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#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
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
5 */
6
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9#include <linux/bio.h>
10#include <linux/sched/signal.h>
11#include <linux/slab.h>
12#include <linux/spinlock.h>
13#include <linux/completion.h>
14#include <linux/buffer_head.h>
15#include <linux/statfs.h>
16#include <linux/seq_file.h>
17#include <linux/mount.h>
18#include <linux/kthread.h>
19#include <linux/delay.h>
20#include <linux/gfs2_ondisk.h>
21#include <linux/crc32.h>
22#include <linux/time.h>
23#include <linux/wait.h>
24#include <linux/writeback.h>
25#include <linux/backing-dev.h>
26#include <linux/kernel.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#include "lops.h"
46
47enum evict_behavior {
48 EVICT_SHOULD_DELETE,
49 EVICT_SHOULD_SKIP_DELETE,
50 EVICT_SHOULD_DEFER_DELETE,
51};
52
53/**
54 * gfs2_jindex_free - Clear all the journal index information
55 * @sdp: The GFS2 superblock
56 *
57 */
58
59void gfs2_jindex_free(struct gfs2_sbd *sdp)
60{
61 struct list_head list;
62 struct gfs2_jdesc *jd;
63
64 spin_lock(&sdp->sd_jindex_spin);
65 list_add(&list, &sdp->sd_jindex_list);
66 list_del_init(&sdp->sd_jindex_list);
67 sdp->sd_journals = 0;
68 spin_unlock(&sdp->sd_jindex_spin);
69
70 down_write(&sdp->sd_log_flush_lock);
71 sdp->sd_jdesc = NULL;
72 up_write(&sdp->sd_log_flush_lock);
73
74 while (!list_empty(&list)) {
75 jd = list_first_entry(&list, struct gfs2_jdesc, jd_list);
76 BUG_ON(jd->jd_log_bio);
77 gfs2_free_journal_extents(jd);
78 list_del(&jd->jd_list);
79 iput(jd->jd_inode);
80 jd->jd_inode = NULL;
81 kfree(jd);
82 }
83}
84
85static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
86{
87 struct gfs2_jdesc *jd;
88
89 list_for_each_entry(jd, head, jd_list) {
90 if (jd->jd_jid == jid)
91 return jd;
92 }
93 return NULL;
94}
95
96struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
97{
98 struct gfs2_jdesc *jd;
99
100 spin_lock(&sdp->sd_jindex_spin);
101 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
102 spin_unlock(&sdp->sd_jindex_spin);
103
104 return jd;
105}
106
107int gfs2_jdesc_check(struct gfs2_jdesc *jd)
108{
109 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
110 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
111 u64 size = i_size_read(jd->jd_inode);
112
113 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
114 return -EIO;
115
116 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
117
118 if (gfs2_write_alloc_required(ip, 0, size)) {
119 gfs2_consist_inode(ip);
120 return -EIO;
121 }
122
123 return 0;
124}
125
126/**
127 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
128 * @sdp: the filesystem
129 *
130 * Returns: errno
131 */
132
133int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
134{
135 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
136 struct gfs2_glock *j_gl = ip->i_gl;
137 struct gfs2_log_header_host head;
138 int error;
139
140 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
141 if (gfs2_withdrawing_or_withdrawn(sdp))
142 return -EIO;
143
144 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
145 if (error) {
146 gfs2_consist(sdp);
147 return error;
148 }
149
150 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
151 gfs2_consist(sdp);
152 return -EIO;
153 }
154
155 /* Initialize some head of the log stuff */
156 sdp->sd_log_sequence = head.lh_sequence + 1;
157 gfs2_log_pointers_init(sdp, head.lh_blkno);
158
159 error = gfs2_quota_init(sdp);
160 if (!error && gfs2_withdrawing_or_withdrawn(sdp))
161 error = -EIO;
162 if (!error)
163 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
164 return error;
165}
166
167void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
168{
169 const struct gfs2_statfs_change *str = buf;
170
171 sc->sc_total = be64_to_cpu(str->sc_total);
172 sc->sc_free = be64_to_cpu(str->sc_free);
173 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
174}
175
176void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
177{
178 struct gfs2_statfs_change *str = buf;
179
180 str->sc_total = cpu_to_be64(sc->sc_total);
181 str->sc_free = cpu_to_be64(sc->sc_free);
182 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
183}
184
185int gfs2_statfs_init(struct gfs2_sbd *sdp)
186{
187 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
188 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
189 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
190 struct buffer_head *m_bh;
191 struct gfs2_holder gh;
192 int error;
193
194 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
195 &gh);
196 if (error)
197 return error;
198
199 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
200 if (error)
201 goto out;
202
203 if (sdp->sd_args.ar_spectator) {
204 spin_lock(&sdp->sd_statfs_spin);
205 gfs2_statfs_change_in(m_sc, m_bh->b_data +
206 sizeof(struct gfs2_dinode));
207 spin_unlock(&sdp->sd_statfs_spin);
208 } else {
209 spin_lock(&sdp->sd_statfs_spin);
210 gfs2_statfs_change_in(m_sc, m_bh->b_data +
211 sizeof(struct gfs2_dinode));
212 gfs2_statfs_change_in(l_sc, sdp->sd_sc_bh->b_data +
213 sizeof(struct gfs2_dinode));
214 spin_unlock(&sdp->sd_statfs_spin);
215
216 }
217
218 brelse(m_bh);
219out:
220 gfs2_glock_dq_uninit(&gh);
221 return 0;
222}
223
224void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
225 s64 dinodes)
226{
227 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
228 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
229 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
230 s64 x, y;
231 int need_sync = 0;
232
233 gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
234
235 spin_lock(&sdp->sd_statfs_spin);
236 l_sc->sc_total += total;
237 l_sc->sc_free += free;
238 l_sc->sc_dinodes += dinodes;
239 gfs2_statfs_change_out(l_sc, sdp->sd_sc_bh->b_data +
240 sizeof(struct gfs2_dinode));
241 if (sdp->sd_args.ar_statfs_percent) {
242 x = 100 * l_sc->sc_free;
243 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
244 if (x >= y || x <= -y)
245 need_sync = 1;
246 }
247 spin_unlock(&sdp->sd_statfs_spin);
248
249 if (need_sync)
250 gfs2_wake_up_statfs(sdp);
251}
252
253void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh)
254{
255 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
256 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
257 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
258 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
259
