Loading...
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
48enum {
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
84static 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
129int 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
286void 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
314static 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
332struct 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
343int 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
369int 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
407fail:
408 t_gh.gh_flags |= GL_NOCACHE;
409 gfs2_glock_dq_uninit(&t_gh);
410
411 return error;
412}
413
414void 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
423static 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
432int 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
471out_m_bh:
472 brelse(m_bh);
473out:
474 gfs2_glock_dq_uninit(&gh);
475 return 0;
476}
477
478void 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
513void 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
536int 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
578out_bh2:
579 brelse(l_bh);
580out_bh:
581 brelse(m_bh);
582out:
583 gfs2_glock_dq_uninit(&gh);
584 return error;
585}
586
587struct 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
602static 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
646out:
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
667int 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
694void 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
704void 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
749static 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 int ret = 0;
756
757 if (wbc->sync_mode == WB_SYNC_ALL)
758 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
759 if (bdi->dirty_exceeded)
760 gfs2_ail1_flush(sdp, wbc);
761 else
762 filemap_fdatawrite(metamapping);
763 if (wbc->sync_mode == WB_SYNC_ALL)
764 ret = filemap_fdatawait(metamapping);
765 if (ret)
766 mark_inode_dirty_sync(inode);
767 return ret;
768}
769
770/**
771 * gfs2_dirty_inode - check for atime updates
772 * @inode: The inode in question
773 * @flags: The type of dirty
774 *
775 * Unfortunately it can be called under any combination of inode
776 * glock and transaction lock, so we have to check carefully.
777 *
778 * At the moment this deals only with atime - it should be possible
779 * to expand that role in future, once a review of the locking has
780 * been carried out.
781 */
782
783static void gfs2_dirty_inode(struct inode *inode, int flags)
784{
785 struct gfs2_inode *ip = GFS2_I(inode);
786 struct gfs2_sbd *sdp = GFS2_SB(inode);
787 struct buffer_head *bh;
788 struct gfs2_holder gh;
789 int need_unlock = 0;
790 int need_endtrans = 0;
791 int ret;
792
793 if (!(flags & (I_DIRTY_DATASYNC|I_DIRTY_SYNC)))
794 return;
795
796 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
797 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
798 if (ret) {
799 fs_err(sdp, "dirty_inode: glock %d\n", ret);
800 return;
801 }
802 need_unlock = 1;
803 }
804
805 if (current->journal_info == NULL) {
806 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
807 if (ret) {
808 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
809 goto out;
810 }
811 need_endtrans = 1;
812 }
813
814 ret = gfs2_meta_inode_buffer(ip, &bh);
815 if (ret == 0) {
816 gfs2_trans_add_bh(ip->i_gl, bh, 1);
817 gfs2_dinode_out(ip, bh->b_data);
818 brelse(bh);
819 }
820
821 if (need_endtrans)
822 gfs2_trans_end(sdp);
823out:
824 if (need_unlock)
825 gfs2_glock_dq_uninit(&gh);
826}
827
828/**
829 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
830 * @sdp: the filesystem
831 *
832 * Returns: errno
833 */
834
835static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
836{
837 struct gfs2_holder t_gh;
838 int error;
839
840 flush_workqueue(gfs2_delete_workqueue);
841 gfs2_quota_sync(sdp->sd_vfs, 0, 1);
842 gfs2_statfs_sync(sdp->sd_vfs, 0);
843
844 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
845 &t_gh);
846 if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
847 return error;
848
849 gfs2_meta_syncfs(sdp);
850 gfs2_log_shutdown(sdp);
851
852 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
853
854 if (t_gh.gh_gl)
855 gfs2_glock_dq_uninit(&t_gh);
856
857 gfs2_quota_cleanup(sdp);
858
859 return error;
860}
861
862static int gfs2_umount_recovery_wait(void *word)
863{
864 schedule();
865 return 0;
866}
867
868/**
869 * gfs2_put_super - Unmount the filesystem
870 * @sb: The VFS superblock
871 *
872 */
873
874static void gfs2_put_super(struct super_block *sb)
875{
876 struct gfs2_sbd *sdp = sb->s_fs_info;
877 int error;
878 struct gfs2_jdesc *jd;
879
880 /* Unfreeze the filesystem, if we need to */
881
882 mutex_lock(&sdp->sd_freeze_lock);
883 if (sdp->sd_freeze_count)
884 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
885 mutex_unlock(&sdp->sd_freeze_lock);
886
887 /* No more recovery requests */
888 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
889 smp_mb();
890
891 /* Wait on outstanding recovery */
892restart:
893 spin_lock(&sdp->sd_jindex_spin);
894 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
895 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
896 continue;
897 spin_unlock(&sdp->sd_jindex_spin);
898 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
899 gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
900 goto restart;
901 }
902 spin_unlock(&sdp->sd_jindex_spin);
903
904 kthread_stop(sdp->sd_quotad_process);
905 kthread_stop(sdp->sd_logd_process);
906
907 if (!(sb->s_flags & MS_RDONLY)) {
908 error = gfs2_make_fs_ro(sdp);
909 if (error)
910 gfs2_io_error(sdp);
911 }
912 /* At this point, we're through modifying the disk */
913
914 /* Release stuff */
915
916 iput(sdp->sd_jindex);
917 iput(sdp->sd_statfs_inode);
918 iput(sdp->sd_rindex);
919 iput(sdp->sd_quota_inode);
920
921 gfs2_glock_put(sdp->sd_rename_gl);
922 gfs2_glock_put(sdp->sd_trans_gl);
923
924 if (!sdp->sd_args.ar_spectator) {
925 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
926 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
927 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
928 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
929 iput(sdp->sd_sc_inode);
930 iput(sdp->sd_qc_inode);
931 }
932
933 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
934 gfs2_clear_rgrpd(sdp);
935 gfs2_jindex_free(sdp);
936 /* Take apart glock structures and buffer lists */
937 gfs2_gl_hash_clear(sdp);
938 /* Unmount the locking protocol */
939 gfs2_lm_unmount(sdp);
940
941 /* At this point, we're through participating in the lockspace */
942 gfs2_sys_fs_del(sdp);
943}
944
945/**
946 * gfs2_sync_fs - sync the filesystem
947 * @sb: the superblock
948 *
949 * Flushes the log to disk.
950 */
951
952static int gfs2_sync_fs(struct super_block *sb, int wait)
953{
954 struct gfs2_sbd *sdp = sb->s_fs_info;
955 if (wait && sdp)
956 gfs2_log_flush(sdp, NULL);
957 return 0;
958}
959
960/**
961 * gfs2_freeze - prevent further writes to the filesystem
962 * @sb: the VFS structure for the filesystem
963 *
964 */
965
966static int gfs2_freeze(struct super_block *sb)
967{
968 struct gfs2_sbd *sdp = sb->s_fs_info;
969 int error;
970
971 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
972 return -EINVAL;
973
974 for (;;) {
975 error = gfs2_freeze_fs(sdp);
976 if (!error)
977 break;
978
979 switch (error) {
980 case -EBUSY:
981 fs_err(sdp, "waiting for recovery before freeze\n");
982 break;
983
984 default:
985 fs_err(sdp, "error freezing FS: %d\n", error);
986 break;
987 }
988
989 fs_err(sdp, "retrying...\n");
990 msleep(1000);
991 }
992 return 0;
993}
994
995/**
996 * gfs2_unfreeze - reallow writes to the filesystem
997 * @sb: the VFS structure for the filesystem
998 *
999 */
1000
1001static int gfs2_unfreeze(struct super_block *sb)
1002{
1003 gfs2_unfreeze_fs(sb->s_fs_info);
1004 return 0;
1005}
1006
1007/**
1008 * statfs_fill - fill in the sg for a given RG
1009 * @rgd: the RG
1010 * @sc: the sc structure
1011 *
1012 * Returns: 0 on success, -ESTALE if the LVB is invalid
1013 */
1014
1015static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1016 struct gfs2_statfs_change_host *sc)
1017{
1018 gfs2_rgrp_verify(rgd);
1019 sc->sc_total += rgd->rd_data;
1020 sc->sc_free += rgd->rd_free;
1021 sc->sc_dinodes += rgd->rd_dinodes;
1022 return 0;
1023}
1024
1025/**
1026 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1027 * @sdp: the filesystem
1028 * @sc: the sc info that will be returned
1029 *
1030 * Any error (other than a signal) will cause this routine to fall back
1031 * to the synchronous version.
1032 *
1033 * FIXME: This really shouldn't busy wait like this.
