xref: /linux/fs/gfs2/super.c (revision 2277ab4a1df50e05bc732fe9488d4e902bb8399a)
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 
25 #include "gfs2.h"
26 #include "incore.h"
27 #include "bmap.h"
28 #include "dir.h"
29 #include "glock.h"
30 #include "glops.h"
31 #include "inode.h"
32 #include "log.h"
33 #include "meta_io.h"
34 #include "quota.h"
35 #include "recovery.h"
36 #include "rgrp.h"
37 #include "super.h"
38 #include "trans.h"
39 #include "util.h"
40 #include "sys.h"
41 #include "eattr.h"
42 
43 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
44 
45 enum {
46 	Opt_lockproto,
47 	Opt_locktable,
48 	Opt_hostdata,
49 	Opt_spectator,
50 	Opt_ignore_local_fs,
51 	Opt_localflocks,
52 	Opt_localcaching,
53 	Opt_debug,
54 	Opt_nodebug,
55 	Opt_upgrade,
56 	Opt_acl,
57 	Opt_noacl,
58 	Opt_quota_off,
59 	Opt_quota_account,
60 	Opt_quota_on,
61 	Opt_quota,
62 	Opt_noquota,
63 	Opt_suiddir,
64 	Opt_nosuiddir,
65 	Opt_data_writeback,
66 	Opt_data_ordered,
67 	Opt_meta,
68 	Opt_discard,
69 	Opt_nodiscard,
70 	Opt_commit,
71 	Opt_error,
72 };
73 
74 static const match_table_t tokens = {
75 	{Opt_lockproto, "lockproto=%s"},
76 	{Opt_locktable, "locktable=%s"},
77 	{Opt_hostdata, "hostdata=%s"},
78 	{Opt_spectator, "spectator"},
79 	{Opt_ignore_local_fs, "ignore_local_fs"},
80 	{Opt_localflocks, "localflocks"},
81 	{Opt_localcaching, "localcaching"},
82 	{Opt_debug, "debug"},
83 	{Opt_nodebug, "nodebug"},
84 	{Opt_upgrade, "upgrade"},
85 	{Opt_acl, "acl"},
86 	{Opt_noacl, "noacl"},
87 	{Opt_quota_off, "quota=off"},
88 	{Opt_quota_account, "quota=account"},
89 	{Opt_quota_on, "quota=on"},
90 	{Opt_quota, "quota"},
91 	{Opt_noquota, "noquota"},
92 	{Opt_suiddir, "suiddir"},
93 	{Opt_nosuiddir, "nosuiddir"},
94 	{Opt_data_writeback, "data=writeback"},
95 	{Opt_data_ordered, "data=ordered"},
96 	{Opt_meta, "meta"},
97 	{Opt_discard, "discard"},
98 	{Opt_nodiscard, "nodiscard"},
99 	{Opt_commit, "commit=%d"},
100 	{Opt_error, NULL}
101 };
102 
103 /**
104  * gfs2_mount_args - Parse mount options
105  * @sdp:
106  * @data:
107  *
108  * Return: errno
109  */
110 
111 int gfs2_mount_args(struct gfs2_sbd *sdp, struct gfs2_args *args, char *options)
112 {
113 	char *o;
114 	int token;
115 	substring_t tmp[MAX_OPT_ARGS];
116 	int rv;
117 
118 	/* Split the options into tokens with the "," character and
119 	   process them */
120 
121 	while (1) {
122 		o = strsep(&options, ",");
123 		if (o == NULL)
124 			break;
125 		if (*o == '\0')
126 			continue;
127 
128 		token = match_token(o, tokens, tmp);
129 		switch (token) {
130 		case Opt_lockproto:
131 			match_strlcpy(args->ar_lockproto, &tmp[0],
132 				      GFS2_LOCKNAME_LEN);
133 			break;
134 		case Opt_locktable:
135 			match_strlcpy(args->ar_locktable, &tmp[0],
136 				      GFS2_LOCKNAME_LEN);
137 			break;
138 		case Opt_hostdata:
139 			match_strlcpy(args->ar_hostdata, &tmp[0],
140 				      GFS2_LOCKNAME_LEN);
141 			break;
142 		case Opt_spectator:
143 			args->ar_spectator = 1;
144 			break;
145 		case Opt_ignore_local_fs:
146 			args->ar_ignore_local_fs = 1;
147 			break;
148 		case Opt_localflocks:
149 			args->ar_localflocks = 1;
150 			break;
151 		case Opt_localcaching:
152 			args->ar_localcaching = 1;
153 			break;
154 		case Opt_debug:
155 			args->ar_debug = 1;
156 			break;
157 		case Opt_nodebug:
158 			args->ar_debug = 0;
159 			break;
160 		case Opt_upgrade:
161 			args->ar_upgrade = 1;
162 			break;
163 		case Opt_acl:
164 			args->ar_posix_acl = 1;
165 			break;
166 		case Opt_noacl:
167 			args->ar_posix_acl = 0;
168 			break;
169 		case Opt_quota_off:
170 		case Opt_noquota:
171 			args->ar_quota = GFS2_QUOTA_OFF;
172 			break;
173 		case Opt_quota_account:
174 			args->ar_quota = GFS2_QUOTA_ACCOUNT;
175 			break;
176 		case Opt_quota_on:
177 		case Opt_quota:
178 			args->ar_quota = GFS2_QUOTA_ON;
179 			break;
180 		case Opt_suiddir:
181 			args->ar_suiddir = 1;
182 			break;
183 		case Opt_nosuiddir:
184 			args->ar_suiddir = 0;
185 			break;
186 		case Opt_data_writeback:
187 			args->ar_data = GFS2_DATA_WRITEBACK;
188 			break;
189 		case Opt_data_ordered:
190 			args->ar_data = GFS2_DATA_ORDERED;
191 			break;
192 		case Opt_meta:
193 			args->ar_meta = 1;
194 			break;
195 		case Opt_discard:
196 			args->ar_discard = 1;
197 			break;
198 		case Opt_nodiscard:
199 			args->ar_discard = 0;
200 			break;
201 		case Opt_commit:
202 			rv = match_int(&tmp[0], &args->ar_commit);
203 			if (rv || args->ar_commit <= 0) {
204 				fs_info(sdp, "commit mount option requires a positive numeric argument\n");
205 				return rv ? rv : -EINVAL;
206 			}
207 			break;
208 		case Opt_error:
209 		default:
210 			fs_info(sdp, "invalid mount option: %s\n", o);
211 			return -EINVAL;
212 		}
213 	}
214 
215 	return 0;
216 }
217 
218 /**
219  * gfs2_jindex_free - Clear all the journal index information