260 gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
261 gfs2_trans_add_meta(m_ip->i_gl, m_bh);
262
263 spin_lock(&sdp->sd_statfs_spin);
264 m_sc->sc_total += l_sc->sc_total;
265 m_sc->sc_free += l_sc->sc_free;
266 m_sc->sc_dinodes += l_sc->sc_dinodes;
267 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
268 memset(sdp->sd_sc_bh->b_data + sizeof(struct gfs2_dinode),
269 0, sizeof(struct gfs2_statfs_change));
270 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
271 spin_unlock(&sdp->sd_statfs_spin);
272}
273
274int gfs2_statfs_sync(struct super_block *sb, int type)
275{
276 struct gfs2_sbd *sdp = sb->s_fs_info;
277 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
278 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
279 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
280 struct gfs2_holder gh;
281 struct buffer_head *m_bh;
282 int error;
283
284 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
285 &gh);
286 if (error)
287 goto out;
288
289 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
290 if (error)
291 goto out_unlock;
292
293 spin_lock(&sdp->sd_statfs_spin);
294 gfs2_statfs_change_in(m_sc, m_bh->b_data +
295 sizeof(struct gfs2_dinode));
296 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
297 spin_unlock(&sdp->sd_statfs_spin);
298 goto out_bh;
299 }
300 spin_unlock(&sdp->sd_statfs_spin);
301
302 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
303 if (error)
304 goto out_bh;
305
306 update_statfs(sdp, m_bh);
307 sdp->sd_statfs_force_sync = 0;
308
309 gfs2_trans_end(sdp);
310
311out_bh:
312 brelse(m_bh);
313out_unlock:
314 gfs2_glock_dq_uninit(&gh);
315out:
316 return error;
317}
318
319struct lfcc {
320 struct list_head list;
321 struct gfs2_holder gh;
322};
323
324/**
325 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
326 * journals are clean
327 * @sdp: the file system
328 *
329 * Returns: errno
330 */
331
332static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
333{
334 struct gfs2_inode *ip;
335 struct gfs2_jdesc *jd;
336 struct lfcc *lfcc;
337 LIST_HEAD(list);
338 struct gfs2_log_header_host lh;
339 int error, error2;
340
341 /*
342 * Grab all the journal glocks in SH mode. We are *probably* doing
343 * that to prevent recovery.
344 */
345
346 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
347 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
348 if (!lfcc) {
349 error = -ENOMEM;
350 goto out;
351 }
352 ip = GFS2_I(jd->jd_inode);
353 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
354 if (error) {
355 kfree(lfcc);
356 goto out;
357 }
358 list_add(&lfcc->list, &list);
359 }
360
361 gfs2_freeze_unlock(sdp);
362
363 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
364 LM_FLAG_NOEXP | GL_NOPID,
365 &sdp->sd_freeze_gh);
366 if (error)
367 goto relock_shared;
368
369 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
370 error = gfs2_jdesc_check(jd);
371 if (error)
372 break;
373 error = gfs2_find_jhead(jd, &lh, false);
374 if (error)
375 break;
376 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
377 error = -EBUSY;
378 break;
379 }
380 }
381
382 if (!error)
383 goto out; /* success */
384
385 gfs2_freeze_unlock(sdp);
386
387relock_shared:
388 error2 = gfs2_freeze_lock_shared(sdp);
389 gfs2_assert_withdraw(sdp, !error2);
390
391out:
392 while (!list_empty(&list)) {
393 lfcc = list_first_entry(&list, struct lfcc, list);
394 list_del(&lfcc->list);
395 gfs2_glock_dq_uninit(&lfcc->gh);
396 kfree(lfcc);
397 }
398 return error;
399}
400
401void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
402{
403 const struct inode *inode = &ip->i_inode;
404 struct gfs2_dinode *str = buf;
405
406 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
407 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
408 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
409 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
410 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
411 str->di_mode = cpu_to_be32(inode->i_mode);
412 str->di_uid = cpu_to_be32(i_uid_read(inode));
413 str->di_gid = cpu_to_be32(i_gid_read(inode));
414 str->di_nlink = cpu_to_be32(inode->i_nlink);
415 str->di_size = cpu_to_be64(i_size_read(inode));
416 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode));
417 str->di_atime = cpu_to_be64(inode_get_atime_sec(inode));
418 str->di_mtime = cpu_to_be64(inode_get_mtime_sec(inode));
419 str->di_ctime = cpu_to_be64(inode_get_ctime_sec(inode));
420
421 str->di_goal_meta = cpu_to_be64(ip->i_goal);
422 str->di_goal_data = cpu_to_be64(ip->i_goal);
423 str->di_generation = cpu_to_be64(ip->i_generation);
424
425 str->di_flags = cpu_to_be32(ip->i_diskflags);
426 str->di_height = cpu_to_be16(ip->i_height);
427 str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) &&
428 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
429 GFS2_FORMAT_DE : 0);
430 str->di_depth = cpu_to_be16(ip->i_depth);
431 str->di_entries = cpu_to_be32(ip->i_entries);
432
433 str->di_eattr = cpu_to_be64(ip->i_eattr);
434 str->di_atime_nsec = cpu_to_be32(inode_get_atime_nsec(inode));
435 str->di_mtime_nsec = cpu_to_be32(inode_get_mtime_nsec(inode));
436 str->di_ctime_nsec = cpu_to_be32(inode_get_ctime_nsec(inode));
437}
438
439/**
440 * gfs2_write_inode - Make sure the inode is stable on the disk
441 * @inode: The inode
442 * @wbc: The writeback control structure
443 *
444 * Returns: errno
445 */
446
447static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
448{
449 struct gfs2_inode *ip = GFS2_I(inode);
450 struct gfs2_sbd *sdp = GFS2_SB(inode);
451 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
452 struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
453 int ret = 0;
454 bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
455
456 if (flush_all)
457 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
458 GFS2_LOG_HEAD_FLUSH_NORMAL |
459 GFS2_LFC_WRITE_INODE);
460 if (bdi->wb.dirty_exceeded)
461 gfs2_ail1_flush(sdp, wbc);
462 else
463 filemap_fdatawrite(metamapping);
464 if (flush_all)
465 ret = filemap_fdatawait(metamapping);
466 if (ret)
467 mark_inode_dirty_sync(inode);
468 else {
469 spin_lock(&inode->i_lock);
470 if (!(inode->i_flags & I_DIRTY))
471 gfs2_ordered_del_inode(ip);
472 spin_unlock(&inode->i_lock);
473 }
474 return ret;
475}
476
477/**
478 * gfs2_dirty_inode - check for atime updates
479 * @inode: The inode in question
480 * @flags: The type of dirty
481 *
482 * Unfortunately it can be called under any combination of inode
483 * glock and freeze glock, so we have to check carefully.
