xref: /linux/fs/gfs2/super.c (revision 2b8232ce512105e28453f301d1510de8363bccd1)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2006 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/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/crc32.h>
16 #include <linux/gfs2_ondisk.h>
17 #include <linux/bio.h>
18 #include <linux/lm_interface.h>
19 
20 #include "gfs2.h"
21 #include "incore.h"
22 #include "bmap.h"
23 #include "dir.h"
24 #include "glock.h"
25 #include "glops.h"
26 #include "inode.h"
27 #include "log.h"
28 #include "meta_io.h"
29 #include "quota.h"
30 #include "recovery.h"
31 #include "rgrp.h"
32 #include "super.h"
33 #include "trans.h"
34 #include "util.h"
35 
36 static const u32 gfs2_old_fs_formats[] = {
37         0
38 };
39 
40 static const u32 gfs2_old_multihost_formats[] = {
41         0
42 };
43 
44 /**
45  * gfs2_tune_init - Fill a gfs2_tune structure with default values
46  * @gt: tune
47  *
48  */
49 
50 void gfs2_tune_init(struct gfs2_tune *gt)
51 {
52 	spin_lock_init(&gt->gt_spin);
53 
54 	gt->gt_ilimit = 100;
55 	gt->gt_ilimit_tries = 3;
56 	gt->gt_ilimit_min = 1;
57 	gt->gt_demote_secs = 300;
58 	gt->gt_incore_log_blocks = 1024;
59 	gt->gt_log_flush_secs = 60;
60 	gt->gt_jindex_refresh_secs = 60;
61 	gt->gt_recoverd_secs = 60;
62 	gt->gt_logd_secs = 1;
63 	gt->gt_quotad_secs = 5;
64 	gt->gt_quota_simul_sync = 64;
65 	gt->gt_quota_warn_period = 10;
66 	gt->gt_quota_scale_num = 1;
67 	gt->gt_quota_scale_den = 1;
68 	gt->gt_quota_cache_secs = 300;
69 	gt->gt_quota_quantum = 60;
70 	gt->gt_atime_quantum = 3600;
71 	gt->gt_new_files_jdata = 0;
72 	gt->gt_new_files_directio = 0;
73 	gt->gt_max_readahead = 1 << 18;
74 	gt->gt_lockdump_size = 131072;
75 	gt->gt_stall_secs = 600;
76 	gt->gt_complain_secs = 10;
77 	gt->gt_reclaim_limit = 5000;
78 	gt->gt_statfs_quantum = 30;
79 	gt->gt_statfs_slow = 0;
80 }
81 
82 /**
83  * gfs2_check_sb - Check superblock
84  * @sdp: the filesystem
85  * @sb: The superblock
86  * @silent: Don't print a message if the check fails
87  *
88  * Checks the version code of the FS is one that we understand how to
89  * read and that the sizes of the various on-disk structures have not
90  * changed.