220  * @sdp: The GFS2 superblock
221  *
222  */
223 
224 void gfs2_jindex_free(struct gfs2_sbd *sdp)
225 {
226 	struct list_head list, *head;
227 	struct gfs2_jdesc *jd;
228 	struct gfs2_journal_extent *jext;
229 
230 	spin_lock(&sdp->sd_jindex_spin);
231 	list_add(&list, &sdp->sd_jindex_list);
232 	list_del_init(&sdp->sd_jindex_list);
233 	sdp->sd_journals = 0;
234 	spin_unlock(&sdp->sd_jindex_spin);
235 
236 	while (!list_empty(&list)) {
237 		jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
238 		head = &jd->extent_list;
239 		while (!list_empty(head)) {
240 			jext = list_entry(head->next,
241 					  struct gfs2_journal_extent,
242 					  extent_list);
243 			list_del(&jext->extent_list);
244 			kfree(jext);
245 		}
246 		list_del(&jd->jd_list);
247 		iput(jd->jd_inode);
248 		kfree(jd);
249 	}
250 }
251 
252 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
253 {
254 	struct gfs2_jdesc *jd;
255 	int found = 0;
256 
257 	list_for_each_entry(jd, head, jd_list) {
258 		if (jd->jd_jid == jid) {
259 			found = 1;
260 			break;
261 		}
262 	}
263 
264 	if (!found)
265 		jd = NULL;
266 
267 	return jd;
268 }
269 
270 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
271 {
272 	struct gfs2_jdesc *jd;
273 
274 	spin_lock(&sdp->sd_jindex_spin);
275 	jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
276 	spin_unlock(&sdp->sd_jindex_spin);
277 
278 	return jd;
279 }
280 
281 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
282 {
283 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
284 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
285 	int ar;
286 	int error;
287 
288 	if (ip->i_disksize < (8 << 20) || ip->i_disksize > (1 << 30) ||
289 	    (ip->i_disksize & (sdp->sd_sb.sb_bsize - 1))) {
290 		gfs2_consist_inode(ip);
291 		return -EIO;
292 	}
293 	jd->jd_blocks = ip->i_disksize >> sdp->sd_sb.sb_bsize_shift;
294 
295 	error = gfs2_write_alloc_required(ip, 0, ip->i_disksize, &ar);
296 	if (!error && ar) {
297 		gfs2_consist_inode(ip);
298 		error = -EIO;
299 	}
300 
301 	return error;
302 }
303 
304 /**
305  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
306  * @sdp: the filesystem
307  *
308  * Returns: errno
309  */
310 
311 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
312 {
313 	struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
314 	struct gfs2_glock *j_gl = ip->i_gl;
315 	struct gfs2_holder t_gh;
316 	struct gfs2_log_header_host head;
317 	int error;
318 
319 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
320 	if (error)
321 		return error;
322 
323 	j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
324 
325 	error = gfs2_find_jhead(sdp->sd_jdesc, &head);
326 	if (error)
327 		goto fail;
328 
329 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
330 		gfs2_consist(sdp);
331 		error = -EIO;
332 		goto fail;
333 	}
334 
335 	/*  Initialize some head of the log stuff  */
336 	sdp->sd_log_sequence = head.lh_sequence + 1;
337 	gfs2_log_pointers_init(sdp, head.lh_blkno);
338 
339 	error = gfs2_quota_init(sdp);
340 	if (error)
341 		goto fail;
342 
343 	set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
344 
345 	gfs2_glock_dq_uninit(&t_gh);
346 
347 	return 0;
348 
349 fail:
350 	t_gh.gh_flags |= GL_NOCACHE;
351 	gfs2_glock_dq_uninit(&t_gh);
352 
353 	return error;
354 }
355 
356 static void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
357 {
358 	const struct gfs2_statfs_change *str = buf;
359 
360 	sc->sc_total = be64_to_cpu(str->sc_total);
361 	sc->sc_free = be64_to_cpu(str->sc_free);
362 	sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
363 }
364 
365 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
366 {
367 	struct gfs2_statfs_change *str = buf;
368 
369 	str->sc_total = cpu_to_be64(sc->sc_total);
370 	str->sc_free = cpu_to_be64(sc->sc_free);
371 	str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
372 }
373 
374 int gfs2_statfs_init(struct gfs2_sbd *sdp)
375 {
376 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
377 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
378 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
379 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
380 	struct buffer_head *m_bh, *l_bh;
381 	struct gfs2_holder gh;
382 	int error;
383 
384 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
385 				   &gh);
386 	if (error)
387 		return error;
388 
389 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
390 	if (error)
391 		goto out;
392 
393 	if (sdp->sd_args.ar_spectator) {
394 		spin_lock(&sdp->sd_statfs_spin);
395 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
396 				      sizeof(struct gfs2_dinode));
397 		spin_unlock(&sdp->sd_statfs_spin);
398 	} else {
399 		error = gfs2_meta_inode_buffer(l_ip, &l_bh);
400 		if (error)
401 			goto out_m_bh;
402 
403 		spin_lock(&sdp->sd_statfs_spin);
