xref: /openbmc/linux/fs/gfs2/super.c (revision ccd51b9f)
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 
47 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
48 
49 enum {
50 	Opt_lockproto,
51 	Opt_locktable,
52 	Opt_hostdata,
53 	Opt_spectator,
54 	Opt_ignore_local_fs,
55 	Opt_localflocks,
56 	Opt_localcaching,
57 	Opt_debug,
58 	Opt_nodebug,
59 	Opt_upgrade,
60 	Opt_acl,
61 	Opt_noacl,
62 	Opt_quota_off,
63 	Opt_quota_account,
64 	Opt_quota_on,
65 	Opt_quota,
66 	Opt_noquota,
67 	Opt_suiddir,
68 	Opt_nosuiddir,
69 	Opt_data_writeback,
70 	Opt_data_ordered,
71 	Opt_meta,
72 	Opt_discard,
73 	Opt_nodiscard,
74 	Opt_commit,
75 	Opt_err_withdraw,
76 	Opt_err_panic,
77 	Opt_statfs_quantum,
78 	Opt_statfs_percent,
79 	Opt_quota_quantum,
80 	Opt_barrier,
81 	Opt_nobarrier,
82 	Opt_rgrplvb,
83 	Opt_norgrplvb,
84 	Opt_loccookie,
85 	Opt_noloccookie,
86 	Opt_error,
87 };
88 
89 static const match_table_t tokens = {
90 	{Opt_lockproto, "lockproto=%s"},
91 	{Opt_locktable, "locktable=%s"},
92 	{Opt_hostdata, "hostdata=%s"},
93 	{Opt_spectator, "spectator"},
94 	{Opt_spectator, "norecovery"},
95 	{Opt_ignore_local_fs, "ignore_local_fs"},
96 	{Opt_localflocks, "localflocks"},
97 	{Opt_localcaching, "localcaching"},
98 	{Opt_debug, "debug"},
99 	{Opt_nodebug, "nodebug"},
100 	{Opt_upgrade, "upgrade"},
101 	{Opt_acl, "acl"},
102 	{Opt_noacl, "noacl"},
103 	{Opt_quota_off, "quota=off"},
104 	{Opt_quota_account, "quota=account"},
105 	{Opt_quota_on, "quota=on"},
106 	{Opt_quota, "quota"},
107 	{Opt_noquota, "noquota"},
108 	{Opt_suiddir, "suiddir"},
109 	{Opt_nosuiddir, "nosuiddir"},
110 	{Opt_data_writeback, "data=writeback"},
111 	{Opt_data_ordered, "data=ordered"},
112 	{Opt_meta, "meta"},
113 	{Opt_discard, "discard"},
114 	{Opt_nodiscard, "nodiscard"},
115 	{Opt_commit, "commit=%d"},
116 	{Opt_err_withdraw, "errors=withdraw"},
117 	{Opt_err_panic, "errors=panic"},
118 	{Opt_statfs_quantum, "statfs_quantum=%d"},
119 	{Opt_statfs_percent, "statfs_percent=%d"},
120 	{Opt_quota_quantum, "quota_quantum=%d"},
121 	{Opt_barrier, "barrier"},
122 	{Opt_nobarrier, "nobarrier"},
123 	{Opt_rgrplvb, "rgrplvb"},
124 	{Opt_norgrplvb, "norgrplvb"},
125 	{Opt_loccookie, "loccookie"},
126 	{Opt_noloccookie, "noloccookie"},
127 	{Opt_error, NULL}
128 };
129 
130 /**
131  * gfs2_mount_args - Parse mount options
132  * @args: The structure into which the parsed options will be written
133  * @options: The options to parse
134  *
135  * Return: errno
136  */
137 
138 int gfs2_mount_args(struct gfs2_args *args, char *options)
139 {
140 	char *o;
141 	int token;
142 	substring_t tmp[MAX_OPT_ARGS];
143 	int rv;
144 
145 	/* Split the options into tokens with the "," character and
146 	   process them */
147 
148 	while (1) {
149 		o = strsep(&options, ",");
150 		if (o == NULL)
151 			break;
152 		if (*o == '\0')
153 			continue;
154 
155 		token = match_token(o, tokens, tmp);
156 		switch (token) {
157 		case Opt_lockproto:
158 			match_strlcpy(args->ar_lockproto, &tmp[0],
159 				      GFS2_LOCKNAME_LEN);
160 			break;
161 		case Opt_locktable:
162 			match_strlcpy(args->ar_locktable, &tmp[0],
163 				      GFS2_LOCKNAME_LEN);
164 			break;
165 		case Opt_hostdata:
166 			match_strlcpy(args->ar_hostdata, &tmp[0],
167 				      GFS2_LOCKNAME_LEN);
168 			break;
169 		case Opt_spectator:
170 			args->ar_spectator = 1;
171 			break;
172 		case Opt_ignore_local_fs:
173 			/* Retained for backwards compat only */
174 			break;
175 		case Opt_localflocks:
176 			args->ar_localflocks = 1;
177 			break;
178 		case Opt_localcaching:
179 			/* Retained for backwards compat only */
180 			break;
181 		case Opt_debug:
182 			if (args->ar_errors == GFS2_ERRORS_PANIC) {
183 				pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
184 				return -EINVAL;
185 			}
186 			args->ar_debug = 1;
187 			break;
188 		case Opt_nodebug:
189 			args->ar_debug = 0;
190 			break;
191 		case Opt_upgrade:
192 			/* Retained for backwards compat only */
193 			break;
194 		case Opt_acl:
195 			args->ar_posix_acl = 1;
196 			break;
197 		case Opt_noacl:
198 			args->ar_posix_acl = 0;
199 			break;
200 		case Opt_quota_off:
201 		case Opt_noquota:
202 			args->ar_quota = GFS2_QUOTA_OFF;
203 			break;
204 		case Opt_quota_account:
205 			args->ar_quota = GFS2_QUOTA_ACCOUNT;
206 			break;
207 		case Opt_quota_on:
208 		case Opt_quota:
209 			args->ar_quota = GFS2_QUOTA_ON;
210 			break;
211 		case Opt_suiddir:
212 			args->ar_suiddir = 1;
213 			break;
214 		case Opt_nosuiddir:
215 			args->ar_suiddir = 0;
216 			break;
217 		case Opt_data_writeback:
218 			args->ar_data = GFS2_DATA_WRITEBACK;
219 			break;
220 		case Opt_data_ordered:
221 			args->ar_data = GFS2_DATA_ORDERED;
222 			break;
223 		case Opt_meta:
224 			args->ar_meta = 1;
225 			break;
226 		case Opt_discard:
227 			args->ar_discard = 1;
228 			break;
229 		case Opt_nodiscard:
230 			args->ar_discard = 0;
231 			break;
232 		case Opt_commit:
233 			rv = match_int(&tmp[0], &args->ar_commit);
234 			if (rv || args->ar_commit <= 0) {
235 				pr_warn("commit mount option requires a positive numeric argument\n");
236 				return rv ? rv : -EINVAL;
237 			}
238 			break;
239 		case Opt_statfs_quantum:
240 			rv = match_int(&tmp[0], &args->ar_statfs_quantum);
