xref: /openbmc/linux/fs/gfs2/super.c (revision 79f08d9e)
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 #include <linux/writeback.h>
26 #include <linux/backing-dev.h>
27 
28 #include "gfs2.h"
29 #include "incore.h"
30 #include "bmap.h"
31 #include "dir.h"
32 #include "glock.h"
33 #include "glops.h"
34 #include "inode.h"
35 #include "log.h"
36 #include "meta_io.h"
37 #include "quota.h"
38 #include "recovery.h"
39 #include "rgrp.h"
40 #include "super.h"
41 #include "trans.h"
42 #include "util.h"
43 #include "sys.h"
44 #include "xattr.h"
45 
46 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
47 
48 enum {
49 	Opt_lockproto,
50 	Opt_locktable,
51 	Opt_hostdata,
52 	Opt_spectator,
53 	Opt_ignore_local_fs,
54 	Opt_localflocks,
55 	Opt_localcaching,
56 	Opt_debug,
57 	Opt_nodebug,
58 	Opt_upgrade,
59 	Opt_acl,
60 	Opt_noacl,
61 	Opt_quota_off,
62 	Opt_quota_account,
63 	Opt_quota_on,
64 	Opt_quota,
65 	Opt_noquota,
66 	Opt_suiddir,
67 	Opt_nosuiddir,
68 	Opt_data_writeback,
69 	Opt_data_ordered,
70 	Opt_meta,
71 	Opt_discard,
72 	Opt_nodiscard,
73 	Opt_commit,
74 	Opt_err_withdraw,
75 	Opt_err_panic,
76 	Opt_statfs_quantum,
77 	Opt_statfs_percent,
78 	Opt_quota_quantum,
79 	Opt_barrier,
80 	Opt_nobarrier,
81 	Opt_rgrplvb,
82 	Opt_norgrplvb,
83 	Opt_error,
84 };
85 
86 static const match_table_t tokens = {
87 	{Opt_lockproto, "lockproto=%s"},
88 	{Opt_locktable, "locktable=%s"},
89 	{Opt_hostdata, "hostdata=%s"},
90 	{Opt_spectator, "spectator"},
91 	{Opt_spectator, "norecovery"},
92 	{Opt_ignore_local_fs, "ignore_local_fs"},
93 	{Opt_localflocks, "localflocks"},
94 	{Opt_localcaching, "localcaching"},
95 	{Opt_debug, "debug"},
96 	{Opt_nodebug, "nodebug"},
97 	{Opt_upgrade, "upgrade"},
98 	{Opt_acl, "acl"},
99 	{Opt_noacl, "noacl"},
100 	{Opt_quota_off, "quota=off"},
101 	{Opt_quota_account, "quota=account"},
102 	{Opt_quota_on, "quota=on"},
103 	{Opt_quota, "quota"},
104 	{Opt_noquota, "noquota"},
105 	{Opt_suiddir, "suiddir"},
106 	{Opt_nosuiddir, "nosuiddir"},
107 	{Opt_data_writeback, "data=writeback"},
108 	{Opt_data_ordered, "data=ordered"},
109 	{Opt_meta, "meta"},
110 	{Opt_discard, "discard"},
111 	{Opt_nodiscard, "nodiscard"},
112 	{Opt_commit, "commit=%d"},
113 	{Opt_err_withdraw, "errors=withdraw"},
114 	{Opt_err_panic, "errors=panic"},
115 	{Opt_statfs_quantum, "statfs_quantum=%d"},
116 	{Opt_statfs_percent, "statfs_percent=%d"},
117 	{Opt_quota_quantum, "quota_quantum=%d"},
118 	{Opt_barrier, "barrier"},
119 	{Opt_nobarrier, "nobarrier"},
120 	{Opt_rgrplvb, "rgrplvb"},
121 	{Opt_norgrplvb, "norgrplvb"},
122 	{Opt_error, NULL}
123 };
124 
125 /**
126  * gfs2_mount_args - Parse mount options
127  * @args: The structure into which the parsed options will be written
128  * @options: The options to parse
129  *
130  * Return: errno
131  */
132 
133 int gfs2_mount_args(struct gfs2_args *args, char *options)
134 {
135 	char *o;
136 	int token;
137 	substring_t tmp[MAX_OPT_ARGS];
138 	int rv;
139 
140 	/* Split the options into tokens with the "," character and
141 	   process them */
142 
143 	while (1) {
144 		o = strsep(&options, ",");
145 		if (o == NULL)
146 			break;
147 		if (*o == '\0')
148 			continue;
149 
150 		token = match_token(o, tokens, tmp);
151 		switch (token) {
152 		case Opt_lockproto:
153 			match_strlcpy(args->ar_lockproto, &tmp[0],
154 				      GFS2_LOCKNAME_LEN);
155 			break;
156 		case Opt_locktable:
157 			match_strlcpy(args->ar_locktable, &tmp[0],
158 				      GFS2_LOCKNAME_LEN);
159 			break;
160 		case Opt_hostdata:
161 			match_strlcpy(args->ar_hostdata, &tmp[0],
162 				      GFS2_LOCKNAME_LEN);
163 			break;
164 		case Opt_spectator:
165 			args->ar_spectator = 1;
166 			break;
167 		case Opt_ignore_local_fs:
168 			/* Retained for backwards compat only */
169 			break;
170 		case Opt_localflocks:
171 			args->ar_localflocks = 1;
172 			break;
173 		case Opt_localcaching:
174 			/* Retained for backwards compat only */
175 			break;
176 		case Opt_debug:
177 			if (args->ar_errors == GFS2_ERRORS_PANIC) {
178 				printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
179 				       "are mutually exclusive.\n");
180 				return -EINVAL;
181 			}
182 			args->ar_debug = 1;
183 			break;
184 		case Opt_nodebug:
185 			args->ar_debug = 0;
186 			break;
187 		case Opt_upgrade:
188 			/* Retained for backwards compat only */
189 			break;
190 		case Opt_acl:
191 			args->ar_posix_acl = 1;
192 			break;
193 		case Opt_noacl:
194 			args->ar_posix_acl = 0;
195 			break;
196 		case Opt_quota_off:
197 		case Opt_noquota:
198 			args->ar_quota = GFS2_QUOTA_OFF;
199 			break;
200 		case Opt_quota_account:
201 			args->ar_quota = GFS2_QUOTA_ACCOUNT;
202 			break;
203 		case Opt_quota_on:
204 		case Opt_quota:
205 			args->ar_quota = GFS2_QUOTA_ON;
206 			break;
207 		case Opt_suiddir:
208 			args->ar_suiddir = 1;
209 			break;
210 		case Opt_nosuiddir:
211 			args->ar_suiddir = 0;
212 			break;
213 		case Opt_data_writeback:
214 			args->ar_data = GFS2_DATA_WRITEBACK;
215 			break;
216 		case Opt_data_ordered:
217 			args->ar_data = GFS2_DATA_ORDERED;
218 			break;
219 		case Opt_meta:
220 			args->ar_meta = 1;
221 			break;
222 		case Opt_discard:
223 			args->ar_discard = 1;
224 			break;
225 		case Opt_nodiscard:
226 			args->ar_discard = 0;
227 			break;
228 		case Opt_commit:
229 			rv = match_int(&tmp[0], &args->ar_commit);
