Merge tag 'for-linus-v3.9-rc1' of git://oss.sgi.com/xfs/xfs
[~shefty/rdma-dev.git] / fs / gfs2 / ops_fstype.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2008 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/blkdev.h>
16 #include <linux/kthread.h>
17 #include <linux/export.h>
18 #include <linux/namei.h>
19 #include <linux/mount.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/quotaops.h>
22 #include <linux/lockdep.h>
23
24 #include "gfs2.h"
25 #include "incore.h"
26 #include "bmap.h"
27 #include "glock.h"
28 #include "glops.h"
29 #include "inode.h"
30 #include "recovery.h"
31 #include "rgrp.h"
32 #include "super.h"
33 #include "sys.h"
34 #include "util.h"
35 #include "log.h"
36 #include "quota.h"
37 #include "dir.h"
38 #include "trace_gfs2.h"
39
40 #define DO 0
41 #define UNDO 1
42
43 /**
44  * gfs2_tune_init - Fill a gfs2_tune structure with default values
45  * @gt: tune
46  *
47  */
48
49 static void gfs2_tune_init(struct gfs2_tune *gt)
50 {
51         spin_lock_init(&gt->gt_spin);
52
53         gt->gt_quota_simul_sync = 64;
54         gt->gt_quota_warn_period = 10;
55         gt->gt_quota_scale_num = 1;
56         gt->gt_quota_scale_den = 1;
57         gt->gt_new_files_jdata = 0;
58         gt->gt_max_readahead = 1 << 18;
59         gt->gt_complain_secs = 10;
60 }
61
62 static struct gfs2_sbd *init_sbd(struct super_block *sb)
63 {
64         struct gfs2_sbd *sdp;
65
66         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
67         if (!sdp)
68                 return NULL;
69
70         sb->s_fs_info = sdp;
71         sdp->sd_vfs = sb;
72         sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
73         if (!sdp->sd_lkstats) {
74                 kfree(sdp);
75                 return NULL;
76         }
77
78         set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
79         gfs2_tune_init(&sdp->sd_tune);
80
81         init_waitqueue_head(&sdp->sd_glock_wait);
82         atomic_set(&sdp->sd_glock_disposal, 0);
83         init_completion(&sdp->sd_locking_init);
84         spin_lock_init(&sdp->sd_statfs_spin);
85
86         spin_lock_init(&sdp->sd_rindex_spin);
87         sdp->sd_rindex_tree.rb_node = NULL;
88
89         INIT_LIST_HEAD(&sdp->sd_jindex_list);
90         spin_lock_init(&sdp->sd_jindex_spin);
91         mutex_init(&sdp->sd_jindex_mutex);
92
93         INIT_LIST_HEAD(&sdp->sd_quota_list);
94         mutex_init(&sdp->sd_quota_mutex);
95         init_waitqueue_head(&sdp->sd_quota_wait);
96         INIT_LIST_HEAD(&sdp->sd_trunc_list);
97         spin_lock_init(&sdp->sd_trunc_lock);
98
99         spin_lock_init(&sdp->sd_log_lock);
100         atomic_set(&sdp->sd_log_pinned, 0);
101         INIT_LIST_HEAD(&sdp->sd_log_le_buf);
102         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
103         INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
104         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
105
106         init_waitqueue_head(&sdp->sd_log_waitq);
107         init_waitqueue_head(&sdp->sd_logd_waitq);
108         spin_lock_init(&sdp->sd_ail_lock);
109         INIT_LIST_HEAD(&sdp->sd_ail1_list);
110         INIT_LIST_HEAD(&sdp->sd_ail2_list);
111
112         init_rwsem(&sdp->sd_log_flush_lock);
113         atomic_set(&sdp->sd_log_in_flight, 0);
114         init_waitqueue_head(&sdp->sd_log_flush_wait);
115
116         INIT_LIST_HEAD(&sdp->sd_revoke_list);
117
118         mutex_init(&sdp->sd_freeze_lock);
119
120         return sdp;
121 }
122
123
124 /**
125  * gfs2_check_sb - Check superblock
126  * @sdp: the filesystem
127  * @sb: The superblock
128  * @silent: Don't print a message if the check fails
129  *
130  * Checks the version code of the FS is one that we understand how to
131  * read and that the sizes of the various on-disk structures have not
132  * changed.
133  */
134
135 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
136 {
137         struct gfs2_sb_host *sb = &sdp->sd_sb;
138
139         if (sb->sb_magic != GFS2_MAGIC ||
140             sb->sb_type != GFS2_METATYPE_SB) {
141                 if (!silent)
142                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
143                 return -EINVAL;
144         }
145
146         /*  If format numbers match exactly, we're done.  */
147
148         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
149             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
150                 return 0;
151
152         fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
153
154         return -EINVAL;
155 }
156
157 static void end_bio_io_page(struct bio *bio, int error)
158 {
159         struct page *page = bio->bi_private;
160
161         if (!error)
162                 SetPageUptodate(page);
163         else
164                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
165         unlock_page(page);
166 }
167
168 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
169 {
170         struct gfs2_sb_host *sb = &sdp->sd_sb;
171         struct super_block *s = sdp->sd_vfs;
172         const struct gfs2_sb *str = buf;
173
174         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
175         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
176         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
177         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
178         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
179         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
180         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
181         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
182         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
183         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
184         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
185
186         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
187         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
188         memcpy(s->s_uuid, str->sb_uuid, 16);
189 }
190
191 /**
192  * gfs2_read_super - Read the gfs2 super block from disk
193  * @sdp: The GFS2 super block
194  * @sector: The location of the super block
195  * @error: The error code to return
196  *
197  * This uses the bio functions to read the super block from disk
198  * because we want to be 100% sure that we never read cached data.