404 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
405 				      sizeof(struct gfs2_dinode));
406 		gfs2_statfs_change_in(l_sc, l_bh->b_data +
407 				      sizeof(struct gfs2_dinode));
408 		spin_unlock(&sdp->sd_statfs_spin);
409 
410 		brelse(l_bh);
411 	}
412 
413 out_m_bh:
414 	brelse(m_bh);
415 out:
416 	gfs2_glock_dq_uninit(&gh);
417 	return 0;
418 }
419 
420 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
421 			s64 dinodes)
422 {
423 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
424 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
425 	struct buffer_head *l_bh;
426 	int error;
427 
428 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
429 	if (error)
430 		return;
431 
432 	gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
433 
434 	spin_lock(&sdp->sd_statfs_spin);
435 	l_sc->sc_total += total;
436 	l_sc->sc_free += free;
437 	l_sc->sc_dinodes += dinodes;
438 	gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
439 	spin_unlock(&sdp->sd_statfs_spin);
440 
441 	brelse(l_bh);
442 }
443 
444 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
445 {
446 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
447 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
448 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
449 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
450 	struct gfs2_holder gh;
451 	struct buffer_head *m_bh, *l_bh;
452 	int error;
453 
454 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
455 				   &gh);
456 	if (error)
457 		return error;
458 
459 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
460 	if (error)
461 		goto out;
462 
463 	spin_lock(&sdp->sd_statfs_spin);
464 	gfs2_statfs_change_in(m_sc, m_bh->b_data +
465 			      sizeof(struct gfs2_dinode));
466 	if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
467 		spin_unlock(&sdp->sd_statfs_spin);
468 		goto out_bh;
469 	}
470 	spin_unlock(&sdp->sd_statfs_spin);
471 
472 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
473 	if (error)
474 		goto out_bh;
475 
476 	error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
477 	if (error)
478 		goto out_bh2;
479 
480 	gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
481 
482 	spin_lock(&sdp->sd_statfs_spin);
483 	m_sc->sc_total += l_sc->sc_total;
484 	m_sc->sc_free += l_sc->sc_free;
485 	m_sc->sc_dinodes += l_sc->sc_dinodes;
486 	memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
487 	memset(l_bh->b_data + sizeof(struct gfs2_dinode),
488 	       0, sizeof(struct gfs2_statfs_change));
489 	spin_unlock(&sdp->sd_statfs_spin);
490 
491 	gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
492 	gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
493 
494 	gfs2_trans_end(sdp);
495 
496 out_bh2:
497 	brelse(l_bh);
498 out_bh:
499 	brelse(m_bh);
500 out:
501 	gfs2_glock_dq_uninit(&gh);
502 	return error;
503 }
504 
505 struct lfcc {
506 	struct list_head list;
507 	struct gfs2_holder gh;
508 };
509 
510 /**
511  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
512  *                            journals are clean
513  * @sdp: the file system
514  * @state: the state to put the transaction lock into
515  * @t_gh: the hold on the transaction lock
516  *
517  * Returns: errno
518  */
519 
520 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
521 				    struct gfs2_holder *t_gh)
522 {
523 	struct gfs2_inode *ip;
524 	struct gfs2_jdesc *jd;
525 	struct lfcc *lfcc;
526 	LIST_HEAD(list);
527 	struct gfs2_log_header_host lh;
528 	int error;
529 
530 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
531 		lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
532 		if (!lfcc) {
533 			error = -ENOMEM;
534 			goto out;
535 		}
536 		ip = GFS2_I(jd->jd_inode);
537 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
538 		if (error) {
539 			kfree(lfcc);
540 			goto out;
541 		}
542 		list_add(&lfcc->list, &list);
543 	}
544 
545 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
546 				   GL_NOCACHE, t_gh);
547 
548 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
549 		error = gfs2_jdesc_check(jd);
550 		if (error)
551 			break;
552 		error = gfs2_find_jhead(jd, &lh);
553 		if (error)
554 			break;
555 		if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
556 			error = -EBUSY;
557 			break;
558 		}
559 	}
560 
561 	if (error)
562 		gfs2_glock_dq_uninit(t_gh);
563 
564 out:
565 	while (!list_empty(&list)) {
566 		lfcc = list_entry(list.next, struct lfcc, list);
567 		list_del(&lfcc->list);
568 		gfs2_glock_dq_uninit(&lfcc->gh);
569 		kfree(lfcc);
570 	}
571 	return error;
572 }
573 
574 /**
575  * gfs2_freeze_fs - freezes the file system
576  * @sdp: the file system
577  *
578  * This function flushes data and meta data for all machines by
579  * aquiring the transaction log exclusively.  All journals are
580  * ensured to be in a clean state as well.