91  */
92 
93 int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
94 {
95 	unsigned int x;
96 
97 	if (sb->sb_magic != GFS2_MAGIC ||
98 	    sb->sb_type != GFS2_METATYPE_SB) {
99 		if (!silent)
100 			printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
101 		return -EINVAL;
102 	}
103 
104 	/*  If format numbers match exactly, we're done.  */
105 
106 	if (sb->sb_fs_format == GFS2_FORMAT_FS &&
107 	    sb->sb_multihost_format == GFS2_FORMAT_MULTI)
108 		return 0;
109 
110 	if (sb->sb_fs_format != GFS2_FORMAT_FS) {
111 		for (x = 0; gfs2_old_fs_formats[x]; x++)
112 			if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
113 				break;
114 
115 		if (!gfs2_old_fs_formats[x]) {
116 			printk(KERN_WARNING
117 			       "GFS2: code version (%u, %u) is incompatible "
118 			       "with ondisk format (%u, %u)\n",
119 			       GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
120 			       sb->sb_fs_format, sb->sb_multihost_format);
121 			printk(KERN_WARNING
122 			       "GFS2: I don't know how to upgrade this FS\n");
123 			return -EINVAL;
124 		}
125 	}
126 
127 	if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
128 		for (x = 0; gfs2_old_multihost_formats[x]; x++)
129 			if (gfs2_old_multihost_formats[x] ==
130 			    sb->sb_multihost_format)
131 				break;
132 
133 		if (!gfs2_old_multihost_formats[x]) {
134 			printk(KERN_WARNING
135 			       "GFS2: code version (%u, %u) is incompatible "
136 			       "with ondisk format (%u, %u)\n",
137 			       GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
138 			       sb->sb_fs_format, sb->sb_multihost_format);
139 			printk(KERN_WARNING
140 			       "GFS2: I don't know how to upgrade this FS\n");
141 			return -EINVAL;
142 		}
143 	}
144 
145 	if (!sdp->sd_args.ar_upgrade) {
146 		printk(KERN_WARNING
147 		       "GFS2: code version (%u, %u) is incompatible "
148 		       "with ondisk format (%u, %u)\n",
149 		       GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
150 		       sb->sb_fs_format, sb->sb_multihost_format);
151 		printk(KERN_INFO
152 		       "GFS2: Use the \"upgrade\" mount option to upgrade "
153 		       "the FS\n");
154 		printk(KERN_INFO "GFS2: See the manual for more details\n");
155 		return -EINVAL;
156 	}
157 
158 	return 0;
159 }
160 
161 
162 static void end_bio_io_page(struct bio *bio, int error)
163 {
164 	struct page *page = bio->bi_private;
165 
166 	if (!error)
167 		SetPageUptodate(page);
168 	else
169 		printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
170 	unlock_page(page);
171 }
172 
173 static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
174 {
175 	const struct gfs2_sb *str = buf;
176 
177 	sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
178 	sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
179 	sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
180 	sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
181 	sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
182 	sb->sb_bsize = be32_to_cpu(str->sb_bsize);
183 	sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
184 	sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
185 	sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
186 	sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
187 	sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
188 
189 	memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
190 	memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
191 }
192 
193 /**
194  * gfs2_read_super - Read the gfs2 super block from disk
195  * @sdp: The GFS2 super block
196  * @sector: The location of the super block
197  * @error: The error code to return
198  *
199  * This uses the bio functions to read the super block from disk
200  * because we want to be 100% sure that we never read cached data.
201  * A super block is read twice only during each GFS2 mount and is
202  * never written to by the filesystem. The first time its read no
203  * locks are held, and the only details which are looked at are those
204  * relating to the locking protocol. Once locking is up and working,
205  * the sb is read again under the lock to establish the location of
