d3c69eb91c74e29c05af120f3938bffb9ebb1d75
[linux-2.6.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/namei.h>
18 #include <linux/mount.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/quotaops.h>
21
22 #include "gfs2.h"
23 #include "incore.h"
24 #include "bmap.h"
25 #include "glock.h"
26 #include "glops.h"
27 #include "inode.h"
28 #include "recovery.h"
29 #include "rgrp.h"
30 #include "super.h"
31 #include "sys.h"
32 #include "util.h"
33 #include "log.h"
34 #include "quota.h"
35 #include "dir.h"
36 #include "trace_gfs2.h"
37
38 #define DO 0
39 #define UNDO 1
40
41 /**
42  * gfs2_tune_init - Fill a gfs2_tune structure with default values
43  * @gt: tune
44  *
45  */
46
47 static void gfs2_tune_init(struct gfs2_tune *gt)
48 {
49         spin_lock_init(&gt->gt_spin);
50
51         gt->gt_quota_simul_sync = 64;
52         gt->gt_quota_warn_period = 10;
53         gt->gt_quota_scale_num = 1;
54         gt->gt_quota_scale_den = 1;
55         gt->gt_new_files_jdata = 0;
56         gt->gt_max_readahead = 1 << 18;
57         gt->gt_complain_secs = 10;
58 }
59
60 static struct gfs2_sbd *init_sbd(struct super_block *sb)
61 {
62         struct gfs2_sbd *sdp;
63
64         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
65         if (!sdp)
66                 return NULL;
67
68         sb->s_fs_info = sdp;
69         sdp->sd_vfs = sb;
70         set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
71         gfs2_tune_init(&sdp->sd_tune);
72
73         init_waitqueue_head(&sdp->sd_glock_wait);
74         atomic_set(&sdp->sd_glock_disposal, 0);
75         spin_lock_init(&sdp->sd_statfs_spin);
76
77         spin_lock_init(&sdp->sd_rindex_spin);
78         mutex_init(&sdp->sd_rindex_mutex);
79         INIT_LIST_HEAD(&sdp->sd_rindex_list);
80         INIT_LIST_HEAD(&sdp->sd_rindex_mru_list);
81
82         INIT_LIST_HEAD(&sdp->sd_jindex_list);
83         spin_lock_init(&sdp->sd_jindex_spin);
84         mutex_init(&sdp->sd_jindex_mutex);
85
86         INIT_LIST_HEAD(&sdp->sd_quota_list);
87         mutex_init(&sdp->sd_quota_mutex);
88         init_waitqueue_head(&sdp->sd_quota_wait);
89         INIT_LIST_HEAD(&sdp->sd_trunc_list);
90         spin_lock_init(&sdp->sd_trunc_lock);
91
92         spin_lock_init(&sdp->sd_log_lock);
93         atomic_set(&sdp->sd_log_pinned, 0);
94         INIT_LIST_HEAD(&sdp->sd_log_le_buf);
95         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
96         INIT_LIST_HEAD(&sdp->sd_log_le_rg);
97         INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
98         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
99
100         init_waitqueue_head(&sdp->sd_log_waitq);
101         init_waitqueue_head(&sdp->sd_logd_waitq);
102         spin_lock_init(&sdp->sd_ail_lock);
103         INIT_LIST_HEAD(&sdp->sd_ail1_list);
104         INIT_LIST_HEAD(&sdp->sd_ail2_list);
105
106         init_rwsem(&sdp->sd_log_flush_lock);
107         atomic_set(&sdp->sd_log_in_flight, 0);
108         init_waitqueue_head(&sdp->sd_log_flush_wait);
109
110         INIT_LIST_HEAD(&sdp->sd_revoke_list);
111
112         mutex_init(&sdp->sd_freeze_lock);
113
114         return sdp;
115 }
116
117
118 /**
119  * gfs2_check_sb - Check superblock
120  * @sdp: the filesystem
121  * @sb: The superblock
122  * @silent: Don't print a message if the check fails
123  *
124  * Checks the version code of the FS is one that we understand how to
125  * read and that the sizes of the various on-disk structures have not
126  * changed.
127  */
128
129 static int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
130 {
131         if (sb->sb_magic != GFS2_MAGIC ||
132             sb->sb_type != GFS2_METATYPE_SB) {
133                 if (!silent)
134                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
135                 return -EINVAL;
136         }
137
138         /*  If format numbers match exactly, we're done.  */
139
140         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
141             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
142                 return 0;
143
144         fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
145
146         return -EINVAL;
147 }
148
149 static void end_bio_io_page(struct bio *bio, int error)
150 {
151         struct page *page = bio->bi_private;
152
153         if (!error)
154                 SetPageUptodate(page);
155         else
156                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
157         unlock_page(page);
158 }
159
160 static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
161 {
162         const struct gfs2_sb *str = buf;
163
164         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
165         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
166         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
167         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
168         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
169         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
170         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
171         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
172         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
173         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
174         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
175
176         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
177         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
178         memcpy(sb->sb_uuid, str->sb_uuid, 16);
179 }
180
181 /**
182  * gfs2_read_super - Read the gfs2 super block from disk
183  * @sdp: The GFS2 super block
184  * @sector: The location of the super block
185  * @error: The error code to return
186  *
187  * This uses the bio functions to read the super block from disk
188  * because we want to be 100% sure that we never read cached data.
