xref: /linux/fs/gfs2/super.c (revision bd628c1bed7902ec1f24ba0fe70758949146abbe)
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/signal.h>
14 #include <linux/slab.h>
15 #include <linux/spinlock.h>
16 #include <linux/completion.h>
17 #include <linux/buffer_head.h>
18 #include <linux/statfs.h>
19 #include <linux/seq_file.h>
20 #include <linux/mount.h>
21 #include <linux/kthread.h>
22 #include <linux/delay.h>
23 #include <linux/gfs2_ondisk.h>
24 #include <linux/crc32.h>
25 #include <linux/time.h>
26 #include <linux/wait.h>
27 #include <linux/writeback.h>
28 #include <linux/backing-dev.h>
29 #include <linux/kernel.h>
30 
31 #include "gfs2.h"
32 #include "incore.h"
33 #include "bmap.h"
34 #include "dir.h"
35 #include "glock.h"
36 #include "glops.h"
37 #include "inode.h"
38 #include "log.h"
39 #include "meta_io.h"
40 #include "quota.h"
41 #include "recovery.h"
42 #include "rgrp.h"
43 #include "super.h"
44 #include "trans.h"
45 #include "util.h"
46 #include "sys.h"
47 #include "xattr.h"
48 #include "lops.h"
49 
50 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
51 
52 enum {
53 	Opt_lockproto,
54 	Opt_locktable,
55 	Opt_hostdata,
56 	Opt_spectator,
57 	Opt_ignore_local_fs,
58 	Opt_localflocks,
59 	Opt_localcaching,
60 	Opt_debug,
61 	Opt_nodebug,
62 	Opt_upgrade,
63 	Opt_acl,
64 	Opt_noacl,
65 	Opt_quota_off,
66 	Opt_quota_account,
67 	Opt_quota_on,
68 	Opt_quota,
69 	Opt_noquota,
70 	Opt_suiddir,
71 	Opt_nosuiddir,
72 	Opt_data_writeback,
73 	Opt_data_ordered,
74 	Opt_meta,
75 	Opt_discard,
76 	Opt_nodiscard,
77 	Opt_commit,
78 	Opt_err_withdraw,
79 	Opt_err_panic,
80 	Opt_statfs_quantum,
81 	Opt_statfs_percent,
82 	Opt_quota_quantum,
83 	Opt_barrier,
84 	Opt_nobarrier,
85 	Opt_rgrplvb,
86 	Opt_norgrplvb,
87 	Opt_loccookie,
88 	Opt_noloccookie,
89 	Opt_error,
90 };
91 
92 static const match_table_t tokens = {
93 	{Opt_lockproto, "lockproto=%s"},
94 	{Opt_locktable, "locktable=%s"},
95 	{Opt_hostdata, "hostdata=%s"},
96 	{Opt_spectator, "spectator"},
97 	{Opt_spectator, "norecovery"},
98 	{Opt_ignore_local_fs, "ignore_local_fs"},
99 	{Opt_localflocks, "localflocks"},
100 	{Opt_localcaching, "localcaching"},
101 	{Opt_debug, "debug"},
102 	{Opt_nodebug, "nodebug"},
103 	{Opt_upgrade, "upgrade"},
104 	{Opt_acl, "acl"},
105 	{Opt_noacl, "noacl"},
106 	{Opt_quota_off, "quota=off"},
107 	{Opt_quota_account, "quota=account"},
108 	{Opt_quota_on, "quota=on"},
109 	{Opt_quota, "quota"},
110 	{Opt_noquota, "noquota"},
111 	{Opt_suiddir, "suiddir"},
112 	{Opt_nosuiddir, "nosuiddir"},
113 	{Opt_data_writeback, "data=writeback"},
114 	{Opt_data_ordered, "data=ordered"},
115 	{Opt_meta, "meta"},
116 	{Opt_discard, "discard"},
117 	{Opt_nodiscard, "nodiscard"},
118 	{Opt_commit, "commit=%d"},
119 	{Opt_err_withdraw, "errors=withdraw"},
120 	{Opt_err_panic, "errors=panic"},
121 	{Opt_statfs_quantum, "statfs_quantum=%d"},
122 	{Opt_statfs_percent, "statfs_percent=%d"},
123 	{Opt_quota_quantum, "quota_quantum=%d"},
124 	{Opt_barrier, "barrier"},
125 	{Opt_nobarrier, "nobarrier"},
126 	{Opt_rgrplvb, "rgrplvb"},
127 	{Opt_norgrplvb, "norgrplvb"},
128 	{Opt_loccookie, "loccookie"},
129 	{Opt_noloccookie, "noloccookie"},
130 	{Opt_error, NULL}
131 };
132 
133 /**
134  * gfs2_mount_args - Parse mount options
135  * @args: The structure into which the parsed options will be written
136  * @options: The options to parse
137  *
138  * Return: errno
139  */
140 
141 int gfs2_mount_args(struct gfs2_args *args, char *options)
142 {
143 	char *o;
144 	int token;
145 	substring_t tmp[MAX_OPT_ARGS];
146 	int rv;
147 
148 	/* Split the options into tokens with the "," character and
149 	   process them */
150 
151 	while (1) {
152 		o = strsep(&options, ",");
153 		if (o == NULL)
154 			break;
155 		if (*o == '\0')
156 			continue;
157 
158 		token = match_token(o, tokens, tmp);
159 		switch (token) {
160 		case Opt_lockproto:
161 			match_strlcpy(args->ar_lockproto, &tmp[0],
162 				      GFS2_LOCKNAME_LEN);
163 			break;
164 		case Opt_locktable:
165 			match_strlcpy(args->ar_locktable, &tmp[0],
166 				      GFS2_LOCKNAME_LEN);
167 			break;
168 		case Opt_hostdata:
169 			match_strlcpy(args->ar_hostdata, &tmp[0],
170 				      GFS2_LOCKNAME_LEN);
171 			break;
172 		case Opt_spectator:
173 			args->ar_spectator = 1;
174 			break;
175 		case Opt_ignore_local_fs:
176 			/* Retained for backwards compat only */
177 			break;
178 		case Opt_localflocks:
179 			args->ar_localflocks = 1;
180 			break;
181 		case Opt_localcaching:
182 			/* Retained for backwards compat only */
183 			break;
184 		case Opt_debug:
185 			if (args->ar_errors == GFS2_ERRORS_PANIC) {
186 				pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
187 				return -EINVAL;
188 			}
189 			args->ar_debug = 1;
190 			break;
191 		case Opt_nodebug:
192 			args->ar_debug = 0;
193 			break;
194 		case Opt_upgrade:
195 			/* Retained for backwards compat only */
196 			break;
197 		case Opt_acl:
198 			args->ar_posix_acl = 1;
199 			break;
200 		case Opt_noacl:
201 			args->ar_posix_acl = 0;
202 			break;
203 		case Opt_quota_off:
204 		case Opt_noquota:
205 			args->ar_quota = GFS2_QUOTA_OFF;
206 			break;
207 		case Opt_quota_account:
208 			args->ar_quota = GFS2_QUOTA_ACCOUNT;
209 			break;
210 		case Opt_quota_on:
211 		case Opt_quota:
212 			args->ar_quota = GFS2_QUOTA_ON;
213 			break;
214 		case Opt_suiddir:
215 			args->ar_suiddir = 1;
216 			break;
217 		case Opt_nosuiddir:
218 			args->ar_suiddir = 0;
219 			break;
220 		case Opt_data_writeback:
221 			args->ar_data = GFS2_DATA_WRITEBACK;
222 			break;
223 		case Opt_data_ordered:
224 			args->ar_data = GFS2_DATA_ORDERED;
225 			break;
226 		case Opt_meta:
227 			args->ar_meta = 1;
228 			break;
229 		case Opt_discard:
230 			args->ar_discard = 1;
231 			break;
232 		case Opt_nodiscard:
233 			args->ar_discard = 0;
234 			break;
235 		case Opt_commit:
236 			rv = match_int(&tmp[0], &args->ar_commit);
237 			if (rv || args->ar_commit <= 0) {
238 				pr_warn("commit mount option requires a positive numeric argument\n");
239 				return rv ? rv : -EINVAL;
240 			}
241 			break;
242 		case Opt_statfs_quantum:
243 			rv = match_int(&tmp[0], &args->ar_statfs_quantum);
244 			if (rv || args->ar_statfs_quantum < 0) {
245 				pr_warn("statfs_quantum mount option requires a non-negative numeric argument\n");
246 				return rv ? rv : -EINVAL;
247 			}
248 			break;
249 		case Opt_quota_quantum:
250 			rv = match_int(&tmp[0], &args->ar_quota_quantum);
251 			if (rv || args->ar_quota_quantum <= 0) {
252 				pr_warn("quota_quantum mount option requires a positive numeric argument\n");
253 				return rv ? rv : -EINVAL;