241 			if (rv || args->ar_statfs_quantum < 0) {
242 				pr_warn("statfs_quantum mount option requires a non-negative numeric argument\n");
243 				return rv ? rv : -EINVAL;
244 			}
245 			break;
246 		case Opt_quota_quantum:
247 			rv = match_int(&tmp[0], &args->ar_quota_quantum);
248 			if (rv || args->ar_quota_quantum <= 0) {
249 				pr_warn("quota_quantum mount option requires a positive numeric argument\n");
250 				return rv ? rv : -EINVAL;
251 			}
252 			break;
253 		case Opt_statfs_percent:
254 			rv = match_int(&tmp[0], &args->ar_statfs_percent);
255 			if (rv || args->ar_statfs_percent < 0 ||
256 			    args->ar_statfs_percent > 100) {
257 				pr_warn("statfs_percent mount option requires a numeric argument between 0 and 100\n");
258 				return rv ? rv : -EINVAL;
259 			}
260 			break;
261 		case Opt_err_withdraw:
262 			args->ar_errors = GFS2_ERRORS_WITHDRAW;
263 			break;
264 		case Opt_err_panic:
265 			if (args->ar_debug) {
266 				pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
267 				return -EINVAL;
268 			}
269 			args->ar_errors = GFS2_ERRORS_PANIC;
270 			break;
271 		case Opt_barrier:
272 			args->ar_nobarrier = 0;
273 			break;
274 		case Opt_nobarrier:
275 			args->ar_nobarrier = 1;
276 			break;
277 		case Opt_rgrplvb:
278 			args->ar_rgrplvb = 1;
279 			break;
280 		case Opt_norgrplvb:
281 			args->ar_rgrplvb = 0;
282 			break;
283 		case Opt_loccookie:
284 			args->ar_loccookie = 1;
285 			break;
286 		case Opt_noloccookie:
287 			args->ar_loccookie = 0;
288 			break;
289 		case Opt_error:
290 		default:
291 			pr_warn("invalid mount option: %s\n", o);
292 			return -EINVAL;
293 		}
294 	}
295 
296 	return 0;
297 }
298 
299 /**
300  * gfs2_jindex_free - Clear all the journal index information
301  * @sdp: The GFS2 superblock
302  *
303  */
304 
305 void gfs2_jindex_free(struct gfs2_sbd *sdp)
306 {
307 	struct list_head list;
308 	struct gfs2_jdesc *jd;
309 
310 	spin_lock(&sdp->sd_jindex_spin);
311 	list_add(&list, &sdp->sd_jindex_list);
312 	list_del_init(&sdp->sd_jindex_list);
313 	sdp->sd_journals = 0;
314 	spin_unlock(&sdp->sd_jindex_spin);
315 
316 	while (!list_empty(&list)) {
317 		jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
318 		gfs2_free_journal_extents(jd);
319 		list_del(&jd->jd_list);
320 		iput(jd->jd_inode);
321 		kfree(jd);
322 	}
323 }
324 
325 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
326 {
327 	struct gfs2_jdesc *jd;
328 	int found = 0;
329 
330 	list_for_each_entry(jd, head, jd_list) {
331 		if (jd->jd_jid == jid) {
332 			found = 1;
333 			break;
334 		}
335 	}
336 
337 	if (!found)
338 		jd = NULL;
339 
340 	return jd;
341 }
342 
343 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
344 {
345 	struct gfs2_jdesc *jd;
346 
347 	spin_lock(&sdp->sd_jindex_spin);
348 	jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
349 	spin_unlock(&sdp->sd_jindex_spin);
350 
351 	return jd;
352 }
353 
354 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
355 {
356 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
357 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
358 	u64 size = i_size_read(jd->jd_inode);
359 
360 	if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
361 		return -EIO;
362 
363 	jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
364 
365 	if (gfs2_write_alloc_required(ip, 0, size)) {
366 		gfs2_consist_inode(ip);
367 		return -EIO;
368 	}
369 
370 	return 0;
371 }
372 
373 static int init_threads(struct gfs2_sbd *sdp)
374 {
375 	struct task_struct *p;
376 	int error = 0;
377 
378 	p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
379 	if (IS_ERR(p)) {
380 		error = PTR_ERR(p);
381 		fs_err(sdp, "can't start logd thread: %d\n", error);
382 		return error;
383 	}
384 	sdp->sd_logd_process = p;
385 
386 	p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
387 	if (IS_ERR(p)) {
388 		error = PTR_ERR(p);
389 		fs_err(sdp, "can't start quotad thread: %d\n", error);
390 		goto fail;
391 	}
392 	sdp->sd_quotad_process = p;
393 	return 0;
394 
395 fail:
396 	kthread_stop(sdp->sd_logd_process);
397 	sdp->sd_logd_process = NULL;
398 	return error;
399 }
400 
401 /**
402  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
403  * @sdp: the filesystem
404  *
405  * Returns: errno
406  */
407 
408 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
409 {
410 	struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
411 	struct gfs2_glock *j_gl = ip->i_gl;
412 	struct gfs2_holder freeze_gh;
413 	struct gfs2_log_header_host head;
414 	int error;
415 
416 	error = init_threads(sdp);
417 	if (error)
418 		return error;
419 
420 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
421 				   &freeze_gh);
422 	if (error)
423 		goto fail_threads;
424 
425 	j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
426 
427 	error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
428 	if (error)
429 		goto fail;
430 
431 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
432 		gfs2_consist(sdp);
433 		error = -EIO;
434 		goto fail;
435 	}
436 
437 	/*  Initialize some head of the log stuff  */
438 	sdp->sd_log_sequence = head.lh_sequence + 1;
439 	gfs2_log_pointers_init(sdp, head.lh_blkno);
440 
441 	error = gfs2_quota_init(sdp);
442 	if (error)
443 		goto fail;
444 