206  * the master directory (contains pointers to journals etc) and the
207  * root directory.
208  *
209  * Returns: 0 on success or error
210  */
211 
212 int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
213 {
214 	struct super_block *sb = sdp->sd_vfs;
215 	struct gfs2_sb *p;
216 	struct page *page;
217 	struct bio *bio;
218 
219 	page = alloc_page(GFP_KERNEL);
220 	if (unlikely(!page))
221 		return -ENOBUFS;
222 
223 	ClearPageUptodate(page);
224 	ClearPageDirty(page);
225 	lock_page(page);
226 
227 	bio = bio_alloc(GFP_KERNEL, 1);
228 	if (unlikely(!bio)) {
229 		__free_page(page);
230 		return -ENOBUFS;
231 	}
232 
233 	bio->bi_sector = sector * (sb->s_blocksize >> 9);
234 	bio->bi_bdev = sb->s_bdev;
235 	bio_add_page(bio, page, PAGE_SIZE, 0);
236 
237 	bio->bi_end_io = end_bio_io_page;
238 	bio->bi_private = page;
239 	submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
240 	wait_on_page_locked(page);
241 	bio_put(bio);
242 	if (!PageUptodate(page)) {
243 		__free_page(page);
244 		return -EIO;
245 	}
246 	p = kmap(page);
247 	gfs2_sb_in(&sdp->sd_sb, p);
248 	kunmap(page);
249 	__free_page(page);
250 	return 0;
251 }
252 
253 /**
254  * gfs2_read_sb - Read super block
255  * @sdp: The GFS2 superblock
256  * @gl: the glock for the superblock (assumed to be held)
257  * @silent: Don't print message if mount fails
258  *
259  */
260 
261 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
262 {
263 	u32 hash_blocks, ind_blocks, leaf_blocks;
264 	u32 tmp_blocks;
265 	unsigned int x;
266 	int error;
267 
268 	error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
269 	if (error) {
270 		if (!silent)
271 			fs_err(sdp, "can't read superblock\n");
272 		return error;
273 	}
274 
275 	error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
276 	if (error)
277 		return error;
278 
279 	sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
280 			       GFS2_BASIC_BLOCK_SHIFT;
281 	sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
282 	sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
283 			  sizeof(struct gfs2_dinode)) / sizeof(u64);
284 	sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
285 			  sizeof(struct gfs2_meta_header)) / sizeof(u64);
286 	sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
287 	sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
288 	sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
289 	sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
290 	sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
291 				sizeof(struct gfs2_meta_header)) /
292 			        sizeof(struct gfs2_quota_change);
293 
294 	/* Compute maximum reservation required to add a entry to a directory */
295 
296 	hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
297 			     sdp->sd_jbsize);
298 
299 	ind_blocks = 0;
300 	for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
301 		tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
302 		ind_blocks += tmp_blocks;
303 	}
304 
305 	leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
306 
307 	sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
308 
309 	sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
310 				sizeof(struct gfs2_dinode);
311 	sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
312 	for (x = 2;; x++) {
313 		u64 space, d;
314 		u32 m;
315 
316 		space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
317 		d = space;
318 		m = do_div(d, sdp->sd_inptrs);
319 
320 		if (d != sdp->sd_heightsize[x - 1] || m)
321 			break;
322 		sdp->sd_heightsize[x] = space;
323 	}
324 	sdp->sd_max_height = x;
325 	gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
326 
327 	sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
328 				 sizeof(struct gfs2_dinode);
329 	sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
330 	for (x = 2;; x++) {
331 		u64 space, d;
332 		u32 m;
333 
334 		space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
335 		d = space;
336 		m = do_div(d, sdp->sd_inptrs);
337 
338 		if (d != sdp->sd_jheightsize[x - 1] || m)
339 			break;
340 		sdp->sd_jheightsize[x] = space;
341 	}
342 	sdp->sd_max_jheight = x;
343 	gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
344 