254 			}
255 			break;
256 		case Opt_statfs_percent:
257 			rv = match_int(&tmp[0], &args->ar_statfs_percent);
258 			if (rv || args->ar_statfs_percent < 0 ||
259 			    args->ar_statfs_percent > 100) {
260 				pr_warn("statfs_percent mount option requires a numeric argument between 0 and 100\n");
261 				return rv ? rv : -EINVAL;
262 			}
263 			break;
264 		case Opt_err_withdraw:
265 			args->ar_errors = GFS2_ERRORS_WITHDRAW;
266 			break;
267 		case Opt_err_panic:
268 			if (args->ar_debug) {
269 				pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
270 				return -EINVAL;
271 			}
272 			args->ar_errors = GFS2_ERRORS_PANIC;
273 			break;
274 		case Opt_barrier:
275 			args->ar_nobarrier = 0;
276 			break;
277 		case Opt_nobarrier:
278 			args->ar_nobarrier = 1;
279 			break;
280 		case Opt_rgrplvb:
281 			args->ar_rgrplvb = 1;
282 			break;
283 		case Opt_norgrplvb:
284 			args->ar_rgrplvb = 0;
285 			break;
286 		case Opt_loccookie:
287 			args->ar_loccookie = 1;
288 			break;
289 		case Opt_noloccookie:
290 			args->ar_loccookie = 0;
291 			break;
292 		case Opt_error:
293 		default:
294 			pr_warn("invalid mount option: %s\n", o);
295 			return -EINVAL;
296 		}
297 	}
298 
299 	return 0;
300 }
301 
302 /**
303  * gfs2_jindex_free - Clear all the journal index information
304  * @sdp: The GFS2 superblock
305  *
306  */
307 
308 void gfs2_jindex_free(struct gfs2_sbd *sdp)
309 {
310 	struct list_head list;
311 	struct gfs2_jdesc *jd;
312 
313 	spin_lock(&sdp->sd_jindex_spin);
314 	list_add(&list, &sdp->sd_jindex_list);
315 	list_del_init(&sdp->sd_jindex_list);
316 	sdp->sd_journals = 0;
317 	spin_unlock(&sdp->sd_jindex_spin);
318 
319 	while (!list_empty(&list)) {
320 		jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
321 		gfs2_free_journal_extents(jd);
322 		list_del(&jd->jd_list);
323 		iput(jd->jd_inode);
324 		kfree(jd);
325 	}
326 }
327 
328 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
329 {
330 	struct gfs2_jdesc *jd;
331 	int found = 0;
332 
333 	list_for_each_entry(jd, head, jd_list) {
334 		if (jd->jd_jid == jid) {
335 			found = 1;
336 			break;
337 		}
338 	}
339 
340 	if (!found)
341 		jd = NULL;
342 
343 	return jd;
344 }
345 
346 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
347 {
348 	struct gfs2_jdesc *jd;
349 
350 	spin_lock(&sdp->sd_jindex_spin);
351 	jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
352 	spin_unlock(&sdp->sd_jindex_spin);
353 
354 	return jd;
355 }
356 
357 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
358 {
359 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
360 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
361 	u64 size = i_size_read(jd->jd_inode);
362 
363 	if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
364 		return -EIO;
365 
366 	jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
367 
368 	if (gfs2_write_alloc_required(ip, 0, size)) {
369 		gfs2_consist_inode(ip);
370 		return -EIO;
371 	}
372 
373 	return 0;
374 }
375 
376 static int init_threads(struct gfs2_sbd *sdp)
377 {
378 	struct task_struct *p;
379 	int error = 0;
380 
381 	p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
382 	if (IS_ERR(p)) {
383 		error = PTR_ERR(p);
384 		fs_err(sdp, "can't start logd thread: %d\n", error);
385 		return error;
386 	}
387 	sdp->sd_logd_process = p;
388 
389 	p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
390 	if (IS_ERR(p)) {
391 		error = PTR_ERR(p);
392 		fs_err(sdp, "can't start quotad thread: %d\n", error);
393 		goto fail;
394 	}
395 	sdp->sd_quotad_process = p;
396 	return 0;
397 
398 fail:
399 	kthread_stop(sdp->sd_logd_process);
400 	return error;
401 }
402 
403 /**
404  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
405  * @sdp: the filesystem
406  *
407  * Returns: errno
408  */
409 
410 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
411 {
412 	struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
413 	struct gfs2_glock *j_gl = ip->i_gl;
414 	struct gfs2_holder freeze_gh;
415 	struct gfs2_log_header_host head;
416 	int error;
417 
418 	error = init_threads(sdp);
419 	if (error)
420 		return error;
421 
422 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
423 				   &freeze_gh);
424 	if (error)
425 		goto fail_threads;
426 
427 	j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
428 
429 	error = gfs2_find_jhead(sdp->sd_jdesc, &head);
430 	if (error)
431 		goto fail;
432 
433 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
434 		gfs2_consist(sdp);
435 		error = -EIO;
436 		goto fail;
437 	}
438 
439 	/*  Initialize some head of the log stuff  */
440 	sdp->sd_log_sequence = head.lh_sequence + 1;
441 	gfs2_log_pointers_init(sdp, head.lh_blkno);
442 
443 	error = gfs2_quota_init(sdp);
444 	if (error)
445 		goto fail;
446 
447 	set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
448 
449 	gfs2_glock_dq_uninit(&freeze_gh);
450 
451 	return 0;
452 
453 fail:
454 	freeze_gh.gh_flags |= GL_NOCACHE;
455 	gfs2_glock_dq_uninit(&freeze_gh);
456 fail_threads:
457 	kthread_stop(sdp->sd_quotad_process);
458 	kthread_stop(sdp->sd_logd_process);
459 	return error;
460 }
461 
462 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
463 {
464 	const struct gfs2_statfs_change *str = buf;
465 
466 	sc->sc_total = be64_to_cpu(str->sc_total);
467 	sc->sc_free = be64_to_cpu(str->sc_free);
468 	sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
469 }
470 
471 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
472 {
473 	struct gfs2_statfs_change *str = buf;
474 
475 	str->sc_total = cpu_to_be64(sc->sc_total);
476 	str->sc_free = cpu_to_be64(sc->sc_free);
477 	str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
478 }
479 
480 int gfs2_statfs_init(struct gfs2_sbd *sdp)
481 {
482 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
483 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
484 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
485 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
486 	struct buffer_head *m_bh, *l_bh;
487 	struct gfs2_holder gh;
488 	int error;
489 
490 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
491 				   &gh);
492 	if (error)
493 		return error;
494 
495 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
496 	if (error)
497 		goto out;
498 
499 	if (sdp->sd_args.ar_spectator) {
500 		spin_lock(&sdp->sd_statfs_spin);
501 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
502 				      sizeof(struct gfs2_dinode));
503 		spin_unlock(&sdp->sd_statfs_spin);
504 	} else {
505 		error = gfs2_meta_inode_buffer(l_ip, &l_bh);
506 		if (error)
507 			goto out_m_bh;
508 
