xref: /openbmc/linux/fs/gfs2/super.c (revision 6dfcd296)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9 
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 
12 #include <linux/bio.h>
13 #include <linux/sched.h>
14 #include <linux/slab.h>
15 #include <linux/spinlock.h>
16 #include <linux/completion.h>
17 #include <linux/buffer_head.h>
18 #include <linux/statfs.h>
19 #include <linux/seq_file.h>
20 #include <linux/mount.h>
21 #include <linux/kthread.h>
22 #include <linux/delay.h>
23 #include <linux/gfs2_ondisk.h>
24 #include <linux/crc32.h>
25 #include <linux/time.h>
26 #include <linux/wait.h>
27 #include <linux/writeback.h>
28 #include <linux/backing-dev.h>
29 #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 
49 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
50 
51 enum {
52 	Opt_lockproto,
53 	Opt_locktable,
54 	Opt_hostdata,
55 	Opt_spectator,
56 	Opt_ignore_local_fs,
57 	Opt_localflocks,
58 	Opt_localcaching,
59 	Opt_debug,
60 	Opt_nodebug,
61 	Opt_upgrade,
62 	Opt_acl,
63 	Opt_noacl,
64 	Opt_quota_off,
65 	Opt_quota_account,
66 	Opt_quota_on,
67 	Opt_quota,
68 	Opt_noquota,
69 	Opt_suiddir,
70 	Opt_nosuiddir,
71 	Opt_data_writeback,
72 	Opt_data_ordered,
73 	Opt_meta,
74 	Opt_discard,
75 	Opt_nodiscard,
76 	Opt_commit,
77 	Opt_err_withdraw,
78 	Opt_err_panic,
79 	Opt_statfs_quantum,
80 	Opt_statfs_percent,
81 	Opt_quota_quantum,
82 	Opt_barrier,
83 	Opt_nobarrier,
84 	Opt_rgrplvb,
85 	Opt_norgrplvb,
86 	Opt_loccookie,
87 	Opt_noloccookie,
88 	Opt_error,
89 };
90 
91 static const match_table_t tokens = {
92 	{Opt_lockproto, "lockproto=%s"},
93 	{Opt_locktable, "locktable=%s"},
94 	{Opt_hostdata, "hostdata=%s"},
95 	{Opt_spectator, "spectator"},
96 	{Opt_spectator, "norecovery"},
97 	{Opt_ignore_local_fs, "ignore_local_fs"},
98 	{Opt_localflocks, "localflocks"},
99 	{Opt_localcaching, "localcaching"},
100 	{Opt_debug, "debug"},
101 	{Opt_nodebug, "nodebug"},
102 	{Opt_upgrade, "upgrade"},
103 	{Opt_acl, "acl"},
104 	{Opt_noacl, "noacl"},
105 	{Opt_quota_off, "quota=off"},
106 	{Opt_quota_account, "quota=account"},
107 	{Opt_quota_on, "quota=on"},
108 	{Opt_quota, "quota"},
109 	{Opt_noquota, "noquota"},
110 	{Opt_suiddir, "suiddir"},
111 	{Opt_nosuiddir, "nosuiddir"},
112 	{Opt_data_writeback, "data=writeback"},
113 	{Opt_data_ordered, "data=ordered"},
114 	{Opt_meta, "meta"},
115 	{Opt_discard, "discard"},
116 	{Opt_nodiscard, "nodiscard"},
117 	{Opt_commit, "commit=%d"},
118 	{Opt_err_withdraw, "errors=withdraw"},
119 	{Opt_err_panic, "errors=panic"},
120 	{Opt_statfs_quantum, "statfs_quantum=%d"},
121 	{Opt_statfs_percent, "statfs_percent=%d"},
122 	{Opt_quota_quantum, "quota_quantum=%d"},
123 	{Opt_barrier, "barrier"},
124 	{Opt_nobarrier, "nobarrier"},
125 	{Opt_rgrplvb, "rgrplvb"},
126 	{Opt_norgrplvb, "norgrplvb"},
127 	{Opt_loccookie, "loccookie"},
128 	{Opt_noloccookie, "noloccookie"},
129 	{Opt_error, NULL}
130 };
131 
132 /**
133  * gfs2_mount_args - Parse mount options
134  * @args: The structure into which the parsed options will be written
135  * @options: The options to parse
136  *
137  * Return: errno
138  */
139 
140 int gfs2_mount_args(struct gfs2_args *args, char *options)
141 {
142 	char *o;
143 	int token;
144 	substring_t tmp[MAX_OPT_ARGS];
145 	int rv;
146 
147 	/* Split the options into tokens with the "," character and
148 	   process them */
149 
150 	while (1) {
151 		o = strsep(&options, ",");
152 		if (o == NULL)
153 			break;
154 		if (*o == '\0')
155 			continue;
156 
157 		token = match_token(o, tokens, tmp);
158 		switch (token) {
159 		case Opt_lockproto:
160 			match_strlcpy(args->ar_lockproto, &tmp[0],
161 				      GFS2_LOCKNAME_LEN);
162 			break;
163 		case Opt_locktable:
164 			match_strlcpy(args->ar_locktable, &tmp[0],
165 				      GFS2_LOCKNAME_LEN);
166 			break;
167 		case Opt_hostdata:
168 			match_strlcpy(args->ar_hostdata, &tmp[0],
169 				      GFS2_LOCKNAME_LEN);
170 			break;
171 		case Opt_spectator:
172 			args->ar_spectator = 1;
173 			break;
174 		case Opt_ignore_local_fs:
175 			/* Retained for backwards compat only */
176 			break;
177 		case Opt_localflocks:
178 			args->ar_localflocks = 1;
179 			break;
180 		case Opt_localcaching:
181 			/* Retained for backwards compat only */
182 			break;
183 		case Opt_debug:
184 			if (args->ar_errors == GFS2_ERRORS_PANIC) {
185 				pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
186 				return -EINVAL;
187 			}
188 			args->ar_debug = 1;
189 			break;
190 		case Opt_nodebug:
191 			args->ar_debug = 0;
192 			break;
193 		case Opt_upgrade:
194 			/* Retained for backwards compat only */
195 			break;
196 		case Opt_acl:
197 			args->ar_posix_acl = 1;
198 			break;
199 		case Opt_noacl:
200 			args->ar_posix_acl = 0;
201 			break;
202 		case Opt_quota_off:
203 		case Opt_noquota:
204 			args->ar_quota = GFS2_QUOTA_OFF;
205 			break;
206 		case Opt_quota_account:
207 			args->ar_quota = GFS2_QUOTA_ACCOUNT;
208 			break;
209 		case Opt_quota_on:
210 		case Opt_quota:
211 			args->ar_quota = GFS2_QUOTA_ON;
212 			break;
213 		case Opt_suiddir:
214 			args->ar_suiddir = 1;
215 			break;
216 		case Opt_nosuiddir:
217 			args->ar_suiddir = 0;
218 			break;
219 		case Opt_data_writeback:
220 			args->ar_data = GFS2_DATA_WRITEBACK;
221 			break;
222 		case Opt_data_ordered:
223 			args->ar_data = GFS2_DATA_ORDERED;
224 			break;
225 		case Opt_meta:
226 			args->ar_meta = 1;
227 			break;
228 		case Opt_discard:
229 			args->ar_discard = 1;
230 			break;
231 		case Opt_nodiscard:
232 			args->ar_discard = 0;
233 			break;
234 		case Opt_commit:
235 			rv = match_int(&tmp[0], &args->ar_commit);
236 			if (rv || args->ar_commit <= 0) {
237 				pr_warn("commit mount option requires a positive numeric argument\n");
238 				return rv ? rv : -EINVAL;
239 			}
240 			break;
241 		case Opt_statfs_quantum:
242 			rv = match_int(&tmp[0], &args->ar_statfs_quantum);
243 			if (rv || args->ar_statfs_quantum < 0) {
244 				pr_warn("statfs_quantum mount option requires a non-negative numeric argument\n");
245 				return rv ? rv : -EINVAL;
246 			}
247 			break;
248 		case Opt_quota_quantum:
249 			rv = match_int(&tmp[0], &args->ar_quota_quantum);
250 			if (rv || args->ar_quota_quantum <= 0) {
251 				pr_warn("quota_quantum mount option requires a positive numeric argument\n");
252 				return rv ? rv : -EINVAL;
253 			}
254 			break;
255 		case Opt_statfs_percent:
256 			rv = match_int(&tmp[0], &args->ar_statfs_percent);