199  * A super block is read twice only during each GFS2 mount and is
200  * never written to by the filesystem. The first time its read no
201  * locks are held, and the only details which are looked at are those
202  * relating to the locking protocol. Once locking is up and working,
203  * the sb is read again under the lock to establish the location of
204  * the master directory (contains pointers to journals etc) and the
205  * root directory.
206  *
207  * Returns: 0 on success or error
208  */
209
210 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
211 {
212         struct super_block *sb = sdp->sd_vfs;
213         struct gfs2_sb *p;
214         struct page *page;
215         struct bio *bio;
216
217         page = alloc_page(GFP_NOFS);
218         if (unlikely(!page))
219                 return -ENOBUFS;
220
221         ClearPageUptodate(page);
222         ClearPageDirty(page);
223         lock_page(page);
224
225         bio = bio_alloc(GFP_NOFS, 1);
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 | REQ_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, p);
241         kunmap(page);
242         __free_page(page);
243         return gfs2_check_sb(sdp, silent);
244 }
245
246 /**
247  * gfs2_read_sb - Read super block
248  * @sdp: The GFS2 superblock
249  * @silent: Don't print message if mount fails
250  *
251  */
252
253 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
254 {
255         u32 hash_blocks, ind_blocks, leaf_blocks;
256         u32 tmp_blocks;
257         unsigned int x;
258         int error;
259
260         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
261         if (error) {
262                 if (!silent)
263                         fs_err(sdp, "can't read superblock\n");
264                 return error;
265         }
266
267         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
268                                GFS2_BASIC_BLOCK_SHIFT;
269         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
270         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
271                           sizeof(struct gfs2_dinode)) / sizeof(u64);
272         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
273                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
274         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
275         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
276         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
277         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
278         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
279                                 sizeof(struct gfs2_meta_header)) /
280                                 sizeof(struct gfs2_quota_change);
281         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
282                                      sizeof(struct gfs2_meta_header))
283                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
284
285         /* Compute maximum reservation required to add a entry to a directory */
286
287         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
288                              sdp->sd_jbsize);
289
290         ind_blocks = 0;
291         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
292                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
293                 ind_blocks += tmp_blocks;
294         }
295
296         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
297
298         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
299
300         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
301                                 sizeof(struct gfs2_dinode);
302         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
303         for (x = 2;; x++) {
304                 u64 space, d;
305                 u32 m;
306
307                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
308                 d = space;
309                 m = do_div(d, sdp->sd_inptrs);
310
311                 if (d != sdp->sd_heightsize[x - 1] || m)
312                         break;
313                 sdp->sd_heightsize[x] = space;
314         }
315         sdp->sd_max_height = x;
316         sdp->sd_heightsize[x] = ~0;
317         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
318
319         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
320                                  sizeof(struct gfs2_dinode);
321         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
322         for (x = 2;; x++) {
323                 u64 space, d;
324                 u32 m;
325
326                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
327                 d = space;
328                 m = do_div(d, sdp->sd_inptrs);
329
330                 if (d != sdp->sd_jheightsize[x - 1] || m)
331                         break;
332                 sdp->sd_jheightsize[x] = space;
333         }
334         sdp->sd_max_jheight = x;
335         sdp->sd_jheightsize[x] = ~0;
336         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
337
338         return 0;
339 }
340
341 static int init_names(struct gfs2_sbd *sdp, int silent)
342 {
343         char *proto, *table;
344         int error = 0;
345
346         proto = sdp->sd_args.ar_lockproto;
347         table = sdp->sd_args.ar_locktable;
348
349         /*  Try to autodetect  */
350
351         if (!proto[0] || !table[0]) {
352                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
353                 if (error)
354                         return error;
355
356                 if (!proto[0])
357                         proto = sdp->sd_sb.sb_lockproto;
358                 if (!table[0])
359                         table = sdp->sd_sb.sb_locktable;
360         }
361
362         if (!table[0])
363                 table = sdp->sd_vfs->s_id;
364
365         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
366         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