509 		spin_lock(&sdp->sd_statfs_spin);
510 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
511 				      sizeof(struct gfs2_dinode));
512 		gfs2_statfs_change_in(l_sc, l_bh->b_data +
513 				      sizeof(struct gfs2_dinode));
514 		spin_unlock(&sdp->sd_statfs_spin);
515 
516 		brelse(l_bh);
517 	}
518 
519 out_m_bh:
520 	brelse(m_bh);
521 out:
522 	gfs2_glock_dq_uninit(&gh);
523 	return 0;
524 }
525 
526 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
527 			s64 dinodes)
528 {
529 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
530 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
531 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
532 	struct buffer_head *l_bh;
533 	s64 x, y;
534 	int need_sync = 0;
535 	int error;
536 
537 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
538 	if (error)
539 		return;
540 
541 	gfs2_trans_add_meta(l_ip->i_gl, l_bh);
542 
543 	spin_lock(&sdp->sd_statfs_spin);
544 	l_sc->sc_total += total;
545 	l_sc->sc_free += free;
546 	l_sc->sc_dinodes += dinodes;
547 	gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
548 	if (sdp->sd_args.ar_statfs_percent) {
549 		x = 100 * l_sc->sc_free;
550 		y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
551 		if (x >= y || x <= -y)
552 			need_sync = 1;
553 	}
554 	spin_unlock(&sdp->sd_statfs_spin);
555 
556 	brelse(l_bh);
557 	if (need_sync)
558 		gfs2_wake_up_statfs(sdp);
559 }
560 
561 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
562 		   struct buffer_head *l_bh)
563 {
564 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
565 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
566 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
567 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
568 
569 	gfs2_trans_add_meta(l_ip->i_gl, l_bh);
570 	gfs2_trans_add_meta(m_ip->i_gl, m_bh);
571 
572 	spin_lock(&sdp->sd_statfs_spin);
573 	m_sc->sc_total += l_sc->sc_total;
574 	m_sc->sc_free += l_sc->sc_free;
575 	m_sc->sc_dinodes += l_sc->sc_dinodes;
576 	memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
577 	memset(l_bh->b_data + sizeof(struct gfs2_dinode),
578 	       0, sizeof(struct gfs2_statfs_change));
579 	gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
580 	spin_unlock(&sdp->sd_statfs_spin);
581 }
582 
583 int gfs2_statfs_sync(struct super_block *sb, int type)
584 {
585 	struct gfs2_sbd *sdp = sb->s_fs_info;
586 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
587 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
588 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
589 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
590 	struct gfs2_holder gh;
591 	struct buffer_head *m_bh, *l_bh;
592 	int error;
593 
594 	sb_start_write(sb);
595 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
596 				   &gh);
597 	if (error)
598 		goto out;
599 
600 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
601 	if (error)
602 		goto out_unlock;
603 
604 	spin_lock(&sdp->sd_statfs_spin);
605 	gfs2_statfs_change_in(m_sc, m_bh->b_data +
606 			      sizeof(struct gfs2_dinode));
607 	if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
608 		spin_unlock(&sdp->sd_statfs_spin);
609 		goto out_bh;
610 	}
611 	spin_unlock(&sdp->sd_statfs_spin);
612 
613 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
614 	if (error)
615 		goto out_bh;
616 
617 	error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
618 	if (error)
619 		goto out_bh2;
620 
621 	update_statfs(sdp, m_bh, l_bh);
622 	sdp->sd_statfs_force_sync = 0;
623 
624 	gfs2_trans_end(sdp);
625 
626 out_bh2:
627 	brelse(l_bh);
628 out_bh:
629 	brelse(m_bh);
630 out_unlock:
631 	gfs2_glock_dq_uninit(&gh);
632 out:
633 	sb_end_write(sb);
634 	return error;
635 }
636 
637 struct lfcc {
638 	struct list_head list;
639 	struct gfs2_holder gh;
640 };
641 
642 /**
643  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
644  *                            journals are clean
645  * @sdp: the file system
646  * @state: the state to put the transaction lock into
647  * @t_gh: the hold on the transaction lock
648  *
649  * Returns: errno
650  */
651 
652 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
653 				    struct gfs2_holder *freeze_gh)
654 {
655 	struct gfs2_inode *ip;
656 	struct gfs2_jdesc *jd;
657 	struct lfcc *lfcc;
658 	LIST_HEAD(list);
659 	struct gfs2_log_header_host lh;
660 	int error;
661 
662 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
663 		lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
664 		if (!lfcc) {
665 			error = -ENOMEM;
666 			goto out;
667 		}
668 		ip = GFS2_I(jd->jd_inode);
669 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
670 		if (error) {
671 			kfree(lfcc);
672 			goto out;
673 		}
674 		list_add(&lfcc->list, &list);
675 	}
676 
677 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
678 				   GL_NOCACHE, freeze_gh);
679 
680 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
681 		error = gfs2_jdesc_check(jd);
682 		if (error)
683 			break;
684 		error = gfs2_find_jhead(jd, &lh);
685 		if (error)
686 			break;
687 		if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
688 			error = -EBUSY;
689 			break;
690 		}
691 	}
692 
693 	if (error)
694 		gfs2_glock_dq_uninit(freeze_gh);
695 
696 out:
697 	while (!list_empty(&list)) {
698 		lfcc = list_entry(list.next, struct lfcc, list);
699 		list_del(&lfcc->list);
700 		gfs2_glock_dq_uninit(&lfcc->gh);
701 		kfree(lfcc);
702 	}
703 	return error;
704 }
705 
706 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
707 {
708 	struct gfs2_dinode *str = buf;
709 
710 	str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
711 	str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
712 	str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
713 	str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
714 	str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
715 	str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
716 	str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
717 	str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
718 	str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
719 	str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
720 	str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
721 	str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
722 	str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
723 	str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
724 
725 	str->di_goal_meta = cpu_to_be64(ip->i_goal);
726 	str->di_goal_data = cpu_to_be64(ip->i_goal);
727 	str->di_generation = cpu_to_be64(ip->i_generation);
728 
