1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9 
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 
12 #include <linux/bio.h>
13 #include <linux/sched/signal.h>
14 #include <linux/slab.h>
15 #include <linux/spinlock.h>
16 #include <linux/completion.h>
17 #include <linux/buffer_head.h>
18 #include <linux/statfs.h>
19 #include <linux/seq_file.h>
20 #include <linux/mount.h>
21 #include <linux/kthread.h>
22 #include <linux/delay.h>
23 #include <linux/gfs2_ondisk.h>
24 #include <linux/crc32.h>
25 #include <linux/time.h>
26 #include <linux/wait.h>
27 #include <linux/writeback.h>
28 #include <linux/backing-dev.h>
29 #include <linux/kernel.h>
30 
31 #include "gfs2.h"
32 #include "incore.h"
33 #include "bmap.h"
34 #include "dir.h"
35 #include "glock.h"
36 #include "glops.h"
37 #include "inode.h"
38 #include "log.h"
39 #include "meta_io.h"
40 #include "quota.h"
41 #include "recovery.h"
42 #include "rgrp.h"
43 #include "super.h"
44 #include "trans.h"
45 #include "util.h"
46 #include "sys.h"
47 #include "xattr.h"
48 
49 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
50 
51 enum {
52 	Opt_lockproto,
53 	Opt_locktable,
54 	Opt_hostdata,
55 	Opt_spectator,
56 	Opt_ignore_local_fs,
57 	Opt_localflocks,
58 	Opt_localcaching,
59 	Opt_debug,
60 	Opt_nodebug,
61 	Opt_upgrade,
62 	Opt_acl,
63 	Opt_noacl,
64 	Opt_quota_off,
65 	Opt_quota_account,
66 	Opt_quota_on,
67 	Opt_quota,
68 	Opt_noquota,
69 	Opt_suiddir,
70 	Opt_nosuiddir,
71 	Opt_data_writeback,
72 	Opt_data_ordered,
73 	Opt_meta,
74 	Opt_discard,
75 	Opt_nodiscard,
76 	Opt_commit,
77 	Opt_err_withdraw,
78 	Opt_err_panic,
79 	Opt_statfs_quantum,
80 	Opt_statfs_percent,
81 	Opt_quota_quantum,
82 	Opt_barrier,
83 	Opt_nobarrier,
84 	Opt_rgrplvb,
85 	Opt_norgrplvb,
86 	Opt_loccookie,
87 	Opt_noloccookie,
88 	Opt_error,
89 };
90 
91 static const match_table_t tokens = {
92 	{Opt_lockproto, "lockproto=%s"},
93 	{Opt_locktable, "locktable=%s"},
94 	{Opt_hostdata, "hostdata=%s"},
95 	{Opt_spectator, "spectator"},
96 	{Opt_spectator, "norecovery"},
97 	{Opt_ignore_local_fs, "ignore_local_fs"},
98 	{Opt_localflocks, "localflocks"},
99 	{Opt_localcaching, "localcaching"},
100 	{Opt_debug, "debug"},
101 	{Opt_nodebug, "nodebug"},
102 	{Opt_upgrade, "upgrade"},
103 	{Opt_acl, "acl"},
104 	{Opt_noacl, "noacl"},
105 	{Opt_quota_off, "quota=off"},
106 	{Opt_quota_account, "quota=account"},
107 	{Opt_quota_on, "quota=on"},
108 	{Opt_quota, "quota"},
109 	{Opt_noquota, "noquota"},
110 	{Opt_suiddir, "suiddir"},
111 	{Opt_nosuiddir, "nosuiddir"},
112 	{Opt_data_writeback, "data=writeback"},
113 	{Opt_data_ordered, "data=ordered"},
114 	{Opt_meta, "meta"},
115 	{Opt_discard, "discard"},
116 	{Opt_nodiscard, "nodiscard"},
117 	{Opt_commit, "commit=%d"},
118 	{Opt_err_withdraw, "errors=withdraw"},
119 	{Opt_err_panic, "errors=panic"},
120 	{Opt_statfs_quantum, "statfs_quantum=%d"},
121 	{Opt_statfs_percent, "statfs_percent=%d"},
122 	{Opt_quota_quantum, "quota_quantum=%d"},
123 	{Opt_barrier, "barrier"},
124 	{Opt_nobarrier, "nobarrier"},
125 	{Opt_rgrplvb, "rgrplvb"},
126 	{Opt_norgrplvb, "norgrplvb"},
127 	{Opt_loccookie, "loccookie"},
128 	{Opt_noloccookie, "noloccookie"},
129 	{Opt_error, NULL}
130 };
131 
132 /**
133  * gfs2_mount_args - Parse mount options
134  * @args: The structure into which the parsed options will be written
135  * @options: The options to parse
136  *
137  * Return: errno
138  */
139 
gfs2_mount_args(struct gfs2_args * args,char * options)140 int gfs2_mount_args(struct gfs2_args *args, char *options)
141 {
142 	char *o;
143 	int token;
144 	substring_t tmp[MAX_OPT_ARGS];
145 	int rv;
146 
147 	/* Split the options into tokens with the "," character and
148 	   process them */
149 
150 	while (1) {
151 		o = strsep(&options, ",");
152 		if (o == NULL)
153 			break;
154 		if (*o == '\0')
155 			continue;
156 
157 		token = match_token(o, tokens, tmp);
158 		switch (token) {
159 		case Opt_lockproto:
160 			match_strlcpy(args->ar_lockproto, &tmp[0],
161 				      GFS2_LOCKNAME_LEN);
162 			break;
163 		case Opt_locktable:
164 			match_strlcpy(args->ar_locktable, &tmp[0],
165 				      GFS2_LOCKNAME_LEN);
166 			break;
167 		case Opt_hostdata:
168 			match_strlcpy(args->ar_hostdata, &tmp[0],
169 				      GFS2_LOCKNAME_LEN);
170 			break;
171 		case Opt_spectator:
172 			args->ar_spectator = 1;
173 			break;
174 		case Opt_ignore_local_fs:
175 			/* Retained for backwards compat only */
176 			break;
177 		case Opt_localflocks:
178 			args->ar_localflocks = 1;
179 			break;
180 		case Opt_localcaching:
181 			/* Retained for backwards compat only */
182 			break;
183 		case Opt_debug:
184 			if (args->ar_errors == GFS2_ERRORS_PANIC) {
185 				pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
186 				return -EINVAL;
187 			}
188 			args->ar_debug = 1;
189 			break;
190 		case Opt_nodebug:
191 			args->ar_debug = 0;
192 			break;
193 		case Opt_upgrade:
194 			/* Retained for backwards compat only */
195 			break;
196 		case Opt_acl:
197 			args->ar_posix_acl = 1;
198 			break;
199 		case Opt_noacl:
200 			args->ar_posix_acl = 0;
201 			break;
202 		case Opt_quota_off:
203 		case Opt_noquota:
204 			args->ar_quota = GFS2_QUOTA_OFF;
205 			break;
206 		case Opt_quota_account:
207 			args->ar_quota = GFS2_QUOTA_ACCOUNT;
208 			break;
209 		case Opt_quota_on:
210 		case Opt_quota:
211 			args->ar_quota = GFS2_QUOTA_ON;
212 			break;
213 		case Opt_suiddir:
214 			args->ar_suiddir = 1;
215 			break;
216 		case Opt_nosuiddir:
217 			args->ar_suiddir = 0;
218 			break;
219 		case Opt_data_writeback:
220 			args->ar_data = GFS2_DATA_WRITEBACK;
221 			break;
222 		case Opt_data_ordered:
223 			args->ar_data = GFS2_DATA_ORDERED;
224 			break;
225 		case Opt_meta:
226 			args->ar_meta = 1;
227 			break;
228 		case Opt_discard:
229 			args->ar_discard = 1;
230 			break;
231 		case Opt_nodiscard:
232 			args->ar_discard = 0;
233 			break;
234 		case Opt_commit:
235 			rv = match_int(&tmp[0], &args->ar_commit);
236 			if (rv || args->ar_commit <= 0) {
237 				pr_warn("commit mount option requires a positive numeric argument\n");
238 				return rv ? rv : -EINVAL;
239 			}
240 			break;
241 		case Opt_statfs_quantum:
242 			rv = match_int(&tmp[0], &args->ar_statfs_quantum);
243 			if (rv || args->ar_statfs_quantum < 0) {
244 				pr_warn("statfs_quantum mount option requires a non-negative numeric argument\n");
245 				return rv ? rv : -EINVAL;
246 			}
247 			break;
248 		case Opt_quota_quantum:
249 			rv = match_int(&tmp[0], &args->ar_quota_quantum);
250 			if (rv || args->ar_quota_quantum <= 0) {
251 				pr_warn("quota_quantum mount option requires a positive numeric argument\n");
252 				return rv ? rv : -EINVAL;
253 			}
254 			break;
255 		case Opt_statfs_percent:
256 			rv = match_int(&tmp[0], &args->ar_statfs_percent);
257 			if (rv || args->ar_statfs_percent < 0 ||
258 			    args->ar_statfs_percent > 100) {