367
368         table = sdp->sd_table_name;
369         while ((table = strchr(table, '/')))
370                 *table = '_';
371
372         return error;
373 }
374
375 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
376                         int undo)
377 {
378         int error = 0;
379
380         if (undo)
381                 goto fail_trans;
382
383         error = gfs2_glock_nq_num(sdp,
384                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
385                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
386                                   mount_gh);
387         if (error) {
388                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
389                 goto fail;
390         }
391
392         error = gfs2_glock_nq_num(sdp,
393                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
394                                   LM_ST_SHARED,
395                                   LM_FLAG_NOEXP | GL_EXACT,
396                                   &sdp->sd_live_gh);
397         if (error) {
398                 fs_err(sdp, "can't acquire live glock: %d\n", error);
399                 goto fail_mount;
400         }
401
402         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
403                                CREATE, &sdp->sd_rename_gl);
404         if (error) {
405                 fs_err(sdp, "can't create rename glock: %d\n", error);
406                 goto fail_live;
407         }
408
409         error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
410                                CREATE, &sdp->sd_trans_gl);
411         if (error) {
412                 fs_err(sdp, "can't create transaction glock: %d\n", error);
413                 goto fail_rename;
414         }
415
416         return 0;
417
418 fail_trans:
419         gfs2_glock_put(sdp->sd_trans_gl);
420 fail_rename:
421         gfs2_glock_put(sdp->sd_rename_gl);
422 fail_live:
423         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
424 fail_mount:
425         gfs2_glock_dq_uninit(mount_gh);
426 fail:
427         return error;
428 }
429
430 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
431                             u64 no_addr, const char *name)
432 {
433         struct gfs2_sbd *sdp = sb->s_fs_info;
434         struct dentry *dentry;
435         struct inode *inode;
436
437         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
438         if (IS_ERR(inode)) {
439                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
440                 return PTR_ERR(inode);
441         }
442         dentry = d_make_root(inode);
443         if (!dentry) {
444                 fs_err(sdp, "can't alloc %s dentry\n", name);
445                 return -ENOMEM;
446         }
447         *dptr = dentry;
448         return 0;
449 }
450
451 static int init_sb(struct gfs2_sbd *sdp, int silent)
452 {
453         struct super_block *sb = sdp->sd_vfs;
454         struct gfs2_holder sb_gh;
455         u64 no_addr;
456         int ret;
457
458         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
459                                 LM_ST_SHARED, 0, &sb_gh);
460         if (ret) {
461                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
462                 return ret;
463         }
464
465         ret = gfs2_read_sb(sdp, silent);
466         if (ret) {
467                 fs_err(sdp, "can't read superblock: %d\n", ret);
468                 goto out;
469         }
470
471         /* Set up the buffer cache and SB for real */
472         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
473                 ret = -EINVAL;
474                 fs_err(sdp, "FS block size (%u) is too small for device "
475                        "block size (%u)\n",
476                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
477                 goto out;
478         }
479         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
480                 ret = -EINVAL;
481                 fs_err(sdp, "FS block size (%u) is too big for machine "
482                        "page size (%u)\n",
483                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
484                 goto out;
485         }
486         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
487
488         /* Get the root inode */
489         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
490         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
491         if (ret)
492                 goto out;
493
494         /* Get the master inode */
495         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
496         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
497         if (ret) {
498                 dput(sdp->sd_root_dir);
499                 goto out;
500         }
501         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
502 out:
503         gfs2_glock_dq_uninit(&sb_gh);
504         return ret;
505 }
506
507 /**
508  * map_journal_extents - create a reusable "extent" mapping from all logical
509  * blocks to all physical blocks for the given journal.  This will save
510  * us time when writing journal blocks.  Most journals will have only one
511  * extent that maps all their logical blocks.  That's because gfs2.mkfs
512  * arranges the journal blocks sequentially to maximize performance.
513  * So the extent would map the first block for the entire file length.
514  * However, gfs2_jadd can happen while file activity is happening, so
515  * those journals may not be sequential.  Less likely is the case where
516  * the users created their own journals by mounting the metafs and
517  * laying it out.  But it's still possible.  These journals might have
518  * several extents.
519  *
520  * TODO: This should be done in bigger chunks rather than one block at a time,
521  *       but since it's only done at mount time, I'm not worried about the
522  *       time it takes.