729 	str->di_flags = cpu_to_be32(ip->i_diskflags);
730 	str->di_height = cpu_to_be16(ip->i_height);
731 	str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
732 					     !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
733 					     GFS2_FORMAT_DE : 0);
734 	str->di_depth = cpu_to_be16(ip->i_depth);
735 	str->di_entries = cpu_to_be32(ip->i_entries);
736 
737 	str->di_eattr = cpu_to_be64(ip->i_eattr);
738 	str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
739 	str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
740 	str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
741 }
742 
743 /**
744  * gfs2_write_inode - Make sure the inode is stable on the disk
745  * @inode: The inode
746  * @wbc: The writeback control structure
747  *
748  * Returns: errno
749  */
750 
751 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
752 {
753 	struct gfs2_inode *ip = GFS2_I(inode);
754 	struct gfs2_sbd *sdp = GFS2_SB(inode);
755 	struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
756 	struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
757 	int ret = 0;
758 	bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
759 
760 	if (flush_all)
761 		gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
762 			       GFS2_LOG_HEAD_FLUSH_NORMAL |
763 			       GFS2_LFC_WRITE_INODE);
764 	if (bdi->wb.dirty_exceeded)
765 		gfs2_ail1_flush(sdp, wbc);
766 	else
767 		filemap_fdatawrite(metamapping);
768 	if (flush_all)
769 		ret = filemap_fdatawait(metamapping);
770 	if (ret)
771 		mark_inode_dirty_sync(inode);
772 	else {
773 		spin_lock(&inode->i_lock);
774 		if (!(inode->i_flags & I_DIRTY))
775 			gfs2_ordered_del_inode(ip);
776 		spin_unlock(&inode->i_lock);
777 	}
778 	return ret;
779 }
780 
781 /**
782  * gfs2_dirty_inode - check for atime updates
783  * @inode: The inode in question
784  * @flags: The type of dirty
785  *
786  * Unfortunately it can be called under any combination of inode
787  * glock and transaction lock, so we have to check carefully.
788  *
789  * At the moment this deals only with atime - it should be possible
790  * to expand that role in future, once a review of the locking has
791  * been carried out.
792  */
793 
794 static void gfs2_dirty_inode(struct inode *inode, int flags)
795 {
796 	struct gfs2_inode *ip = GFS2_I(inode);
797 	struct gfs2_sbd *sdp = GFS2_SB(inode);
798 	struct buffer_head *bh;
799 	struct gfs2_holder gh;
800 	int need_unlock = 0;
801 	int need_endtrans = 0;
802 	int ret;
803 
804 	if (!(flags & I_DIRTY_INODE))
805 		return;
806 	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
807 		return;
808 	if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
809 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
810 		if (ret) {
811 			fs_err(sdp, "dirty_inode: glock %d\n", ret);
812 			return;
813 		}
814 		need_unlock = 1;
815 	} else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
816 		return;
817 
818 	if (current->journal_info == NULL) {
819 		ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
820 		if (ret) {
821 			fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
822 			goto out;
823 		}
824 		need_endtrans = 1;
825 	}
826 
827 	ret = gfs2_meta_inode_buffer(ip, &bh);
828 	if (ret == 0) {
829 		gfs2_trans_add_meta(ip->i_gl, bh);
830 		gfs2_dinode_out(ip, bh->b_data);
831 		brelse(bh);
832 	}
833 
834 	if (need_endtrans)
835 		gfs2_trans_end(sdp);
836 out:
837 	if (need_unlock)
838 		gfs2_glock_dq_uninit(&gh);
839 }
840 
841 /**
842  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
843  * @sdp: the filesystem
844  *
845  * Returns: errno
846  */
847 
848 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
849 {
850 	struct gfs2_holder freeze_gh;
851 	int error;
852 
853 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, GL_NOCACHE,
854 				   &freeze_gh);
855 	if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
856 		return error;
857 
858 	flush_workqueue(gfs2_delete_workqueue);
859 	kthread_stop(sdp->sd_quotad_process);
860 	kthread_stop(sdp->sd_logd_process);
861 
862 	gfs2_quota_sync(sdp->sd_vfs, 0);
863 	gfs2_statfs_sync(sdp->sd_vfs, 0);
864 
865 	gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
866 		       GFS2_LFC_MAKE_FS_RO);
867 	wait_event(sdp->sd_reserving_log_wait, atomic_read(&sdp->sd_reserving_log) == 0);
868 	gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
869 
870 	if (gfs2_holder_initialized(&freeze_gh))
871 		gfs2_glock_dq_uninit(&freeze_gh);
872 
873 	gfs2_quota_cleanup(sdp);
874 
875 	return error;
876 }
877 
878 /**
879  * gfs2_put_super - Unmount the filesystem
880  * @sb: The VFS superblock
881  *
882  */
883 
884 static void gfs2_put_super(struct super_block *sb)
885 {
886 	struct gfs2_sbd *sdp = sb->s_fs_info;
887 	int error;
888 	struct gfs2_jdesc *jd;
889 
890 	/* No more recovery requests */
891 	set_bit(SDF_NORECOVERY, &sdp->sd_flags);
892 	smp_mb();
893 
894 	/* Wait on outstanding recovery */
895 restart:
896 	spin_lock(&sdp->sd_jindex_spin);
897 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
898 		if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
899 			continue;
900 		spin_unlock(&sdp->sd_jindex_spin);
901 		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
902 			    TASK_UNINTERRUPTIBLE);
903 		goto restart;
904 	}
905 	spin_unlock(&sdp->sd_jindex_spin);
906 
907 	if (!sb_rdonly(sb)) {
908 		error = gfs2_make_fs_ro(sdp);
909 		if (error)
910 			gfs2_io_error(sdp);
911 	}
912 	/*  At this point, we're through modifying the disk  */
913 
914 	/*  Release stuff  */
915 
916 	iput(sdp->sd_jindex);
917 	iput(sdp->sd_statfs_inode);
918 	iput(sdp->sd_rindex);
919 	iput(sdp->sd_quota_inode);
920 
921 	gfs2_glock_put(sdp->sd_rename_gl);
922 	gfs2_glock_put(sdp->sd_freeze_gl);
923 
924 	if (!sdp->sd_args.ar_spectator) {
925 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
926 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
927 		gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
928 		gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
929 		iput(sdp->sd_sc_inode);
930 		iput(sdp->sd_qc_inode);
931 	}
932 
933 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
934 	gfs2_clear_rgrpd(sdp);
935 	gfs2_jindex_free(sdp);
936 	/*  Take apart glock structures and buffer lists  */
937 	gfs2_gl_hash_clear(sdp);
938 	gfs2_delete_debugfs_file(sdp);
939 	/*  Unmount the locking protocol  */
940 	gfs2_lm_unmount(sdp);
941 
942 	/*  At this point, we're through participating in the lockspace  */
943 	gfs2_sys_fs_del(sdp);
944 }
945 
946 /**
947  * gfs2_sync_fs - sync the filesystem
948  * @sb: the superblock
949  *
950  * Flushes the log to disk.
