (wait && sdp)
930                 gfs2_log_flush(sdp, NULL);
931         return 0;
932 }
933
934 /**
935  * gfs2_freeze - prevent further writes to the filesystem
936  * @sb: the VFS structure for the filesystem
937  *
938  */
939
940 static int gfs2_freeze(struct super_block *sb)
941 {
942         struct gfs2_sbd *sdp = sb->s_fs_info;
943         int error;
944
945         if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
946                 return -EINVAL;
947
948         for (;;) {
949                 error = gfs2_freeze_fs(sdp);
950                 if (!error)
951                         break;
952
953                 switch (error) {
954                 case -EBUSY:
955                         fs_err(sdp, "waiting for recovery before freeze\n");
956                         break;
957
958                 default:
959                         fs_err(sdp, "error freezing FS: %d\n", error);
960                         break;
961                 }
962
963                 fs_err(sdp, "retrying...\n");
964                 msleep(1000);
965         }
966         return 0;
967 }
968
969 /**
970  * gfs2_unfreeze - reallow writes to the filesystem
971  * @sb: the VFS structure for the filesystem
972  *
973  */
974
975 static int gfs2_unfreeze(struct super_block *sb)
976 {
977         gfs2_unfreeze_fs(sb->s_fs_info);
978         return 0;
979 }
980
981 /**
982  * statfs_fill - fill in the sg for a given RG
983  * @rgd: the RG
984  * @sc: the sc structure
985  *
986  * Returns: 0 on success, -ESTALE if the LVB is invalid
987  */
988
989 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
990                             struct gfs2_statfs_change_host *sc)
991 {
992         gfs2_rgrp_verify(rgd);
993         sc->sc_total += rgd->rd_data;
994         sc->sc_free += rgd->rd_free;
995         sc->sc_dinodes += rgd->rd_dinodes;
996         return 0;
997 }
998
999 /**
1000  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1001  * @sdp: the filesystem
1002  * @sc: the sc info that will be returned
1003  *
1004  * Any error (other than a signal) will cause this routine to fall back
1005  * to the synchronous version.
1006  *
1007  * FIXME: This really shouldn't busy wait like this.
1008  *
1009  * Returns: errno
1010  */
1011
1012 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1013 {
1014         struct gfs2_holder ri_gh;
1015         struct gfs2_rgrpd *rgd_next;
1016         struct gfs2_holder *gha, *gh;
1017         unsigned int slots = 64;
1018         unsigned int x;
1019         int done;
1020         int error = 0, err;
1021
1022         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1023         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1024         if (!gha)
1025                 return -ENOMEM;
1026
1027         error = gfs2_rindex_hold(sdp, &ri_gh);
1028         if (error)
1029                 goto out;
1030
1031         rgd_next = gfs2_rgrpd_get_first(sdp);
1032
1033         for (;;) {
1034                 done = 1;
1035
1036                 for (x = 0; x < slots; x++) {
1037                         gh = gha + x;
1038
1039                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
1040                                 err = gfs2_glock_wait(gh);
1041                                 if (err) {
1042                                         gfs2_holder_uninit(gh);
1043                                         error = err;
1044                                 } else {
1045                                         if (!error)
1046                                                 error = statfs_slow_fill(
1047                                                         gh->gh_gl->gl_object, sc);
1048                                         gfs2_glock_dq_uninit(gh);
1049                                 }
1050                         }
1051
1052                         if (gh->gh_gl)
1053                                 done = 0;
1054                         else if (rgd_next && !error) {
1055                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1056                                                            LM_ST_SHARED,
1057                                                            GL_ASYNC,
1058                                                            gh);
1059                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1060                                 done = 0;
1061                         }
1062
1063                         if (signal_pending(current))
1064                                 error = -ERESTARTSYS;
1065                 }
1066
1067                 if (done)
1068                         break;
1069