581  *
582  * Returns: errno
583  */
584 
585 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
586 {
587 	int error = 0;
588 
589 	mutex_lock(&sdp->sd_freeze_lock);
590 
591 	if (!sdp->sd_freeze_count++) {
592 		error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
593 		if (error)
594 			sdp->sd_freeze_count--;
595 	}
596 
597 	mutex_unlock(&sdp->sd_freeze_lock);
598 
599 	return error;
600 }
601 
602 /**
603  * gfs2_unfreeze_fs - unfreezes the file system
604  * @sdp: the file system
605  *
606  * This function allows the file system to proceed by unlocking
607  * the exclusively held transaction lock.  Other GFS2 nodes are
608  * now free to acquire the lock shared and go on with their lives.
609  *
610  */
611 
612 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
613 {
614 	mutex_lock(&sdp->sd_freeze_lock);
615 
616 	if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
617 		gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
618 
619 	mutex_unlock(&sdp->sd_freeze_lock);
620 }
621 
622 
623 /**
624  * gfs2_write_inode - Make sure the inode is stable on the disk
625  * @inode: The inode
626  * @sync: synchronous write flag
627  *
628  * Returns: errno
629  */
630 
631 static int gfs2_write_inode(struct inode *inode, int sync)
632 {
633 	struct gfs2_inode *ip = GFS2_I(inode);
634 	struct gfs2_sbd *sdp = GFS2_SB(inode);
635 	struct gfs2_holder gh;
636 	struct buffer_head *bh;
637 	struct timespec atime;
638 	struct gfs2_dinode *di;
639 	int ret = 0;
640 
641 	/* Check this is a "normal" inode, etc */
642 	if (!test_bit(GIF_USER, &ip->i_flags) ||
643 	    (current->flags & PF_MEMALLOC))
644 		return 0;
645 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
646 	if (ret)
647 		goto do_flush;
648 	ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
649 	if (ret)
650 		goto do_unlock;
651 	ret = gfs2_meta_inode_buffer(ip, &bh);
652 	if (ret == 0) {
653 		di = (struct gfs2_dinode *)bh->b_data;
654 		atime.tv_sec = be64_to_cpu(di->di_atime);
655 		atime.tv_nsec = be32_to_cpu(di->di_atime_nsec);
656 		if (timespec_compare(&inode->i_atime, &atime) > 0) {
657 			gfs2_trans_add_bh(ip->i_gl, bh, 1);
658 			gfs2_dinode_out(ip, bh->b_data);
659 		}
660 		brelse(bh);
661 	}
662 	gfs2_trans_end(sdp);
663 do_unlock:
664 	gfs2_glock_dq_uninit(&gh);
665 do_flush:
666 	if (sync != 0)
667 		gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
668 	return ret;
669 }
670 
671 /**
672  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
673  * @sdp: the filesystem
674  *
675  * Returns: errno
676  */
677 
678 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
679 {
680 	struct gfs2_holder t_gh;
681 	int error;
682 
683 	gfs2_quota_sync(sdp);
684 	gfs2_statfs_sync(sdp);
685 
686 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
687 				   &t_gh);
688 	if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
689 		return error;
690 
691 	gfs2_meta_syncfs(sdp);
692 	gfs2_log_shutdown(sdp);
693 
694 	clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
695 
696 	if (t_gh.gh_gl)
697 		gfs2_glock_dq_uninit(&t_gh);
698 
699 	gfs2_quota_cleanup(sdp);
700 
701 	return error;
702 }
703 
704 static int gfs2_umount_recovery_wait(void *word)
705 {
706 	schedule();
707 	return 0;
708 }
709 
710 /**
711  * gfs2_put_super - Unmount the filesystem
712  * @sb: The VFS superblock
713  *
714  */
715 
716 static void gfs2_put_super(struct super_block *sb)
717 {
718 	struct gfs2_sbd *sdp = sb->s_fs_info;
719 	int error;
720 	struct gfs2_jdesc *jd;
721 
722 	/*  Unfreeze the filesystem, if we need to  */
723 
724 	mutex_lock(&sdp->sd_freeze_lock);
725 	if (sdp->sd_freeze_count)
726 		gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
727 	mutex_unlock(&sdp->sd_freeze_lock);
728 
729 	/* No more recovery requests */
730 	set_bit(SDF_NORECOVERY, &sdp->sd_flags);
731 	smp_mb();
732 
733 	/* Wait on outstanding recovery */