951  */
952 
953 static int gfs2_sync_fs(struct super_block *sb, int wait)
954 {
955 	struct gfs2_sbd *sdp = sb->s_fs_info;
956 
957 	gfs2_quota_sync(sb, -1);
958 	if (wait)
959 		gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
960 			       GFS2_LFC_SYNC_FS);
961 	return sdp->sd_log_error;
962 }
963 
964 void gfs2_freeze_func(struct work_struct *work)
965 {
966 	int error;
967 	struct gfs2_holder freeze_gh;
968 	struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
969 	struct super_block *sb = sdp->sd_vfs;
970 
971 	atomic_inc(&sb->s_active);
972 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
973 				   &freeze_gh);
974 	if (error) {
975 		printk(KERN_INFO "GFS2: couldn't get freeze lock : %d\n", error);
976 		gfs2_assert_withdraw(sdp, 0);
977 	}
978 	else {
979 		atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
980 		error = thaw_super(sb);
981 		if (error) {
982 			printk(KERN_INFO "GFS2: couldn't thaw filesystem: %d\n",
983 			       error);
984 			gfs2_assert_withdraw(sdp, 0);
985 		}
986 		if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
987 			freeze_gh.gh_flags |= GL_NOCACHE;
988 		gfs2_glock_dq_uninit(&freeze_gh);
989 	}
990 	deactivate_super(sb);
991 	return;
992 }
993 
994 /**
995  * gfs2_freeze - prevent further writes to the filesystem
996  * @sb: the VFS structure for the filesystem
997  *
998  */
999 
1000 static int gfs2_freeze(struct super_block *sb)
1001 {
1002 	struct gfs2_sbd *sdp = sb->s_fs_info;
1003 	int error = 0;
1004 
1005 	mutex_lock(&sdp->sd_freeze_mutex);
1006 	if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN)
1007 		goto out;
1008 
1009 	if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
1010 		error = -EINVAL;
1011 		goto out;
1012 	}
1013 
1014 	for (;;) {
1015 		error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
1016 		if (!error)
1017 			break;
1018 
1019 		switch (error) {
1020 		case -EBUSY:
1021 			fs_err(sdp, "waiting for recovery before freeze\n");
1022 			break;
1023 
1024 		default:
1025 			fs_err(sdp, "error freezing FS: %d\n", error);
1026 			break;
1027 		}
1028 
1029 		fs_err(sdp, "retrying...\n");
1030 		msleep(1000);
1031 	}
1032 	error = 0;
1033 out:
1034 	mutex_unlock(&sdp->sd_freeze_mutex);
1035 	return error;
1036 }
1037 
1038 /**
1039  * gfs2_unfreeze - reallow writes to the filesystem
1040  * @sb: the VFS structure for the filesystem
1041  *
1042  */
1043 
1044 static int gfs2_unfreeze(struct super_block *sb)
1045 {
1046 	struct gfs2_sbd *sdp = sb->s_fs_info;
1047 
1048 	mutex_lock(&sdp->sd_freeze_mutex);
1049         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
1050 	    !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
1051 		mutex_unlock(&sdp->sd_freeze_mutex);
1052                 return 0;
1053 	}
1054 
1055 	gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1056 	mutex_unlock(&sdp->sd_freeze_mutex);
1057 	return 0;
1058 }
1059 
1060 /**
1061  * statfs_fill - fill in the sg for a given RG
1062  * @rgd: the RG
1063  * @sc: the sc structure
1064  *
1065  * Returns: 0 on success, -ESTALE if the LVB is invalid
1066  */
1067 
1068 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1069 			    struct gfs2_statfs_change_host *sc)
1070 {
1071 	gfs2_rgrp_verify(rgd);
1072 	sc->sc_total += rgd->rd_data;
1073 	sc->sc_free += rgd->rd_free;
1074 	sc->sc_dinodes += rgd->rd_dinodes;
1075 	return 0;
1076 }
1077 
1078 /**
1079  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1080  * @sdp: the filesystem
1081  * @sc: the sc info that will be returned
1082  *
1083  * Any error (other than a signal) will cause this routine to fall back
1084  * to the synchronous version.
1085  *
1086  * FIXME: This really shouldn't busy wait like this.