484 *
485 * At the moment this deals only with atime - it should be possible
486 * to expand that role in future, once a review of the locking has
487 * been carried out.
488 */
489
490static void gfs2_dirty_inode(struct inode *inode, int flags)
491{
492 struct gfs2_inode *ip = GFS2_I(inode);
493 struct gfs2_sbd *sdp = GFS2_SB(inode);
494 struct buffer_head *bh;
495 struct gfs2_holder gh;
496 int need_unlock = 0;
497 int need_endtrans = 0;
498 int ret;
499
500 if (unlikely(!ip->i_gl)) {
501 /* This can only happen during incomplete inode creation. */
502 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
503 return;
504 }
505
506 if (gfs2_withdrawing_or_withdrawn(sdp))
507 return;
508 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
509 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
510 if (ret) {
511 fs_err(sdp, "dirty_inode: glock %d\n", ret);
512 gfs2_dump_glock(NULL, ip->i_gl, true);
513 return;
514 }
515 need_unlock = 1;
516 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
517 return;
518
519 if (current->journal_info == NULL) {
520 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
521 if (ret) {
522 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
523 goto out;
524 }
525 need_endtrans = 1;
526 }
527
528 ret = gfs2_meta_inode_buffer(ip, &bh);
529 if (ret == 0) {
530 gfs2_trans_add_meta(ip->i_gl, bh);
531 gfs2_dinode_out(ip, bh->b_data);
532 brelse(bh);
533 }
534
535 if (need_endtrans)
536 gfs2_trans_end(sdp);
537out:
538 if (need_unlock)
539 gfs2_glock_dq_uninit(&gh);
540}
541
542/**
543 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
544 * @sdp: the filesystem
545 *
546 * Returns: errno
547 */
548
549void gfs2_make_fs_ro(struct gfs2_sbd *sdp)
550{
551 int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
552
553 if (!test_bit(SDF_KILL, &sdp->sd_flags))
554 gfs2_flush_delete_work(sdp);
555
556 gfs2_destroy_threads(sdp);
557
558 if (log_write_allowed) {
559 gfs2_quota_sync(sdp->sd_vfs, 0);
560 gfs2_statfs_sync(sdp->sd_vfs, 0);
561
562 /* We do two log flushes here. The first one commits dirty inodes
563 * and rgrps to the journal, but queues up revokes to the ail list.
564 * The second flush writes out and removes the revokes.
565 *
566 * The first must be done before the FLUSH_SHUTDOWN code
567 * clears the LIVE flag, otherwise it will not be able to start
568 * a transaction to write its revokes, and the error will cause
569 * a withdraw of the file system. */
570 gfs2_log_flush(sdp, NULL, GFS2_LFC_MAKE_FS_RO);
571 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
572 GFS2_LFC_MAKE_FS_RO);
573 wait_event_timeout(sdp->sd_log_waitq,
574 gfs2_log_is_empty(sdp),
575 HZ * 5);
576 gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
577 }
578 gfs2_quota_cleanup(sdp);
579}
580
581/**
582 * gfs2_put_super - Unmount the filesystem
583 * @sb: The VFS superblock
584 *
585 */
586
587static void gfs2_put_super(struct super_block *sb)
588{
589 struct gfs2_sbd *sdp = sb->s_fs_info;
590 struct gfs2_jdesc *jd;
591
592 /* No more recovery requests */
593 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
594 smp_mb();
595
596 /* Wait on outstanding recovery */
597restart:
598 spin_lock(&sdp->sd_jindex_spin);
599 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
600 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
601 continue;
602 spin_unlock(&sdp->sd_jindex_spin);
603 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
604 TASK_UNINTERRUPTIBLE);
605 goto restart;
606 }
607 spin_unlock(&sdp->sd_jindex_spin);
608
609 if (!sb_rdonly(sb))
610 gfs2_make_fs_ro(sdp);
611 else {
612 if (gfs2_withdrawing_or_withdrawn(sdp))
613 gfs2_destroy_threads(sdp);
614
615 gfs2_quota_cleanup(sdp);
616 }
617
618 WARN_ON(gfs2_withdrawing(sdp));
619
620 /* At this point, we're through modifying the disk */
621
622 /* Release stuff */
623
624 gfs2_freeze_unlock(sdp);
625
626 iput(sdp->sd_jindex);
627 iput(sdp->sd_statfs_inode);
628 iput(sdp->sd_rindex);
629 iput(sdp->sd_quota_inode);
630
631 gfs2_glock_put(sdp->sd_rename_gl);
632 gfs2_glock_put(sdp->sd_freeze_gl);
633
634 if (!sdp->sd_args.ar_spectator) {
635 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
636 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
637 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
638 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
639 brelse(sdp->sd_sc_bh);
640 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
641 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
642 free_local_statfs_inodes(sdp);
643 iput(sdp->sd_qc_inode);
644 }
645
646 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
647 gfs2_clear_rgrpd(sdp);
648 gfs2_jindex_free(sdp);
649 /* Take apart glock structures and buffer lists */
650 gfs2_gl_hash_clear(sdp);
651 truncate_inode_pages_final(&sdp->sd_aspace);
652 gfs2_delete_debugfs_file(sdp);
653
654 gfs2_sys_fs_del(sdp);
655 free_sbd(sdp);
656}
657
658/**
659 * gfs2_sync_fs - sync the filesystem
660 * @sb: the superblock
661 * @wait: true to wait for completion
662 *
663 * Flushes the log to disk.
