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 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);
786do_unlock:
787 if (unlock_required)
788 gfs2_glock_dq_uninit(&gh);
789do_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
809static 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
836static 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
848static 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 */
866restart:
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
926static 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
940static 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
975static 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
989static 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
1012static 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
1075out:
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
1088static 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
1120static 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
1156static 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_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_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_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_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
1240static 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
1252static 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
1270static 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
1363static 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
1377static 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
1424out_rg_gunlock:
1425 gfs2_glock_dq_uninit(&al->al_rgd_gh);
1426out_rindex_relse:
1427 gfs2_glock_dq_uninit(&al->al_ri_gh);
1428out_qs:
1429 gfs2_quota_unhold(ip);
1430out:
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
1456static 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
1515out_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
1524out_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);
1532out:
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
1547static 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
1559static 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
1566static void gfs2_destroy_inode(struct inode *inode)
1567{
1568 call_rcu(&inode->i_rcu, gfs2_i_callback);
1569}
1570
1571const 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
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
30#include "gfs2.h"
31#include "incore.h"
32#include "bmap.h"
33#include "dir.h"
34#include "glock.h"
35#include "glops.h"
36#include "inode.h"
37#include "log.h"
38#include "meta_io.h"
39#include "quota.h"
40#include "recovery.h"
41#include "rgrp.h"
42#include "super.h"
43#include "trans.h"
44#include "util.h"
45#include "sys.h"
46#include "xattr.h"
47
48#define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
49
50enum {
51 Opt_lockproto,
52 Opt_locktable,
53 Opt_hostdata,
54 Opt_spectator,
55 Opt_ignore_local_fs,
56 Opt_localflocks,
57 Opt_localcaching,
58 Opt_debug,
59 Opt_nodebug,
60 Opt_upgrade,
61 Opt_acl,
62 Opt_noacl,
63 Opt_quota_off,
64 Opt_quota_account,
65 Opt_quota_on,
66 Opt_quota,
67 Opt_noquota,
68 Opt_suiddir,
69 Opt_nosuiddir,
70 Opt_data_writeback,
71 Opt_data_ordered,
72 Opt_meta,
73 Opt_discard,
74 Opt_nodiscard,
75 Opt_commit,
76 Opt_err_withdraw,
77 Opt_err_panic,
78 Opt_statfs_quantum,
79 Opt_statfs_percent,
80 Opt_quota_quantum,
81 Opt_barrier,
82 Opt_nobarrier,
83 Opt_rgrplvb,
84 Opt_norgrplvb,
85 Opt_error,
86};
87
88static const match_table_t tokens = {
89 {Opt_lockproto, "lockproto=%s"},
90 {Opt_locktable, "locktable=%s"},
91 {Opt_hostdata, "hostdata=%s"},
92 {Opt_spectator, "spectator"},
93 {Opt_spectator, "norecovery"},
94 {Opt_ignore_local_fs, "ignore_local_fs"},
95 {Opt_localflocks, "localflocks"},
96 {Opt_localcaching, "localcaching"},
97 {Opt_debug, "debug"},
98 {Opt_nodebug, "nodebug"},
99 {Opt_upgrade, "upgrade"},
100 {Opt_acl, "acl"},
101 {Opt_noacl, "noacl"},
102 {Opt_quota_off, "quota=off"},
103 {Opt_quota_account, "quota=account"},
104 {Opt_quota_on, "quota=on"},
105 {Opt_quota, "quota"},
106 {Opt_noquota, "noquota"},
107 {Opt_suiddir, "suiddir"},
108 {Opt_nosuiddir, "nosuiddir"},
109 {Opt_data_writeback, "data=writeback"},
110 {Opt_data_ordered, "data=ordered"},
111 {Opt_meta, "meta"},
112 {Opt_discard, "discard"},
113 {Opt_nodiscard, "nodiscard"},
114 {Opt_commit, "commit=%d"},
115 {Opt_err_withdraw, "errors=withdraw"},
116 {Opt_err_panic, "errors=panic"},
117 {Opt_statfs_quantum, "statfs_quantum=%d"},
118 {Opt_statfs_percent, "statfs_percent=%d"},
119 {Opt_quota_quantum, "quota_quantum=%d"},
120 {Opt_barrier, "barrier"},
121 {Opt_nobarrier, "nobarrier"},
122 {Opt_rgrplvb, "rgrplvb"},
123 {Opt_norgrplvb, "norgrplvb"},
124 {Opt_error, NULL}
125};
126
127/**
128 * gfs2_mount_args - Parse mount options
129 * @args: The structure into which the parsed options will be written
130 * @options: The options to parse
131 *
132 * Return: errno
133 */
134
135int gfs2_mount_args(struct gfs2_args *args, char *options)
136{
137 char *o;
138 int token;
139 substring_t tmp[MAX_OPT_ARGS];
140 int rv;
141
142 /* Split the options into tokens with the "," character and
143 process them */
144
145 while (1) {
146 o = strsep(&options, ",");
147 if (o == NULL)
148 break;
149 if (*o == '\0')
150 continue;
151
152 token = match_token(o, tokens, tmp);
153 switch (token) {
154 case Opt_lockproto:
155 match_strlcpy(args->ar_lockproto, &tmp[0],
156 GFS2_LOCKNAME_LEN);