1034 *
1035 * Returns: errno
1036 */
1037
1038static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1039{
1040 struct gfs2_rgrpd *rgd_next;
1041 struct gfs2_holder *gha, *gh;
1042 unsigned int slots = 64;
1043 unsigned int x;
1044 int done;
1045 int error = 0, err;
1046
1047 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1048 gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1049 if (!gha)
1050 return -ENOMEM;
1051
1052 rgd_next = gfs2_rgrpd_get_first(sdp);
1053
1054 for (;;) {
1055 done = 1;
1056
1057 for (x = 0; x < slots; x++) {
1058 gh = gha + x;
1059
1060 if (gh->gh_gl && gfs2_glock_poll(gh)) {
1061 err = gfs2_glock_wait(gh);
1062 if (err) {
1063 gfs2_holder_uninit(gh);
1064 error = err;
1065 } else {
1066 if (!error)
1067 error = statfs_slow_fill(
1068 gh->gh_gl->gl_object, sc);
1069 gfs2_glock_dq_uninit(gh);
1070 }
1071 }
1072
1073 if (gh->gh_gl)
1074 done = 0;
1075 else if (rgd_next && !error) {
1076 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1077 LM_ST_SHARED,
1078 GL_ASYNC,
1079 gh);
1080 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1081 done = 0;
1082 }
1083
1084 if (signal_pending(current))
1085 error = -ERESTARTSYS;
1086 }
1087
1088 if (done)
1089 break;
1090
1091 yield();
1092 }
1093
1094 kfree(gha);
1095 return error;
1096}
1097
1098/**
1099 * gfs2_statfs_i - Do a statfs
1100 * @sdp: the filesystem
1101 * @sg: the sg structure
1102 *
1103 * Returns: errno
1104 */
1105
1106static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1107{
1108 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1109 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1110
1111 spin_lock(&sdp->sd_statfs_spin);
1112
1113 *sc = *m_sc;
1114 sc->sc_total += l_sc->sc_total;
1115 sc->sc_free += l_sc->sc_free;
1116 sc->sc_dinodes += l_sc->sc_dinodes;
1117
1118 spin_unlock(&sdp->sd_statfs_spin);
1119
1120 if (sc->sc_free < 0)
1121 sc->sc_free = 0;
1122 if (sc->sc_free > sc->sc_total)
1123 sc->sc_free = sc->sc_total;
1124 if (sc->sc_dinodes < 0)
1125 sc->sc_dinodes = 0;
1126
1127 return 0;
1128}
1129
1130/**
1131 * gfs2_statfs - Gather and return stats about the filesystem
1132 * @sb: The superblock
1133 * @statfsbuf: The buffer
1134 *
1135 * Returns: 0 on success or error code
1136 */
1137
1138static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1139{
1140 struct super_block *sb = dentry->d_inode->i_sb;
1141 struct gfs2_sbd *sdp = sb->s_fs_info;
1142 struct gfs2_statfs_change_host sc;
1143 int error;
1144
1145 error = gfs2_rindex_update(sdp);
1146 if (error)
1147 return error;
1148
1149 if (gfs2_tune_get(sdp, gt_statfs_slow))
1150 error = gfs2_statfs_slow(sdp, &sc);
1151 else
1152 error = gfs2_statfs_i(sdp, &sc);
1153
1154 if (error)
1155 return error;
1156
1157 buf->f_type = GFS2_MAGIC;
1158 buf->f_bsize = sdp->sd_sb.sb_bsize;
1159 buf->f_blocks = sc.sc_total;
1160 buf->f_bfree = sc.sc_free;
1161 buf->f_bavail = sc.sc_free;
1162 buf->f_files = sc.sc_dinodes + sc.sc_free;
1163 buf->f_ffree = sc.sc_free;
1164 buf->f_namelen = GFS2_FNAMESIZE;
1165
1166 return 0;
1167}
1168
1169/**
1170 * gfs2_remount_fs - called when the FS is remounted
1171 * @sb: the filesystem
1172 * @flags: the remount flags
1173 * @data: extra data passed in (not used right now)
1174 *
1175 * Returns: errno
1176 */
1177
1178static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1179{
1180 struct gfs2_sbd *sdp = sb->s_fs_info;
1181 struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1182 struct gfs2_tune *gt = &sdp->sd_tune;
1183 int error;
1184
1185 spin_lock(>->gt_spin);
1186 args.ar_commit = gt->gt_logd_secs;
1187 args.ar_quota_quantum = gt->gt_quota_quantum;
1188 if (gt->gt_statfs_slow)
1189 args.ar_statfs_quantum = 0;
1190 else
1191 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1192 spin_unlock(>->gt_spin);
1193 error = gfs2_mount_args(&args, data);
1194 if (error)
1195 return error;
1196
1197 /* Not allowed to change locking details */
1198 if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1199 strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1200 strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1201 return -EINVAL;
1202
1203 /* Some flags must not be changed */
1204 if (args_neq(&args, &sdp->sd_args, spectator) ||
1205 args_neq(&args, &sdp->sd_args, localflocks) ||
1206 args_neq(&args, &sdp->sd_args, meta))
1207 return -EINVAL;
1208
1209 if (sdp->sd_args.ar_spectator)
1210 *flags |= MS_RDONLY;
1211
1212 if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1213 if (*flags & MS_RDONLY)
1214 error = gfs2_make_fs_ro(sdp);
1215 else
1216 error = gfs2_make_fs_rw(sdp);
1217 if (error)
1218 return error;
1219 }
1220
1221 sdp->sd_args = args;
1222 if (sdp->sd_args.ar_posix_acl)
1223 sb->s_flags |= MS_POSIXACL;
1224 else
1225 sb->s_flags &= ~MS_POSIXACL;
1226 if (sdp->sd_args.ar_nobarrier)
1227 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1228 else
1229 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1230 spin_lock(>->gt_spin);
1231 gt->gt_logd_secs = args.ar_commit;
1232 gt->gt_quota_quantum = args.ar_quota_quantum;
1233 if (args.ar_statfs_quantum) {
1234 gt->gt_statfs_slow = 0;
1235 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1236 }
1237 else {
1238 gt->gt_statfs_slow = 1;
1239 gt->gt_statfs_quantum = 30;
1240 }
1241 spin_unlock(>->gt_spin);
1242
1243 gfs2_online_uevent(sdp);
1244 return 0;
1245}
1246
1247/**
1248 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1249 * @inode: The inode to drop
1250 *
1251 * If we've received a callback on an iopen lock then its because a
1252 * remote node tried to deallocate the inode but failed due to this node
1253 * still having the inode open. Here we mark the link count zero
1254 * since we know that it must have reached zero if the GLF_DEMOTE flag
1255 * is set on the iopen glock. If we didn't do a disk read since the
1256 * remote node removed the final link then we might otherwise miss
1257 * this event. This check ensures that this node will deallocate the
1258 * inode's blocks, or alternatively pass the baton on to another
1259 * node for later deallocation.
1260 */
1261
1262static int gfs2_drop_inode(struct inode *inode)
1263{
1264 struct gfs2_inode *ip = GFS2_I(inode);
1265
1266 if (inode->i_nlink) {
1267 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1268 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1269 clear_nlink(inode);
1270 }
1271 return generic_drop_inode(inode);
1272}
1273
1274static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1275{
1276 do {
1277 if (d1 == d2)
1278 return 1;
1279 d1 = d1->d_parent;
1280 } while (!IS_ROOT(d1));
1281 return 0;
1282}
1283
1284/**
1285 * gfs2_show_options - Show mount options for /proc/mounts
1286 * @s: seq_file structure
1287 * @root: root of this (sub)tree
1288 *
1289 * Returns: 0 on success or error code
1290 */
1291
1292static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1293{
1294 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1295 struct gfs2_args *args = &sdp->sd_args;
1296 int val;
1297
1298 if (is_ancestor(root, sdp->sd_master_dir))
1299 seq_printf(s, ",meta");
1300 if (args->ar_lockproto[0])
1301 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1302 if (args->ar_locktable[0])
1303 seq_printf(s, ",locktable=%s", args->ar_locktable);
1304 if (args->ar_hostdata[0])
1305 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1306 if (args->ar_spectator)
1307 seq_printf(s, ",spectator");
1308 if (args->ar_localflocks)
1309 seq_printf(s, ",localflocks");
1310 if (args->ar_debug)
1311 seq_printf(s, ",debug");
1312 if (args->ar_posix_acl)
1313 seq_printf(s, ",acl");
1314 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1315 char *state;
1316 switch (args->ar_quota) {
1317 case GFS2_QUOTA_OFF:
1318 state = "off";
1319 break;
1320 case GFS2_QUOTA_ACCOUNT:
1321 state = "account";
1322 break;
1323 case GFS2_QUOTA_ON:
1324 state = "on";
1325 break;
1326 default:
1327 state = "unknown";
1328 break;
1329 }
1330 seq_printf(s, ",quota=%s", state);
1331 }
1332 if (args->ar_suiddir)
1333 seq_printf(s, ",suiddir");
1334 if (args->ar_data != GFS2_DATA_DEFAULT) {
1335 char *state;
1336 switch (args->ar_data) {
1337 case GFS2_DATA_WRITEBACK:
1338 state = "writeback";
1339 break;
1340 case GFS2_DATA_ORDERED:
1341 state = "ordered";
1342 break;
1343 default:
1344 state = "unknown";
1345 break;
1346 }
1347 seq_printf(s, ",data=%s", state);
1348 }
1349 if (args->ar_discard)
1350 seq_printf(s, ",discard");
1351 val = sdp->sd_tune.gt_logd_secs;
1352 if (val != 30)
1353 seq_printf(s, ",commit=%d", val);
1354 val = sdp->sd_tune.gt_statfs_quantum;
1355 if (val != 30)
1356 seq_printf(s, ",statfs_quantum=%d", val);
1357 val = sdp->sd_tune.gt_quota_quantum;
1358 if (val != 60)
1359 seq_printf(s, ",quota_quantum=%d", val);
1360 if (args->ar_statfs_percent)
1361 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1362 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1363 const char *state;
1364
1365 switch (args->ar_errors) {
1366 case GFS2_ERRORS_WITHDRAW:
1367 state = "withdraw";
1368 break;
1369 case GFS2_ERRORS_PANIC:
1370 state = "panic";
1371 break;
1372 default:
1373 state = "unknown";
1374 break;
1375 }
1376 seq_printf(s, ",errors=%s", state);
1377 }
1378 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1379 seq_printf(s, ",nobarrier");
1380 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1381 seq_printf(s, ",demote_interface_used");
1382 return 0;
1383}
1384
1385static void gfs2_final_release_pages(struct gfs2_inode *ip)
1386{
1387 struct inode *inode = &ip->i_inode;
1388 struct gfs2_glock *gl = ip->i_gl;
1389
1390 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1391 truncate_inode_pages(&inode->i_data, 0);
1392
1393 if (atomic_read(&gl->gl_revokes) == 0) {
1394 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1395 clear_bit(GLF_DIRTY, &gl->gl_flags);
1396 }
1397}
1398
1399static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1400{
1401 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1402 struct gfs2_qadata *qa;
1403 struct gfs2_rgrpd *rgd;
1404 struct gfs2_holder gh;
1405 int error;
1406
1407 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1408 gfs2_consist_inode(ip);
1409 return -EIO;
1410 }
1411
1412 qa = gfs2_qadata_get(ip);
1413 if (!qa)
1414 return -ENOMEM;
1415
1416 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1417 if (error)
1418 goto out;
1419
1420 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1421 if (!rgd) {
1422 gfs2_consist_inode(ip);
1423 error = -EIO;
1424 goto out_qs;
1425 }
1426
1427 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1428 if (error)
1429 goto out_qs;
1430
1431 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1432 sdp->sd_jdesc->jd_blocks);
1433 if (error)
1434 goto out_rg_gunlock;
1435
1436 gfs2_free_di(rgd, ip);
1437
1438 gfs2_final_release_pages(ip);
1439
1440 gfs2_trans_end(sdp);
1441
1442out_rg_gunlock:
1443 gfs2_glock_dq_uninit(&gh);
1444out_qs:
1445 gfs2_quota_unhold(ip);
1446out:
1447 gfs2_qadata_put(ip);
1448 return error;
1449}
1450
1451/**
1452 * gfs2_evict_inode - Remove an inode from cache
1453 * @inode: The inode to evict
1454 *
1455 * There are three cases to consider:
1456 * 1. i_nlink == 0, we are final opener (and must deallocate)
1457 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1458 * 3. i_nlink > 0
1459 *
1460 * If the fs is read only, then we have to treat all cases as per #3
1461 * since we are unable to do any deallocation. The inode will be
1462 * deallocated by the next read/write node to attempt an allocation
1463 * in the same resource group
1464 *
1465 * We have to (at the moment) hold the inodes main lock to cover
1466 * the gap between unlocking the shared lock on the iopen lock and
1467 * taking the exclusive lock. I'd rather do a shared -> exclusive
1468 * conversion on the iopen lock, but we can change that later. This
1469 * is safe, just less efficient.