259 				pr_warn("statfs_percent mount option requires a numeric argument between 0 and 100\n");
260 				return rv ? rv : -EINVAL;
261 			}
262 			break;
263 		case Opt_err_withdraw:
264 			args->ar_errors = GFS2_ERRORS_WITHDRAW;
265 			break;
266 		case Opt_err_panic:
267 			if (args->ar_debug) {
268 				pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
269 				return -EINVAL;
270 			}
271 			args->ar_errors = GFS2_ERRORS_PANIC;
272 			break;
273 		case Opt_barrier:
274 			args->ar_nobarrier = 0;
275 			break;
276 		case Opt_nobarrier:
277 			args->ar_nobarrier = 1;
278 			break;
279 		case Opt_rgrplvb:
280 			args->ar_rgrplvb = 1;
281 			break;
282 		case Opt_norgrplvb:
283 			args->ar_rgrplvb = 0;
284 			break;
285 		case Opt_loccookie:
286 			args->ar_loccookie = 1;
287 			break;
288 		case Opt_noloccookie:
289 			args->ar_loccookie = 0;
290 			break;
291 		case Opt_error:
292 		default:
293 			pr_warn("invalid mount option: %s\n", o);
294 			return -EINVAL;
295 		}
296 	}
297 
298 	return 0;
299 }
300 
301 /**
302  * gfs2_jindex_free - Clear all the journal index information
303  * @sdp: The GFS2 superblock
304  *
305  */
306 
gfs2_jindex_free(struct gfs2_sbd * sdp)307 void gfs2_jindex_free(struct gfs2_sbd *sdp)
308 {
309 	struct list_head list;
310 	struct gfs2_jdesc *jd;
311 
312 	spin_lock(&sdp->sd_jindex_spin);
313 	list_add(&list, &sdp->sd_jindex_list);
314 	list_del_init(&sdp->sd_jindex_list);
315 	sdp->sd_journals = 0;
316 	spin_unlock(&sdp->sd_jindex_spin);
317 
318 	while (!list_empty(&list)) {
319 		jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
320 		gfs2_free_journal_extents(jd);
321 		list_del(&jd->jd_list);
322 		iput(jd->jd_inode);
323 		kfree(jd);
324 	}
325 }
326 
jdesc_find_i(struct list_head * head,unsigned int jid)327 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
328 {
329 	struct gfs2_jdesc *jd;
330 	int found = 0;
331 
332 	list_for_each_entry(jd, head, jd_list) {
333 		if (jd->jd_jid == jid) {
334 			found = 1;
335 			break;
336 		}
337 	}
338 
339 	if (!found)
340 		jd = NULL;
341 
342 	return jd;
343 }
344 
gfs2_jdesc_find(struct gfs2_sbd * sdp,unsigned int jid)345 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
346 {
347 	struct gfs2_jdesc *jd;
348 
349 	spin_lock(&sdp->sd_jindex_spin);
350 	jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
351 	spin_unlock(&sdp->sd_jindex_spin);
352 
353 	return jd;
354 }
355 
gfs2_jdesc_check(struct gfs2_jdesc * jd)356 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
357 {
358 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
359 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
360 	u64 size = i_size_read(jd->jd_inode);
361 
362 	if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
363 		return -EIO;
364 
365 	jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
366 
367 	if (gfs2_write_alloc_required(ip, 0, size)) {
368 		gfs2_consist_inode(ip);
369 		return -EIO;
370 	}
371 
372 	return 0;
373 }
374 
init_threads(struct gfs2_sbd * sdp)375 static int init_threads(struct gfs2_sbd *sdp)
376 {
377 	struct task_struct *p;
378 	int error = 0;
379 
380 	p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
381 	if (IS_ERR(p)) {
382 		error = PTR_ERR(p);
383 		fs_err(sdp, "can't start logd thread: %d\n", error);
384 		return error;
385 	}
386 	sdp->sd_logd_process = p;
387 
388 	p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
389 	if (IS_ERR(p)) {
390 		error = PTR_ERR(p);
391 		fs_err(sdp, "can't start quotad thread: %d\n", error);
392 		goto fail;
393 	}
394 	sdp->sd_quotad_process = p;
395 	return 0;
396 
397 fail:
398 	kthread_stop(sdp->sd_logd_process);
399 	return error;
400 }
401 
402 /**
403  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
404  * @sdp: the filesystem
405  *
406  * Returns: errno
407  */
408 
gfs2_make_fs_rw(struct gfs2_sbd * sdp)409 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
410 {
411 	struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
412 	struct gfs2_glock *j_gl = ip->i_gl;
413 	struct gfs2_holder freeze_gh;
414 	struct gfs2_log_header_host head;
415 	int error;
416 
417 	error = init_threads(sdp);
418 	if (error)
419 		return error;
420 
421 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
422 				   &freeze_gh);
423 	if (error)
424 		goto fail_threads;
425 
426 	j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
427 
428 	error = gfs2_find_jhead(sdp->sd_jdesc, &head);
429 	if (error)
430 		goto fail;
431 
432 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
433 		gfs2_consist(sdp);
434 		error = -EIO;
435 		goto fail;
436 	}
437 
438 	/*  Initialize some head of the log stuff  */
439 	sdp->sd_log_sequence = head.lh_sequence + 1;
440 	gfs2_log_pointers_init(sdp, head.lh_blkno);
441 
442 	error = gfs2_quota_init(sdp);
443 	if (error)
444 		goto fail;
445 
446 	set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
447 
448 	gfs2_glock_dq_uninit(&freeze_gh);
449 
450 	return 0;
451 
452 fail:
453 	freeze_gh.gh_flags |= GL_NOCACHE;
454 	gfs2_glock_dq_uninit(&freeze_gh);
455 fail_threads:
456 	kthread_stop(sdp->sd_quotad_process);
457 	kthread_stop(sdp->sd_logd_process);
458 	return error;
459 }
460 
gfs2_statfs_change_in(struct gfs2_statfs_change_host * sc,const void * buf)461 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
462 {
463 	const struct gfs2_statfs_change *str = buf;
464 
465 	sc->sc_total = be64_to_cpu(str->sc_total);
466 	sc->sc_free = be64_to_cpu(str->sc_free);
467 	sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
468 }
469 
gfs2_statfs_change_out(const struct gfs2_statfs_change_host * sc,void * buf)470 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
471 {
472 	struct gfs2_statfs_change *str = buf;
473 
474 	str->sc_total = cpu_to_be64(sc->sc_total);
475 	str->sc_free = cpu_to_be64(sc->sc_free);
476 	str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
477 }
478 
gfs2_statfs_init(struct gfs2_sbd * sdp)479 int gfs2_statfs_init(struct gfs2_sbd *sdp)
480 {
481 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
482 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
483 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
484 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
485 	struct buffer_head *m_bh, *l_bh;
486 	struct gfs2_holder gh;
487 	int error;
488 
489 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
490 				   &gh);
491 	if (error)
492 		return error;
493 
494 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
495 	if (error)
496 		goto out;
497 
498 	if (sdp->sd_args.ar_spectator) {
499 		spin_lock(&sdp->sd_statfs_spin);
500 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
501 				      sizeof(struct gfs2_dinode));
502 		spin_unlock(&sdp->sd_statfs_spin);
503 	} else {
504 		error = gfs2_meta_inode_buffer(l_ip, &l_bh);
505 		if (error)
506 			goto out_m_bh;
507 
508 		spin_lock(&sdp->sd_statfs_spin);