157 break;
158 case Opt_locktable:
159 match_strlcpy(args->ar_locktable, &tmp[0],
160 GFS2_LOCKNAME_LEN);
161 break;
162 case Opt_hostdata:
163 match_strlcpy(args->ar_hostdata, &tmp[0],
164 GFS2_LOCKNAME_LEN);
165 break;
166 case Opt_spectator:
167 args->ar_spectator = 1;
168 break;
169 case Opt_ignore_local_fs:
170 /* Retained for backwards compat only */
171 break;
172 case Opt_localflocks:
173 args->ar_localflocks = 1;
174 break;
175 case Opt_localcaching:
176 /* Retained for backwards compat only */
177 break;
178 case Opt_debug:
179 if (args->ar_errors == GFS2_ERRORS_PANIC) {
180 pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
181 return -EINVAL;
182 }
183 args->ar_debug = 1;
184 break;
185 case Opt_nodebug:
186 args->ar_debug = 0;
187 break;
188 case Opt_upgrade:
189 /* Retained for backwards compat only */
190 break;
191 case Opt_acl:
192 args->ar_posix_acl = 1;
193 break;
194 case Opt_noacl:
195 args->ar_posix_acl = 0;
196 break;
197 case Opt_quota_off:
198 case Opt_noquota:
199 args->ar_quota = GFS2_QUOTA_OFF;
200 break;
201 case Opt_quota_account:
202 args->ar_quota = GFS2_QUOTA_ACCOUNT;
203 break;
204 case Opt_quota_on:
205 case Opt_quota:
206 args->ar_quota = GFS2_QUOTA_ON;
207 break;
208 case Opt_suiddir:
209 args->ar_suiddir = 1;
210 break;
211 case Opt_nosuiddir:
212 args->ar_suiddir = 0;
213 break;
214 case Opt_data_writeback:
215 args->ar_data = GFS2_DATA_WRITEBACK;
216 break;
217 case Opt_data_ordered:
218 args->ar_data = GFS2_DATA_ORDERED;
219 break;
220 case Opt_meta:
221 args->ar_meta = 1;
222 break;
223 case Opt_discard:
224 args->ar_discard = 1;
225 break;
226 case Opt_nodiscard:
227 args->ar_discard = 0;
228 break;
229 case Opt_commit:
230 rv = match_int(&tmp[0], &args->ar_commit);
231 if (rv || args->ar_commit <= 0) {
232 pr_warn("commit mount option requires a positive numeric argument\n");
233 return rv ? rv : -EINVAL;
234 }
235 break;
236 case Opt_statfs_quantum:
237 rv = match_int(&tmp[0], &args->ar_statfs_quantum);
238 if (rv || args->ar_statfs_quantum < 0) {
239 pr_warn("statfs_quantum mount option requires a non-negative numeric argument\n");
240 return rv ? rv : -EINVAL;
241 }
242 break;
243 case Opt_quota_quantum:
244 rv = match_int(&tmp[0], &args->ar_quota_quantum);
245 if (rv || args->ar_quota_quantum <= 0) {
246 pr_warn("quota_quantum mount option requires a positive numeric argument\n");
247 return rv ? rv : -EINVAL;
248 }
249 break;
250 case Opt_statfs_percent:
251 rv = match_int(&tmp[0], &args->ar_statfs_percent);
252 if (rv || args->ar_statfs_percent < 0 ||
253 args->ar_statfs_percent > 100) {
254 pr_warn("statfs_percent mount option requires a numeric argument between 0 and 100\n");
255 return rv ? rv : -EINVAL;
256 }
257 break;
258 case Opt_err_withdraw:
259 args->ar_errors = GFS2_ERRORS_WITHDRAW;
260 break;
261 case Opt_err_panic:
262 if (args->ar_debug) {
263 pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
264 return -EINVAL;
265 }
266 args->ar_errors = GFS2_ERRORS_PANIC;
267 break;
268 case Opt_barrier:
269 args->ar_nobarrier = 0;
270 break;
271 case Opt_nobarrier:
272 args->ar_nobarrier = 1;
273 break;
274 case Opt_rgrplvb:
275 args->ar_rgrplvb = 1;
276 break;
277 case Opt_norgrplvb:
278 args->ar_rgrplvb = 0;
279 break;
280 case Opt_error:
281 default:
282 pr_warn("invalid mount option: %s\n", o);
283 return -EINVAL;
284 }
285 }
286
287 return 0;
288}
289
290/**
291 * gfs2_jindex_free - Clear all the journal index information
292 * @sdp: The GFS2 superblock
293 *
294 */
295
296void gfs2_jindex_free(struct gfs2_sbd *sdp)
297{
298 struct list_head list;
299 struct gfs2_jdesc *jd;
300
301 spin_lock(&sdp->sd_jindex_spin);
302 list_add(&list, &sdp->sd_jindex_list);
303 list_del_init(&sdp->sd_jindex_list);
304 sdp->sd_journals = 0;
305 spin_unlock(&sdp->sd_jindex_spin);
306
307 while (!list_empty(&list)) {
308 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
309 gfs2_free_journal_extents(jd);
310 list_del(&jd->jd_list);
311 iput(jd->jd_inode);
312 kfree(jd);
313 }
314}
315
316static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
317{
318 struct gfs2_jdesc *jd;
319 int found = 0;
320
321 list_for_each_entry(jd, head, jd_list) {
322 if (jd->jd_jid == jid) {
323 found = 1;
324 break;
325 }
326 }
327
328 if (!found)
329 jd = NULL;
330
331 return jd;
332}
333
334struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
335{
336 struct gfs2_jdesc *jd;
337
338 spin_lock(&sdp->sd_jindex_spin);
339 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
340 spin_unlock(&sdp->sd_jindex_spin);
341
342 return jd;
343}
344
345int gfs2_jdesc_check(struct gfs2_jdesc *jd)
346{
347 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
348 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
349 u64 size = i_size_read(jd->jd_inode);
350
351 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, 1 << 30))
352 return -EIO;
353
354 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
355
356 if (gfs2_write_alloc_required(ip, 0, size)) {
357 gfs2_consist_inode(ip);
358 return -EIO;
359 }
360
361 return 0;
362}
363
364static int init_threads(struct gfs2_sbd *sdp)
365{
366 struct task_struct *p;
367 int error = 0;
368
369 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
370 if (IS_ERR(p)) {
371 error = PTR_ERR(p);
372 fs_err(sdp, "can't start logd thread: %d\n", error);
373 return error;
374 }
375 sdp->sd_logd_process = p;
376
377 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
378 if (IS_ERR(p)) {
379 error = PTR_ERR(p);