257 			if (rv || args->ar_statfs_percent < 0 ||
258 			    args->ar_statfs_percent > 100) {
259 				pr_warn("statfs_percent mount option requires a numeric argument between 0 and 100\n");
260 				return rv ? rv : -EINVAL;
261 			}
262 			break;
263 		case Opt_err_withdraw:
264 			args->ar_errors = GFS2_ERRORS_WITHDRAW;
265 			break;
266 		case Opt_err_panic:
267 			if (args->ar_debug) {
268 				pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
269 				return -EINVAL;
270 			}
271 			args->ar_errors = GFS2_ERRORS_PANIC;
272 			break;
273 		case Opt_barrier:
274 			args->ar_nobarrier = 0;
275 			break;
276 		case Opt_nobarrier:
277 			args->ar_nobarrier = 1;
278 			break;
279 		case Opt_rgrplvb:
280 			args->ar_rgrplvb = 1;
281 			break;
282 		case Opt_norgrplvb:
283 			args->ar_rgrplvb = 0;
284 			break;
285 		case Opt_loccookie:
286 			args->ar_loccookie = 1;
287 			break;
288 		case Opt_noloccookie:
289 			args->ar_loccookie = 0;
290 			break;
291 		case Opt_error:
292 		default:
293 			pr_warn("invalid mount option: %s\n", o);
294 			return -EINVAL;
295 		}
296 	}
297 
298 	return 0;
299 }
300 
301 /**
302  * gfs2_jindex_free - Clear all the journal index information
303  * @sdp: The GFS2 superblock
304  *
305  */
306 
307 void gfs2_jindex_free(struct gfs2_sbd *sdp)
308 {
309 	struct list_head list;
310 	struct gfs2_jdesc *jd;
311 
312 	spin_lock(&sdp->sd_jindex_spin);
313 	list_add(&list, &sdp->sd_jindex_list);
314 	list_del_init(&sdp->sd_jindex_list);
315 	sdp->sd_journals = 0;
316 	spin_unlock(&sdp->sd_jindex_spin);
317 
318 	while (!list_empty(&list)) {
319 		jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
320 		gfs2_free_journal_extents(jd);
321 		list_del(&jd->jd_list);
322 		iput(jd->jd_inode);
323 		kfree(jd);
324 	}
325 }
326 
327 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
328 {
329 	struct gfs2_jdesc *jd;
330 	int found = 0;
331 
332 	list_for_each_entry(jd, head, jd_list) {
333 		if (jd->jd_jid == jid) {
334 			found = 1;
335 			break;
336 		}
337 	}
338 
339 	if (!found)
340 		jd = NULL;
341 
342 	return jd;
343 }
344 
345 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
346 {
347 	struct gfs2_jdesc *jd;
348 
349 	spin_lock(&sdp->sd_jindex_spin);
350 	jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
351 	spin_unlock(&sdp->sd_jindex_spin);
352 
353 	return jd;
354 }
355 
356 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
357 {
358 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
359 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
360 	u64 size = i_size_read(jd->jd_inode);
361 
362 	if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
363 		return -EIO;
364 
365 	jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
366 
367 	if (gfs2_write_alloc_required(ip, 0, size)) {
368 		gfs2_consist_inode(ip);
369 		return -EIO;
370 	}
371 
372 	return 0;
373 }
374 
375 static int init_threads(struct gfs2_sbd *sdp)
376 {
377 	struct task_struct *p;
378 	int error = 0;
379 
380 	p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
381 	if (IS_ERR(p)) {
382 		error = PTR_ERR(p);
383 		fs_err(sdp, "can't start logd thread: %d\n", error);
384 		return error;
385 	}
386 	sdp->sd_logd_process = p;
387 
388 	p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
389 	if (IS_ERR(p)) {
390 		error = PTR_ERR(p);
391 		fs_err(sdp, "can't start quotad thread: %d\n", error);
392 		goto fail;
393 	}
394 	sdp->sd_quotad_process = p;
395 	return 0;
396 
397 fail:
398 	kthread_stop(sdp->sd_logd_process);
399 	return error;
400 }
401 
402 /**
403  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
404  * @sdp: the filesystem
405  *
406  * Returns: errno
407  */
408 
409 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
410 {
411 	struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
412 	struct gfs2_glock *j_gl = ip->i_gl;
413 	struct gfs2_holder freeze_gh;
414 	struct gfs2_log_header_host head;
415 	int error;
416 
417 	error = init_threads(sdp);
418 	if (error)
419 		return error;
420 
421 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
422 				   &freeze_gh);
423 	if (error)
424 		goto fail_threads;
425 
426 	j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
427 
428 	error = gfs2_find_jhead(sdp->sd_jdesc, &head);
429 	if (error)
430 		goto fail;
431 
432 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
433 		gfs2_consist(sdp);
434 		error = -EIO;
435 		goto fail;
436 	}
437 
438 	/*  Initialize some head of the log stuff  */
439 	sdp->sd_log_sequence = head.lh_sequence + 1;
440 	gfs2_log_pointers_init(sdp, head.lh_blkno);
441 
442 	error = gfs2_quota_init(sdp);
443 	if (error)
444 		goto fail;
445 
446 	set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
447 
448 	gfs2_glock_dq_uninit(&freeze_gh);
449 
450 	return 0;
451 
452 fail:
453 	freeze_gh.gh_flags |= GL_NOCACHE;
454 	gfs2_glock_dq_uninit(&freeze_gh);
455 fail_threads:
456 	kthread_stop(sdp->sd_quotad_process);
457 	kthread_stop(sdp->sd_logd_process);
458 	return error;
459 }
460 
461 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
462 {
463 	const struct gfs2_statfs_change *str = buf;
464 
465 	sc->sc_total = be64_to_cpu(str->sc_total);
466 	sc->sc_free = be64_to_cpu(str->sc_free);
467 	sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
468 }
469 
470 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
471 {
472 	struct gfs2_statfs_change *str = buf;
473 
474 	str->sc_total = cpu_to_be64(sc->sc_total);
475 	str->sc_free = cpu_to_be64(sc->sc_free);
476 	str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
477 }
478 
479 int gfs2_statfs_init(struct gfs2_sbd *sdp)
480 {
481 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
482 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
483 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
484 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
485 	struct buffer_head *m_bh, *l_bh;
486 	struct gfs2_holder gh;
487 	int error;
488 
489 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
490 				   &gh);
491 	if (error)
492 		return error;
493 
494 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
495 	if (error)
496 		goto out;
497 
498 	if (sdp->sd_args.ar_spectator) {
499 		spin_lock(&sdp->sd_statfs_spin);
500 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
501 				      sizeof(struct gfs2_dinode));
502 		spin_unlock(&sdp->sd_statfs_spin);
503 	} else {
504 		error = gfs2_meta_inode_buffer(l_ip, &l_bh);