1087  *
1088  * Returns: errno
1089  */
1090 
1091 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1092 {
1093 	struct gfs2_rgrpd *rgd_next;
1094 	struct gfs2_holder *gha, *gh;
1095 	unsigned int slots = 64;
1096 	unsigned int x;
1097 	int done;
1098 	int error = 0, err;
1099 
1100 	memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1101 	gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1102 	if (!gha)
1103 		return -ENOMEM;
1104 	for (x = 0; x < slots; x++)
1105 		gfs2_holder_mark_uninitialized(gha + x);
1106 
1107 	rgd_next = gfs2_rgrpd_get_first(sdp);
1108 
1109 	for (;;) {
1110 		done = 1;
1111 
1112 		for (x = 0; x < slots; x++) {
1113 			gh = gha + x;
1114 
1115 			if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
1116 				err = gfs2_glock_wait(gh);
1117 				if (err) {
1118 					gfs2_holder_uninit(gh);
1119 					error = err;
1120 				} else {
1121 					if (!error) {
1122 						struct gfs2_rgrpd *rgd =
1123 							gfs2_glock2rgrp(gh->gh_gl);
1124 
1125 						error = statfs_slow_fill(rgd, sc);
1126 					}
1127 					gfs2_glock_dq_uninit(gh);
1128 				}
1129 			}
1130 
1131 			if (gfs2_holder_initialized(gh))
1132 				done = 0;
1133 			else if (rgd_next && !error) {
1134 				error = gfs2_glock_nq_init(rgd_next->rd_gl,
1135 							   LM_ST_SHARED,
1136 							   GL_ASYNC,
1137 							   gh);
1138 				rgd_next = gfs2_rgrpd_get_next(rgd_next);
1139 				done = 0;
1140 			}
1141 
1142 			if (signal_pending(current))
1143 				error = -ERESTARTSYS;
1144 		}
1145 
1146 		if (done)
1147 			break;
1148 
1149 		yield();
1150 	}
1151 
1152 	kfree(gha);
1153 	return error;
1154 }
1155 
1156 /**
1157  * gfs2_statfs_i - Do a statfs
1158  * @sdp: the filesystem
1159  * @sg: the sg structure
1160  *
1161  * Returns: errno
1162  */
1163 
1164 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1165 {
1166 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1167 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1168 
1169 	spin_lock(&sdp->sd_statfs_spin);
1170 
1171 	*sc = *m_sc;
1172 	sc->sc_total += l_sc->sc_total;
1173 	sc->sc_free += l_sc->sc_free;
1174 	sc->sc_dinodes += l_sc->sc_dinodes;
1175 
1176 	spin_unlock(&sdp->sd_statfs_spin);
1177 
1178 	if (sc->sc_free < 0)
1179 		sc->sc_free = 0;
1180 	if (sc->sc_free > sc->sc_total)
1181 		sc->sc_free = sc->sc_total;
1182 	if (sc->sc_dinodes < 0)
1183 		sc->sc_dinodes = 0;
1184 
1185 	return 0;
1186 }
1187 
1188 /**
1189  * gfs2_statfs - Gather and return stats about the filesystem
1190  * @sb: The superblock
1191  * @statfsbuf: The buffer
1192  *
1193  * Returns: 0 on success or error code
1194  */
1195 
1196 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1197 {
1198 	struct super_block *sb = dentry->d_sb;
1199 	struct gfs2_sbd *sdp = sb->s_fs_info;
1200 	struct gfs2_statfs_change_host sc;
1201 	int error;
1202 
1203 	error = gfs2_rindex_update(sdp);
1204 	if (error)
1205 		return error;
1206 
1207 	if (gfs2_tune_get(sdp, gt_statfs_slow))
1208 		error = gfs2_statfs_slow(sdp, &sc);
1209 	else
1210 		error = gfs2_statfs_i(sdp, &sc);
1211 
1212 	if (error)
1213 		return error;
1214 
1215 	buf->f_type = GFS2_MAGIC;
1216 	buf->f_bsize = sdp->sd_sb.sb_bsize;
1217 	buf->f_blocks = sc.sc_total;
1218 	buf->f_bfree = sc.sc_free;
1219 	buf->f_bavail = sc.sc_free;
1220 	buf->f_files = sc.sc_dinodes + sc.sc_free;
1221 	buf->f_ffree = sc.sc_free;
1222 	buf->f_namelen = GFS2_FNAMESIZE;
1223 
1224 	return 0;
1225 }
1226 
1227 /**
1228  * gfs2_remount_fs - called when the FS is remounted
1229  * @sb:  the filesystem
1230  * @flags:  the remount flags
1231  * @data:  extra data passed in (not used right now)
1232  *
1233  * Returns: errno
1234  */
1235 
1236 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1237 {
1238 	struct gfs2_sbd *sdp = sb->s_fs_info;
1239 	struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1240 	struct gfs2_tune *gt = &sdp->sd_tune;
1241 	int error;
1242 
1243 	sync_filesystem(sb);
1244 
1245 	spin_lock(&gt->gt_spin);
1246 	args.ar_commit = gt->gt_logd_secs;
1247 	args.ar_quota_quantum = gt->gt_quota_quantum;
1248 	if (gt->gt_statfs_slow)
1249 		args.ar_statfs_quantum = 0;
1250 	else
1251 		args.ar_statfs_quantum = gt->gt_statfs_quantum;
1252 	spin_unlock(&gt->gt_spin);
1253 	error = gfs2_mount_args(&args, data);
1254 	if (error)
1255 		return error;
1256 
1257 	/* Not allowed to change locking details */
1258 	if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1259 	    strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1260 	    strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1261 		return -EINVAL;
1262 
1263 	/* Some flags must not be changed */
1264 	if (args_neq(&args, &sdp->sd_args, spectator) ||
1265 	    args_neq(&args, &sdp->sd_args, localflocks) ||
1266 	    args_neq(&args, &sdp->sd_args, meta))
1267 		return -EINVAL;
1268 
1269 	if (sdp->sd_args.ar_spectator)
1270 		*flags |= SB_RDONLY;
1271 
1272 	if ((sb->s_flags ^ *flags) & SB_RDONLY) {
1273 		if (*flags & SB_RDONLY)
1274 			error = gfs2_make_fs_ro(sdp);
1275 		else
1276 			error = gfs2_make_fs_rw(sdp);
1277 		if (error)
1278 			return error;
1279 	}
1280 
1281 	sdp->sd_args = args;
1282 	if (sdp->sd_args.ar_posix_acl)
1283 		sb->s_flags |= SB_POSIXACL;
1284 	else
1285 		sb->s_flags &= ~SB_POSIXACL;
1286 	if (sdp->sd_args.ar_nobarrier)
1287 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1288 	else
1289 		clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1290 	spin_lock(&gt->gt_spin);
1291 	gt->gt_logd_secs = args.ar_commit;
1292 	gt->gt_quota_quantum = args.ar_quota_quantum;
1293 	if (args.ar_statfs_quantum) {
1294 		gt->gt_statfs_slow = 0;
1295 		gt->gt_statfs_quantum = args.ar_statfs_quantum;
1296 	}
1297 	else {
1298 		gt->gt_statfs_slow = 1;
1299 		gt->gt_statfs_quantum = 30;
1300 	}
1301 	spin_unlock(&gt->gt_spin);
1302 
1303 	gfs2_online_uevent(sdp);
1304 	return 0;
1305 }
1306 
1307 /**
1308  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1309  * @inode: The inode to drop
1310  *
1311  * If we've received a callback on an iopen lock then it's because a
1312  * remote node tried to deallocate the inode but failed due to this node
1313  * still having the inode open. Here we mark the link count zero
1314  * since we know that it must have reached zero if the GLF_DEMOTE flag
1315  * is set on the iopen glock. If we didn't do a disk read since the
1316  * remote node removed the final link then we might otherwise miss
1317  * this event. This check ensures that this node will deallocate the
1318  * inode's blocks, or alternatively pass the baton on to another
1319  * node for later deallocation.
1320  */
1321 
1322 static int gfs2_drop_inode(struct inode *inode)
1323 {
1324 	struct gfs2_inode *ip = GFS2_I(inode);
1325 
1326 	if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
1327 	    inode->i_nlink &&
1328 	    gfs2_holder_initialized(&ip->i_iopen_gh)) {
1329 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1330 		if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1331 			clear_nlink(inode);
1332 	}
1333 
1334 	/*
1335 	 * When under memory pressure when an inode's link count has dropped to
1336 	 * zero, defer deleting the inode to the delete workqueue.  This avoids
1337 	 * calling into DLM under memory pressure, which can deadlock.
