xref: /openbmc/linux/fs/gfs2/super.c (revision 009d851837ab26cab18adda6169a813f70b0b21b)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9 
10 #include <linux/bio.h>
11 #include <linux/sched.h>
12 #include <linux/slab.h>
13 #include <linux/spinlock.h>
14 #include <linux/completion.h>
15 #include <linux/buffer_head.h>
16 #include <linux/statfs.h>
17 #include <linux/seq_file.h>
18 #include <linux/mount.h>
19 #include <linux/kthread.h>
20 #include <linux/delay.h>
21 #include <linux/gfs2_ondisk.h>
22 #include <linux/crc32.h>
23 #include <linux/time.h>
24 #include <linux/wait.h>
25 
26 #include "gfs2.h"
27 #include "incore.h"
28 #include "bmap.h"
29 #include "dir.h"
30 #include "glock.h"
31 #include "glops.h"
32 #include "inode.h"
33 #include "log.h"
34 #include "meta_io.h"
35 #include "quota.h"
36 #include "recovery.h"
37 #include "rgrp.h"
38 #include "super.h"
39 #include "trans.h"
40 #include "util.h"
41 #include "sys.h"
42 #include "xattr.h"
43 
44 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
45 
46 enum {
47 	Opt_lockproto,
48 	Opt_locktable,
49 	Opt_hostdata,
50 	Opt_spectator,
51 	Opt_ignore_local_fs,
52 	Opt_localflocks,
53 	Opt_localcaching,
54 	Opt_debug,
55 	Opt_nodebug,
56 	Opt_upgrade,
57 	Opt_acl,
58 	Opt_noacl,
59 	Opt_quota_off,
60 	Opt_quota_account,
61 	Opt_quota_on,
62 	Opt_quota,
63 	Opt_noquota,
64 	Opt_suiddir,
65 	Opt_nosuiddir,
66 	Opt_data_writeback,
67 	Opt_data_ordered,
68 	Opt_meta,
69 	Opt_discard,
70 	Opt_nodiscard,
71 	Opt_commit,
72 	Opt_err_withdraw,
73 	Opt_err_panic,
74 	Opt_statfs_quantum,
75 	Opt_statfs_percent,
76 	Opt_quota_quantum,
77 	Opt_barrier,
78 	Opt_nobarrier,
79 	Opt_error,
80 };
81 
82 static const match_table_t tokens = {
83 	{Opt_lockproto, "lockproto=%s"},
84 	{Opt_locktable, "locktable=%s"},
85 	{Opt_hostdata, "hostdata=%s"},
86 	{Opt_spectator, "spectator"},
87 	{Opt_ignore_local_fs, "ignore_local_fs"},
88 	{Opt_localflocks, "localflocks"},
89 	{Opt_localcaching, "localcaching"},
90 	{Opt_debug, "debug"},
91 	{Opt_nodebug, "nodebug"},
92 	{Opt_upgrade, "upgrade"},
93 	{Opt_acl, "acl"},
94 	{Opt_noacl, "noacl"},
95 	{Opt_quota_off, "quota=off"},
96 	{Opt_quota_account, "quota=account"},
97 	{Opt_quota_on, "quota=on"},
98 	{Opt_quota, "quota"},
99 	{Opt_noquota, "noquota"},
100 	{Opt_suiddir, "suiddir"},
101 	{Opt_nosuiddir, "nosuiddir"},
102 	{Opt_data_writeback, "data=writeback"},
103 	{Opt_data_ordered, "data=ordered"},
104 	{Opt_meta, "meta"},
105 	{Opt_discard, "discard"},
106 	{Opt_nodiscard, "nodiscard"},
107 	{Opt_commit, "commit=%d"},
108 	{Opt_err_withdraw, "errors=withdraw"},
109 	{Opt_err_panic, "errors=panic"},
110 	{Opt_statfs_quantum, "statfs_quantum=%d"},
111 	{Opt_statfs_percent, "statfs_percent=%d"},
112 	{Opt_quota_quantum, "quota_quantum=%d"},
113 	{Opt_barrier, "barrier"},
114 	{Opt_nobarrier, "nobarrier"},
115 	{Opt_error, NULL}
116 };
117 
118 /**
119  * gfs2_mount_args - Parse mount options
120  * @args: The structure into which the parsed options will be written
121  * @options: The options to parse
122  *
123  * Return: errno
124  */
125 
126 int gfs2_mount_args(struct gfs2_args *args, char *options)
127 {
128 	char *o;
129 	int token;
130 	substring_t tmp[MAX_OPT_ARGS];
131 	int rv;
132 
133 	/* Split the options into tokens with the "," character and
134 	   process them */
135 
136 	while (1) {
137 		o = strsep(&options, ",");
138 		if (o == NULL)
139 			break;
140 		if (*o == '\0')
141 			continue;
142 
143 		token = match_token(o, tokens, tmp);
144 		switch (token) {
145 		case Opt_lockproto:
146 			match_strlcpy(args->ar_lockproto, &tmp[0],
147 				      GFS2_LOCKNAME_LEN);
148 			break;
149 		case Opt_locktable:
150 			match_strlcpy(args->ar_locktable, &tmp[0],
151 				      GFS2_LOCKNAME_LEN);
152 			break;
153 		case Opt_hostdata:
154 			match_strlcpy(args->ar_hostdata, &tmp[0],
155 				      GFS2_LOCKNAME_LEN);
156 			break;
157 		case Opt_spectator:
158 			args->ar_spectator = 1;
159 			break;
160 		case Opt_ignore_local_fs:
161 			args->ar_ignore_local_fs = 1;
162 			break;
163 		case Opt_localflocks:
164 			args->ar_localflocks = 1;
165 			break;
166 		case Opt_localcaching:
167 			args->ar_localcaching = 1;
168 			break;
169 		case Opt_debug:
170 			if (args->ar_errors == GFS2_ERRORS_PANIC) {
171 				printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
172 				       "are mutually exclusive.\n");
173 				return -EINVAL;
174 			}
175 			args->ar_debug = 1;
176 			break;
177 		case Opt_nodebug:
178 			args->ar_debug = 0;
179 			break;
180 		case Opt_upgrade:
181 			args->ar_upgrade = 1;
182 			break;
183 		case Opt_acl:
184 			args->ar_posix_acl = 1;
185 			break;
186 		case Opt_noacl:
187 			args->ar_posix_acl = 0;
188 			break;
189 		case Opt_quota_off:
190 		case Opt_noquota:
191 			args->ar_quota = GFS2_QUOTA_OFF;
192 			break;
193 		case Opt_quota_account:
194 			args->ar_quota = GFS2_QUOTA_ACCOUNT;
195 			break;
196 		case Opt_quota_on:
197 		case Opt_quota:
198 			args->ar_quota = GFS2_QUOTA_ON;
199 			break;
200 		case Opt_suiddir:
201 			args->ar_suiddir = 1;
202 			break;
203 		case Opt_nosuiddir:
204 			args->ar_suiddir = 0;
205 			break;
206 		case Opt_data_writeback:
207 			args->ar_data = GFS2_DATA_WRITEBACK;
208 			break;
209 		case Opt_data_ordered:
210 			args->ar_data = GFS2_DATA_ORDERED;
211 			break;
212 		case Opt_meta:
213 			args->ar_meta = 1;
214 			break;
215 		case Opt_discard:
216 			args->ar_discard = 1;
217 			break;
218 		case Opt_nodiscard:
219 			args->ar_discard = 0;
220 			break;
221 		case Opt_commit:
222 			rv = match_int(&tmp[0], &args->ar_commit);
223 			if (rv || args->ar_commit <= 0) {
224 				printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n");
225 				return rv ? rv : -EINVAL;
226 			}
227 			break;
228 		case Opt_statfs_quantum:
229 			rv = match_int(&tmp[0], &args->ar_statfs_quantum);
230 			if (rv || args->ar_statfs_quantum < 0) {
231 				printk(KERN_WARNING "GFS2: statfs_quantum mount option requires a non-negative numeric argument\n");
232 				return rv ? rv : -EINVAL;
233 			}
234 			break;
235 		case Opt_quota_quantum:
236 			rv = match_int(&tmp[0], &args->ar_quota_quantum);
237 			if (rv || args->ar_quota_quantum <= 0) {
238 				printk(KERN_WARNING "GFS2: quota_quantum mount option requires a positive numeric argument\n");
239 				return rv ? rv : -EINVAL;
240 			}
241 			break;
242 		case Opt_statfs_percent:
243 			rv = match_int(&tmp[0], &args->ar_statfs_percent);
244 			if (rv || args->ar_statfs_percent < 0 ||
245 			    args->ar_statfs_percent > 100) {
246 				printk(KERN_WARNING "statfs_percent mount option requires a numeric argument between 0 and 100\n");
247 				return rv ? rv : -EINVAL;
248 			}
249 			break;
250 		case Opt_err_withdraw:
251 			args->ar_errors = GFS2_ERRORS_WITHDRAW;
252 			break;
253 		case Opt_err_panic:
254 			if (args->ar_debug) {
255 				printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
256 					"are mutually exclusive.\n");
257 				return -EINVAL;
258 			}
259 			args->ar_errors = GFS2_ERRORS_PANIC;
260 			break;
261 		case Opt_barrier:
262 			args->ar_nobarrier = 0;
263 			break;
264 		case Opt_nobarrier:
265 			args->ar_nobarrier = 1;
266 			break;
267 		case Opt_error:
268 		default:
269 			printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o);
270 			return -EINVAL;
271 		}
272 	}
273 
274 	return 0;
275 }
276 
277 /**
278  * gfs2_jindex_free - Clear all the journal index information
279  * @sdp: The GFS2 superblock
280  *
281  */
282 
283 void gfs2_jindex_free(struct gfs2_sbd *sdp)
284 {
285 	struct list_head list, *head;
286 	struct gfs2_jdesc *jd;
287 	struct gfs2_journal_extent *jext;
288 
289 	spin_lock(&sdp->sd_jindex_spin);
290 	list_add(&list, &sdp->sd_jindex_list);
291 	list_del_init(&sdp->sd_jindex_list);
292 	sdp->sd_journals = 0;
293 	spin_unlock(&sdp->sd_jindex_spin);
294 
295 	while (!list_empty(&list)) {
296 		jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
297 		head = &jd->extent_list;
298 		while (!list_empty(head)) {
299 			jext = list_entry(head->next,
300 					  struct gfs2_journal_extent,
301 					  extent_list);
302 			list_del(&jext->extent_list);
303 			kfree(jext);
304 		}
305 		list_del(&jd->jd_list);
306 		iput(jd->jd_inode);
307 		kfree(jd);
308 	}
309 }
310 
311 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
312 {
313 	struct gfs2_jdesc *jd;
314 	int found = 0;
315 
316 	list_for_each_entry(jd, head, jd_list) {
317 		if (jd->jd_jid == jid) {
318 			found = 1;
319 			break;
320 		}
321 	}
322 
323 	if (!found)
324 		jd = NULL;
325 
326 	return jd;
327 }
328 
329 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
330 {
331 	struct gfs2_jdesc *jd;
332 
333 	spin_lock(&sdp->sd_jindex_spin);
334 	jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
335 	spin_unlock(&sdp->sd_jindex_spin);
336 
337 	return jd;
338 }
339 
340 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
341 {
342 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
343 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
344 	int ar;
345 	int error;
346 
347 	if (ip->i_disksize < (8 << 20) || ip->i_disksize > (1 << 30) ||
348 	    (ip->i_disksize & (sdp->sd_sb.sb_bsize - 1))) {
349 		gfs2_consist_inode(ip);
350 		return -EIO;
351 	}
352 	jd->jd_blocks = ip->i_disksize >> sdp->sd_sb.sb_bsize_shift;
353 
354 	error = gfs2_write_alloc_required(ip, 0, ip->i_disksize, &ar);
355 	if (!error && ar) {
356 		gfs2_consist_inode(ip);
357 		error = -EIO;
358 	}
359 
360 	return error;
361 }
362 
363 /**
364  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
365  * @sdp: the filesystem
366  *
367  * Returns: errno
368  */
369 
370 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
371 {
372 	struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
373 	struct gfs2_glock *j_gl = ip->i_gl;
374 	struct gfs2_holder t_gh;
375 	struct gfs2_log_header_host head;
376 	int error;
377 
378 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
379 	if (error)
380 		return error;
381 
382 	j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
383 
384 	error = gfs2_find_jhead(sdp->sd_jdesc, &head);
385 	if (error)
386 		goto fail;
387 
388 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