345 	return 0;
346 }
347 
348 /**
349  * gfs2_jindex_hold - Grab a lock on the jindex
350  * @sdp: The GFS2 superblock
351  * @ji_gh: the holder for the jindex glock
352  *
353  * This is very similar to the gfs2_rindex_hold() function, except that
354  * in general we hold the jindex lock for longer periods of time and
355  * we grab it far less frequently (in general) then the rgrp lock.
356  *
357  * Returns: errno
358  */
359 
360 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
361 {
362 	struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
363 	struct qstr name;
364 	char buf[20];
365 	struct gfs2_jdesc *jd;
366 	int error;
367 
368 	name.name = buf;
369 
370 	mutex_lock(&sdp->sd_jindex_mutex);
371 
372 	for (;;) {
373 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
374 		if (error)
375 			break;
376 
377 		name.len = sprintf(buf, "journal%u", sdp->sd_journals);
378 		name.hash = gfs2_disk_hash(name.name, name.len);
379 
380 		error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
381 		if (error == -ENOENT) {
382 			error = 0;
383 			break;
384 		}
385 
386 		gfs2_glock_dq_uninit(ji_gh);
387 
388 		if (error)
389 			break;
390 
391 		error = -ENOMEM;
392 		jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
393 		if (!jd)
394 			break;
395 
396 		jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
397 		if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
398 			if (!jd->jd_inode)
399 				error = -ENOENT;
400 			else
401 				error = PTR_ERR(jd->jd_inode);
402 			kfree(jd);
403 			break;
404 		}
405 
406 		spin_lock(&sdp->sd_jindex_spin);
407 		jd->jd_jid = sdp->sd_journals++;
408 		list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
409 		spin_unlock(&sdp->sd_jindex_spin);
410 	}
411 
412 	mutex_unlock(&sdp->sd_jindex_mutex);
413 
414 	return error;
415 }
416 
417 /**
418  * gfs2_jindex_free - Clear all the journal index information
419  * @sdp: The GFS2 superblock
420  *
421  */
422 
423 void gfs2_jindex_free(struct gfs2_sbd *sdp)
424 {
425 	struct list_head list;
426 	struct gfs2_jdesc *jd;
427 
428 	spin_lock(&sdp->sd_jindex_spin);
429 	list_add(&list, &sdp->sd_jindex_list);
430 	list_del_init(&sdp->sd_jindex_list);
431 	sdp->sd_journals = 0;
432 	spin_unlock(&sdp->sd_jindex_spin);
433 
434 	while (!list_empty(&list)) {
435 		jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
436 		list_del(&jd->jd_list);
437 		iput(jd->jd_inode);
438 		kfree(jd);
439 	}
440 }
441 
442 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
443 {
444 	struct gfs2_jdesc *jd;
445 	int found = 0;
446 
447 	list_for_each_entry(jd, head, jd_list) {
448 		if (jd->jd_jid == jid) {
449 			found = 1;
450 			break;
451 		}
452 	}
453 
454 	if (!found)
455 		jd = NULL;
456 
457 	return jd;
458 }
459 
460 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
461 {
462 	struct gfs2_jdesc *jd;
463 
464 	spin_lock(&sdp->sd_jindex_spin);
465 	jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
466 	spin_unlock(&sdp->sd_jindex_spin);
467 
468 	return jd;
469 }
470 
471 void gfs2_jdesc_make_dirty(struct gfs2_sbd *sdp, unsigned int jid)
472 {
473 	struct gfs2_jdesc *jd;
474 
475 	spin_lock(&sdp->sd_jindex_spin);
476 	jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
477 	if (jd)
478 		jd->jd_dirty = 1;
479 	spin_unlock(&sdp->sd_jindex_spin);
480 }
481 
482 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
483 {
484 	struct gfs2_jdesc *jd;
485 	int found = 0;
486 
487 	spin_lock(&sdp->sd_jindex_spin);
488 
489 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
490 		if (jd->jd_dirty) {
491 			jd->jd_dirty = 0;
492 			found = 1;
493 			break;
494 		}
495 	}
496 	spin_unlock(&sdp->sd_jindex_spin);
497 
498 	if (!found)
499 		jd = NULL;
500 
501 	return jd;
502 }
503 
504 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
505 {
506 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
507 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
508 	int ar;
509 	int error;
510 
511 	if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
512 	    (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
513 		gfs2_consist_inode(ip);
514 		return -EIO;
515 	}
