xref: /linux/fs/gfs2/super.c (revision c537b994505099b7197e7d3125b942ecbcc51eb6)
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_scand_secs = 15;
62 	gt->gt_recoverd_secs = 60;
63 	gt->gt_logd_secs = 1;
64 	gt->gt_quotad_secs = 5;
65 	gt->gt_quota_simul_sync = 64;
66 	gt->gt_quota_warn_period = 10;
67 	gt->gt_quota_scale_num = 1;
68 	gt->gt_quota_scale_den = 1;
69 	gt->gt_quota_cache_secs = 300;
70 	gt->gt_quota_quantum = 60;
71 	gt->gt_atime_quantum = 3600;
72 	gt->gt_new_files_jdata = 0;
73 	gt->gt_new_files_directio = 0;
74 	gt->gt_max_readahead = 1 << 18;
75 	gt->gt_lockdump_size = 131072;
76 	gt->gt_stall_secs = 600;
77 	gt->gt_complain_secs = 10;
78 	gt->gt_reclaim_limit = 5000;
79 	gt->gt_statfs_quantum = 30;
80 	gt->gt_statfs_slow = 0;
81 }
82 
83 /**
84  * gfs2_check_sb - Check superblock
85  * @sdp: the filesystem
86  * @sb: The superblock
87  * @silent: Don't print a message if the check fails
88  *
89  * Checks the version code of the FS is one that we understand how to
90  * read and that the sizes of the various on-disk structures have not
91  * changed.
92  */
93 
94 int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
95 {
96 	unsigned int x;
97 
98 	if (sb->sb_header.mh_magic != GFS2_MAGIC ||
99 	    sb->sb_header.mh_type != GFS2_METATYPE_SB) {
100 		if (!silent)
101 			printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
102 		return -EINVAL;
103 	}
104 
105 	/*  If format numbers match exactly, we're done.  */
106 
107 	if (sb->sb_fs_format == GFS2_FORMAT_FS &&
108 	    sb->sb_multihost_format == GFS2_FORMAT_MULTI)
109 		return 0;
110 
111 	if (sb->sb_fs_format != GFS2_FORMAT_FS) {
112 		for (x = 0; gfs2_old_fs_formats[x]; x++)
113 			if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
114 				break;
115 
116 		if (!gfs2_old_fs_formats[x]) {
117 			printk(KERN_WARNING
118 			       "GFS2: code version (%u, %u) is incompatible "
119 			       "with ondisk format (%u, %u)\n",
120 			       GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
121 			       sb->sb_fs_format, sb->sb_multihost_format);
122 			printk(KERN_WARNING
123 			       "GFS2: I don't know how to upgrade this FS\n");
124 			return -EINVAL;
125 		}
126 	}
127 
128 	if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
129 		for (x = 0; gfs2_old_multihost_formats[x]; x++)
130 			if (gfs2_old_multihost_formats[x] ==
131 			    sb->sb_multihost_format)
132 				break;
133 
134 		if (!gfs2_old_multihost_formats[x]) {
135 			printk(KERN_WARNING
136 			       "GFS2: code version (%u, %u) is incompatible "
137 			       "with ondisk format (%u, %u)\n",
138 			       GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
139 			       sb->sb_fs_format, sb->sb_multihost_format);
140 			printk(KERN_WARNING
141 			       "GFS2: I don't know how to upgrade this FS\n");
142 			return -EINVAL;
143 		}
144 	}
145 
146 	if (!sdp->sd_args.ar_upgrade) {
147 		printk(KERN_WARNING
148 		       "GFS2: code version (%u, %u) is incompatible "
149 		       "with ondisk format (%u, %u)\n",
150 		       GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
151 		       sb->sb_fs_format, sb->sb_multihost_format);
152 		printk(KERN_INFO
153 		       "GFS2: Use the \"upgrade\" mount option to upgrade "
154 		       "the FS\n");
155 		printk(KERN_INFO "GFS2: See the manual for more details\n");
156 		return -EINVAL;
157 	}
158 
159 	return 0;
160 }
161 
162 
163 static int end_bio_io_page(struct bio *bio, unsigned int bytes_done, int error)
164 {
165 	struct page *page = bio->bi_private;
166 	if (bio->bi_size)
167 		return 1;
168 
169 	if (!error)
170 		SetPageUptodate(page);
171 	else
172 		printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
173 	unlock_page(page);
174 	return 0;
175 }
176 
177 /**
178  * gfs2_read_super - Read the gfs2 super block from disk
179  * @sb: The VFS super block
180  * @sector: The location of the super block
181  *
182  * This uses the bio functions to read the super block from disk
183  * because we want to be 100% sure that we never read cached data.
