GFS2: Make writeback more responsive to system conditions
[deliverable/linux.git] / fs / gfs2 / super.c
1 /*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10 #include <linux/bio.h>
11 #include <linux/sched.h>
12 #include <linux/slab.h>
13 #include <linux/spinlock.h>
14 #include <linux/completion.h>
15 #include <linux/buffer_head.h>
16 #include <linux/statfs.h>
17 #include <linux/seq_file.h>
18 #include <linux/mount.h>
19 #include <linux/kthread.h>
20 #include <linux/delay.h>
21 #include <linux/gfs2_ondisk.h>
22 #include <linux/crc32.h>
23 #include <linux/time.h>
24 #include <linux/wait.h>
25 #include <linux/writeback.h>
26 #include <linux/backing-dev.h>
27
28 #include "gfs2.h"
29 #include "incore.h"
30 #include "bmap.h"
31 #include "dir.h"
32 #include "glock.h"
33 #include "glops.h"
34 #include "inode.h"
35 #include "log.h"
36 #include "meta_io.h"
37 #include "quota.h"
38 #include "recovery.h"
39 #include "rgrp.h"
40 #include "super.h"
41 #include "trans.h"
42 #include "util.h"
43 #include "sys.h"
44 #include "xattr.h"
45
46 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
47
48 enum {
49 Opt_lockproto,
50 Opt_locktable,
51 Opt_hostdata,
52 Opt_spectator,
53 Opt_ignore_local_fs,
54 Opt_localflocks,
55 Opt_localcaching,
56 Opt_debug,
57 Opt_nodebug,
58 Opt_upgrade,
59 Opt_acl,
60 Opt_noacl,
61 Opt_quota_off,
62 Opt_quota_account,
63 Opt_quota_on,
64 Opt_quota,
65 Opt_noquota,
66 Opt_suiddir,
67 Opt_nosuiddir,
68 Opt_data_writeback,
69 Opt_data_ordered,
70 Opt_meta,
71 Opt_discard,
72 Opt_nodiscard,
73 Opt_commit,
74 Opt_err_withdraw,
75 Opt_err_panic,
76 Opt_statfs_quantum,
77 Opt_statfs_percent,
78 Opt_quota_quantum,
79 Opt_barrier,
80 Opt_nobarrier,
81 Opt_error,
82 };
83
84 static const match_table_t tokens = {
85 {Opt_lockproto, "lockproto=%s"},
86 {Opt_locktable, "locktable=%s"},
87 {Opt_hostdata, "hostdata=%s"},
88 {Opt_spectator, "spectator"},
89 {Opt_spectator, "norecovery"},
90 {Opt_ignore_local_fs, "ignore_local_fs"},
91 {Opt_localflocks, "localflocks"},
92 {Opt_localcaching, "localcaching"},
93 {Opt_debug, "debug"},
94 {Opt_nodebug, "nodebug"},
95 {Opt_upgrade, "upgrade"},
96 {Opt_acl, "acl"},
97 {Opt_noacl, "noacl"},
98 {Opt_quota_off, "quota=off"},
99 {Opt_quota_account, "quota=account"},
100 {Opt_quota_on, "quota=on"},
101 {Opt_quota, "quota"},
102 {Opt_noquota, "noquota"},
103 {Opt_suiddir, "suiddir"},
104 {Opt_nosuiddir, "nosuiddir"},
105 {Opt_data_writeback, "data=writeback"},
106 {Opt_data_ordered, "data=ordered"},
107 {Opt_meta, "meta"},
108 {Opt_discard, "discard"},
109 {Opt_nodiscard, "nodiscard"},
110 {Opt_commit, "commit=%d"},
111 {Opt_err_withdraw, "errors=withdraw"},
112 {Opt_err_panic, "errors=panic"},
113 {Opt_statfs_quantum, "statfs_quantum=%d"},
114 {Opt_statfs_percent, "statfs_percent=%d"},
115 {Opt_quota_quantum, "quota_quantum=%d"},
116 {Opt_barrier, "barrier"},
117 {Opt_nobarrier, "nobarrier"},
118 {Opt_error, NULL}
119 };
120
121 /**
122 * gfs2_mount_args - Parse mount options
123 * @args: The structure into which the parsed options will be written
124 * @options: The options to parse
125 *
126 * Return: errno
127 */
128
129 int gfs2_mount_args(struct gfs2_args *args, char *options)
130 {
131 char *o;
132 int token;
133 substring_t tmp[MAX_OPT_ARGS];
134 int rv;
135
136 /* Split the options into tokens with the "," character and
137 process them */
138
139 while (1) {
140 o = strsep(&options, ",");
141 if (o == NULL)
142 break;
143 if (*o == '\0')
144 continue;
145
146 token = match_token(o, tokens, tmp);
147 switch (token) {
148 case Opt_lockproto:
149 match_strlcpy(args->ar_lockproto, &tmp[0],
150 GFS2_LOCKNAME_LEN);
151 break;
152 case Opt_locktable:
153 match_strlcpy(args->ar_locktable, &tmp[0],
154 GFS2_LOCKNAME_LEN);
155 break;
156 case Opt_hostdata:
157 match_strlcpy(args->ar_hostdata, &tmp[0],
158 GFS2_LOCKNAME_LEN);
159 break;
160 case Opt_spectator:
161 args->ar_spectator = 1;
162 break;
163 case Opt_ignore_local_fs:
164 /* Retained for backwards compat only */
165 break;
166 case Opt_localflocks:
167 args->ar_localflocks = 1;
168 break;
169 case Opt_localcaching:
170 /* Retained for backwards compat only */
171 break;
172 case Opt_debug:
173 if (args->ar_errors == GFS2_ERRORS_PANIC) {
174 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
175 "are mutually exclusive.\n");
176 return -EINVAL;
177 }
178 args->ar_debug = 1;
179 break;
180 case Opt_nodebug:
181 args->ar_debug = 0;
182 break;
183 case Opt_upgrade:
184 /* Retained for backwards compat only */
185 break;
186 case Opt_acl:
187 args->ar_posix_acl = 1;
188 break;
189 case Opt_noacl:
190 args->ar_posix_acl = 0;
191 break;
192 case Opt_quota_off:
193 case Opt_noquota:
194 args->ar_quota = GFS2_QUOTA_OFF;
195 break;
196 case Opt_quota_account:
197 args->ar_quota = GFS2_QUOTA_ACCOUNT;
198 break;
199 case Opt_quota_on:
200 case Opt_quota:
201 args->ar_quota = GFS2_QUOTA_ON;
202 break;
203 case Opt_suiddir:
204 args->ar_suiddir = 1;