380 fs_err(sdp, "can't start quotad thread: %d\n", error);
381 goto fail;
382 }
383 sdp->sd_quotad_process = p;
384 return 0;
385
386fail:
387 kthread_stop(sdp->sd_logd_process);
388 return error;
389}
390
391/**
392 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
393 * @sdp: the filesystem
394 *
395 * Returns: errno
396 */
397
398int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
399{
400 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
401 struct gfs2_glock *j_gl = ip->i_gl;
402 struct gfs2_holder t_gh;
403 struct gfs2_log_header_host head;
404 int error;
405
406 error = init_threads(sdp);
407 if (error)
408 return error;
409
410 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
411 if (error)
412 goto fail_threads;
413
414 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
415
416 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
417 if (error)
418 goto fail;
419
420 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
421 gfs2_consist(sdp);
422 error = -EIO;
423 goto fail;
424 }
425
426 /* Initialize some head of the log stuff */
427 sdp->sd_log_sequence = head.lh_sequence + 1;
428 gfs2_log_pointers_init(sdp, head.lh_blkno);
429
430 error = gfs2_quota_init(sdp);
431 if (error)
432 goto fail;
433
434 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
435
436 gfs2_glock_dq_uninit(&t_gh);
437
438 return 0;
439
440fail:
441 t_gh.gh_flags |= GL_NOCACHE;
442 gfs2_glock_dq_uninit(&t_gh);
443fail_threads:
444 kthread_stop(sdp->sd_quotad_process);
445 kthread_stop(sdp->sd_logd_process);
446 return error;
447}
448
449void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
450{
451 const struct gfs2_statfs_change *str = buf;
452
453 sc->sc_total = be64_to_cpu(str->sc_total);
454 sc->sc_free = be64_to_cpu(str->sc_free);
455 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
456}
457
458static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
459{
460 struct gfs2_statfs_change *str = buf;
461
462 str->sc_total = cpu_to_be64(sc->sc_total);
463 str->sc_free = cpu_to_be64(sc->sc_free);
464 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
465}
466
467int gfs2_statfs_init(struct gfs2_sbd *sdp)
468{
469 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
470 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
471 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
472 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
473 struct buffer_head *m_bh, *l_bh;
474 struct gfs2_holder gh;
475 int error;
476
477 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
478 &gh);
479 if (error)
480 return error;
481
482 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
483 if (error)
484 goto out;
485
486 if (sdp->sd_args.ar_spectator) {
487 spin_lock(&sdp->sd_statfs_spin);
488 gfs2_statfs_change_in(m_sc, m_bh->b_data +
489 sizeof(struct gfs2_dinode));
490 spin_unlock(&sdp->sd_statfs_spin);
491 } else {
492 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
493 if (error)
494 goto out_m_bh;
495
496 spin_lock(&sdp->sd_statfs_spin);
497 gfs2_statfs_change_in(m_sc, m_bh->b_data +
498 sizeof(struct gfs2_dinode));
499 gfs2_statfs_change_in(l_sc, l_bh->b_data +
500 sizeof(struct gfs2_dinode));
501 spin_unlock(&sdp->sd_statfs_spin);
502
503 brelse(l_bh);
504 }
505
506out_m_bh:
507 brelse(m_bh);
508out:
509 gfs2_glock_dq_uninit(&gh);
510 return 0;
511}
512
513void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
514 s64 dinodes)
515{
516 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
517 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
518 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
519 struct buffer_head *l_bh;
520 s64 x, y;
521 int need_sync = 0;
522 int error;
523
524 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
525 if (error)
526 return;
527
528 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
529
530 spin_lock(&sdp->sd_statfs_spin);
531 l_sc->sc_total += total;
532 l_sc->sc_free += free;
533 l_sc->sc_dinodes += dinodes;
534 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
535 if (sdp->sd_args.ar_statfs_percent) {
536 x = 100 * l_sc->sc_free;
537 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
538 if (x >= y || x <= -y)
539 need_sync = 1;
540 }
541 spin_unlock(&sdp->sd_statfs_spin);
542
543 brelse(l_bh);
544 if (need_sync)
545 gfs2_wake_up_statfs(sdp);
546}
547
548void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
549 struct buffer_head *l_bh)
550{
551 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
552 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
553 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
554 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
555
556 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
557
558 spin_lock(&sdp->sd_statfs_spin);
559 m_sc->sc_total += l_sc->sc_total;
560 m_sc->sc_free += l_sc->sc_free;
561 m_sc->sc_dinodes += l_sc->sc_dinodes;
562 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
563 memset(l_bh->b_data + sizeof(struct gfs2_dinode),
564 0, sizeof(struct gfs2_statfs_change));
565 spin_unlock(&sdp->sd_statfs_spin);
566
567 gfs2_trans_add_meta(m_ip->i_gl, m_bh);
568 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
569}
570
571int gfs2_statfs_sync(struct super_block *sb, int type)
572{
573 struct gfs2_sbd *sdp = sb->s_fs_info;
574 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
575 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
576 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
577 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