445 	set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
446 
447 	gfs2_glock_dq_uninit(&freeze_gh);
448 
449 	return 0;
450 
451 fail:
452 	freeze_gh.gh_flags |= GL_NOCACHE;
453 	gfs2_glock_dq_uninit(&freeze_gh);
454 fail_threads:
455 	if (sdp->sd_quotad_process)
456 		kthread_stop(sdp->sd_quotad_process);
457 	sdp->sd_quotad_process = NULL;
458 	if (sdp->sd_logd_process)
459 		kthread_stop(sdp->sd_logd_process);
460 	sdp->sd_logd_process = NULL;
461 	return error;
462 }
463 
464 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
465 {
466 	const struct gfs2_statfs_change *str = buf;
467 
468 	sc->sc_total = be64_to_cpu(str->sc_total);
469 	sc->sc_free = be64_to_cpu(str->sc_free);
470 	sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
471 }
472 
473 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
474 {
475 	struct gfs2_statfs_change *str = buf;
476 
477 	str->sc_total = cpu_to_be64(sc->sc_total);
478 	str->sc_free = cpu_to_be64(sc->sc_free);
479 	str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
480 }
481 
482 int gfs2_statfs_init(struct gfs2_sbd *sdp)
483 {
484 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
485 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
486 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
487 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
488 	struct buffer_head *m_bh, *l_bh;
489 	struct gfs2_holder gh;
490 	int error;
491 
492 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
493 				   &gh);
494 	if (error)
495 		return error;
496 
497 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
498 	if (error)
499 		goto out;
500 
501 	if (sdp->sd_args.ar_spectator) {
502 		spin_lock(&sdp->sd_statfs_spin);
503 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
504 				      sizeof(struct gfs2_dinode));
505 		spin_unlock(&sdp->sd_statfs_spin);
506 	} else {
507 		error = gfs2_meta_inode_buffer(l_ip, &l_bh);
508 		if (error)
509 			goto out_m_bh;
510 
511 		spin_lock(&sdp->sd_statfs_spin);
512 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
513 				      sizeof(struct gfs2_dinode));
514 		gfs2_statfs_change_in(l_sc, l_bh->b_data +
515 				      sizeof(struct gfs2_dinode));
516 		spin_unlock(&sdp->sd_statfs_spin);
517 
518 		brelse(l_bh);
519 	}
520 
521 out_m_bh:
522 	brelse(m_bh);
523 out:
524 	gfs2_glock_dq_uninit(&gh);
525 	return 0;
526 }
527 
528 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
529 			s64 dinodes)
530 {
531 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
532 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
533 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
534 	struct buffer_head *l_bh;
535 	s64 x, y;
536 	int need_sync = 0;
537 	int error;
538 
539 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
540 	if (error)
541 		return;
542 
543 	gfs2_trans_add_meta(l_ip->i_gl, l_bh);
544 
545 	spin_lock(&sdp->sd_statfs_spin);
546 	l_sc->sc_total += total;
547 	l_sc->sc_free += free;
548 	l_sc->sc_dinodes += dinodes;
549 	gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
550 	if (sdp->sd_args.ar_statfs_percent) {
551 		x = 100 * l_sc->sc_free;
552 		y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
553 		if (x >= y || x <= -y)
554 			need_sync = 1;
555 	}
556 	spin_unlock(&sdp->sd_statfs_spin);
557 
558 	brelse(l_bh);
559 	if (need_sync)
560 		gfs2_wake_up_statfs(sdp);
561 }
562 
563 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
564 		   struct buffer_head *l_bh)
565 {
566 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
567 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
568 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
569 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
570 
571 	gfs2_trans_add_meta(l_ip->i_gl, l_bh);
572 	gfs2_trans_add_meta(m_ip->i_gl, m_bh);
573 
574 	spin_lock(&sdp->sd_statfs_spin);
575 	m_sc->sc_total += l_sc->sc_total;
576 	m_sc->sc_free += l_sc->sc_free;
577 	m_sc->sc_dinodes += l_sc->sc_dinodes;
578 	memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
579 	memset(l_bh->b_data + sizeof(struct gfs2_dinode),
580 	       0, sizeof(struct gfs2_statfs_change));
581 	gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
582 	spin_unlock(&sdp->sd_statfs_spin);
583 }
584 
585 int gfs2_statfs_sync(struct super_block *sb, int type)
586 {
587 	struct gfs2_sbd *sdp = sb->s_fs_info;
588 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
589 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
590 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
591 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
592 	struct gfs2_holder gh;
593 	struct buffer_head *m_bh, *l_bh;
594 	int error;
595 
596 	sb_start_write(sb);
597 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
598 				   &gh);
599 	if (error)
600 		goto out;
601 
602 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
603 	if (error)
604 		goto out_unlock;
605 
606 	spin_lock(&sdp->sd_statfs_spin);
607 	gfs2_statfs_change_in(m_sc, m_bh->b_data +
608 			      sizeof(struct gfs2_dinode));
609 	if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
610 		spin_unlock(&sdp->sd_statfs_spin);
611 		goto out_bh;
612 	}
613 	spin_unlock(&sdp->sd_statfs_spin);
614 
615 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
616 	if (error)
617 		goto out_bh;
618 