505 		if (error)
506 			goto out_m_bh;
507 
508 		spin_lock(&sdp->sd_statfs_spin);
509 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
510 				      sizeof(struct gfs2_dinode));
511 		gfs2_statfs_change_in(l_sc, l_bh->b_data +
512 				      sizeof(struct gfs2_dinode));
513 		spin_unlock(&sdp->sd_statfs_spin);
514 
515 		brelse(l_bh);
516 	}
517 
518 out_m_bh:
519 	brelse(m_bh);
520 out:
521 	gfs2_glock_dq_uninit(&gh);
522 	return 0;
523 }
524 
525 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
526 			s64 dinodes)
527 {
528 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
529 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
530 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
531 	struct buffer_head *l_bh;
532 	s64 x, y;
533 	int need_sync = 0;
534 	int error;
535 
536 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
537 	if (error)
538 		return;
539 
540 	gfs2_trans_add_meta(l_ip->i_gl, l_bh);
541 
542 	spin_lock(&sdp->sd_statfs_spin);
543 	l_sc->sc_total += total;
544 	l_sc->sc_free += free;
545 	l_sc->sc_dinodes += dinodes;
546 	gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
547 	if (sdp->sd_args.ar_statfs_percent) {
548 		x = 100 * l_sc->sc_free;
549 		y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
550 		if (x >= y || x <= -y)
551 			need_sync = 1;
552 	}
553 	spin_unlock(&sdp->sd_statfs_spin);
554 
555 	brelse(l_bh);
556 	if (need_sync)
557 		gfs2_wake_up_statfs(sdp);
558 }
559 
560 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
561 		   struct buffer_head *l_bh)
562 {
563 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
564 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
565 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
566 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
567 
568 	gfs2_trans_add_meta(l_ip->i_gl, l_bh);
569 	gfs2_trans_add_meta(m_ip->i_gl, m_bh);
570 
571 	spin_lock(&sdp->sd_statfs_spin);
572 	m_sc->sc_total += l_sc->sc_total;
573 	m_sc->sc_free += l_sc->sc_free;
574 	m_sc->sc_dinodes += l_sc->sc_dinodes;
575 	memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
576 	memset(l_bh->b_data + sizeof(struct gfs2_dinode),
577 	       0, sizeof(struct gfs2_statfs_change));
578 	gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
579 	spin_unlock(&sdp->sd_statfs_spin);
580 }
581 
582 int gfs2_statfs_sync(struct super_block *sb, int type)
583 {
584 	struct gfs2_sbd *sdp = sb->s_fs_info;
585 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
586 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
587 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
588 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
589 	struct gfs2_holder gh;
590 	struct buffer_head *m_bh, *l_bh;
591 	int error;
592 
593 	sb_start_write(sb);
594 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
595 				   &gh);
596 	if (error)
597 		goto out;
598 
599 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
600 	if (error)
601 		goto out_unlock;
602 
603 	spin_lock(&sdp->sd_statfs_spin);
604 	gfs2_statfs_change_in(m_sc, m_bh->b_data +
605 			      sizeof(struct gfs2_dinode));
606 	if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
607 		spin_unlock(&sdp->sd_statfs_spin);
608 		goto out_bh;
609 	}
610 	spin_unlock(&sdp->sd_statfs_spin);
611 
612 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
613 	if (error)
614 		goto out_bh;
615 
616 	error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
617 	if (error)
618 		goto out_bh2;
619 
620 	update_statfs(sdp, m_bh, l_bh);
621 	sdp->sd_statfs_force_sync = 0;
622 
623 	gfs2_trans_end(sdp);
624 
625 out_bh2:
626 	brelse(l_bh);
627 out_bh:
628 	brelse(m_bh);
629 out_unlock:
630 	gfs2_glock_dq_uninit(&gh);
631 out:
632 	sb_end_write(sb);
633 	return error;
634 }
635 
636 struct lfcc {
637 	struct list_head list;
638 	struct gfs2_holder gh;
639 };
640 
641 /**
642  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
643  *                            journals are clean
644  * @sdp: the file system
645  * @state: the state to put the transaction lock into
646  * @t_gh: the hold on the transaction lock
647  *
648  * Returns: errno
649  */
650 
651 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
652 				    struct gfs2_holder *freeze_gh)
653 {
654 	struct gfs2_inode *ip;
655 	struct gfs2_jdesc *jd;
656 	struct lfcc *lfcc;
657 	LIST_HEAD(list);
658 	struct gfs2_log_header_host lh;
659 	int error;
660 
661 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
662 		lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
663 		if (!lfcc) {
664 			error = -ENOMEM;
665 			goto out;
666 		}
667 		ip = GFS2_I(jd->jd_inode);
668 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
669 		if (error) {
670 			kfree(lfcc);
671 			goto out;
672 		}
673 		list_add(&lfcc->list, &list);
674 	}
675 
676 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
677 				   GL_NOCACHE, freeze_gh);
678 
679 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
680 		error = gfs2_jdesc_check(jd);
681 		if (error)
682 			break;
683 		error = gfs2_find_jhead(jd, &lh);
684 		if (error)
685 			break;
686 		if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
687 			error = -EBUSY;
688 			break;
689 		}
690 	}
691 
692 	if (error)
693 		gfs2_glock_dq_uninit(freeze_gh);
694 
695 out:
696 	while (!list_empty(&list)) {
697 		lfcc = list_entry(list.next, struct lfcc, list);
698 		list_del(&lfcc->list);
699 		gfs2_glock_dq_uninit(&lfcc->gh);
700 		kfree(lfcc);
701 	}
702 	return error;
703 }
704 
705 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
706 {
707 	struct gfs2_dinode *str = buf;
708 
709 	str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
710 	str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
711 	str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
712 	str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
713 	str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
714 	str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
715 	str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
716 	str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
717 	str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
718 	str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
719 	str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
720 	str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
721 	str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
722 	str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