1070                 yield();
1071         }
1072
1073         gfs2_glock_dq_uninit(&ri_gh);
1074
1075 out:
1076         kfree(gha);
1077         return error;
1078 }
1079
1080 /**
1081  * gfs2_statfs_i - Do a statfs
1082  * @sdp: the filesystem
1083  * @sg: the sg structure
1084  *
1085  * Returns: errno
1086  */
1087
1088 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1089 {
1090         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1091         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1092
1093         spin_lock(&sdp->sd_statfs_spin);
1094
1095         *sc = *m_sc;
1096         sc->sc_total += l_sc->sc_total;
1097         sc->sc_free += l_sc->sc_free;
1098         sc->sc_dinodes += l_sc->sc_dinodes;
1099
1100         spin_unlock(&sdp->sd_statfs_spin);
1101
1102         if (sc->sc_free < 0)
1103                 sc->sc_free = 0;
1104         if (sc->sc_free > sc->sc_total)
1105                 sc->sc_free = sc->sc_total;
1106         if (sc->sc_dinodes < 0)
1107                 sc->sc_dinodes = 0;
1108
1109         return 0;
1110 }
1111
1112 /**
1113  * gfs2_statfs - Gather and return stats about the filesystem
1114  * @sb: The superblock
1115  * @statfsbuf: The buffer
1116  *
1117  * Returns: 0 on success or error code
1118  */
1119
1120 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1121 {
1122         struct super_block *sb = dentry->d_inode->i_sb;
1123         struct gfs2_sbd *sdp = sb->s_fs_info;
1124         struct gfs2_statfs_change_host sc;
1125         int error;
1126
1127         if (gfs2_tune_get(sdp, gt_statfs_slow))
1128                 error = gfs2_statfs_slow(sdp, &sc);
1129         else
1130                 error = gfs2_statfs_i(sdp, &sc);
1131
1132         if (error)
1133                 return error;
1134
1135         buf->f_type = GFS2_MAGIC;
1136         buf->f_bsize = sdp->sd_sb.sb_bsize;
1137         buf->f_blocks = sc.sc_total;
1138         buf->f_bfree = sc.sc_free;
1139         buf->f_bavail = sc.sc_free;
1140         buf->f_files = sc.sc_dinodes + sc.sc_free;
1141         buf->f_ffree = sc.sc_free;
1142         buf->f_namelen = GFS2_FNAMESIZE;
1143
1144         return 0;
1145 }
1146
1147 /**
1148  * gfs2_remount_fs - called when the FS is remounted
1149  * @sb:  the filesystem
1150  * @flags:  the remount flags
1151  * @data:  extra data passed in (not used right now)
1152  *
1153  * Returns: errno
1154  */
1155
1156 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1157 {
1158         struct gfs2_sbd *sdp = sb->s_fs_info;
1159         struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1160         struct gfs2_tune *gt = &sdp->sd_tune;
1161         int error;
1162
1163         spin_lock(&gt->gt_spin);
1164         args.ar_commit = gt->gt_logd_secs;
1165         args.ar_quota_quantum = gt->gt_quota_quantum;
1166         if (gt->gt_statfs_slow)
1167                 args.ar_statfs_quantum = 0;
1168         else
1169                 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1170         spin_unlock(&gt->gt_spin);
1171         error = gfs2_mount_args(&args, data);
1172         if (error)
1173                 return error;
1174
1175         /* Not allowed to change locking details */
1176         if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1177             strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1178             strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1179                 return -EINVAL;
1180
1181         /* Some flags must not be changed */
1182         if (args_neq(&args, &sdp->sd_args, spectator) ||
1183             args_neq(&args, &sdp->sd_args, localflocks) ||
1184             args_neq(&args, &sdp->sd_args, meta))
1185                 return -EINVAL;
1186
1187         if (sdp->sd_args.ar_spectator)
1188                 *flags |= MS_RDONLY;
1189
1190         if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1191                 if (*flags & MS_RDONLY)
1192                         error = gfs2_make_fs_ro(sdp);
1193                 else
1194                         error = gfs2_make_fs_rw(sdp);
1195                 if (error)
1196                         return error;
1197         }
1198
1199         sdp->sd_args = args;
1200         if (sdp->sd_args.ar_posix_acl)
1201                 sb->s_flags |= MS_POSIXACL;
1202         else
1203                 sb->s_flags &= ~MS_POSIXACL;
1204         if (sdp->sd_args.ar_nobarrier)
1205                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1206         else
1207                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1208         spin_lock(&gt->gt_spin);
1209         gt->gt_logd_secs = args.ar_commit;