389 		gfs2_consist(sdp);
390 		error = -EIO;
391 		goto fail;
392 	}
393 
394 	/*  Initialize some head of the log stuff  */
395 	sdp->sd_log_sequence = head.lh_sequence + 1;
396 	gfs2_log_pointers_init(sdp, head.lh_blkno);
397 
398 	error = gfs2_quota_init(sdp);
399 	if (error)
400 		goto fail;
401 
402 	set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
403 
404 	gfs2_glock_dq_uninit(&t_gh);
405 
406 	return 0;
407 
408 fail:
409 	t_gh.gh_flags |= GL_NOCACHE;
410 	gfs2_glock_dq_uninit(&t_gh);
411 
412 	return error;
413 }
414 
415 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
416 {
417 	const struct gfs2_statfs_change *str = buf;
418 
419 	sc->sc_total = be64_to_cpu(str->sc_total);
420 	sc->sc_free = be64_to_cpu(str->sc_free);
421 	sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
422 }
423 
424 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
425 {
426 	struct gfs2_statfs_change *str = buf;
427 
428 	str->sc_total = cpu_to_be64(sc->sc_total);
429 	str->sc_free = cpu_to_be64(sc->sc_free);
430 	str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
431 }
432 
433 int gfs2_statfs_init(struct gfs2_sbd *sdp)
434 {
435 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
436 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
437 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
438 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
439 	struct buffer_head *m_bh, *l_bh;
440 	struct gfs2_holder gh;
441 	int error;
442 
443 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
444 				   &gh);
445 	if (error)
446 		return error;
447 
448 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
449 	if (error)
450 		goto out;
451 
452 	if (sdp->sd_args.ar_spectator) {
453 		spin_lock(&sdp->sd_statfs_spin);
454 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
455 				      sizeof(struct gfs2_dinode));
456 		spin_unlock(&sdp->sd_statfs_spin);
457 	} else {
458 		error = gfs2_meta_inode_buffer(l_ip, &l_bh);
459 		if (error)
460 			goto out_m_bh;
461 
462 		spin_lock(&sdp->sd_statfs_spin);
463 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
464 				      sizeof(struct gfs2_dinode));
465 		gfs2_statfs_change_in(l_sc, l_bh->b_data +
466 				      sizeof(struct gfs2_dinode));
467 		spin_unlock(&sdp->sd_statfs_spin);
468 
469 		brelse(l_bh);
470 	}
471 
472 out_m_bh:
473 	brelse(m_bh);
474 out:
475 	gfs2_glock_dq_uninit(&gh);
476 	return 0;
477 }
478 
479 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
480 			s64 dinodes)
481 {
482 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
483 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
484 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
485 	struct buffer_head *l_bh;
486 	s64 x, y;
487 	int need_sync = 0;
488 	int error;
489 
490 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
491 	if (error)
492 		return;
493 
494 	gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
495 
496 	spin_lock(&sdp->sd_statfs_spin);
497 	l_sc->sc_total += total;
498 	l_sc->sc_free += free;
499 	l_sc->sc_dinodes += dinodes;
500 	gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
501 	if (sdp->sd_args.ar_statfs_percent) {
502 		x = 100 * l_sc->sc_free;
503 		y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
504 		if (x >= y || x <= -y)
505 			need_sync = 1;
506 	}
507 	spin_unlock(&sdp->sd_statfs_spin);
508 
509 	brelse(l_bh);
510 	if (need_sync)
511 		gfs2_wake_up_statfs(sdp);
512 }
513 
514 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
515 		   struct buffer_head *l_bh)
516 {
517 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
518 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
519 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
520 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
521 
522 	gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
523 
524 	spin_lock(&sdp->sd_statfs_spin);
525 	m_sc->sc_total += l_sc->sc_total;
526 	m_sc->sc_free += l_sc->sc_free;
527 	m_sc->sc_dinodes += l_sc->sc_dinodes;
528 	memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
529 	memset(l_bh->b_data + sizeof(struct gfs2_dinode),
530 	       0, sizeof(struct gfs2_statfs_change));
531 	spin_unlock(&sdp->sd_statfs_spin);
532 
533 	gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
534 	gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
535 }
536 
537 int gfs2_statfs_sync(struct super_block *sb, int type)
538 {
539 	struct gfs2_sbd *sdp = sb->s_fs_info;
540 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
541 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
542 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
543 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
544 	struct gfs2_holder gh;
545 	struct buffer_head *m_bh, *l_bh;
546 	int error;
547 
548 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
549 				   &gh);
550 	if (error)
551 		return error;
552 
553 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
554 	if (error)
555 		goto out;
556 
557 	spin_lock(&sdp->sd_statfs_spin);
558 	gfs2_statfs_change_in(m_sc, m_bh->b_data +
559 			      sizeof(struct gfs2_dinode));