509 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
510 				      sizeof(struct gfs2_dinode));
511 		gfs2_statfs_change_in(l_sc, l_bh->b_data +
512 				      sizeof(struct gfs2_dinode));
513 		spin_unlock(&sdp->sd_statfs_spin);
514 
515 		brelse(l_bh);
516 	}
517 
518 out_m_bh:
519 	brelse(m_bh);
520 out:
521 	gfs2_glock_dq_uninit(&gh);
522 	return 0;
523 }
524 
gfs2_statfs_change(struct gfs2_sbd * sdp,s64 total,s64 free,s64 dinodes)525 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
526 			s64 dinodes)
527 {
528 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
529 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
530 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
531 	struct buffer_head *l_bh;
532 	s64 x, y;
533 	int need_sync = 0;
534 	int error;
535 
536 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
537 	if (error)
538 		return;
539 
540 	gfs2_trans_add_meta(l_ip->i_gl, l_bh);
541 
542 	spin_lock(&sdp->sd_statfs_spin);
543 	l_sc->sc_total += total;
544 	l_sc->sc_free += free;
545 	l_sc->sc_dinodes += dinodes;
546 	gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
547 	if (sdp->sd_args.ar_statfs_percent) {
548 		x = 100 * l_sc->sc_free;
549 		y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
550 		if (x >= y || x <= -y)
551 			need_sync = 1;
552 	}
553 	spin_unlock(&sdp->sd_statfs_spin);
554 
555 	brelse(l_bh);
556 	if (need_sync)
557 		gfs2_wake_up_statfs(sdp);
558 }
559 
update_statfs(struct gfs2_sbd * sdp,struct buffer_head * m_bh,struct buffer_head * l_bh)560 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
561 		   struct buffer_head *l_bh)
562 {
563 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
564 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
565 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
566 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
567 
568 	gfs2_trans_add_meta(l_ip->i_gl, l_bh);
569 	gfs2_trans_add_meta(m_ip->i_gl, m_bh);
570 
571 	spin_lock(&sdp->sd_statfs_spin);
572 	m_sc->sc_total += l_sc->sc_total;
573 	m_sc->sc_free += l_sc->sc_free;
574 	m_sc->sc_dinodes += l_sc->sc_dinodes;
575 	memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
576 	memset(l_bh->b_data + sizeof(struct gfs2_dinode),
577 	       0, sizeof(struct gfs2_statfs_change));
578 	gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
579 	spin_unlock(&sdp->sd_statfs_spin);
580 }
581 
gfs2_statfs_sync(struct super_block * sb,int type)582 int gfs2_statfs_sync(struct super_block *sb, int type)
583 {
584 	struct gfs2_sbd *sdp = sb->s_fs_info;
585 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
586 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
587 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
588 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
589 	struct gfs2_holder gh;
590 	struct buffer_head *m_bh, *l_bh;
591 	int error;
592 
593 	sb_start_write(sb);
594 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
595 				   &gh);
596 	if (error)
597 		goto out;
598 
599 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
600 	if (error)
601 		goto out_unlock;
602 
603 	spin_lock(&sdp->sd_statfs_spin);
604 	gfs2_statfs_change_in(m_sc, m_bh->b_data +
605 			      sizeof(struct gfs2_dinode));
606 	if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
607 		spin_unlock(&sdp->sd_statfs_spin);
608 		goto out_bh;
609 	}
610 	spin_unlock(&sdp->sd_statfs_spin);
611 
612 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
613 	if (error)
614 		goto out_bh;
615 
616 	error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
617 	if (error)
618 		goto out_bh2;
619 
620 	update_statfs(sdp, m_bh, l_bh);
621 	sdp->sd_statfs_force_sync = 0;
622 
623 	gfs2_trans_end(sdp);
624 
625 out_bh2:
626 	brelse(l_bh);
627 out_bh:
628 	brelse(m_bh);
629 out_unlock:
630 	gfs2_glock_dq_uninit(&gh);
631 out:
632 	sb_end_write(sb);
633 	return error;
634 }
635 
636 struct lfcc {
637 	struct list_head list;
638 	struct gfs2_holder gh;
639 };
640 
641 /**
642  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
643  *                            journals are clean
644  * @sdp: the file system
645  * @state: the state to put the transaction lock into
646  * @t_gh: the hold on the transaction lock
647  *
648  * Returns: errno
649  */
650 
gfs2_lock_fs_check_clean(struct gfs2_sbd * sdp,struct gfs2_holder * freeze_gh)651 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
652 				    struct gfs2_holder *freeze_gh)
653 {
654 	struct gfs2_inode *ip;
655 	struct gfs2_jdesc *jd;
656 	struct lfcc *lfcc;
657 	LIST_HEAD(list);
658 	struct gfs2_log_header_host lh;
659 	int error;
660 
661 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
662 		lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
663 		if (!lfcc) {
664 			error = -ENOMEM;
665 			goto out;
666 		}
667 		ip = GFS2_I(jd->jd_inode);
668 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
669 		if (error) {
670 			kfree(lfcc);
671 			goto out;
672 		}
673 		list_add(&lfcc->list, &list);
674 	}
675 
676 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
677 				   GL_NOCACHE, freeze_gh);
678 
679 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
680 		error = gfs2_jdesc_check(jd);
681 		if (error)
682 			break;
683 		error = gfs2_find_jhead(jd, &lh);
684 		if (error)
685 			break;
686 		if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
687 			error = -EBUSY;
688 			break;
689 		}
690 	}
691 
692 	if (error)
693 		gfs2_glock_dq_uninit(freeze_gh);
694 
695 out:
696 	while (!list_empty(&list)) {
697 		lfcc = list_entry(list.next, struct lfcc, list);
698 		list_del(&lfcc->list);
699 		gfs2_glock_dq_uninit(&lfcc->gh);
700 		kfree(lfcc);
701 	}
702 	return error;
703 }
704 
gfs2_dinode_out(const struct gfs2_inode * ip,void * buf)705 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
706 {
707 	struct gfs2_dinode *str = buf;
708 
709 	str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
710 	str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
711 	str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
712 	str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
713 	str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
714 	str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
715 	str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
716 	str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
717 	str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
718 	str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
719 	str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
720 	str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
721 	str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
722 	str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
723 
724 	str->di_goal_meta = cpu_to_be64(ip->i_goal);
725 	str->di_goal_data = cpu_to_be64(ip->i_goal);
726 	str->di_generation = cpu_to_be64(ip->i_generation);