230 			if (rv || args->ar_commit <= 0) {
231 				printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n");
232 				return rv ? rv : -EINVAL;
233 			}
234 			break;
235 		case Opt_statfs_quantum:
236 			rv = match_int(&tmp[0], &args->ar_statfs_quantum);
237 			if (rv || args->ar_statfs_quantum < 0) {
238 				printk(KERN_WARNING "GFS2: statfs_quantum mount option requires a non-negative numeric argument\n");
239 				return rv ? rv : -EINVAL;
240 			}
241 			break;
242 		case Opt_quota_quantum:
243 			rv = match_int(&tmp[0], &args->ar_quota_quantum);
244 			if (rv || args->ar_quota_quantum <= 0) {
245 				printk(KERN_WARNING "GFS2: quota_quantum mount option requires a positive numeric argument\n");
246 				return rv ? rv : -EINVAL;
247 			}
248 			break;
249 		case Opt_statfs_percent:
250 			rv = match_int(&tmp[0], &args->ar_statfs_percent);
251 			if (rv || args->ar_statfs_percent < 0 ||
252 			    args->ar_statfs_percent > 100) {
253 				printk(KERN_WARNING "statfs_percent mount option requires a numeric argument between 0 and 100\n");
254 				return rv ? rv : -EINVAL;
255 			}
256 			break;
257 		case Opt_err_withdraw:
258 			args->ar_errors = GFS2_ERRORS_WITHDRAW;
259 			break;
260 		case Opt_err_panic:
261 			if (args->ar_debug) {
262 				printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
263 					"are mutually exclusive.\n");
264 				return -EINVAL;
265 			}
266 			args->ar_errors = GFS2_ERRORS_PANIC;
267 			break;
268 		case Opt_barrier:
269 			args->ar_nobarrier = 0;
270 			break;
271 		case Opt_nobarrier:
272 			args->ar_nobarrier = 1;
273 			break;
274 		case Opt_rgrplvb:
275 			args->ar_rgrplvb = 1;
276 			break;
277 		case Opt_norgrplvb:
278 			args->ar_rgrplvb = 0;
279 			break;
280 		case Opt_error:
281 		default:
282 			printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o);
283 			return -EINVAL;
284 		}
285 	}
286 
287 	return 0;
288 }
289 
290 /**
291  * gfs2_jindex_free - Clear all the journal index information
292  * @sdp: The GFS2 superblock
293  *
294  */
295 
296 void gfs2_jindex_free(struct gfs2_sbd *sdp)
297 {
298 	struct list_head list, *head;
299 	struct gfs2_jdesc *jd;
300 	struct gfs2_journal_extent *jext;
301 
302 	spin_lock(&sdp->sd_jindex_spin);
303 	list_add(&list, &sdp->sd_jindex_list);
304 	list_del_init(&sdp->sd_jindex_list);
305 	sdp->sd_journals = 0;
306 	spin_unlock(&sdp->sd_jindex_spin);
307 
308 	while (!list_empty(&list)) {
309 		jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
310 		head = &jd->extent_list;
311 		while (!list_empty(head)) {
312 			jext = list_entry(head->next,
313 					  struct gfs2_journal_extent,
314 					  extent_list);
315 			list_del(&jext->extent_list);
316 			kfree(jext);
317 		}
318 		list_del(&jd->jd_list);
319 		iput(jd->jd_inode);
320 		kfree(jd);
321 	}
322 }
323 
324 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
325 {
326 	struct gfs2_jdesc *jd;
327 	int found = 0;
328 
329 	list_for_each_entry(jd, head, jd_list) {
330 		if (jd->jd_jid == jid) {
331 			found = 1;
332 			break;
333 		}
334 	}
335 
336 	if (!found)
337 		jd = NULL;
338 
339 	return jd;
340 }
341 
342 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
343 {
344 	struct gfs2_jdesc *jd;
345 
346 	spin_lock(&sdp->sd_jindex_spin);
347 	jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
348 	spin_unlock(&sdp->sd_jindex_spin);
349 
350 	return jd;
351 }
352 
353 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
354 {
355 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
356 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
357 	u64 size = i_size_read(jd->jd_inode);
358 
359 	if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, 1 << 30))
360 		return -EIO;
361 
362 	jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
363 
364 	if (gfs2_write_alloc_required(ip, 0, size)) {
365 		gfs2_consist_inode(ip);
366 		return -EIO;
367 	}
368 
369 	return 0;
370 }
371 
372 /**
373  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
374  * @sdp: the filesystem
375  *
376  * Returns: errno
377  */
378 
379 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
380 {
381 	struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
382 	struct gfs2_glock *j_gl = ip->i_gl;
383 	struct gfs2_holder t_gh;
384 	struct gfs2_log_header_host head;
385 	int error;
386 
387 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
388 	if (error)
389 		return error;
390 
391 	j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
392 
393 	error = gfs2_find_jhead(sdp->sd_jdesc, &head);
394 	if (error)
395 		goto fail;
396 
397 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
398 		gfs2_consist(sdp);
399 		error = -EIO;
400 		goto fail;
401 	}
402 
403 	/*  Initialize some head of the log stuff  */
404 	sdp->sd_log_sequence = head.lh_sequence + 1;
405 	gfs2_log_pointers_init(sdp, head.lh_blkno);
406 
407 	error = gfs2_quota_init(sdp);
408 	if (error)
409 		goto fail;
410 
411 	set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
412 
413 	gfs2_glock_dq_uninit(&t_gh);
414 
415 	return 0;
416 
417 fail:
418 	t_gh.gh_flags |= GL_NOCACHE;
419 	gfs2_glock_dq_uninit(&t_gh);
420 
421 	return error;
422 }
423 
424 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
425 {
426 	const struct gfs2_statfs_change *str = buf;
427 
428 	sc->sc_total = be64_to_cpu(str->sc_total);
429 	sc->sc_free = be64_to_cpu(str->sc_free);