734 restart:
735 	spin_lock(&sdp->sd_jindex_spin);
736 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
737 		if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
738 			continue;
739 		spin_unlock(&sdp->sd_jindex_spin);
740 		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
741 			    gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
742 		goto restart;
743 	}
744 	spin_unlock(&sdp->sd_jindex_spin);
745 
746 	kthread_stop(sdp->sd_quotad_process);
747 	kthread_stop(sdp->sd_logd_process);
748 
749 	if (!(sb->s_flags & MS_RDONLY)) {
750 		error = gfs2_make_fs_ro(sdp);
751 		if (error)
752 			gfs2_io_error(sdp);
753 	}
754 	/*  At this point, we're through modifying the disk  */
755 
756 	/*  Release stuff  */
757 
758 	iput(sdp->sd_jindex);
759 	iput(sdp->sd_inum_inode);
760 	iput(sdp->sd_statfs_inode);
761 	iput(sdp->sd_rindex);
762 	iput(sdp->sd_quota_inode);
763 
764 	gfs2_glock_put(sdp->sd_rename_gl);
765 	gfs2_glock_put(sdp->sd_trans_gl);
766 
767 	if (!sdp->sd_args.ar_spectator) {
768 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
769 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
770 		gfs2_glock_dq_uninit(&sdp->sd_ir_gh);
771 		gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
772 		gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
773 		iput(sdp->sd_ir_inode);
774 		iput(sdp->sd_sc_inode);
775 		iput(sdp->sd_qc_inode);
776 	}
777 
778 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
779 	gfs2_clear_rgrpd(sdp);
780 	gfs2_jindex_free(sdp);
781 	/*  Take apart glock structures and buffer lists  */
782 	gfs2_gl_hash_clear(sdp);
783 	/*  Unmount the locking protocol  */
784 	gfs2_lm_unmount(sdp);
785 
786 	/*  At this point, we're through participating in the lockspace  */
787 	gfs2_sys_fs_del(sdp);
788 }
789 
790 /**
791  * gfs2_sync_fs - sync the filesystem
792  * @sb: the superblock
793  *
794  * Flushes the log to disk.
795  */
796 
797 static int gfs2_sync_fs(struct super_block *sb, int wait)
798 {
799 	if (wait && sb->s_fs_info)
800 		gfs2_log_flush(sb->s_fs_info, NULL);
801 	return 0;
802 }
803 
804 /**
805  * gfs2_freeze - prevent further writes to the filesystem
806  * @sb: the VFS structure for the filesystem
807  *
808  */
809 
810 static int gfs2_freeze(struct super_block *sb)
811 {
812 	struct gfs2_sbd *sdp = sb->s_fs_info;
813 	int error;
814 
815 	if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
816 		return -EINVAL;
817 
818 	for (;;) {
819 		error = gfs2_freeze_fs(sdp);
820 		if (!error)
821 			break;
822 
823 		switch (error) {
824 		case -EBUSY:
825 			fs_err(sdp, "waiting for recovery before freeze\n");
826 			break;
827 
828 		default:
829 			fs_err(sdp, "error freezing FS: %d\n", error);
830 			break;
831 		}
832 
833 		fs_err(sdp, "retrying...\n");
834 		msleep(1000);
835 	}
836 	return 0;
837 }
838 
839 /**
840  * gfs2_unfreeze - reallow writes to the filesystem
841  * @sb: the VFS structure for the filesystem
842  *
843  */
844 
845 static int gfs2_unfreeze(struct super_block *sb)
846 {
847 	gfs2_unfreeze_fs(sb->s_fs_info);
848 	return 0;
849 }
850 
851 /**
852  * statfs_fill - fill in the sg for a given RG
853  * @rgd: the RG
854  * @sc: the sc structure
855  *
856  * Returns: 0 on success, -ESTALE if the LVB is invalid
857  */
858 
859 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
860 			    struct gfs2_statfs_change_host *sc)
861 {
862 	gfs2_rgrp_verify(rgd);
863 	sc->sc_total += rgd->rd_data;
864 	sc->sc_free += rgd->rd_free;
865 	sc->sc_dinodes += rgd->rd_dinodes;
866 	return 0;
867 }
868 
869 /**
870  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
871  * @sdp: the filesystem
872  * @sc: the sc info that will be returned
873  *
874  * Any error (other than a signal) will cause this routine to fall back
875  * to the synchronous version.
876  *
877  * FIXME: This really shouldn't busy wait like this.