1338 	 */
1339 	if (!inode->i_nlink &&
1340 	    unlikely(current->flags & PF_MEMALLOC) &&
1341 	    gfs2_holder_initialized(&ip->i_iopen_gh)) {
1342 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1343 
1344 		gfs2_glock_hold(gl);
1345 		if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
1346 			gfs2_glock_queue_put(gl);
1347 		return false;
1348 	}
1349 
1350 	return generic_drop_inode(inode);
1351 }
1352 
1353 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1354 {
1355 	do {
1356 		if (d1 == d2)
1357 			return 1;
1358 		d1 = d1->d_parent;
1359 	} while (!IS_ROOT(d1));
1360 	return 0;
1361 }
1362 
1363 /**
1364  * gfs2_show_options - Show mount options for /proc/mounts
1365  * @s: seq_file structure
1366  * @root: root of this (sub)tree
1367  *
1368  * Returns: 0 on success or error code
1369  */
1370 
1371 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1372 {
1373 	struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1374 	struct gfs2_args *args = &sdp->sd_args;
1375 	int val;
1376 
1377 	if (is_ancestor(root, sdp->sd_master_dir))
1378 		seq_puts(s, ",meta");
1379 	if (args->ar_lockproto[0])
1380 		seq_show_option(s, "lockproto", args->ar_lockproto);
1381 	if (args->ar_locktable[0])
1382 		seq_show_option(s, "locktable", args->ar_locktable);
1383 	if (args->ar_hostdata[0])
1384 		seq_show_option(s, "hostdata", args->ar_hostdata);
1385 	if (args->ar_spectator)
1386 		seq_puts(s, ",spectator");
1387 	if (args->ar_localflocks)
1388 		seq_puts(s, ",localflocks");
1389 	if (args->ar_debug)
1390 		seq_puts(s, ",debug");
1391 	if (args->ar_posix_acl)
1392 		seq_puts(s, ",acl");
1393 	if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1394 		char *state;
1395 		switch (args->ar_quota) {
1396 		case GFS2_QUOTA_OFF:
1397 			state = "off";
1398 			break;
1399 		case GFS2_QUOTA_ACCOUNT:
1400 			state = "account";
1401 			break;
1402 		case GFS2_QUOTA_ON:
1403 			state = "on";
1404 			break;
1405 		default:
1406 			state = "unknown";
1407 			break;
1408 		}
1409 		seq_printf(s, ",quota=%s", state);
1410 	}
1411 	if (args->ar_suiddir)
1412 		seq_puts(s, ",suiddir");
1413 	if (args->ar_data != GFS2_DATA_DEFAULT) {
1414 		char *state;
1415 		switch (args->ar_data) {
1416 		case GFS2_DATA_WRITEBACK:
1417 			state = "writeback";
1418 			break;
1419 		case GFS2_DATA_ORDERED:
1420 			state = "ordered";
1421 			break;
1422 		default:
1423 			state = "unknown";
1424 			break;
1425 		}
1426 		seq_printf(s, ",data=%s", state);
1427 	}
1428 	if (args->ar_discard)
1429 		seq_puts(s, ",discard");
1430 	val = sdp->sd_tune.gt_logd_secs;
1431 	if (val != 30)
1432 		seq_printf(s, ",commit=%d", val);
1433 	val = sdp->sd_tune.gt_statfs_quantum;
1434 	if (val != 30)
1435 		seq_printf(s, ",statfs_quantum=%d", val);
1436 	else if (sdp->sd_tune.gt_statfs_slow)
1437 		seq_puts(s, ",statfs_quantum=0");
1438 	val = sdp->sd_tune.gt_quota_quantum;
1439 	if (val != 60)
1440 		seq_printf(s, ",quota_quantum=%d", val);
1441 	if (args->ar_statfs_percent)
1442 		seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1443 	if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1444 		const char *state;
1445 
1446 		switch (args->ar_errors) {
1447 		case GFS2_ERRORS_WITHDRAW:
1448 			state = "withdraw";
1449 			break;
1450 		case GFS2_ERRORS_PANIC:
1451 			state = "panic";
1452 			break;
1453 		default:
1454 			state = "unknown";
1455 			break;
1456 		}
1457 		seq_printf(s, ",errors=%s", state);
1458 	}
1459 	if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1460 		seq_puts(s, ",nobarrier");
1461 	if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1462 		seq_puts(s, ",demote_interface_used");
1463 	if (args->ar_rgrplvb)
1464 		seq_puts(s, ",rgrplvb");
1465 	if (args->ar_loccookie)
1466 		seq_puts(s, ",loccookie");
1467 	return 0;
1468 }
1469 
1470 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1471 {
1472 	struct inode *inode = &ip->i_inode;
1473 	struct gfs2_glock *gl = ip->i_gl;
1474 
1475 	truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1476 	truncate_inode_pages(&inode->i_data, 0);
1477 
1478 	if (atomic_read(&gl->gl_revokes) == 0) {
1479 		clear_bit(GLF_LFLUSH, &gl->gl_flags);
1480 		clear_bit(GLF_DIRTY, &gl->gl_flags);
1481 	}
1482 }
1483 
1484 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1485 {
1486 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1487 	struct gfs2_rgrpd *rgd;
1488 	struct gfs2_holder gh;
1489 	int error;
1490 
1491 	if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1492 		gfs2_consist_inode(ip);
1493 		return -EIO;
1494 	}
1495 
1496 	error = gfs2_rindex_update(sdp);
1497 	if (error)
1498 		return error;
1499 
1500 	error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1501 	if (error)
1502 		return error;
1503 
1504 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1505 	if (!rgd) {
1506 		gfs2_consist_inode(ip);
1507 		error = -EIO;
1508 		goto out_qs;
1509 	}
1510 
1511 	error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1512 	if (error)
1513 		goto out_qs;
1514 
1515 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1516 				 sdp->sd_jdesc->jd_blocks);
1517 	if (error)
1518 		goto out_rg_gunlock;
1519 
1520 	gfs2_free_di(rgd, ip);
1521 
1522 	gfs2_final_release_pages(ip);
1523 
1524 	gfs2_trans_end(sdp);
1525 
1526 out_rg_gunlock:
1527 	gfs2_glock_dq_uninit(&gh);
1528 out_qs:
1529 	gfs2_quota_unhold(ip);
1530 	return error;
1531 }
1532 
1533 /**
1534  * gfs2_glock_put_eventually
1535  * @gl:	The glock to put
1536  *
1537  * When under memory pressure, trigger a deferred glock put to make sure we
1538  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1539  */
1540 
1541 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1542 {
1543 	if (current->flags & PF_MEMALLOC)
1544 		gfs2_glock_queue_put(gl);
1545 	else
1546 		gfs2_glock_put(gl);
1547 }
1548 
1549 /**
1550  * gfs2_evict_inode - Remove an inode from cache
1551  * @inode: The inode to evict
1552  *
1553  * There are three cases to consider:
1554  * 1. i_nlink == 0, we are final opener (and must deallocate)
1555  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1556  * 3. i_nlink > 0
1557  *
1558  * If the fs is read only, then we have to treat all cases as per #3
1559  * since we are unable to do any deallocation. The inode will be
1560  * deallocated by the next read/write node to attempt an allocation
1561  * in the same resource group
1562  *
1563  * We have to (at the moment) hold the inodes main lock to cover
1564  * the gap between unlocking the shared lock on the iopen lock and
1565  * taking the exclusive lock. I'd rather do a shared -> exclusive
1566  * conversion on the iopen lock, but we can change that later. This
1567  * is safe, just less efficient.