430 	sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
431 }
432 
433 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
434 {
435 	struct gfs2_statfs_change *str = buf;
436 
437 	str->sc_total = cpu_to_be64(sc->sc_total);
438 	str->sc_free = cpu_to_be64(sc->sc_free);
439 	str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
440 }
441 
442 int gfs2_statfs_init(struct gfs2_sbd *sdp)
443 {
444 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
445 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
446 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
447 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
448 	struct buffer_head *m_bh, *l_bh;
449 	struct gfs2_holder gh;
450 	int error;
451 
452 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
453 				   &gh);
454 	if (error)
455 		return error;
456 
457 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
458 	if (error)
459 		goto out;
460 
461 	if (sdp->sd_args.ar_spectator) {
462 		spin_lock(&sdp->sd_statfs_spin);
463 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
464 				      sizeof(struct gfs2_dinode));
465 		spin_unlock(&sdp->sd_statfs_spin);
466 	} else {
467 		error = gfs2_meta_inode_buffer(l_ip, &l_bh);
468 		if (error)
469 			goto out_m_bh;
470 
471 		spin_lock(&sdp->sd_statfs_spin);
472 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
473 				      sizeof(struct gfs2_dinode));
474 		gfs2_statfs_change_in(l_sc, l_bh->b_data +
475 				      sizeof(struct gfs2_dinode));
476 		spin_unlock(&sdp->sd_statfs_spin);
477 
478 		brelse(l_bh);
479 	}
480 
481 out_m_bh:
482 	brelse(m_bh);
483 out:
484 	gfs2_glock_dq_uninit(&gh);
485 	return 0;
486 }
487 
488 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
489 			s64 dinodes)
490 {
491 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
492 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
493 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
494 	struct buffer_head *l_bh;
495 	s64 x, y;
496 	int need_sync = 0;
497 	int error;
498 
499 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
500 	if (error)
501 		return;
502 
503 	gfs2_trans_add_meta(l_ip->i_gl, l_bh);
504 
505 	spin_lock(&sdp->sd_statfs_spin);
506 	l_sc->sc_total += total;
507 	l_sc->sc_free += free;
508 	l_sc->sc_dinodes += dinodes;
509 	gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
510 	if (sdp->sd_args.ar_statfs_percent) {
511 		x = 100 * l_sc->sc_free;
512 		y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
513 		if (x >= y || x <= -y)
514 			need_sync = 1;
515 	}
516 	spin_unlock(&sdp->sd_statfs_spin);
517 
518 	brelse(l_bh);
519 	if (need_sync)
520 		gfs2_wake_up_statfs(sdp);
521 }
522 
523 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
524 		   struct buffer_head *l_bh)
525 {
526 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
527 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
528 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
529 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
530 
531 	gfs2_trans_add_meta(l_ip->i_gl, l_bh);
532 
533 	spin_lock(&sdp->sd_statfs_spin);
534 	m_sc->sc_total += l_sc->sc_total;
535 	m_sc->sc_free += l_sc->sc_free;
536 	m_sc->sc_dinodes += l_sc->sc_dinodes;
537 	memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
538 	memset(l_bh->b_data + sizeof(struct gfs2_dinode),
539 	       0, sizeof(struct gfs2_statfs_change));
540 	spin_unlock(&sdp->sd_statfs_spin);
541 
542 	gfs2_trans_add_meta(m_ip->i_gl, m_bh);
543 	gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
544 }
545 
546 int gfs2_statfs_sync(struct super_block *sb, int type)
547 {
548 	struct gfs2_sbd *sdp = sb->s_fs_info;
549 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
550 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
551 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
552 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
553 	struct gfs2_holder gh;
554 	struct buffer_head *m_bh, *l_bh;
555 	int error;
556 
557 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
558 				   &gh);
559 	if (error)
560 		return error;
561 
562 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
563 	if (error)
564 		goto out;
565 
566 	spin_lock(&sdp->sd_statfs_spin);
567 	gfs2_statfs_change_in(m_sc, m_bh->b_data +
568 			      sizeof(struct gfs2_dinode));
569 	if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
570 		spin_unlock(&sdp->sd_statfs_spin);
571 		goto out_bh;
572 	}
573 	spin_unlock(&sdp->sd_statfs_spin);
574 
575 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
576 	if (error)
577 		goto out_bh;
578 
579 	error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
580 	if (error)
581 		goto out_bh2;
582 
583 	update_statfs(sdp, m_bh, l_bh);
584 	sdp->sd_statfs_force_sync = 0;
585 
586 	gfs2_trans_end(sdp);
587 
588 out_bh2:
589 	brelse(l_bh);
590 out_bh:
591 	brelse(m_bh);
592 out:
593 	gfs2_glock_dq_uninit(&gh);
594 	return error;
595 }
596 
597 struct lfcc {
598 	struct list_head list;
599 	struct gfs2_holder gh;
600 };
601 
602 /**
603  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
604  *                            journals are clean
605  * @sdp: the file system
606  * @state: the state to put the transaction lock into
607  * @t_gh: the hold on the transaction lock
608  *
609  * Returns: errno
610  */
611 
612 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
613 				    struct gfs2_holder *t_gh)