664 */
665
666static int gfs2_sync_fs(struct super_block *sb, int wait)
667{
668 struct gfs2_sbd *sdp = sb->s_fs_info;
669
670 gfs2_quota_sync(sb, -1);
671 if (wait)
672 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
673 GFS2_LFC_SYNC_FS);
674 return sdp->sd_log_error;
675}
676
677static int gfs2_do_thaw(struct gfs2_sbd *sdp)
678{
679 struct super_block *sb = sdp->sd_vfs;
680 int error;
681
682 error = gfs2_freeze_lock_shared(sdp);
683 if (error)
684 goto fail;
685 error = thaw_super(sb, FREEZE_HOLDER_USERSPACE);
686 if (!error)
687 return 0;
688
689fail:
690 fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n", error);
691 gfs2_assert_withdraw(sdp, 0);
692 return error;
693}
694
695void gfs2_freeze_func(struct work_struct *work)
696{
697 struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
698 struct super_block *sb = sdp->sd_vfs;
699 int error;
700
701 mutex_lock(&sdp->sd_freeze_mutex);
702 error = -EBUSY;
703 if (test_bit(SDF_FROZEN, &sdp->sd_flags))
704 goto freeze_failed;
705
706 error = freeze_super(sb, FREEZE_HOLDER_USERSPACE);
707 if (error)
708 goto freeze_failed;
709
710 gfs2_freeze_unlock(sdp);
711 set_bit(SDF_FROZEN, &sdp->sd_flags);
712
713 error = gfs2_do_thaw(sdp);
714 if (error)
715 goto out;
716
717 clear_bit(SDF_FROZEN, &sdp->sd_flags);
718 goto out;
719
720freeze_failed:
721 fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n", error);
722
723out:
724 mutex_unlock(&sdp->sd_freeze_mutex);
725 deactivate_super(sb);
726}
727
728/**
729 * gfs2_freeze_super - prevent further writes to the filesystem
730 * @sb: the VFS structure for the filesystem
731 *
732 */
733
734static int gfs2_freeze_super(struct super_block *sb, enum freeze_holder who)
735{
736 struct gfs2_sbd *sdp = sb->s_fs_info;
737 int error;
738
739 if (!mutex_trylock(&sdp->sd_freeze_mutex))
740 return -EBUSY;
741 if (test_bit(SDF_FROZEN, &sdp->sd_flags)) {
742 mutex_unlock(&sdp->sd_freeze_mutex);
743 return -EBUSY;
744 }
745
746 for (;;) {
747 error = freeze_super(sb, FREEZE_HOLDER_USERSPACE);
748 if (error) {
749 fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
750 error);
751 goto out;
752 }
753
754 error = gfs2_lock_fs_check_clean(sdp);
755 if (!error) {
756 set_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags);
757 set_bit(SDF_FROZEN, &sdp->sd_flags);
758 break;
759 }
760
761 error = gfs2_do_thaw(sdp);
762 if (error)
763 goto out;
764
765 if (error == -EBUSY)
766 fs_err(sdp, "waiting for recovery before freeze\n");
767 else if (error == -EIO) {
768 fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
769 "to recovery error.\n");
770 goto out;
771 } else {
772 fs_err(sdp, "error freezing FS: %d\n", error);
773 }
774 fs_err(sdp, "retrying...\n");
775 msleep(1000);
776 }
777
778out:
779 mutex_unlock(&sdp->sd_freeze_mutex);
780 return error;
781}
782
783static int gfs2_freeze_fs(struct super_block *sb)
784{
785 struct gfs2_sbd *sdp = sb->s_fs_info;
786
787 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
788 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
789 GFS2_LFC_FREEZE_GO_SYNC);
790 if (gfs2_withdrawing_or_withdrawn(sdp))
791 return -EIO;
792 }
793 return 0;
794}
795
796/**
797 * gfs2_thaw_super - reallow writes to the filesystem
798 * @sb: the VFS structure for the filesystem
799 *
800 */
801
802static int gfs2_thaw_super(struct super_block *sb, enum freeze_holder who)
803{
804 struct gfs2_sbd *sdp = sb->s_fs_info;
805 int error;
806
807 if (!mutex_trylock(&sdp->sd_freeze_mutex))
808 return -EBUSY;
809 if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags)) {
810 mutex_unlock(&sdp->sd_freeze_mutex);
811 return -EINVAL;
812 }
813
814 atomic_inc(&sb->s_active);
815 gfs2_freeze_unlock(sdp);
816
817 error = gfs2_do_thaw(sdp);
818
819 if (!error) {
820 clear_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags);
821 clear_bit(SDF_FROZEN, &sdp->sd_flags);
822 }
823 mutex_unlock(&sdp->sd_freeze_mutex);
824 deactivate_super(sb);
825 return error;
826}
827
828void gfs2_thaw_freeze_initiator(struct super_block *sb)
829{
830 struct gfs2_sbd *sdp = sb->s_fs_info;
831
832 mutex_lock(&sdp->sd_freeze_mutex);
833 if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags))
834 goto out;
835
836 gfs2_freeze_unlock(sdp);
837
838out:
839 mutex_unlock(&sdp->sd_freeze_mutex);
840}
841
842/**
843 * statfs_slow_fill - fill in the sg for a given RG
844 * @rgd: the RG
845 * @sc: the sc structure
846 *
847 * Returns: 0 on success, -ESTALE if the LVB is invalid
848 */
849
850static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
851 struct gfs2_statfs_change_host *sc)
852{
853 gfs2_rgrp_verify(rgd);
854 sc->sc_total += rgd->rd_data;
855 sc->sc_free += rgd->rd_free;
856 sc->sc_dinodes += rgd->rd_dinodes;
857 return 0;
858}
859
860/**
861 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
862 * @sdp: the filesystem
863 * @sc: the sc info that will be returned
864 *
865 * Any error (other than a signal) will cause this routine to fall back
866 * to the synchronous version.
867 *
868 * FIXME: This really shouldn't busy wait like this.