578 struct gfs2_holder gh;
579 struct buffer_head *m_bh, *l_bh;
580 int error;
581
582 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
583 &gh);
584 if (error)
585 return error;
586
587 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
588 if (error)
589 goto out;
590
591 spin_lock(&sdp->sd_statfs_spin);
592 gfs2_statfs_change_in(m_sc, m_bh->b_data +
593 sizeof(struct gfs2_dinode));
594 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
595 spin_unlock(&sdp->sd_statfs_spin);
596 goto out_bh;
597 }
598 spin_unlock(&sdp->sd_statfs_spin);
599
600 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
601 if (error)
602 goto out_bh;
603
604 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
605 if (error)
606 goto out_bh2;
607
608 update_statfs(sdp, m_bh, l_bh);
609 sdp->sd_statfs_force_sync = 0;
610
611 gfs2_trans_end(sdp);
612
613out_bh2:
614 brelse(l_bh);
615out_bh:
616 brelse(m_bh);
617out:
618 gfs2_glock_dq_uninit(&gh);
619 return error;
620}
621
622struct lfcc {
623 struct list_head list;
624 struct gfs2_holder gh;
625};
626
627/**
628 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
629 * journals are clean
630 * @sdp: the file system
631 * @state: the state to put the transaction lock into
632 * @t_gh: the hold on the transaction lock
633 *
634 * Returns: errno
635 */
636
637static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
638 struct gfs2_holder *t_gh)
639{
640 struct gfs2_inode *ip;
641 struct gfs2_jdesc *jd;
642 struct lfcc *lfcc;
643 LIST_HEAD(list);
644 struct gfs2_log_header_host lh;
645 int error;
646
647 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
648 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
649 if (!lfcc) {
650 error = -ENOMEM;
651 goto out;
652 }
653 ip = GFS2_I(jd->jd_inode);
654 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
655 if (error) {
656 kfree(lfcc);
657 goto out;
658 }
659 list_add(&lfcc->list, &list);
660 }
661
662 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
663 GL_NOCACHE, t_gh);
664
665 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
666 error = gfs2_jdesc_check(jd);
667 if (error)
668 break;
669 error = gfs2_find_jhead(jd, &lh);
670 if (error)
671 break;
672 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
673 error = -EBUSY;
674 break;
675 }
676 }
677
678 if (error)
679 gfs2_glock_dq_uninit(t_gh);
680
681out:
682 while (!list_empty(&list)) {
683 lfcc = list_entry(list.next, struct lfcc, list);
684 list_del(&lfcc->list);
685 gfs2_glock_dq_uninit(&lfcc->gh);
686 kfree(lfcc);
687 }
688 return error;
689}
690
691void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
692{
693 struct gfs2_dinode *str = buf;
694
695 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
696 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
697 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
698 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
699 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
700 str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
701 str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
702 str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
703 str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
704 str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
705 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
706 str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
707 str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
708 str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
709
710 str->di_goal_meta = cpu_to_be64(ip->i_goal);
711 str->di_goal_data = cpu_to_be64(ip->i_goal);
712 str->di_generation = cpu_to_be64(ip->i_generation);
713
714 str->di_flags = cpu_to_be32(ip->i_diskflags);
715 str->di_height = cpu_to_be16(ip->i_height);
716 str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
717 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
718 GFS2_FORMAT_DE : 0);
719 str->di_depth = cpu_to_be16(ip->i_depth);
720 str->di_entries = cpu_to_be32(ip->i_entries);
721
722 str->di_eattr = cpu_to_be64(ip->i_eattr);
723 str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
724 str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
725 str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
726}
727
728/**
729 * gfs2_write_inode - Make sure the inode is stable on the disk
730 * @inode: The inode
731 * @wbc: The writeback control structure
732 *
733 * Returns: errno
734 */
735
736static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
737{
738 struct gfs2_inode *ip = GFS2_I(inode);
739 struct gfs2_sbd *sdp = GFS2_SB(inode);
740 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
741 struct backing_dev_info *bdi = metamapping->backing_dev_info;
742 int ret = 0;
743
744 if (wbc->sync_mode == WB_SYNC_ALL)
745 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
746 if (bdi->dirty_exceeded)
747 gfs2_ail1_flush(sdp, wbc);
748 else
749 filemap_fdatawrite(metamapping);
750 if (wbc->sync_mode == WB_SYNC_ALL)
751 ret = filemap_fdatawait(metamapping);
752 if (ret)
753 mark_inode_dirty_sync(inode);
754 return ret;
755}
756
757/**
758 * gfs2_dirty_inode - check for atime updates
759 * @inode: The inode in question
760 * @flags: The type of dirty
761 *
762 * Unfortunately it can be called under any combination of inode
763 * glock and transaction lock, so we have to check carefully.
764 *
765 * At the moment this deals only with atime - it should be possible
766 * to expand that role in future, once a review of the locking has
767 * been carried out.