205 break;
206 case Opt_nosuiddir:
207 args->ar_suiddir = 0;
208 break;
209 case Opt_data_writeback:
210 args->ar_data = GFS2_DATA_WRITEBACK;
211 break;
212 case Opt_data_ordered:
213 args->ar_data = GFS2_DATA_ORDERED;
214 break;
215 case Opt_meta:
216 args->ar_meta = 1;
217 break;
218 case Opt_discard:
219 args->ar_discard = 1;
220 break;
221 case Opt_nodiscard:
222 args->ar_discard = 0;
223 break;
224 case Opt_commit:
225 rv = match_int(&tmp[0], &args->ar_commit);
226 if (rv || args->ar_commit <= 0) {
227 printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n");
228 return rv ? rv : -EINVAL;
229 }
230 break;
231 case Opt_statfs_quantum:
232 rv = match_int(&tmp[0], &args->ar_statfs_quantum);
233 if (rv || args->ar_statfs_quantum < 0) {
234 printk(KERN_WARNING "GFS2: statfs_quantum mount option requires a non-negative numeric argument\n");
235 return rv ? rv : -EINVAL;
236 }
237 break;
238 case Opt_quota_quantum:
239 rv = match_int(&tmp[0], &args->ar_quota_quantum);
240 if (rv || args->ar_quota_quantum <= 0) {
241 printk(KERN_WARNING "GFS2: quota_quantum mount option requires a positive numeric argument\n");
242 return rv ? rv : -EINVAL;
243 }
244 break;
245 case Opt_statfs_percent:
246 rv = match_int(&tmp[0], &args->ar_statfs_percent);
247 if (rv || args->ar_statfs_percent < 0 ||
248 args->ar_statfs_percent > 100) {
249 printk(KERN_WARNING "statfs_percent mount option requires a numeric argument between 0 and 100\n");
250 return rv ? rv : -EINVAL;
251 }
252 break;
253 case Opt_err_withdraw:
254 args->ar_errors = GFS2_ERRORS_WITHDRAW;
255 break;
256 case Opt_err_panic:
257 if (args->ar_debug) {
258 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
259 "are mutually exclusive.\n");
260 return -EINVAL;
261 }
262 args->ar_errors = GFS2_ERRORS_PANIC;
263 break;
264 case Opt_barrier:
265 args->ar_nobarrier = 0;
266 break;
267 case Opt_nobarrier:
268 args->ar_nobarrier = 1;
269 break;
270 case Opt_error:
271 default:
272 printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o);
273 return -EINVAL;
274 }
275 }
276
277 return 0;
278 }
279
280 /**
281 * gfs2_jindex_free - Clear all the journal index information
282 * @sdp: The GFS2 superblock
283 *
284 */
285
286 void gfs2_jindex_free(struct gfs2_sbd *sdp)
287 {
288 struct list_head list, *head;
289 struct gfs2_jdesc *jd;
290 struct gfs2_journal_extent *jext;
291
292 spin_lock(&sdp->sd_jindex_spin);
293 list_add(&list, &sdp->sd_jindex_list);
294 list_del_init(&sdp->sd_jindex_list);
295 sdp->sd_journals = 0;
296 spin_unlock(&sdp->sd_jindex_spin);
297
298 while (!list_empty(&list)) {
299 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
300 head = &jd->extent_list;
301 while (!list_empty(head)) {
302 jext = list_entry(head->next,
303 struct gfs2_journal_extent,
304 extent_list);
305 list_del(&jext->extent_list);
306 kfree(jext);
307 }
308 list_del(&jd->jd_list);
309 iput(jd->jd_inode);
310 kfree(jd);
311 }
312 }
313
314 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
315 {
316 struct gfs2_jdesc *jd;
317 int found = 0;
318
319 list_for_each_entry(jd, head, jd_list) {
320 if (jd->jd_jid == jid) {
321 found = 1;
322 break;
323 }
324 }
325
326 if (!found)
327 jd = NULL;
328
329 return jd;
330 }
331
332 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
333 {
334 struct gfs2_jdesc *jd;
335
336 spin_lock(&sdp->sd_jindex_spin);
337 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
338 spin_unlock(&sdp->sd_jindex_spin);
339
340 return jd;
341 }
342
343 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
344 {
345 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
346 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
347 u64 size = i_size_read(jd->jd_inode);
348
349 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, 1 << 30))
350 return -EIO;
351
352 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
353
354 if (gfs2_write_alloc_required(ip, 0, size)) {
355 gfs2_consist_inode(ip);
356 return -EIO;
357 }
358
359 return 0;
360 }
361
362 /**
363 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
364 * @sdp: the filesystem
365 *
366 * Returns: errno
367 */
368
369 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
370 {
371 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
372 struct gfs2_glock *j_gl = ip->i_gl;
373 struct gfs2_holder t_gh;
374 struct gfs2_log_header_host head;
375 int error;
376
377 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
378 if (error)
379 return error;
380
381 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
382
383 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
384 if (error)
385 goto fail;
386
387 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
388 gfs2_consist(sdp);
389 error = -EIO;
390 goto fail;
391 }
392
393 /* Initialize some head of the log stuff */
394 sdp->sd_log_sequence = head.lh_sequence + 1;
395 gfs2_log_pointers_init(sdp, head.lh_blkno);
396
397 error = gfs2_quota_init(sdp);
398 if (error)