560 	if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
561 		spin_unlock(&sdp->sd_statfs_spin);
562 		goto out_bh;
563 	}
564 	spin_unlock(&sdp->sd_statfs_spin);
565 
566 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
567 	if (error)
568 		goto out_bh;
569 
570 	error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
571 	if (error)
572 		goto out_bh2;
573 
574 	update_statfs(sdp, m_bh, l_bh);
575 	sdp->sd_statfs_force_sync = 0;
576 
577 	gfs2_trans_end(sdp);
578 
579 out_bh2:
580 	brelse(l_bh);
581 out_bh:
582 	brelse(m_bh);
583 out:
584 	gfs2_glock_dq_uninit(&gh);
585 	return error;
586 }
587 
588 struct lfcc {
589 	struct list_head list;
590 	struct gfs2_holder gh;
591 };
592 
593 /**
594  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
595  *                            journals are clean
596  * @sdp: the file system
597  * @state: the state to put the transaction lock into
598  * @t_gh: the hold on the transaction lock
599  *
600  * Returns: errno
601  */
602 
603 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
604 				    struct gfs2_holder *t_gh)
605 {
606 	struct gfs2_inode *ip;
607 	struct gfs2_jdesc *jd;
608 	struct lfcc *lfcc;
609 	LIST_HEAD(list);
610 	struct gfs2_log_header_host lh;
611 	int error;
612 
613 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
614 		lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
615 		if (!lfcc) {
616 			error = -ENOMEM;
617 			goto out;
618 		}
619 		ip = GFS2_I(jd->jd_inode);
620 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
621 		if (error) {
622 			kfree(lfcc);
623 			goto out;
624 		}
625 		list_add(&lfcc->list, &list);
626 	}
627 
628 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
629 				   GL_NOCACHE, t_gh);
630 
631 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
632 		error = gfs2_jdesc_check(jd);
633 		if (error)
634 			break;
635 		error = gfs2_find_jhead(jd, &lh);
636 		if (error)
637 			break;
638 		if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
639 			error = -EBUSY;
640 			break;
641 		}
642 	}
643 
644 	if (error)
645 		gfs2_glock_dq_uninit(t_gh);
646 
647 out:
648 	while (!list_empty(&list)) {
649 		lfcc = list_entry(list.next, struct lfcc, list);
650 		list_del(&lfcc->list);
651 		gfs2_glock_dq_uninit(&lfcc->gh);
652 		kfree(lfcc);
653 	}
654 	return error;
655 }
656 
657 /**
658  * gfs2_freeze_fs - freezes the file system
659  * @sdp: the file system
660  *
661  * This function flushes data and meta data for all machines by
662  * aquiring the transaction log exclusively.  All journals are
663  * ensured to be in a clean state as well.
664  *
665  * Returns: errno
666  */
667 
668 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
669 {
670 	int error = 0;
671 
672 	mutex_lock(&sdp->sd_freeze_lock);
673 
674 	if (!sdp->sd_freeze_count++) {
675 		error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
676 		if (error)
677 			sdp->sd_freeze_count--;
678 	}
679 
680 	mutex_unlock(&sdp->sd_freeze_lock);
681 
682 	return error;
683 }
684 
685 /**
686  * gfs2_unfreeze_fs - unfreezes the file system
687  * @sdp: the file system
688  *
689  * This function allows the file system to proceed by unlocking
690  * the exclusively held transaction lock.  Other GFS2 nodes are
691  * now free to acquire the lock shared and go on with their lives.
692  *
693  */
694 
695 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
696 {
697 	mutex_lock(&sdp->sd_freeze_lock);
698 
699 	if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
700 		gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
701 
702 	mutex_unlock(&sdp->sd_freeze_lock);
703 }
704 
705 
706 /**
707  * gfs2_write_inode - Make sure the inode is stable on the disk
708  * @inode: The inode
709  * @sync: synchronous write flag
710  *
711  * Returns: errno
712  */
713 
714 static int gfs2_write_inode(struct inode *inode, int sync)
715 {
716 	struct gfs2_inode *ip = GFS2_I(inode);
717 	struct gfs2_sbd *sdp = GFS2_SB(inode);
718 	struct gfs2_holder gh;
719 	struct buffer_head *bh;
720 	struct timespec atime;
721 	struct gfs2_dinode *di;
722 	int ret = 0;
723 
724 	/* Check this is a "normal" inode, etc */
725 	if (current->flags & PF_MEMALLOC)
726 		return 0;
727 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
728 	if (ret)
729 		goto do_flush;
730 	ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
731 	if (ret)
732 		goto do_unlock;
733 	ret = gfs2_meta_inode_buffer(ip, &bh);
734 	if (ret == 0) {
735 		di = (struct gfs2_dinode *)bh->b_data;
736 		atime.tv_sec = be64_to_cpu(di->di_atime);
737 		atime.tv_nsec = be32_to_cpu(di->di_atime_nsec);
738 		if (timespec_compare(&inode->i_atime, &atime) > 0) {
739 			gfs2_trans_add_bh(ip->i_gl, bh, 1);
740 			gfs2_dinode_out(ip, bh->b_data);
741 		}
742 		brelse(bh);
743 	}
744 	gfs2_trans_end(sdp);
745 do_unlock:
746 	gfs2_glock_dq_uninit(&gh);
747 do_flush:
748 	if (sync != 0)
749 		gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
750 	return ret;
751 }
752 
753 /**
754  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
755  * @sdp: the filesystem
756  *
757  * Returns: errno
758  */