184  * A super block is read twice only during each GFS2 mount and is
185  * never written to by the filesystem. The first time its read no
186  * locks are held, and the only details which are looked at are those
187  * relating to the locking protocol. Once locking is up and working,
188  * the sb is read again under the lock to establish the location of
189  * the master directory (contains pointers to journals etc) and the
190  * root directory.
191  *
192  * Returns: A page containing the sb or NULL
193  */
194 
195 struct page *gfs2_read_super(struct super_block *sb, sector_t sector)
196 {
197 	struct page *page;
198 	struct bio *bio;
199 
200 	page = alloc_page(GFP_KERNEL);
201 	if (unlikely(!page))
202 		return NULL;
203 
204 	ClearPageUptodate(page);
205 	ClearPageDirty(page);
206 	lock_page(page);
207 
208 	bio = bio_alloc(GFP_KERNEL, 1);
209 	if (unlikely(!bio)) {
210 		__free_page(page);
211 		return NULL;
212 	}
213 
214 	bio->bi_sector = sector * (sb->s_blocksize >> 9);
215 	bio->bi_bdev = sb->s_bdev;
216 	bio_add_page(bio, page, PAGE_SIZE, 0);
217 
218 	bio->bi_end_io = end_bio_io_page;
219 	bio->bi_private = page;
220 	submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
221 	wait_on_page_locked(page);
222 	bio_put(bio);
223 	if (!PageUptodate(page)) {
224 		__free_page(page);
225 		return NULL;
226 	}
227 	return page;
228 }
229 
230 /**
231  * gfs2_read_sb - Read super block
232  * @sdp: The GFS2 superblock
233  * @gl: the glock for the superblock (assumed to be held)
234  * @silent: Don't print message if mount fails
235  *
236  */
237 
238 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
239 {
240 	u32 hash_blocks, ind_blocks, leaf_blocks;
241 	u32 tmp_blocks;
242 	unsigned int x;
243 	int error;
244 	struct page *page;
245 	char *sb;
246 
247 	page = gfs2_read_super(sdp->sd_vfs, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
248 	if (!page) {
249 		if (!silent)
250 			fs_err(sdp, "can't read superblock\n");
251 		return -EIO;
252 	}
253 	sb = kmap(page);
254 	gfs2_sb_in(&sdp->sd_sb, sb);
255 	kunmap(page);
256 	__free_page(page);
257 
258 	error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
259 	if (error)
260 		return error;
261 
262 	sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
263 			       GFS2_BASIC_BLOCK_SHIFT;
264 	sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
265 	sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
266 			  sizeof(struct gfs2_dinode)) / sizeof(u64);
267 	sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
268 			  sizeof(struct gfs2_meta_header)) / sizeof(u64);
269 	sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
270 	sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
271 	sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
272 	sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
273 	sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
274 				sizeof(struct gfs2_meta_header)) /
275 			        sizeof(struct gfs2_quota_change);
276 
277 	/* Compute maximum reservation required to add a entry to a directory */
278 
279 	hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
280 			     sdp->sd_jbsize);
281 
282 	ind_blocks = 0;
283 	for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
284 		tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
285 		ind_blocks += tmp_blocks;
286 	}
287 
288 	leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
289 
290 	sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
291 
292 	sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
293 				sizeof(struct gfs2_dinode);
294 	sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
295 	for (x = 2;; x++) {
296 		u64 space, d;
297 		u32 m;
298 
299 		space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
300 		d = space;
301 		m = do_div(d, sdp->sd_inptrs);
302 
303 		if (d != sdp->sd_heightsize[x - 1] || m)
304 			break;
305 		sdp->sd_heightsize[x] = space;
306 	}
307 	sdp->sd_max_height = x;
308 	gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
309 
310 	sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
311 				 sizeof(struct gfs2_dinode);
312 	sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
313 	for (x = 2;; x++) {
314 		u64 space, d;
315 		u32 m;
316 
317 		space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
318 		d = space;
319 		m = do_div(d, sdp->sd_inptrs);
320 
321 		if (d != sdp->sd_jheightsize[x - 1] || m)
322 			break;
323 		sdp->sd_jheightsize[x] = space;
324 	}
325 	sdp->sd_max_jheight = x;
326 	gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
327 
328 	return 0;
329 }
330 
331 /**
332  * gfs2_jindex_hold - Grab a lock on the jindex
333  * @sdp: The GFS2 superblock
334  * @ji_gh: the holder for the jindex glock
335  *
336  * This is very similar to the gfs2_rindex_hold() function, except that
337  * in general we hold the jindex lock for longer periods of time and
338  * we grab it far less frequently (in general) then the rgrp lock.