878  *
879  * Returns: errno
880  */
881 
882 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
883 {
884 	struct gfs2_holder ri_gh;
885 	struct gfs2_rgrpd *rgd_next;
886 	struct gfs2_holder *gha, *gh;
887 	unsigned int slots = 64;
888 	unsigned int x;
889 	int done;
890 	int error = 0, err;
891 
892 	memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
893 	gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
894 	if (!gha)
895 		return -ENOMEM;
896 
897 	error = gfs2_rindex_hold(sdp, &ri_gh);
898 	if (error)
899 		goto out;
900 
901 	rgd_next = gfs2_rgrpd_get_first(sdp);
902 
903 	for (;;) {
904 		done = 1;
905 
906 		for (x = 0; x < slots; x++) {
907 			gh = gha + x;
908 
909 			if (gh->gh_gl && gfs2_glock_poll(gh)) {
910 				err = gfs2_glock_wait(gh);
911 				if (err) {
912 					gfs2_holder_uninit(gh);
913 					error = err;
914 				} else {
915 					if (!error)
916 						error = statfs_slow_fill(
917 							gh->gh_gl->gl_object, sc);
918 					gfs2_glock_dq_uninit(gh);
919 				}
920 			}
921 
922 			if (gh->gh_gl)
923 				done = 0;
924 			else if (rgd_next && !error) {
925 				error = gfs2_glock_nq_init(rgd_next->rd_gl,
926 							   LM_ST_SHARED,
927 							   GL_ASYNC,
928 							   gh);
929 				rgd_next = gfs2_rgrpd_get_next(rgd_next);
930 				done = 0;
931 			}
932 
933 			if (signal_pending(current))
934 				error = -ERESTARTSYS;
935 		}
936 
937 		if (done)
938 			break;
939 
940 		yield();
941 	}
942 
943 	gfs2_glock_dq_uninit(&ri_gh);
944 
945 out:
946 	kfree(gha);
947 	return error;
948 }
949 
950 /**
951  * gfs2_statfs_i - Do a statfs
952  * @sdp: the filesystem
953  * @sg: the sg structure
954  *
955  * Returns: errno
956  */
957 
958 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
959 {
960 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
961 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
962 
963 	spin_lock(&sdp->sd_statfs_spin);
964 
965 	*sc = *m_sc;
966 	sc->sc_total += l_sc->sc_total;
967 	sc->sc_free += l_sc->sc_free;
968 	sc->sc_dinodes += l_sc->sc_dinodes;
969 
970 	spin_unlock(&sdp->sd_statfs_spin);
971 
972 	if (sc->sc_free < 0)
973 		sc->sc_free = 0;
974 	if (sc->sc_free > sc->sc_total)
975 		sc->sc_free = sc->sc_total;
976 	if (sc->sc_dinodes < 0)
977 		sc->sc_dinodes = 0;
978 
979 	return 0;
980 }
981 
982 /**
983  * gfs2_statfs - Gather and return stats about the filesystem
984  * @sb: The superblock
985  * @statfsbuf: The buffer
986  *
987  * Returns: 0 on success or error code
988  */
989 
990 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
991 {
992 	struct super_block *sb = dentry->d_inode->i_sb;
993 	struct gfs2_sbd *sdp = sb->s_fs_info;
994 	struct gfs2_statfs_change_host sc;
995 	int error;
996 
997 	if (gfs2_tune_get(sdp, gt_statfs_slow))
998 		error = gfs2_statfs_slow(sdp, &sc);
999 	else
1000 		error = gfs2_statfs_i(sdp, &sc);
1001 
1002 	if (error)
1003 		return error;
1004 
1005 	buf->f_type = GFS2_MAGIC;
1006 	buf->f_bsize = sdp->sd_sb.sb_bsize;
1007 	buf->f_blocks = sc.sc_total;
1008 	buf->f_bfree = sc.sc_free;
1009 	buf->f_bavail = sc.sc_free;
1010 	buf->f_files = sc.sc_dinodes + sc.sc_free;
1011 	buf->f_ffree = sc.sc_free;
1012 	buf->f_namelen = GFS2_FNAMESIZE;
1013 
1014 	return 0;
1015 }
1016 
1017 /**
1018  * gfs2_remount_fs - called when the FS is remounted
1019  * @sb:  the filesystem
1020  * @flags:  the remount flags
1021  * @data:  extra data passed in (not used right now)
1022  *
1023  * Returns: errno
1024  */
1025 
1026 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1027 {
1028 	struct gfs2_sbd *sdp = sb->s_fs_info;
1029 	struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1030 	struct gfs2_tune *gt = &sdp->sd_tune;
1031 	int error;
1032 
1033 	spin_lock(&gt->gt_spin);
1034 	args.ar_commit = gt->gt_log_flush_secs;
1035 	spin_unlock(&gt->gt_spin);
1036 	error = gfs2_mount_args(sdp, &args, data);
1037 	if (error)
1038 		return error;
1039 
1040 	/* Not allowed to change locking details */
1041 	if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1042 	    strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1043 	    strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1044 		return -EINVAL;
1045 
1046 	/* Some flags must not be changed */
1047 	if (args_neq(&args, &sdp->sd_args, spectator) ||
1048 	    args_neq(&args, &sdp->sd_args, ignore_local_fs) ||
1049 	    args_neq(&args, &sdp->sd_args, localflocks) ||
1050 	    args_neq(&args, &sdp->sd_args, localcaching) ||
1051 	    args_neq(&args, &sdp->sd_args, meta))
1052 		return -EINVAL;
1053 
1054 	if (sdp->sd_args.ar_spectator)
1055 		*flags |= MS_RDONLY;
1056 
1057 	if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1058 		if (*flags & MS_RDONLY)
1059 			error = gfs2_make_fs_ro(sdp);
1060 		else
1061 			error = gfs2_make_fs_rw(sdp);
1062 		if (error)
1063 			return error;
1064 	}
1065 
1066 	sdp->sd_args = args;
1067 	if (sdp->sd_args.ar_posix_acl)
1068 		sb->s_flags |= MS_POSIXACL;