523  */
524 static int map_journal_extents(struct gfs2_sbd *sdp)
525 {
526         struct gfs2_jdesc *jd = sdp->sd_jdesc;
527         unsigned int lb;
528         u64 db, prev_db; /* logical block, disk block, prev disk block */
529         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
530         struct gfs2_journal_extent *jext = NULL;
531         struct buffer_head bh;
532         int rc = 0;
533
534         prev_db = 0;
535
536         for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
537                 bh.b_state = 0;
538                 bh.b_blocknr = 0;
539                 bh.b_size = 1 << ip->i_inode.i_blkbits;
540                 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
541                 db = bh.b_blocknr;
542                 if (rc || !db) {
543                         printk(KERN_INFO "GFS2 journal mapping error %d: lb="
544                                "%u db=%llu\n", rc, lb, (unsigned long long)db);
545                         break;
546                 }
547                 if (!prev_db || db != prev_db + 1) {
548                         jext = kzalloc(sizeof(struct gfs2_journal_extent),
549                                        GFP_KERNEL);
550                         if (!jext) {
551                                 printk(KERN_INFO "GFS2 error: out of memory "
552                                        "mapping journal extents.\n");
553                                 rc = -ENOMEM;
554                                 break;
555                         }
556                         jext->dblock = db;
557                         jext->lblock = lb;
558                         jext->blocks = 1;
559                         list_add_tail(&jext->extent_list, &jd->extent_list);
560                 } else {
561                         jext->blocks++;
562                 }
563                 prev_db = db;
564         }
565         return rc;
566 }
567
568 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
569 {
570         char *message = "FIRSTMOUNT=Done";
571         char *envp[] = { message, NULL };
572
573         fs_info(sdp, "first mount done, others may mount\n");
574
575         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
576                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
577
578         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
579 }
580
581 /**
582  * gfs2_jindex_hold - Grab a lock on the jindex
583  * @sdp: The GFS2 superblock
584  * @ji_gh: the holder for the jindex glock
585  *
586  * Returns: errno
587  */
588
589 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
590 {
591         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
592         struct qstr name;
593         char buf[20];
594         struct gfs2_jdesc *jd;
595         int error;
596
597         name.name = buf;
598
599         mutex_lock(&sdp->sd_jindex_mutex);
600
601         for (;;) {
602                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
603                 if (error)
604                         break;
605
606                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
607                 name.hash = gfs2_disk_hash(name.name, name.len);
608
609                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
610                 if (error == -ENOENT) {
611                         error = 0;
612                         break;
613                 }
614
615                 gfs2_glock_dq_uninit(ji_gh);
616
617                 if (error)
618                         break;
619
620                 error = -ENOMEM;
621                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
622                 if (!jd)
623                         break;
624
625                 INIT_LIST_HEAD(&jd->extent_list);
626                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
627                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
628                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
629                         if (!jd->jd_inode)
630                                 error = -ENOENT;
631                         else
632                                 error = PTR_ERR(jd->jd_inode);
633                         kfree(jd);
634                         break;
635                 }
636
637                 spin_lock(&sdp->sd_jindex_spin);
638                 jd->jd_jid = sdp->sd_journals++;
639                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
640                 spin_unlock(&sdp->sd_jindex_spin);
641         }
642
643         mutex_unlock(&sdp->sd_jindex_mutex);
644
645         return error;
646 }
647
648 static int init_journal(struct gfs2_sbd *sdp, int undo)
649 {
650         struct inode *master = sdp->sd_master_dir->d_inode;
651         struct gfs2_holder ji_gh;
652         struct gfs2_inode *ip;
653         int jindex = 1;
654         int error = 0;
655
656         if (undo) {
657                 jindex = 0;
658                 goto fail_jinode_gh;
659         }
660
661         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
662         if (IS_ERR(sdp->sd_jindex)) {
663                 fs_err(sdp, "can't lookup journal index: %d\n", error);
664                 return PTR_ERR(sdp->sd_jindex);
665         }
666
667         /* Load in the journal index special file */
668
669         error = gfs2_jindex_hold(sdp, &ji_gh);
670         if (error) {
671                 fs_err(sdp, "can't read journal index: %d\n", error);
672                 goto fail;
673         }
674
675         error = -EUSERS;
676         if (!gfs2_jindex_size(sdp)) {
677                 fs_err(sdp, "no journals!\n");
678                 goto fail_jindex;
679         }
680
681         if (sdp->sd_args.ar_spectator) {
682                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
683                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
684                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
685                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
686         } else {
687                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
688                         fs_err(sdp, "can't mount journal #%u\n",
689                                sdp->sd_lockstruct.ls_jid);
690                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
691                                gfs2_jindex_size(sdp),
692                                gfs2_jindex_size(sdp) - 1);
693                         goto fail_jindex;
694                 }
695                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
696
697                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
698                                           &gfs2_journal_glops,
699                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
700                                           &sdp->sd_journal_gh);
701                 if (error) {
702                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
703                         goto fail_jindex;
704                 }
705
706                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
707                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
708                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
709                                            &sdp->sd_jinode_gh);
710                 if (error) {
711                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
712                                error);
713                         goto fail_journal_gh;
714                 }
715
716                 error = gfs2_jdesc_check(sdp->sd_jdesc);
717                 if (error) {
718                         fs_err(sdp, "my journal (%u) is bad: %d\n",
719                                sdp->sd_jdesc->jd_jid, error);