768 */
769
770static void gfs2_dirty_inode(struct inode *inode, int flags)
771{
772 struct gfs2_inode *ip = GFS2_I(inode);
773 struct gfs2_sbd *sdp = GFS2_SB(inode);
774 struct buffer_head *bh;
775 struct gfs2_holder gh;
776 int need_unlock = 0;
777 int need_endtrans = 0;
778 int ret;
779
780 if (!(flags & (I_DIRTY_DATASYNC|I_DIRTY_SYNC)))
781 return;
782
783 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
784 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
785 if (ret) {
786 fs_err(sdp, "dirty_inode: glock %d\n", ret);
787 return;
788 }
789 need_unlock = 1;
790 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
791 return;
792
793 if (current->journal_info == NULL) {
794 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
795 if (ret) {
796 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
797 goto out;
798 }
799 need_endtrans = 1;
800 }
801
802 ret = gfs2_meta_inode_buffer(ip, &bh);
803 if (ret == 0) {
804 gfs2_trans_add_meta(ip->i_gl, bh);
805 gfs2_dinode_out(ip, bh->b_data);
806 brelse(bh);
807 }
808
809 if (need_endtrans)
810 gfs2_trans_end(sdp);
811out:
812 if (need_unlock)
813 gfs2_glock_dq_uninit(&gh);
814}
815
816/**
817 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
818 * @sdp: the filesystem
819 *
820 * Returns: errno
821 */
822
823static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
824{
825 struct gfs2_holder t_gh;
826 int error;
827
828 kthread_stop(sdp->sd_quotad_process);
829 kthread_stop(sdp->sd_logd_process);
830
831 flush_workqueue(gfs2_delete_workqueue);
832 gfs2_quota_sync(sdp->sd_vfs, 0);
833 gfs2_statfs_sync(sdp->sd_vfs, 0);
834
835 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
836 &t_gh);
837 if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
838 return error;
839
840 gfs2_meta_syncfs(sdp);
841 gfs2_log_shutdown(sdp);
842
843 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
844
845 if (t_gh.gh_gl)
846 gfs2_glock_dq_uninit(&t_gh);
847
848 gfs2_quota_cleanup(sdp);
849
850 return error;
851}
852
853static int gfs2_umount_recovery_wait(void *word)
854{
855 schedule();
856 return 0;
857}
858
859/**
860 * gfs2_put_super - Unmount the filesystem
861 * @sb: The VFS superblock
862 *
863 */
864
865static void gfs2_put_super(struct super_block *sb)
866{
867 struct gfs2_sbd *sdp = sb->s_fs_info;
868 int error;
869 struct gfs2_jdesc *jd;
870
871 /* No more recovery requests */
872 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
873 smp_mb();
874
875 /* Wait on outstanding recovery */
876restart:
877 spin_lock(&sdp->sd_jindex_spin);
878 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
879 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
880 continue;
881 spin_unlock(&sdp->sd_jindex_spin);
882 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
883 gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
884 goto restart;
885 }
886 spin_unlock(&sdp->sd_jindex_spin);
887
888 if (!(sb->s_flags & MS_RDONLY)) {
889 error = gfs2_make_fs_ro(sdp);
890 if (error)
891 gfs2_io_error(sdp);
892 }
893 /* At this point, we're through modifying the disk */
894
895 /* Release stuff */
896
897 iput(sdp->sd_jindex);
898 iput(sdp->sd_statfs_inode);
899 iput(sdp->sd_rindex);
900 iput(sdp->sd_quota_inode);
901
902 gfs2_glock_put(sdp->sd_rename_gl);
903 gfs2_glock_put(sdp->sd_trans_gl);
904
905 if (!sdp->sd_args.ar_spectator) {
906 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
907 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
908 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
909 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
910 iput(sdp->sd_sc_inode);
911 iput(sdp->sd_qc_inode);
912 }
913
914 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
915 gfs2_clear_rgrpd(sdp);
916 gfs2_jindex_free(sdp);
917 /* Take apart glock structures and buffer lists */
918 gfs2_gl_hash_clear(sdp);
919 /* Unmount the locking protocol */
920 gfs2_lm_unmount(sdp);
921
922 /* At this point, we're through participating in the lockspace */
923 gfs2_sys_fs_del(sdp);
924}
925
926/**
927 * gfs2_sync_fs - sync the filesystem
928 * @sb: the superblock
929 *
930 * Flushes the log to disk.
931 */
932
933static int gfs2_sync_fs(struct super_block *sb, int wait)
934{
935 struct gfs2_sbd *sdp = sb->s_fs_info;
936
937 gfs2_quota_sync(sb, -1);
938 if (wait && sdp)
939 gfs2_log_flush(sdp, NULL);
940 return 0;
941}
942
943/**
944 * gfs2_freeze - prevent further writes to the filesystem
945 * @sb: the VFS structure for the filesystem
946 *
947 */
948
949static int gfs2_freeze(struct super_block *sb)
950{
951 struct gfs2_sbd *sdp = sb->s_fs_info;
952 int error;
953
954 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
955 return -EINVAL;
956
957 for (;;) {
958 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
959 if (!error)
960 break;
961
962 switch (error) {
963 case -EBUSY:
964 fs_err(sdp, "waiting for recovery before freeze\n");
965 break;
966
967 default:
968 fs_err(sdp, "error freezing FS: %d\n", error);
969 break;
970 }
971
972 fs_err(sdp, "retrying...\n");
973 msleep(1000);
974 }
975 return 0;
976}
977
978/**
979 * gfs2_unfreeze - reallow writes to the filesystem
980 * @sb: the VFS structure for the filesystem
981 *
982 */
983
984static int gfs2_unfreeze(struct super_block *sb)
985{
986 struct gfs2_sbd *sdp = sb->s_fs_info;
987
988 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
989 return 0;
990}
991
992/**
993 * statfs_fill - fill in the sg for a given RG
994 * @rgd: the RG
995 * @sc: the sc structure
996 *
997 * Returns: 0 on success, -ESTALE if the LVB is invalid
998 */
999
1000static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1001 struct gfs2_statfs_change_host *sc)
1002{
1003 gfs2_rgrp_verify(rgd);
1004 sc->sc_total += rgd->rd_data;
1005 sc->sc_free += rgd->rd_free;
1006 sc->sc_dinodes += rgd->rd_dinodes;
1007 return 0;
1008}
1009
1010/**
1011 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1012 * @sdp: the filesystem
1013 * @sc: the sc info that will be returned
1014 *
1015 * Any error (other than a signal) will cause this routine to fall back
1016 * to the synchronous version.
1017 *
1018 * FIXME: This really shouldn't busy wait like this.