1210         gt->gt_quota_quantum = args.ar_quota_quantum;
1211         if (args.ar_statfs_quantum) {
1212                 gt->gt_statfs_slow = 0;
1213                 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1214         }
1215         else {
1216                 gt->gt_statfs_slow = 1;
1217                 gt->gt_statfs_quantum = 30;
1218         }
1219         spin_unlock(&gt->gt_spin);
1220
1221         gfs2_online_uevent(sdp);
1222         return 0;
1223 }
1224
1225 /**
1226  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1227  * @inode: The inode to drop
1228  *
1229  * If we've received a callback on an iopen lock then its because a
1230  * remote node tried to deallocate the inode but failed due to this node
1231  * still having the inode open. Here we mark the link count zero
1232  * since we know that it must have reached zero if the GLF_DEMOTE flag
1233  * is set on the iopen glock. If we didn't do a disk read since the
1234  * remote node removed the final link then we might otherwise miss
1235  * this event. This check ensures that this node will deallocate the
1236  * inode's blocks, or alternatively pass the baton on to another
1237  * node for later deallocation.
1238  */
1239
1240 static int gfs2_drop_inode(struct inode *inode)
1241 {
1242         struct gfs2_inode *ip = GFS2_I(inode);
1243
1244         if (inode->i_nlink) {
1245                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1246                 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1247                         clear_nlink(inode);
1248         }
1249         return generic_drop_inode(inode);
1250 }
1251
1252 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1253 {
1254         do {
1255                 if (d1 == d2)
1256                         return 1;
1257                 d1 = d1->d_parent;
1258         } while (!IS_ROOT(d1));
1259         return 0;
1260 }
1261
1262 /**
1263  * gfs2_show_options - Show mount options for /proc/mounts
1264  * @s: seq_file structure
1265  * @mnt: vfsmount
1266  *
1267  * Returns: 0 on success or error code
1268  */
1269
1270 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1271 {
1272         struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
1273         struct gfs2_args *args = &sdp->sd_args;
1274         int val;
1275
1276         if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir))
1277                 seq_printf(s, ",meta");
1278         if (args->ar_lockproto[0])
1279                 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1280         if (args->ar_locktable[0])
1281                 seq_printf(s, ",locktable=%s", args->ar_locktable);
1282         if (args->ar_hostdata[0])
1283                 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1284         if (args->ar_spectator)
1285                 seq_printf(s, ",spectator");
1286         if (args->ar_localflocks)
1287                 seq_printf(s, ",localflocks");
1288         if (args->ar_debug)
1289                 seq_printf(s, ",debug");
1290         if (args->ar_posix_acl)
1291                 seq_printf(s, ",acl");
1292         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1293                 char *state;
1294                 switch (args->ar_quota) {
1295                 case GFS2_QUOTA_OFF:
1296                         state = "off";
1297                         break;
1298                 case GFS2_QUOTA_ACCOUNT:
1299                         state = "account";
1300                         break;
1301                 case GFS2_QUOTA_ON:
1302                         state = "on";
1303                         break;
1304                 default:
1305                         state = "unknown";
1306                         break;
1307                 }
1308                 seq_printf(s, ",quota=%s", state);
1309         }
1310         if (args->ar_suiddir)
1311                 seq_printf(s, ",suiddir");
1312         if (args->ar_data != GFS2_DATA_DEFAULT) {
1313                 char *state;
1314                 switch (args->ar_data) {
1315                 case GFS2_DATA_WRITEBACK:
1316                         state = "writeback";
1317                         break;
1318                 case GFS2_DATA_ORDERED:
1319                         state = "ordered";
1320                         break;
1321                 default:
1322                         state = "unknown";
1323                         break;
1324                 }
1325                 seq_printf(s, ",data=%s", state);
1326         }
1327         if (args->ar_discard)
1328                 seq_printf(s, ",discard");