727 
728 	str->di_flags = cpu_to_be32(ip->i_diskflags);
729 	str->di_height = cpu_to_be16(ip->i_height);
730 	str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
731 					     !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
732 					     GFS2_FORMAT_DE : 0);
733 	str->di_depth = cpu_to_be16(ip->i_depth);
734 	str->di_entries = cpu_to_be32(ip->i_entries);
735 
736 	str->di_eattr = cpu_to_be64(ip->i_eattr);
737 	str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
738 	str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
739 	str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
740 }
741 
742 /**
743  * gfs2_write_inode - Make sure the inode is stable on the disk
744  * @inode: The inode
745  * @wbc: The writeback control structure
746  *
747  * Returns: errno
748  */
749 
gfs2_write_inode(struct inode * inode,struct writeback_control * wbc)750 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
751 {
752 	struct gfs2_inode *ip = GFS2_I(inode);
753 	struct gfs2_sbd *sdp = GFS2_SB(inode);
754 	struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
755 	struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
756 	int ret = 0;
757 	bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
758 
759 	if (flush_all)
760 		gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
761 			       GFS2_LOG_HEAD_FLUSH_NORMAL |
762 			       GFS2_LFC_WRITE_INODE);
763 	if (bdi->wb.dirty_exceeded)
764 		gfs2_ail1_flush(sdp, wbc);
765 	else
766 		filemap_fdatawrite(metamapping);
767 	if (flush_all)
768 		ret = filemap_fdatawait(metamapping);
769 	if (ret)
770 		mark_inode_dirty_sync(inode);
771 	else {
772 		spin_lock(&inode->i_lock);
773 		if (!(inode->i_flags & I_DIRTY))
774 			gfs2_ordered_del_inode(ip);
775 		spin_unlock(&inode->i_lock);
776 	}
777 	return ret;
778 }
779 
780 /**
781  * gfs2_dirty_inode - check for atime updates
782  * @inode: The inode in question
783  * @flags: The type of dirty
784  *
785  * Unfortunately it can be called under any combination of inode
786  * glock and transaction lock, so we have to check carefully.
787  *
788  * At the moment this deals only with atime - it should be possible
789  * to expand that role in future, once a review of the locking has
790  * been carried out.
791  */
792 
gfs2_dirty_inode(struct inode * inode,int flags)793 static void gfs2_dirty_inode(struct inode *inode, int flags)
794 {
795 	struct gfs2_inode *ip = GFS2_I(inode);
796 	struct gfs2_sbd *sdp = GFS2_SB(inode);
797 	struct buffer_head *bh;
798 	struct gfs2_holder gh;
799 	int need_unlock = 0;
800 	int need_endtrans = 0;
801 	int ret;
802 
803 	if (!(flags & I_DIRTY_INODE))
804 		return;
805 	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
806 		return;
807 	if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
808 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
809 		if (ret) {
810 			fs_err(sdp, "dirty_inode: glock %d\n", ret);
811 			return;
812 		}
813 		need_unlock = 1;
814 	} else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
815 		return;
816 
817 	if (current->journal_info == NULL) {
818 		ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
819 		if (ret) {
820 			fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
821 			goto out;
822 		}
823 		need_endtrans = 1;
824 	}
825 
826 	ret = gfs2_meta_inode_buffer(ip, &bh);
827 	if (ret == 0) {
828 		gfs2_trans_add_meta(ip->i_gl, bh);
829 		gfs2_dinode_out(ip, bh->b_data);
830 		brelse(bh);
831 	}
832 
833 	if (need_endtrans)
834 		gfs2_trans_end(sdp);
835 out:
836 	if (need_unlock)
837 		gfs2_glock_dq_uninit(&gh);
838 }
839 
840 /**
841  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
842  * @sdp: the filesystem
843  *
844  * Returns: errno
845  */
846 
gfs2_make_fs_ro(struct gfs2_sbd * sdp)847 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
848 {
849 	struct gfs2_holder freeze_gh;
850 	int error;
851 
852 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, GL_NOCACHE,
853 				   &freeze_gh);
854 	if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
855 		return error;
856 
857 	flush_workqueue(gfs2_delete_workqueue);
858 	kthread_stop(sdp->sd_quotad_process);
859 	kthread_stop(sdp->sd_logd_process);
860 
861 	gfs2_quota_sync(sdp->sd_vfs, 0);
862 	gfs2_statfs_sync(sdp->sd_vfs, 0);
863 
864 	gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
865 		       GFS2_LFC_MAKE_FS_RO);
866 	wait_event(sdp->sd_reserving_log_wait, atomic_read(&sdp->sd_reserving_log) == 0);
867 	gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
868 
869 	if (gfs2_holder_initialized(&freeze_gh))
870 		gfs2_glock_dq_uninit(&freeze_gh);
871 
872 	gfs2_quota_cleanup(sdp);
873 
874 	return error;
875 }
876 
877 /**
878  * gfs2_put_super - Unmount the filesystem
879  * @sb: The VFS superblock
880  *
881  */
882 
gfs2_put_super(struct super_block * sb)883 static void gfs2_put_super(struct super_block *sb)
884 {
885 	struct gfs2_sbd *sdp = sb->s_fs_info;
886 	int error;
887 	struct gfs2_jdesc *jd;
888 
889 	/* No more recovery requests */
890 	set_bit(SDF_NORECOVERY, &sdp->sd_flags);
891 	smp_mb();
892 
893 	/* Wait on outstanding recovery */
894 restart:
895 	spin_lock(&sdp->sd_jindex_spin);
896 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
897 		if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
898 			continue;
899 		spin_unlock(&sdp->sd_jindex_spin);
900 		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
901 			    TASK_UNINTERRUPTIBLE);
902 		goto restart;
903 	}
904 	spin_unlock(&sdp->sd_jindex_spin);
905 
906 	if (!sb_rdonly(sb)) {
907 		error = gfs2_make_fs_ro(sdp);
908 		if (error)
909 			gfs2_io_error(sdp);
910 	}
911 	/*  At this point, we're through modifying the disk  */
912 
913 	/*  Release stuff  */
914 
915 	iput(sdp->sd_jindex);
916 	iput(sdp->sd_statfs_inode);
917 	iput(sdp->sd_rindex);
918 	iput(sdp->sd_quota_inode);
919 
920 	gfs2_glock_put(sdp->sd_rename_gl);
921 	gfs2_glock_put(sdp->sd_freeze_gl);
922 
923 	if (!sdp->sd_args.ar_spectator) {
924 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
925 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
926 		gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
927 		gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
928 		iput(sdp->sd_sc_inode);
929 		iput(sdp->sd_qc_inode);
930 	}
931 
932 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
933 	gfs2_clear_rgrpd(sdp);
934 	gfs2_jindex_free(sdp);
935 	/*  Take apart glock structures and buffer lists  */
936 	gfs2_gl_hash_clear(sdp);
937 	truncate_inode_pages_final(&sdp->sd_aspace);
938 	gfs2_delete_debugfs_file(sdp);
939 	/*  Unmount the locking protocol  */
940 	gfs2_lm_unmount(sdp);
941 
942 	/*  At this point, we're through participating in the lockspace  */
943 	gfs2_sys_fs_del(sdp);
944 }
945 
946 /**
947  * gfs2_sync_fs - sync the filesystem
948  * @sb: the superblock
949  *
950  * Flushes the log to disk.