339  *
340  * Returns: errno
341  */
342 
343 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
344 {
345 	struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
346 	struct qstr name;
347 	char buf[20];
348 	struct gfs2_jdesc *jd;
349 	int error;
350 
351 	name.name = buf;
352 
353 	mutex_lock(&sdp->sd_jindex_mutex);
354 
355 	for (;;) {
356 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
357 		if (error)
358 			break;
359 
360 		name.len = sprintf(buf, "journal%u", sdp->sd_journals);
361 		name.hash = gfs2_disk_hash(name.name, name.len);
362 
363 		error = gfs2_dir_search(sdp->sd_jindex, &name, NULL, NULL);
364 		if (error == -ENOENT) {
365 			error = 0;
366 			break;
367 		}
368 
369 		gfs2_glock_dq_uninit(ji_gh);
370 
371 		if (error)
372 			break;
373 
374 		error = -ENOMEM;
375 		jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
376 		if (!jd)
377 			break;
378 
379 		jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
380 		if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
381 			if (!jd->jd_inode)
382 				error = -ENOENT;
383 			else
384 				error = PTR_ERR(jd->jd_inode);
385 			kfree(jd);
386 			break;
387 		}
388 
389 		spin_lock(&sdp->sd_jindex_spin);
390 		jd->jd_jid = sdp->sd_journals++;
391 		list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
392 		spin_unlock(&sdp->sd_jindex_spin);
393 	}
394 
395 	mutex_unlock(&sdp->sd_jindex_mutex);
396 
397 	return error;
398 }
399 
400 /**
401  * gfs2_jindex_free - Clear all the journal index information
402  * @sdp: The GFS2 superblock
403  *
404  */
405 
406 void gfs2_jindex_free(struct gfs2_sbd *sdp)
407 {
408 	struct list_head list;
409 	struct gfs2_jdesc *jd;
410 
411 	spin_lock(&sdp->sd_jindex_spin);
412 	list_add(&list, &sdp->sd_jindex_list);
413 	list_del_init(&sdp->sd_jindex_list);
414 	sdp->sd_journals = 0;
415 	spin_unlock(&sdp->sd_jindex_spin);
416 
417 	while (!list_empty(&list)) {
418 		jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
419 		list_del(&jd->jd_list);
420 		iput(jd->jd_inode);
421 		kfree(jd);
422 	}
423 }
424 
425 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
426 {
427 	struct gfs2_jdesc *jd;
428 	int found = 0;
429 
430 	list_for_each_entry(jd, head, jd_list) {
431 		if (jd->jd_jid == jid) {
432 			found = 1;
433 			break;
434 		}
435 	}
436 
437 	if (!found)
438 		jd = NULL;
439 
440 	return jd;
441 }
442 
443 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
444 {
445 	struct gfs2_jdesc *jd;
446 
447 	spin_lock(&sdp->sd_jindex_spin);
448 	jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
449 	spin_unlock(&sdp->sd_jindex_spin);
450 
451 	return jd;
452 }
453 
454 void gfs2_jdesc_make_dirty(struct gfs2_sbd *sdp, unsigned int jid)
455 {
456 	struct gfs2_jdesc *jd;
457 
458 	spin_lock(&sdp->sd_jindex_spin);
459 	jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
460 	if (jd)
461 		jd->jd_dirty = 1;
462 	spin_unlock(&sdp->sd_jindex_spin);
463 }
464 
465 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
466 {
467 	struct gfs2_jdesc *jd;
468 	int found = 0;
469 
470 	spin_lock(&sdp->sd_jindex_spin);
471 
472 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
473 		if (jd->jd_dirty) {
474 			jd->jd_dirty = 0;
475 			found = 1;
476 			break;
477 		}
478 	}
479 	spin_unlock(&sdp->sd_jindex_spin);
480 
481 	if (!found)
482 		jd = NULL;
483 
484 	return jd;
485 }
486 
487 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
488 {
489 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
490 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
491 	int ar;
492 	int error;
493 
494 	if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
495 	    (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
496 		gfs2_consist_inode(ip);
497 		return -EIO;
498 	}
499 	jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
500 