1329         val = sdp->sd_tune.gt_logd_secs;
1330         if (val != 30)
1331                 seq_printf(s, ",commit=%d", val);
1332         val = sdp->sd_tune.gt_statfs_quantum;
1333         if (val != 30)
1334                 seq_printf(s, ",statfs_quantum=%d", val);
1335         val = sdp->sd_tune.gt_quota_quantum;
1336         if (val != 60)
1337                 seq_printf(s, ",quota_quantum=%d", val);
1338         if (args->ar_statfs_percent)
1339                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1340         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1341                 const char *state;
1342
1343                 switch (args->ar_errors) {
1344                 case GFS2_ERRORS_WITHDRAW:
1345                         state = "withdraw";
1346                         break;
1347                 case GFS2_ERRORS_PANIC:
1348                         state = "panic";
1349                         break;
1350                 default:
1351                         state = "unknown";
1352                         break;
1353                 }
1354                 seq_printf(s, ",errors=%s", state);
1355         }
1356         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1357                 seq_printf(s, ",nobarrier");
1358         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1359                 seq_printf(s, ",demote_interface_used");
1360         return 0;
1361 }
1362
1363 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1364 {
1365         struct inode *inode = &ip->i_inode;
1366         struct gfs2_glock *gl = ip->i_gl;
1367
1368         truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1369         truncate_inode_pages(&inode->i_data, 0);
1370
1371         if (atomic_read(&gl->gl_revokes) == 0) {
1372                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1373                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1374         }
1375 }
1376
1377 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1378 {
1379         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1380         struct gfs2_alloc *al;
1381         struct gfs2_rgrpd *rgd;
1382         int error;
1383
1384         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1385                 gfs2_consist_inode(ip);
1386                 return -EIO;
1387         }
1388
1389         al = gfs2_alloc_get(ip);
1390         if (!al)
1391                 return -ENOMEM;
1392
1393         error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1394         if (error)
1395                 goto out;
1396
1397         error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
1398         if (error)
1399                 goto out_qs;
1400
1401         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr);
1402         if (!rgd) {
1403                 gfs2_consist_inode(ip);
1404                 error = -EIO;
1405                 goto out_rindex_relse;
1406         }
1407
1408         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0,
1409                                    &al->al_rgd_gh);
1410         if (error)
1411                 goto out_rindex_relse;
1412
1413         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1414                                  sdp->sd_jdesc->jd_blocks);
1415         if (error)
1416                 goto out_rg_gunlock;
1417
1418         gfs2_free_di(rgd, ip);
1419
1420         gfs2_final_release_pages(ip);
1421
1422         gfs2_trans_end(sdp);
1423
1424 out_rg_gunlock:
1425         gfs2_glock_dq_uninit(&al->al_rgd_gh);
1426 out_rindex_relse:
1427         gfs2_glock_dq_uninit(&al->al_ri_gh);
1428 out_qs:
1429         gfs2_quota_unhold(ip);
1430 out:
1431         gfs2_alloc_put(ip);
1432         return error;
1433 }
1434
1435 /**
1436  * gfs2_evict_inode - Remove an inode from cache
1437  * @inode: The inode to evict
1438  *
1439  * There are three cases to consider:
1440  * 1. i_nlink == 0, we are final opener (and must deallocate)
1441  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1442  * 3. i_nlink > 0
1443  *
1444  * If the fs is read only, then we have to treat all cases as per #3
1445  * since we are unable to do any deallocation. The inode will be
1446  * deallocated by the next read/write node to attempt an allocation
1447  * in the same resource group
1448  *
1449  * We have to (at the moment) hold the inodes main lock to cover
1450  * the gap between unlocking the shared lock on the iopen lock and
1451  * taking the exclusive lock. I'd rather do a shared -> exclusive
1452  * conversion on the iopen lock, but we can change that later. This
1453  * is safe, just less efficient.