614 {
615 	struct gfs2_inode *ip;
616 	struct gfs2_jdesc *jd;
617 	struct lfcc *lfcc;
618 	LIST_HEAD(list);
619 	struct gfs2_log_header_host lh;
620 	int error;
621 
622 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
623 		lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
624 		if (!lfcc) {
625 			error = -ENOMEM;
626 			goto out;
627 		}
628 		ip = GFS2_I(jd->jd_inode);
629 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
630 		if (error) {
631 			kfree(lfcc);
632 			goto out;
633 		}
634 		list_add(&lfcc->list, &list);
635 	}
636 
637 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
638 				   GL_NOCACHE, t_gh);
639 
640 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
641 		error = gfs2_jdesc_check(jd);
642 		if (error)
643 			break;
644 		error = gfs2_find_jhead(jd, &lh);
645 		if (error)
646 			break;
647 		if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
648 			error = -EBUSY;
649 			break;
650 		}
651 	}
652 
653 	if (error)
654 		gfs2_glock_dq_uninit(t_gh);
655 
656 out:
657 	while (!list_empty(&list)) {
658 		lfcc = list_entry(list.next, struct lfcc, list);
659 		list_del(&lfcc->list);
660 		gfs2_glock_dq_uninit(&lfcc->gh);
661 		kfree(lfcc);
662 	}
663 	return error;
664 }
665 
666 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
667 {
668 	struct gfs2_dinode *str = buf;
669 
670 	str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
671 	str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
672 	str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
673 	str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
674 	str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
675 	str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
676 	str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
677 	str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
678 	str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
679 	str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
680 	str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
681 	str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
682 	str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
683 	str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
684 
685 	str->di_goal_meta = cpu_to_be64(ip->i_goal);
686 	str->di_goal_data = cpu_to_be64(ip->i_goal);
687 	str->di_generation = cpu_to_be64(ip->i_generation);
688 
689 	str->di_flags = cpu_to_be32(ip->i_diskflags);
690 	str->di_height = cpu_to_be16(ip->i_height);
691 	str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
692 					     !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
693 					     GFS2_FORMAT_DE : 0);
694 	str->di_depth = cpu_to_be16(ip->i_depth);
695 	str->di_entries = cpu_to_be32(ip->i_entries);
696 
697 	str->di_eattr = cpu_to_be64(ip->i_eattr);
698 	str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
699 	str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
700 	str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
701 }
702 
703 /**
704  * gfs2_write_inode - Make sure the inode is stable on the disk
705  * @inode: The inode
706  * @wbc: The writeback control structure
707  *
708  * Returns: errno
709  */
710 
711 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
712 {
713 	struct gfs2_inode *ip = GFS2_I(inode);
714 	struct gfs2_sbd *sdp = GFS2_SB(inode);
715 	struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
716 	struct backing_dev_info *bdi = metamapping->backing_dev_info;
717 	int ret = 0;
718 
719 	if (wbc->sync_mode == WB_SYNC_ALL)
720 		gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
721 	if (bdi->dirty_exceeded)
722 		gfs2_ail1_flush(sdp, wbc);
723 	else
724 		filemap_fdatawrite(metamapping);
725 	if (wbc->sync_mode == WB_SYNC_ALL)
726 		ret = filemap_fdatawait(metamapping);
727 	if (ret)
728 		mark_inode_dirty_sync(inode);
729 	return ret;
730 }
731 
732 /**
733  * gfs2_dirty_inode - check for atime updates
734  * @inode: The inode in question
735  * @flags: The type of dirty
736  *
737  * Unfortunately it can be called under any combination of inode
738  * glock and transaction lock, so we have to check carefully.
739  *
740  * At the moment this deals only with atime - it should be possible
741  * to expand that role in future, once a review of the locking has
742  * been carried out.
743  */
744 
745 static void gfs2_dirty_inode(struct inode *inode, int flags)
746 {
747 	struct gfs2_inode *ip = GFS2_I(inode);
748 	struct gfs2_sbd *sdp = GFS2_SB(inode);
749 	struct buffer_head *bh;
750 	struct gfs2_holder gh;
751 	int need_unlock = 0;
752 	int need_endtrans = 0;
753 	int ret;
754 
755 	if (!(flags & (I_DIRTY_DATASYNC|I_DIRTY_SYNC)))
756 		return;
757 
758 	if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
759 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
760 		if (ret) {
761 			fs_err(sdp, "dirty_inode: glock %d\n", ret);
762 			return;
763 		}
764 		need_unlock = 1;
765 	} else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
766 		return;
767 
768 	if (current->journal_info == NULL) {
769 		ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
770 		if (ret) {
771 			fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
772 			goto out;
773 		}
774 		need_endtrans = 1;
775 	}
776 
777 	ret = gfs2_meta_inode_buffer(ip, &bh);
778 	if (ret == 0) {
779 		gfs2_trans_add_meta(ip->i_gl, bh);
780 		gfs2_dinode_out(ip, bh->b_data);