869 *
870 * Returns: errno
871 */
872
873static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
874{
875 struct gfs2_rgrpd *rgd_next;
876 struct gfs2_holder *gha, *gh;
877 unsigned int slots = 64;
878 unsigned int x;
879 int done;
880 int error = 0, err;
881
882 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
883 gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
884 if (!gha)
885 return -ENOMEM;
886 for (x = 0; x < slots; x++)
887 gfs2_holder_mark_uninitialized(gha + x);
888
889 rgd_next = gfs2_rgrpd_get_first(sdp);
890
891 for (;;) {
892 done = 1;
893
894 for (x = 0; x < slots; x++) {
895 gh = gha + x;
896
897 if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
898 err = gfs2_glock_wait(gh);
899 if (err) {
900 gfs2_holder_uninit(gh);
901 error = err;
902 } else {
903 if (!error) {
904 struct gfs2_rgrpd *rgd =
905 gfs2_glock2rgrp(gh->gh_gl);
906
907 error = statfs_slow_fill(rgd, sc);
908 }
909 gfs2_glock_dq_uninit(gh);
910 }
911 }
912
913 if (gfs2_holder_initialized(gh))
914 done = 0;
915 else if (rgd_next && !error) {
916 error = gfs2_glock_nq_init(rgd_next->rd_gl,
917 LM_ST_SHARED,
918 GL_ASYNC,
919 gh);
920 rgd_next = gfs2_rgrpd_get_next(rgd_next);
921 done = 0;
922 }
923
924 if (signal_pending(current))
925 error = -ERESTARTSYS;
926 }
927
928 if (done)
929 break;
930
931 yield();
932 }
933
934 kfree(gha);
935 return error;
936}
937
938/**
939 * gfs2_statfs_i - Do a statfs
940 * @sdp: the filesystem
941 * @sc: the sc structure
942 *
943 * Returns: errno
944 */
945
946static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
947{
948 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
949 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
950
951 spin_lock(&sdp->sd_statfs_spin);
952
953 *sc = *m_sc;
954 sc->sc_total += l_sc->sc_total;
955 sc->sc_free += l_sc->sc_free;
956 sc->sc_dinodes += l_sc->sc_dinodes;
957
958 spin_unlock(&sdp->sd_statfs_spin);
959
960 if (sc->sc_free < 0)
961 sc->sc_free = 0;
962 if (sc->sc_free > sc->sc_total)
963 sc->sc_free = sc->sc_total;
964 if (sc->sc_dinodes < 0)
965 sc->sc_dinodes = 0;
966
967 return 0;
968}
969
970/**
971 * gfs2_statfs - Gather and return stats about the filesystem
972 * @dentry: The name of the link
973 * @buf: The buffer
974 *
975 * Returns: 0 on success or error code
976 */
977
978static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
979{
980 struct super_block *sb = dentry->d_sb;
981 struct gfs2_sbd *sdp = sb->s_fs_info;
982 struct gfs2_statfs_change_host sc;
983 int error;
984
985 error = gfs2_rindex_update(sdp);
986 if (error)
987 return error;
988
989 if (gfs2_tune_get(sdp, gt_statfs_slow))
990 error = gfs2_statfs_slow(sdp, &sc);
991 else
992 error = gfs2_statfs_i(sdp, &sc);
993
994 if (error)
995 return error;
996
997 buf->f_type = GFS2_MAGIC;
998 buf->f_bsize = sdp->sd_sb.sb_bsize;
999 buf->f_blocks = sc.sc_total;
1000 buf->f_bfree = sc.sc_free;
1001 buf->f_bavail = sc.sc_free;
1002 buf->f_files = sc.sc_dinodes + sc.sc_free;
1003 buf->f_ffree = sc.sc_free;
1004 buf->f_namelen = GFS2_FNAMESIZE;
1005 buf->f_fsid = uuid_to_fsid(sb->s_uuid.b);
1006
1007 return 0;
1008}
1009
1010/**
1011 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1012 * @inode: The inode to drop
1013 *
1014 * If we've received a callback on an iopen lock then it's because a
1015 * remote node tried to deallocate the inode but failed due to this node
1016 * still having the inode open. Here we mark the link count zero
1017 * since we know that it must have reached zero if the GLF_DEMOTE flag
1018 * is set on the iopen glock. If we didn't do a disk read since the
1019 * remote node removed the final link then we might otherwise miss
1020 * this event. This check ensures that this node will deallocate the
1021 * inode's blocks, or alternatively pass the baton on to another
1022 * node for later deallocation.
1023 */
1024
1025static int gfs2_drop_inode(struct inode *inode)
1026{
1027 struct gfs2_inode *ip = GFS2_I(inode);
1028 struct gfs2_sbd *sdp = GFS2_SB(inode);
1029
1030 if (inode->i_nlink &&
1031 gfs2_holder_initialized(&ip->i_iopen_gh)) {
1032 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1033 if (glock_needs_demote(gl))
1034 clear_nlink(inode);
1035 }
1036
1037 /*
1038 * When under memory pressure when an inode's link count has dropped to
1039 * zero, defer deleting the inode to the delete workqueue. This avoids
1040 * calling into DLM under memory pressure, which can deadlock.
1041 */
1042 if (!inode->i_nlink &&
1043 unlikely(current->flags & PF_MEMALLOC) &&
1044 gfs2_holder_initialized(&ip->i_iopen_gh)) {
1045 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1046
1047 gfs2_glock_hold(gl);
1048 if (!gfs2_queue_verify_delete(gl, true))
1049 gfs2_glock_put_async(gl);
1050 return 0;
1051 }
1052
1053 /*
1054 * No longer cache inodes when trying to evict them all.