720                         goto fail_jinode_gh;
721                 }
722                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
723                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
724                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
725
726                 /* Map the extents for this journal's blocks */
727                 map_journal_extents(sdp);
728         }
729         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
730
731         if (sdp->sd_lockstruct.ls_first) {
732                 unsigned int x;
733                 for (x = 0; x < sdp->sd_journals; x++) {
734                         error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x),
735                                                      true);
736                         if (error) {
737                                 fs_err(sdp, "error recovering journal %u: %d\n",
738                                        x, error);
739                                 goto fail_jinode_gh;
740                         }
741                 }
742
743                 gfs2_others_may_mount(sdp);
744         } else if (!sdp->sd_args.ar_spectator) {
745                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
746                 if (error) {
747                         fs_err(sdp, "error recovering my journal: %d\n", error);
748                         goto fail_jinode_gh;
749                 }
750         }
751
752         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
753         gfs2_glock_dq_uninit(&ji_gh);
754         jindex = 0;
755
756         return 0;
757
758 fail_jinode_gh:
759         if (!sdp->sd_args.ar_spectator)
760                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
761 fail_journal_gh:
762         if (!sdp->sd_args.ar_spectator)
763                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
764 fail_jindex:
765         gfs2_jindex_free(sdp);
766         if (jindex)
767                 gfs2_glock_dq_uninit(&ji_gh);
768 fail:
769         iput(sdp->sd_jindex);
770         return error;
771 }
772
773 static struct lock_class_key gfs2_quota_imutex_key;
774
775 static int init_inodes(struct gfs2_sbd *sdp, int undo)
776 {
777         int error = 0;
778         struct inode *master = sdp->sd_master_dir->d_inode;
779
780         if (undo)
781                 goto fail_qinode;
782
783         error = init_journal(sdp, undo);
784         if (error)
785                 goto fail;
786
787         /* Read in the master statfs inode */
788         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
789         if (IS_ERR(sdp->sd_statfs_inode)) {
790                 error = PTR_ERR(sdp->sd_statfs_inode);
791                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
792                 goto fail_journal;
793         }
794
795         /* Read in the resource index inode */
796         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
797         if (IS_ERR(sdp->sd_rindex)) {
798                 error = PTR_ERR(sdp->sd_rindex);
799                 fs_err(sdp, "can't get resource index inode: %d\n", error);
800                 goto fail_statfs;
801         }
802         sdp->sd_rindex_uptodate = 0;
803
804         /* Read in the quota inode */
805         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
806         if (IS_ERR(sdp->sd_quota_inode)) {
807                 error = PTR_ERR(sdp->sd_quota_inode);
808                 fs_err(sdp, "can't get quota file inode: %d\n", error);
809                 goto fail_rindex;
810         }
811         /*
812          * i_mutex on quota files is special. Since this inode is hidden system
813          * file, we are safe to define locking ourselves.
814          */
815         lockdep_set_class(&sdp->sd_quota_inode->i_mutex,
816                           &gfs2_quota_imutex_key);
817
818         error = gfs2_rindex_update(sdp);
819         if (error)
820                 goto fail_qinode;
821
822         return 0;
823
824 fail_qinode:
825         iput(sdp->sd_quota_inode);
826 fail_rindex:
827         gfs2_clear_rgrpd(sdp);
828         iput(sdp->sd_rindex);
829 fail_statfs:
830         iput(sdp->sd_statfs_inode);
831 fail_journal:
832         init_journal(sdp, UNDO);
833 fail:
834         return error;
835 }
836
837 static int init_per_node(struct gfs2_sbd *sdp, int undo)
838 {
839         struct inode *pn = NULL;
840         char buf[30];
841         int error = 0;
842         struct gfs2_inode *ip;
843         struct inode *master = sdp->sd_master_dir->d_inode;
844
845         if (sdp->sd_args.ar_spectator)
846                 return 0;
847
848         if (undo)
849                 goto fail_qc_gh;
850
851         pn = gfs2_lookup_simple(master, "per_node");
852         if (IS_ERR(pn)) {
853                 error = PTR_ERR(pn);
854                 fs_err(sdp, "can't find per_node directory: %d\n", error);
855                 return error;
856         }
857
858         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
859         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
860         if (IS_ERR(sdp->sd_sc_inode)) {
861                 error = PTR_ERR(sdp->sd_sc_inode);
862                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
863                 goto fail;
864         }
865
866         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
867         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
868         if (IS_ERR(sdp->sd_qc_inode)) {
869                 error = PTR_ERR(sdp->sd_qc_inode);
870                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
871                 goto fail_ut_i;
872         }
873
874         iput(pn);
875         pn = NULL;
876
877         ip = GFS2_I(sdp->sd_sc_inode);
878         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
879                                    &sdp->sd_sc_gh);
880         if (error) {
881                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
882                 goto fail_qc_i;
883         }
884
885         ip = GFS2_I(sdp->sd_qc_inode);
886         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
887                                    &sdp->sd_qc_gh);
888         if (error) {
889                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
890                 goto fail_ut_gh;
891         }
892
893         return 0;
894
895 fail_qc_gh:
896         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
897 fail_ut_gh:
898         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
899 fail_qc_i:
900         iput(sdp->sd_qc_inode);
901 fail_ut_i:
902         iput(sdp->sd_sc_inode);
903 fail:
904         if (pn)
905                 iput(pn);
906         return error;
907 }
908
909 static int init_threads(struct gfs2_sbd *sdp, int undo)
910 {
911         struct task_struct *p;
912         int error = 0;
913
914         if (undo)
915                 goto fail_quotad;
916
917         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
918         error = IS_ERR(p);
919         if (error) {
920                 fs_err(sdp, "can't start logd thread: %d\n", error);
921                 return error;
922         }
923         sdp->sd_logd_process = p;
924
925         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
926         error = IS_ERR(p);
927         if (error) {
928                 fs_err(sdp, "can't start quotad thread: %d\n", error);
929                 goto fail;
930         }
931         sdp->sd_quotad_process = p;
932
933         return 0;
934
935
936 fail_quotad:
937         kthread_stop(sdp->sd_quotad_process);
938 fail:
939         kthread_stop(sdp->sd_logd_process);
940         return error;
941 }
942
943 static const match_table_t nolock_tokens = {