1454  */
1455
1456 static void gfs2_evict_inode(struct inode *inode)
1457 {
1458         struct super_block *sb = inode->i_sb;
1459         struct gfs2_sbd *sdp = sb->s_fs_info;
1460         struct gfs2_inode *ip = GFS2_I(inode);
1461         struct gfs2_holder gh;
1462         int error;
1463
1464         if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1465                 goto out;
1466
1467         /* Must not read inode block until block type has been verified */
1468         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1469         if (unlikely(error)) {
1470                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1471                 goto out;
1472         }
1473
1474         error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1475         if (error)
1476                 goto out_truncate;
1477
1478         if (test_bit(GIF_INVALID, &ip->i_flags)) {
1479                 error = gfs2_inode_refresh(ip);
1480                 if (error)
1481                         goto out_truncate;
1482         }
1483
1484         ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1485         gfs2_glock_dq_wait(&ip->i_iopen_gh);
1486         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1487         error = gfs2_glock_nq(&ip->i_iopen_gh);
1488         if (error)
1489                 goto out_truncate;
1490
1491         /* Case 1 starts here */
1492
1493         if (S_ISDIR(inode->i_mode) &&
1494             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1495                 error = gfs2_dir_exhash_dealloc(ip);
1496                 if (error)
1497                         goto out_unlock;
1498         }
1499
1500         if (ip->i_eattr) {
1501                 error = gfs2_ea_dealloc(ip);
1502                 if (error)
1503                         goto out_unlock;
1504         }
1505
1506         if (!gfs2_is_stuffed(ip)) {
1507                 error = gfs2_file_dealloc(ip);
1508                 if (error)
1509                         goto out_unlock;
1510         }
1511
1512         error = gfs2_dinode_dealloc(ip);
1513         goto out_unlock;
1514
1515 out_truncate:
1516         /* Case 2 starts here */
1517         error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1518         if (error)
1519                 goto out_unlock;
1520         /* Needs to be done before glock release & also in a transaction */
1521         truncate_inode_pages(&inode->i_data, 0);
1522         gfs2_trans_end(sdp);
1523
1524 out_unlock:
1525         /* Error path for case 1 */
1526         if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
1527                 gfs2_glock_dq(&ip->i_iopen_gh);
1528         gfs2_holder_uninit(&ip->i_iopen_gh);
1529         gfs2_glock_dq_uninit(&gh);
1530         if (error && error != GLR_TRYFAILED && error != -EROFS)
1531                 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1532 out:
1533         /* Case 3 starts here */
1534         truncate_inode_pages(&inode->i_data, 0);
1535         end_writeback(inode);
1536         gfs2_dir_hash_inval(ip);
1537         ip->i_gl->gl_object = NULL;
1538         gfs2_glock_add_to_lru(ip->i_gl);
1539         gfs2_glock_put(ip->i_gl);
1540         ip->i_gl = NULL;
1541         if (ip->i_iopen_gh.gh_gl) {
1542                 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1543                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1544         }
1545 }
1546
1547 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1548 {
1549         struct gfs2_inode *ip;
1550
1551         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1552         if (ip) {
1553                 ip->i_flags = 0;
1554                 ip->i_gl = NULL;
1555         }
1556         return &ip->i_inode;
1557 }
1558
1559 static void gfs2_i_callback(struct rcu_head *head)
1560 {
1561         struct inode *inode = container_of(head, struct inode, i_rcu);
1562         INIT_LIST_HEAD(&inode->i_dentry);
1563         kmem_cache_free(gfs2_inode_cachep, inode);
1564 }
1565
1566 static void gfs2_destroy_inode(struct inode *inode)
1567 {
1568         call_rcu(&inode->i_rcu, gfs2_i_callback);
1569 }
1570
1571 const struct super_operations gfs2_super_ops = {
1572         .alloc_inode            = gfs2_alloc_inode,
1573         .destroy_inode          = gfs2_destroy_inode,
1574         .write_inode            = gfs2_write_inode,
1575         .evict_inode            = gfs2_evict_inode,
1576         .put_super              = gfs2_put_super,
1577         .sync_fs                = gfs2_sync_fs,
1578         .freeze_fs              = gfs2_freeze,
1579         .unfreeze_fs            = gfs2_unfreeze,
1580         .statfs                 = gfs2_statfs,
1581         .remount_fs             = gfs2_remount_fs,
1582         .drop_inode             = gfs2_drop_inode,
1583         .show_options           = gfs2_show_options,
1584 };
1585