723 
724 	str->di_goal_meta = cpu_to_be64(ip->i_goal);
725 	str->di_goal_data = cpu_to_be64(ip->i_goal);
726 	str->di_generation = cpu_to_be64(ip->i_generation);
727 
728 	str->di_flags = cpu_to_be32(ip->i_diskflags);
729 	str->di_height = cpu_to_be16(ip->i_height);
730 	str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
731 					     !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
732 					     GFS2_FORMAT_DE : 0);
733 	str->di_depth = cpu_to_be16(ip->i_depth);
734 	str->di_entries = cpu_to_be32(ip->i_entries);
735 
736 	str->di_eattr = cpu_to_be64(ip->i_eattr);
737 	str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
738 	str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
739 	str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
740 }
741 
742 /**
743  * gfs2_write_inode - Make sure the inode is stable on the disk
744  * @inode: The inode
745  * @wbc: The writeback control structure
746  *
747  * Returns: errno
748  */
749 
750 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
751 {
752 	struct gfs2_inode *ip = GFS2_I(inode);
753 	struct gfs2_sbd *sdp = GFS2_SB(inode);
754 	struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
755 	struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
756 	int ret = 0;
757 
758 	if (wbc->sync_mode == WB_SYNC_ALL)
759 		gfs2_log_flush(GFS2_SB(inode), ip->i_gl, NORMAL_FLUSH);
760 	if (bdi->wb.dirty_exceeded)
761 		gfs2_ail1_flush(sdp, wbc);
762 	else
763 		filemap_fdatawrite(metamapping);
764 	if (wbc->sync_mode == WB_SYNC_ALL)
765 		ret = filemap_fdatawait(metamapping);
766 	if (ret)
767 		mark_inode_dirty_sync(inode);
768 	return ret;
769 }
770 
771 /**
772  * gfs2_dirty_inode - check for atime updates
773  * @inode: The inode in question
774  * @flags: The type of dirty
775  *
776  * Unfortunately it can be called under any combination of inode
777  * glock and transaction lock, so we have to check carefully.
778  *
779  * At the moment this deals only with atime - it should be possible
780  * to expand that role in future, once a review of the locking has
781  * been carried out.
782  */
783 
784 static void gfs2_dirty_inode(struct inode *inode, int flags)
785 {
786 	struct gfs2_inode *ip = GFS2_I(inode);
787 	struct gfs2_sbd *sdp = GFS2_SB(inode);
788 	struct buffer_head *bh;
789 	struct gfs2_holder gh;
790 	int need_unlock = 0;
791 	int need_endtrans = 0;
792 	int ret;
793 
794 	if (!(flags & (I_DIRTY_DATASYNC|I_DIRTY_SYNC)))
795 		return;
796 
797 	if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
798 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
799 		if (ret) {
800 			fs_err(sdp, "dirty_inode: glock %d\n", ret);
801 			return;
802 		}
803 		need_unlock = 1;
804 	} else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
805 		return;
806 
807 	if (current->journal_info == NULL) {
808 		ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
809 		if (ret) {
810 			fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
811 			goto out;
812 		}
813 		need_endtrans = 1;
814 	}
815 
816 	ret = gfs2_meta_inode_buffer(ip, &bh);
817 	if (ret == 0) {
818 		gfs2_trans_add_meta(ip->i_gl, bh);
819 		gfs2_dinode_out(ip, bh->b_data);
820 		brelse(bh);
821 	}
822 
823 	if (need_endtrans)
824 		gfs2_trans_end(sdp);
825 out:
826 	if (need_unlock)
827 		gfs2_glock_dq_uninit(&gh);
828 }
829 
830 /**
831  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
832  * @sdp: the filesystem
833  *
834  * Returns: errno
835  */
836 
837 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
838 {
839 	struct gfs2_holder freeze_gh;
840 	int error;
841 
842 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, GL_NOCACHE,
843 				   &freeze_gh);
844 	if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
845 		return error;
846 
847 	kthread_stop(sdp->sd_quotad_process);
848 	kthread_stop(sdp->sd_logd_process);
849 
850 	flush_workqueue(gfs2_delete_workqueue);
851 	gfs2_quota_sync(sdp->sd_vfs, 0);
852 	gfs2_statfs_sync(sdp->sd_vfs, 0);
853 
854 	gfs2_log_flush(sdp, NULL, SHUTDOWN_FLUSH);
855 	wait_event(sdp->sd_reserving_log_wait, atomic_read(&sdp->sd_reserving_log) == 0);
856 	gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
857 
858 	if (gfs2_holder_initialized(&freeze_gh))
859 		gfs2_glock_dq_uninit(&freeze_gh);
860 
861 	gfs2_quota_cleanup(sdp);
862 
863 	return error;
864 }
865 
866 /**
867  * gfs2_put_super - Unmount the filesystem
868  * @sb: The VFS superblock
869  *
870  */
871 
872 static void gfs2_put_super(struct super_block *sb)
873 {
874 	struct gfs2_sbd *sdp = sb->s_fs_info;
875 	int error;
876 	struct gfs2_jdesc *jd;
877 
878 	/* No more recovery requests */
879 	set_bit(SDF_NORECOVERY, &sdp->sd_flags);
880 	smp_mb();
881 
882 	/* Wait on outstanding recovery */
883 restart:
884 	spin_lock(&sdp->sd_jindex_spin);
885 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
886 		if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
887 			continue;
888 		spin_unlock(&sdp->sd_jindex_spin);
889 		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
890 			    TASK_UNINTERRUPTIBLE);
891 		goto restart;
892 	}
893 	spin_unlock(&sdp->sd_jindex_spin);
894 
895 	if (!(sb->s_flags & MS_RDONLY)) {
896 		error = gfs2_make_fs_ro(sdp);
897 		if (error)
898 			gfs2_io_error(sdp);
899 	}
900 	/*  At this point, we're through modifying the disk  */
901 
902 	/*  Release stuff  */
903 
904 	iput(sdp->sd_jindex);
905 	iput(sdp->sd_statfs_inode);
906 	iput(sdp->sd_rindex);
907 	iput(sdp->sd_quota_inode);
908 
909 	gfs2_glock_put(sdp->sd_rename_gl);
910 	gfs2_glock_put(sdp->sd_freeze_gl);
911 
912 	if (!sdp->sd_args.ar_spectator) {
913 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
914 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
915 		gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
916 		gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
917 		iput(sdp->sd_sc_inode);
918 		iput(sdp->sd_qc_inode);
919 	}
920 
921 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
922 	gfs2_clear_rgrpd(sdp);
923 	gfs2_jindex_free(sdp);
924 	/*  Take apart glock structures and buffer lists  */
925 	gfs2_gl_hash_clear(sdp);
926 	/*  Unmount the locking protocol  */
927 	gfs2_lm_unmount(sdp);
928 
929 	/*  At this point, we're through participating in the lockspace  */
930 	gfs2_sys_fs_del(sdp);
931 }
932 
933 /**
934  * gfs2_sync_fs - sync the filesystem
935  * @sb: the superblock
936  *
937  * Flushes the log to disk.