944         { Opt_jid, "jid=%d\n", },
945         { Opt_err, NULL },
946 };
947
948 static const struct lm_lockops nolock_ops = {
949         .lm_proto_name = "lock_nolock",
950         .lm_put_lock = gfs2_glock_free,
951         .lm_tokens = &nolock_tokens,
952 };
953
954 /**
955  * gfs2_lm_mount - mount a locking protocol
956  * @sdp: the filesystem
957  * @args: mount arguments
958  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
959  *
960  * Returns: errno
961  */
962
963 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
964 {
965         const struct lm_lockops *lm;
966         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
967         struct gfs2_args *args = &sdp->sd_args;
968         const char *proto = sdp->sd_proto_name;
969         const char *table = sdp->sd_table_name;
970         char *o, *options;
971         int ret;
972
973         if (!strcmp("lock_nolock", proto)) {
974                 lm = &nolock_ops;
975                 sdp->sd_args.ar_localflocks = 1;
976 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
977         } else if (!strcmp("lock_dlm", proto)) {
978                 lm = &gfs2_dlm_ops;
979 #endif
980         } else {
981                 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
982                 return -ENOENT;
983         }
984
985         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
986
987         ls->ls_ops = lm;
988         ls->ls_first = 1;
989
990         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
991                 substring_t tmp[MAX_OPT_ARGS];
992                 int token, option;
993
994                 if (!o || !*o)
995                         continue;
996
997                 token = match_token(o, *lm->lm_tokens, tmp);
998                 switch (token) {
999                 case Opt_jid:
1000                         ret = match_int(&tmp[0], &option);
1001                         if (ret || option < 0) 
1002                                 goto hostdata_error;
1003                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1004                                 ls->ls_jid = option;
1005                         break;
1006                 case Opt_id:
1007                 case Opt_nodir:
1008                         /* Obsolete, but left for backward compat purposes */
1009                         break;
1010                 case Opt_first:
1011                         ret = match_int(&tmp[0], &option);
1012                         if (ret || (option != 0 && option != 1))
1013                                 goto hostdata_error;
1014                         ls->ls_first = option;
1015                         break;
1016                 case Opt_err:
1017                 default:
1018 hostdata_error:
1019                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1020                         return -EINVAL;
1021                 }
1022         }
1023
1024         if (lm->lm_mount == NULL) {
1025                 fs_info(sdp, "Now mounting FS...\n");
1026                 complete_all(&sdp->sd_locking_init);
1027                 return 0;
1028         }
1029         ret = lm->lm_mount(sdp, table);
1030         if (ret == 0)
1031                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1032         complete_all(&sdp->sd_locking_init);
1033         return ret;
1034 }
1035
1036 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1037 {
1038         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1039         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1040             lm->lm_unmount)
1041                 lm->lm_unmount(sdp);
1042 }
1043
1044 static int gfs2_journalid_wait(void *word)
1045 {
1046         if (signal_pending(current))
1047                 return -EINTR;
1048         schedule();
1049         return 0;
1050 }
1051
1052 static int wait_on_journal(struct gfs2_sbd *sdp)
1053 {
1054         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1055                 return 0;
1056
1057         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1058 }
1059
1060 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1061 {
1062         struct super_block *sb = sdp->sd_vfs;
1063         char ro[20];
1064         char spectator[20];
1065         char *envp[] = { ro, spectator, NULL };
1066         sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1067         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1068         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1069 }
1070
1071 /**
1072  * fill_super - Read in superblock
1073  * @sb: The VFS superblock
1074  * @data: Mount options
1075  * @silent: Don't complain if it's not a GFS2 filesystem
1076  *
1077  * Returns: errno
1078  */
1079
1080 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1081 {
1082         struct gfs2_sbd *sdp;
1083         struct gfs2_holder mount_gh;
1084         int error;
1085
1086         sdp = init_sbd(sb);
1087         if (!sdp) {
1088                 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1089                 return -ENOMEM;
1090         }
1091         sdp->sd_args = *args;
1092
1093         if (sdp->sd_args.ar_spectator) {
1094                 sb->s_flags |= MS_RDONLY;
1095                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1096         }
1097         if (sdp->sd_args.ar_posix_acl)
1098                 sb->s_flags |= MS_POSIXACL;
1099         if (sdp->sd_args.ar_nobarrier)
1100                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1101
1102         sb->s_flags |= MS_NOSEC;
1103         sb->s_magic = GFS2_MAGIC;
1104         sb->s_op = &gfs2_super_ops;
1105         sb->s_d_op = &gfs2_dops;
1106         sb->s_export_op = &gfs2_export_ops;
1107         sb->s_xattr = gfs2_xattr_handlers;
1108         sb->s_qcop = &gfs2_quotactl_ops;
1109         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1110         sb->s_time_gran = 1;
1111         sb->s_maxbytes = MAX_LFS_FILESIZE;
1112
1113         /* Set up the buffer cache and fill in some fake block size values
1114            to allow us to read-in the on-disk superblock. */
1115         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1116         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1117         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1118                                GFS2_BASIC_BLOCK_SHIFT;
1119         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1120
1121         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1122         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1123         if (sdp->sd_args.ar_statfs_quantum) {
1124                 sdp->sd_tune.gt_statfs_slow = 0;
1125                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1126         } else {
1127                 sdp->sd_tune.gt_statfs_slow = 1;
1128                 sdp->sd_tune.gt_statfs_quantum = 30;
1129         }
1130
1131         error = init_names(sdp, silent);
1132         if (error) {
1133                 /* In this case, we haven't initialized sysfs, so we have to
1134                    manually free the sdp. */
1135                 free_percpu(sdp->sd_lkstats);
1136                 kfree(sdp);
1137                 sb->s_fs_info = NULL;
1138                 return error;
1139         }
1140
1141         snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1142
1143         error = gfs2_sys_fs_add(sdp);
1144         /*
1145          * If we hit an error here, gfs2_sys_fs_add will have called function
1146          * kobject_put which causes the sysfs usage count to go to zero, which
1147          * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1148          * Subsequent error paths here will call gfs2_sys_fs_del, which also
1149          * kobject_put to free sdp.