1055 */
1056 if (test_bit(SDF_EVICTING, &sdp->sd_flags))
1057 return 1;
1058
1059 return generic_drop_inode(inode);
1060}
1061
1062/**
1063 * gfs2_show_options - Show mount options for /proc/mounts
1064 * @s: seq_file structure
1065 * @root: root of this (sub)tree
1066 *
1067 * Returns: 0 on success or error code
1068 */
1069
1070static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1071{
1072 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1073 struct gfs2_args *args = &sdp->sd_args;
1074 unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum;
1075
1076 spin_lock(&sdp->sd_tune.gt_spin);
1077 logd_secs = sdp->sd_tune.gt_logd_secs;
1078 quota_quantum = sdp->sd_tune.gt_quota_quantum;
1079 statfs_quantum = sdp->sd_tune.gt_statfs_quantum;
1080 statfs_slow = sdp->sd_tune.gt_statfs_slow;
1081 spin_unlock(&sdp->sd_tune.gt_spin);
1082
1083 if (is_subdir(root, sdp->sd_master_dir))
1084 seq_puts(s, ",meta");
1085 if (args->ar_lockproto[0])
1086 seq_show_option(s, "lockproto", args->ar_lockproto);
1087 if (args->ar_locktable[0])
1088 seq_show_option(s, "locktable", args->ar_locktable);
1089 if (args->ar_hostdata[0])
1090 seq_show_option(s, "hostdata", args->ar_hostdata);
1091 if (args->ar_spectator)
1092 seq_puts(s, ",spectator");
1093 if (args->ar_localflocks)
1094 seq_puts(s, ",localflocks");
1095 if (args->ar_debug)
1096 seq_puts(s, ",debug");
1097 if (args->ar_posix_acl)
1098 seq_puts(s, ",acl");
1099 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1100 char *state;
1101 switch (args->ar_quota) {
1102 case GFS2_QUOTA_OFF:
1103 state = "off";
1104 break;
1105 case GFS2_QUOTA_ACCOUNT:
1106 state = "account";
1107 break;
1108 case GFS2_QUOTA_ON:
1109 state = "on";
1110 break;
1111 case GFS2_QUOTA_QUIET:
1112 state = "quiet";
1113 break;
1114 default:
1115 state = "unknown";
1116 break;
1117 }
1118 seq_printf(s, ",quota=%s", state);
1119 }
1120 if (args->ar_suiddir)
1121 seq_puts(s, ",suiddir");
1122 if (args->ar_data != GFS2_DATA_DEFAULT) {
1123 char *state;
1124 switch (args->ar_data) {
1125 case GFS2_DATA_WRITEBACK:
1126 state = "writeback";
1127 break;
1128 case GFS2_DATA_ORDERED:
1129 state = "ordered";
1130 break;
1131 default:
1132 state = "unknown";
1133 break;
1134 }
1135 seq_printf(s, ",data=%s", state);
1136 }
1137 if (args->ar_discard)
1138 seq_puts(s, ",discard");
1139 if (logd_secs != 30)
1140 seq_printf(s, ",commit=%d", logd_secs);
1141 if (statfs_quantum != 30)
1142 seq_printf(s, ",statfs_quantum=%d", statfs_quantum);
1143 else if (statfs_slow)
1144 seq_puts(s, ",statfs_quantum=0");
1145 if (quota_quantum != 60)
1146 seq_printf(s, ",quota_quantum=%d", quota_quantum);
1147 if (args->ar_statfs_percent)
1148 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1149 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1150 const char *state;
1151
1152 switch (args->ar_errors) {
1153 case GFS2_ERRORS_WITHDRAW:
1154 state = "withdraw";
1155 break;
1156 case GFS2_ERRORS_PANIC:
1157 state = "panic";
1158 break;
1159 default:
1160 state = "unknown";
1161 break;
1162 }
1163 seq_printf(s, ",errors=%s", state);
1164 }
1165 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1166 seq_puts(s, ",nobarrier");
1167 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1168 seq_puts(s, ",demote_interface_used");
1169 if (args->ar_rgrplvb)
1170 seq_puts(s, ",rgrplvb");
1171 if (args->ar_loccookie)
1172 seq_puts(s, ",loccookie");
1173 return 0;
1174}
1175
1176static void gfs2_final_release_pages(struct gfs2_inode *ip)
1177{
1178 struct inode *inode = &ip->i_inode;
1179 struct gfs2_glock *gl = ip->i_gl;
1180
1181 if (unlikely(!gl)) {
1182 /* This can only happen during incomplete inode creation. */
1183 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
1184 return;
1185 }
1186
1187 truncate_inode_pages(gfs2_glock2aspace(gl), 0);
1188 truncate_inode_pages(&inode->i_data, 0);
1189
1190 if (atomic_read(&gl->gl_revokes) == 0) {
1191 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1192 clear_bit(GLF_DIRTY, &gl->gl_flags);
1193 }
1194}
1195
1196static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1197{
1198 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1199 struct gfs2_rgrpd *rgd;
1200 struct gfs2_holder gh;
1201 int error;
1202
1203 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1204 gfs2_consist_inode(ip);
1205 return -EIO;
1206 }
1207
1208 gfs2_rindex_update(sdp);
1209
1210 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1211 if (error)
1212 return error;
1213
1214 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1215 if (!rgd) {
1216 gfs2_consist_inode(ip);
1217 error = -EIO;
1218 goto out_qs;
1219 }
1220
1221 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1222 LM_FLAG_NODE_SCOPE, &gh);
1223 if (error)
1224 goto out_qs;
1225
1226 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1227 sdp->sd_jdesc->jd_blocks);
1228 if (error)
1229 goto out_rg_gunlock;
1230
1231 gfs2_free_di(rgd, ip);
1232
1233 gfs2_final_release_pages(ip);
1234
1235 gfs2_trans_end(sdp);
1236
1237out_rg_gunlock:
1238 gfs2_glock_dq_uninit(&gh);
1239out_qs:
1240 gfs2_quota_unhold(ip);
1241 return error;
1242}
1243
1244/**
1245 * gfs2_glock_put_eventually
1246 * @gl: The glock to put
1247 *
1248 * When under memory pressure, trigger a deferred glock put to make sure we
1249 * won't call into DLM and deadlock. Otherwise, put the glock directly.
1250 */
1251
1252static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1253{
1254 if (current->flags & PF_MEMALLOC)
1255 gfs2_glock_put_async(gl);
1256 else
1257 gfs2_glock_put(gl);
1258}
1259
1260static enum evict_behavior gfs2_upgrade_iopen_glock(struct inode *inode)
1261{
1262 struct gfs2_inode *ip = GFS2_I(inode);
1263 struct gfs2_sbd *sdp = GFS2_SB(inode);
1264 struct gfs2_holder *gh = &ip->i_iopen_gh;
1265 int error;
1266
1267 gh->gh_flags |= GL_NOCACHE;
1268 gfs2_glock_dq_wait(gh);
1269
1270 /*
1271 * If there are no other lock holders, we will immediately get
1272 * exclusive access to the iopen glock here.