781 		brelse(bh);
782 	}
783 
784 	if (need_endtrans)
785 		gfs2_trans_end(sdp);
786 out:
787 	if (need_unlock)
788 		gfs2_glock_dq_uninit(&gh);
789 }
790 
791 /**
792  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
793  * @sdp: the filesystem
794  *
795  * Returns: errno
796  */
797 
798 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
799 {
800 	struct gfs2_holder t_gh;
801 	int error;
802 
803 	flush_workqueue(gfs2_delete_workqueue);
804 	gfs2_quota_sync(sdp->sd_vfs, 0);
805 	gfs2_statfs_sync(sdp->sd_vfs, 0);
806 
807 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
808 				   &t_gh);
809 	if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
810 		return error;
811 
812 	gfs2_meta_syncfs(sdp);
813 	gfs2_log_shutdown(sdp);
814 
815 	clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
816 
817 	if (t_gh.gh_gl)
818 		gfs2_glock_dq_uninit(&t_gh);
819 
820 	gfs2_quota_cleanup(sdp);
821 
822 	return error;
823 }
824 
825 static int gfs2_umount_recovery_wait(void *word)
826 {
827 	schedule();
828 	return 0;
829 }
830 
831 /**
832  * gfs2_put_super - Unmount the filesystem
833  * @sb: The VFS superblock
834  *
835  */
836 
837 static void gfs2_put_super(struct super_block *sb)
838 {
839 	struct gfs2_sbd *sdp = sb->s_fs_info;
840 	int error;
841 	struct gfs2_jdesc *jd;
842 
843 	/* No more recovery requests */
844 	set_bit(SDF_NORECOVERY, &sdp->sd_flags);
845 	smp_mb();
846 
847 	/* Wait on outstanding recovery */
848 restart:
849 	spin_lock(&sdp->sd_jindex_spin);
850 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
851 		if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
852 			continue;
853 		spin_unlock(&sdp->sd_jindex_spin);
854 		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
855 			    gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
856 		goto restart;
857 	}
858 	spin_unlock(&sdp->sd_jindex_spin);
859 
860 	kthread_stop(sdp->sd_quotad_process);
861 	kthread_stop(sdp->sd_logd_process);
862 
863 	if (!(sb->s_flags & MS_RDONLY)) {
864 		error = gfs2_make_fs_ro(sdp);
865 		if (error)
866 			gfs2_io_error(sdp);
867 	}
868 	/*  At this point, we're through modifying the disk  */
869 
870 	/*  Release stuff  */
871 
872 	iput(sdp->sd_jindex);
873 	iput(sdp->sd_statfs_inode);
874 	iput(sdp->sd_rindex);
875 	iput(sdp->sd_quota_inode);
876 
877 	gfs2_glock_put(sdp->sd_rename_gl);
878 	gfs2_glock_put(sdp->sd_trans_gl);
879 
880 	if (!sdp->sd_args.ar_spectator) {
881 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
882 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
883 		gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
884 		gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
885 		iput(sdp->sd_sc_inode);
886 		iput(sdp->sd_qc_inode);
887 	}
888 
889 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
890 	gfs2_clear_rgrpd(sdp);
891 	gfs2_jindex_free(sdp);
892 	/*  Take apart glock structures and buffer lists  */
893 	gfs2_gl_hash_clear(sdp);
894 	/*  Unmount the locking protocol  */
895 	gfs2_lm_unmount(sdp);
896 
897 	/*  At this point, we're through participating in the lockspace  */
898 	gfs2_sys_fs_del(sdp);
899 }
900 
901 /**
902  * gfs2_sync_fs - sync the filesystem
903  * @sb: the superblock
904  *
905  * Flushes the log to disk.
906  */
907 
908 static int gfs2_sync_fs(struct super_block *sb, int wait)
909 {
910 	struct gfs2_sbd *sdp = sb->s_fs_info;
911 
912 	gfs2_quota_sync(sb, -1);
913 	if (wait && sdp)
914 		gfs2_log_flush(sdp, NULL);
915 	return 0;
916 }
917 
918 /**
919  * gfs2_freeze - prevent further writes to the filesystem
920  * @sb: the VFS structure for the filesystem
921  *
922  */
923 
924 static int gfs2_freeze(struct super_block *sb)
925 {
926 	struct gfs2_sbd *sdp = sb->s_fs_info;
927 	int error;
928 
929 	if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
930 		return -EINVAL;
931 
932 	for (;;) {
933 		error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
934 		if (!error)
935 			break;
936 
937 		switch (error) {
938 		case -EBUSY:
939 			fs_err(sdp, "waiting for recovery before freeze\n");
940 			break;
941 
942 		default:
943 			fs_err(sdp, "error freezing FS: %d\n", error);
944 			break;
945 		}
946 
947 		fs_err(sdp, "retrying...\n");
948 		msleep(1000);
949 	}
950 	return 0;
951 }
952 
953 /**
954  * gfs2_unfreeze - reallow writes to the filesystem
955  * @sb: the VFS structure for the filesystem
956  *
957  */
958 
959 static int gfs2_unfreeze(struct super_block *sb)
960 {
961 	struct gfs2_sbd *sdp = sb->s_fs_info;
962 
963 	gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
964 	return 0;
965 }
966 
967 /**
968  * statfs_fill - fill in the sg for a given RG
969  * @rgd: the RG
970  * @sc: the sc structure
971  *
972  * Returns: 0 on success, -ESTALE if the LVB is invalid
973  */
974 
975 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
976 			    struct gfs2_statfs_change_host *sc)
977 {
978 	gfs2_rgrp_verify(rgd);
979 	sc->sc_total += rgd->rd_data;
980 	sc->sc_free += rgd->rd_free;
981 	sc->sc_dinodes += rgd->rd_dinodes;
982 	return 0;
983 }
984 
985 /**
986  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
987  * @sdp: the filesystem
988  * @sc: the sc info that will be returned
989  *
990  * Any error (other than a signal) will cause this routine to fall back
991  * to the synchronous version.
992  *
993  * FIXME: This really shouldn't busy wait like this.