1470 */
1471
1472static void gfs2_evict_inode(struct inode *inode)
1473{
1474 struct super_block *sb = inode->i_sb;
1475 struct gfs2_sbd *sdp = sb->s_fs_info;
1476 struct gfs2_inode *ip = GFS2_I(inode);
1477 struct gfs2_holder gh;
1478 int error;
1479
1480 if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1481 goto out;
1482
1483 /* Must not read inode block until block type has been verified */
1484 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1485 if (unlikely(error)) {
1486 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1487 goto out;
1488 }
1489
1490 if (!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1491 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1492 if (error)
1493 goto out_truncate;
1494 }
1495
1496 if (test_bit(GIF_INVALID, &ip->i_flags)) {
1497 error = gfs2_inode_refresh(ip);
1498 if (error)
1499 goto out_truncate;
1500 }
1501
1502 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1503 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1504 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1505 error = gfs2_glock_nq(&ip->i_iopen_gh);
1506 if (error)
1507 goto out_truncate;
1508
1509 /* Case 1 starts here */
1510
1511 if (S_ISDIR(inode->i_mode) &&
1512 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1513 error = gfs2_dir_exhash_dealloc(ip);
1514 if (error)
1515 goto out_unlock;
1516 }
1517
1518 if (ip->i_eattr) {
1519 error = gfs2_ea_dealloc(ip);
1520 if (error)
1521 goto out_unlock;
1522 }
1523
1524 if (!gfs2_is_stuffed(ip)) {
1525 error = gfs2_file_dealloc(ip);
1526 if (error)
1527 goto out_unlock;
1528 }
1529
1530 error = gfs2_dinode_dealloc(ip);
1531 goto out_unlock;
1532
1533out_truncate:
1534 gfs2_log_flush(sdp, ip->i_gl);
1535 write_inode_now(inode, 1);
1536 gfs2_ail_flush(ip->i_gl, 0);
1537
1538 /* Case 2 starts here */
1539 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1540 if (error)
1541 goto out_unlock;
1542 /* Needs to be done before glock release & also in a transaction */
1543 truncate_inode_pages(&inode->i_data, 0);
1544 gfs2_trans_end(sdp);
1545
1546out_unlock:
1547 /* Error path for case 1 */
1548 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
1549 gfs2_glock_dq(&ip->i_iopen_gh);
1550 gfs2_holder_uninit(&ip->i_iopen_gh);
1551 gfs2_glock_dq_uninit(&gh);
1552 if (error && error != GLR_TRYFAILED && error != -EROFS)
1553 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1554out:
1555 /* Case 3 starts here */
1556 truncate_inode_pages(&inode->i_data, 0);
1557 clear_inode(inode);
1558 gfs2_dir_hash_inval(ip);
1559 ip->i_gl->gl_object = NULL;
1560 flush_delayed_work_sync(&ip->i_gl->gl_work);
1561 gfs2_glock_add_to_lru(ip->i_gl);
1562 gfs2_glock_put(ip->i_gl);
1563 ip->i_gl = NULL;
1564 if (ip->i_iopen_gh.gh_gl) {
1565 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1566 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1567 }
1568}
1569
1570static struct inode *gfs2_alloc_inode(struct super_block *sb)
1571{
1572 struct gfs2_inode *ip;
1573
1574 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1575 if (ip) {
1576 ip->i_flags = 0;
1577 ip->i_gl = NULL;
1578 ip->i_rgd = NULL;
1579 }
1580 return &ip->i_inode;
1581}
1582
1583static void gfs2_i_callback(struct rcu_head *head)
1584{
1585 struct inode *inode = container_of(head, struct inode, i_rcu);
1586 kmem_cache_free(gfs2_inode_cachep, inode);
1587}
1588
1589static void gfs2_destroy_inode(struct inode *inode)
1590{
1591 call_rcu(&inode->i_rcu, gfs2_i_callback);
1592}
1593
1594const struct super_operations gfs2_super_ops = {
1595 .alloc_inode = gfs2_alloc_inode,
1596 .destroy_inode = gfs2_destroy_inode,
1597 .write_inode = gfs2_write_inode,
1598 .dirty_inode = gfs2_dirty_inode,
1599 .evict_inode = gfs2_evict_inode,
1600 .put_super = gfs2_put_super,
1601 .sync_fs = gfs2_sync_fs,
1602 .freeze_fs = gfs2_freeze,
1603 .unfreeze_fs = gfs2_unfreeze,
1604 .statfs = gfs2_statfs,
1605 .remount_fs = gfs2_remount_fs,
1606 .drop_inode = gfs2_drop_inode,
1607 .show_options = gfs2_show_options,
1608};
1609
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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12#include <linux/bio.h>
13#include <linux/sched.h>
14#include <linux/slab.h>
15#include <linux/spinlock.h>
16#include <linux/completion.h>
17#include <linux/buffer_head.h>
18#include <linux/statfs.h>
19#include <linux/seq_file.h>
20#include <linux/mount.h>
21#include <linux/kthread.h>
22#include <linux/delay.h>
23#include <linux/gfs2_ondisk.h>
24#include <linux/crc32.h>
25#include <linux/time.h>
26#include <linux/wait.h>
27#include <linux/writeback.h>
28#include <linux/backing-dev.h>
29#include <linux/kernel.h>
30
31#include "gfs2.h"
32#include "incore.h"
33#include "bmap.h"
34#include "dir.h"
35#include "glock.h"
36#include "glops.h"
37#include "inode.h"
38#include "log.h"
39#include "meta_io.h"
40#include "quota.h"
41#include "recovery.h"
42#include "rgrp.h"
43#include "super.h"
44#include "trans.h"
45#include "util.h"
46#include "sys.h"
47#include "xattr.h"
48
49#define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
50
51enum {
52 Opt_lockproto,
53 Opt_locktable,
54 Opt_hostdata,
55 Opt_spectator,
56 Opt_ignore_local_fs,
57 Opt_localflocks,
58 Opt_localcaching,
59 Opt_debug,
60 Opt_nodebug,
61 Opt_upgrade,
62 Opt_acl,
63 Opt_noacl,
64 Opt_quota_off,
65 Opt_quota_account,
66 Opt_quota_on,
67 Opt_quota,
68 Opt_noquota,
69 Opt_suiddir,
70 Opt_nosuiddir,
71 Opt_data_writeback,
72 Opt_data_ordered,
73 Opt_meta,
74 Opt_discard,
75 Opt_nodiscard,
76 Opt_commit,
77 Opt_err_withdraw,
78 Opt_err_panic,
79 Opt_statfs_quantum,
80 Opt_statfs_percent,
81 Opt_quota_quantum,
82 Opt_barrier,
83 Opt_nobarrier,
84 Opt_rgrplvb,
85 Opt_norgrplvb,
86 Opt_loccookie,
87 Opt_noloccookie,
88 Opt_error,
89};
90
91static const match_table_t tokens = {
92 {Opt_lockproto, "lockproto=%s"},
93 {Opt_locktable, "locktable=%s"},
94 {Opt_hostdata, "hostdata=%s"},
95 {Opt_spectator, "spectator"},
96 {Opt_spectator, "norecovery"},
97 {Opt_ignore_local_fs, "ignore_local_fs"},
98 {Opt_localflocks, "localflocks"},
99 {Opt_localcaching, "localcaching"},
100 {Opt_debug, "debug"},
101 {Opt_nodebug, "nodebug"},
102 {Opt_upgrade, "upgrade"},
103 {Opt_acl, "acl"},
104 {Opt_noacl, "noacl"},
105 {Opt_quota_off, "quota=off"},
106 {Opt_quota_account, "quota=account"},
107 {Opt_quota_on, "quota=on"},
108 {Opt_quota, "quota"},
109 {Opt_noquota, "noquota"},
110 {Opt_suiddir, "suiddir"},
111 {Opt_nosuiddir, "nosuiddir"},
112 {Opt_data_writeback, "data=writeback"},
113 {Opt_data_ordered, "data=ordered"},
114 {Opt_meta, "meta"},
115 {Opt_discard, "discard"},
116 {Opt_nodiscard, "nodiscard"},
117 {Opt_commit, "commit=%d"},
118 {Opt_err_withdraw, "errors=withdraw"},
119 {Opt_err_panic, "errors=panic"},
120 {Opt_statfs_quantum, "statfs_quantum=%d"},
121 {Opt_statfs_percent, "statfs_percent=%d"},
122 {Opt_quota_quantum, "quota_quantum=%d"},
123 {Opt_barrier, "barrier"},
124 {Opt_nobarrier, "nobarrier"},
125 {Opt_rgrplvb, "rgrplvb"},
126 {Opt_norgrplvb, "norgrplvb"},
127 {Opt_loccookie, "loccookie"},
128 {Opt_noloccookie, "noloccookie"},
129 {Opt_error, NULL}
130};
131
132/**
133 * gfs2_mount_args - Parse mount options
134 * @args: The structure into which the parsed options will be written
135 * @options: The options to parse
136 *
137 * Return: errno
138 */
139
140int gfs2_mount_args(struct gfs2_args *args, char *options)
141{
142 char *o;
143 int token;
144 substring_t tmp[MAX_OPT_ARGS];
145 int rv;
146
147 /* Split the options into tokens with the "," character and
148 process them */
149
150 while (1) {
151 o = strsep(&options, ",");
152 if (o == NULL)
153 break;
154 if (*o == '\0')
155 continue;
156
157 token = match_token(o, tokens, tmp);
158 switch (token) {
159 case Opt_lockproto:
160 match_strlcpy(args->ar_lockproto, &tmp[0],
161 GFS2_LOCKNAME_LEN);
162 break;
163 case Opt_locktable:
164 match_strlcpy(args->ar_locktable, &tmp[0],
165 GFS2_LOCKNAME_LEN);
166 break;
167 case Opt_hostdata:
168 match_strlcpy(args->ar_hostdata, &tmp[0],
169 GFS2_LOCKNAME_LEN);
170 break;
171 case Opt_spectator:
172 args->ar_spectator = 1;
173 break;
174 case Opt_ignore_local_fs:
175 /* Retained for backwards compat only */
176 break;
177 case Opt_localflocks:
178 args->ar_localflocks = 1;
179 break;
180 case Opt_localcaching:
181 /* Retained for backwards compat only */
182 break;
183 case Opt_debug:
184 if (args->ar_errors == GFS2_ERRORS_PANIC) {
185 pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
186 return -EINVAL;
187 }
188 args->ar_debug = 1;
189 break;
190 case Opt_nodebug:
191 args->ar_debug = 0;
192 break;
193 case Opt_upgrade:
194 /* Retained for backwards compat only */
195 break;
196 case Opt_acl:
197 args->ar_posix_acl = 1;
198 break;
199 case Opt_noacl:
200 args->ar_posix_acl = 0;
201 break;
202 case Opt_quota_off:
203 case Opt_noquota:
204 args->ar_quota = GFS2_QUOTA_OFF;
205 break;
206 case Opt_quota_account:
207 args->ar_quota = GFS2_QUOTA_ACCOUNT;
208 break;
209 case Opt_quota_on:
210 case Opt_quota:
211 args->ar_quota = GFS2_QUOTA_ON;
212 break;
213 case Opt_suiddir:
214 args->ar_suiddir = 1;
215 break;
216 case Opt_nosuiddir:
217 args->ar_suiddir = 0;
218 break;
219 case Opt_data_writeback:
220 args->ar_data = GFS2_DATA_WRITEBACK;
221 break;
222 case Opt_data_ordered:
223 args->ar_data = GFS2_DATA_ORDERED;
224 break;
225 case Opt_meta:
226 args->ar_meta = 1;
227 break;
228 case Opt_discard:
229 args->ar_discard = 1;
230 break;
231 case Opt_nodiscard:
232 args->ar_discard = 0;
233 break;
234 case Opt_commit:
235 rv = match_int(&tmp[0], &args->ar_commit);
236 if (rv || args->ar_commit <= 0) {
237 pr_warn("commit mount option requires a positive numeric argument\n");
238 return rv ? rv : -EINVAL;
239 }
240 break;
241 case Opt_statfs_quantum:
242 rv = match_int(&tmp[0], &args->ar_statfs_quantum);
243 if (rv || args->ar_statfs_quantum < 0) {
244 pr_warn("statfs_quantum mount option requires a non-negative numeric argument\n");
245 return rv ? rv : -EINVAL;
246 }
247 break;
248 case Opt_quota_quantum:
249 rv = match_int(&tmp[0], &args->ar_quota_quantum);
250 if (rv || args->ar_quota_quantum <= 0) {
251 pr_warn("quota_quantum mount option requires a positive numeric argument\n");
252 return rv ? rv : -EINVAL;
253 }
254 break;
255 case Opt_statfs_percent:
256 rv = match_int(&tmp[0], &args->ar_statfs_percent);
257 if (rv || args->ar_statfs_percent < 0 ||
258 args->ar_statfs_percent > 100) {
259 pr_warn("statfs_percent mount option requires a numeric argument between 0 and 100\n");
260 return rv ? rv : -EINVAL;
261 }
262 break;
263 case Opt_err_withdraw:
264 args->ar_errors = GFS2_ERRORS_WITHDRAW;
265 break;
266 case Opt_err_panic:
267 if (args->ar_debug) {
268 pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
269 return -EINVAL;
270 }
271 args->ar_errors = GFS2_ERRORS_PANIC;
272 break;
273 case Opt_barrier:
274 args->ar_nobarrier = 0;
275 break;
276 case Opt_nobarrier:
277 args->ar_nobarrier = 1;
278 break;
279 case Opt_rgrplvb:
280 args->ar_rgrplvb = 1;
281 break;
282 case Opt_norgrplvb:
283 args->ar_rgrplvb = 0;
284 break;
285 case Opt_loccookie:
286 args->ar_loccookie = 1;
287 break;
288 case Opt_noloccookie:
289 args->ar_loccookie = 0;
290 break;
291 case Opt_error:
292 default:
293 pr_warn("invalid mount option: %s\n", o);
294 return -EINVAL;
295 }
296 }
297
298 return 0;
299}
300
301/**
302 * gfs2_jindex_free - Clear all the journal index information
303 * @sdp: The GFS2 superblock
304 *
305 */
306
307void gfs2_jindex_free(struct gfs2_sbd *sdp)
308{
309 struct list_head list;
310 struct gfs2_jdesc *jd;
311
312 spin_lock(&sdp->sd_jindex_spin);
313 list_add(&list, &sdp->sd_jindex_list);
314 list_del_init(&sdp->sd_jindex_list);
315 sdp->sd_journals = 0;
316 spin_unlock(&sdp->sd_jindex_spin);
317
318 while (!list_empty(&list)) {
319 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
320 gfs2_free_journal_extents(jd);
321 list_del(&jd->jd_list);
322 iput(jd->jd_inode);
323 kfree(jd);
324 }
325}
326
327static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
328{
329 struct gfs2_jdesc *jd;
330 int found = 0;
331
332 list_for_each_entry(jd, head, jd_list) {
333 if (jd->jd_jid == jid) {
334 found = 1;
335 break;
336 }
337 }
338
339 if (!found)
340 jd = NULL;
341
342 return jd;
343}
344
345struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
346{
347 struct gfs2_jdesc *jd;
348
349 spin_lock(&sdp->sd_jindex_spin);
350 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
351 spin_unlock(&sdp->sd_jindex_spin);
352
353 return jd;
354}
355
356int gfs2_jdesc_check(struct gfs2_jdesc *jd)
357{
358 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
359 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
360 u64 size = i_size_read(jd->jd_inode);
361
362 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
363 return -EIO;
364
365 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
366
367 if (gfs2_write_alloc_required(ip, 0, size)) {
368 gfs2_consist_inode(ip);
369 return -EIO;
370 }
371
372 return 0;
373}
374
375static int init_threads(struct gfs2_sbd *sdp)
376{
377 struct task_struct *p;
378 int error = 0;
379
380 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
381 if (IS_ERR(p)) {
382 error = PTR_ERR(p);
383 fs_err(sdp, "can't start logd thread: %d\n", error);
384 return error;
385 }
386 sdp->sd_logd_process = p;
387
388 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
389 if (IS_ERR(p)) {
390 error = PTR_ERR(p);
391 fs_err(sdp, "can't start quotad thread: %d\n", error);
392 goto fail;
393 }
394 sdp->sd_quotad_process = p;
395 return 0;
396
397fail:
398 kthread_stop(sdp->sd_logd_process);
399 return error;
400}
401
402/**
403 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
404 * @sdp: the filesystem
405 *
406 * Returns: errno
407 */
408
409int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
410{
411 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
412 struct gfs2_glock *j_gl = ip->i_gl;
413 struct gfs2_holder freeze_gh;
414 struct gfs2_log_header_host head;
415 int error;
416
417 error = init_threads(sdp);
418 if (error)
419 return error;
420
421 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
422 &freeze_gh);
423 if (error)
424 goto fail_threads;
425
426 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
427
428 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
429 if (error)
430 goto fail;
431
432 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
433 gfs2_consist(sdp);
434 error = -EIO;
435 goto fail;
436 }
437
438 /* Initialize some head of the log stuff */
439 sdp->sd_log_sequence = head.lh_sequence + 1;
440 gfs2_log_pointers_init(sdp, head.lh_blkno);
441
442 error = gfs2_quota_init(sdp);
443 if (error)
444 goto fail;
445
446 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
447
448 gfs2_glock_dq_uninit(&freeze_gh);
449
450 return 0;
451
452fail:
453 freeze_gh.gh_flags |= GL_NOCACHE;
454 gfs2_glock_dq_uninit(&freeze_gh);
455fail_threads:
456 kthread_stop(sdp->sd_quotad_process);
457 kthread_stop(sdp->sd_logd_process);
458 return error;
459}
460
461void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
462{
463 const struct gfs2_statfs_change *str = buf;
464
465 sc->sc_total = be64_to_cpu(str->sc_total);
466 sc->sc_free = be64_to_cpu(str->sc_free);
467 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
468}
469
470static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
471{
472 struct gfs2_statfs_change *str = buf;
473
474 str->sc_total = cpu_to_be64(sc->sc_total);
475 str->sc_free = cpu_to_be64(sc->sc_free);
476 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
477}
478
479int gfs2_statfs_init(struct gfs2_sbd *sdp)
480{
481 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
482 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
483 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
484 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
485 struct buffer_head *m_bh, *l_bh;
486 struct gfs2_holder gh;
487 int error;
488
489 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
490 &gh);
491 if (error)
492 return error;
493
494 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
495 if (error)
496 goto out;
497
498 if (sdp->sd_args.ar_spectator) {
499 spin_lock(&sdp->sd_statfs_spin);
500 gfs2_statfs_change_in(m_sc, m_bh->b_data +