1273 *
1274 * Otherwise, the other nodes holding the lock will be notified about
1275 * our locking request (see iopen_go_callback()). If they do not have
1276 * the inode open, they are expected to evict the cached inode and
1277 * release the lock, allowing us to proceed.
1278 *
1279 * Otherwise, if they cannot evict the inode, they are expected to poke
1280 * the inode glock (note: not the iopen glock). We will notice that
1281 * and stop waiting for the iopen glock immediately. The other node(s)
1282 * are then expected to take care of deleting the inode when they no
1283 * longer use it.
1284 *
1285 * As a last resort, if another node keeps holding the iopen glock
1286 * without showing any activity on the inode glock, we will eventually
1287 * time out and fail the iopen glock upgrade.
1288 */
1289
1290 gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1291 error = gfs2_glock_nq(gh);
1292 if (error)
1293 return EVICT_SHOULD_SKIP_DELETE;
1294
1295 wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1296 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1297 glock_needs_demote(ip->i_gl),
1298 5 * HZ);
1299 if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1300 gfs2_glock_dq(gh);
1301 if (glock_needs_demote(ip->i_gl))
1302 return EVICT_SHOULD_SKIP_DELETE;
1303 return EVICT_SHOULD_DEFER_DELETE;
1304 }
1305 error = gfs2_glock_holder_ready(gh);
1306 if (error)
1307 return EVICT_SHOULD_SKIP_DELETE;
1308 return EVICT_SHOULD_DELETE;
1309}
1310
1311/**
1312 * evict_should_delete - determine whether the inode is eligible for deletion
1313 * @inode: The inode to evict
1314 * @gh: The glock holder structure
1315 *
1316 * This function determines whether the evicted inode is eligible to be deleted
1317 * and locks the inode glock.
1318 *
1319 * Returns: the fate of the dinode
1320 */
1321static enum evict_behavior evict_should_delete(struct inode *inode,
1322 struct gfs2_holder *gh)
1323{
1324 struct gfs2_inode *ip = GFS2_I(inode);
1325 struct super_block *sb = inode->i_sb;
1326 struct gfs2_sbd *sdp = sb->s_fs_info;
1327 int ret;
1328
1329 if (unlikely(test_bit(GIF_ALLOC_FAILED, &ip->i_flags)))
1330 goto should_delete;
1331
1332 if (test_bit(GIF_DEFER_DELETE, &ip->i_flags))
1333 return EVICT_SHOULD_DEFER_DELETE;
1334
1335 /* Deletes should never happen under memory pressure anymore. */
1336 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1337 return EVICT_SHOULD_DEFER_DELETE;
1338
1339 /* Must not read inode block until block type has been verified */
1340 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1341 if (unlikely(ret)) {
1342 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1343 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1344 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1345 return EVICT_SHOULD_DEFER_DELETE;
1346 }
1347
1348 if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1349 return EVICT_SHOULD_SKIP_DELETE;
1350 ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1351 if (ret)
1352 return EVICT_SHOULD_SKIP_DELETE;
1353
1354 ret = gfs2_instantiate(gh);
1355 if (ret)
1356 return EVICT_SHOULD_SKIP_DELETE;
1357
1358 /*
1359 * The inode may have been recreated in the meantime.
1360 */
1361 if (inode->i_nlink)
1362 return EVICT_SHOULD_SKIP_DELETE;
1363
1364should_delete:
1365 if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1366 test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1367 enum evict_behavior behavior =
1368 gfs2_upgrade_iopen_glock(inode);
1369
1370 if (behavior != EVICT_SHOULD_DELETE) {
1371 gfs2_holder_uninit(&ip->i_iopen_gh);
1372 return behavior;
1373 }
1374 }
1375 return EVICT_SHOULD_DELETE;
1376}
1377
1378/**
1379 * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1380 * @inode: The inode to evict
1381 */
1382static int evict_unlinked_inode(struct inode *inode)
1383{
1384 struct gfs2_inode *ip = GFS2_I(inode);
1385 int ret;
1386
1387 if (S_ISDIR(inode->i_mode) &&
1388 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1389 ret = gfs2_dir_exhash_dealloc(ip);
1390 if (ret)
1391 goto out;
1392 }
1393
1394 if (ip->i_eattr) {
1395 ret = gfs2_ea_dealloc(ip);
1396 if (ret)
1397 goto out;
1398 }
1399
1400 if (!gfs2_is_stuffed(ip)) {
1401 ret = gfs2_file_dealloc(ip);
1402 if (ret)
1403 goto out;
1404 }
1405
1406 /*
1407 * As soon as we clear the bitmap for the dinode, gfs2_create_inode()
1408 * can get called to recreate it, or even gfs2_inode_lookup() if the
1409 * inode was recreated on another node in the meantime.
1410 *
1411 * However, inserting the new inode into the inode hash table will not
1412 * succeed until the old inode is removed, and that only happens after
1413 * ->evict_inode() returns. The new inode is attached to its inode and
1414 * iopen glocks after inserting it into the inode hash table, so at
1415 * that point we can be sure that both glocks are unused.