399 goto fail;
400
401 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
402
403 gfs2_glock_dq_uninit(&t_gh);
404
405 return 0;
406
407 fail:
408 t_gh.gh_flags |= GL_NOCACHE;
409 gfs2_glock_dq_uninit(&t_gh);
410
411 return error;
412 }
413
414 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
415 {
416 const struct gfs2_statfs_change *str = buf;
417
418 sc->sc_total = be64_to_cpu(str->sc_total);
419 sc->sc_free = be64_to_cpu(str->sc_free);
420 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
421 }
422
423 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
424 {
425 struct gfs2_statfs_change *str = buf;
426
427 str->sc_total = cpu_to_be64(sc->sc_total);
428 str->sc_free = cpu_to_be64(sc->sc_free);
429 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
430 }
431
432 int gfs2_statfs_init(struct gfs2_sbd *sdp)
433 {
434 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
435 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
436 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
437 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
438 struct buffer_head *m_bh, *l_bh;
439 struct gfs2_holder gh;
440 int error;
441
442 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
443 &gh);
444 if (error)
445 return error;
446
447 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
448 if (error)
449 goto out;
450
451 if (sdp->sd_args.ar_spectator) {
452 spin_lock(&sdp->sd_statfs_spin);
453 gfs2_statfs_change_in(m_sc, m_bh->b_data +
454 sizeof(struct gfs2_dinode));
455 spin_unlock(&sdp->sd_statfs_spin);
456 } else {
457 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
458 if (error)
459 goto out_m_bh;
460
461 spin_lock(&sdp->sd_statfs_spin);
462 gfs2_statfs_change_in(m_sc, m_bh->b_data +
463 sizeof(struct gfs2_dinode));
464 gfs2_statfs_change_in(l_sc, l_bh->b_data +
465 sizeof(struct gfs2_dinode));
466 spin_unlock(&sdp->sd_statfs_spin);
467
468 brelse(l_bh);
469 }
470
471 out_m_bh:
472 brelse(m_bh);
473 out:
474 gfs2_glock_dq_uninit(&gh);
475 return 0;
476 }
477
478 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
479 s64 dinodes)
480 {
481 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
482 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
483 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
484 struct buffer_head *l_bh;
485 s64 x, y;
486 int need_sync = 0;
487 int error;
488
489 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
490 if (error)
491 return;
492
493 gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
494
495 spin_lock(&sdp->sd_statfs_spin);
496 l_sc->sc_total += total;
497 l_sc->sc_free += free;
498 l_sc->sc_dinodes += dinodes;
499 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
500 if (sdp->sd_args.ar_statfs_percent) {
501 x = 100 * l_sc->sc_free;
502 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
503 if (x >= y || x <= -y)
504 need_sync = 1;
505 }
506 spin_unlock(&sdp->sd_statfs_spin);
507
508 brelse(l_bh);
509 if (need_sync)
510 gfs2_wake_up_statfs(sdp);
511 }
512
513 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
514 struct buffer_head *l_bh)
515 {
516 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
517 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
518 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
519 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
520
521 gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
522
523 spin_lock(&sdp->sd_statfs_spin);
524 m_sc->sc_total += l_sc->sc_total;
525 m_sc->sc_free += l_sc->sc_free;
526 m_sc->sc_dinodes += l_sc->sc_dinodes;
527 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
528 memset(l_bh->b_data + sizeof(struct gfs2_dinode),
529 0, sizeof(struct gfs2_statfs_change));
530 spin_unlock(&sdp->sd_statfs_spin);
531
532 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
533 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
534 }
535
536 int gfs2_statfs_sync(struct super_block *sb, int type)
537 {
538 struct gfs2_sbd *sdp = sb->s_fs_info;
539 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
540 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
541 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
542 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
543 struct gfs2_holder gh;
544 struct buffer_head *m_bh, *l_bh;
545 int error;
546
547 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
548 &gh);
549 if (error)
550 return error;
551
552 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
553 if (error)
554 goto out;
555
556 spin_lock(&sdp->sd_statfs_spin);
557 gfs2_statfs_change_in(m_sc, m_bh->b_data +
558 sizeof(struct gfs2_dinode));
559 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
560 spin_unlock(&sdp->sd_statfs_spin);
561 goto out_bh;
562 }
563 spin_unlock(&sdp->sd_statfs_spin);
564
565 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
566 if (error)
567 goto out_bh;
568
569 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
570 if (error)