501 sizeof(struct gfs2_dinode));
502 spin_unlock(&sdp->sd_statfs_spin);
503 } else {
504 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
505 if (error)
506 goto out_m_bh;
507
508 spin_lock(&sdp->sd_statfs_spin);
509 gfs2_statfs_change_in(m_sc, m_bh->b_data +
510 sizeof(struct gfs2_dinode));
511 gfs2_statfs_change_in(l_sc, l_bh->b_data +
512 sizeof(struct gfs2_dinode));
513 spin_unlock(&sdp->sd_statfs_spin);
514
515 brelse(l_bh);
516 }
517
518out_m_bh:
519 brelse(m_bh);
520out:
521 gfs2_glock_dq_uninit(&gh);
522 return 0;
523}
524
525void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
526 s64 dinodes)
527{
528 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
529 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
530 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
531 struct buffer_head *l_bh;
532 s64 x, y;
533 int need_sync = 0;
534 int error;
535
536 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
537 if (error)
538 return;
539
540 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
541
542 spin_lock(&sdp->sd_statfs_spin);
543 l_sc->sc_total += total;
544 l_sc->sc_free += free;
545 l_sc->sc_dinodes += dinodes;
546 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
547 if (sdp->sd_args.ar_statfs_percent) {
548 x = 100 * l_sc->sc_free;
549 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
550 if (x >= y || x <= -y)
551 need_sync = 1;
552 }
553 spin_unlock(&sdp->sd_statfs_spin);
554
555 brelse(l_bh);
556 if (need_sync)
557 gfs2_wake_up_statfs(sdp);
558}
559
560void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
561 struct buffer_head *l_bh)
562{
563 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
564 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
565 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
566 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
567
568 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
569 gfs2_trans_add_meta(m_ip->i_gl, m_bh);
570
571 spin_lock(&sdp->sd_statfs_spin);
572 m_sc->sc_total += l_sc->sc_total;
573 m_sc->sc_free += l_sc->sc_free;
574 m_sc->sc_dinodes += l_sc->sc_dinodes;
575 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
576 memset(l_bh->b_data + sizeof(struct gfs2_dinode),
577 0, sizeof(struct gfs2_statfs_change));
578 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
579 spin_unlock(&sdp->sd_statfs_spin);
580}
581
582int gfs2_statfs_sync(struct super_block *sb, int type)
583{
584 struct gfs2_sbd *sdp = sb->s_fs_info;
585 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
586 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
587 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
588 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
589 struct gfs2_holder gh;
590 struct buffer_head *m_bh, *l_bh;
591 int error;
592
593 sb_start_write(sb);
594 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
595 &gh);
596 if (error)
597 goto out;
598
599 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
600 if (error)
601 goto out_unlock;
602
603 spin_lock(&sdp->sd_statfs_spin);
604 gfs2_statfs_change_in(m_sc, m_bh->b_data +
605 sizeof(struct gfs2_dinode));
606 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
607 spin_unlock(&sdp->sd_statfs_spin);
608 goto out_bh;
609 }
610 spin_unlock(&sdp->sd_statfs_spin);
611
612 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
613 if (error)
614 goto out_bh;
615
616 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
617 if (error)
618 goto out_bh2;
619
620 update_statfs(sdp, m_bh, l_bh);
621 sdp->sd_statfs_force_sync = 0;
622
623 gfs2_trans_end(sdp);
624
625out_bh2:
626 brelse(l_bh);
627out_bh:
628 brelse(m_bh);
629out_unlock:
630 gfs2_glock_dq_uninit(&gh);
631out:
632 sb_end_write(sb);
633 return error;
634}
635
636struct lfcc {
637 struct list_head list;
638 struct gfs2_holder gh;
639};
640
641/**
642 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
643 * journals are clean
644 * @sdp: the file system
645 * @state: the state to put the transaction lock into
646 * @t_gh: the hold on the transaction lock
647 *
648 * Returns: errno
649 */
650
651static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
652 struct gfs2_holder *freeze_gh)
653{
654 struct gfs2_inode *ip;
655 struct gfs2_jdesc *jd;
656 struct lfcc *lfcc;
657 LIST_HEAD(list);
658 struct gfs2_log_header_host lh;
659 int error;
660
661 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
662 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
663 if (!lfcc) {
664 error = -ENOMEM;
665 goto out;
666 }
667 ip = GFS2_I(jd->jd_inode);
668 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
669 if (error) {
670 kfree(lfcc);
671 goto out;
672 }
673 list_add(&lfcc->list, &list);
674 }
675
676 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
677 GL_NOCACHE, freeze_gh);
678
679 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
680 error = gfs2_jdesc_check(jd);
681 if (error)
682 break;
683 error = gfs2_find_jhead(jd, &lh);
684 if (error)
685 break;
686 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
687 error = -EBUSY;
688 break;
689 }
690 }
691
692 if (error)
693 gfs2_glock_dq_uninit(freeze_gh);
694
695out:
696 while (!list_empty(&list)) {
697 lfcc = list_entry(list.next, struct lfcc, list);
698 list_del(&lfcc->list);
699 gfs2_glock_dq_uninit(&lfcc->gh);
700 kfree(lfcc);
701 }
702 return error;
703}
704
705void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
706{
707 struct gfs2_dinode *str = buf;
708
709 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
710 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
711 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
712 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
713 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
714 str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
715 str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
716 str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
717 str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
718 str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
719 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
720 str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
721 str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
722 str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
723
724 str->di_goal_meta = cpu_to_be64(ip->i_goal);
725 str->di_goal_data = cpu_to_be64(ip->i_goal);
726 str->di_generation = cpu_to_be64(ip->i_generation);
727
728 str->di_flags = cpu_to_be32(ip->i_diskflags);
729 str->di_height = cpu_to_be16(ip->i_height);
730 str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
731 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
732 GFS2_FORMAT_DE : 0);
733 str->di_depth = cpu_to_be16(ip->i_depth);
734 str->di_entries = cpu_to_be32(ip->i_entries);
735
736 str->di_eattr = cpu_to_be64(ip->i_eattr);
737 str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
738 str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
739 str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
740}
741
742/**
743 * gfs2_write_inode - Make sure the inode is stable on the disk
744 * @inode: The inode
745 * @wbc: The writeback control structure
746 *
747 * Returns: errno
748 */
749
750static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
751{
752 struct gfs2_inode *ip = GFS2_I(inode);
753 struct gfs2_sbd *sdp = GFS2_SB(inode);
754 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
755 struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
756 int ret = 0;
757
758 if (wbc->sync_mode == WB_SYNC_ALL)
759 gfs2_log_flush(GFS2_SB(inode), ip->i_gl, NORMAL_FLUSH);
760 if (bdi->wb.dirty_exceeded)
761 gfs2_ail1_flush(sdp, wbc);
762 else
763 filemap_fdatawrite(metamapping);
764 if (wbc->sync_mode == WB_SYNC_ALL)
765 ret = filemap_fdatawait(metamapping);
766 if (ret)
767 mark_inode_dirty_sync(inode);
768 return ret;
769}
770
771/**
772 * gfs2_dirty_inode - check for atime updates
773 * @inode: The inode in question
774 * @flags: The type of dirty
775 *
776 * Unfortunately it can be called under any combination of inode
777 * glock and transaction lock, so we have to check carefully.