516 	jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
517 
518 	error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
519 	if (!error && ar) {
520 		gfs2_consist_inode(ip);
521 		error = -EIO;
522 	}
523 
524 	return error;
525 }
526 
527 /**
528  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
529  * @sdp: the filesystem
530  *
531  * Returns: errno
532  */
533 
534 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
535 {
536 	struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
537 	struct gfs2_glock *j_gl = ip->i_gl;
538 	struct gfs2_holder t_gh;
539 	struct gfs2_log_header_host head;
540 	int error;
541 
542 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
543 	if (error)
544 		return error;
545 
546 	gfs2_meta_cache_flush(ip);
547 	j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
548 
549 	error = gfs2_find_jhead(sdp->sd_jdesc, &head);
550 	if (error)
551 		goto fail;
552 
553 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
554 		gfs2_consist(sdp);
555 		error = -EIO;
556 		goto fail;
557 	}
558 
559 	/*  Initialize some head of the log stuff  */
560 	sdp->sd_log_sequence = head.lh_sequence + 1;
561 	gfs2_log_pointers_init(sdp, head.lh_blkno);
562 
563 	error = gfs2_quota_init(sdp);
564 	if (error)
565 		goto fail;
566 
567 	set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
568 
569 	gfs2_glock_dq_uninit(&t_gh);
570 
571 	return 0;
572 
573 fail:
574 	t_gh.gh_flags |= GL_NOCACHE;
575 	gfs2_glock_dq_uninit(&t_gh);
576 
577 	return error;
578 }
579 
580 /**
581  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
582  * @sdp: the filesystem
583  *
584  * Returns: errno
585  */
586 
587 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
588 {
589 	struct gfs2_holder t_gh;
590 	int error;
591 
592 	gfs2_quota_sync(sdp);
593 	gfs2_statfs_sync(sdp);
594 
595 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
596 				   &t_gh);
597 	if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
598 		return error;
599 
600 	gfs2_meta_syncfs(sdp);
601 	gfs2_log_shutdown(sdp);
602 
603 	clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
604 
605 	if (t_gh.gh_gl)
606 		gfs2_glock_dq_uninit(&t_gh);
607 
608 	gfs2_quota_cleanup(sdp);
609 
610 	return error;
611 }
612 
613 static void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
614 {
615 	const struct gfs2_statfs_change *str = buf;
616 
617 	sc->sc_total = be64_to_cpu(str->sc_total);
618 	sc->sc_free = be64_to_cpu(str->sc_free);
619 	sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
620 }
621 
622 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
623 {
624 	struct gfs2_statfs_change *str = buf;
625 
626 	str->sc_total = cpu_to_be64(sc->sc_total);
627 	str->sc_free = cpu_to_be64(sc->sc_free);
628 	str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
629 }
630 
631 int gfs2_statfs_init(struct gfs2_sbd *sdp)
632 {
633 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
634 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
635 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
636 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
637 	struct buffer_head *m_bh, *l_bh;
638 	struct gfs2_holder gh;
639 	int error;
640 
641 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
642 				   &gh);
643 	if (error)
644 		return error;
645 
646 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
647 	if (error)
648 		goto out;
649 
650 	if (sdp->sd_args.ar_spectator) {
651 		spin_lock(&sdp->sd_statfs_spin);
652 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
653 				      sizeof(struct gfs2_dinode));
654 		spin_unlock(&sdp->sd_statfs_spin);
655 	} else {
656 		error = gfs2_meta_inode_buffer(l_ip, &l_bh);
657 		if (error)
658 			goto out_m_bh;
659 
660 		spin_lock(&sdp->sd_statfs_spin);
661 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
662 				      sizeof(struct gfs2_dinode));
663 		gfs2_statfs_change_in(l_sc, l_bh->b_data +
664 				      sizeof(struct gfs2_dinode));
665 		spin_unlock(&sdp->sd_statfs_spin);
666 
667 		brelse(l_bh);
668 	}
669 
670 out_m_bh:
671 	brelse(m_bh);
672 out:
673 	gfs2_glock_dq_uninit(&gh);
674 	return 0;
675 }
676 
677 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