1568  */
1569 
1570 static void gfs2_evict_inode(struct inode *inode)
1571 {
1572 	struct super_block *sb = inode->i_sb;
1573 	struct gfs2_sbd *sdp = sb->s_fs_info;
1574 	struct gfs2_inode *ip = GFS2_I(inode);
1575 	struct gfs2_holder gh;
1576 	struct address_space *metamapping;
1577 	int error;
1578 
1579 	if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1580 		clear_inode(inode);
1581 		return;
1582 	}
1583 
1584 	if (inode->i_nlink || sb_rdonly(sb))
1585 		goto out;
1586 
1587 	if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1588 		BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1589 		gfs2_holder_mark_uninitialized(&gh);
1590 		goto alloc_failed;
1591 	}
1592 
1593 	/* Deletes should never happen under memory pressure anymore.  */
1594 	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1595 		goto out;
1596 
1597 	/* Must not read inode block until block type has been verified */
1598 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1599 	if (unlikely(error)) {
1600 		glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1601 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1602 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1603 		goto out;
1604 	}
1605 
1606 	error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1607 	if (error)
1608 		goto out_truncate;
1609 
1610 	if (test_bit(GIF_INVALID, &ip->i_flags)) {
1611 		error = gfs2_inode_refresh(ip);
1612 		if (error)
1613 			goto out_truncate;
1614 	}
1615 
1616 	/*
1617 	 * The inode may have been recreated in the meantime.
1618 	 */
1619 	if (inode->i_nlink)
1620 		goto out_truncate;
1621 
1622 alloc_failed:
1623 	if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1624 	    test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1625 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1626 		gfs2_glock_dq_wait(&ip->i_iopen_gh);
1627 		gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE,
1628 				   &ip->i_iopen_gh);
1629 		error = gfs2_glock_nq(&ip->i_iopen_gh);
1630 		if (error)
1631 			goto out_truncate;
1632 	}
1633 
1634 	/* Case 1 starts here */
1635 
1636 	if (S_ISDIR(inode->i_mode) &&
1637 	    (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1638 		error = gfs2_dir_exhash_dealloc(ip);
1639 		if (error)
1640 			goto out_unlock;
1641 	}
1642 
1643 	if (ip->i_eattr) {
1644 		error = gfs2_ea_dealloc(ip);
1645 		if (error)
1646 			goto out_unlock;
1647 	}
1648 
1649 	if (!gfs2_is_stuffed(ip)) {
1650 		error = gfs2_file_dealloc(ip);
1651 		if (error)
1652 			goto out_unlock;
1653 	}
1654 
1655 	/* We're about to clear the bitmap for the dinode, but as soon as we
1656 	   do, gfs2_create_inode can create another inode at the same block
1657 	   location and try to set gl_object again. We clear gl_object here so
1658 	   that subsequent inode creates don't see an old gl_object. */
1659 	glock_clear_object(ip->i_gl, ip);
1660 	error = gfs2_dinode_dealloc(ip);
1661 	goto out_unlock;
1662 
1663 out_truncate:
1664 	gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1665 		       GFS2_LFC_EVICT_INODE);
1666 	metamapping = gfs2_glock2aspace(ip->i_gl);
1667 	if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1668 		filemap_fdatawrite(metamapping);
1669 		filemap_fdatawait(metamapping);
1670 	}
1671 	write_inode_now(inode, 1);
1672 	gfs2_ail_flush(ip->i_gl, 0);
1673 
1674 	/* Case 2 starts here */
1675 	error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1676 	if (error)
1677 		goto out_unlock;
1678 	/* Needs to be done before glock release & also in a transaction */
1679 	truncate_inode_pages(&inode->i_data, 0);
1680 	truncate_inode_pages(metamapping, 0);
1681 	gfs2_trans_end(sdp);
1682 
1683 out_unlock:
1684 	/* Error path for case 1 */
1685 	if (gfs2_rs_active(&ip->i_res))
1686 		gfs2_rs_deltree(&ip->i_res);
1687 
1688 	if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1689 		glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1690 		if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1691 			ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1692 			gfs2_glock_dq(&ip->i_iopen_gh);
1693 		}
1694 		gfs2_holder_uninit(&ip->i_iopen_gh);
1695 	}
1696 	if (gfs2_holder_initialized(&gh)) {
1697 		glock_clear_object(ip->i_gl, ip);
1698 		gfs2_glock_dq_uninit(&gh);
1699 	}
1700 	if (error && error != GLR_TRYFAILED && error != -EROFS)
1701 		fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1702 out:
1703 	/* Case 3 starts here */
1704 	truncate_inode_pages_final(&inode->i_data);
1705 	gfs2_rsqa_delete(ip, NULL);
1706 	gfs2_ordered_del_inode(ip);
1707 	clear_inode(inode);
1708 	gfs2_dir_hash_inval(ip);
1709 	glock_clear_object(ip->i_gl, ip);
1710 	wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1711 	gfs2_glock_add_to_lru(ip->i_gl);
1712 	gfs2_glock_put_eventually(ip->i_gl);
1713 	ip->i_gl = NULL;
1714 	if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1715 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1716 
1717 		glock_clear_object(gl, ip);
1718 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1719 		gfs2_glock_hold(gl);
1720 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1721 		gfs2_glock_put_eventually(gl);
1722 	}
1723 }
1724 
1725 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1726 {
1727 	struct gfs2_inode *ip;
1728 
1729 	ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1730 	if (ip) {
1731 		ip->i_flags = 0;
1732 		ip->i_gl = NULL;
1733 		memset(&ip->i_res, 0, sizeof(ip->i_res));
1734 		RB_CLEAR_NODE(&ip->i_res.rs_node);
1735 		ip->i_rahead = 0;
1736 	}
1737 	return &ip->i_inode;
1738 }
1739 
1740 static void gfs2_i_callback(struct rcu_head *head)
1741 {
1742 	struct inode *inode = container_of(head, struct inode, i_rcu);
1743 	kmem_cache_free(gfs2_inode_cachep, inode);
1744 }
1745 
1746 static void gfs2_destroy_inode(struct inode *inode)
1747 {
1748 	call_rcu(&inode->i_rcu, gfs2_i_callback);
1749 }
1750 
1751 const struct super_operations gfs2_super_ops = {
1752 	.alloc_inode		= gfs2_alloc_inode,
1753 	.destroy_inode		= gfs2_destroy_inode,
1754 	.write_inode		= gfs2_write_inode,
1755 	.dirty_inode		= gfs2_dirty_inode,
1756 	.evict_inode		= gfs2_evict_inode,
1757 	.put_super		= gfs2_put_super,
1758 	.sync_fs		= gfs2_sync_fs,
1759 	.freeze_super		= gfs2_freeze,
1760 	.thaw_super		= gfs2_unfreeze,
1761 	.statfs			= gfs2_statfs,
1762 	.remount_fs		= gfs2_remount_fs,
1763 	.drop_inode		= gfs2_drop_inode,
1764 	.show_options		= gfs2_show_options,
1765 };
1766 
1767