Merge branch 'linus' into x86/cleanups
[linux-2.6] / fs / gfs2 / super.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9
10 #include <linux/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_demote_secs = 300;
55         gt->gt_incore_log_blocks = 1024;
56         gt->gt_log_flush_secs = 60;
57         gt->gt_recoverd_secs = 60;
58         gt->gt_logd_secs = 1;
59         gt->gt_quotad_secs = 5;
60         gt->gt_quota_simul_sync = 64;
61         gt->gt_quota_warn_period = 10;
62         gt->gt_quota_scale_num = 1;
63         gt->gt_quota_scale_den = 1;
64         gt->gt_quota_cache_secs = 300;
65         gt->gt_quota_quantum = 60;
66         gt->gt_atime_quantum = 3600;
67         gt->gt_new_files_jdata = 0;
68         gt->gt_max_readahead = 1 << 18;
69         gt->gt_stall_secs = 600;
70         gt->gt_complain_secs = 10;
71         gt->gt_statfs_quantum = 30;
72         gt->gt_statfs_slow = 0;
73 }
74
75 /**
76  * gfs2_check_sb - Check superblock
77  * @sdp: the filesystem
78  * @sb: The superblock
79  * @silent: Don't print a message if the check fails
80  *
81  * Checks the version code of the FS is one that we understand how to
82  * read and that the sizes of the various on-disk structures have not
83  * changed.
84  */
85
86 int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
87 {
88         unsigned int x;
89
90         if (sb->sb_magic != GFS2_MAGIC ||
91             sb->sb_type != GFS2_METATYPE_SB) {
92                 if (!silent)
93                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
94                 return -EINVAL;
95         }
96
97         /*  If format numbers match exactly, we're done.  */
98
99         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
100             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
101                 return 0;
102
103         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
104                 for (x = 0; gfs2_old_fs_formats[x]; x++)
105                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
106                                 break;
107
108                 if (!gfs2_old_fs_formats[x]) {
109                         printk(KERN_WARNING
110                                "GFS2: code version (%u, %u) is incompatible "
111                                "with ondisk format (%u, %u)\n",
112                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
113                                sb->sb_fs_format, sb->sb_multihost_format);
114                         printk(KERN_WARNING
115                                "GFS2: I don't know how to upgrade this FS\n");
116                         return -EINVAL;
117                 }
118         }
119
120         if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
121                 for (x = 0; gfs2_old_multihost_formats[x]; x++)
122                         if (gfs2_old_multihost_formats[x] ==
123                             sb->sb_multihost_format)
124                                 break;
125
126                 if (!gfs2_old_multihost_formats[x]) {
127                         printk(KERN_WARNING
128                                "GFS2: code version (%u, %u) is incompatible "
129                                "with ondisk format (%u, %u)\n",
130                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
131                                sb->sb_fs_format, sb->sb_multihost_format);
132                         printk(KERN_WARNING
133                                "GFS2: I don't know how to upgrade this FS\n");
134                         return -EINVAL;
135                 }
136         }
137
138         if (!sdp->sd_args.ar_upgrade) {
139                 printk(KERN_WARNING
140                        "GFS2: code version (%u, %u) is incompatible "
141                        "with ondisk format (%u, %u)\n",
142                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
143                        sb->sb_fs_format, sb->sb_multihost_format);
144                 printk(KERN_INFO
145                        "GFS2: Use the \"upgrade\" mount option to upgrade "
146                        "the FS\n");
147                 printk(KERN_INFO "GFS2: See the manual for more details\n");
148                 return -EINVAL;
149         }
150
151         return 0;
152 }
153
154
155 static void end_bio_io_page(struct bio *bio, int error)
156 {
157         struct page *page = bio->bi_private;
158
159         if (!error)
160                 SetPageUptodate(page);
161         else
162                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
163         unlock_page(page);
164 }
165
166 static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
167 {
168         const struct gfs2_sb *str = buf;
169
170         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
171         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
172         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
173         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
174         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
175         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
176         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
177         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
178         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
179         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
180         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
181
182         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
183         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
184 }
185
186 /**
187  * gfs2_read_super - Read the gfs2 super block from disk
188  * @sdp: The GFS2 super block
189  * @sector: The location of the super block
190  * @error: The error code to return
191  *
192  * This uses the bio functions to read the super block from disk
193  * because we want to be 100% sure that we never read cached data.