1416 */
1417
1418 ret = gfs2_dinode_dealloc(ip);
1419 if (!ret && ip->i_gl)
1420 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1421
1422out:
1423 return ret;
1424}
1425
1426/*
1427 * evict_linked_inode - evict an inode whose dinode has not been unlinked
1428 * @inode: The inode to evict
1429 */
1430static int evict_linked_inode(struct inode *inode)
1431{
1432 struct super_block *sb = inode->i_sb;
1433 struct gfs2_sbd *sdp = sb->s_fs_info;
1434 struct gfs2_inode *ip = GFS2_I(inode);
1435 struct address_space *metamapping;
1436 int ret;
1437
1438 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1439 GFS2_LFC_EVICT_INODE);
1440 metamapping = gfs2_glock2aspace(ip->i_gl);
1441 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1442 filemap_fdatawrite(metamapping);
1443 filemap_fdatawait(metamapping);
1444 }
1445 write_inode_now(inode, 1);
1446 gfs2_ail_flush(ip->i_gl, 0);
1447
1448 ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1449 if (ret)
1450 return ret;
1451
1452 /* Needs to be done before glock release & also in a transaction */
1453 truncate_inode_pages(&inode->i_data, 0);
1454 truncate_inode_pages(metamapping, 0);
1455 gfs2_trans_end(sdp);
1456 return 0;
1457}
1458
1459/**
1460 * gfs2_evict_inode - Remove an inode from cache
1461 * @inode: The inode to evict
1462 *
1463 * There are three cases to consider:
1464 * 1. i_nlink == 0, we are final opener (and must deallocate)
1465 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1466 * 3. i_nlink > 0
1467 *
1468 * If the fs is read only, then we have to treat all cases as per #3
1469 * since we are unable to do any deallocation. The inode will be
1470 * deallocated by the next read/write node to attempt an allocation
1471 * in the same resource group
1472 *
1473 * We have to (at the moment) hold the inodes main lock to cover
1474 * the gap between unlocking the shared lock on the iopen lock and
1475 * taking the exclusive lock. I'd rather do a shared -> exclusive
1476 * conversion on the iopen lock, but we can change that later. This
1477 * is safe, just less efficient.
1478 */
1479
1480static void gfs2_evict_inode(struct inode *inode)
1481{
1482 struct super_block *sb = inode->i_sb;
1483 struct gfs2_sbd *sdp = sb->s_fs_info;
1484 struct gfs2_inode *ip = GFS2_I(inode);
1485 struct gfs2_holder gh;
1486 enum evict_behavior behavior;
1487 int ret;
1488
1489 gfs2_holder_mark_uninitialized(&gh);
1490 if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr)
1491 goto out;
1492
1493 /*
1494 * In case of an incomplete mount, gfs2_evict_inode() may be called for
1495 * system files without having an active journal to write to. In that
1496 * case, skip the filesystem evict.
1497 */
1498 if (!sdp->sd_jdesc)
1499 goto out;
1500
1501 behavior = evict_should_delete(inode, &gh);
1502 if (behavior == EVICT_SHOULD_DEFER_DELETE &&
1503 !test_bit(SDF_KILL, &sdp->sd_flags)) {
1504 struct gfs2_glock *io_gl = ip->i_iopen_gh.gh_gl;
1505
1506 if (io_gl) {
1507 gfs2_glock_hold(io_gl);
1508 if (!gfs2_queue_verify_delete(io_gl, true))
1509 gfs2_glock_put(io_gl);
1510 goto out;
1511 }
1512 behavior = EVICT_SHOULD_DELETE;
1513 }
1514 if (behavior == EVICT_SHOULD_DELETE)
1515 ret = evict_unlinked_inode(inode);
1516 else
1517 ret = evict_linked_inode(inode);
1518
1519 if (gfs2_rs_active(&ip->i_res))
1520 gfs2_rs_deltree(&ip->i_res);
1521
1522 if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1523 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1524out:
1525 if (gfs2_holder_initialized(&gh))
1526 gfs2_glock_dq_uninit(&gh);
1527 truncate_inode_pages_final(&inode->i_data);
1528 if (ip->i_qadata)
1529 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1530 gfs2_rs_deltree(&ip->i_res);
1531 gfs2_ordered_del_inode(ip);
1532 clear_inode(inode);
1533 gfs2_dir_hash_inval(ip);
1534 if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1535 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1536
1537 glock_clear_object(gl, ip);
1538 gfs2_glock_hold(gl);
1539 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1540 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1541 gfs2_glock_put_eventually(gl);
1542 }
1543 if (ip->i_gl) {
1544 glock_clear_object(ip->i_gl, ip);
1545 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1546 gfs2_glock_put_eventually(ip->i_gl);
1547 rcu_assign_pointer(ip->i_gl, NULL);
1548 }
1549}
1550
1551static struct inode *gfs2_alloc_inode(struct super_block *sb)
1552{
1553 struct gfs2_inode *ip;
1554
1555 ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL);
1556 if (!ip)
1557 return NULL;
1558 ip->i_no_addr = 0;
1559 ip->i_no_formal_ino = 0;
1560 ip->i_flags = 0;
1561 ip->i_gl = NULL;
1562 gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1563 memset(&ip->i_res, 0, sizeof(ip->i_res));
1564 RB_CLEAR_NODE(&ip->i_res.rs_node);
1565 ip->i_diskflags = 0;
1566 ip->i_rahead = 0;
1567 return &ip->i_inode;
1568}
1569
1570static void gfs2_free_inode(struct inode *inode)
1571{
1572 kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1573}
1574
1575void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1576{
1577 struct local_statfs_inode *lsi, *safe;
1578
1579 /* Run through the statfs inodes list to iput and free memory */
1580 list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1581 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1582 sdp->sd_sc_inode = NULL; /* belongs to this node */
1583 if (lsi->si_sc_inode)
1584 iput(lsi->si_sc_inode);
1585 list_del(&lsi->si_list);
1586 kfree(lsi);
1587 }
1588}
1589
1590struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1591 unsigned int index)
1592{
1593 struct local_statfs_inode *lsi;
1594
1595 /* Return the local (per node) statfs inode in the
1596 * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1597 list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1598 if (lsi->si_jid == index)
1599 return lsi->si_sc_inode;
1600 }
1601 return NULL;
1602}
1603
1604const struct super_operations gfs2_super_ops = {
1605 .alloc_inode = gfs2_alloc_inode,
1606 .free_inode = gfs2_free_inode,
1607 .write_inode = gfs2_write_inode,
1608 .dirty_inode = gfs2_dirty_inode,
1609 .evict_inode = gfs2_evict_inode,
1610 .put_super = gfs2_put_super,
1611 .sync_fs = gfs2_sync_fs,
1612 .freeze_super = gfs2_freeze_super,
1613 .freeze_fs = gfs2_freeze_fs,
1614 .thaw_super = gfs2_thaw_super,
1615 .statfs = gfs2_statfs,
1616 .drop_inode = gfs2_drop_inode,
1617 .show_options = gfs2_show_options,
1618};
1619