571 goto out_bh2;
572
573 update_statfs(sdp, m_bh, l_bh);
574 sdp->sd_statfs_force_sync = 0;
575
576 gfs2_trans_end(sdp);
577
578 out_bh2:
579 brelse(l_bh);
580 out_bh:
581 brelse(m_bh);
582 out:
583 gfs2_glock_dq_uninit(&gh);
584 return error;
585 }
586
587 struct lfcc {
588 struct list_head list;
589 struct gfs2_holder gh;
590 };
591
592 /**
593 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
594 * journals are clean
595 * @sdp: the file system
596 * @state: the state to put the transaction lock into
597 * @t_gh: the hold on the transaction lock
598 *
599 * Returns: errno
600 */
601
602 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
603 struct gfs2_holder *t_gh)
604 {
605 struct gfs2_inode *ip;
606 struct gfs2_jdesc *jd;
607 struct lfcc *lfcc;
608 LIST_HEAD(list);
609 struct gfs2_log_header_host lh;
610 int error;
611
612 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
613 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
614 if (!lfcc) {
615 error = -ENOMEM;
616 goto out;
617 }
618 ip = GFS2_I(jd->jd_inode);
619 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
620 if (error) {
621 kfree(lfcc);
622 goto out;
623 }
624 list_add(&lfcc->list, &list);
625 }
626
627 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
628 GL_NOCACHE, t_gh);
629
630 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
631 error = gfs2_jdesc_check(jd);
632 if (error)
633 break;
634 error = gfs2_find_jhead(jd, &lh);
635 if (error)
636 break;
637 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
638 error = -EBUSY;
639 break;
640 }
641 }
642
643 if (error)
644 gfs2_glock_dq_uninit(t_gh);
645
646 out:
647 while (!list_empty(&list)) {
648 lfcc = list_entry(list.next, struct lfcc, list);
649 list_del(&lfcc->list);
650 gfs2_glock_dq_uninit(&lfcc->gh);
651 kfree(lfcc);
652 }
653 return error;
654 }
655
656 /**
657 * gfs2_freeze_fs - freezes the file system
658 * @sdp: the file system
659 *
660 * This function flushes data and meta data for all machines by
661 * acquiring the transaction log exclusively. All journals are
662 * ensured to be in a clean state as well.
663 *
664 * Returns: errno
665 */
666
667 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
668 {
669 int error = 0;
670
671 mutex_lock(&sdp->sd_freeze_lock);
672
673 if (!sdp->sd_freeze_count++) {
674 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
675 if (error)
676 sdp->sd_freeze_count--;
677 }
678
679 mutex_unlock(&sdp->sd_freeze_lock);
680
681 return error;
682 }
683
684 /**
685 * gfs2_unfreeze_fs - unfreezes the file system
686 * @sdp: the file system
687 *
688 * This function allows the file system to proceed by unlocking
689 * the exclusively held transaction lock. Other GFS2 nodes are
690 * now free to acquire the lock shared and go on with their lives.
691 *
692 */
693
694 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
695 {
696 mutex_lock(&sdp->sd_freeze_lock);
697
698 if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
699 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
700
701 mutex_unlock(&sdp->sd_freeze_lock);
702 }
703
704
705 /**
706 * gfs2_write_inode - Make sure the inode is stable on the disk
707 * @inode: The inode
708 * @wbc: The writeback control structure
709 *
710 * Returns: errno
711 */
712
713 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
714 {
715 struct gfs2_inode *ip = GFS2_I(inode);
716 struct gfs2_sbd *sdp = GFS2_SB(inode);
717 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
718 struct backing_dev_info *bdi = metamapping->backing_dev_info;
719 struct gfs2_holder gh;
720 struct buffer_head *bh;
721 struct timespec atime;
722 struct gfs2_dinode *di;
723 int ret = -EAGAIN;
724
725 /* Skip timestamp update, if this is from a memalloc */
726 if (current->flags & PF_MEMALLOC)
727 goto do_flush;
728 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
729 if (ret)
730 goto do_flush;
731 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
732 if (ret)
733 goto do_unlock;
734 ret = gfs2_meta_inode_buffer(ip, &bh);
735 if (ret == 0) {
736 di = (struct gfs2_dinode *)bh->b_data;
737 atime.tv_sec = be64_to_cpu(di->di_atime);
738 atime.tv_nsec = be32_to_cpu(di->di_atime_nsec);
739 if (timespec_compare(&inode->i_atime, &atime) > 0) {
740 gfs2_trans_add_bh(ip->i_gl, bh, 1);
741 gfs2_dinode_out(ip, bh->b_data);
742 }
743 brelse(bh);
744 }
745 gfs2_trans_end(sdp);
746 do_unlock:
747 gfs2_glock_dq_uninit(&gh);
748 do_flush:
749 if (wbc->sync_mode == WB_SYNC_ALL)
750 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
751 filemap_fdatawrite(metamapping);
752 if (bdi->dirty_exceeded)
753 gfs2_ail1_flush(sdp, wbc);
754 if (!ret && (wbc->sync_mode == WB_SYNC_ALL))
755 ret = filemap_fdatawait(metamapping);
756 if (ret)
757 mark_inode_dirty_sync(inode);
758 return ret;
759 }
760
761 /**
762 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
763 * @sdp: the filesystem
764 *
765 * Returns: errno
766 */
767