759 
760 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
761 {
762 	struct gfs2_holder t_gh;
763 	int error;
764 
765 	flush_workqueue(gfs2_delete_workqueue);
766 	gfs2_quota_sync(sdp->sd_vfs, 0);
767 	gfs2_statfs_sync(sdp->sd_vfs, 0);
768 
769 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
770 				   &t_gh);
771 	if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
772 		return error;
773 
774 	gfs2_meta_syncfs(sdp);
775 	gfs2_log_shutdown(sdp);
776 
777 	clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
778 
779 	if (t_gh.gh_gl)
780 		gfs2_glock_dq_uninit(&t_gh);
781 
782 	gfs2_quota_cleanup(sdp);
783 
784 	return error;
785 }
786 
787 static int gfs2_umount_recovery_wait(void *word)
788 {
789 	schedule();
790 	return 0;
791 }
792 
793 /**
794  * gfs2_put_super - Unmount the filesystem
795  * @sb: The VFS superblock
796  *
797  */
798 
799 static void gfs2_put_super(struct super_block *sb)
800 {
801 	struct gfs2_sbd *sdp = sb->s_fs_info;
802 	int error;
803 	struct gfs2_jdesc *jd;
804 
805 	/*  Unfreeze the filesystem, if we need to  */
806 
807 	mutex_lock(&sdp->sd_freeze_lock);
808 	if (sdp->sd_freeze_count)
809 		gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
810 	mutex_unlock(&sdp->sd_freeze_lock);
811 
812 	/* No more recovery requests */
813 	set_bit(SDF_NORECOVERY, &sdp->sd_flags);
814 	smp_mb();
815 
816 	/* Wait on outstanding recovery */
817 restart:
818 	spin_lock(&sdp->sd_jindex_spin);
819 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
820 		if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
821 			continue;
822 		spin_unlock(&sdp->sd_jindex_spin);
823 		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
824 			    gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
825 		goto restart;
826 	}
827 	spin_unlock(&sdp->sd_jindex_spin);
828 
829 	kthread_stop(sdp->sd_quotad_process);
830 	kthread_stop(sdp->sd_logd_process);
831 
832 	if (!(sb->s_flags & MS_RDONLY)) {
833 		error = gfs2_make_fs_ro(sdp);
834 		if (error)
835 			gfs2_io_error(sdp);
836 	}
837 	/*  At this point, we're through modifying the disk  */
838 
839 	/*  Release stuff  */
840 
841 	iput(sdp->sd_jindex);
842 	iput(sdp->sd_statfs_inode);
843 	iput(sdp->sd_rindex);
844 	iput(sdp->sd_quota_inode);
845 
846 	gfs2_glock_put(sdp->sd_rename_gl);
847 	gfs2_glock_put(sdp->sd_trans_gl);
848 
849 	if (!sdp->sd_args.ar_spectator) {
850 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
851 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
852 		gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
853 		gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
854 		iput(sdp->sd_sc_inode);
855 		iput(sdp->sd_qc_inode);
856 	}
857 
858 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
859 	gfs2_clear_rgrpd(sdp);
860 	gfs2_jindex_free(sdp);
861 	/*  Take apart glock structures and buffer lists  */
862 	gfs2_gl_hash_clear(sdp);
863 	/*  Unmount the locking protocol  */
864 	gfs2_lm_unmount(sdp);
865 
866 	/*  At this point, we're through participating in the lockspace  */
867 	gfs2_sys_fs_del(sdp);
868 }
869 
870 /**
871  * gfs2_sync_fs - sync the filesystem
872  * @sb: the superblock
873  *
874  * Flushes the log to disk.
875  */
876 
877 static int gfs2_sync_fs(struct super_block *sb, int wait)
878 {
879 	if (wait && sb->s_fs_info)
880 		gfs2_log_flush(sb->s_fs_info, NULL);
881 	return 0;
882 }
883 
884 /**
885  * gfs2_freeze - prevent further writes to the filesystem
886  * @sb: the VFS structure for the filesystem
887  *
888  */
889 
890 static int gfs2_freeze(struct super_block *sb)
891 {
892 	struct gfs2_sbd *sdp = sb->s_fs_info;
893 	int error;
894 
895 	if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
896 		return -EINVAL;
897 
898 	for (;;) {
899 		error = gfs2_freeze_fs(sdp);
900 		if (!error)
901 			break;
902 
903 		switch (error) {
904 		case -EBUSY:
905 			fs_err(sdp, "waiting for recovery before freeze\n");
906 			break;
907 
908 		default:
909 			fs_err(sdp, "error freezing FS: %d\n", error);
910 			break;
911 		}
912 
913 		fs_err(sdp, "retrying...\n");
914 		msleep(1000);
915 	}
916 	return 0;
917 }
918 
919 /**
920  * gfs2_unfreeze - reallow writes to the filesystem
921  * @sb: the VFS structure for the filesystem
922  *
923  */
924 
925 static int gfs2_unfreeze(struct super_block *sb)
926 {
927 	gfs2_unfreeze_fs(sb->s_fs_info);
928 	return 0;
929 }
930 
931 /**
932  * statfs_fill - fill in the sg for a given RG
933  * @rgd: the RG
934  * @sc: the sc structure
935  *
936  * Returns: 0 on success, -ESTALE if the LVB is invalid
937  */
938 
939 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
940 			    struct gfs2_statfs_change_host *sc)
941 {
942 	gfs2_rgrp_verify(rgd);
943 	sc->sc_total += rgd->rd_data;
944 	sc->sc_free += rgd->rd_free;
945 	sc->sc_dinodes += rgd->rd_dinodes;
946 	return 0;
947 }
948 
949 /**
950  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
951  * @sdp: the filesystem
952  * @sc: the sc info that will be returned
953  *
954  * Any error (other than a signal) will cause this routine to fall back
955  * to the synchronous version.