1069 	else
1070 		sb->s_flags &= ~MS_POSIXACL;
1071 	spin_lock(&gt->gt_spin);
1072 	gt->gt_log_flush_secs = args.ar_commit;
1073 	spin_unlock(&gt->gt_spin);
1074 
1075 	return 0;
1076 }
1077 
1078 /**
1079  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1080  * @inode: The inode to drop
1081  *
1082  * If we've received a callback on an iopen lock then its because a
1083  * remote node tried to deallocate the inode but failed due to this node
1084  * still having the inode open. Here we mark the link count zero
1085  * since we know that it must have reached zero if the GLF_DEMOTE flag
1086  * is set on the iopen glock. If we didn't do a disk read since the
1087  * remote node removed the final link then we might otherwise miss
1088  * this event. This check ensures that this node will deallocate the
1089  * inode's blocks, or alternatively pass the baton on to another
1090  * node for later deallocation.
1091  */
1092 
1093 static void gfs2_drop_inode(struct inode *inode)
1094 {
1095 	struct gfs2_inode *ip = GFS2_I(inode);
1096 
1097 	if (test_bit(GIF_USER, &ip->i_flags) && inode->i_nlink) {
1098 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1099 		if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1100 			clear_nlink(inode);
1101 	}
1102 	generic_drop_inode(inode);
1103 }
1104 
1105 /**
1106  * gfs2_clear_inode - Deallocate an inode when VFS is done with it
1107  * @inode: The VFS inode
1108  *
1109  */
1110 
1111 static void gfs2_clear_inode(struct inode *inode)
1112 {
1113 	struct gfs2_inode *ip = GFS2_I(inode);
1114 
1115 	/* This tells us its a "real" inode and not one which only
1116 	 * serves to contain an address space (see rgrp.c, meta_io.c)
1117 	 * which therefore doesn't have its own glocks.
1118 	 */
1119 	if (test_bit(GIF_USER, &ip->i_flags)) {
1120 		ip->i_gl->gl_object = NULL;
1121 		gfs2_glock_put(ip->i_gl);
1122 		ip->i_gl = NULL;
1123 		if (ip->i_iopen_gh.gh_gl) {
1124 			ip->i_iopen_gh.gh_gl->gl_object = NULL;
1125 			gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1126 		}
1127 	}
1128 }
1129 
1130 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1131 {
1132 	do {
1133 		if (d1 == d2)
1134 			return 1;
1135 		d1 = d1->d_parent;
1136 	} while (!IS_ROOT(d1));
1137 	return 0;
1138 }
1139 
1140 /**
1141  * gfs2_show_options - Show mount options for /proc/mounts
1142  * @s: seq_file structure
1143  * @mnt: vfsmount
1144  *
1145  * Returns: 0 on success or error code
1146  */
1147 
1148 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1149 {
1150 	struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
1151 	struct gfs2_args *args = &sdp->sd_args;
1152 	int lfsecs;
1153 
1154 	if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir))
1155 		seq_printf(s, ",meta");
1156 	if (args->ar_lockproto[0])
1157 		seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1158 	if (args->ar_locktable[0])
1159 		seq_printf(s, ",locktable=%s", args->ar_locktable);
1160 	if (args->ar_hostdata[0])
1161 		seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1162 	if (args->ar_spectator)
1163 		seq_printf(s, ",spectator");
1164 	if (args->ar_ignore_local_fs)
1165 		seq_printf(s, ",ignore_local_fs");
1166 	if (args->ar_localflocks)
1167 		seq_printf(s, ",localflocks");
1168 	if (args->ar_localcaching)
1169 		seq_printf(s, ",localcaching");
1170 	if (args->ar_debug)
1171 		seq_printf(s, ",debug");
1172 	if (args->ar_upgrade)
1173 		seq_printf(s, ",upgrade");
1174 	if (args->ar_posix_acl)
1175 		seq_printf(s, ",acl");
1176 	if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1177 		char *state;
1178 		switch (args->ar_quota) {
1179 		case GFS2_QUOTA_OFF:
1180 			state = "off";
1181 			break;
1182 		case GFS2_QUOTA_ACCOUNT:
1183 			state = "account";
1184 			break;
1185 		case GFS2_QUOTA_ON:
1186 			state = "on";
1187 			break;
1188 		default:
1189 			state = "unknown";
1190 			break;
1191 		}
1192 		seq_printf(s, ",quota=%s", state);
1193 	}
1194 	if (args->ar_suiddir)
1195 		seq_printf(s, ",suiddir");
1196 	if (args->ar_data != GFS2_DATA_DEFAULT) {
1197 		char *state;
1198 		switch (args->ar_data) {
1199 		case GFS2_DATA_WRITEBACK:
1200 			state = "writeback";
1201 			break;
1202 		case GFS2_DATA_ORDERED:
1203 			state = "ordered";
1204 			break;
1205 		default:
1206 			state = "unknown";
1207 			break;
1208 		}
1209 		seq_printf(s, ",data=%s", state);
1210 	}
1211 	if (args->ar_discard)
1212 		seq_printf(s, ",discard");
1213 	lfsecs = sdp->sd_tune.gt_log_flush_secs;
1214 	if (lfsecs != 60)
1215 		seq_printf(s, ",commit=%d", lfsecs);
1216 	return 0;
1217 }
1218 
1219 /*
1220  * We have to (at the moment) hold the inodes main lock to cover
1221  * the gap between unlocking the shared lock on the iopen lock and
1222  * taking the exclusive lock. I'd rather do a shared -> exclusive
1223  * conversion on the iopen lock, but we can change that later. This
1224  * is safe, just less efficient.