768 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
769 {
770 struct gfs2_holder t_gh;
771 int error;
772
773 flush_workqueue(gfs2_delete_workqueue);
774 gfs2_quota_sync(sdp->sd_vfs, 0, 1);
775 gfs2_statfs_sync(sdp->sd_vfs, 0);
776
777 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
778 &t_gh);
779 if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
780 return error;
781
782 gfs2_meta_syncfs(sdp);
783 gfs2_log_shutdown(sdp);
784
785 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
786
787 if (t_gh.gh_gl)
788 gfs2_glock_dq_uninit(&t_gh);
789
790 gfs2_quota_cleanup(sdp);
791
792 return error;
793 }
794
795 static int gfs2_umount_recovery_wait(void *word)
796 {
797 schedule();
798 return 0;
799 }
800
801 /**
802 * gfs2_put_super - Unmount the filesystem
803 * @sb: The VFS superblock
804 *
805 */
806
807 static void gfs2_put_super(struct super_block *sb)
808 {
809 struct gfs2_sbd *sdp = sb->s_fs_info;
810 int error;
811 struct gfs2_jdesc *jd;
812
813 /* Unfreeze the filesystem, if we need to */
814
815 mutex_lock(&sdp->sd_freeze_lock);
816 if (sdp->sd_freeze_count)
817 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
818 mutex_unlock(&sdp->sd_freeze_lock);
819
820 /* No more recovery requests */
821 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
822 smp_mb();
823
824 /* Wait on outstanding recovery */
825 restart:
826 spin_lock(&sdp->sd_jindex_spin);
827 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
828 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
829 continue;
830 spin_unlock(&sdp->sd_jindex_spin);
831 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
832 gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
833 goto restart;
834 }
835 spin_unlock(&sdp->sd_jindex_spin);
836
837 kthread_stop(sdp->sd_quotad_process);
838 kthread_stop(sdp->sd_logd_process);
839
840 if (!(sb->s_flags & MS_RDONLY)) {
841 error = gfs2_make_fs_ro(sdp);
842 if (error)
843 gfs2_io_error(sdp);
844 }
845 /* At this point, we're through modifying the disk */
846
847 /* Release stuff */
848
849 iput(sdp->sd_jindex);
850 iput(sdp->sd_statfs_inode);
851 iput(sdp->sd_rindex);
852 iput(sdp->sd_quota_inode);
853
854 gfs2_glock_put(sdp->sd_rename_gl);
855 gfs2_glock_put(sdp->sd_trans_gl);
856
857 if (!sdp->sd_args.ar_spectator) {
858 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
859 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
860 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
861 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
862 iput(sdp->sd_sc_inode);
863 iput(sdp->sd_qc_inode);
864 }
865
866 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
867 gfs2_clear_rgrpd(sdp);
868 gfs2_jindex_free(sdp);
869 /* Take apart glock structures and buffer lists */
870 gfs2_gl_hash_clear(sdp);
871 /* Unmount the locking protocol */
872 gfs2_lm_unmount(sdp);
873
874 /* At this point, we're through participating in the lockspace */
875 gfs2_sys_fs_del(sdp);
876 }
877
878 /**
879 * gfs2_sync_fs - sync the filesystem
880 * @sb: the superblock
881 *
882 * Flushes the log to disk.
883 */
884
885 static int gfs2_sync_fs(struct super_block *sb, int wait)
886 {
887 struct gfs2_sbd *sdp = sb->s_fs_info;
888 if (wait && sdp)
889 gfs2_log_flush(sdp, NULL);
890 return 0;
891 }
892
893 /**
894 * gfs2_freeze - prevent further writes to the filesystem
895 * @sb: the VFS structure for the filesystem
896 *
897 */
898
899 static int gfs2_freeze(struct super_block *sb)
900 {
901 struct gfs2_sbd *sdp = sb->s_fs_info;
902 int error;
903
904 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
905 return -EINVAL;
906
907 for (;;) {
908 error = gfs2_freeze_fs(sdp);
909 if (!error)
910 break;
911
912 switch (error) {
913 case -EBUSY:
914 fs_err(sdp, "waiting for recovery before freeze\n");
915 break;
916
917 default:
918 fs_err(sdp, "error freezing FS: %d\n", error);
919 break;
920 }
921
922 fs_err(sdp, "retrying...\n");
923 msleep(1000);
924 }
925 return 0;
926 }
927
928 /**
929 * gfs2_unfreeze - reallow writes to the filesystem
930 * @sb: the VFS structure for the filesystem
931 *
932 */
933
934 static int gfs2_unfreeze(struct super_block *sb)
935 {
936 gfs2_unfreeze_fs(sb->s_fs_info);
937 return 0;
938 }
939
940 /**
941 * statfs_fill - fill in the sg for a given RG
942 * @rgd: the RG
943 * @sc: the sc structure
944 *
945 * Returns: 0 on success, -ESTALE if the LVB is invalid
946 */
947
948 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
949 struct gfs2_statfs_change_host *sc)
950 {
951 gfs2_rgrp_verify(rgd);
952 sc->sc_total += rgd->rd_data;
953 sc->sc_free += rgd->rd_free;
954 sc->sc_dinodes += rgd->rd_dinodes;
955 return 0;
956 }
957
958 /**
959 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
960 * @sdp: the filesystem
961 * @sc: the sc info that will be returned
962 *
963 * Any error (other than a signal) will cause this routine to fall back
964 * to the synchronous version.
965 *
966 * FIXME: This really shouldn't busy wait like this.