189  * A super block is read twice only during each GFS2 mount and is
190  * never written to by the filesystem. The first time its read no
191  * locks are held, and the only details which are looked at are those
192  * relating to the locking protocol. Once locking is up and working,
193  * the sb is read again under the lock to establish the location of
194  * the master directory (contains pointers to journals etc) and the
195  * root directory.
196  *
197  * Returns: 0 on success or error
198  */
199
200 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
201 {
202         struct super_block *sb = sdp->sd_vfs;
203         struct gfs2_sb *p;
204         struct page *page;
205         struct bio *bio;
206
207         page = alloc_page(GFP_NOFS);
208         if (unlikely(!page))
209                 return -ENOBUFS;
210
211         ClearPageUptodate(page);
212         ClearPageDirty(page);
213         lock_page(page);
214
215         bio = bio_alloc(GFP_NOFS, 1);
216         bio->bi_sector = sector * (sb->s_blocksize >> 9);
217         bio->bi_bdev = sb->s_bdev;
218         bio_add_page(bio, page, PAGE_SIZE, 0);
219
220         bio->bi_end_io = end_bio_io_page;
221         bio->bi_private = page;
222         submit_bio(READ_SYNC | REQ_META, bio);
223         wait_on_page_locked(page);
224         bio_put(bio);
225         if (!PageUptodate(page)) {
226                 __free_page(page);
227                 return -EIO;
228         }
229         p = kmap(page);
230         gfs2_sb_in(&sdp->sd_sb, p);
231         kunmap(page);
232         __free_page(page);
233         return 0;
234 }
235
236 /**
237  * gfs2_read_sb - Read super block
238  * @sdp: The GFS2 superblock
239  * @silent: Don't print message if mount fails
240  *
241  */
242
243 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
244 {
245         u32 hash_blocks, ind_blocks, leaf_blocks;
246         u32 tmp_blocks;
247         unsigned int x;
248         int error;
249
250         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
251         if (error) {
252                 if (!silent)
253                         fs_err(sdp, "can't read superblock\n");
254                 return error;
255         }
256
257         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
258         if (error)
259                 return error;
260
261         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
262                                GFS2_BASIC_BLOCK_SHIFT;
263         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
264         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
265                           sizeof(struct gfs2_dinode)) / sizeof(u64);
266         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
267                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
268         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
269         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
270         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
271         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
272         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
273                                 sizeof(struct gfs2_meta_header)) /
274                                 sizeof(struct gfs2_quota_change);
275
276         /* Compute maximum reservation required to add a entry to a directory */
277
278         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
279                              sdp->sd_jbsize);
280
281         ind_blocks = 0;
282         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
283                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
284                 ind_blocks += tmp_blocks;
285         }
286
287         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
288
289         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
290
291         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
292                                 sizeof(struct gfs2_dinode);
293         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
294         for (x = 2;; x++) {
295                 u64 space, d;
296                 u32 m;
297
298                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
299                 d = space;
300                 m = do_div(d, sdp->sd_inptrs);
301
302                 if (d != sdp->sd_heightsize[x - 1] || m)
303                         break;
304                 sdp->sd_heightsize[x] = space;
305         }
306         sdp->sd_max_height = x;
307         sdp->sd_heightsize[x] = ~0;
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         sdp->sd_jheightsize[x] = ~0;
327         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
328
329         return 0;
330 }
331
332 static int init_names(struct gfs2_sbd *sdp, int silent)
333 {
334         char *proto, *table;
335         int error = 0;
336
337         proto = sdp->sd_args.ar_lockproto;
338         table = sdp->sd_args.ar_locktable;
339
340         /*  Try to autodetect  */
341
342         if (!proto[0] || !table[0]) {
343                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
344                 if (error)
345                         return error;
346
347                 error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
348                 if (error)
349                         goto out;
350
351                 if (!proto[0])
352                         proto = sdp->sd_sb.sb_lockproto;
353                 if (!table[0])
354                         table = sdp->sd_sb.sb_locktable;
355         }
356
357         if (!table[0])
358                 table = sdp->sd_vfs->s_id;
359
360         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
361         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
362
363         table = sdp->sd_table_name;
364         while ((table = strchr(table, '/')))