994  *
995  * Returns: errno
996  */
997 
998 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
999 {
1000 	struct gfs2_rgrpd *rgd_next;
1001 	struct gfs2_holder *gha, *gh;
1002 	unsigned int slots = 64;
1003 	unsigned int x;
1004 	int done;
1005 	int error = 0, err;
1006 
1007 	memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1008 	gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1009 	if (!gha)
1010 		return -ENOMEM;
1011 
1012 	rgd_next = gfs2_rgrpd_get_first(sdp);
1013 
1014 	for (;;) {
1015 		done = 1;
1016 
1017 		for (x = 0; x < slots; x++) {
1018 			gh = gha + x;
1019 
1020 			if (gh->gh_gl && gfs2_glock_poll(gh)) {
1021 				err = gfs2_glock_wait(gh);
1022 				if (err) {
1023 					gfs2_holder_uninit(gh);
1024 					error = err;
1025 				} else {
1026 					if (!error)
1027 						error = statfs_slow_fill(
1028 							gh->gh_gl->gl_object, sc);
1029 					gfs2_glock_dq_uninit(gh);
1030 				}
1031 			}
1032 
1033 			if (gh->gh_gl)
1034 				done = 0;
1035 			else if (rgd_next && !error) {
1036 				error = gfs2_glock_nq_init(rgd_next->rd_gl,
1037 							   LM_ST_SHARED,
1038 							   GL_ASYNC,
1039 							   gh);
1040 				rgd_next = gfs2_rgrpd_get_next(rgd_next);
1041 				done = 0;
1042 			}
1043 
1044 			if (signal_pending(current))
1045 				error = -ERESTARTSYS;
1046 		}
1047 
1048 		if (done)
1049 			break;
1050 
1051 		yield();
1052 	}
1053 
1054 	kfree(gha);
1055 	return error;
1056 }
1057 
1058 /**
1059  * gfs2_statfs_i - Do a statfs
1060  * @sdp: the filesystem
1061  * @sg: the sg structure
1062  *
1063  * Returns: errno
1064  */
1065 
1066 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1067 {
1068 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1069 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1070 
1071 	spin_lock(&sdp->sd_statfs_spin);
1072 
1073 	*sc = *m_sc;
1074 	sc->sc_total += l_sc->sc_total;
1075 	sc->sc_free += l_sc->sc_free;
1076 	sc->sc_dinodes += l_sc->sc_dinodes;
1077 
1078 	spin_unlock(&sdp->sd_statfs_spin);
1079 
1080 	if (sc->sc_free < 0)
1081 		sc->sc_free = 0;
1082 	if (sc->sc_free > sc->sc_total)
1083 		sc->sc_free = sc->sc_total;
1084 	if (sc->sc_dinodes < 0)
1085 		sc->sc_dinodes = 0;
1086 
1087 	return 0;
1088 }
1089 
1090 /**
1091  * gfs2_statfs - Gather and return stats about the filesystem
1092  * @sb: The superblock
1093  * @statfsbuf: The buffer
1094  *
1095  * Returns: 0 on success or error code
1096  */
1097 
1098 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1099 {
1100 	struct super_block *sb = dentry->d_inode->i_sb;
1101 	struct gfs2_sbd *sdp = sb->s_fs_info;
1102 	struct gfs2_statfs_change_host sc;
1103 	int error;
1104 
1105 	error = gfs2_rindex_update(sdp);
1106 	if (error)
1107 		return error;
1108 
1109 	if (gfs2_tune_get(sdp, gt_statfs_slow))
1110 		error = gfs2_statfs_slow(sdp, &sc);
1111 	else
1112 		error = gfs2_statfs_i(sdp, &sc);
1113 
1114 	if (error)
1115 		return error;
1116 
1117 	buf->f_type = GFS2_MAGIC;
1118 	buf->f_bsize = sdp->sd_sb.sb_bsize;
1119 	buf->f_blocks = sc.sc_total;
1120 	buf->f_bfree = sc.sc_free;
1121 	buf->f_bavail = sc.sc_free;
1122 	buf->f_files = sc.sc_dinodes + sc.sc_free;
1123 	buf->f_ffree = sc.sc_free;
1124 	buf->f_namelen = GFS2_FNAMESIZE;
1125 
1126 	return 0;
1127 }
1128 
1129 /**
1130  * gfs2_remount_fs - called when the FS is remounted
1131  * @sb:  the filesystem
1132  * @flags:  the remount flags
1133  * @data:  extra data passed in (not used right now)
1134  *
1135  * Returns: errno
1136  */
1137 
1138 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1139 {
1140 	struct gfs2_sbd *sdp = sb->s_fs_info;
1141 	struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1142 	struct gfs2_tune *gt = &sdp->sd_tune;
1143 	int error;
1144 
1145 	spin_lock(&gt->gt_spin);
1146 	args.ar_commit = gt->gt_logd_secs;
1147 	args.ar_quota_quantum = gt->gt_quota_quantum;
1148 	if (gt->gt_statfs_slow)
1149 		args.ar_statfs_quantum = 0;
1150 	else
1151 		args.ar_statfs_quantum = gt->gt_statfs_quantum;
1152 	spin_unlock(&gt->gt_spin);
1153 	error = gfs2_mount_args(&args, data);
1154 	if (error)
1155 		return error;
1156 
1157 	/* Not allowed to change locking details */
1158 	if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1159 	    strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1160 	    strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1161 		return -EINVAL;
1162 
1163 	/* Some flags must not be changed */
1164 	if (args_neq(&args, &sdp->sd_args, spectator) ||
1165 	    args_neq(&args, &sdp->sd_args, localflocks) ||
1166 	    args_neq(&args, &sdp->sd_args, meta))
1167 		return -EINVAL;
1168 
1169 	if (sdp->sd_args.ar_spectator)
1170 		*flags |= MS_RDONLY;
1171 
1172 	if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1173 		if (*flags & MS_RDONLY)
1174 			error = gfs2_make_fs_ro(sdp);
1175 		else
1176 			error = gfs2_make_fs_rw(sdp);
1177 		if (error)
1178 			return error;
1179 	}
1180 
1181 	sdp->sd_args = args;
1182 	if (sdp->sd_args.ar_posix_acl)
1183 		sb->s_flags |= MS_POSIXACL;
1184 	else
1185 		sb->s_flags &= ~MS_POSIXACL;
1186 	if (sdp->sd_args.ar_nobarrier)
1187 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1188 	else
1189 		clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1190 	spin_lock(&gt->gt_spin);
1191 	gt->gt_logd_secs = args.ar_commit;
1192 	gt->gt_quota_quantum = args.ar_quota_quantum;
1193 	if (args.ar_statfs_quantum) {
1194 		gt->gt_statfs_slow = 0;
1195 		gt->gt_statfs_quantum = args.ar_statfs_quantum;
1196 	}
1197 	else {
1198 		gt->gt_statfs_slow = 1;
1199 		gt->gt_statfs_quantum = 30;
1200 	}
1201 	spin_unlock(&gt->gt_spin);
1202 
1203 	gfs2_online_uevent(sdp);
1204 	return 0;
1205 }
1206 
1207 /**
1208  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1209  * @inode: The inode to drop
1210  *
1211  * If we've received a callback on an iopen lock then its because a
1212  * remote node tried to deallocate the inode but failed due to this node
1213  * still having the inode open. Here we mark the link count zero
1214  * since we know that it must have reached zero if the GLF_DEMOTE flag
1215  * is set on the iopen glock. If we didn't do a disk read since the
1216  * remote node removed the final link then we might otherwise miss
1217  * this event. This check ensures that this node will deallocate the
1218  * inode's blocks, or alternatively pass the baton on to another
1219  * node for later deallocation.