967 *
968 * Returns: errno
969 */
970
971 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
972 {
973 struct gfs2_holder ri_gh;
974 struct gfs2_rgrpd *rgd_next;
975 struct gfs2_holder *gha, *gh;
976 unsigned int slots = 64;
977 unsigned int x;
978 int done;
979 int error = 0, err;
980
981 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
982 gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
983 if (!gha)
984 return -ENOMEM;
985
986 error = gfs2_rindex_hold(sdp, &ri_gh);
987 if (error)
988 goto out;
989
990 rgd_next = gfs2_rgrpd_get_first(sdp);
991
992 for (;;) {
993 done = 1;
994
995 for (x = 0; x < slots; x++) {
996 gh = gha + x;
997
998 if (gh->gh_gl && gfs2_glock_poll(gh)) {
999 err = gfs2_glock_wait(gh);
1000 if (err) {
1001 gfs2_holder_uninit(gh);
1002 error = err;
1003 } else {
1004 if (!error)
1005 error = statfs_slow_fill(
1006 gh->gh_gl->gl_object, sc);
1007 gfs2_glock_dq_uninit(gh);
1008 }
1009 }
1010
1011 if (gh->gh_gl)
1012 done = 0;
1013 else if (rgd_next && !error) {
1014 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1015 LM_ST_SHARED,
1016 GL_ASYNC,
1017 gh);
1018 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1019 done = 0;
1020 }
1021
1022 if (signal_pending(current))
1023 error = -ERESTARTSYS;
1024 }
1025
1026 if (done)
1027 break;
1028
1029 yield();
1030 }
1031
1032 gfs2_glock_dq_uninit(&ri_gh);
1033
1034 out:
1035 kfree(gha);
1036 return error;
1037 }
1038
1039 /**
1040 * gfs2_statfs_i - Do a statfs
1041 * @sdp: the filesystem
1042 * @sg: the sg structure
1043 *
1044 * Returns: errno
1045 */
1046
1047 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1048 {
1049 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1050 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1051
1052 spin_lock(&sdp->sd_statfs_spin);
1053
1054 *sc = *m_sc;
1055 sc->sc_total += l_sc->sc_total;
1056 sc->sc_free += l_sc->sc_free;
1057 sc->sc_dinodes += l_sc->sc_dinodes;
1058
1059 spin_unlock(&sdp->sd_statfs_spin);
1060
1061 if (sc->sc_free < 0)
1062 sc->sc_free = 0;
1063 if (sc->sc_free > sc->sc_total)
1064 sc->sc_free = sc->sc_total;
1065 if (sc->sc_dinodes < 0)
1066 sc->sc_dinodes = 0;
1067
1068 return 0;
1069 }
1070
1071 /**
1072 * gfs2_statfs - Gather and return stats about the filesystem
1073 * @sb: The superblock
1074 * @statfsbuf: The buffer
1075 *
1076 * Returns: 0 on success or error code
1077 */
1078
1079 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1080 {
1081 struct super_block *sb = dentry->d_inode->i_sb;
1082 struct gfs2_sbd *sdp = sb->s_fs_info;
1083 struct gfs2_statfs_change_host sc;
1084 int error;
1085
1086 if (gfs2_tune_get(sdp, gt_statfs_slow))
1087 error = gfs2_statfs_slow(sdp, &sc);
1088 else
1089 error = gfs2_statfs_i(sdp, &sc);
1090
1091 if (error)
1092 return error;
1093
1094 buf->f_type = GFS2_MAGIC;
1095 buf->f_bsize = sdp->sd_sb.sb_bsize;
1096 buf->f_blocks = sc.sc_total;
1097 buf->f_bfree = sc.sc_free;
1098 buf->f_bavail = sc.sc_free;
1099 buf->f_files = sc.sc_dinodes + sc.sc_free;
1100 buf->f_ffree = sc.sc_free;
1101 buf->f_namelen = GFS2_FNAMESIZE;
1102
1103 return 0;
1104 }
1105
1106 /**
1107 * gfs2_remount_fs - called when the FS is remounted
1108 * @sb: the filesystem
1109 * @flags: the remount flags
1110 * @data: extra data passed in (not used right now)
1111 *
1112 * Returns: errno
1113 */
1114
1115 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1116 {
1117 struct gfs2_sbd *sdp = sb->s_fs_info;
1118 struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1119 struct gfs2_tune *gt = &sdp->sd_tune;
1120 int error;
1121
1122 spin_lock(&gt->gt_spin);
1123 args.ar_commit = gt->gt_logd_secs;
1124 args.ar_quota_quantum = gt->gt_quota_quantum;
1125 if (gt->gt_statfs_slow)
1126 args.ar_statfs_quantum = 0;
1127 else
1128 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1129 spin_unlock(&gt->gt_spin);
1130 error = gfs2_mount_args(&args, data);
1131 if (error)
1132 return error;
1133
1134 /* Not allowed to change locking details */
1135 if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1136 strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1137 strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1138 return -EINVAL;
1139
1140 /* Some flags must not be changed */
1141 if (args_neq(&args, &sdp->sd_args, spectator) ||
1142 args_neq(&args, &sdp->sd_args, localflocks) ||
1143 args_neq(&args, &sdp->sd_args, meta))
1144 return -EINVAL;
1145
1146 if (sdp->sd_args.ar_spectator)
1147 *flags |= MS_RDONLY;
1148
1149 if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1150 if (*flags & MS_RDONLY)
1151 error = gfs2_make_fs_ro(sdp);
1152 else
1153 error = gfs2_make_fs_rw(sdp);
1154 if (error)
1155 return error;
1156 }
1157
1158 sdp->sd_args = args;
1159 if (sdp->sd_args.ar_posix_acl)
1160 sb->s_flags |= MS_POSIXACL;
1161 else
1162 sb->s_flags &= ~MS_POSIXACL;
1163 if (sdp->sd_args.ar_nobarrier)
1164 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1165 else
1166 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1167 spin_lock(&gt->gt_spin);
1168 gt->gt_logd_secs = args.ar_commit;
1169 gt->gt_quota_quantum = args.ar_quota_quantum;
1170 if (args.ar_statfs_quantum) {
1171 gt->gt_statfs_slow = 0;
1172 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1173 }
1174 else {
1175 gt->gt_statfs_slow = 1;
1176 gt->gt_statfs_quantum = 30;
1177 }
1178 spin_unlock(&gt->gt_spin);
1179
1180 gfs2_online_uevent(sdp);
1181 return 0;
1182 }
1183
1184 /**
1185 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1186 * @inode: The inode to drop
1187 *
1188 * If we've received a callback on an iopen lock then its because a
1189 * remote node tried to deallocate the inode but failed due to this node
1190 * still having the inode open. Here we mark the link count zero
1191 * since we know that it must have reached zero if the GLF_DEMOTE flag
1192 * is set on the iopen glock. If we didn't do a disk read since the
1193 * remote node removed the final link then we might otherwise miss
1194 * this event. This check ensures that this node will deallocate the
1195 * inode's blocks, or alternatively pass the baton on to another
1196 * node for later deallocation.