501 	error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
502 	if (!error && ar) {
503 		gfs2_consist_inode(ip);
504 		error = -EIO;
505 	}
506 
507 	return error;
508 }
509 
510 /**
511  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
512  * @sdp: the filesystem
513  *
514  * Returns: errno
515  */
516 
517 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
518 {
519 	struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
520 	struct gfs2_glock *j_gl = ip->i_gl;
521 	struct gfs2_holder t_gh;
522 	struct gfs2_log_header_host head;
523 	int error;
524 
525 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
526 	if (error)
527 		return error;
528 
529 	gfs2_meta_cache_flush(ip);
530 	j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
531 
532 	error = gfs2_find_jhead(sdp->sd_jdesc, &head);
533 	if (error)
534 		goto fail;
535 
536 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
537 		gfs2_consist(sdp);
538 		error = -EIO;
539 		goto fail;
540 	}
541 
542 	/*  Initialize some head of the log stuff  */
543 	sdp->sd_log_sequence = head.lh_sequence + 1;
544 	gfs2_log_pointers_init(sdp, head.lh_blkno);
545 
546 	error = gfs2_quota_init(sdp);
547 	if (error)
548 		goto fail;
549 
550 	set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
551 
552 	gfs2_glock_dq_uninit(&t_gh);
553 
554 	return 0;
555 
556 fail:
557 	t_gh.gh_flags |= GL_NOCACHE;
558 	gfs2_glock_dq_uninit(&t_gh);
559 
560 	return error;
561 }
562 
563 /**
564  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
565  * @sdp: the filesystem
566  *
567  * Returns: errno
568  */
569 
570 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
571 {
572 	struct gfs2_holder t_gh;
573 	int error;
574 
575 	gfs2_quota_sync(sdp);
576 	gfs2_statfs_sync(sdp);
577 
578 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
579 				   &t_gh);
580 	if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
581 		return error;
582 
583 	gfs2_meta_syncfs(sdp);
584 	gfs2_log_shutdown(sdp);
585 
586 	clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
587 
588 	if (t_gh.gh_gl)
589 		gfs2_glock_dq_uninit(&t_gh);
590 
591 	gfs2_quota_cleanup(sdp);
592 
593 	return error;
594 }
595 
596 int gfs2_statfs_init(struct gfs2_sbd *sdp)
597 {
598 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
599 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
600 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
601 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
602 	struct buffer_head *m_bh, *l_bh;
603 	struct gfs2_holder gh;
604 	int error;
605 
606 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
607 				   &gh);
608 	if (error)
609 		return error;
610 
611 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
612 	if (error)
613 		goto out;
614 
615 	if (sdp->sd_args.ar_spectator) {
616 		spin_lock(&sdp->sd_statfs_spin);
617 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
618 				      sizeof(struct gfs2_dinode));
619 		spin_unlock(&sdp->sd_statfs_spin);
620 	} else {
621 		error = gfs2_meta_inode_buffer(l_ip, &l_bh);
622 		if (error)
623 			goto out_m_bh;
624 
625 		spin_lock(&sdp->sd_statfs_spin);
626 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
627 				      sizeof(struct gfs2_dinode));
628 		gfs2_statfs_change_in(l_sc, l_bh->b_data +
629 				      sizeof(struct gfs2_dinode));
630 		spin_unlock(&sdp->sd_statfs_spin);
631 
632 		brelse(l_bh);
633 	}
634 
635 out_m_bh:
636 	brelse(m_bh);
637 out:
638 	gfs2_glock_dq_uninit(&gh);
639 	return 0;
640 }
641 
642 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
643 			s64 dinodes)
644 {
645 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
646 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
647 	struct buffer_head *l_bh;
648 	int error;
649 
650 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
651 	if (error)
652 		return;
653 
654 	mutex_lock(&sdp->sd_statfs_mutex);