619 	error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
620 	if (error)
621 		goto out_bh2;
622 
623 	update_statfs(sdp, m_bh, l_bh);
624 	sdp->sd_statfs_force_sync = 0;
625 
626 	gfs2_trans_end(sdp);
627 
628 out_bh2:
629 	brelse(l_bh);
630 out_bh:
631 	brelse(m_bh);
632 out_unlock:
633 	gfs2_glock_dq_uninit(&gh);
634 out:
635 	sb_end_write(sb);
636 	return error;
637 }
638 
639 struct lfcc {
640 	struct list_head list;
641 	struct gfs2_holder gh;
642 };
643 
644 /**
645  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
646  *                            journals are clean
647  * @sdp: the file system
648  * @state: the state to put the transaction lock into
649  * @t_gh: the hold on the transaction lock
650  *
651  * Returns: errno
652  */
653 
654 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
655 				    struct gfs2_holder *freeze_gh)
656 {
657 	struct gfs2_inode *ip;
658 	struct gfs2_jdesc *jd;
659 	struct lfcc *lfcc;
660 	LIST_HEAD(list);
661 	struct gfs2_log_header_host lh;
662 	int error;
663 
664 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
665 		lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
666 		if (!lfcc) {
667 			error = -ENOMEM;
668 			goto out;
669 		}
670 		ip = GFS2_I(jd->jd_inode);
671 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
672 		if (error) {
673 			kfree(lfcc);
674 			goto out;
675 		}
676 		list_add(&lfcc->list, &list);
677 	}
678 
679 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
680 				   GL_NOCACHE, freeze_gh);
681 
682 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
683 		error = gfs2_jdesc_check(jd);
684 		if (error)
685 			break;
686 		error = gfs2_find_jhead(jd, &lh, false);
687 		if (error)
688 			break;
689 		if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
690 			error = -EBUSY;
691 			break;
692 		}
693 	}
694 
695 	if (error)
696 		gfs2_glock_dq_uninit(freeze_gh);
697 
698 out:
699 	while (!list_empty(&list)) {
700 		lfcc = list_entry(list.next, struct lfcc, list);
701 		list_del(&lfcc->list);
702 		gfs2_glock_dq_uninit(&lfcc->gh);
703 		kfree(lfcc);
704 	}
705 	return error;
706 }
707 
708 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
709 {
710 	struct gfs2_dinode *str = buf;
711 
712 	str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
713 	str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
714 	str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
715 	str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
716 	str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
717 	str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
718 	str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
719 	str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
720 	str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
721 	str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
722 	str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
723 	str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
724 	str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
725 	str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
726 
727 	str->di_goal_meta = cpu_to_be64(ip->i_goal);
728 	str->di_goal_data = cpu_to_be64(ip->i_goal);
729 	str->di_generation = cpu_to_be64(ip->i_generation);
730 
731 	str->di_flags = cpu_to_be32(ip->i_diskflags);
732 	str->di_height = cpu_to_be16(ip->i_height);
733 	str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
734 					     !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
735 					     GFS2_FORMAT_DE : 0);
736 	str->di_depth = cpu_to_be16(ip->i_depth);
737 	str->di_entries = cpu_to_be32(ip->i_entries);
738 
739 	str->di_eattr = cpu_to_be64(ip->i_eattr);
740 	str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
741 	str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
742 	str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
743 }
744 
745 /**
746  * gfs2_write_inode - Make sure the inode is stable on the disk
747  * @inode: The inode
748  * @wbc: The writeback control structure
749  *
750  * Returns: errno
751  */
752 
753 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
754 {
755 	struct gfs2_inode *ip = GFS2_I(inode);
756 	struct gfs2_sbd *sdp = GFS2_SB(inode);
757 	struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
758 	struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
759 	int ret = 0;
760 	bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
761 
762 	if (flush_all)
763 		gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
764 			       GFS2_LOG_HEAD_FLUSH_NORMAL |
765 			       GFS2_LFC_WRITE_INODE);
766 	if (bdi->wb.dirty_exceeded)
767 		gfs2_ail1_flush(sdp, wbc);
768 	else
769 		filemap_fdatawrite(metamapping);
770 	if (flush_all)
771 		ret = filemap_fdatawait(metamapping);
772 	if (ret)
773 		mark_inode_dirty_sync(inode);
774 	else {
775 		spin_lock(&inode->i_lock);
776 		if (!(inode->i_flags & I_DIRTY))
777 			gfs2_ordered_del_inode(ip);
778 		spin_unlock(&inode->i_lock);
779 	}
780 	return ret;
781 }
782 
783 /**
784  * gfs2_dirty_inode - check for atime updates
785  * @inode: The inode in question
786  * @flags: The type of dirty
787  *
788  * Unfortunately it can be called under any combination of inode
789  * glock and transaction lock, so we have to check carefully.