1220  */
1221 
1222 static int gfs2_drop_inode(struct inode *inode)
1223 {
1224 	struct gfs2_inode *ip = GFS2_I(inode);
1225 
1226 	if (inode->i_nlink) {
1227 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1228 		if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1229 			clear_nlink(inode);
1230 	}
1231 	return generic_drop_inode(inode);
1232 }
1233 
1234 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1235 {
1236 	do {
1237 		if (d1 == d2)
1238 			return 1;
1239 		d1 = d1->d_parent;
1240 	} while (!IS_ROOT(d1));
1241 	return 0;
1242 }
1243 
1244 /**
1245  * gfs2_show_options - Show mount options for /proc/mounts
1246  * @s: seq_file structure
1247  * @root: root of this (sub)tree
1248  *
1249  * Returns: 0 on success or error code
1250  */
1251 
1252 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1253 {
1254 	struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1255 	struct gfs2_args *args = &sdp->sd_args;
1256 	int val;
1257 
1258 	if (is_ancestor(root, sdp->sd_master_dir))
1259 		seq_printf(s, ",meta");
1260 	if (args->ar_lockproto[0])
1261 		seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1262 	if (args->ar_locktable[0])
1263 		seq_printf(s, ",locktable=%s", args->ar_locktable);
1264 	if (args->ar_hostdata[0])
1265 		seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1266 	if (args->ar_spectator)
1267 		seq_printf(s, ",spectator");
1268 	if (args->ar_localflocks)
1269 		seq_printf(s, ",localflocks");
1270 	if (args->ar_debug)
1271 		seq_printf(s, ",debug");
1272 	if (args->ar_posix_acl)
1273 		seq_printf(s, ",acl");
1274 	if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1275 		char *state;
1276 		switch (args->ar_quota) {
1277 		case GFS2_QUOTA_OFF:
1278 			state = "off";
1279 			break;
1280 		case GFS2_QUOTA_ACCOUNT:
1281 			state = "account";
1282 			break;
1283 		case GFS2_QUOTA_ON:
1284 			state = "on";
1285 			break;
1286 		default:
1287 			state = "unknown";
1288 			break;
1289 		}
1290 		seq_printf(s, ",quota=%s", state);
1291 	}
1292 	if (args->ar_suiddir)
1293 		seq_printf(s, ",suiddir");
1294 	if (args->ar_data != GFS2_DATA_DEFAULT) {
1295 		char *state;
1296 		switch (args->ar_data) {
1297 		case GFS2_DATA_WRITEBACK:
1298 			state = "writeback";
1299 			break;
1300 		case GFS2_DATA_ORDERED:
1301 			state = "ordered";
1302 			break;
1303 		default:
1304 			state = "unknown";
1305 			break;
1306 		}
1307 		seq_printf(s, ",data=%s", state);
1308 	}
1309 	if (args->ar_discard)
1310 		seq_printf(s, ",discard");
1311 	val = sdp->sd_tune.gt_logd_secs;
1312 	if (val != 30)
1313 		seq_printf(s, ",commit=%d", val);
1314 	val = sdp->sd_tune.gt_statfs_quantum;
1315 	if (val != 30)
1316 		seq_printf(s, ",statfs_quantum=%d", val);
1317 	else if (sdp->sd_tune.gt_statfs_slow)
1318 		seq_puts(s, ",statfs_quantum=0");
1319 	val = sdp->sd_tune.gt_quota_quantum;
1320 	if (val != 60)
1321 		seq_printf(s, ",quota_quantum=%d", val);
1322 	if (args->ar_statfs_percent)
1323 		seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1324 	if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1325 		const char *state;
1326 
1327 		switch (args->ar_errors) {
1328 		case GFS2_ERRORS_WITHDRAW:
1329 			state = "withdraw";
1330 			break;
1331 		case GFS2_ERRORS_PANIC:
1332 			state = "panic";
1333 			break;
1334 		default:
1335 			state = "unknown";
1336 			break;
1337 		}
1338 		seq_printf(s, ",errors=%s", state);
1339 	}
1340 	if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1341 		seq_printf(s, ",nobarrier");
1342 	if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1343 		seq_printf(s, ",demote_interface_used");
1344 	if (args->ar_rgrplvb)
1345 		seq_printf(s, ",rgrplvb");
1346 	return 0;
1347 }
1348 
1349 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1350 {
1351 	struct inode *inode = &ip->i_inode;
1352 	struct gfs2_glock *gl = ip->i_gl;
1353 
1354 	truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1355 	truncate_inode_pages(&inode->i_data, 0);
1356 
1357 	if (atomic_read(&gl->gl_revokes) == 0) {
1358 		clear_bit(GLF_LFLUSH, &gl->gl_flags);
1359 		clear_bit(GLF_DIRTY, &gl->gl_flags);
1360 	}
1361 }
1362 
1363 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1364 {
1365 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1366 	struct gfs2_rgrpd *rgd;
1367 	struct gfs2_holder gh;
1368 	int error;
1369 
1370 	if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1371 		gfs2_consist_inode(ip);
1372 		return -EIO;
1373 	}
1374 
1375 	error = gfs2_rindex_update(sdp);
1376 	if (error)
1377 		return error;
1378 
1379 	error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1380 	if (error)
1381 		return error;
1382 
1383 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1384 	if (!rgd) {
1385 		gfs2_consist_inode(ip);
1386 		error = -EIO;
1387 		goto out_qs;
1388 	}
1389 
1390 	error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1391 	if (error)
1392 		goto out_qs;
1393 
1394 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1395 				 sdp->sd_jdesc->jd_blocks);
1396 	if (error)
1397 		goto out_rg_gunlock;
1398 
1399 	gfs2_free_di(rgd, ip);
1400 
1401 	gfs2_final_release_pages(ip);
1402 
1403 	gfs2_trans_end(sdp);
1404 
1405 out_rg_gunlock:
1406 	gfs2_glock_dq_uninit(&gh);
1407 out_qs:
1408 	gfs2_quota_unhold(ip);
1409 	return error;
1410 }
1411 
1412 /**
1413  * gfs2_evict_inode - Remove an inode from cache
1414  * @inode: The inode to evict
1415  *
1416  * There are three cases to consider:
1417  * 1. i_nlink == 0, we are final opener (and must deallocate)
1418  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1419  * 3. i_nlink > 0
1420  *
1421  * If the fs is read only, then we have to treat all cases as per #3
1422  * since we are unable to do any deallocation. The inode will be
1423  * deallocated by the next read/write node to attempt an allocation
1424  * in the same resource group
1425  *
1426  * We have to (at the moment) hold the inodes main lock to cover
1427  * the gap between unlocking the shared lock on the iopen lock and
1428  * taking the exclusive lock. I'd rather do a shared -> exclusive
1429  * conversion on the iopen lock, but we can change that later. This
1430  * is safe, just less efficient.