951  */
952 
gfs2_sync_fs(struct super_block * sb,int wait)953 static int gfs2_sync_fs(struct super_block *sb, int wait)
954 {
955 	struct gfs2_sbd *sdp = sb->s_fs_info;
956 
957 	gfs2_quota_sync(sb, -1);
958 	if (wait)
959 		gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
960 			       GFS2_LFC_SYNC_FS);
961 	return sdp->sd_log_error;
962 }
963 
gfs2_freeze_func(struct work_struct * work)964 void gfs2_freeze_func(struct work_struct *work)
965 {
966 	int error;
967 	struct gfs2_holder freeze_gh;
968 	struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
969 	struct super_block *sb = sdp->sd_vfs;
970 
971 	atomic_inc(&sb->s_active);
972 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
973 				   &freeze_gh);
974 	if (error) {
975 		printk(KERN_INFO "GFS2: couln't get freeze lock : %d\n", error);
976 		gfs2_assert_withdraw(sdp, 0);
977 	}
978 	else {
979 		atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
980 		error = thaw_super(sb);
981 		if (error) {
982 			printk(KERN_INFO "GFS2: couldn't thaw filesystem: %d\n",
983 			       error);
984 			gfs2_assert_withdraw(sdp, 0);
985 		}
986 		if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
987 			freeze_gh.gh_flags |= GL_NOCACHE;
988 		gfs2_glock_dq_uninit(&freeze_gh);
989 	}
990 	deactivate_super(sb);
991 	return;
992 }
993 
994 /**
995  * gfs2_freeze - prevent further writes to the filesystem
996  * @sb: the VFS structure for the filesystem
997  *
998  */
999 
gfs2_freeze(struct super_block * sb)1000 static int gfs2_freeze(struct super_block *sb)
1001 {
1002 	struct gfs2_sbd *sdp = sb->s_fs_info;
1003 	int error;
1004 
1005 	mutex_lock(&sdp->sd_freeze_mutex);
1006 	if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) {
1007 		error = -EBUSY;
1008 		goto out;
1009 	}
1010 
1011 	if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
1012 		error = -EINVAL;
1013 		goto out;
1014 	}
1015 
1016 	for (;;) {
1017 		error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
1018 		if (!error)
1019 			break;
1020 
1021 		switch (error) {
1022 		case -EBUSY:
1023 			fs_err(sdp, "waiting for recovery before freeze\n");
1024 			break;
1025 
1026 		default:
1027 			fs_err(sdp, "error freezing FS: %d\n", error);
1028 			break;
1029 		}
1030 
1031 		fs_err(sdp, "retrying...\n");
1032 		msleep(1000);
1033 	}
1034 	error = 0;
1035 out:
1036 	mutex_unlock(&sdp->sd_freeze_mutex);
1037 	return error;
1038 }
1039 
1040 /**
1041  * gfs2_unfreeze - reallow writes to the filesystem
1042  * @sb: the VFS structure for the filesystem
1043  *
1044  */
1045 
gfs2_unfreeze(struct super_block * sb)1046 static int gfs2_unfreeze(struct super_block *sb)
1047 {
1048 	struct gfs2_sbd *sdp = sb->s_fs_info;
1049 
1050 	mutex_lock(&sdp->sd_freeze_mutex);
1051 	if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
1052 	    !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
1053 		mutex_unlock(&sdp->sd_freeze_mutex);
1054 		return -EINVAL;
1055 	}
1056 
1057 	gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1058 	mutex_unlock(&sdp->sd_freeze_mutex);
1059 	return 0;
1060 }
1061 
1062 /**
1063  * statfs_fill - fill in the sg for a given RG
1064  * @rgd: the RG
1065  * @sc: the sc structure
1066  *
1067  * Returns: 0 on success, -ESTALE if the LVB is invalid
1068  */
1069 
statfs_slow_fill(struct gfs2_rgrpd * rgd,struct gfs2_statfs_change_host * sc)1070 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1071 			    struct gfs2_statfs_change_host *sc)
1072 {
1073 	gfs2_rgrp_verify(rgd);
1074 	sc->sc_total += rgd->rd_data;
1075 	sc->sc_free += rgd->rd_free;
1076 	sc->sc_dinodes += rgd->rd_dinodes;
1077 	return 0;
1078 }
1079 
1080 /**
1081  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1082  * @sdp: the filesystem
1083  * @sc: the sc info that will be returned
1084  *
1085  * Any error (other than a signal) will cause this routine to fall back
1086  * to the synchronous version.
1087  *
1088  * FIXME: This really shouldn't busy wait like this.
1089  *
1090  * Returns: errno
1091  */
1092 
gfs2_statfs_slow(struct gfs2_sbd * sdp,struct gfs2_statfs_change_host * sc)1093 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1094 {
1095 	struct gfs2_rgrpd *rgd_next;
1096 	struct gfs2_holder *gha, *gh;
1097 	unsigned int slots = 64;
1098 	unsigned int x;
1099 	int done;
1100 	int error = 0, err;
1101 
1102 	memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1103 	gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1104 	if (!gha)
1105 		return -ENOMEM;
1106 	for (x = 0; x < slots; x++)
1107 		gfs2_holder_mark_uninitialized(gha + x);
1108 
1109 	rgd_next = gfs2_rgrpd_get_first(sdp);
1110 
1111 	for (;;) {
1112 		done = 1;
1113 
1114 		for (x = 0; x < slots; x++) {
1115 			gh = gha + x;
1116 
1117 			if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
1118 				err = gfs2_glock_wait(gh);
1119 				if (err) {
1120 					gfs2_holder_uninit(gh);
1121 					error = err;
1122 				} else {
1123 					if (!error) {
1124 						struct gfs2_rgrpd *rgd =
1125 							gfs2_glock2rgrp(gh->gh_gl);
1126 
1127 						error = statfs_slow_fill(rgd, sc);
1128 					}
1129 					gfs2_glock_dq_uninit(gh);
1130 				}
1131 			}
1132 
1133 			if (gfs2_holder_initialized(gh))
1134 				done = 0;
1135 			else if (rgd_next && !error) {
1136 				error = gfs2_glock_nq_init(rgd_next->rd_gl,
1137 							   LM_ST_SHARED,
1138 							   GL_ASYNC,
1139 							   gh);
1140 				rgd_next = gfs2_rgrpd_get_next(rgd_next);
1141 				done = 0;
1142 			}
1143 
1144 			if (signal_pending(current))
1145 				error = -ERESTARTSYS;
1146 		}
1147 
1148 		if (done)
1149 			break;
1150 
1151 		yield();
1152 	}
1153 
1154 	kfree(gha);
1155 	return error;
1156 }
1157 
1158 /**
1159  * gfs2_statfs_i - Do a statfs
1160  * @sdp: the filesystem
1161  * @sg: the sg structure