956  *
957  * FIXME: This really shouldn't busy wait like this.
958  *
959  * Returns: errno
960  */
961 
962 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
963 {
964 	struct gfs2_holder ri_gh;
965 	struct gfs2_rgrpd *rgd_next;
966 	struct gfs2_holder *gha, *gh;
967 	unsigned int slots = 64;
968 	unsigned int x;
969 	int done;
970 	int error = 0, err;
971 
972 	memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
973 	gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
974 	if (!gha)
975 		return -ENOMEM;
976 
977 	error = gfs2_rindex_hold(sdp, &ri_gh);
978 	if (error)
979 		goto out;
980 
981 	rgd_next = gfs2_rgrpd_get_first(sdp);
982 
983 	for (;;) {
984 		done = 1;
985 
986 		for (x = 0; x < slots; x++) {
987 			gh = gha + x;
988 
989 			if (gh->gh_gl && gfs2_glock_poll(gh)) {
990 				err = gfs2_glock_wait(gh);
991 				if (err) {
992 					gfs2_holder_uninit(gh);
993 					error = err;
994 				} else {
995 					if (!error)
996 						error = statfs_slow_fill(
997 							gh->gh_gl->gl_object, sc);
998 					gfs2_glock_dq_uninit(gh);
999 				}
1000 			}
1001 
1002 			if (gh->gh_gl)
1003 				done = 0;
1004 			else if (rgd_next && !error) {
1005 				error = gfs2_glock_nq_init(rgd_next->rd_gl,
1006 							   LM_ST_SHARED,
1007 							   GL_ASYNC,
1008 							   gh);
1009 				rgd_next = gfs2_rgrpd_get_next(rgd_next);
1010 				done = 0;
1011 			}
1012 
1013 			if (signal_pending(current))
1014 				error = -ERESTARTSYS;
1015 		}
1016 
1017 		if (done)
1018 			break;
1019 
1020 		yield();
1021 	}
1022 
1023 	gfs2_glock_dq_uninit(&ri_gh);
1024 
1025 out:
1026 	kfree(gha);
1027 	return error;
1028 }
1029 
1030 /**
1031  * gfs2_statfs_i - Do a statfs
1032  * @sdp: the filesystem
1033  * @sg: the sg structure
1034  *
1035  * Returns: errno
1036  */
1037 
1038 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1039 {
1040 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1041 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1042 
1043 	spin_lock(&sdp->sd_statfs_spin);
1044 
1045 	*sc = *m_sc;
1046 	sc->sc_total += l_sc->sc_total;
1047 	sc->sc_free += l_sc->sc_free;
1048 	sc->sc_dinodes += l_sc->sc_dinodes;
1049 
1050 	spin_unlock(&sdp->sd_statfs_spin);
1051 
1052 	if (sc->sc_free < 0)
1053 		sc->sc_free = 0;
1054 	if (sc->sc_free > sc->sc_total)
1055 		sc->sc_free = sc->sc_total;
1056 	if (sc->sc_dinodes < 0)
1057 		sc->sc_dinodes = 0;
1058 
1059 	return 0;
1060 }
1061 
1062 /**
1063  * gfs2_statfs - Gather and return stats about the filesystem
1064  * @sb: The superblock
1065  * @statfsbuf: The buffer
1066  *
1067  * Returns: 0 on success or error code
1068  */
1069 
1070 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1071 {
1072 	struct super_block *sb = dentry->d_inode->i_sb;
1073 	struct gfs2_sbd *sdp = sb->s_fs_info;
1074 	struct gfs2_statfs_change_host sc;
1075 	int error;
1076 
1077 	if (gfs2_tune_get(sdp, gt_statfs_slow))
1078 		error = gfs2_statfs_slow(sdp, &sc);
1079 	else
1080 		error = gfs2_statfs_i(sdp, &sc);
1081 
1082 	if (error)
1083 		return error;
1084 
1085 	buf->f_type = GFS2_MAGIC;
1086 	buf->f_bsize = sdp->sd_sb.sb_bsize;
1087 	buf->f_blocks = sc.sc_total;
1088 	buf->f_bfree = sc.sc_free;
1089 	buf->f_bavail = sc.sc_free;
1090 	buf->f_files = sc.sc_dinodes + sc.sc_free;
1091 	buf->f_ffree = sc.sc_free;
1092 	buf->f_namelen = GFS2_FNAMESIZE;
1093 
1094 	return 0;
1095 }
1096 
1097 /**
1098  * gfs2_remount_fs - called when the FS is remounted
1099  * @sb:  the filesystem
1100  * @flags:  the remount flags
1101  * @data:  extra data passed in (not used right now)
1102  *
1103  * Returns: errno
1104  */
1105 
1106 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1107 {
1108 	struct gfs2_sbd *sdp = sb->s_fs_info;
1109 	struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1110 	struct gfs2_tune *gt = &sdp->sd_tune;
1111 	int error;
1112 
1113 	spin_lock(&gt->gt_spin);
1114 	args.ar_commit = gt->gt_log_flush_secs;
1115 	args.ar_quota_quantum = gt->gt_quota_quantum;
1116 	if (gt->gt_statfs_slow)
1117 		args.ar_statfs_quantum = 0;
1118 	else
1119 		args.ar_statfs_quantum = gt->gt_statfs_quantum;
1120 	spin_unlock(&gt->gt_spin);
1121 	error = gfs2_mount_args(&args, data);
1122 	if (error)
1123 		return error;
1124 
1125 	/* Not allowed to change locking details */
1126 	if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1127 	    strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1128 	    strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1129 		return -EINVAL;
1130 
1131 	/* Some flags must not be changed */
1132 	if (args_neq(&args, &sdp->sd_args, spectator) ||
1133 	    args_neq(&args, &sdp->sd_args, ignore_local_fs) ||