1150          */
1151         if (error)
1152                 return error;
1153
1154         gfs2_create_debugfs_file(sdp);
1155
1156         error = gfs2_lm_mount(sdp, silent);
1157         if (error)
1158                 goto fail_debug;
1159
1160         error = init_locking(sdp, &mount_gh, DO);
1161         if (error)
1162                 goto fail_lm;
1163
1164         error = init_sb(sdp, silent);
1165         if (error)
1166                 goto fail_locking;
1167
1168         error = wait_on_journal(sdp);
1169         if (error)
1170                 goto fail_sb;
1171
1172         /*
1173          * If user space has failed to join the cluster or some similar
1174          * failure has occurred, then the journal id will contain a
1175          * negative (error) number. This will then be returned to the
1176          * caller (of the mount syscall). We do this even for spectator
1177          * mounts (which just write a jid of 0 to indicate "ok" even though
1178          * the jid is unused in the spectator case)
1179          */
1180         if (sdp->sd_lockstruct.ls_jid < 0) {
1181                 error = sdp->sd_lockstruct.ls_jid;
1182                 sdp->sd_lockstruct.ls_jid = 0;
1183                 goto fail_sb;
1184         }
1185
1186         if (sdp->sd_args.ar_spectator)
1187                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1188                          sdp->sd_table_name);
1189         else
1190                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1191                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1192
1193         error = init_inodes(sdp, DO);
1194         if (error)
1195                 goto fail_sb;
1196
1197         error = init_per_node(sdp, DO);
1198         if (error)
1199                 goto fail_inodes;
1200
1201         error = gfs2_statfs_init(sdp);
1202         if (error) {
1203                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1204                 goto fail_per_node;
1205         }
1206
1207         error = init_threads(sdp, DO);
1208         if (error)
1209                 goto fail_per_node;
1210
1211         if (!(sb->s_flags & MS_RDONLY)) {
1212                 error = gfs2_make_fs_rw(sdp);
1213                 if (error) {
1214                         fs_err(sdp, "can't make FS RW: %d\n", error);
1215                         goto fail_threads;
1216                 }
1217         }
1218
1219         gfs2_glock_dq_uninit(&mount_gh);
1220         gfs2_online_uevent(sdp);
1221         return 0;
1222
1223 fail_threads:
1224         init_threads(sdp, UNDO);
1225 fail_per_node:
1226         init_per_node(sdp, UNDO);
1227 fail_inodes:
1228         init_inodes(sdp, UNDO);
1229 fail_sb:
1230         if (sdp->sd_root_dir)
1231                 dput(sdp->sd_root_dir);
1232         if (sdp->sd_master_dir)
1233                 dput(sdp->sd_master_dir);
1234         if (sb->s_root)
1235                 dput(sb->s_root);
1236         sb->s_root = NULL;
1237 fail_locking:
1238         init_locking(sdp, &mount_gh, UNDO);
1239 fail_lm:
1240         gfs2_gl_hash_clear(sdp);
1241         gfs2_lm_unmount(sdp);
1242 fail_debug:
1243         gfs2_delete_debugfs_file(sdp);
1244         free_percpu(sdp->sd_lkstats);
1245         /* gfs2_sys_fs_del must be the last thing we do, since it causes
1246          * sysfs to call function gfs2_sbd_release, which frees sdp. */
1247         gfs2_sys_fs_del(sdp);
1248         sb->s_fs_info = NULL;
1249         return error;
1250 }
1251
1252 static int set_gfs2_super(struct super_block *s, void *data)
1253 {
1254         s->s_bdev = data;
1255         s->s_dev = s->s_bdev->bd_dev;
1256
1257         /*
1258          * We set the bdi here to the queue backing, file systems can
1259          * overwrite this in ->fill_super()
1260          */
1261         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1262         return 0;
1263 }
1264
1265 static int test_gfs2_super(struct super_block *s, void *ptr)
1266 {
1267         struct block_device *bdev = ptr;
1268         return (bdev == s->s_bdev);
1269 }
1270
1271 /**
1272  * gfs2_mount - Get the GFS2 superblock
1273  * @fs_type: The GFS2 filesystem type
1274  * @flags: Mount flags
1275  * @dev_name: The name of the device
1276  * @data: The mount arguments
1277  *
1278  * Q. Why not use get_sb_bdev() ?