1019 *
1020 * Returns: errno
1021 */
1022
1023static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1024{
1025 struct gfs2_rgrpd *rgd_next;
1026 struct gfs2_holder *gha, *gh;
1027 unsigned int slots = 64;
1028 unsigned int x;
1029 int done;
1030 int error = 0, err;
1031
1032 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1033 gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1034 if (!gha)
1035 return -ENOMEM;
1036
1037 rgd_next = gfs2_rgrpd_get_first(sdp);
1038
1039 for (;;) {
1040 done = 1;
1041
1042 for (x = 0; x < slots; x++) {
1043 gh = gha + x;
1044
1045 if (gh->gh_gl && gfs2_glock_poll(gh)) {
1046 err = gfs2_glock_wait(gh);
1047 if (err) {
1048 gfs2_holder_uninit(gh);
1049 error = err;
1050 } else {
1051 if (!error)
1052 error = statfs_slow_fill(
1053 gh->gh_gl->gl_object, sc);
1054 gfs2_glock_dq_uninit(gh);
1055 }
1056 }
1057
1058 if (gh->gh_gl)
1059 done = 0;
1060 else if (rgd_next && !error) {
1061 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1062 LM_ST_SHARED,
1063 GL_ASYNC,
1064 gh);
1065 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1066 done = 0;
1067 }
1068
1069 if (signal_pending(current))
1070 error = -ERESTARTSYS;
1071 }
1072
1073 if (done)
1074 break;
1075
1076 yield();
1077 }
1078
1079 kfree(gha);
1080 return error;
1081}
1082
1083/**
1084 * gfs2_statfs_i - Do a statfs
1085 * @sdp: the filesystem
1086 * @sg: the sg structure
1087 *
1088 * Returns: errno
1089 */
1090
1091static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1092{
1093 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1094 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1095
1096 spin_lock(&sdp->sd_statfs_spin);
1097
1098 *sc = *m_sc;
1099 sc->sc_total += l_sc->sc_total;
1100 sc->sc_free += l_sc->sc_free;
1101 sc->sc_dinodes += l_sc->sc_dinodes;
1102
1103 spin_unlock(&sdp->sd_statfs_spin);
1104
1105 if (sc->sc_free < 0)
1106 sc->sc_free = 0;
1107 if (sc->sc_free > sc->sc_total)
1108 sc->sc_free = sc->sc_total;
1109 if (sc->sc_dinodes < 0)
1110 sc->sc_dinodes = 0;
1111
1112 return 0;
1113}
1114
1115/**
1116 * gfs2_statfs - Gather and return stats about the filesystem
1117 * @sb: The superblock
1118 * @statfsbuf: The buffer
1119 *
1120 * Returns: 0 on success or error code
1121 */
1122
1123static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1124{
1125 struct super_block *sb = dentry->d_inode->i_sb;
1126 struct gfs2_sbd *sdp = sb->s_fs_info;
1127 struct gfs2_statfs_change_host sc;
1128 int error;
1129
1130 error = gfs2_rindex_update(sdp);
1131 if (error)
1132 return error;
1133
1134 if (gfs2_tune_get(sdp, gt_statfs_slow))
1135 error = gfs2_statfs_slow(sdp, &sc);
1136 else
1137 error = gfs2_statfs_i(sdp, &sc);
1138
1139 if (error)
1140 return error;
1141
1142 buf->f_type = GFS2_MAGIC;
1143 buf->f_bsize = sdp->sd_sb.sb_bsize;
1144 buf->f_blocks = sc.sc_total;
1145 buf->f_bfree = sc.sc_free;
1146 buf->f_bavail = sc.sc_free;
1147 buf->f_files = sc.sc_dinodes + sc.sc_free;
1148 buf->f_ffree = sc.sc_free;
1149 buf->f_namelen = GFS2_FNAMESIZE;
1150
1151 return 0;
1152}
1153
1154/**
1155 * gfs2_remount_fs - called when the FS is remounted
1156 * @sb: the filesystem
1157 * @flags: the remount flags
1158 * @data: extra data passed in (not used right now)
1159 *
1160 * Returns: errno
1161 */
1162
1163static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1164{
1165 struct gfs2_sbd *sdp = sb->s_fs_info;
1166 struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1167 struct gfs2_tune *gt = &sdp->sd_tune;
1168 int error;
1169
1170 sync_filesystem(sb);
1171
1172 spin_lock(>->gt_spin);
1173 args.ar_commit = gt->gt_logd_secs;
1174 args.ar_quota_quantum = gt->gt_quota_quantum;
1175 if (gt->gt_statfs_slow)
1176 args.ar_statfs_quantum = 0;
1177 else
1178 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1179 spin_unlock(>->gt_spin);
1180 error = gfs2_mount_args(&args, data);
1181 if (error)
1182 return error;
1183
1184 /* Not allowed to change locking details */
1185 if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1186 strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1187 strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1188 return -EINVAL;
1189
1190 /* Some flags must not be changed */
1191 if (args_neq(&args, &sdp->sd_args, spectator) ||
1192 args_neq(&args, &sdp->sd_args, localflocks) ||
1193 args_neq(&args, &sdp->sd_args, meta))
1194 return -EINVAL;
1195
1196 if (sdp->sd_args.ar_spectator)
1197 *flags |= MS_RDONLY;
1198
1199 if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1200 if (*flags & MS_RDONLY)
1201 error = gfs2_make_fs_ro(sdp);
1202 else
1203 error = gfs2_make_fs_rw(sdp);
1204 if (error)
1205 return error;
1206 }
1207
1208 sdp->sd_args = args;
1209 if (sdp->sd_args.ar_posix_acl)
1210 sb->s_flags |= MS_POSIXACL;
1211 else
1212 sb->s_flags &= ~MS_POSIXACL;
1213 if (sdp->sd_args.ar_nobarrier)
1214 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1215 else
1216 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1217 spin_lock(>->gt_spin);
1218 gt->gt_logd_secs = args.ar_commit;
1219 gt->gt_quota_quantum = args.ar_quota_quantum;
1220 if (args.ar_statfs_quantum) {
1221 gt->gt_statfs_slow = 0;
1222 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1223 }
1224 else {
1225 gt->gt_statfs_slow = 1;
1226 gt->gt_statfs_quantum = 30;
1227 }
1228 spin_unlock(>->gt_spin);
1229
1230 gfs2_online_uevent(sdp);
1231 return 0;
1232}
1233
1234/**
1235 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1236 * @inode: The inode to drop
1237 *
1238 * If we've received a callback on an iopen lock then its because a
1239 * remote node tried to deallocate the inode but failed due to this node
1240 * still having the inode open. Here we mark the link count zero
1241 * since we know that it must have reached zero if the GLF_DEMOTE flag
1242 * is set on the iopen glock. If we didn't do a disk read since the
1243 * remote node removed the final link then we might otherwise miss
1244 * this event. This check ensures that this node will deallocate the
1245 * inode's blocks, or alternatively pass the baton on to another
1246 * node for later deallocation.