194  * A super block is read twice only during each GFS2 mount and is
195  * never written to by the filesystem. The first time its read no
196  * locks are held, and the only details which are looked at are those
197  * relating to the locking protocol. Once locking is up and working,
198  * the sb is read again under the lock to establish the location of
199  * the master directory (contains pointers to journals etc) and the
200  * root directory.
201  *
202  * Returns: 0 on success or error
203  */
204
205 int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
206 {
207         struct super_block *sb = sdp->sd_vfs;
208         struct gfs2_sb *p;
209         struct page *page;
210         struct bio *bio;
211
212         page = alloc_page(GFP_NOFS);
213         if (unlikely(!page))
214                 return -ENOBUFS;
215
216         ClearPageUptodate(page);
217         ClearPageDirty(page);
218         lock_page(page);
219
220         bio = bio_alloc(GFP_NOFS, 1);
221         if (unlikely(!bio)) {
222                 __free_page(page);
223                 return -ENOBUFS;
224         }
225
226         bio->bi_sector = sector * (sb->s_blocksize >> 9);
227         bio->bi_bdev = sb->s_bdev;
228         bio_add_page(bio, page, PAGE_SIZE, 0);
229
230         bio->bi_end_io = end_bio_io_page;
231         bio->bi_private = page;
232         submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
233         wait_on_page_locked(page);
234         bio_put(bio);
235         if (!PageUptodate(page)) {
236                 __free_page(page);
237                 return -EIO;
238         }
239         p = kmap(page);
240         gfs2_sb_in(&sdp->sd_sb, p);
241         kunmap(page);
242         __free_page(page);
243         return 0;
244 }
245
246 /**
247  * gfs2_read_sb - Read super block
248  * @sdp: The GFS2 superblock
249  * @gl: the glock for the superblock (assumed to be held)
250  * @silent: Don't print message if mount fails
251  *
252  */
253
254 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
255 {
256         u32 hash_blocks, ind_blocks, leaf_blocks;
257         u32 tmp_blocks;
258         unsigned int x;
259         int error;
260
261         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
262         if (error) {
263                 if (!silent)
264                         fs_err(sdp, "can't read superblock\n");
265                 return error;
266         }
267
268         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
269         if (error)
270                 return error;
271
272         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
273                                GFS2_BASIC_BLOCK_SHIFT;
274         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
275         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
276                           sizeof(struct gfs2_dinode)) / sizeof(u64);
277         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
278                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
279         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
280         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
281         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
282         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
283         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
284                                 sizeof(struct gfs2_meta_header)) /
285                                 sizeof(struct gfs2_quota_change);
286
287         /* Compute maximum reservation required to add a entry to a directory */
288
289         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
290                              sdp->sd_jbsize);
291
292         ind_blocks = 0;
293         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
294                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
295                 ind_blocks += tmp_blocks;
296         }
297
298         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
299
300         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
301
302         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
303                                 sizeof(struct gfs2_dinode);
304         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
305         for (x = 2;; x++) {
306                 u64 space, d;
307                 u32 m;
308
309                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
310                 d = space;
311                 m = do_div(d, sdp->sd_inptrs);
312
313                 if (d != sdp->sd_heightsize[x - 1] || m)
314                         break;
315                 sdp->sd_heightsize[x] = space;
316         }
317         sdp->sd_max_height = x;
318         sdp->sd_heightsize[x] = ~0;
319         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
320
321         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
322                                  sizeof(struct gfs2_dinode);
323         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
324         for (x = 2;; x++) {
325                 u64 space, d;
326                 u32 m;
327
328                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
329                 d = space;
330                 m = do_div(d, sdp->sd_inptrs);
331
332                 if (d != sdp->sd_jheightsize[x - 1] || m)
333                         break;
334                 sdp->sd_jheightsize[x] = space;
335         }
336         sdp->sd_max_jheight = x;
337         sdp->sd_jheightsize[x] = ~0;
338         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
339
340         return 0;
341 }
342
343 /**
344  * gfs2_jindex_hold - Grab a lock on the jindex
345  * @sdp: The GFS2 superblock
346  * @ji_gh: the holder for the jindex glock
347  *
348  * This is very similar to the gfs2_rindex_hold() function, except that
349  * in general we hold the jindex lock for longer periods of time and
350  * we grab it far less frequently (in general) then the rgrp lock.