790  *
791  * At the moment this deals only with atime - it should be possible
792  * to expand that role in future, once a review of the locking has
793  * been carried out.
794  */
795 
796 static void gfs2_dirty_inode(struct inode *inode, int flags)
797 {
798 	struct gfs2_inode *ip = GFS2_I(inode);
799 	struct gfs2_sbd *sdp = GFS2_SB(inode);
800 	struct buffer_head *bh;
801 	struct gfs2_holder gh;
802 	int need_unlock = 0;
803 	int need_endtrans = 0;
804 	int ret;
805 
806 	if (!(flags & I_DIRTY_INODE))
807 		return;
808 	if (unlikely(test_bit(SDF_WITHDRAWN, &sdp->sd_flags)))
809 		return;
810 	if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
811 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
812 		if (ret) {
813 			fs_err(sdp, "dirty_inode: glock %d\n", ret);
814 			return;
815 		}
816 		need_unlock = 1;
817 	} else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
818 		return;
819 
820 	if (current->journal_info == NULL) {
821 		ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
822 		if (ret) {
823 			fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
824 			goto out;
825 		}
826 		need_endtrans = 1;
827 	}
828 
829 	ret = gfs2_meta_inode_buffer(ip, &bh);
830 	if (ret == 0) {
831 		gfs2_trans_add_meta(ip->i_gl, bh);
832 		gfs2_dinode_out(ip, bh->b_data);
833 		brelse(bh);
834 	}
835 
836 	if (need_endtrans)
837 		gfs2_trans_end(sdp);
838 out:
839 	if (need_unlock)
840 		gfs2_glock_dq_uninit(&gh);
841 }
842 
843 /**
844  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
845  * @sdp: the filesystem
846  *
847  * Returns: errno
848  */
849 
850 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
851 {
852 	struct gfs2_holder freeze_gh;
853 	int error;
854 
855 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, GL_NOCACHE,
856 				   &freeze_gh);
857 	if (error && !test_bit(SDF_WITHDRAWN, &sdp->sd_flags))
858 		return error;
859 
860 	flush_workqueue(gfs2_delete_workqueue);
861 	if (sdp->sd_quotad_process)
862 		kthread_stop(sdp->sd_quotad_process);
863 	sdp->sd_quotad_process = NULL;
864 	if (sdp->sd_logd_process)
865 		kthread_stop(sdp->sd_logd_process);
866 	sdp->sd_logd_process = NULL;
867 
868 	gfs2_quota_sync(sdp->sd_vfs, 0);
869 	gfs2_statfs_sync(sdp->sd_vfs, 0);
870 
871 	gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
872 		       GFS2_LFC_MAKE_FS_RO);
873 	wait_event(sdp->sd_reserving_log_wait, atomic_read(&sdp->sd_reserving_log) == 0);
874 	gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
875 
876 	if (gfs2_holder_initialized(&freeze_gh))
877 		gfs2_glock_dq_uninit(&freeze_gh);
878 
879 	gfs2_quota_cleanup(sdp);
880 
881 	return error;
882 }
883 
884 /**
885  * gfs2_put_super - Unmount the filesystem
886  * @sb: The VFS superblock
887  *
888  */
889 
890 static void gfs2_put_super(struct super_block *sb)
891 {
892 	struct gfs2_sbd *sdp = sb->s_fs_info;
893 	int error;
894 	struct gfs2_jdesc *jd;
895 
896 	/* No more recovery requests */
897 	set_bit(SDF_NORECOVERY, &sdp->sd_flags);
898 	smp_mb();
899 
900 	/* Wait on outstanding recovery */
901 restart:
902 	spin_lock(&sdp->sd_jindex_spin);
903 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
904 		if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
905 			continue;
906 		spin_unlock(&sdp->sd_jindex_spin);
907 		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
908 			    TASK_UNINTERRUPTIBLE);
909 		goto restart;
910 	}
911 	spin_unlock(&sdp->sd_jindex_spin);
912 
913 	if (!sb_rdonly(sb)) {
914 		error = gfs2_make_fs_ro(sdp);
915 		if (error)
916 			gfs2_io_error(sdp);
917 	}
918 	/*  At this point, we're through modifying the disk  */
919 
920 	/*  Release stuff  */
921 
922 	iput(sdp->sd_jindex);
923 	iput(sdp->sd_statfs_inode);
924 	iput(sdp->sd_rindex);
925 	iput(sdp->sd_quota_inode);
926 
927 	gfs2_glock_put(sdp->sd_rename_gl);
928 	gfs2_glock_put(sdp->sd_freeze_gl);
929 
930 	if (!sdp->sd_args.ar_spectator) {
931 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
932 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
933 		gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
934 		gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
935 		iput(sdp->sd_sc_inode);
936 		iput(sdp->sd_qc_inode);
937 	}
938 
939 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
940 	gfs2_clear_rgrpd(sdp);
941 	gfs2_jindex_free(sdp);
942 	/*  Take apart glock structures and buffer lists  */
943 	gfs2_gl_hash_clear(sdp);
944 	gfs2_delete_debugfs_file(sdp);
945 	/*  Unmount the locking protocol  */
946 	gfs2_lm_unmount(sdp);
947 
948 	/*  At this point, we're through participating in the lockspace  */
949 	gfs2_sys_fs_del(sdp);
950 }
951 
952 /**
953  * gfs2_sync_fs - sync the filesystem
954  * @sb: the superblock
955  *
956  * Flushes the log to disk.