938  */
939 
940 static int gfs2_sync_fs(struct super_block *sb, int wait)
941 {
942 	struct gfs2_sbd *sdp = sb->s_fs_info;
943 
944 	gfs2_quota_sync(sb, -1);
945 	if (wait && sdp)
946 		gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
947 	return 0;
948 }
949 
950 void gfs2_freeze_func(struct work_struct *work)
951 {
952 	int error;
953 	struct gfs2_holder freeze_gh;
954 	struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
955 	struct super_block *sb = sdp->sd_vfs;
956 
957 	atomic_inc(&sb->s_active);
958 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
959 				   &freeze_gh);
960 	if (error) {
961 		printk(KERN_INFO "GFS2: couln't get freeze lock : %d\n", error);
962 		gfs2_assert_withdraw(sdp, 0);
963 	}
964 	else {
965 		atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
966 		error = thaw_super(sb);
967 		if (error) {
968 			printk(KERN_INFO "GFS2: couldn't thaw filesystem: %d\n",
969 			       error);
970 			gfs2_assert_withdraw(sdp, 0);
971 		}
972 		if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
973 			freeze_gh.gh_flags |= GL_NOCACHE;
974 		gfs2_glock_dq_uninit(&freeze_gh);
975 	}
976 	deactivate_super(sb);
977 	return;
978 }
979 
980 /**
981  * gfs2_freeze - prevent further writes to the filesystem
982  * @sb: the VFS structure for the filesystem
983  *
984  */
985 
986 static int gfs2_freeze(struct super_block *sb)
987 {
988 	struct gfs2_sbd *sdp = sb->s_fs_info;
989 	int error = 0;
990 
991 	mutex_lock(&sdp->sd_freeze_mutex);
992 	if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN)
993 		goto out;
994 
995 	if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
996 		error = -EINVAL;
997 		goto out;
998 	}
999 
1000 	for (;;) {
1001 		error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
1002 		if (!error)
1003 			break;
1004 
1005 		switch (error) {
1006 		case -EBUSY:
1007 			fs_err(sdp, "waiting for recovery before freeze\n");
1008 			break;
1009 
1010 		default:
1011 			fs_err(sdp, "error freezing FS: %d\n", error);
1012 			break;
1013 		}
1014 
1015 		fs_err(sdp, "retrying...\n");
1016 		msleep(1000);
1017 	}
1018 	error = 0;
1019 out:
1020 	mutex_unlock(&sdp->sd_freeze_mutex);
1021 	return error;
1022 }
1023 
1024 /**
1025  * gfs2_unfreeze - reallow writes to the filesystem
1026  * @sb: the VFS structure for the filesystem
1027  *
1028  */
1029 
1030 static int gfs2_unfreeze(struct super_block *sb)
1031 {
1032 	struct gfs2_sbd *sdp = sb->s_fs_info;
1033 
1034 	mutex_lock(&sdp->sd_freeze_mutex);
1035         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
1036 	    !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
1037 		mutex_unlock(&sdp->sd_freeze_mutex);
1038                 return 0;
1039 	}
1040 
1041 	gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1042 	mutex_unlock(&sdp->sd_freeze_mutex);
1043 	return 0;
1044 }
1045 
1046 /**
1047  * statfs_fill - fill in the sg for a given RG
1048  * @rgd: the RG
1049  * @sc: the sc structure
1050  *
1051  * Returns: 0 on success, -ESTALE if the LVB is invalid
1052  */
1053 
1054 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1055 			    struct gfs2_statfs_change_host *sc)
1056 {
1057 	gfs2_rgrp_verify(rgd);
1058 	sc->sc_total += rgd->rd_data;
1059 	sc->sc_free += rgd->rd_free;
1060 	sc->sc_dinodes += rgd->rd_dinodes;
1061 	return 0;
1062 }
1063 
1064 /**
1065  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1066  * @sdp: the filesystem
1067  * @sc: the sc info that will be returned
1068  *
1069  * Any error (other than a signal) will cause this routine to fall back
1070  * to the synchronous version.
1071  *
1072  * FIXME: This really shouldn't busy wait like this.