1197 */
1198
1199 static int gfs2_drop_inode(struct inode *inode)
1200 {
1201 struct gfs2_inode *ip = GFS2_I(inode);
1202
1203 if (inode->i_nlink) {
1204 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1205 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1206 clear_nlink(inode);
1207 }
1208 return generic_drop_inode(inode);
1209 }
1210
1211 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1212 {
1213 do {
1214 if (d1 == d2)
1215 return 1;
1216 d1 = d1->d_parent;
1217 } while (!IS_ROOT(d1));
1218 return 0;
1219 }
1220
1221 /**
1222 * gfs2_show_options - Show mount options for /proc/mounts
1223 * @s: seq_file structure
1224 * @mnt: vfsmount
1225 *
1226 * Returns: 0 on success or error code
1227 */
1228
1229 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1230 {
1231 struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
1232 struct gfs2_args *args = &sdp->sd_args;
1233 int val;
1234
1235 if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir))
1236 seq_printf(s, ",meta");
1237 if (args->ar_lockproto[0])
1238 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1239 if (args->ar_locktable[0])
1240 seq_printf(s, ",locktable=%s", args->ar_locktable);
1241 if (args->ar_hostdata[0])
1242 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1243 if (args->ar_spectator)
1244 seq_printf(s, ",spectator");
1245 if (args->ar_localflocks)
1246 seq_printf(s, ",localflocks");
1247 if (args->ar_debug)
1248 seq_printf(s, ",debug");
1249 if (args->ar_posix_acl)
1250 seq_printf(s, ",acl");
1251 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1252 char *state;
1253 switch (args->ar_quota) {
1254 case GFS2_QUOTA_OFF:
1255 state = "off";
1256 break;
1257 case GFS2_QUOTA_ACCOUNT:
1258 state = "account";
1259 break;
1260 case GFS2_QUOTA_ON:
1261 state = "on";
1262 break;
1263 default:
1264 state = "unknown";
1265 break;
1266 }
1267 seq_printf(s, ",quota=%s", state);
1268 }
1269 if (args->ar_suiddir)
1270 seq_printf(s, ",suiddir");
1271 if (args->ar_data != GFS2_DATA_DEFAULT) {
1272 char *state;
1273 switch (args->ar_data) {
1274 case GFS2_DATA_WRITEBACK:
1275 state = "writeback";
1276 break;
1277 case GFS2_DATA_ORDERED:
1278 state = "ordered";
1279 break;
1280 default:
1281 state = "unknown";
1282 break;
1283 }
1284 seq_printf(s, ",data=%s", state);
1285 }
1286 if (args->ar_discard)
1287 seq_printf(s, ",discard");
1288 val = sdp->sd_tune.gt_logd_secs;
1289 if (val != 30)
1290 seq_printf(s, ",commit=%d", val);
1291 val = sdp->sd_tune.gt_statfs_quantum;
1292 if (val != 30)
1293 seq_printf(s, ",statfs_quantum=%d", val);
1294 val = sdp->sd_tune.gt_quota_quantum;
1295 if (val != 60)
1296 seq_printf(s, ",quota_quantum=%d", val);
1297 if (args->ar_statfs_percent)
1298 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1299 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1300 const char *state;
1301
1302 switch (args->ar_errors) {
1303 case GFS2_ERRORS_WITHDRAW:
1304 state = "withdraw";
1305 break;
1306 case GFS2_ERRORS_PANIC:
1307 state = "panic";
1308 break;
1309 default:
1310 state = "unknown";
1311 break;
1312 }
1313 seq_printf(s, ",errors=%s", state);
1314 }
1315 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1316 seq_printf(s, ",nobarrier");
1317 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1318 seq_printf(s, ",demote_interface_used");
1319 return 0;
1320 }
1321
1322 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1323 {
1324 struct inode *inode = &ip->i_inode;
1325 struct gfs2_glock *gl = ip->i_gl;
1326
1327 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1328 truncate_inode_pages(&inode->i_data, 0);
1329
1330 if (atomic_read(&gl->gl_revokes) == 0) {
1331 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1332 clear_bit(GLF_DIRTY, &gl->gl_flags);
1333 }
1334 }
1335
1336 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1337 {
1338 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1339 struct gfs2_alloc *al;
1340 struct gfs2_rgrpd *rgd;
1341 int error;
1342
1343 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1344 if (gfs2_consist_inode(ip))
1345 gfs2_dinode_print(ip);
1346 return -EIO;
1347 }
1348
1349 al = gfs2_alloc_get(ip);
1350 if (!al)
1351 return -ENOMEM;
1352
1353 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1354 if (error)
1355 goto out;
1356
1357 error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
1358 if (error)
1359 goto out_qs;
1360
1361 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr);
1362 if (!rgd) {
1363 gfs2_consist_inode(ip);
1364 error = -EIO;
1365 goto out_rindex_relse;
1366 }
1367
1368 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0,
1369 &al->al_rgd_gh);
1370 if (error)
1371 goto out_rindex_relse;
1372
1373 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1374 sdp->sd_jdesc->jd_blocks);
1375 if (error)
1376 goto out_rg_gunlock;
1377
1378 gfs2_free_di(rgd, ip);
1379
1380 gfs2_final_release_pages(ip);
1381
1382 gfs2_trans_end(sdp);
1383
1384 out_rg_gunlock:
1385 gfs2_glock_dq_uninit(&al->al_rgd_gh);
1386 out_rindex_relse:
1387 gfs2_glock_dq_uninit(&al->al_ri_gh);
1388 out_qs:
1389 gfs2_quota_unhold(ip);
1390 out:
1391 gfs2_alloc_put(ip);
1392 return error;
1393 }
1394
1395 /*
1396 * We have to (at the moment) hold the inodes main lock to cover
1397 * the gap between unlocking the shared lock on the iopen lock and
1398 * taking the exclusive lock. I'd rather do a shared -> exclusive
1399 * conversion on the iopen lock, but we can change that later. This
1400 * is safe, just less efficient.