365                 *table = '_';
366
367 out:
368         return error;
369 }
370
371 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
372                         int undo)
373 {
374         int error = 0;
375
376         if (undo)
377                 goto fail_trans;
378
379         error = gfs2_glock_nq_num(sdp,
380                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
381                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
382                                   mount_gh);
383         if (error) {
384                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
385                 goto fail;
386         }
387
388         error = gfs2_glock_nq_num(sdp,
389                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
390                                   LM_ST_SHARED,
391                                   LM_FLAG_NOEXP | GL_EXACT,
392                                   &sdp->sd_live_gh);
393         if (error) {
394                 fs_err(sdp, "can't acquire live glock: %d\n", error);
395                 goto fail_mount;
396         }
397
398         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
399                                CREATE, &sdp->sd_rename_gl);
400         if (error) {
401                 fs_err(sdp, "can't create rename glock: %d\n", error);
402                 goto fail_live;
403         }
404
405         error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
406                                CREATE, &sdp->sd_trans_gl);
407         if (error) {
408                 fs_err(sdp, "can't create transaction glock: %d\n", error);
409                 goto fail_rename;
410         }
411
412         return 0;
413
414 fail_trans:
415         gfs2_glock_put(sdp->sd_trans_gl);
416 fail_rename:
417         gfs2_glock_put(sdp->sd_rename_gl);
418 fail_live:
419         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
420 fail_mount:
421         gfs2_glock_dq_uninit(mount_gh);
422 fail:
423         return error;
424 }
425
426 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
427                             u64 no_addr, const char *name)
428 {
429         struct gfs2_sbd *sdp = sb->s_fs_info;
430         struct dentry *dentry;
431         struct inode *inode;
432
433         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
434         if (IS_ERR(inode)) {
435                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
436                 return PTR_ERR(inode);
437         }
438         dentry = d_alloc_root(inode);
439         if (!dentry) {
440                 fs_err(sdp, "can't alloc %s dentry\n", name);
441                 iput(inode);
442                 return -ENOMEM;
443         }
444         *dptr = dentry;
445         return 0;
446 }
447
448 static int init_sb(struct gfs2_sbd *sdp, int silent)
449 {
450         struct super_block *sb = sdp->sd_vfs;
451         struct gfs2_holder sb_gh;
452         u64 no_addr;
453         int ret;
454
455         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
456                                 LM_ST_SHARED, 0, &sb_gh);
457         if (ret) {
458                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
459                 return ret;
460         }
461
462         ret = gfs2_read_sb(sdp, silent);
463         if (ret) {
464                 fs_err(sdp, "can't read superblock: %d\n", ret);
465                 goto out;
466         }
467
468         /* Set up the buffer cache and SB for real */
469         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
470                 ret = -EINVAL;
471                 fs_err(sdp, "FS block size (%u) is too small for device "
472                        "block size (%u)\n",
473                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
474                 goto out;
475         }
476         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
477                 ret = -EINVAL;
478                 fs_err(sdp, "FS block size (%u) is too big for machine "
479                        "page size (%u)\n",
480                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
481                 goto out;
482         }
483         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
484
485         /* Get the root inode */
486         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
487         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
488         if (ret)
489                 goto out;
490
491         /* Get the master inode */
492         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
493         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
494         if (ret) {
495                 dput(sdp->sd_root_dir);
496                 goto out;
497         }
498         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
499 out:
500         gfs2_glock_dq_uninit(&sb_gh);
501         return ret;
502 }
503
504 /**
505  * map_journal_extents - create a reusable "extent" mapping from all logical
506  * blocks to all physical blocks for the given journal.  This will save
507  * us time when writing journal blocks.  Most journals will have only one
508  * extent that maps all their logical blocks.  That's because gfs2.mkfs
509  * arranges the journal blocks sequentially to maximize performance.
510  * So the extent would map the first block for the entire file length.
511  * However, gfs2_jadd can happen while file activity is happening, so
512  * those journals may not be sequential.  Less likely is the case where
513  * the users created their own journals by mounting the metafs and
514  * laying it out.  But it's still possible.  These journals might have
515  * several extents.
516  *
517  * TODO: This should be done in bigger chunks rather than one block at a time,
518  *       but since it's only done at mount time, I'm not worried about the
519  *       time it takes.