1073  *
1074  * Returns: errno
1075  */
1076 
1077 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1078 {
1079 	struct gfs2_rgrpd *rgd_next;
1080 	struct gfs2_holder *gha, *gh;
1081 	unsigned int slots = 64;
1082 	unsigned int x;
1083 	int done;
1084 	int error = 0, err;
1085 
1086 	memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1087 	gha = kmalloc(slots * sizeof(struct gfs2_holder), GFP_KERNEL);
1088 	if (!gha)
1089 		return -ENOMEM;
1090 	for (x = 0; x < slots; x++)
1091 		gfs2_holder_mark_uninitialized(gha + x);
1092 
1093 	rgd_next = gfs2_rgrpd_get_first(sdp);
1094 
1095 	for (;;) {
1096 		done = 1;
1097 
1098 		for (x = 0; x < slots; x++) {
1099 			gh = gha + x;
1100 
1101 			if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
1102 				err = gfs2_glock_wait(gh);
1103 				if (err) {
1104 					gfs2_holder_uninit(gh);
1105 					error = err;
1106 				} else {
1107 					if (!error)
1108 						error = statfs_slow_fill(
1109 							gh->gh_gl->gl_object, sc);
1110 					gfs2_glock_dq_uninit(gh);
1111 				}
1112 			}
1113 
1114 			if (gfs2_holder_initialized(gh))
1115 				done = 0;
1116 			else if (rgd_next && !error) {
1117 				error = gfs2_glock_nq_init(rgd_next->rd_gl,
1118 							   LM_ST_SHARED,
1119 							   GL_ASYNC,
1120 							   gh);
1121 				rgd_next = gfs2_rgrpd_get_next(rgd_next);
1122 				done = 0;
1123 			}
1124 
1125 			if (signal_pending(current))
1126 				error = -ERESTARTSYS;
1127 		}
1128 
1129 		if (done)
1130 			break;
1131 
1132 		yield();
1133 	}
1134 
1135 	kfree(gha);
1136 	return error;
1137 }
1138 
1139 /**
1140  * gfs2_statfs_i - Do a statfs
1141  * @sdp: the filesystem
1142  * @sg: the sg structure
1143  *
1144  * Returns: errno
1145  */
1146 
1147 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1148 {
1149 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1150 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1151 
1152 	spin_lock(&sdp->sd_statfs_spin);
1153 
1154 	*sc = *m_sc;
1155 	sc->sc_total += l_sc->sc_total;
1156 	sc->sc_free += l_sc->sc_free;
1157 	sc->sc_dinodes += l_sc->sc_dinodes;
1158 
1159 	spin_unlock(&sdp->sd_statfs_spin);
1160 
1161 	if (sc->sc_free < 0)
1162 		sc->sc_free = 0;
1163 	if (sc->sc_free > sc->sc_total)
1164 		sc->sc_free = sc->sc_total;
1165 	if (sc->sc_dinodes < 0)
1166 		sc->sc_dinodes = 0;
1167 
1168 	return 0;
1169 }
1170 
1171 /**
1172  * gfs2_statfs - Gather and return stats about the filesystem
1173  * @sb: The superblock
1174  * @statfsbuf: The buffer
1175  *
1176  * Returns: 0 on success or error code
1177  */
1178 
1179 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1180 {
1181 	struct super_block *sb = dentry->d_sb;
1182 	struct gfs2_sbd *sdp = sb->s_fs_info;
1183 	struct gfs2_statfs_change_host sc;
1184 	int error;
1185 
1186 	error = gfs2_rindex_update(sdp);
1187 	if (error)
1188 		return error;
1189 
1190 	if (gfs2_tune_get(sdp, gt_statfs_slow))
1191 		error = gfs2_statfs_slow(sdp, &sc);
1192 	else
1193 		error = gfs2_statfs_i(sdp, &sc);
1194 
1195 	if (error)
1196 		return error;
1197 
1198 	buf->f_type = GFS2_MAGIC;
1199 	buf->f_bsize = sdp->sd_sb.sb_bsize;
1200 	buf->f_blocks = sc.sc_total;
1201 	buf->f_bfree = sc.sc_free;
1202 	buf->f_bavail = sc.sc_free;
1203 	buf->f_files = sc.sc_dinodes + sc.sc_free;
1204 	buf->f_ffree = sc.sc_free;
1205 	buf->f_namelen = GFS2_FNAMESIZE;
1206 
1207 	return 0;
1208 }
1209 
1210 /**
1211  * gfs2_remount_fs - called when the FS is remounted
1212  * @sb:  the filesystem
1213  * @flags:  the remount flags
1214  * @data:  extra data passed in (not used right now)
1215  *
1216  * Returns: errno
1217  */
1218 
1219 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1220 {
1221 	struct gfs2_sbd *sdp = sb->s_fs_info;
1222 	struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1223 	struct gfs2_tune *gt = &sdp->sd_tune;
1224 	int error;
1225 
1226 	sync_filesystem(sb);
1227 
1228 	spin_lock(&gt->gt_spin);
1229 	args.ar_commit = gt->gt_logd_secs;
1230 	args.ar_quota_quantum = gt->gt_quota_quantum;
1231 	if (gt->gt_statfs_slow)
1232 		args.ar_statfs_quantum = 0;
1233 	else
1234 		args.ar_statfs_quantum = gt->gt_statfs_quantum;
1235 	spin_unlock(&gt->gt_spin);
1236 	error = gfs2_mount_args(&args, data);
1237 	if (error)
1238 		return error;
1239 
1240 	/* Not allowed to change locking details */
1241 	if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1242 	    strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1243 	    strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1244 		return -EINVAL;
1245 
1246 	/* Some flags must not be changed */
1247 	if (args_neq(&args, &sdp->sd_args, spectator) ||
1248 	    args_neq(&args, &sdp->sd_args, localflocks) ||
1249 	    args_neq(&args, &sdp->sd_args, meta))
1250 		return -EINVAL;
1251 
1252 	if (sdp->sd_args.ar_spectator)
1253 		*flags |= MS_RDONLY;
1254 
1255 	if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1256 		if (*flags & MS_RDONLY)
1257 			error = gfs2_make_fs_ro(sdp);
1258 		else
1259 			error = gfs2_make_fs_rw(sdp);
1260 		if (error)
1261 			return error;
1262 	}
1263 
1264 	sdp->sd_args = args;
1265 	if (sdp->sd_args.ar_posix_acl)
1266 		sb->s_flags |= MS_POSIXACL;
1267 	else
1268 		sb->s_flags &= ~MS_POSIXACL;
1269 	if (sdp->sd_args.ar_nobarrier)
1270 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1271 	else
1272 		clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1273 	spin_lock(&gt->gt_spin);
1274 	gt->gt_logd_secs = args.ar_commit;
1275 	gt->gt_quota_quantum = args.ar_quota_quantum;
1276 	if (args.ar_statfs_quantum) {
1277 		gt->gt_statfs_slow = 0;
1278 		gt->gt_statfs_quantum = args.ar_statfs_quantum;
1279 	}
1280 	else {
1281 		gt->gt_statfs_slow = 1;
1282 		gt->gt_statfs_quantum = 30;
1283 	}
1284 	spin_unlock(&gt->gt_spin);
1285 
1286 	gfs2_online_uevent(sdp);
1287 	return 0;
1288 }
1289 
1290 /**
1291  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1292  * @inode: The inode to drop
1293  *
1294  * If we've received a callback on an iopen lock then its because a
1295  * remote node tried to deallocate the inode but failed due to this node
1296  * still having the inode open. Here we mark the link count zero
1297  * since we know that it must have reached zero if the GLF_DEMOTE flag
1298  * is set on the iopen glock. If we didn't do a disk read since the
1299  * remote node removed the final link then we might otherwise miss
1300  * this event. This check ensures that this node will deallocate the
1301  * inode's blocks, or alternatively pass the baton on to another
1302  * node for later deallocation.