778 *
779 * At the moment this deals only with atime - it should be possible
780 * to expand that role in future, once a review of the locking has
781 * been carried out.
782 */
783
784static void gfs2_dirty_inode(struct inode *inode, int flags)
785{
786 struct gfs2_inode *ip = GFS2_I(inode);
787 struct gfs2_sbd *sdp = GFS2_SB(inode);
788 struct buffer_head *bh;
789 struct gfs2_holder gh;
790 int need_unlock = 0;
791 int need_endtrans = 0;
792 int ret;
793
794 if (!(flags & (I_DIRTY_DATASYNC|I_DIRTY_SYNC)))
795 return;
796
797 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
798 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
799 if (ret) {
800 fs_err(sdp, "dirty_inode: glock %d\n", ret);
801 return;
802 }
803 need_unlock = 1;
804 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
805 return;
806
807 if (current->journal_info == NULL) {
808 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
809 if (ret) {
810 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
811 goto out;
812 }
813 need_endtrans = 1;
814 }
815
816 ret = gfs2_meta_inode_buffer(ip, &bh);
817 if (ret == 0) {
818 gfs2_trans_add_meta(ip->i_gl, bh);
819 gfs2_dinode_out(ip, bh->b_data);
820 brelse(bh);
821 }
822
823 if (need_endtrans)
824 gfs2_trans_end(sdp);
825out:
826 if (need_unlock)
827 gfs2_glock_dq_uninit(&gh);
828}
829
830/**
831 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
832 * @sdp: the filesystem
833 *
834 * Returns: errno
835 */
836
837static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
838{
839 struct gfs2_holder freeze_gh;
840 int error;
841
842 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, GL_NOCACHE,
843 &freeze_gh);
844 if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
845 return error;
846
847 kthread_stop(sdp->sd_quotad_process);
848 kthread_stop(sdp->sd_logd_process);
849
850 flush_workqueue(gfs2_delete_workqueue);
851 gfs2_quota_sync(sdp->sd_vfs, 0);
852 gfs2_statfs_sync(sdp->sd_vfs, 0);
853
854 gfs2_log_flush(sdp, NULL, SHUTDOWN_FLUSH);
855 wait_event(sdp->sd_reserving_log_wait, atomic_read(&sdp->sd_reserving_log) == 0);
856 gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
857
858 if (gfs2_holder_initialized(&freeze_gh))
859 gfs2_glock_dq_uninit(&freeze_gh);
860
861 gfs2_quota_cleanup(sdp);
862
863 return error;
864}
865
866/**
867 * gfs2_put_super - Unmount the filesystem
868 * @sb: The VFS superblock
869 *
870 */
871
872static void gfs2_put_super(struct super_block *sb)
873{
874 struct gfs2_sbd *sdp = sb->s_fs_info;
875 int error;
876 struct gfs2_jdesc *jd;
877
878 /* No more recovery requests */
879 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
880 smp_mb();
881
882 /* Wait on outstanding recovery */
883restart:
884 spin_lock(&sdp->sd_jindex_spin);
885 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
886 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
887 continue;
888 spin_unlock(&sdp->sd_jindex_spin);
889 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
890 TASK_UNINTERRUPTIBLE);
891 goto restart;
892 }
893 spin_unlock(&sdp->sd_jindex_spin);
894
895 if (!(sb->s_flags & MS_RDONLY)) {
896 error = gfs2_make_fs_ro(sdp);
897 if (error)
898 gfs2_io_error(sdp);
899 }
900 /* At this point, we're through modifying the disk */
901
902 /* Release stuff */
903
904 iput(sdp->sd_jindex);
905 iput(sdp->sd_statfs_inode);
906 iput(sdp->sd_rindex);
907 iput(sdp->sd_quota_inode);
908
909 gfs2_glock_put(sdp->sd_rename_gl);
910 gfs2_glock_put(sdp->sd_freeze_gl);
911
912 if (!sdp->sd_args.ar_spectator) {
913 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
914 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
915 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
916 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
917 iput(sdp->sd_sc_inode);
918 iput(sdp->sd_qc_inode);
919 }
920
921 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
922 gfs2_clear_rgrpd(sdp);
923 gfs2_jindex_free(sdp);
924 /* Take apart glock structures and buffer lists */
925 gfs2_gl_hash_clear(sdp);
926 /* Unmount the locking protocol */
927 gfs2_lm_unmount(sdp);
928
929 /* At this point, we're through participating in the lockspace */
930 gfs2_sys_fs_del(sdp);
931}
932
933/**
934 * gfs2_sync_fs - sync the filesystem
935 * @sb: the superblock
936 *
937 * Flushes the log to disk.
938 */
939
940static int gfs2_sync_fs(struct super_block *sb, int wait)
941{
942 struct gfs2_sbd *sdp = sb->s_fs_info;
943
944 gfs2_quota_sync(sb, -1);
945 if (wait && sdp)
946 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
947 return 0;
948}
949
950void gfs2_freeze_func(struct work_struct *work)
951{
952 int error;
953 struct gfs2_holder freeze_gh;
954 struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
955 struct super_block *sb = sdp->sd_vfs;
956
957 atomic_inc(&sb->s_active);
958 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
959 &freeze_gh);
960 if (error) {
961 printk(KERN_INFO "GFS2: couln't get freeze lock : %d\n", error);
962 gfs2_assert_withdraw(sdp, 0);
963 }
964 else {
965 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
966 error = thaw_super(sb);
967 if (error) {
968 printk(KERN_INFO "GFS2: couldn't thaw filesystem: %d\n",
969 error);
970 gfs2_assert_withdraw(sdp, 0);
971 }
972 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
973 freeze_gh.gh_flags |= GL_NOCACHE;
974 gfs2_glock_dq_uninit(&freeze_gh);
975 }
976 deactivate_super(sb);
977 return;
978}
979
980/**
981 * gfs2_freeze - prevent further writes to the filesystem
982 * @sb: the VFS structure for the filesystem
983 *
984 */
985
986static int gfs2_freeze(struct super_block *sb)
987{
988 struct gfs2_sbd *sdp = sb->s_fs_info;
989 int error = 0;
990
991 mutex_lock(&sdp->sd_freeze_mutex);
992 if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN)
993 goto out;
994
995 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
996 error = -EINVAL;
997 goto out;
998 }
999
1000 for (;;) {
1001 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
1002 if (!error)
1003 break;
1004
1005 switch (error) {
1006 case -EBUSY:
1007 fs_err(sdp, "waiting for recovery before freeze\n");
1008 break;
1009
1010 default:
1011 fs_err(sdp, "error freezing FS: %d\n", error);
1012 break;
1013 }
1014
1015 fs_err(sdp, "retrying...\n");
1016 msleep(1000);
1017 }
1018 error = 0;
1019out:
1020 mutex_unlock(&sdp->sd_freeze_mutex);
1021 return error;
1022}
1023
1024/**
1025 * gfs2_unfreeze - reallow writes to the filesystem
1026 * @sb: the VFS structure for the filesystem
1027 *
1028 */
1029
1030static int gfs2_unfreeze(struct super_block *sb)
1031{
1032 struct gfs2_sbd *sdp = sb->s_fs_info;
1033
1034 mutex_lock(&sdp->sd_freeze_mutex);
1035 if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
1036 !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
1037 mutex_unlock(&sdp->sd_freeze_mutex);
1038 return 0;
1039 }
1040
1041 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1042 mutex_unlock(&sdp->sd_freeze_mutex);
1043 return 0;
1044}
1045
1046/**
1047 * statfs_fill - fill in the sg for a given RG
1048 * @rgd: the RG
1049 * @sc: the sc structure
1050 *
1051 * Returns: 0 on success, -ESTALE if the LVB is invalid
1052 */
1053
1054static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1055 struct gfs2_statfs_change_host *sc)
1056{
1057 gfs2_rgrp_verify(rgd);
1058 sc->sc_total += rgd->rd_data;
1059 sc->sc_free += rgd->rd_free;
1060 sc->sc_dinodes += rgd->rd_dinodes;
1061 return 0;
1062}
1063
1064/**
1065 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1066 * @sdp: the filesystem
1067 * @sc: the sc info that will be returned
1068 *
1069 * Any error (other than a signal) will cause this routine to fall back
1070 * to the synchronous version.
1071 *
1072 * FIXME: This really shouldn't busy wait like this.