1162  *
1163  * Returns: errno
1164  */
1165 
gfs2_statfs_i(struct gfs2_sbd * sdp,struct gfs2_statfs_change_host * sc)1166 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1167 {
1168 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1169 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1170 
1171 	spin_lock(&sdp->sd_statfs_spin);
1172 
1173 	*sc = *m_sc;
1174 	sc->sc_total += l_sc->sc_total;
1175 	sc->sc_free += l_sc->sc_free;
1176 	sc->sc_dinodes += l_sc->sc_dinodes;
1177 
1178 	spin_unlock(&sdp->sd_statfs_spin);
1179 
1180 	if (sc->sc_free < 0)
1181 		sc->sc_free = 0;
1182 	if (sc->sc_free > sc->sc_total)
1183 		sc->sc_free = sc->sc_total;
1184 	if (sc->sc_dinodes < 0)
1185 		sc->sc_dinodes = 0;
1186 
1187 	return 0;
1188 }
1189 
1190 /**
1191  * gfs2_statfs - Gather and return stats about the filesystem
1192  * @sb: The superblock
1193  * @statfsbuf: The buffer
1194  *
1195  * Returns: 0 on success or error code
1196  */
1197 
gfs2_statfs(struct dentry * dentry,struct kstatfs * buf)1198 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1199 {
1200 	struct super_block *sb = dentry->d_sb;
1201 	struct gfs2_sbd *sdp = sb->s_fs_info;
1202 	struct gfs2_statfs_change_host sc;
1203 	int error;
1204 
1205 	error = gfs2_rindex_update(sdp);
1206 	if (error)
1207 		return error;
1208 
1209 	if (gfs2_tune_get(sdp, gt_statfs_slow))
1210 		error = gfs2_statfs_slow(sdp, &sc);
1211 	else
1212 		error = gfs2_statfs_i(sdp, &sc);
1213 
1214 	if (error)
1215 		return error;
1216 
1217 	buf->f_type = GFS2_MAGIC;
1218 	buf->f_bsize = sdp->sd_sb.sb_bsize;
1219 	buf->f_blocks = sc.sc_total;
1220 	buf->f_bfree = sc.sc_free;
1221 	buf->f_bavail = sc.sc_free;
1222 	buf->f_files = sc.sc_dinodes + sc.sc_free;
1223 	buf->f_ffree = sc.sc_free;
1224 	buf->f_namelen = GFS2_FNAMESIZE;
1225 
1226 	return 0;
1227 }
1228 
1229 /**
1230  * gfs2_remount_fs - called when the FS is remounted
1231  * @sb:  the filesystem
1232  * @flags:  the remount flags
1233  * @data:  extra data passed in (not used right now)
1234  *
1235  * Returns: errno
1236  */
1237 
gfs2_remount_fs(struct super_block * sb,int * flags,char * data)1238 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1239 {
1240 	struct gfs2_sbd *sdp = sb->s_fs_info;
1241 	struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1242 	struct gfs2_tune *gt = &sdp->sd_tune;
1243 	int error;
1244 
1245 	sync_filesystem(sb);
1246 
1247 	spin_lock(&gt->gt_spin);
1248 	args.ar_commit = gt->gt_logd_secs;
1249 	args.ar_quota_quantum = gt->gt_quota_quantum;
1250 	if (gt->gt_statfs_slow)
1251 		args.ar_statfs_quantum = 0;
1252 	else
1253 		args.ar_statfs_quantum = gt->gt_statfs_quantum;
1254 	spin_unlock(&gt->gt_spin);
1255 	error = gfs2_mount_args(&args, data);
1256 	if (error)
1257 		return error;
1258 
1259 	/* Not allowed to change locking details */
1260 	if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1261 	    strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1262 	    strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1263 		return -EINVAL;
1264 
1265 	/* Some flags must not be changed */
1266 	if (args_neq(&args, &sdp->sd_args, spectator) ||
1267 	    args_neq(&args, &sdp->sd_args, localflocks) ||
1268 	    args_neq(&args, &sdp->sd_args, meta))
1269 		return -EINVAL;
1270 
1271 	if (sdp->sd_args.ar_spectator)
1272 		*flags |= SB_RDONLY;
1273 
1274 	if ((sb->s_flags ^ *flags) & SB_RDONLY) {
1275 		if (*flags & SB_RDONLY)
1276 			error = gfs2_make_fs_ro(sdp);
1277 		else
1278 			error = gfs2_make_fs_rw(sdp);
1279 		if (error)
1280 			return error;
1281 	}
1282 
1283 	sdp->sd_args = args;
1284 	if (sdp->sd_args.ar_posix_acl)
1285 		sb->s_flags |= SB_POSIXACL;
1286 	else
1287 		sb->s_flags &= ~SB_POSIXACL;
1288 	if (sdp->sd_args.ar_nobarrier)
1289 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1290 	else
1291 		clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1292 	spin_lock(&gt->gt_spin);
1293 	gt->gt_logd_secs = args.ar_commit;
1294 	gt->gt_quota_quantum = args.ar_quota_quantum;
1295 	if (args.ar_statfs_quantum) {
1296 		gt->gt_statfs_slow = 0;
1297 		gt->gt_statfs_quantum = args.ar_statfs_quantum;
1298 	}
1299 	else {
1300 		gt->gt_statfs_slow = 1;
1301 		gt->gt_statfs_quantum = 30;
1302 	}
1303 	spin_unlock(&gt->gt_spin);
1304 
1305 	gfs2_online_uevent(sdp);
1306 	return 0;
1307 }
1308 
1309 /**
1310  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1311  * @inode: The inode to drop
1312  *
1313  * If we've received a callback on an iopen lock then it's because a
1314  * remote node tried to deallocate the inode but failed due to this node
1315  * still having the inode open. Here we mark the link count zero
1316  * since we know that it must have reached zero if the GLF_DEMOTE flag
1317  * is set on the iopen glock. If we didn't do a disk read since the
1318  * remote node removed the final link then we might otherwise miss
1319  * this event. This check ensures that this node will deallocate the
1320  * inode's blocks, or alternatively pass the baton on to another
1321  * node for later deallocation.
1322  */
1323 
gfs2_drop_inode(struct inode * inode)1324 static int gfs2_drop_inode(struct inode *inode)
1325 {
1326 	struct gfs2_inode *ip = GFS2_I(inode);
1327 
1328 	if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
1329 	    inode->i_nlink &&
1330 	    gfs2_holder_initialized(&ip->i_iopen_gh)) {
1331 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1332 		if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1333 			clear_nlink(inode);
1334 	}
1335 
1336 	/*
1337 	 * When under memory pressure when an inode's link count has dropped to
1338 	 * zero, defer deleting the inode to the delete workqueue.  This avoids
1339 	 * calling into DLM under memory pressure, which can deadlock.