655 	gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
656 	mutex_unlock(&sdp->sd_statfs_mutex);
657 
658 	spin_lock(&sdp->sd_statfs_spin);
659 	l_sc->sc_total += total;
660 	l_sc->sc_free += free;
661 	l_sc->sc_dinodes += dinodes;
662 	gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
663 	spin_unlock(&sdp->sd_statfs_spin);
664 
665 	brelse(l_bh);
666 }
667 
668 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
669 {
670 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
671 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
672 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
673 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
674 	struct gfs2_holder gh;
675 	struct buffer_head *m_bh, *l_bh;
676 	int error;
677 
678 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
679 				   &gh);
680 	if (error)
681 		return error;
682 
683 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
684 	if (error)
685 		goto out;
686 
687 	spin_lock(&sdp->sd_statfs_spin);
688 	gfs2_statfs_change_in(m_sc, m_bh->b_data +
689 			      sizeof(struct gfs2_dinode));
690 	if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
691 		spin_unlock(&sdp->sd_statfs_spin);
692 		goto out_bh;
693 	}
694 	spin_unlock(&sdp->sd_statfs_spin);
695 
696 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
697 	if (error)
698 		goto out_bh;
699 
700 	error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
701 	if (error)
702 		goto out_bh2;
703 
704 	mutex_lock(&sdp->sd_statfs_mutex);
705 	gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
706 	mutex_unlock(&sdp->sd_statfs_mutex);
707 
708 	spin_lock(&sdp->sd_statfs_spin);
709 	m_sc->sc_total += l_sc->sc_total;
710 	m_sc->sc_free += l_sc->sc_free;
711 	m_sc->sc_dinodes += l_sc->sc_dinodes;
712 	memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
713 	memset(l_bh->b_data + sizeof(struct gfs2_dinode),
714 	       0, sizeof(struct gfs2_statfs_change));
715 	spin_unlock(&sdp->sd_statfs_spin);
716 
717 	gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
718 	gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
719 
720 	gfs2_trans_end(sdp);
721 
722 out_bh2:
723 	brelse(l_bh);
724 out_bh:
725 	brelse(m_bh);
726 out:
727 	gfs2_glock_dq_uninit(&gh);
728 	return error;
729 }
730 
731 /**
732  * gfs2_statfs_i - Do a statfs
733  * @sdp: the filesystem
734  * @sg: the sg structure
735  *
736  * Returns: errno
737  */
738 
739 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
740 {
741 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
742 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
743 
744 	spin_lock(&sdp->sd_statfs_spin);
745 
746 	*sc = *m_sc;
747 	sc->sc_total += l_sc->sc_total;
748 	sc->sc_free += l_sc->sc_free;
749 	sc->sc_dinodes += l_sc->sc_dinodes;
750 
751 	spin_unlock(&sdp->sd_statfs_spin);
752 
753 	if (sc->sc_free < 0)
754 		sc->sc_free = 0;
755 	if (sc->sc_free > sc->sc_total)
756 		sc->sc_free = sc->sc_total;
757 	if (sc->sc_dinodes < 0)
758 		sc->sc_dinodes = 0;
759 
760 	return 0;
761 }
762 
763 /**
764  * statfs_fill - fill in the sg for a given RG
765  * @rgd: the RG
766  * @sc: the sc structure
767  *
768  * Returns: 0 on success, -ESTALE if the LVB is invalid
769  */
770 
771 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
772 			    struct gfs2_statfs_change_host *sc)
773 {
774 	gfs2_rgrp_verify(rgd);
775 	sc->sc_total += rgd->rd_ri.ri_data;
776 	sc->sc_free += rgd->rd_rg.rg_free;
777 	sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
778 	return 0;
779 }
780 
781 /**
782  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
783  * @sdp: the filesystem
784  * @sc: the sc info that will be returned
785  *
786  * Any error (other than a signal) will cause this routine to fall back
787  * to the synchronous version.
788  *
789  * FIXME: This really shouldn't busy wait like this.