1303  */
1304 
1305 static int gfs2_drop_inode(struct inode *inode)
1306 {
1307 	struct gfs2_inode *ip = GFS2_I(inode);
1308 
1309 	if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
1310 	    inode->i_nlink &&
1311 	    gfs2_holder_initialized(&ip->i_iopen_gh)) {
1312 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1313 		if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1314 			clear_nlink(inode);
1315 	}
1316 	return generic_drop_inode(inode);
1317 }
1318 
1319 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1320 {
1321 	do {
1322 		if (d1 == d2)
1323 			return 1;
1324 		d1 = d1->d_parent;
1325 	} while (!IS_ROOT(d1));
1326 	return 0;
1327 }
1328 
1329 /**
1330  * gfs2_show_options - Show mount options for /proc/mounts
1331  * @s: seq_file structure
1332  * @root: root of this (sub)tree
1333  *
1334  * Returns: 0 on success or error code
1335  */
1336 
1337 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1338 {
1339 	struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1340 	struct gfs2_args *args = &sdp->sd_args;
1341 	int val;
1342 
1343 	if (is_ancestor(root, sdp->sd_master_dir))
1344 		seq_puts(s, ",meta");
1345 	if (args->ar_lockproto[0])
1346 		seq_show_option(s, "lockproto", args->ar_lockproto);
1347 	if (args->ar_locktable[0])
1348 		seq_show_option(s, "locktable", args->ar_locktable);
1349 	if (args->ar_hostdata[0])
1350 		seq_show_option(s, "hostdata", args->ar_hostdata);
1351 	if (args->ar_spectator)
1352 		seq_puts(s, ",spectator");
1353 	if (args->ar_localflocks)
1354 		seq_puts(s, ",localflocks");
1355 	if (args->ar_debug)
1356 		seq_puts(s, ",debug");
1357 	if (args->ar_posix_acl)
1358 		seq_puts(s, ",acl");
1359 	if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1360 		char *state;
1361 		switch (args->ar_quota) {
1362 		case GFS2_QUOTA_OFF:
1363 			state = "off";
1364 			break;
1365 		case GFS2_QUOTA_ACCOUNT:
1366 			state = "account";
1367 			break;
1368 		case GFS2_QUOTA_ON:
1369 			state = "on";
1370 			break;
1371 		default:
1372 			state = "unknown";
1373 			break;
1374 		}
1375 		seq_printf(s, ",quota=%s", state);
1376 	}
1377 	if (args->ar_suiddir)
1378 		seq_puts(s, ",suiddir");
1379 	if (args->ar_data != GFS2_DATA_DEFAULT) {
1380 		char *state;
1381 		switch (args->ar_data) {
1382 		case GFS2_DATA_WRITEBACK:
1383 			state = "writeback";
1384 			break;
1385 		case GFS2_DATA_ORDERED:
1386 			state = "ordered";
1387 			break;
1388 		default:
1389 			state = "unknown";
1390 			break;
1391 		}
1392 		seq_printf(s, ",data=%s", state);
1393 	}
1394 	if (args->ar_discard)
1395 		seq_puts(s, ",discard");
1396 	val = sdp->sd_tune.gt_logd_secs;
1397 	if (val != 30)
1398 		seq_printf(s, ",commit=%d", val);
1399 	val = sdp->sd_tune.gt_statfs_quantum;
1400 	if (val != 30)
1401 		seq_printf(s, ",statfs_quantum=%d", val);
1402 	else if (sdp->sd_tune.gt_statfs_slow)
1403 		seq_puts(s, ",statfs_quantum=0");
1404 	val = sdp->sd_tune.gt_quota_quantum;
1405 	if (val != 60)
1406 		seq_printf(s, ",quota_quantum=%d", val);
1407 	if (args->ar_statfs_percent)
1408 		seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1409 	if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1410 		const char *state;
1411 
1412 		switch (args->ar_errors) {
1413 		case GFS2_ERRORS_WITHDRAW:
1414 			state = "withdraw";
1415 			break;
1416 		case GFS2_ERRORS_PANIC:
1417 			state = "panic";
1418 			break;
1419 		default:
1420 			state = "unknown";
1421 			break;
1422 		}
1423 		seq_printf(s, ",errors=%s", state);
1424 	}
1425 	if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1426 		seq_puts(s, ",nobarrier");
1427 	if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1428 		seq_puts(s, ",demote_interface_used");
1429 	if (args->ar_rgrplvb)
1430 		seq_puts(s, ",rgrplvb");
1431 	if (args->ar_loccookie)
1432 		seq_puts(s, ",loccookie");
1433 	return 0;
1434 }
1435 
1436 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1437 {
1438 	struct inode *inode = &ip->i_inode;
1439 	struct gfs2_glock *gl = ip->i_gl;
1440 
1441 	truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1442 	truncate_inode_pages(&inode->i_data, 0);
1443 
1444 	if (atomic_read(&gl->gl_revokes) == 0) {
1445 		clear_bit(GLF_LFLUSH, &gl->gl_flags);
1446 		clear_bit(GLF_DIRTY, &gl->gl_flags);
1447 	}
1448 }
1449 
1450 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1451 {
1452 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1453 	struct gfs2_rgrpd *rgd;
1454 	struct gfs2_holder gh;
1455 	int error;
1456 
1457 	if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1458 		gfs2_consist_inode(ip);
1459 		return -EIO;
1460 	}
1461 
1462 	error = gfs2_rindex_update(sdp);
1463 	if (error)
1464 		return error;
1465 
1466 	error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1467 	if (error)
1468 		return error;
1469 
1470 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1471 	if (!rgd) {
1472 		gfs2_consist_inode(ip);
1473 		error = -EIO;
1474 		goto out_qs;
1475 	}
1476 
1477 	error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1478 	if (error)
1479 		goto out_qs;
1480 
1481 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1482 				 sdp->sd_jdesc->jd_blocks);
1483 	if (error)
1484 		goto out_rg_gunlock;
1485 
1486 	gfs2_free_di(rgd, ip);
1487 
1488 	gfs2_final_release_pages(ip);
1489 
1490 	gfs2_trans_end(sdp);
1491 
1492 out_rg_gunlock:
1493 	gfs2_glock_dq_uninit(&gh);
1494 out_qs:
1495 	gfs2_quota_unhold(ip);
1496 	return error;
1497 }
1498 
1499 /**
1500  * gfs2_evict_inode - Remove an inode from cache
1501  * @inode: The inode to evict
1502  *
1503  * There are three cases to consider:
1504  * 1. i_nlink == 0, we are final opener (and must deallocate)
1505  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1506  * 3. i_nlink > 0
1507  *
1508  * If the fs is read only, then we have to treat all cases as per #3
1509  * since we are unable to do any deallocation. The inode will be
1510  * deallocated by the next read/write node to attempt an allocation
1511  * in the same resource group
1512  *
1513  * We have to (at the moment) hold the inodes main lock to cover
1514  * the gap between unlocking the shared lock on the iopen lock and
1515  * taking the exclusive lock. I'd rather do a shared -> exclusive
1516  * conversion on the iopen lock, but we can change that later. This
1517  * is safe, just less efficient.