957  */
958 
959 static int gfs2_sync_fs(struct super_block *sb, int wait)
960 {
961 	struct gfs2_sbd *sdp = sb->s_fs_info;
962 
963 	gfs2_quota_sync(sb, -1);
964 	if (wait)
965 		gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
966 			       GFS2_LFC_SYNC_FS);
967 	return sdp->sd_log_error;
968 }
969 
970 void gfs2_freeze_func(struct work_struct *work)
971 {
972 	int error;
973 	struct gfs2_holder freeze_gh;
974 	struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
975 	struct super_block *sb = sdp->sd_vfs;
976 
977 	atomic_inc(&sb->s_active);
978 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
979 				   &freeze_gh);
980 	if (error) {
981 		fs_info(sdp, "GFS2: couldn't get freeze lock : %d\n", error);
982 		gfs2_assert_withdraw(sdp, 0);
983 	} else {
984 		atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
985 		error = thaw_super(sb);
986 		if (error) {
987 			fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
988 				error);
989 			gfs2_assert_withdraw(sdp, 0);
990 		}
991 		if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
992 			freeze_gh.gh_flags |= GL_NOCACHE;
993 		gfs2_glock_dq_uninit(&freeze_gh);
994 	}
995 	deactivate_super(sb);
996 	clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
997 	wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
998 	return;
999 }
1000 
1001 /**
1002  * gfs2_freeze - prevent further writes to the filesystem
1003  * @sb: the VFS structure for the filesystem
1004  *
1005  */
1006 
1007 static int gfs2_freeze(struct super_block *sb)
1008 {
1009 	struct gfs2_sbd *sdp = sb->s_fs_info;
1010 	int error = 0;
1011 
1012 	mutex_lock(&sdp->sd_freeze_mutex);
1013 	if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN)
1014 		goto out;
1015 
1016 	if (test_bit(SDF_WITHDRAWN, &sdp->sd_flags)) {
1017 		error = -EINVAL;
1018 		goto out;
1019 	}
1020 
1021 	for (;;) {
1022 		error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
1023 		if (!error)
1024 			break;
1025 
1026 		if (error == -EBUSY)
1027 			fs_err(sdp, "waiting for recovery before freeze\n");
1028 		else
1029 			fs_err(sdp, "error freezing FS: %d\n", error);
1030 
1031 		fs_err(sdp, "retrying...\n");
1032 		msleep(1000);
1033 	}
1034 	set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
1035 out:
1036 	mutex_unlock(&sdp->sd_freeze_mutex);
1037 	return error;
1038 }
1039 
1040 /**
1041  * gfs2_unfreeze - reallow writes to the filesystem
1042  * @sb: the VFS structure for the filesystem
1043  *
1044  */
1045 
1046 static int gfs2_unfreeze(struct super_block *sb)
1047 {
1048 	struct gfs2_sbd *sdp = sb->s_fs_info;
1049 
1050 	mutex_lock(&sdp->sd_freeze_mutex);
1051         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
1052 	    !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
1053 		mutex_unlock(&sdp->sd_freeze_mutex);
1054                 return 0;
1055 	}
1056 
1057 	gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1058 	mutex_unlock(&sdp->sd_freeze_mutex);
1059 	return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
1060 }
1061 
1062 /**
1063  * statfs_fill - fill in the sg for a given RG
1064  * @rgd: the RG
1065  * @sc: the sc structure
1066  *
1067  * Returns: 0 on success, -ESTALE if the LVB is invalid
1068  */
1069 
1070 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1071 			    struct gfs2_statfs_change_host *sc)
1072 {
1073 	gfs2_rgrp_verify(rgd);
1074 	sc->sc_total += rgd->rd_data;
1075 	sc->sc_free += rgd->rd_free;
1076 	sc->sc_dinodes += rgd->rd_dinodes;
1077 	return 0;
1078 }
1079 
1080 /**
1081  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1082  * @sdp: the filesystem
1083  * @sc: the sc info that will be returned
1084  *
1085  * Any error (other than a signal) will cause this routine to fall back
1086  * to the synchronous version.
1087  *
1088  * FIXME: This really shouldn't busy wait like this.