1247 */
1248
1249static int gfs2_drop_inode(struct inode *inode)
1250{
1251 struct gfs2_inode *ip = GFS2_I(inode);
1252
1253 if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) && inode->i_nlink) {
1254 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1255 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1256 clear_nlink(inode);
1257 }
1258 return generic_drop_inode(inode);
1259}
1260
1261static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1262{
1263 do {
1264 if (d1 == d2)
1265 return 1;
1266 d1 = d1->d_parent;
1267 } while (!IS_ROOT(d1));
1268 return 0;
1269}
1270
1271/**
1272 * gfs2_show_options - Show mount options for /proc/mounts
1273 * @s: seq_file structure
1274 * @root: root of this (sub)tree
1275 *
1276 * Returns: 0 on success or error code
1277 */
1278
1279static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1280{
1281 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1282 struct gfs2_args *args = &sdp->sd_args;
1283 int val;
1284
1285 if (is_ancestor(root, sdp->sd_master_dir))
1286 seq_printf(s, ",meta");
1287 if (args->ar_lockproto[0])
1288 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1289 if (args->ar_locktable[0])
1290 seq_printf(s, ",locktable=%s", args->ar_locktable);
1291 if (args->ar_hostdata[0])
1292 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1293 if (args->ar_spectator)
1294 seq_printf(s, ",spectator");
1295 if (args->ar_localflocks)
1296 seq_printf(s, ",localflocks");
1297 if (args->ar_debug)
1298 seq_printf(s, ",debug");
1299 if (args->ar_posix_acl)
1300 seq_printf(s, ",acl");
1301 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1302 char *state;
1303 switch (args->ar_quota) {
1304 case GFS2_QUOTA_OFF:
1305 state = "off";
1306 break;
1307 case GFS2_QUOTA_ACCOUNT:
1308 state = "account";
1309 break;
1310 case GFS2_QUOTA_ON:
1311 state = "on";
1312 break;
1313 default:
1314 state = "unknown";
1315 break;
1316 }
1317 seq_printf(s, ",quota=%s", state);
1318 }
1319 if (args->ar_suiddir)
1320 seq_printf(s, ",suiddir");
1321 if (args->ar_data != GFS2_DATA_DEFAULT) {
1322 char *state;
1323 switch (args->ar_data) {
1324 case GFS2_DATA_WRITEBACK:
1325 state = "writeback";
1326 break;
1327 case GFS2_DATA_ORDERED:
1328 state = "ordered";
1329 break;
1330 default:
1331 state = "unknown";
1332 break;
1333 }
1334 seq_printf(s, ",data=%s", state);
1335 }
1336 if (args->ar_discard)
1337 seq_printf(s, ",discard");
1338 val = sdp->sd_tune.gt_logd_secs;
1339 if (val != 30)
1340 seq_printf(s, ",commit=%d", val);
1341 val = sdp->sd_tune.gt_statfs_quantum;
1342 if (val != 30)
1343 seq_printf(s, ",statfs_quantum=%d", val);
1344 else if (sdp->sd_tune.gt_statfs_slow)
1345 seq_puts(s, ",statfs_quantum=0");
1346 val = sdp->sd_tune.gt_quota_quantum;
1347 if (val != 60)
1348 seq_printf(s, ",quota_quantum=%d", val);
1349 if (args->ar_statfs_percent)
1350 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1351 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1352 const char *state;
1353
1354 switch (args->ar_errors) {
1355 case GFS2_ERRORS_WITHDRAW:
1356 state = "withdraw";
1357 break;
1358 case GFS2_ERRORS_PANIC:
1359 state = "panic";
1360 break;
1361 default:
1362 state = "unknown";
1363 break;
1364 }
1365 seq_printf(s, ",errors=%s", state);
1366 }
1367 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1368 seq_printf(s, ",nobarrier");
1369 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1370 seq_printf(s, ",demote_interface_used");
1371 if (args->ar_rgrplvb)
1372 seq_printf(s, ",rgrplvb");
1373 return 0;
1374}
1375
1376static void gfs2_final_release_pages(struct gfs2_inode *ip)
1377{
1378 struct inode *inode = &ip->i_inode;
1379 struct gfs2_glock *gl = ip->i_gl;
1380
1381 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1382 truncate_inode_pages(&inode->i_data, 0);
1383
1384 if (atomic_read(&gl->gl_revokes) == 0) {
1385 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1386 clear_bit(GLF_DIRTY, &gl->gl_flags);
1387 }
1388}
1389
1390static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1391{
1392 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1393 struct gfs2_rgrpd *rgd;
1394 struct gfs2_holder gh;
1395 int error;
1396
1397 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1398 gfs2_consist_inode(ip);
1399 return -EIO;
1400 }
1401
1402 error = gfs2_rindex_update(sdp);
1403 if (error)
1404 return error;
1405
1406 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1407 if (error)
1408 return error;
1409
1410 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1411 if (!rgd) {
1412 gfs2_consist_inode(ip);
1413 error = -EIO;
1414 goto out_qs;
1415 }
1416
1417 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1418 if (error)
1419 goto out_qs;
1420
1421 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1422 sdp->sd_jdesc->jd_blocks);
1423 if (error)
1424 goto out_rg_gunlock;
1425
1426 gfs2_free_di(rgd, ip);
1427
1428 gfs2_final_release_pages(ip);
1429
1430 gfs2_trans_end(sdp);
1431
1432out_rg_gunlock:
1433 gfs2_glock_dq_uninit(&gh);
1434out_qs:
1435 gfs2_quota_unhold(ip);
1436 return error;
1437}
1438
1439/**
1440 * gfs2_evict_inode - Remove an inode from cache
1441 * @inode: The inode to evict
1442 *
1443 * There are three cases to consider:
1444 * 1. i_nlink == 0, we are final opener (and must deallocate)
1445 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1446 * 3. i_nlink > 0
1447 *
1448 * If the fs is read only, then we have to treat all cases as per #3
1449 * since we are unable to do any deallocation. The inode will be
1450 * deallocated by the next read/write node to attempt an allocation
1451 * in the same resource group
1452 *
1453 * We have to (at the moment) hold the inodes main lock to cover
1454 * the gap between unlocking the shared lock on the iopen lock and
1455 * taking the exclusive lock. I'd rather do a shared -> exclusive
1456 * conversion on the iopen lock, but we can change that later. This
1457 * is safe, just less efficient.