1431  */
1432 
1433 static void gfs2_evict_inode(struct inode *inode)
1434 {
1435 	struct super_block *sb = inode->i_sb;
1436 	struct gfs2_sbd *sdp = sb->s_fs_info;
1437 	struct gfs2_inode *ip = GFS2_I(inode);
1438 	struct gfs2_holder gh;
1439 	int error;
1440 
1441 	if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1442 		goto out;
1443 
1444 	/* Must not read inode block until block type has been verified */
1445 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1446 	if (unlikely(error)) {
1447 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1448 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1449 		goto out;
1450 	}
1451 
1452 	if (!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1453 		error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1454 		if (error)
1455 			goto out_truncate;
1456 	}
1457 
1458 	if (test_bit(GIF_INVALID, &ip->i_flags)) {
1459 		error = gfs2_inode_refresh(ip);
1460 		if (error)
1461 			goto out_truncate;
1462 	}
1463 
1464 	ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1465 	gfs2_glock_dq_wait(&ip->i_iopen_gh);
1466 	gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1467 	error = gfs2_glock_nq(&ip->i_iopen_gh);
1468 	if (error)
1469 		goto out_truncate;
1470 
1471 	/* Case 1 starts here */
1472 
1473 	if (S_ISDIR(inode->i_mode) &&
1474 	    (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1475 		error = gfs2_dir_exhash_dealloc(ip);
1476 		if (error)
1477 			goto out_unlock;
1478 	}
1479 
1480 	if (ip->i_eattr) {
1481 		error = gfs2_ea_dealloc(ip);
1482 		if (error)
1483 			goto out_unlock;
1484 	}
1485 
1486 	if (!gfs2_is_stuffed(ip)) {
1487 		error = gfs2_file_dealloc(ip);
1488 		if (error)
1489 			goto out_unlock;
1490 	}
1491 
1492 	error = gfs2_dinode_dealloc(ip);
1493 	goto out_unlock;
1494 
1495 out_truncate:
1496 	gfs2_log_flush(sdp, ip->i_gl);
1497 	if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1498 		struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
1499 		filemap_fdatawrite(metamapping);
1500 		filemap_fdatawait(metamapping);
1501 	}
1502 	write_inode_now(inode, 1);
1503 	gfs2_ail_flush(ip->i_gl, 0);
1504 
1505 	/* Case 2 starts here */
1506 	error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1507 	if (error)
1508 		goto out_unlock;
1509 	/* Needs to be done before glock release & also in a transaction */
1510 	truncate_inode_pages(&inode->i_data, 0);
1511 	gfs2_trans_end(sdp);
1512 
1513 out_unlock:
1514 	/* Error path for case 1 */
1515 	if (gfs2_rs_active(ip->i_res))
1516 		gfs2_rs_deltree(ip->i_res);
1517 
1518 	if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1519 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1520 		gfs2_glock_dq(&ip->i_iopen_gh);
1521 	}
1522 	gfs2_holder_uninit(&ip->i_iopen_gh);
1523 	gfs2_glock_dq_uninit(&gh);
1524 	if (error && error != GLR_TRYFAILED && error != -EROFS)
1525 		fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1526 out:
1527 	/* Case 3 starts here */
1528 	truncate_inode_pages(&inode->i_data, 0);
1529 	gfs2_rs_delete(ip, NULL);
1530 	gfs2_ordered_del_inode(ip);
1531 	clear_inode(inode);
1532 	gfs2_dir_hash_inval(ip);
1533 	ip->i_gl->gl_object = NULL;
1534 	flush_delayed_work(&ip->i_gl->gl_work);
1535 	gfs2_glock_add_to_lru(ip->i_gl);
1536 	gfs2_glock_put(ip->i_gl);
1537 	ip->i_gl = NULL;
1538 	if (ip->i_iopen_gh.gh_gl) {
1539 		ip->i_iopen_gh.gh_gl->gl_object = NULL;
1540 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1541 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1542 	}
1543 }
1544 
1545 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1546 {
1547 	struct gfs2_inode *ip;
1548 
1549 	ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1550 	if (ip) {
1551 		ip->i_flags = 0;
1552 		ip->i_gl = NULL;
1553 		ip->i_rgd = NULL;
1554 		ip->i_res = NULL;
1555 	}
1556 	return &ip->i_inode;
1557 }
1558 
1559 static void gfs2_i_callback(struct rcu_head *head)
1560 {
1561 	struct inode *inode = container_of(head, struct inode, i_rcu);
1562 	kmem_cache_free(gfs2_inode_cachep, inode);
1563 }
1564 
1565 static void gfs2_destroy_inode(struct inode *inode)
1566 {
1567 	call_rcu(&inode->i_rcu, gfs2_i_callback);
1568 }
1569 
1570 const struct super_operations gfs2_super_ops = {
1571 	.alloc_inode		= gfs2_alloc_inode,
1572 	.destroy_inode		= gfs2_destroy_inode,
1573 	.write_inode		= gfs2_write_inode,
1574 	.dirty_inode		= gfs2_dirty_inode,
1575 	.evict_inode		= gfs2_evict_inode,
1576 	.put_super		= gfs2_put_super,
1577 	.sync_fs		= gfs2_sync_fs,
1578 	.freeze_fs 		= gfs2_freeze,
1579 	.unfreeze_fs		= gfs2_unfreeze,
1580 	.statfs			= gfs2_statfs,
1581 	.remount_fs		= gfs2_remount_fs,
1582 	.drop_inode		= gfs2_drop_inode,
1583 	.show_options		= gfs2_show_options,
1584 };
1585 
1586