1089  *
1090  * Returns: errno
1091  */
1092 
1093 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1094 {
1095 	struct gfs2_rgrpd *rgd_next;
1096 	struct gfs2_holder *gha, *gh;
1097 	unsigned int slots = 64;
1098 	unsigned int x;
1099 	int done;
1100 	int error = 0, err;
1101 
1102 	memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1103 	gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1104 	if (!gha)
1105 		return -ENOMEM;
1106 	for (x = 0; x < slots; x++)
1107 		gfs2_holder_mark_uninitialized(gha + x);
1108 
1109 	rgd_next = gfs2_rgrpd_get_first(sdp);
1110 
1111 	for (;;) {
1112 		done = 1;
1113 
1114 		for (x = 0; x < slots; x++) {
1115 			gh = gha + x;
1116 
1117 			if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
1118 				err = gfs2_glock_wait(gh);
1119 				if (err) {
1120 					gfs2_holder_uninit(gh);
1121 					error = err;
1122 				} else {
1123 					if (!error) {
1124 						struct gfs2_rgrpd *rgd =
1125 							gfs2_glock2rgrp(gh->gh_gl);
1126 
1127 						error = statfs_slow_fill(rgd, sc);
1128 					}
1129 					gfs2_glock_dq_uninit(gh);
1130 				}
1131 			}
1132 
1133 			if (gfs2_holder_initialized(gh))
1134 				done = 0;
1135 			else if (rgd_next && !error) {
1136 				error = gfs2_glock_nq_init(rgd_next->rd_gl,
1137 							   LM_ST_SHARED,
1138 							   GL_ASYNC,
1139 							   gh);
1140 				rgd_next = gfs2_rgrpd_get_next(rgd_next);
1141 				done = 0;
1142 			}
1143 
1144 			if (signal_pending(current))
1145 				error = -ERESTARTSYS;
1146 		}
1147 
1148 		if (done)
1149 			break;
1150 
1151 		yield();
1152 	}
1153 
1154 	kfree(gha);
1155 	return error;
1156 }
1157 
1158 /**
1159  * gfs2_statfs_i - Do a statfs
1160  * @sdp: the filesystem
1161  * @sg: the sg structure
1162  *
1163  * Returns: errno
1164  */
1165 
1166 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1167 {
1168 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1169 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1170 
1171 	spin_lock(&sdp->sd_statfs_spin);
1172 
1173 	*sc = *m_sc;
1174 	sc->sc_total += l_sc->sc_total;
1175 	sc->sc_free += l_sc->sc_free;
1176 	sc->sc_dinodes += l_sc->sc_dinodes;
1177 
1178 	spin_unlock(&sdp->sd_statfs_spin);
1179 
1180 	if (sc->sc_free < 0)
1181 		sc->sc_free = 0;
1182 	if (sc->sc_free > sc->sc_total)
1183 		sc->sc_free = sc->sc_total;
1184 	if (sc->sc_dinodes < 0)
1185 		sc->sc_dinodes = 0;
1186 
1187 	return 0;
1188 }
1189 
1190 /**
1191  * gfs2_statfs - Gather and return stats about the filesystem
1192  * @sb: The superblock
1193  * @statfsbuf: The buffer
1194  *
1195  * Returns: 0 on success or error code
1196  */
1197 
1198 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1199 {
1200 	struct super_block *sb = dentry->d_sb;
1201 	struct gfs2_sbd *sdp = sb->s_fs_info;
1202 	struct gfs2_statfs_change_host sc;
1203 	int error;
1204 
1205 	error = gfs2_rindex_update(sdp);
1206 	if (error)
1207 		return error;
1208 
1209 	if (gfs2_tune_get(sdp, gt_statfs_slow))
1210 		error = gfs2_statfs_slow(sdp, &sc);
1211 	else
1212 		error = gfs2_statfs_i(sdp, &sc);
1213 
1214 	if (error)
1215 		return error;
1216 
1217 	buf->f_type = GFS2_MAGIC;
1218 	buf->f_bsize = sdp->sd_sb.sb_bsize;
1219 	buf->f_blocks = sc.sc_total;
1220 	buf->f_bfree = sc.sc_free;
1221 	buf->f_bavail = sc.sc_free;
1222 	buf->f_files = sc.sc_dinodes + sc.sc_free;
1223 	buf->f_ffree = sc.sc_free;
1224 	buf->f_namelen = GFS2_FNAMESIZE;
1225 
1226 	return 0;
1227 }
1228 
1229 /**
1230  * gfs2_remount_fs - called when the FS is remounted
1231  * @sb:  the filesystem
1232  * @flags:  the remount flags
1233  * @data:  extra data passed in (not used right now)
1234  *
1235  * Returns: errno
1236  */
1237 
1238 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1239 {
1240 	struct gfs2_sbd *sdp = sb->s_fs_info;
1241 	struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1242 	struct gfs2_tune *gt = &sdp->sd_tune;
1243 	int error;
1244 
1245 	sync_filesystem(sb);
1246 
1247 	spin_lock(&gt->gt_spin);
1248 	args.ar_commit = gt->gt_logd_secs;
1249 	args.ar_quota_quantum = gt->gt_quota_quantum;
1250 	if (gt->gt_statfs_slow)
1251 		args.ar_statfs_quantum = 0;
1252 	else
1253 		args.ar_statfs_quantum = gt->gt_statfs_quantum;
1254 	spin_unlock(&gt->gt_spin);
1255 	error = gfs2_mount_args(&args, data);
1256 	if (error)
1257 		return error;
1258 
1259 	/* Not allowed to change locking details */
1260 	if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1261 	    strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1262 	    strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1263 		return -EINVAL;
1264 
1265 	/* Some flags must not be changed */