790  *
791  * Returns: errno
792  */
793 
794 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
795 {
796 	struct gfs2_holder ri_gh;
797 	struct gfs2_rgrpd *rgd_next;
798 	struct gfs2_holder *gha, *gh;
799 	unsigned int slots = 64;
800 	unsigned int x;
801 	int done;
802 	int error = 0, err;
803 
804 	memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
805 	gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
806 	if (!gha)
807 		return -ENOMEM;
808 
809 	error = gfs2_rindex_hold(sdp, &ri_gh);
810 	if (error)
811 		goto out;
812 
813 	rgd_next = gfs2_rgrpd_get_first(sdp);
814 
815 	for (;;) {
816 		done = 1;
817 
818 		for (x = 0; x < slots; x++) {
819 			gh = gha + x;
820 
821 			if (gh->gh_gl && gfs2_glock_poll(gh)) {
822 				err = gfs2_glock_wait(gh);
823 				if (err) {
824 					gfs2_holder_uninit(gh);
825 					error = err;
826 				} else {
827 					if (!error)
828 						error = statfs_slow_fill(
829 							gh->gh_gl->gl_object, sc);
830 					gfs2_glock_dq_uninit(gh);
831 				}
832 			}
833 
834 			if (gh->gh_gl)
835 				done = 0;
836 			else if (rgd_next && !error) {
837 				error = gfs2_glock_nq_init(rgd_next->rd_gl,
838 							   LM_ST_SHARED,
839 							   GL_ASYNC,
840 							   gh);
841 				rgd_next = gfs2_rgrpd_get_next(rgd_next);
842 				done = 0;
843 			}
844 
845 			if (signal_pending(current))
846 				error = -ERESTARTSYS;
847 		}
848 
849 		if (done)
850 			break;
851 
852 		yield();
853 	}
854 
855 	gfs2_glock_dq_uninit(&ri_gh);
856 
857 out:
858 	kfree(gha);
859 	return error;
860 }
861 
862 struct lfcc {
863 	struct list_head list;
864 	struct gfs2_holder gh;
865 };
866 
867 /**
868  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
869  *                            journals are clean
870  * @sdp: the file system
871  * @state: the state to put the transaction lock into
872  * @t_gh: the hold on the transaction lock
873  *
874  * Returns: errno
875  */
876 
877 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
878 				    struct gfs2_holder *t_gh)
879 {
880 	struct gfs2_inode *ip;
881 	struct gfs2_holder ji_gh;
882 	struct gfs2_jdesc *jd;
883 	struct lfcc *lfcc;
884 	LIST_HEAD(list);
885 	struct gfs2_log_header_host lh;
886 	int error;
887 
888 	error = gfs2_jindex_hold(sdp, &ji_gh);
889 	if (error)
890 		return error;
891 
892 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
893 		lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
894 		if (!lfcc) {
895 			error = -ENOMEM;
896 			goto out;
897 		}
898 		ip = GFS2_I(jd->jd_inode);
899 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
900 		if (error) {
901 			kfree(lfcc);
902 			goto out;
903 		}
904 		list_add(&lfcc->list, &list);
905 	}
906 
907 	error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
908 			       LM_FLAG_PRIORITY | GL_NOCACHE,
909 			       t_gh);
910 
911 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
912 		error = gfs2_jdesc_check(jd);
913 		if (error)
914 			break;
915 		error = gfs2_find_jhead(jd, &lh);
916 		if (error)
917 			break;
918 		if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
919 			error = -EBUSY;
920 			break;
921 		}
922 	}
923 
924 	if (error)
925 		gfs2_glock_dq_uninit(t_gh);
926 
927 out:
928 	while (!list_empty(&list)) {
929 		lfcc = list_entry(list.next, struct lfcc, list);
930 		list_del(&lfcc->list);
931 		gfs2_glock_dq_uninit(&lfcc->gh);
932 		kfree(lfcc);
933 	}
934 	gfs2_glock_dq_uninit(&ji_gh);
935 	return error;
936 }
937 
938 /**
939  * gfs2_freeze_fs - freezes the file system
940  * @sdp: the file system
941  *
942  * This function flushes data and meta data for all machines by
943  * aquiring the transaction log exclusively.  All journals are
944  * ensured to be in a clean state as well.
945  *
946  * Returns: errno
947  */
948 
949 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
950 {
951 	int error = 0;
952 
953 	mutex_lock(&sdp->sd_freeze_lock);
954 
955 	if (!sdp->sd_freeze_count++) {
956 		error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
957 		if (error)
958 			sdp->sd_freeze_count--;
959 	}
960 
961 	mutex_unlock(&sdp->sd_freeze_lock);
962 
963 	return error;
964 }
965 
966 /**
967  * gfs2_unfreeze_fs - unfreezes the file system
968  * @sdp: the file system
969  *
970  * This function allows the file system to proceed by unlocking
971  * the exclusively held transaction lock.  Other GFS2 nodes are
972  * now free to acquire the lock shared and go on with their lives.
973  *
974  */
975 
976 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
977 {
978 	mutex_lock(&sdp->sd_freeze_lock);
979 
980 	if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
981 		gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
982 
983 	mutex_unlock(&sdp->sd_freeze_lock);
984 }
985 
986