1134 	    args_neq(&args, &sdp->sd_args, localflocks) ||
1135 	    args_neq(&args, &sdp->sd_args, localcaching) ||
1136 	    args_neq(&args, &sdp->sd_args, meta))
1137 		return -EINVAL;
1138 
1139 	if (sdp->sd_args.ar_spectator)
1140 		*flags |= MS_RDONLY;
1141 
1142 	if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1143 		if (*flags & MS_RDONLY)
1144 			error = gfs2_make_fs_ro(sdp);
1145 		else
1146 			error = gfs2_make_fs_rw(sdp);
1147 		if (error)
1148 			return error;
1149 	}
1150 
1151 	sdp->sd_args = args;
1152 	if (sdp->sd_args.ar_posix_acl)
1153 		sb->s_flags |= MS_POSIXACL;
1154 	else
1155 		sb->s_flags &= ~MS_POSIXACL;
1156 	if (sdp->sd_args.ar_nobarrier)
1157 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1158 	else
1159 		clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1160 	spin_lock(&gt->gt_spin);
1161 	gt->gt_log_flush_secs = args.ar_commit;
1162 	gt->gt_quota_quantum = args.ar_quota_quantum;
1163 	if (args.ar_statfs_quantum) {
1164 		gt->gt_statfs_slow = 0;
1165 		gt->gt_statfs_quantum = args.ar_statfs_quantum;
1166 	}
1167 	else {
1168 		gt->gt_statfs_slow = 1;
1169 		gt->gt_statfs_quantum = 30;
1170 	}
1171 	spin_unlock(&gt->gt_spin);
1172 
1173 	gfs2_online_uevent(sdp);
1174 	return 0;
1175 }
1176 
1177 /**
1178  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1179  * @inode: The inode to drop
1180  *
1181  * If we've received a callback on an iopen lock then its because a
1182  * remote node tried to deallocate the inode but failed due to this node
1183  * still having the inode open. Here we mark the link count zero
1184  * since we know that it must have reached zero if the GLF_DEMOTE flag
1185  * is set on the iopen glock. If we didn't do a disk read since the
1186  * remote node removed the final link then we might otherwise miss
1187  * this event. This check ensures that this node will deallocate the
1188  * inode's blocks, or alternatively pass the baton on to another
1189  * node for later deallocation.
1190  */
1191 
1192 static void gfs2_drop_inode(struct inode *inode)
1193 {
1194 	struct gfs2_inode *ip = GFS2_I(inode);
1195 
1196 	if (inode->i_nlink) {
1197 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1198 		if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1199 			clear_nlink(inode);
1200 	}
1201 	generic_drop_inode(inode);
1202 }
1203 
1204 /**
1205  * gfs2_clear_inode - Deallocate an inode when VFS is done with it
1206  * @inode: The VFS inode
1207  *
1208  */
1209 
1210 static void gfs2_clear_inode(struct inode *inode)
1211 {
1212 	struct gfs2_inode *ip = GFS2_I(inode);
1213 
1214 	ip->i_gl->gl_object = NULL;
1215 	gfs2_glock_put(ip->i_gl);
1216 	ip->i_gl = NULL;
1217 	if (ip->i_iopen_gh.gh_gl) {
1218 		ip->i_iopen_gh.gh_gl->gl_object = NULL;
1219 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1220 	}
1221 }
1222 
1223 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1224 {
1225 	do {
1226 		if (d1 == d2)
1227 			return 1;
1228 		d1 = d1->d_parent;
1229 	} while (!IS_ROOT(d1));
1230 	return 0;
1231 }
1232 
1233 /**
1234  * gfs2_show_options - Show mount options for /proc/mounts
1235  * @s: seq_file structure
1236  * @mnt: vfsmount
1237  *
1238  * Returns: 0 on success or error code
1239  */
1240 
1241 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1242 {
1243 	struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
1244 	struct gfs2_args *args = &sdp->sd_args;
1245 	int val;
1246 
1247 	if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir))
1248 		seq_printf(s, ",meta");
1249 	if (args->ar_lockproto[0])
1250 		seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1251 	if (args->ar_locktable[0])
1252 		seq_printf(s, ",locktable=%s", args->ar_locktable);
1253 	if (args->ar_hostdata[0])
1254 		seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1255 	if (args->ar_spectator)
1256 		seq_printf(s, ",spectator");
1257 	if (args->ar_ignore_local_fs)
1258 		seq_printf(s, ",ignore_local_fs");
1259 	if (args->ar_localflocks)
1260 		seq_printf(s, ",localflocks");
1261 	if (args->ar_localcaching)
1262 		seq_printf(s, ",localcaching");
1263 	if (args->ar_debug)
1264 		seq_printf(s, ",debug");
1265 	if (args->ar_upgrade)
1266 		seq_printf(s, ",upgrade");
1267 	if (args->ar_posix_acl)
1268 		seq_printf(s, ",acl");
1269 	if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1270 		char *state;
1271 		switch (args->ar_quota) {
1272 		case GFS2_QUOTA_OFF:
1273 			state = "off";
1274 			break;
1275 		case GFS2_QUOTA_ACCOUNT:
1276 			state = "account";
1277 			break;
1278 		case GFS2_QUOTA_ON:
1279 			state = "on";
1280 			break;
1281 		default:
1282 			state = "unknown";
1283 			break;
1284 		}
1285 		seq_printf(s, ",quota=%s", state);
1286 	}