1266 	if (args_neq(&args, &sdp->sd_args, spectator) ||
1267 	    args_neq(&args, &sdp->sd_args, localflocks) ||
1268 	    args_neq(&args, &sdp->sd_args, meta))
1269 		return -EINVAL;
1270 
1271 	if (sdp->sd_args.ar_spectator)
1272 		*flags |= SB_RDONLY;
1273 
1274 	if ((sb->s_flags ^ *flags) & SB_RDONLY) {
1275 		if (*flags & SB_RDONLY)
1276 			error = gfs2_make_fs_ro(sdp);
1277 		else
1278 			error = gfs2_make_fs_rw(sdp);
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 error;
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 	if (S_ISDIR(inode->i_mode) &&
1635 	    (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1636 		error = gfs2_dir_exhash_dealloc(ip);
1637 		if (error)
1638 			goto out_unlock;
1639 	}
1640 
1641 	if (ip->i_eattr) {
1642 		error = gfs2_ea_dealloc(ip);
1643 		if (error)
1644 			goto out_unlock;
1645 	}
1646 
1647 	if (!gfs2_is_stuffed(ip)) {
1648 		error = gfs2_file_dealloc(ip);
1649 		if (error)
1650 			goto out_unlock;
1651 	}
1652 
1653 	/* We're about to clear the bitmap for the dinode, but as soon as we
1654 	   do, gfs2_create_inode can create another inode at the same block
1655 	   location and try to set gl_object again. We clear gl_object here so
1656 	   that subsequent inode creates don't see an old gl_object. */
1657 	glock_clear_object(ip->i_gl, ip);
1658 	error = gfs2_dinode_dealloc(ip);
1659 	goto out_unlock;
1660 
1661 out_truncate:
1662 	gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1663 		       GFS2_LFC_EVICT_INODE);
1664 	metamapping = gfs2_glock2aspace(ip->i_gl);
1665 	if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1666 		filemap_fdatawrite(metamapping);
1667 		filemap_fdatawait(metamapping);
1668 	}
1669 	write_inode_now(inode, 1);
1670 	gfs2_ail_flush(ip->i_gl, 0);
1671 
1672 	error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1673 	if (error)
1674 		goto out_unlock;
1675 	/* Needs to be done before glock release & also in a transaction */
1676 	truncate_inode_pages(&inode->i_data, 0);
1677 	truncate_inode_pages(metamapping, 0);
1678 	gfs2_trans_end(sdp);
1679 
1680 out_unlock:
1681 	if (gfs2_rs_active(&ip->i_res))
1682 		gfs2_rs_deltree(&ip->i_res);
1683 
1684 	if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1685 		glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1686 		if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1687 			ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1688 			gfs2_glock_dq(&ip->i_iopen_gh);
1689 		}
1690 		gfs2_holder_uninit(&ip->i_iopen_gh);
1691 	}
1692 	if (gfs2_holder_initialized(&gh)) {
1693 		glock_clear_object(ip->i_gl, ip);
1694 		gfs2_glock_dq_uninit(&gh);
1695 	}
1696 	if (error && error != GLR_TRYFAILED && error != -EROFS)
1697 		fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1698 out:
1699 	truncate_inode_pages_final(&inode->i_data);
1700 	gfs2_rsqa_delete(ip, NULL);
1701 	gfs2_ordered_del_inode(ip);
1702 	clear_inode(inode);
1703 	gfs2_dir_hash_inval(ip);
1704 	glock_clear_object(ip->i_gl, ip);
1705 	wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1706 	gfs2_glock_add_to_lru(ip->i_gl);
1707 	gfs2_glock_put_eventually(ip->i_gl);
1708 	ip->i_gl = NULL;
1709 	if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1710 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1711 
1712 		glock_clear_object(gl, ip);
1713 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1714 		gfs2_glock_hold(gl);
1715 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1716 		gfs2_glock_put_eventually(gl);
1717 	}
1718 }
1719 
1720 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1721 {
1722 	struct gfs2_inode *ip;
1723 
1724 	ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1725 	if (ip) {
1726 		ip->i_flags = 0;
1727 		ip->i_gl = NULL;
1728 		memset(&ip->i_res, 0, sizeof(ip->i_res));
1729 		RB_CLEAR_NODE(&ip->i_res.rs_node);
1730 		ip->i_rahead = 0;
1731 	}
1732 	return &ip->i_inode;
1733 }
1734 
1735 static void gfs2_free_inode(struct inode *inode)
1736 {
1737 	kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1738 }
1739 
1740 const struct super_operations gfs2_super_ops = {
1741 	.alloc_inode		= gfs2_alloc_inode,
1742 	.free_inode		= gfs2_free_inode,
1743 	.write_inode		= gfs2_write_inode,
1744 	.dirty_inode		= gfs2_dirty_inode,
1745 	.evict_inode		= gfs2_evict_inode,
1746 	.put_super		= gfs2_put_super,
1747 	.sync_fs		= gfs2_sync_fs,
1748 	.freeze_super		= gfs2_freeze,
1749 	.thaw_super		= gfs2_unfreeze,
1750 	.statfs			= gfs2_statfs,
1751 	.remount_fs		= gfs2_remount_fs,
1752 	.drop_inode		= gfs2_drop_inode,
1753 	.show_options		= gfs2_show_options,
1754 };
1755 
1756