1279  * A. We need to select one of two root directories to mount, independent
1280  *    of whether this is the initial, or subsequent, mount of this sb
1281  *
1282  * Returns: 0 or -ve on error
1283  */
1284
1285 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1286                        const char *dev_name, void *data)
1287 {
1288         struct block_device *bdev;
1289         struct super_block *s;
1290         fmode_t mode = FMODE_READ | FMODE_EXCL;
1291         int error;
1292         struct gfs2_args args;
1293         struct gfs2_sbd *sdp;
1294
1295         if (!(flags & MS_RDONLY))
1296                 mode |= FMODE_WRITE;
1297
1298         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1299         if (IS_ERR(bdev))
1300                 return ERR_CAST(bdev);
1301
1302         /*
1303          * once the super is inserted into the list by sget, s_umount
1304          * will protect the lockfs code from trying to start a snapshot
1305          * while we are mounting
1306          */
1307         mutex_lock(&bdev->bd_fsfreeze_mutex);
1308         if (bdev->bd_fsfreeze_count > 0) {
1309                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1310                 error = -EBUSY;
1311                 goto error_bdev;
1312         }
1313         s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1314         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1315         error = PTR_ERR(s);
1316         if (IS_ERR(s))
1317                 goto error_bdev;
1318
1319         if (s->s_root)
1320                 blkdev_put(bdev, mode);
1321
1322         memset(&args, 0, sizeof(args));
1323         args.ar_quota = GFS2_QUOTA_DEFAULT;
1324         args.ar_data = GFS2_DATA_DEFAULT;
1325         args.ar_commit = 30;
1326         args.ar_statfs_quantum = 30;
1327         args.ar_quota_quantum = 60;
1328         args.ar_errors = GFS2_ERRORS_DEFAULT;
1329
1330         error = gfs2_mount_args(&args, data);
1331         if (error) {
1332                 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1333                 goto error_super;
1334         }
1335
1336         if (s->s_root) {
1337                 error = -EBUSY;
1338                 if ((flags ^ s->s_flags) & MS_RDONLY)
1339                         goto error_super;
1340         } else {
1341                 char b[BDEVNAME_SIZE];
1342
1343                 s->s_mode = mode;
1344                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1345                 sb_set_blocksize(s, block_size(bdev));
1346                 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1347                 if (error)
1348                         goto error_super;
1349                 s->s_flags |= MS_ACTIVE;
1350                 bdev->bd_super = s;
1351         }
1352
1353         sdp = s->s_fs_info;
1354         if (args.ar_meta)
1355                 return dget(sdp->sd_master_dir);
1356         else
1357                 return dget(sdp->sd_root_dir);
1358
1359 error_super:
1360         deactivate_locked_super(s);
1361         return ERR_PTR(error);
1362 error_bdev:
1363         blkdev_put(bdev, mode);
1364         return ERR_PTR(error);
1365 }
1366
1367 static int set_meta_super(struct super_block *s, void *ptr)
1368 {
1369         return -EINVAL;
1370 }
1371
1372 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1373                         int flags, const char *dev_name, void *data)
1374 {
1375         struct super_block *s;
1376         struct gfs2_sbd *sdp;
1377         struct path path;
1378         int error;
1379
1380         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1381         if (error) {
1382                 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1383                        dev_name, error);
1384                 return ERR_PTR(error);
1385         }
1386         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1387                  path.dentry->d_inode->i_sb->s_bdev);
1388         path_put(&path);
1389         if (IS_ERR(s)) {
1390                 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1391                 return ERR_CAST(s);
1392         }
1393         if ((flags ^ s->s_flags) & MS_RDONLY) {
1394                 deactivate_locked_super(s);
1395                 return ERR_PTR(-EBUSY);
1396         }
1397         sdp = s->s_fs_info;
1398         return dget(sdp->sd_master_dir);
1399 }
1400
1401 static void gfs2_kill_sb(struct super_block *sb)
1402 {
1403         struct gfs2_sbd *sdp = sb->s_fs_info;
1404
1405         if (sdp == NULL) {
1406                 kill_block_super(sb);
1407                 return;
1408         }
1409
1410         gfs2_meta_syncfs(sdp);
1411         dput(sdp->sd_root_dir);
1412         dput(sdp->sd_master_dir);
1413         sdp->sd_root_dir = NULL;
1414         sdp->sd_master_dir = NULL;
1415         shrink_dcache_sb(sb);
1416         gfs2_delete_debugfs_file(sdp);
1417         free_percpu(sdp->sd_lkstats);
1418         kill_block_super(sb);
1419 }
1420
1421 struct file_system_type gfs2_fs_type = {
1422         .name = "gfs2",
1423         .fs_flags = FS_REQUIRES_DEV,
1424         .mount = gfs2_mount,
1425         .kill_sb = gfs2_kill_sb,
1426         .owner = THIS_MODULE,
1427 };
1428
1429 struct file_system_type gfs2meta_fs_type = {
1430         .name = "gfs2meta",
1431         .fs_flags = FS_REQUIRES_DEV,
1432         .mount = gfs2_mount_meta,
1433         .owner = THIS_MODULE,
1434 };
1435