1287 	if (args->ar_suiddir)
1288 		seq_printf(s, ",suiddir");
1289 	if (args->ar_data != GFS2_DATA_DEFAULT) {
1290 		char *state;
1291 		switch (args->ar_data) {
1292 		case GFS2_DATA_WRITEBACK:
1293 			state = "writeback";
1294 			break;
1295 		case GFS2_DATA_ORDERED:
1296 			state = "ordered";
1297 			break;
1298 		default:
1299 			state = "unknown";
1300 			break;
1301 		}
1302 		seq_printf(s, ",data=%s", state);
1303 	}
1304 	if (args->ar_discard)
1305 		seq_printf(s, ",discard");
1306 	val = sdp->sd_tune.gt_log_flush_secs;
1307 	if (val != 60)
1308 		seq_printf(s, ",commit=%d", val);
1309 	val = sdp->sd_tune.gt_statfs_quantum;
1310 	if (val != 30)
1311 		seq_printf(s, ",statfs_quantum=%d", val);
1312 	val = sdp->sd_tune.gt_quota_quantum;
1313 	if (val != 60)
1314 		seq_printf(s, ",quota_quantum=%d", val);
1315 	if (args->ar_statfs_percent)
1316 		seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1317 	if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1318 		const char *state;
1319 
1320 		switch (args->ar_errors) {
1321 		case GFS2_ERRORS_WITHDRAW:
1322 			state = "withdraw";
1323 			break;
1324 		case GFS2_ERRORS_PANIC:
1325 			state = "panic";
1326 			break;
1327 		default:
1328 			state = "unknown";
1329 			break;
1330 		}
1331 		seq_printf(s, ",errors=%s", state);
1332 	}
1333 	if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1334 		seq_printf(s, ",nobarrier");
1335 
1336 	return 0;
1337 }
1338 
1339 /*
1340  * We have to (at the moment) hold the inodes main lock to cover
1341  * the gap between unlocking the shared lock on the iopen lock and
1342  * taking the exclusive lock. I'd rather do a shared -> exclusive
1343  * conversion on the iopen lock, but we can change that later. This
1344  * is safe, just less efficient.
1345  */
1346 
1347 static void gfs2_delete_inode(struct inode *inode)
1348 {
1349 	struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
1350 	struct gfs2_inode *ip = GFS2_I(inode);
1351 	struct gfs2_holder gh;
1352 	int error;
1353 
1354 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1355 	if (unlikely(error)) {
1356 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1357 		goto out;
1358 	}
1359 
1360 	error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1361 	if (error)
1362 		goto out_truncate;
1363 
1364 	gfs2_glock_dq_wait(&ip->i_iopen_gh);
1365 	gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1366 	error = gfs2_glock_nq(&ip->i_iopen_gh);
1367 	if (error)
1368 		goto out_truncate;
1369 
1370 	if (S_ISDIR(inode->i_mode) &&
1371 	    (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1372 		error = gfs2_dir_exhash_dealloc(ip);
1373 		if (error)
1374 			goto out_unlock;
1375 	}
1376 
1377 	if (ip->i_eattr) {
1378 		error = gfs2_ea_dealloc(ip);
1379 		if (error)
1380 			goto out_unlock;
1381 	}
1382 
1383 	if (!gfs2_is_stuffed(ip)) {
1384 		error = gfs2_file_dealloc(ip);
1385 		if (error)
1386 			goto out_unlock;
1387 	}
1388 
1389 	error = gfs2_dinode_dealloc(ip);
1390 	if (error)
1391 		goto out_unlock;
1392 
1393 out_truncate:
1394 	error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1395 	if (error)
1396 		goto out_unlock;
1397 	/* Needs to be done before glock release & also in a transaction */
1398 	truncate_inode_pages(&inode->i_data, 0);
1399 	gfs2_trans_end(sdp);
1400 
1401 out_unlock:
1402 	if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
1403 		gfs2_glock_dq(&ip->i_iopen_gh);
1404 	gfs2_holder_uninit(&ip->i_iopen_gh);
1405 	gfs2_glock_dq_uninit(&gh);
1406 	if (error && error != GLR_TRYFAILED && error != -EROFS)
1407 		fs_warn(sdp, "gfs2_delete_inode: %d\n", error);
1408 out:
1409 	truncate_inode_pages(&inode->i_data, 0);
1410 	clear_inode(inode);
1411 }
1412 
1413 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1414 {
1415 	struct gfs2_inode *ip;
1416 
1417 	ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1418 	if (ip) {
1419 		ip->i_flags = 0;
1420 		ip->i_gl = NULL;
1421 	}
1422 	return &ip->i_inode;
1423 }
1424 
1425 static void gfs2_destroy_inode(struct inode *inode)
1426 {
1427 	kmem_cache_free(gfs2_inode_cachep, inode);
1428 }
1429 
1430 const struct super_operations gfs2_super_ops = {
1431 	.alloc_inode		= gfs2_alloc_inode,
1432 	.destroy_inode		= gfs2_destroy_inode,
1433 	.write_inode		= gfs2_write_inode,
1434 	.delete_inode		= gfs2_delete_inode,
1435 	.put_super		= gfs2_put_super,
1436 	.sync_fs		= gfs2_sync_fs,
1437 	.freeze_fs 		= gfs2_freeze,
1438 	.unfreeze_fs		= gfs2_unfreeze,
1439 	.statfs			= gfs2_statfs,
1440 	.remount_fs		= gfs2_remount_fs,
1441 	.clear_inode		= gfs2_clear_inode,
1442 	.drop_inode		= gfs2_drop_inode,
1443 	.show_options		= gfs2_show_options,
1444 };
1445 
1446