520  */
521 static int map_journal_extents(struct gfs2_sbd *sdp)
522 {
523         struct gfs2_jdesc *jd = sdp->sd_jdesc;
524         unsigned int lb;
525         u64 db, prev_db; /* logical block, disk block, prev disk block */
526         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
527         struct gfs2_journal_extent *jext = NULL;
528         struct buffer_head bh;
529         int rc = 0;
530
531         prev_db = 0;
532
533         for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
534                 bh.b_state = 0;
535                 bh.b_blocknr = 0;
536                 bh.b_size = 1 << ip->i_inode.i_blkbits;
537                 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
538                 db = bh.b_blocknr;
539                 if (rc || !db) {
540                         printk(KERN_INFO "GFS2 journal mapping error %d: lb="
541                                "%u db=%llu\n", rc, lb, (unsigned long long)db);
542                         break;
543                 }
544                 if (!prev_db || db != prev_db + 1) {
545                         jext = kzalloc(sizeof(struct gfs2_journal_extent),
546                                        GFP_KERNEL);
547                         if (!jext) {
548                                 printk(KERN_INFO "GFS2 error: out of memory "
549                                        "mapping journal extents.\n");
550                                 rc = -ENOMEM;
551                                 break;
552                         }
553                         jext->dblock = db;
554                         jext->lblock = lb;
555                         jext->blocks = 1;
556                         list_add_tail(&jext->extent_list, &jd->extent_list);
557                 } else {
558                         jext->blocks++;
559                 }
560                 prev_db = db;
561         }
562         return rc;
563 }
564
565 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
566 {
567         char *message = "FIRSTMOUNT=Done";
568         char *envp[] = { message, NULL };
569         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
570         ls->ls_first_done = 1;
571         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
572 }
573
574 /**
575  * gfs2_jindex_hold - Grab a lock on the jindex
576  * @sdp: The GFS2 superblock
577  * @ji_gh: the holder for the jindex glock
578  *
579  * Returns: errno
580  */
581
582 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
583 {
584         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
585         struct qstr name;
586         char buf[20];
587         struct gfs2_jdesc *jd;
588         int error;
589
590         name.name = buf;
591
592         mutex_lock(&sdp->sd_jindex_mutex);
593
594         for (;;) {
595                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
596                 if (error)
597                         break;
598
599                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
600                 name.hash = gfs2_disk_hash(name.name, name.len);
601
602                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
603                 if (error == -ENOENT) {
604                         error = 0;
605                         break;
606                 }
607
608                 gfs2_glock_dq_uninit(ji_gh);
609
610                 if (error)
611                         break;
612
613                 error = -ENOMEM;
614                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
615                 if (!jd)
616                         break;
617
618                 INIT_LIST_HEAD(&jd->extent_list);
619                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
620                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
621                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
622                         if (!jd->jd_inode)
623                                 error = -ENOENT;
624                         else
625                                 error = PTR_ERR(jd->jd_inode);
626                         kfree(jd);
627                         break;
628                 }
629
630                 spin_lock(&sdp->sd_jindex_spin);
631                 jd->jd_jid = sdp->sd_journals++;
632                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
633                 spin_unlock(&sdp->sd_jindex_spin);
634         }
635
636         mutex_unlock(&sdp->sd_jindex_mutex);
637
638         return error;
639 }
640
641 static int init_journal(struct gfs2_sbd *sdp, int undo)
642 {
643         struct inode *master = sdp->sd_master_dir->d_inode;
644         struct gfs2_holder ji_gh;
645         struct gfs2_inode *ip;
646         int jindex = 1;
647         int error = 0;
648
649         if (undo) {
650                 jindex = 0;
651                 goto fail_jinode_gh;
652         }
653
654         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
655         if (IS_ERR(sdp->sd_jindex)) {
656                 fs_err(sdp, "can't lookup journal index: %d\n", error);
657                 return PTR_ERR(sdp->sd_jindex);
658         }
659         ip = GFS2_I(sdp->sd_jindex);
660
661         /* Load in the journal index special file */
662
663         error = gfs2_jindex_hold(sdp, &ji_gh);
664         if (error) {
665                 fs_err(sdp, "can't read journal index: %d\n", error);
666                 goto fail;
667         }
668
669         error = -EUSERS;
670         if (!gfs2_jindex_size(sdp)) {
671                 fs_err(sdp, "no journals!\n");
672                 goto fail_jindex;
673         }
674
675         if (sdp->sd_args.ar_spectator) {
676                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
677                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
678                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
679                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
680         } else {
681                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