1340 	 */
1341 	if (!inode->i_nlink &&
1342 	    unlikely(current->flags & PF_MEMALLOC) &&
1343 	    gfs2_holder_initialized(&ip->i_iopen_gh)) {
1344 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1345 
1346 		gfs2_glock_hold(gl);
1347 		if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
1348 			gfs2_glock_queue_put(gl);
1349 		return false;
1350 	}
1351 
1352 	return generic_drop_inode(inode);
1353 }
1354 
is_ancestor(const struct dentry * d1,const struct dentry * d2)1355 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1356 {
1357 	do {
1358 		if (d1 == d2)
1359 			return 1;
1360 		d1 = d1->d_parent;
1361 	} while (!IS_ROOT(d1));
1362 	return 0;
1363 }
1364 
1365 /**
1366  * gfs2_show_options - Show mount options for /proc/mounts
1367  * @s: seq_file structure
1368  * @root: root of this (sub)tree
1369  *
1370  * Returns: 0 on success or error code
1371  */
1372 
gfs2_show_options(struct seq_file * s,struct dentry * root)1373 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1374 {
1375 	struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1376 	struct gfs2_args *args = &sdp->sd_args;
1377 	unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum;
1378 
1379 	spin_lock(&sdp->sd_tune.gt_spin);
1380 	logd_secs = sdp->sd_tune.gt_logd_secs;
1381 	quota_quantum = sdp->sd_tune.gt_quota_quantum;
1382 	statfs_quantum = sdp->sd_tune.gt_statfs_quantum;
1383 	statfs_slow = sdp->sd_tune.gt_statfs_slow;
1384 	spin_unlock(&sdp->sd_tune.gt_spin);
1385 
1386 	if (is_ancestor(root, sdp->sd_master_dir))
1387 		seq_puts(s, ",meta");
1388 	if (args->ar_lockproto[0])
1389 		seq_show_option(s, "lockproto", args->ar_lockproto);
1390 	if (args->ar_locktable[0])
1391 		seq_show_option(s, "locktable", args->ar_locktable);
1392 	if (args->ar_hostdata[0])
1393 		seq_show_option(s, "hostdata", args->ar_hostdata);
1394 	if (args->ar_spectator)
1395 		seq_puts(s, ",spectator");
1396 	if (args->ar_localflocks)
1397 		seq_puts(s, ",localflocks");
1398 	if (args->ar_debug)
1399 		seq_puts(s, ",debug");
1400 	if (args->ar_posix_acl)
1401 		seq_puts(s, ",acl");
1402 	if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1403 		char *state;
1404 		switch (args->ar_quota) {
1405 		case GFS2_QUOTA_OFF:
1406 			state = "off";
1407 			break;
1408 		case GFS2_QUOTA_ACCOUNT:
1409 			state = "account";
1410 			break;
1411 		case GFS2_QUOTA_ON:
1412 			state = "on";
1413 			break;
1414 		default:
1415 			state = "unknown";
1416 			break;
1417 		}
1418 		seq_printf(s, ",quota=%s", state);
1419 	}
1420 	if (args->ar_suiddir)
1421 		seq_puts(s, ",suiddir");
1422 	if (args->ar_data != GFS2_DATA_DEFAULT) {
1423 		char *state;
1424 		switch (args->ar_data) {
1425 		case GFS2_DATA_WRITEBACK:
1426 			state = "writeback";
1427 			break;
1428 		case GFS2_DATA_ORDERED:
1429 			state = "ordered";
1430 			break;
1431 		default:
1432 			state = "unknown";
1433 			break;
1434 		}
1435 		seq_printf(s, ",data=%s", state);
1436 	}
1437 	if (args->ar_discard)
1438 		seq_puts(s, ",discard");
1439 	if (logd_secs != 30)
1440 		seq_printf(s, ",commit=%d", logd_secs);
1441 	if (statfs_quantum != 30)
1442 		seq_printf(s, ",statfs_quantum=%d", statfs_quantum);
1443 	else if (statfs_slow)
1444 		seq_puts(s, ",statfs_quantum=0");
1445 	if (quota_quantum != 60)
1446 		seq_printf(s, ",quota_quantum=%d", quota_quantum);
1447 	if (args->ar_statfs_percent)
1448 		seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1449 	if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1450 		const char *state;
1451 
1452 		switch (args->ar_errors) {
1453 		case GFS2_ERRORS_WITHDRAW:
1454 			state = "withdraw";
1455 			break;
1456 		case GFS2_ERRORS_PANIC:
1457 			state = "panic";
1458 			break;
1459 		default:
1460 			state = "unknown";
1461 			break;
1462 		}
1463 		seq_printf(s, ",errors=%s", state);
1464 	}
1465 	if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1466 		seq_puts(s, ",nobarrier");
1467 	if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1468 		seq_puts(s, ",demote_interface_used");
1469 	if (args->ar_rgrplvb)
1470 		seq_puts(s, ",rgrplvb");
1471 	if (args->ar_loccookie)
1472 		seq_puts(s, ",loccookie");
1473 	return 0;
1474 }
1475 
gfs2_final_release_pages(struct gfs2_inode * ip)1476 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1477 {
1478 	struct inode *inode = &ip->i_inode;
1479 	struct gfs2_glock *gl = ip->i_gl;
1480 
1481 	truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1482 	truncate_inode_pages(&inode->i_data, 0);
1483 
1484 	if (atomic_read(&gl->gl_revokes) == 0) {
1485 		clear_bit(GLF_LFLUSH, &gl->gl_flags);
1486 		clear_bit(GLF_DIRTY, &gl->gl_flags);
1487 	}
1488 }
1489 
gfs2_dinode_dealloc(struct gfs2_inode * ip)1490 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1491 {
1492 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1493 	struct gfs2_rgrpd *rgd;
1494 	struct gfs2_holder gh;
1495 	int error;
1496 
1497 	if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1498 		gfs2_consist_inode(ip);
1499 		return -EIO;
1500 	}
1501 
1502 	error = gfs2_rindex_update(sdp);
1503 	if (error)
1504 		return error;
1505 
1506 	error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1507 	if (error)
1508 		return error;
1509 
1510 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1511 	if (!rgd) {
1512 		gfs2_consist_inode(ip);
1513 		error = -EIO;
1514 		goto out_qs;
1515 	}
1516 
1517 	error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1518 	if (error)
1519 		goto out_qs;
1520 
1521 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1522 				 sdp->sd_jdesc->jd_blocks);
1523 	if (error)
1524 		goto out_rg_gunlock;
1525 
1526 	gfs2_free_di(rgd, ip);
1527 
1528 	gfs2_final_release_pages(ip);
1529 
1530 	gfs2_trans_end(sdp);
1531 
1532 out_rg_gunlock:
1533 	gfs2_glock_dq_uninit(&gh);
1534 out_qs:
1535 	gfs2_quota_unhold(ip);
1536 	return error;
1537 }
1538 
1539 /**
1540  * gfs2_glock_put_eventually
1541  * @gl:	The glock to put
1542  *
1543  * When under memory pressure, trigger a deferred glock put to make sure we
1544  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1545  */
1546 
gfs2_glock_put_eventually(struct gfs2_glock * gl)1547 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1548 {
1549 	if (current->flags & PF_MEMALLOC)
1550 		gfs2_glock_queue_put(gl);
1551 	else
1552 		gfs2_glock_put(gl);
1553 }
1554 
1555 /**
1556  * gfs2_evict_inode - Remove an inode from cache
1557  * @inode: The inode to evict
1558  *
1559  * There are three cases to consider:
1560  * 1. i_nlink == 0, we are final opener (and must deallocate)
1561  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1562  * 3. i_nlink > 0
1563  *
1564  * If the fs is read only, then we have to treat all cases as per #3
1565  * since we are unable to do any deallocation. The inode will be
1566  * deallocated by the next read/write node to attempt an allocation
1567  * in the same resource group
1568  *
1569  * We have to (at the moment) hold the inodes main lock to cover
1570  * the gap between unlocking the shared lock on the iopen lock and
1571  * taking the exclusive lock. I'd rather do a shared -> exclusive
1572  * conversion on the iopen lock, but we can change that later. This
1573  * is safe, just less efficient.