1073 *
1074 * Returns: errno
1075 */
1076
1077static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1078{
1079 struct gfs2_rgrpd *rgd_next;
1080 struct gfs2_holder *gha, *gh;
1081 unsigned int slots = 64;
1082 unsigned int x;
1083 int done;
1084 int error = 0, err;
1085
1086 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1087 gha = kmalloc(slots * sizeof(struct gfs2_holder), GFP_KERNEL);
1088 if (!gha)
1089 return -ENOMEM;
1090 for (x = 0; x < slots; x++)
1091 gfs2_holder_mark_uninitialized(gha + x);
1092
1093 rgd_next = gfs2_rgrpd_get_first(sdp);
1094
1095 for (;;) {
1096 done = 1;
1097
1098 for (x = 0; x < slots; x++) {
1099 gh = gha + x;
1100
1101 if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
1102 err = gfs2_glock_wait(gh);
1103 if (err) {
1104 gfs2_holder_uninit(gh);
1105 error = err;
1106 } else {
1107 if (!error)
1108 error = statfs_slow_fill(
1109 gh->gh_gl->gl_object, sc);
1110 gfs2_glock_dq_uninit(gh);
1111 }
1112 }
1113
1114 if (gfs2_holder_initialized(gh))
1115 done = 0;
1116 else if (rgd_next && !error) {
1117 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1118 LM_ST_SHARED,
1119 GL_ASYNC,
1120 gh);
1121 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1122 done = 0;
1123 }
1124
1125 if (signal_pending(current))
1126 error = -ERESTARTSYS;
1127 }
1128
1129 if (done)
1130 break;
1131
1132 yield();
1133 }
1134
1135 kfree(gha);
1136 return error;
1137}
1138
1139/**
1140 * gfs2_statfs_i - Do a statfs
1141 * @sdp: the filesystem
1142 * @sg: the sg structure
1143 *
1144 * Returns: errno
1145 */
1146
1147static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1148{
1149 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1150 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1151
1152 spin_lock(&sdp->sd_statfs_spin);
1153
1154 *sc = *m_sc;
1155 sc->sc_total += l_sc->sc_total;
1156 sc->sc_free += l_sc->sc_free;
1157 sc->sc_dinodes += l_sc->sc_dinodes;
1158
1159 spin_unlock(&sdp->sd_statfs_spin);
1160
1161 if (sc->sc_free < 0)
1162 sc->sc_free = 0;
1163 if (sc->sc_free > sc->sc_total)
1164 sc->sc_free = sc->sc_total;
1165 if (sc->sc_dinodes < 0)
1166 sc->sc_dinodes = 0;
1167
1168 return 0;
1169}
1170
1171/**
1172 * gfs2_statfs - Gather and return stats about the filesystem
1173 * @sb: The superblock
1174 * @statfsbuf: The buffer
1175 *
1176 * Returns: 0 on success or error code
1177 */
1178
1179static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1180{
1181 struct super_block *sb = dentry->d_sb;
1182 struct gfs2_sbd *sdp = sb->s_fs_info;
1183 struct gfs2_statfs_change_host sc;
1184 int error;
1185
1186 error = gfs2_rindex_update(sdp);
1187 if (error)
1188 return error;
1189
1190 if (gfs2_tune_get(sdp, gt_statfs_slow))
1191 error = gfs2_statfs_slow(sdp, &sc);
1192 else
1193 error = gfs2_statfs_i(sdp, &sc);
1194
1195 if (error)
1196 return error;
1197
1198 buf->f_type = GFS2_MAGIC;
1199 buf->f_bsize = sdp->sd_sb.sb_bsize;
1200 buf->f_blocks = sc.sc_total;
1201 buf->f_bfree = sc.sc_free;
1202 buf->f_bavail = sc.sc_free;
1203 buf->f_files = sc.sc_dinodes + sc.sc_free;
1204 buf->f_ffree = sc.sc_free;
1205 buf->f_namelen = GFS2_FNAMESIZE;
1206
1207 return 0;
1208}
1209
1210/**
1211 * gfs2_remount_fs - called when the FS is remounted
1212 * @sb: the filesystem
1213 * @flags: the remount flags
1214 * @data: extra data passed in (not used right now)
1215 *
1216 * Returns: errno
1217 */
1218
1219static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1220{
1221 struct gfs2_sbd *sdp = sb->s_fs_info;
1222 struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1223 struct gfs2_tune *gt = &sdp->sd_tune;
1224 int error;
1225
1226 sync_filesystem(sb);
1227
1228 spin_lock(>->gt_spin);
1229 args.ar_commit = gt->gt_logd_secs;
1230 args.ar_quota_quantum = gt->gt_quota_quantum;
1231 if (gt->gt_statfs_slow)
1232 args.ar_statfs_quantum = 0;
1233 else
1234 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1235 spin_unlock(>->gt_spin);
1236 error = gfs2_mount_args(&args, data);
1237 if (error)
1238 return error;
1239
1240 /* Not allowed to change locking details */
1241 if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1242 strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1243 strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1244 return -EINVAL;
1245
1246 /* Some flags must not be changed */
1247 if (args_neq(&args, &sdp->sd_args, spectator) ||
1248 args_neq(&args, &sdp->sd_args, localflocks) ||
1249 args_neq(&args, &sdp->sd_args, meta))
1250 return -EINVAL;
1251
1252 if (sdp->sd_args.ar_spectator)
1253 *flags |= MS_RDONLY;
1254
1255 if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1256 if (*flags & MS_RDONLY)
1257 error = gfs2_make_fs_ro(sdp);
1258 else
1259 error = gfs2_make_fs_rw(sdp);
1260 if (error)
1261 return error;
1262 }
1263
1264 sdp->sd_args = args;
1265 if (sdp->sd_args.ar_posix_acl)
1266 sb->s_flags |= MS_POSIXACL;
1267 else
1268 sb->s_flags &= ~MS_POSIXACL;
1269 if (sdp->sd_args.ar_nobarrier)
1270 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1271 else
1272 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1273 spin_lock(>->gt_spin);
1274 gt->gt_logd_secs = args.ar_commit;
1275 gt->gt_quota_quantum = args.ar_quota_quantum;
1276 if (args.ar_statfs_quantum) {
1277 gt->gt_statfs_slow = 0;
1278 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1279 }
1280 else {
1281 gt->gt_statfs_slow = 1;
1282 gt->gt_statfs_quantum = 30;
1283 }
1284 spin_unlock(>->gt_spin);
1285
1286 gfs2_online_uevent(sdp);
1287 return 0;
1288}
1289
1290/**
1291 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1292 * @inode: The inode to drop
1293 *
1294 * If we've received a callback on an iopen lock then its because a
1295 * remote node tried to deallocate the inode but failed due to this node
1296 * still having the inode open. Here we mark the link count zero
1297 * since we know that it must have reached zero if the GLF_DEMOTE flag
1298 * is set on the iopen glock. If we didn't do a disk read since the
1299 * remote node removed the final link then we might otherwise miss
1300 * this event. This check ensures that this node will deallocate the
1301 * inode's blocks, or alternatively pass the baton on to another
1302 * node for later deallocation.