682                         fs_err(sdp, "can't mount journal #%u\n",
683                                sdp->sd_lockstruct.ls_jid);
684                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
685                                gfs2_jindex_size(sdp),
686                                gfs2_jindex_size(sdp) - 1);
687                         goto fail_jindex;
688                 }
689                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
690
691                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
692                                           &gfs2_journal_glops,
693                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
694                                           &sdp->sd_journal_gh);
695                 if (error) {
696                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
697                         goto fail_jindex;
698                 }
699
700                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
701                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
702                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
703                                            &sdp->sd_jinode_gh);
704                 if (error) {
705                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
706                                error);
707                         goto fail_journal_gh;
708                 }
709
710                 error = gfs2_jdesc_check(sdp->sd_jdesc);
711                 if (error) {
712                         fs_err(sdp, "my journal (%u) is bad: %d\n",
713                                sdp->sd_jdesc->jd_jid, error);
714                         goto fail_jinode_gh;
715                 }
716                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
717                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
718                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
719
720                 /* Map the extents for this journal's blocks */
721                 map_journal_extents(sdp);
722         }
723         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
724
725         if (sdp->sd_lockstruct.ls_first) {
726                 unsigned int x;
727                 for (x = 0; x < sdp->sd_journals; x++) {
728                         error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x),
729                                                      true);
730                         if (error) {
731                                 fs_err(sdp, "error recovering journal %u: %d\n",
732                                        x, error);
733                                 goto fail_jinode_gh;
734                         }
735                 }
736
737                 gfs2_others_may_mount(sdp);
738         } else if (!sdp->sd_args.ar_spectator) {
739                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
740                 if (error) {
741                         fs_err(sdp, "error recovering my journal: %d\n", error);
742                         goto fail_jinode_gh;
743                 }
744         }
745
746         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
747         gfs2_glock_dq_uninit(&ji_gh);
748         jindex = 0;
749
750         return 0;
751
752 fail_jinode_gh:
753         if (!sdp->sd_args.ar_spectator)
754                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
755 fail_journal_gh:
756         if (!sdp->sd_args.ar_spectator)
757                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
758 fail_jindex:
759         gfs2_jindex_free(sdp);
760         if (jindex)
761                 gfs2_glock_dq_uninit(&ji_gh);
762 fail:
763         iput(sdp->sd_jindex);
764         return error;
765 }
766
767
768 static int init_inodes(struct gfs2_sbd *sdp, int undo)
769 {
770         int error = 0;
771         struct gfs2_inode *ip;
772         struct inode *master = sdp->sd_master_dir->d_inode;
773
774         if (undo)
775                 goto fail_qinode;
776
777         error = init_journal(sdp, undo);
778         if (error)
779                 goto fail;
780
781         /* Read in the master statfs inode */
782         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
783         if (IS_ERR(sdp->sd_statfs_inode)) {
784                 error = PTR_ERR(sdp->sd_statfs_inode);
785                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
786                 goto fail_journal;
787         }
788
789         /* Read in the resource index inode */
790         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
791         if (IS_ERR(sdp->sd_rindex)) {
792                 error = PTR_ERR(sdp->sd_rindex);
793                 fs_err(sdp, "can't get resource index inode: %d\n", error);
794                 goto fail_statfs;
795         }
796         ip = GFS2_I(sdp->sd_rindex);
797         sdp->sd_rindex_uptodate = 0;
798
799         /* Read in the quota inode */
800         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
801         if (IS_ERR(sdp->sd_quota_inode)) {
802                 error = PTR_ERR(sdp->sd_quota_inode);
803                 fs_err(sdp, "can't get quota file inode: %d\n", error);
804                 goto fail_rindex;
805         }
806         return 0;
807
808 fail_qinode:
809         iput(sdp->sd_quota_inode);
810 fail_rindex:
811         gfs2_clear_rgrpd(sdp);
812         iput(sdp->sd_rindex);
813 fail_statfs:
814         iput(sdp->sd_statfs_inode);
815 fail_journal:
816         init_journal(sdp, UNDO);
817 fail:
818         return error;
819 }
820
821 static int init_per_node(struct gfs2_sbd *sdp, int undo)
822 {
823         struct inode *pn = NULL;
824         char buf[30];
825         int error = 0;
826         struct gfs2_inode *ip;
827         struct inode *master = sdp->sd_master_dir->d_inode;
828
829         if (sdp->sd_args.ar_spectator)
830                 return 0;
831
832         if (undo)
833                 goto fail_qc_gh;
834