1574  */
1575 
gfs2_evict_inode(struct inode * inode)1576 static void gfs2_evict_inode(struct inode *inode)
1577 {
1578 	struct super_block *sb = inode->i_sb;
1579 	struct gfs2_sbd *sdp = sb->s_fs_info;
1580 	struct gfs2_inode *ip = GFS2_I(inode);
1581 	struct gfs2_holder gh;
1582 	struct address_space *metamapping;
1583 	int error;
1584 
1585 	if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1586 		clear_inode(inode);
1587 		return;
1588 	}
1589 
1590 	if (inode->i_nlink || sb_rdonly(sb))
1591 		goto out;
1592 
1593 	/*
1594 	 * In case of an incomplete mount, gfs2_evict_inode() may be called for
1595 	 * system files without having an active journal to write to.  In that
1596 	 * case, skip the filesystem evict.
1597 	 */
1598 	if (!sdp->sd_jdesc)
1599 		goto out;
1600 
1601 	if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1602 		BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1603 		gfs2_holder_mark_uninitialized(&gh);
1604 		goto alloc_failed;
1605 	}
1606 
1607 	/* Deletes should never happen under memory pressure anymore.  */
1608 	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1609 		goto out;
1610 
1611 	/* Must not read inode block until block type has been verified */
1612 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1613 	if (unlikely(error)) {
1614 		glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1615 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1616 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1617 		goto out;
1618 	}
1619 
1620 	error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1621 	if (error)
1622 		goto out_truncate;
1623 
1624 	if (test_bit(GIF_INVALID, &ip->i_flags)) {
1625 		error = gfs2_inode_refresh(ip);
1626 		if (error)
1627 			goto out_truncate;
1628 	}
1629 
1630 	/*
1631 	 * The inode may have been recreated in the meantime.
1632 	 */
1633 	if (inode->i_nlink)
1634 		goto out_truncate;
1635 
1636 alloc_failed:
1637 	if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1638 	    test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1639 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1640 		gfs2_glock_dq_wait(&ip->i_iopen_gh);
1641 		gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE,
1642 				   &ip->i_iopen_gh);
1643 		error = gfs2_glock_nq(&ip->i_iopen_gh);
1644 		if (error)
1645 			goto out_truncate;
1646 	}
1647 
1648 	/* Case 1 starts here */
1649 
1650 	if (S_ISDIR(inode->i_mode) &&
1651 	    (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1652 		error = gfs2_dir_exhash_dealloc(ip);
1653 		if (error)
1654 			goto out_unlock;
1655 	}
1656 
1657 	if (ip->i_eattr) {
1658 		error = gfs2_ea_dealloc(ip);
1659 		if (error)
1660 			goto out_unlock;
1661 	}
1662 
1663 	if (!gfs2_is_stuffed(ip)) {
1664 		error = gfs2_file_dealloc(ip);
1665 		if (error)
1666 			goto out_unlock;
1667 	}
1668 
1669 	/* We're about to clear the bitmap for the dinode, but as soon as we
1670 	   do, gfs2_create_inode can create another inode at the same block
1671 	   location and try to set gl_object again. We clear gl_object here so
1672 	   that subsequent inode creates don't see an old gl_object. */
1673 	glock_clear_object(ip->i_gl, ip);
1674 	error = gfs2_dinode_dealloc(ip);
1675 	goto out_unlock;
1676 
1677 out_truncate:
1678 	gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1679 		       GFS2_LFC_EVICT_INODE);
1680 	metamapping = gfs2_glock2aspace(ip->i_gl);
1681 	if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1682 		filemap_fdatawrite(metamapping);
1683 		filemap_fdatawait(metamapping);
1684 	}
1685 	write_inode_now(inode, 1);
1686 	gfs2_ail_flush(ip->i_gl, 0);
1687 
1688 	/* Case 2 starts here */
1689 	error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1690 	if (error)
1691 		goto out_unlock;
1692 	/* Needs to be done before glock release & also in a transaction */
1693 	truncate_inode_pages(&inode->i_data, 0);
1694 	truncate_inode_pages(metamapping, 0);
1695 	gfs2_trans_end(sdp);
1696 
1697 out_unlock:
1698 	/* Error path for case 1 */
1699 	if (gfs2_rs_active(&ip->i_res))
1700 		gfs2_rs_deltree(&ip->i_res);
1701 
1702 	if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1703 		glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1704 		if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1705 			ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1706 			gfs2_glock_dq(&ip->i_iopen_gh);
1707 		}
1708 		gfs2_holder_uninit(&ip->i_iopen_gh);
1709 	}
1710 	if (gfs2_holder_initialized(&gh)) {
1711 		glock_clear_object(ip->i_gl, ip);
1712 		gfs2_glock_dq_uninit(&gh);
1713 	}
1714 	if (error && error != GLR_TRYFAILED && error != -EROFS)
1715 		fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1716 out:
1717 	/* Case 3 starts here */
1718 	truncate_inode_pages_final(&inode->i_data);
1719 	gfs2_rsqa_delete(ip, NULL);
1720 	gfs2_ordered_del_inode(ip);
1721 	clear_inode(inode);
1722 	gfs2_dir_hash_inval(ip);
1723 	glock_clear_object(ip->i_gl, ip);
1724 	wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1725 	gfs2_glock_add_to_lru(ip->i_gl);
1726 	gfs2_glock_put_eventually(ip->i_gl);
1727 	ip->i_gl = NULL;
1728 	if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1729 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1730 
1731 		glock_clear_object(gl, ip);
1732 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1733 		gfs2_glock_hold(gl);
1734 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1735 		gfs2_glock_put_eventually(gl);
1736 	}
1737 }
1738 
gfs2_alloc_inode(struct super_block * sb)1739 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1740 {
1741 	struct gfs2_inode *ip;
1742 
1743 	ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1744 	if (ip) {
1745 		ip->i_flags = 0;
1746 		ip->i_gl = NULL;
1747 		memset(&ip->i_res, 0, sizeof(ip->i_res));
1748 		RB_CLEAR_NODE(&ip->i_res.rs_node);
1749 		ip->i_rahead = 0;
1750 	}
1751 	return &ip->i_inode;
1752 }
1753 
gfs2_i_callback(struct rcu_head * head)1754 static void gfs2_i_callback(struct rcu_head *head)
1755 {
1756 	struct inode *inode = container_of(head, struct inode, i_rcu);
1757 	kmem_cache_free(gfs2_inode_cachep, inode);
1758 }
1759 
gfs2_destroy_inode(struct inode * inode)1760 static void gfs2_destroy_inode(struct inode *inode)
1761 {
1762 	call_rcu(&inode->i_rcu, gfs2_i_callback);
1763 }
1764 
1765 const struct super_operations gfs2_super_ops = {
1766 	.alloc_inode		= gfs2_alloc_inode,
1767 	.destroy_inode		= gfs2_destroy_inode,
1768 	.write_inode		= gfs2_write_inode,
1769 	.dirty_inode		= gfs2_dirty_inode,
1770 	.evict_inode		= gfs2_evict_inode,
1771 	.put_super		= gfs2_put_super,
1772 	.sync_fs		= gfs2_sync_fs,
1773 	.freeze_super		= gfs2_freeze,
1774 	.thaw_super		= gfs2_unfreeze,
1775 	.statfs			= gfs2_statfs,
1776 	.remount_fs		= gfs2_remount_fs,
1777 	.drop_inode		= gfs2_drop_inode,
1778 	.show_options		= gfs2_show_options,
1779 };
1780 
1781