678 			s64 dinodes)
679 {
680 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
681 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
682 	struct buffer_head *l_bh;
683 	int error;
684 
685 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
686 	if (error)
687 		return;
688 
689 	mutex_lock(&sdp->sd_statfs_mutex);
690 	gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
691 	mutex_unlock(&sdp->sd_statfs_mutex);
692 
693 	spin_lock(&sdp->sd_statfs_spin);
694 	l_sc->sc_total += total;
695 	l_sc->sc_free += free;
696 	l_sc->sc_dinodes += dinodes;
697 	gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
698 	spin_unlock(&sdp->sd_statfs_spin);
699 
700 	brelse(l_bh);
701 }
702 
703 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
704 {
705 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
706 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
707 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
708 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
709 	struct gfs2_holder gh;
710 	struct buffer_head *m_bh, *l_bh;
711 	int error;
712 
713 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
714 				   &gh);
715 	if (error)
716 		return error;
717 
718 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
719 	if (error)
720 		goto out;
721 
722 	spin_lock(&sdp->sd_statfs_spin);
723 	gfs2_statfs_change_in(m_sc, m_bh->b_data +
724 			      sizeof(struct gfs2_dinode));
725 	if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
726 		spin_unlock(&sdp->sd_statfs_spin);
727 		goto out_bh;
728 	}
729 	spin_unlock(&sdp->sd_statfs_spin);
730 
731 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
732 	if (error)
733 		goto out_bh;
734 
735 	error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
736 	if (error)
737 		goto out_bh2;
738 
739 	mutex_lock(&sdp->sd_statfs_mutex);
740 	gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
741 	mutex_unlock(&sdp->sd_statfs_mutex);
742 
743 	spin_lock(&sdp->sd_statfs_spin);
744 	m_sc->sc_total += l_sc->sc_total;
745 	m_sc->sc_free += l_sc->sc_free;
746 	m_sc->sc_dinodes += l_sc->sc_dinodes;
747 	memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
748 	memset(l_bh->b_data + sizeof(struct gfs2_dinode),
749 	       0, sizeof(struct gfs2_statfs_change));
750 	spin_unlock(&sdp->sd_statfs_spin);
751 
752 	gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
753 	gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
754 
755 	gfs2_trans_end(sdp);
756 
757 out_bh2:
758 	brelse(l_bh);
759 out_bh:
760 	brelse(m_bh);
761 out:
762 	gfs2_glock_dq_uninit(&gh);
763 	return error;
764 }
765 
766 /**
767  * gfs2_statfs_i - Do a statfs
768  * @sdp: the filesystem
769  * @sg: the sg structure
770  *
771  * Returns: errno
772  */
773 
774 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
775 {
776 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
777 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
778 
779 	spin_lock(&sdp->sd_statfs_spin);
780 
781 	*sc = *m_sc;
782 	sc->sc_total += l_sc->sc_total;
783 	sc->sc_free += l_sc->sc_free;
784 	sc->sc_dinodes += l_sc->sc_dinodes;
785 
786 	spin_unlock(&sdp->sd_statfs_spin);
787 
788 	if (sc->sc_free < 0)
789 		sc->sc_free = 0;
790 	if (sc->sc_free > sc->sc_total)
791 		sc->sc_free = sc->sc_total;
792 	if (sc->sc_dinodes < 0)
793 		sc->sc_dinodes = 0;
794 
795 	return 0;
796 }
797 
798 /**
799  * statfs_fill - fill in the sg for a given RG
800  * @rgd: the RG
801  * @sc: the sc structure
802  *
803  * Returns: 0 on success, -ESTALE if the LVB is invalid
804  */
805 
806 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
807 			    struct gfs2_statfs_change_host *sc)
808 {
809 	gfs2_rgrp_verify(rgd);
810 	sc->sc_total += rgd->rd_data;
811 	sc->sc_free += rgd->rd_rg.rg_free;
812 	sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
813 	return 0;
814 }
815 
816 /**
817  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
818  * @sdp: the filesystem
819  * @sc: the sc info that will be returned
820  *
821  * Any error (other than a signal) will cause this routine to fall back
822  * to the synchronous version.
823  *
824  * FIXME: This really shouldn't busy wait like this.