835         pn = gfs2_lookup_simple(master, "per_node");
836         if (IS_ERR(pn)) {
837                 error = PTR_ERR(pn);
838                 fs_err(sdp, "can't find per_node directory: %d\n", error);
839                 return error;
840         }
841
842         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
843         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
844         if (IS_ERR(sdp->sd_sc_inode)) {
845                 error = PTR_ERR(sdp->sd_sc_inode);
846                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
847                 goto fail;
848         }
849
850         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
851         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
852         if (IS_ERR(sdp->sd_qc_inode)) {
853                 error = PTR_ERR(sdp->sd_qc_inode);
854                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
855                 goto fail_ut_i;
856         }
857
858         iput(pn);
859         pn = NULL;
860
861         ip = GFS2_I(sdp->sd_sc_inode);
862         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
863                                    &sdp->sd_sc_gh);
864         if (error) {
865                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
866                 goto fail_qc_i;
867         }
868
869         ip = GFS2_I(sdp->sd_qc_inode);
870         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
871                                    &sdp->sd_qc_gh);
872         if (error) {
873                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
874                 goto fail_ut_gh;
875         }
876
877         return 0;
878
879 fail_qc_gh:
880         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
881 fail_ut_gh:
882         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
883 fail_qc_i:
884         iput(sdp->sd_qc_inode);
885 fail_ut_i:
886         iput(sdp->sd_sc_inode);
887 fail:
888         if (pn)
889                 iput(pn);
890         return error;
891 }
892
893 static int init_threads(struct gfs2_sbd *sdp, int undo)
894 {
895         struct task_struct *p;
896         int error = 0;
897
898         if (undo)
899                 goto fail_quotad;
900
901         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
902         error = IS_ERR(p);
903         if (error) {
904                 fs_err(sdp, "can't start logd thread: %d\n", error);
905                 return error;
906         }
907         sdp->sd_logd_process = p;
908
909         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
910         error = IS_ERR(p);
911         if (error) {
912                 fs_err(sdp, "can't start quotad thread: %d\n", error);
913                 goto fail;
914         }
915         sdp->sd_quotad_process = p;
916
917         return 0;
918
919
920 fail_quotad:
921         kthread_stop(sdp->sd_quotad_process);
922 fail:
923         kthread_stop(sdp->sd_logd_process);
924         return error;
925 }
926
927 static const match_table_t nolock_tokens = {
928         { Opt_jid, "jid=%d\n", },
929         { Opt_err, NULL },
930 };
931
932 static const struct lm_lockops nolock_ops = {
933         .lm_proto_name = "lock_nolock",
934         .lm_put_lock = gfs2_glock_free,
935         .lm_tokens = &nolock_tokens,
936 };
937
938 /**
939  * gfs2_lm_mount - mount a locking protocol
940  * @sdp: the filesystem
941  * @args: mount arguments
942  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
943  *
944  * Returns: errno
945  */
946
947 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
948 {
949         const struct lm_lockops *lm;
950         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
951         struct gfs2_args *args = &sdp->sd_args;
952         const char *proto = sdp->sd_proto_name;
953         const char *table = sdp->sd_table_name;
954         const char *fsname;
955         char *o, *options;
956         int ret;
957
958         if (!strcmp("lock_nolock", proto)) {
959                 lm = &nolock_ops;
960                 sdp->sd_args.ar_localflocks = 1;
961 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
962         } else if (!strcmp("lock_dlm", proto)) {
963                 lm = &gfs2_dlm_ops;
964 #endif
965         } else {
966                 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
967                 return -ENOENT;
968         }
969
970         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
971
972         ls->ls_ops = lm;
973         ls->ls_first = 1;
974
975         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
976                 substring_t tmp[MAX_OPT_ARGS];
977                 int token, option;
978
979                 if (!o || !*o)
980                         continue;
981
982                 token = match_token(o, *lm->lm_tokens, tmp);
983                 switch (token) {
984                 case Opt_jid:
985                         ret = match_int(&tmp[0], &option);
986                         if (ret || option < 0) 
987                                 goto hostdata_error;
988                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
989                                 ls->ls_jid = option;
990                         break;
991                 case Opt_id:
992                         /* Obsolete, but left for backward compat purposes */
993                         break;
994                 case Opt_first:
995                         ret = match_int(&tmp[0], &option);
996                         if (ret || (option != 0 && option != 1))
997                                 goto hostdata_error;
998                         ls->ls_first = option;
999                         break;
1000                 case Opt_nodir:
1001                         ret = match_int(&tmp[0], &option);
1002                         if (ret || (option != 0 && option != 1))
1003                                 goto hostdata_error;
1004                         ls->ls_nodir = option;
1005                         break;