351  *
352  * Returns: errno
353  */
354
355 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
356 {
357         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
358         struct qstr name;
359         char buf[20];
360         struct gfs2_jdesc *jd;
361         int error;
362
363         name.name = buf;
364
365         mutex_lock(&sdp->sd_jindex_mutex);
366
367         for (;;) {
368                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
369                 if (error)
370                         break;
371
372                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
373                 name.hash = gfs2_disk_hash(name.name, name.len);
374
375                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
376                 if (error == -ENOENT) {
377                         error = 0;
378                         break;
379                 }
380
381                 gfs2_glock_dq_uninit(ji_gh);
382
383                 if (error)
384                         break;
385
386                 error = -ENOMEM;
387                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
388                 if (!jd)
389                         break;
390
391                 INIT_LIST_HEAD(&jd->extent_list);
392                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
393                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
394                         if (!jd->jd_inode)
395                                 error = -ENOENT;
396                         else
397                                 error = PTR_ERR(jd->jd_inode);
398                         kfree(jd);
399                         break;
400                 }
401
402                 spin_lock(&sdp->sd_jindex_spin);
403                 jd->jd_jid = sdp->sd_journals++;
404                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
405                 spin_unlock(&sdp->sd_jindex_spin);
406         }
407
408         mutex_unlock(&sdp->sd_jindex_mutex);
409
410         return error;
411 }
412
413 /**
414  * gfs2_jindex_free - Clear all the journal index information
415  * @sdp: The GFS2 superblock
416  *
417  */
418
419 void gfs2_jindex_free(struct gfs2_sbd *sdp)
420 {
421         struct list_head list, *head;
422         struct gfs2_jdesc *jd;
423         struct gfs2_journal_extent *jext;
424
425         spin_lock(&sdp->sd_jindex_spin);
426         list_add(&list, &sdp->sd_jindex_list);
427         list_del_init(&sdp->sd_jindex_list);
428         sdp->sd_journals = 0;
429         spin_unlock(&sdp->sd_jindex_spin);
430
431         while (!list_empty(&list)) {
432                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
433                 head = &jd->extent_list;
434                 while (!list_empty(head)) {
435                         jext = list_entry(head->next,
436                                           struct gfs2_journal_extent,
437                                           extent_list);
438                         list_del(&jext->extent_list);
439                         kfree(jext);
440                 }
441                 list_del(&jd->jd_list);
442                 iput(jd->jd_inode);
443                 kfree(jd);
444         }
445 }
446
447 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
448 {
449         struct gfs2_jdesc *jd;
450         int found = 0;
451
452         list_for_each_entry(jd, head, jd_list) {
453                 if (jd->jd_jid == jid) {
454                         found = 1;
455                         break;
456                 }
457         }
458
459         if (!found)
460                 jd = NULL;
461
462         return jd;
463 }
464
465 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
466 {
467         struct gfs2_jdesc *jd;
468
469         spin_lock(&sdp->sd_jindex_spin);
470         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
471         spin_unlock(&sdp->sd_jindex_spin);
472
473         return jd;
474 }
475
476 void gfs2_jdesc_make_dirty(struct gfs2_sbd *sdp, unsigned int jid)
477 {
478         struct gfs2_jdesc *jd;
479
480         spin_lock(&sdp->sd_jindex_spin);
481         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
482         if (jd)
483                 jd->jd_dirty = 1;
484         spin_unlock(&sdp->sd_jindex_spin);
485 }
486
487 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
488 {
489         struct gfs2_jdesc *jd;
490         int found = 0;
491
492         spin_lock(&sdp->sd_jindex_spin);
493
494         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
495                 if (jd->jd_dirty) {
496                         jd->jd_dirty = 0;
497                         found = 1;
498                         break;
499                 }
500         }
501         spin_unlock(&sdp->sd_jindex_spin);
502
503         if (!found)
504                 jd = NULL;
505
506         return jd;
507 }
508
509 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
510 {
511         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