1303 */
1304
1305static int gfs2_drop_inode(struct inode *inode)
1306{
1307 struct gfs2_inode *ip = GFS2_I(inode);
1308
1309 if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
1310 inode->i_nlink &&
1311 gfs2_holder_initialized(&ip->i_iopen_gh)) {
1312 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1313 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1314 clear_nlink(inode);
1315 }
1316 return generic_drop_inode(inode);
1317}
1318
1319static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1320{
1321 do {
1322 if (d1 == d2)
1323 return 1;
1324 d1 = d1->d_parent;
1325 } while (!IS_ROOT(d1));
1326 return 0;
1327}
1328
1329/**
1330 * gfs2_show_options - Show mount options for /proc/mounts
1331 * @s: seq_file structure
1332 * @root: root of this (sub)tree
1333 *
1334 * Returns: 0 on success or error code
1335 */
1336
1337static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1338{
1339 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1340 struct gfs2_args *args = &sdp->sd_args;
1341 int val;
1342
1343 if (is_ancestor(root, sdp->sd_master_dir))
1344 seq_puts(s, ",meta");
1345 if (args->ar_lockproto[0])
1346 seq_show_option(s, "lockproto", args->ar_lockproto);
1347 if (args->ar_locktable[0])
1348 seq_show_option(s, "locktable", args->ar_locktable);
1349 if (args->ar_hostdata[0])
1350 seq_show_option(s, "hostdata", args->ar_hostdata);
1351 if (args->ar_spectator)
1352 seq_puts(s, ",spectator");
1353 if (args->ar_localflocks)
1354 seq_puts(s, ",localflocks");
1355 if (args->ar_debug)
1356 seq_puts(s, ",debug");
1357 if (args->ar_posix_acl)
1358 seq_puts(s, ",acl");
1359 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1360 char *state;
1361 switch (args->ar_quota) {
1362 case GFS2_QUOTA_OFF:
1363 state = "off";
1364 break;
1365 case GFS2_QUOTA_ACCOUNT:
1366 state = "account";
1367 break;
1368 case GFS2_QUOTA_ON:
1369 state = "on";
1370 break;
1371 default:
1372 state = "unknown";
1373 break;
1374 }
1375 seq_printf(s, ",quota=%s", state);
1376 }
1377 if (args->ar_suiddir)
1378 seq_puts(s, ",suiddir");
1379 if (args->ar_data != GFS2_DATA_DEFAULT) {
1380 char *state;
1381 switch (args->ar_data) {
1382 case GFS2_DATA_WRITEBACK:
1383 state = "writeback";
1384 break;
1385 case GFS2_DATA_ORDERED:
1386 state = "ordered";
1387 break;
1388 default:
1389 state = "unknown";
1390 break;
1391 }
1392 seq_printf(s, ",data=%s", state);
1393 }
1394 if (args->ar_discard)
1395 seq_puts(s, ",discard");
1396 val = sdp->sd_tune.gt_logd_secs;
1397 if (val != 30)
1398 seq_printf(s, ",commit=%d", val);
1399 val = sdp->sd_tune.gt_statfs_quantum;
1400 if (val != 30)
1401 seq_printf(s, ",statfs_quantum=%d", val);
1402 else if (sdp->sd_tune.gt_statfs_slow)
1403 seq_puts(s, ",statfs_quantum=0");
1404 val = sdp->sd_tune.gt_quota_quantum;
1405 if (val != 60)
1406 seq_printf(s, ",quota_quantum=%d", val);
1407 if (args->ar_statfs_percent)
1408 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1409 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1410 const char *state;
1411
1412 switch (args->ar_errors) {
1413 case GFS2_ERRORS_WITHDRAW:
1414 state = "withdraw";
1415 break;
1416 case GFS2_ERRORS_PANIC:
1417 state = "panic";
1418 break;
1419 default:
1420 state = "unknown";
1421 break;
1422 }
1423 seq_printf(s, ",errors=%s", state);
1424 }
1425 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1426 seq_puts(s, ",nobarrier");
1427 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1428 seq_puts(s, ",demote_interface_used");
1429 if (args->ar_rgrplvb)
1430 seq_puts(s, ",rgrplvb");
1431 if (args->ar_loccookie)
1432 seq_puts(s, ",loccookie");
1433 return 0;
1434}
1435
1436static void gfs2_final_release_pages(struct gfs2_inode *ip)
1437{
1438 struct inode *inode = &ip->i_inode;
1439 struct gfs2_glock *gl = ip->i_gl;
1440
1441 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1442 truncate_inode_pages(&inode->i_data, 0);
1443
1444 if (atomic_read(&gl->gl_revokes) == 0) {
1445 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1446 clear_bit(GLF_DIRTY, &gl->gl_flags);
1447 }
1448}
1449
1450static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1451{
1452 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1453 struct gfs2_rgrpd *rgd;
1454 struct gfs2_holder gh;
1455 int error;
1456
1457 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1458 gfs2_consist_inode(ip);
1459 return -EIO;
1460 }
1461
1462 error = gfs2_rindex_update(sdp);
1463 if (error)
1464 return error;
1465
1466 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1467 if (error)
1468 return error;
1469
1470 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1471 if (!rgd) {
1472 gfs2_consist_inode(ip);
1473 error = -EIO;
1474 goto out_qs;
1475 }
1476
1477 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1478 if (error)
1479 goto out_qs;
1480
1481 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1482 sdp->sd_jdesc->jd_blocks);
1483 if (error)
1484 goto out_rg_gunlock;
1485
1486 gfs2_free_di(rgd, ip);
1487
1488 gfs2_final_release_pages(ip);
1489
1490 gfs2_trans_end(sdp);
1491
1492out_rg_gunlock:
1493 gfs2_glock_dq_uninit(&gh);
1494out_qs:
1495 gfs2_quota_unhold(ip);
1496 return error;
1497}
1498
1499/**
1500 * gfs2_evict_inode - Remove an inode from cache
1501 * @inode: The inode to evict
1502 *
1503 * There are three cases to consider:
1504 * 1. i_nlink == 0, we are final opener (and must deallocate)
1505 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1506 * 3. i_nlink > 0
1507 *
1508 * If the fs is read only, then we have to treat all cases as per #3
1509 * since we are unable to do any deallocation. The inode will be
1510 * deallocated by the next read/write node to attempt an allocation
1511 * in the same resource group
1512 *
1513 * We have to (at the moment) hold the inodes main lock to cover
1514 * the gap between unlocking the shared lock on the iopen lock and
1515 * taking the exclusive lock. I'd rather do a shared -> exclusive
1516 * conversion on the iopen lock, but we can change that later. This
1517 * is safe, just less efficient.
1518 */
1519
1520static void gfs2_evict_inode(struct inode *inode)
1521{
1522 struct super_block *sb = inode->i_sb;
1523 struct gfs2_sbd *sdp = sb->s_fs_info;
1524 struct gfs2_inode *ip = GFS2_I(inode);
1525 struct gfs2_holder gh;
1526 struct address_space *metamapping;
1527 int error;
1528
1529 if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1530 clear_inode(inode);
1531 return;
1532 }
1533
1534 if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1535 goto out;
1536
1537 /* Must not read inode block until block type has been verified */
1538 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1539 if (unlikely(error)) {
1540 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1541 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1542 gfs2_holder_uninit(&ip->i_iopen_gh);
1543 goto out;
1544 }
1545
1546 if (!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1547 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1548 if (error)
1549 goto out_truncate;
1550 }
1551
1552 if (test_bit(GIF_INVALID, &ip->i_flags)) {
1553 error = gfs2_inode_refresh(ip);
1554 if (error)
1555 goto out_truncate;
1556 }
1557
1558 if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1559 test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1560 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1561 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1562 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE,
1563 &ip->i_iopen_gh);
1564 error = gfs2_glock_nq(&ip->i_iopen_gh);
1565 if (error)
1566 goto out_truncate;
1567 }
1568
1569 /* Case 1 starts here */
1570
1571 if (S_ISDIR(inode->i_mode) &&
1572 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1573 error = gfs2_dir_exhash_dealloc(ip);
1574 if (error)
1575 goto out_unlock;
1576 }
1577
1578 if (ip->i_eattr) {
1579 error = gfs2_ea_dealloc(ip);
1580 if (error)
1581 goto out_unlock;
1582 }
1583
1584 if (!gfs2_is_stuffed(ip)) {
1585 error = gfs2_file_dealloc(ip);
1586 if (error)
1587 goto out_unlock;
1588 }
1589
1590 error = gfs2_dinode_dealloc(ip);
1591 goto out_unlock;
1592
1593out_truncate:
1594 gfs2_log_flush(sdp, ip->i_gl, NORMAL_FLUSH);
1595 metamapping = gfs2_glock2aspace(ip->i_gl);
1596 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1597 filemap_fdatawrite(metamapping);
1598 filemap_fdatawait(metamapping);
1599 }
1600 write_inode_now(inode, 1);
1601 gfs2_ail_flush(ip->i_gl, 0);
1602
1603 /* Case 2 starts here */
1604 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1605 if (error)
1606 goto out_unlock;
1607 /* Needs to be done before glock release & also in a transaction */
1608 truncate_inode_pages(&inode->i_data, 0);
1609 truncate_inode_pages(metamapping, 0);
1610 gfs2_trans_end(sdp);
1611
1612out_unlock:
1613 /* Error path for case 1 */
1614 if (gfs2_rs_active(&ip->i_res))
1615 gfs2_rs_deltree(&ip->i_res);
1616
1617 if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1618 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1619 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1620 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1621 }
1622 gfs2_holder_uninit(&ip->i_iopen_gh);
1623 }
1624 gfs2_glock_dq_uninit(&gh);
1625 if (error && error != GLR_TRYFAILED && error != -EROFS)
1626 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1627out:
1628 /* Case 3 starts here */
1629 truncate_inode_pages_final(&inode->i_data);
1630 gfs2_rsqa_delete(ip, NULL);
1631 gfs2_ordered_del_inode(ip);
1632 clear_inode(inode);
1633 gfs2_dir_hash_inval(ip);
1634 ip->i_gl->gl_object = NULL;
1635 flush_delayed_work(&ip->i_gl->gl_work);
1636 gfs2_glock_add_to_lru(ip->i_gl);
1637 gfs2_glock_put(ip->i_gl);
1638 ip->i_gl = NULL;
1639 if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1640 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1641 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1642 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1643 gfs2_holder_uninit(&ip->i_iopen_gh);
1644 }
1645}
1646
1647static struct inode *gfs2_alloc_inode(struct super_block *sb)
1648{
1649 struct gfs2_inode *ip;
1650
1651 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1652 if (ip) {
1653 ip->i_flags = 0;
1654 ip->i_gl = NULL;
1655 ip->i_rgd = NULL;
1656 memset(&ip->i_res, 0, sizeof(ip->i_res));
1657 RB_CLEAR_NODE(&ip->i_res.rs_node);
1658 ip->i_rahead = 0;
1659 }
1660 return &ip->i_inode;
1661}
1662
1663static void gfs2_i_callback(struct rcu_head *head)
1664{
1665 struct inode *inode = container_of(head, struct inode, i_rcu);
1666 kmem_cache_free(gfs2_inode_cachep, inode);
1667}
1668
1669static void gfs2_destroy_inode(struct inode *inode)
1670{
1671 call_rcu(&inode->i_rcu, gfs2_i_callback);
1672}
1673
1674const struct super_operations gfs2_super_ops = {
1675 .alloc_inode = gfs2_alloc_inode,
1676 .destroy_inode = gfs2_destroy_inode,
1677 .write_inode = gfs2_write_inode,
1678 .dirty_inode = gfs2_dirty_inode,
1679 .evict_inode = gfs2_evict_inode,
1680 .put_super = gfs2_put_super,
1681 .sync_fs = gfs2_sync_fs,
1682 .freeze_super = gfs2_freeze,
1683 .thaw_super = gfs2_unfreeze,
1684 .statfs = gfs2_statfs,
1685 .remount_fs = gfs2_remount_fs,
1686 .drop_inode = gfs2_drop_inode,
1687 .show_options = gfs2_show_options,
1688};
1689