1006                 case Opt_err:
1007                 default:
1008 hostdata_error:
1009                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1010                         return -EINVAL;
1011                 }
1012         }
1013
1014         if (sdp->sd_args.ar_spectator)
1015                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s", table);
1016         else
1017                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u", table,
1018                          sdp->sd_lockstruct.ls_jid);
1019
1020         fsname = strchr(table, ':');
1021         if (fsname)
1022                 fsname++;
1023         if (lm->lm_mount == NULL) {
1024                 fs_info(sdp, "Now mounting FS...\n");
1025                 return 0;
1026         }
1027         ret = lm->lm_mount(sdp, fsname);
1028         if (ret == 0)
1029                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1030         return ret;
1031 }
1032
1033 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1034 {
1035         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1036         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1037             lm->lm_unmount)
1038                 lm->lm_unmount(sdp);
1039 }
1040
1041 static int gfs2_journalid_wait(void *word)
1042 {
1043         if (signal_pending(current))
1044                 return -EINTR;
1045         schedule();
1046         return 0;
1047 }
1048
1049 static int wait_on_journal(struct gfs2_sbd *sdp)
1050 {
1051         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1052                 return 0;
1053
1054         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1055 }
1056
1057 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1058 {
1059         struct super_block *sb = sdp->sd_vfs;
1060         char ro[20];
1061         char spectator[20];
1062         char *envp[] = { ro, spectator, NULL };
1063         sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1064         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1065         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1066 }
1067
1068 /**
1069  * fill_super - Read in superblock
1070  * @sb: The VFS superblock
1071  * @data: Mount options
1072  * @silent: Don't complain if it's not a GFS2 filesystem
1073  *
1074  * Returns: errno
1075  */
1076
1077 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1078 {
1079         struct gfs2_sbd *sdp;
1080         struct gfs2_holder mount_gh;
1081         int error;
1082
1083         sdp = init_sbd(sb);
1084         if (!sdp) {
1085                 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1086                 return -ENOMEM;
1087         }
1088         sdp->sd_args = *args;
1089
1090         if (sdp->sd_args.ar_spectator) {
1091                 sb->s_flags |= MS_RDONLY;
1092                 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
1093         }
1094         if (sdp->sd_args.ar_posix_acl)
1095                 sb->s_flags |= MS_POSIXACL;
1096         if (sdp->sd_args.ar_nobarrier)
1097                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1098
1099         sb->s_magic = GFS2_MAGIC;
1100         sb->s_op = &gfs2_super_ops;
1101         sb->s_d_op = &gfs2_dops;
1102         sb->s_export_op = &gfs2_export_ops;
1103         sb->s_xattr = gfs2_xattr_handlers;
1104         sb->s_qcop = &gfs2_quotactl_ops;
1105         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1106         sb->s_time_gran = 1;
1107         sb->s_maxbytes = MAX_LFS_FILESIZE;
1108
1109         /* Set up the buffer cache and fill in some fake block size values
1110            to allow us to read-in the on-disk superblock. */
1111         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1112         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1113         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1114                                GFS2_BASIC_BLOCK_SHIFT;
1115         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1116
1117         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1118         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1119         if (sdp->sd_args.ar_statfs_quantum) {
1120                 sdp->sd_tune.gt_statfs_slow = 0;
1121                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1122         }
1123         else {
1124                 sdp->sd_tune.gt_statfs_slow = 1;
1125                 sdp->sd_tune.gt_statfs_quantum = 30;
1126         }
1127
1128         error = init_names(sdp, silent);
1129         if (error)
1130                 goto fail;
1131
1132         gfs2_create_debugfs_file(sdp);
1133
1134         error = gfs2_sys_fs_add(sdp);
1135         if (error)
1136                 goto fail;
1137
1138         error = gfs2_lm_mount(sdp, silent);
1139         if (error)
1140                 goto fail_sys;
1141
1142         error = init_locking(sdp, &mount_gh, DO);
1143         if (error)
1144                 goto fail_lm;
1145
1146         error = init_sb(sdp, silent);
1147         if (error)
1148                 goto fail_locking;
1149
1150         error = wait_on_journal(sdp);
1151         if (error)
1152                 goto fail_sb;
1153
1154         /*
1155          * If user space has failed to join the cluster or some similar
1156          * failure has occurred, then the journal id will contain a
1157          * negative (error) number. This will then be returned to the
1158          * caller (of the mount syscall). We do this even for spectator
1159          * mounts (which just write a jid of 0 to indicate "ok" even though
1160          * the jid is unused in the spectator case)
1161          */
1162         if (sdp->sd_lockstruct.ls_jid < 0) {
1163                 error = sdp->sd_lockstruct.ls_jid;
1164                 sdp->sd_lockstruct.ls_jid = 0;
1165                 goto fail_sb;
1166         }
1167
1168         error = init_inodes(sdp, DO);
1169         if (error)
1170                 goto fail_sb;
1171
1172         error = init_per_node(sdp, DO);