825  *
826  * Returns: errno
827  */
828 
829 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
830 {
831 	struct gfs2_holder ri_gh;
832 	struct gfs2_rgrpd *rgd_next;
833 	struct gfs2_holder *gha, *gh;
834 	unsigned int slots = 64;
835 	unsigned int x;
836 	int done;
837 	int error = 0, err;
838 
839 	memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
840 	gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
841 	if (!gha)
842 		return -ENOMEM;
843 
844 	error = gfs2_rindex_hold(sdp, &ri_gh);
845 	if (error)
846 		goto out;
847 
848 	rgd_next = gfs2_rgrpd_get_first(sdp);
849 
850 	for (;;) {
851 		done = 1;
852 
853 		for (x = 0; x < slots; x++) {
854 			gh = gha + x;
855 
856 			if (gh->gh_gl && gfs2_glock_poll(gh)) {
857 				err = gfs2_glock_wait(gh);
858 				if (err) {
859 					gfs2_holder_uninit(gh);
860 					error = err;
861 				} else {
862 					if (!error)
863 						error = statfs_slow_fill(
864 							gh->gh_gl->gl_object, sc);
865 					gfs2_glock_dq_uninit(gh);
866 				}
867 			}
868 
869 			if (gh->gh_gl)
870 				done = 0;
871 			else if (rgd_next && !error) {
872 				error = gfs2_glock_nq_init(rgd_next->rd_gl,
873 							   LM_ST_SHARED,
874 							   GL_ASYNC,
875 							   gh);
876 				rgd_next = gfs2_rgrpd_get_next(rgd_next);
877 				done = 0;
878 			}
879 
880 			if (signal_pending(current))
881 				error = -ERESTARTSYS;
882 		}
883 
884 		if (done)
885 			break;
886 
887 		yield();
888 	}
889 
890 	gfs2_glock_dq_uninit(&ri_gh);
891 
892 out:
893 	kfree(gha);
894 	return error;
895 }
896 
897 struct lfcc {
898 	struct list_head list;
899 	struct gfs2_holder gh;
900 };
901 
902 /**
903  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
904  *                            journals are clean
905  * @sdp: the file system
906  * @state: the state to put the transaction lock into
907  * @t_gh: the hold on the transaction lock
908  *
909  * Returns: errno
910  */
911 
912 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
913 				    struct gfs2_holder *t_gh)
914 {
915 	struct gfs2_inode *ip;
916 	struct gfs2_holder ji_gh;
917 	struct gfs2_jdesc *jd;
918 	struct lfcc *lfcc;
919 	LIST_HEAD(list);
920 	struct gfs2_log_header_host lh;
921 	int error;
922 
923 	error = gfs2_jindex_hold(sdp, &ji_gh);
924 	if (error)
925 		return error;
926 
927 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
928 		lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
929 		if (!lfcc) {
930 			error = -ENOMEM;
931 			goto out;
932 		}
933 		ip = GFS2_I(jd->jd_inode);
934 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
935 		if (error) {
936 			kfree(lfcc);
937 			goto out;
938 		}
939 		list_add(&lfcc->list, &list);
940 	}
941 
942 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
943 			       LM_FLAG_PRIORITY | GL_NOCACHE,
944 			       t_gh);
945 
946 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
947 		error = gfs2_jdesc_check(jd);
948 		if (error)
949 			break;
950 		error = gfs2_find_jhead(jd, &lh);
951 		if (error)
952 			break;
953 		if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
954 			error = -EBUSY;
955 			break;
956 		}
957 	}
958 
959 	if (error)
960 		gfs2_glock_dq_uninit(t_gh);
961 
962 out:
963 	while (!list_empty(&list)) {
964 		lfcc = list_entry(list.next, struct lfcc, list);
965 		list_del(&lfcc->list);
966 		gfs2_glock_dq_uninit(&lfcc->gh);
967 		kfree(lfcc);
968 	}
969 	gfs2_glock_dq_uninit(&ji_gh);
970 	return error;
971 }
972 
973 /**
974  * gfs2_freeze_fs - freezes the file system
975  * @sdp: the file system
976  *
977  * This function flushes data and meta data for all machines by
978  * aquiring the transaction log exclusively.  All journals are
979  * ensured to be in a clean state as well.
980  *
981  * Returns: errno
982  */
983 
984 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
985 {
986 	int error = 0;
987 
988 	mutex_lock(&sdp->sd_freeze_lock);
989 
990 	if (!sdp->sd_freeze_count++) {
991 		error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
992 		if (error)
993 			sdp->sd_freeze_count--;
994 	}
995 
996 	mutex_unlock(&sdp->sd_freeze_lock);
997 
998 	return error;
999 }
1000 
1001 /**
1002  * gfs2_unfreeze_fs - unfreezes the file system
1003  * @sdp: the file system
1004  *
1005  * This function allows the file system to proceed by unlocking
1006  * the exclusively held transaction lock.  Other GFS2 nodes are
1007  * now free to acquire the lock shared and go on with their lives.
1008  *
1009  */
1010 
1011 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
1012 {
1013 	mutex_lock(&sdp->sd_freeze_lock);
1014 
1015 	if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
1016 		gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1017 
1018 	mutex_unlock(&sdp->sd_freeze_lock);
1019 }
1020 
1021