1458 */
1459
1460static void gfs2_evict_inode(struct inode *inode)
1461{
1462 struct super_block *sb = inode->i_sb;
1463 struct gfs2_sbd *sdp = sb->s_fs_info;
1464 struct gfs2_inode *ip = GFS2_I(inode);
1465 struct gfs2_holder gh;
1466 int error;
1467
1468 if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1469 clear_inode(inode);
1470 return;
1471 }
1472
1473 if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1474 goto out;
1475
1476 /* Must not read inode block until block type has been verified */
1477 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1478 if (unlikely(error)) {
1479 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1480 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1481 goto out;
1482 }
1483
1484 if (!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1485 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1486 if (error)
1487 goto out_truncate;
1488 }
1489
1490 if (test_bit(GIF_INVALID, &ip->i_flags)) {
1491 error = gfs2_inode_refresh(ip);
1492 if (error)
1493 goto out_truncate;
1494 }
1495
1496 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1497 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1498 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1499 error = gfs2_glock_nq(&ip->i_iopen_gh);
1500 if (error)
1501 goto out_truncate;
1502
1503 /* Case 1 starts here */
1504
1505 if (S_ISDIR(inode->i_mode) &&
1506 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1507 error = gfs2_dir_exhash_dealloc(ip);
1508 if (error)
1509 goto out_unlock;
1510 }
1511
1512 if (ip->i_eattr) {
1513 error = gfs2_ea_dealloc(ip);
1514 if (error)
1515 goto out_unlock;
1516 }
1517
1518 if (!gfs2_is_stuffed(ip)) {
1519 error = gfs2_file_dealloc(ip);
1520 if (error)
1521 goto out_unlock;
1522 }
1523
1524 error = gfs2_dinode_dealloc(ip);
1525 goto out_unlock;
1526
1527out_truncate:
1528 gfs2_log_flush(sdp, ip->i_gl);
1529 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1530 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
1531 filemap_fdatawrite(metamapping);
1532 filemap_fdatawait(metamapping);
1533 }
1534 write_inode_now(inode, 1);
1535 gfs2_ail_flush(ip->i_gl, 0);
1536
1537 /* Case 2 starts here */
1538 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1539 if (error)
1540 goto out_unlock;
1541 /* Needs to be done before glock release & also in a transaction */
1542 truncate_inode_pages(&inode->i_data, 0);
1543 gfs2_trans_end(sdp);
1544
1545out_unlock:
1546 /* Error path for case 1 */
1547 if (gfs2_rs_active(ip->i_res))
1548 gfs2_rs_deltree(ip->i_res);
1549
1550 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1551 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1552 gfs2_glock_dq(&ip->i_iopen_gh);
1553 }
1554 gfs2_holder_uninit(&ip->i_iopen_gh);
1555 gfs2_glock_dq_uninit(&gh);
1556 if (error && error != GLR_TRYFAILED && error != -EROFS)
1557 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1558out:
1559 /* Case 3 starts here */
1560 truncate_inode_pages_final(&inode->i_data);
1561 gfs2_rs_delete(ip, NULL);
1562 gfs2_ordered_del_inode(ip);
1563 clear_inode(inode);
1564 gfs2_dir_hash_inval(ip);
1565 ip->i_gl->gl_object = NULL;
1566 flush_delayed_work(&ip->i_gl->gl_work);
1567 gfs2_glock_add_to_lru(ip->i_gl);
1568 gfs2_glock_put(ip->i_gl);
1569 ip->i_gl = NULL;
1570 if (ip->i_iopen_gh.gh_gl) {
1571 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1572 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1573 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1574 }
1575}
1576
1577static struct inode *gfs2_alloc_inode(struct super_block *sb)
1578{
1579 struct gfs2_inode *ip;
1580
1581 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1582 if (ip) {
1583 ip->i_flags = 0;
1584 ip->i_gl = NULL;
1585 ip->i_rgd = NULL;
1586 ip->i_res = NULL;
1587 }
1588 return &ip->i_inode;
1589}
1590
1591static void gfs2_i_callback(struct rcu_head *head)
1592{
1593 struct inode *inode = container_of(head, struct inode, i_rcu);
1594 kmem_cache_free(gfs2_inode_cachep, inode);
1595}
1596
1597static void gfs2_destroy_inode(struct inode *inode)
1598{
1599 call_rcu(&inode->i_rcu, gfs2_i_callback);
1600}
1601
1602const struct super_operations gfs2_super_ops = {
1603 .alloc_inode = gfs2_alloc_inode,
1604 .destroy_inode = gfs2_destroy_inode,
1605 .write_inode = gfs2_write_inode,
1606 .dirty_inode = gfs2_dirty_inode,
1607 .evict_inode = gfs2_evict_inode,
1608 .put_super = gfs2_put_super,
1609 .sync_fs = gfs2_sync_fs,
1610 .freeze_fs = gfs2_freeze,
1611 .unfreeze_fs = gfs2_unfreeze,
1612 .statfs = gfs2_statfs,
1613 .remount_fs = gfs2_remount_fs,
1614 .drop_inode = gfs2_drop_inode,
1615 .show_options = gfs2_show_options,
1616};
1617