512         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
513         int ar;
514         int error;
515
516         if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
517             (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
518                 gfs2_consist_inode(ip);
519                 return -EIO;
520         }
521         jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
522
523         error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
524         if (!error && ar) {
525                 gfs2_consist_inode(ip);
526                 error = -EIO;
527         }
528
529         return error;
530 }
531
532 /**
533  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
534  * @sdp: the filesystem
535  *
536  * Returns: errno
537  */
538
539 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
540 {
541         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
542         struct gfs2_glock *j_gl = ip->i_gl;
543         struct gfs2_holder t_gh;
544         struct gfs2_log_header_host head;
545         int error;
546
547         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
548         if (error)
549                 return error;
550
551         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
552
553         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
554         if (error)
555                 goto fail;
556
557         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
558                 gfs2_consist(sdp);
559                 error = -EIO;
560                 goto fail;
561         }
562
563         /*  Initialize some head of the log stuff  */
564         sdp->sd_log_sequence = head.lh_sequence + 1;
565         gfs2_log_pointers_init(sdp, head.lh_blkno);
566
567         error = gfs2_quota_init(sdp);
568         if (error)
569                 goto fail;
570
571         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
572
573         gfs2_glock_dq_uninit(&t_gh);
574
575         return 0;
576
577 fail:
578         t_gh.gh_flags |= GL_NOCACHE;
579         gfs2_glock_dq_uninit(&t_gh);
580
581         return error;
582 }
583
584 /**
585  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
586  * @sdp: the filesystem
587  *
588  * Returns: errno
589  */
590
591 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
592 {
593         struct gfs2_holder t_gh;
594         int error;
595
596         gfs2_quota_sync(sdp);
597         gfs2_statfs_sync(sdp);
598
599         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
600                                    &t_gh);
601         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
602                 return error;
603
604         gfs2_meta_syncfs(sdp);
605         gfs2_log_shutdown(sdp);
606
607         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
608
609         if (t_gh.gh_gl)
610                 gfs2_glock_dq_uninit(&t_gh);
611
612         gfs2_quota_cleanup(sdp);
613
614         return error;
615 }
616
617 static void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
618 {
619         const struct gfs2_statfs_change *str = buf;
620
621         sc->sc_total = be64_to_cpu(str->sc_total);
622         sc->sc_free = be64_to_cpu(str->sc_free);
623         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
624 }
625
626 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
627 {
628         struct gfs2_statfs_change *str = buf;
629
630         str->sc_total = cpu_to_be64(sc->sc_total);
631         str->sc_free = cpu_to_be64(sc->sc_free);
632         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
633 }
634
635 int gfs2_statfs_init(struct gfs2_sbd *sdp)
636 {
637         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
638         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
639         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
640         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
641         struct buffer_head *m_bh, *l_bh;
642         struct gfs2_holder gh;
643         int error;
644
645         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
646                                    &gh);
647         if (error)
648                 return error;
649
650         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
651         if (error)
652                 goto out;
653
654         if (sdp->sd_args.ar_spectator) {
655                 spin_lock(&sdp->sd_statfs_spin);
656                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
657                                       sizeof(struct gfs2_dinode));
658                 spin_unlock(&sdp->sd_statfs_spin);
659         } else {
660                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
661                 if (error)
662                         goto out_m_bh;
663
664                 spin_lock(&sdp->sd_statfs_spin);
665                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
666                                       sizeof(struct gfs2_dinode));