1225  */
1226 
1227 static void gfs2_delete_inode(struct inode *inode)
1228 {
1229 	struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
1230 	struct gfs2_inode *ip = GFS2_I(inode);
1231 	struct gfs2_holder gh;
1232 	int error;
1233 
1234 	if (!test_bit(GIF_USER, &ip->i_flags))
1235 		goto out;
1236 
1237 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1238 	if (unlikely(error)) {
1239 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1240 		goto out;
1241 	}
1242 
1243 	gfs2_glock_dq_wait(&ip->i_iopen_gh);
1244 	gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1245 	error = gfs2_glock_nq(&ip->i_iopen_gh);
1246 	if (error)
1247 		goto out_truncate;
1248 
1249 	if (S_ISDIR(inode->i_mode) &&
1250 	    (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1251 		error = gfs2_dir_exhash_dealloc(ip);
1252 		if (error)
1253 			goto out_unlock;
1254 	}
1255 
1256 	if (ip->i_eattr) {
1257 		error = gfs2_ea_dealloc(ip);
1258 		if (error)
1259 			goto out_unlock;
1260 	}
1261 
1262 	if (!gfs2_is_stuffed(ip)) {
1263 		error = gfs2_file_dealloc(ip);
1264 		if (error)
1265 			goto out_unlock;
1266 	}
1267 
1268 	error = gfs2_dinode_dealloc(ip);
1269 	if (error)
1270 		goto out_unlock;
1271 
1272 out_truncate:
1273 	error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1274 	if (error)
1275 		goto out_unlock;
1276 	/* Needs to be done before glock release & also in a transaction */
1277 	truncate_inode_pages(&inode->i_data, 0);
1278 	gfs2_trans_end(sdp);
1279 
1280 out_unlock:
1281 	if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
1282 		gfs2_glock_dq(&ip->i_iopen_gh);
1283 	gfs2_holder_uninit(&ip->i_iopen_gh);
1284 	gfs2_glock_dq_uninit(&gh);
1285 	if (error && error != GLR_TRYFAILED && error != -EROFS)
1286 		fs_warn(sdp, "gfs2_delete_inode: %d\n", error);
1287 out:
1288 	truncate_inode_pages(&inode->i_data, 0);
1289 	clear_inode(inode);
1290 }
1291 
1292 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1293 {
1294 	struct gfs2_inode *ip;
1295 
1296 	ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1297 	if (ip) {
1298 		ip->i_flags = 0;
1299 		ip->i_gl = NULL;
1300 	}
1301 	return &ip->i_inode;
1302 }
1303 
1304 static void gfs2_destroy_inode(struct inode *inode)
1305 {
1306 	kmem_cache_free(gfs2_inode_cachep, inode);
1307 }
1308 
1309 const struct super_operations gfs2_super_ops = {
1310 	.alloc_inode		= gfs2_alloc_inode,
1311 	.destroy_inode		= gfs2_destroy_inode,
1312 	.write_inode		= gfs2_write_inode,
1313 	.delete_inode		= gfs2_delete_inode,
1314 	.put_super		= gfs2_put_super,
1315 	.sync_fs		= gfs2_sync_fs,
1316 	.freeze_fs 		= gfs2_freeze,
1317 	.unfreeze_fs		= gfs2_unfreeze,
1318 	.statfs			= gfs2_statfs,
1319 	.remount_fs		= gfs2_remount_fs,
1320 	.clear_inode		= gfs2_clear_inode,
1321 	.drop_inode		= gfs2_drop_inode,
1322 	.show_options		= gfs2_show_options,
1323 };
1324 
1325