1401 */
1402
1403 static void gfs2_evict_inode(struct inode *inode)
1404 {
1405 struct super_block *sb = inode->i_sb;
1406 struct gfs2_sbd *sdp = sb->s_fs_info;
1407 struct gfs2_inode *ip = GFS2_I(inode);
1408 struct gfs2_holder gh;
1409 int error;
1410
1411 if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1412 goto out;
1413
1414 /* Must not read inode block until block type has been verified */
1415 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1416 if (unlikely(error)) {
1417 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1418 goto out;
1419 }
1420
1421 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1422 if (error)
1423 goto out_truncate;
1424
1425 if (test_bit(GIF_INVALID, &ip->i_flags)) {
1426 error = gfs2_inode_refresh(ip);
1427 if (error)
1428 goto out_truncate;
1429 }
1430
1431 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1432 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1433 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1434 error = gfs2_glock_nq(&ip->i_iopen_gh);
1435 if (error)
1436 goto out_truncate;
1437
1438 if (S_ISDIR(inode->i_mode) &&
1439 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1440 error = gfs2_dir_exhash_dealloc(ip);
1441 if (error)
1442 goto out_unlock;
1443 }
1444
1445 if (ip->i_eattr) {
1446 error = gfs2_ea_dealloc(ip);
1447 if (error)
1448 goto out_unlock;
1449 }
1450
1451 if (!gfs2_is_stuffed(ip)) {
1452 error = gfs2_file_dealloc(ip);
1453 if (error)
1454 goto out_unlock;
1455 }
1456
1457 error = gfs2_dinode_dealloc(ip);
1458 goto out_unlock;
1459
1460 out_truncate:
1461 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1462 if (error)
1463 goto out_unlock;
1464 gfs2_final_release_pages(ip);
1465 gfs2_trans_end(sdp);
1466
1467 out_unlock:
1468 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
1469 gfs2_glock_dq(&ip->i_iopen_gh);
1470 gfs2_holder_uninit(&ip->i_iopen_gh);
1471 gfs2_glock_dq_uninit(&gh);
1472 if (error && error != GLR_TRYFAILED && error != -EROFS)
1473 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1474 out:
1475 truncate_inode_pages(&inode->i_data, 0);
1476 end_writeback(inode);
1477
1478 ip->i_gl->gl_object = NULL;
1479 gfs2_glock_add_to_lru(ip->i_gl);
1480 gfs2_glock_put(ip->i_gl);
1481 ip->i_gl = NULL;
1482 if (ip->i_iopen_gh.gh_gl) {
1483 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1484 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1485 }
1486 }
1487
1488 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1489 {
1490 struct gfs2_inode *ip;
1491
1492 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1493 if (ip) {
1494 ip->i_flags = 0;
1495 ip->i_gl = NULL;
1496 }
1497 return &ip->i_inode;
1498 }
1499
1500 static void gfs2_i_callback(struct rcu_head *head)
1501 {
1502 struct inode *inode = container_of(head, struct inode, i_rcu);
1503 INIT_LIST_HEAD(&inode->i_dentry);
1504 kmem_cache_free(gfs2_inode_cachep, inode);
1505 }
1506
1507 static void gfs2_destroy_inode(struct inode *inode)
1508 {
1509 call_rcu(&inode->i_rcu, gfs2_i_callback);
1510 }
1511
1512 const struct super_operations gfs2_super_ops = {
1513 .alloc_inode = gfs2_alloc_inode,
1514 .destroy_inode = gfs2_destroy_inode,
1515 .write_inode = gfs2_write_inode,
1516 .evict_inode = gfs2_evict_inode,
1517 .put_super = gfs2_put_super,
1518 .sync_fs = gfs2_sync_fs,
1519 .freeze_fs = gfs2_freeze,
1520 .unfreeze_fs = gfs2_unfreeze,
1521 .statfs = gfs2_statfs,
1522 .remount_fs = gfs2_remount_fs,
1523 .drop_inode = gfs2_drop_inode,
1524 .show_options = gfs2_show_options,
1525 };
1526
This page took 0.067223 seconds and 5 git commands to generate.