667                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
668                                       sizeof(struct gfs2_dinode));
669                 spin_unlock(&sdp->sd_statfs_spin);
670
671                 brelse(l_bh);
672         }
673
674 out_m_bh:
675         brelse(m_bh);
676 out:
677         gfs2_glock_dq_uninit(&gh);
678         return 0;
679 }
680
681 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
682                         s64 dinodes)
683 {
684         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
685         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
686         struct buffer_head *l_bh;
687         int error;
688
689         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
690         if (error)
691                 return;
692
693         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
694
695         spin_lock(&sdp->sd_statfs_spin);
696         l_sc->sc_total += total;
697         l_sc->sc_free += free;
698         l_sc->sc_dinodes += dinodes;
699         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
700         spin_unlock(&sdp->sd_statfs_spin);
701
702         brelse(l_bh);
703 }
704
705 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
706 {
707         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
708         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
709         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
710         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
711         struct gfs2_holder gh;
712         struct buffer_head *m_bh, *l_bh;
713         int error;
714
715         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
716                                    &gh);
717         if (error)
718                 return error;
719
720         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
721         if (error)
722                 goto out;
723
724         spin_lock(&sdp->sd_statfs_spin);
725         gfs2_statfs_change_in(m_sc, m_bh->b_data +
726                               sizeof(struct gfs2_dinode));
727         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
728                 spin_unlock(&sdp->sd_statfs_spin);
729                 goto out_bh;
730         }
731         spin_unlock(&sdp->sd_statfs_spin);
732
733         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
734         if (error)
735                 goto out_bh;
736
737         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
738         if (error)
739                 goto out_bh2;
740
741         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
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                                    GL_NOCACHE, t_gh);
944
945         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
946                 error = gfs2_jdesc_check(jd);
947                 if (error)
948                         break;
949                 error = gfs2_find_jhead(jd, &lh);
950                 if (error)
951                         break;
952                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
953                         error = -EBUSY;
954                         break;
955                 }
956         }
957
958         if (error)
959                 gfs2_glock_dq_uninit(t_gh);
960
961 out:
962         while (!list_empty(&list)) {
963                 lfcc = list_entry(list.next, struct lfcc, list);
964                 list_del(&lfcc->list);
965                 gfs2_glock_dq_uninit(&lfcc->gh);
966                 kfree(lfcc);
967         }
968         gfs2_glock_dq_uninit(&ji_gh);
969         return error;
970 }
971
972 /**
973  * gfs2_freeze_fs - freezes the file system
974  * @sdp: the file system
975  *
976  * This function flushes data and meta data for all machines by
977  * aquiring the transaction log exclusively.  All journals are
978  * ensured to be in a clean state as well.
979  *
980  * Returns: errno
981  */
982
983 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
984 {
985         int error = 0;
986
987         mutex_lock(&sdp->sd_freeze_lock);
988
989         if (!sdp->sd_freeze_count++) {
990                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
991                 if (error)
992                         sdp->sd_freeze_count--;
993         }
994
995         mutex_unlock(&sdp->sd_freeze_lock);
996
997         return error;
998 }
999
1000 /**
1001  * gfs2_unfreeze_fs - unfreezes the file system
1002  * @sdp: the file system
1003  *
1004  * This function allows the file system to proceed by unlocking
1005  * the exclusively held transaction lock.  Other GFS2 nodes are
1006  * now free to acquire the lock shared and go on with their lives.
1007  *
1008  */
1009
1010 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
1011 {
1012         mutex_lock(&sdp->sd_freeze_lock);
1013
1014         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
1015                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1016
1017         mutex_unlock(&sdp->sd_freeze_lock);
1018 }
1019