1518  */
1519 
1520 static void gfs2_evict_inode(struct inode *inode)
1521 {
1522 	struct super_block *sb = inode->i_sb;
1523 	struct gfs2_sbd *sdp = sb->s_fs_info;
1524 	struct gfs2_inode *ip = GFS2_I(inode);
1525 	struct gfs2_holder gh;
1526 	struct address_space *metamapping;
1527 	int error;
1528 
1529 	if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1530 		clear_inode(inode);
1531 		return;
1532 	}
1533 
1534 	if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1535 		goto out;
1536 
1537 	/* Must not read inode block until block type has been verified */
1538 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1539 	if (unlikely(error)) {
1540 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1541 		gfs2_glock_dq_wait(&ip->i_iopen_gh);
1542 		gfs2_holder_uninit(&ip->i_iopen_gh);
1543 		goto out;
1544 	}
1545 
1546 	if (!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1547 		error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1548 		if (error)
1549 			goto out_truncate;
1550 	}
1551 
1552 	if (test_bit(GIF_INVALID, &ip->i_flags)) {
1553 		error = gfs2_inode_refresh(ip);
1554 		if (error)
1555 			goto out_truncate;
1556 	}
1557 
1558 	if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1559 	    test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1560 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1561 		gfs2_glock_dq_wait(&ip->i_iopen_gh);
1562 		gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE,
1563 				   &ip->i_iopen_gh);
1564 		error = gfs2_glock_nq(&ip->i_iopen_gh);
1565 		if (error)
1566 			goto out_truncate;
1567 	}
1568 
1569 	/* Case 1 starts here */
1570 
1571 	if (S_ISDIR(inode->i_mode) &&
1572 	    (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1573 		error = gfs2_dir_exhash_dealloc(ip);
1574 		if (error)
1575 			goto out_unlock;
1576 	}
1577 
1578 	if (ip->i_eattr) {
1579 		error = gfs2_ea_dealloc(ip);
1580 		if (error)
1581 			goto out_unlock;
1582 	}
1583 
1584 	if (!gfs2_is_stuffed(ip)) {
1585 		error = gfs2_file_dealloc(ip);
1586 		if (error)
1587 			goto out_unlock;
1588 	}
1589 
1590 	error = gfs2_dinode_dealloc(ip);
1591 	goto out_unlock;
1592 
1593 out_truncate:
1594 	gfs2_log_flush(sdp, ip->i_gl, NORMAL_FLUSH);
1595 	metamapping = gfs2_glock2aspace(ip->i_gl);
1596 	if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1597 		filemap_fdatawrite(metamapping);
1598 		filemap_fdatawait(metamapping);
1599 	}
1600 	write_inode_now(inode, 1);
1601 	gfs2_ail_flush(ip->i_gl, 0);
1602 
1603 	/* Case 2 starts here */
1604 	error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1605 	if (error)
1606 		goto out_unlock;
1607 	/* Needs to be done before glock release & also in a transaction */
1608 	truncate_inode_pages(&inode->i_data, 0);
1609 	truncate_inode_pages(metamapping, 0);
1610 	gfs2_trans_end(sdp);
1611 
1612 out_unlock:
1613 	/* Error path for case 1 */
1614 	if (gfs2_rs_active(&ip->i_res))
1615 		gfs2_rs_deltree(&ip->i_res);
1616 
1617 	if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1618 		if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1619 			ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1620 			gfs2_glock_dq_wait(&ip->i_iopen_gh);
1621 		}
1622 		gfs2_holder_uninit(&ip->i_iopen_gh);
1623 	}
1624 	gfs2_glock_dq_uninit(&gh);
1625 	if (error && error != GLR_TRYFAILED && error != -EROFS)
1626 		fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1627 out:
1628 	/* Case 3 starts here */
1629 	truncate_inode_pages_final(&inode->i_data);
1630 	gfs2_rsqa_delete(ip, NULL);
1631 	gfs2_ordered_del_inode(ip);
1632 	clear_inode(inode);
1633 	gfs2_dir_hash_inval(ip);
1634 	ip->i_gl->gl_object = NULL;
1635 	flush_delayed_work(&ip->i_gl->gl_work);
1636 	gfs2_glock_add_to_lru(ip->i_gl);
1637 	gfs2_glock_put(ip->i_gl);
1638 	ip->i_gl = NULL;
1639 	if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1640 		ip->i_iopen_gh.gh_gl->gl_object = NULL;
1641 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1642 		gfs2_glock_dq_wait(&ip->i_iopen_gh);
1643 		gfs2_holder_uninit(&ip->i_iopen_gh);
1644 	}
1645 }
1646 
1647 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1648 {
1649 	struct gfs2_inode *ip;
1650 
1651 	ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1652 	if (ip) {
1653 		ip->i_flags = 0;
1654 		ip->i_gl = NULL;
1655 		ip->i_rgd = NULL;
1656 		memset(&ip->i_res, 0, sizeof(ip->i_res));
1657 		RB_CLEAR_NODE(&ip->i_res.rs_node);
1658 		ip->i_rahead = 0;
1659 	}
1660 	return &ip->i_inode;
1661 }
1662 
1663 static void gfs2_i_callback(struct rcu_head *head)
1664 {
1665 	struct inode *inode = container_of(head, struct inode, i_rcu);
1666 	kmem_cache_free(gfs2_inode_cachep, inode);
1667 }
1668 
1669 static void gfs2_destroy_inode(struct inode *inode)
1670 {
1671 	call_rcu(&inode->i_rcu, gfs2_i_callback);
1672 }
1673 
1674 const struct super_operations gfs2_super_ops = {
1675 	.alloc_inode		= gfs2_alloc_inode,
1676 	.destroy_inode		= gfs2_destroy_inode,
1677 	.write_inode		= gfs2_write_inode,
1678 	.dirty_inode		= gfs2_dirty_inode,
1679 	.evict_inode		= gfs2_evict_inode,
1680 	.put_super		= gfs2_put_super,
1681 	.sync_fs		= gfs2_sync_fs,
1682 	.freeze_super		= gfs2_freeze,
1683 	.thaw_super		= gfs2_unfreeze,
1684 	.statfs			= gfs2_statfs,
1685 	.remount_fs		= gfs2_remount_fs,
1686 	.drop_inode		= gfs2_drop_inode,
1687 	.show_options		= gfs2_show_options,
1688 };
1689 
1690