1173         if (error)
1174                 goto fail_inodes;
1175
1176         error = gfs2_statfs_init(sdp);
1177         if (error) {
1178                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1179                 goto fail_per_node;
1180         }
1181
1182         error = init_threads(sdp, DO);
1183         if (error)
1184                 goto fail_per_node;
1185
1186         if (!(sb->s_flags & MS_RDONLY)) {
1187                 error = gfs2_make_fs_rw(sdp);
1188                 if (error) {
1189                         fs_err(sdp, "can't make FS RW: %d\n", error);
1190                         goto fail_threads;
1191                 }
1192         }
1193
1194         gfs2_glock_dq_uninit(&mount_gh);
1195         gfs2_online_uevent(sdp);
1196         return 0;
1197
1198 fail_threads:
1199         init_threads(sdp, UNDO);
1200 fail_per_node:
1201         init_per_node(sdp, UNDO);
1202 fail_inodes:
1203         init_inodes(sdp, UNDO);
1204 fail_sb:
1205         if (sdp->sd_root_dir)
1206                 dput(sdp->sd_root_dir);
1207         if (sdp->sd_master_dir)
1208                 dput(sdp->sd_master_dir);
1209         if (sb->s_root)
1210                 dput(sb->s_root);
1211         sb->s_root = NULL;
1212 fail_locking:
1213         init_locking(sdp, &mount_gh, UNDO);
1214 fail_lm:
1215         gfs2_gl_hash_clear(sdp);
1216         gfs2_lm_unmount(sdp);
1217 fail_sys:
1218         gfs2_sys_fs_del(sdp);
1219 fail:
1220         gfs2_delete_debugfs_file(sdp);
1221         kfree(sdp);
1222         sb->s_fs_info = NULL;
1223         return error;
1224 }
1225
1226 static int set_gfs2_super(struct super_block *s, void *data)
1227 {
1228         s->s_bdev = data;
1229         s->s_dev = s->s_bdev->bd_dev;
1230
1231         /*
1232          * We set the bdi here to the queue backing, file systems can
1233          * overwrite this in ->fill_super()
1234          */
1235         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1236         return 0;
1237 }
1238
1239 static int test_gfs2_super(struct super_block *s, void *ptr)
1240 {
1241         struct block_device *bdev = ptr;
1242         return (bdev == s->s_bdev);
1243 }
1244
1245 /**
1246  * gfs2_mount - Get the GFS2 superblock
1247  * @fs_type: The GFS2 filesystem type
1248  * @flags: Mount flags
1249  * @dev_name: The name of the device
1250  * @data: The mount arguments
1251  *
1252  * Q. Why not use get_sb_bdev() ?
1253  * A. We need to select one of two root directories to mount, independent
1254  *    of whether this is the initial, or subsequent, mount of this sb
1255  *
1256  * Returns: 0 or -ve on error
1257  */
1258
1259 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1260                        const char *dev_name, void *data)
1261 {
1262         struct block_device *bdev;
1263         struct super_block *s;
1264         fmode_t mode = FMODE_READ | FMODE_EXCL;
1265         int error;
1266         struct gfs2_args args;
1267         struct gfs2_sbd *sdp;
1268
1269         if (!(flags & MS_RDONLY))
1270                 mode |= FMODE_WRITE;
1271
1272         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1273         if (IS_ERR(bdev))
1274                 return ERR_CAST(bdev);
1275
1276         /*
1277          * once the super is inserted into the list by sget, s_umount
1278          * will protect the lockfs code from trying to start a snapshot
1279          * while we are mounting
1280          */
1281         mutex_lock(&bdev->bd_fsfreeze_mutex);
1282         if (bdev->bd_fsfreeze_count > 0) {
1283                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1284                 error = -EBUSY;
1285                 goto error_bdev;
1286         }
1287         s = sget(fs_type, test_gfs2_super, set_gfs2_super, bdev);
1288         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1289         error = PTR_ERR(s);
1290         if (IS_ERR(s))
1291                 goto error_bdev;
1292
1293         if (s->s_root)
1294                 blkdev_put(bdev, mode);
1295
1296         memset(&args, 0, sizeof(args));
1297         args.ar_quota = GFS2_QUOTA_DEFAULT;
1298         args.ar_data = GFS2_DATA_DEFAULT;
1299         args.ar_commit = 30;
1300         args.ar_statfs_quantum = 30;
1301         args.ar_quota_quantum = 60;
1302         args.ar_errors = GFS2_ERRORS_DEFAULT;
1303
1304         error = gfs2_mount_args(&args, data);
1305         if (error) {
1306                 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1307                 goto error_super;
1308         }
1309
1310         if (s->s_root) {
1311                 error = -EBUSY;
1312                 if ((flags ^ s->s_flags) & MS_RDONLY)
1313                         goto error_super;
1314         } else {
1315                 char b[BDEVNAME_SIZE];
1316
1317                 s->s_flags = flags;
1318                 s->s_mode = mode;
1319                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1320                 sb_set_blocksize(s, block_size(bdev));
1321                 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1322                 if (error)
1323                         goto error_super;
1324                 s->s_flags |= MS_ACTIVE;
1325                 bdev->bd_super = s;
1326         }
1327
1328         sdp = s->s_fs_info;
1329         if (args.ar_meta)
1330                 return dget(sdp->sd_master_dir);
1331         else
1332                 return dget(sdp->sd_root_dir);
1333
1334 error_super:
1335         deactivate_locked_super(s);
1336         return ERR_PTR(error);
1337 error_bdev:
1338         blkdev_put(bdev, mode);
1339         return ERR_PTR(error);
1340 }
1341
1342 static int set_meta_super(struct super_block *s, void *ptr)
1343 {
1344         return -EINVAL;
1345 }
1346
1347 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1348                         int flags, const char *dev_name, void *data)
1349 {
1350         struct super_block *s;
1351         struct gfs2_sbd *sdp;
1352         struct path path;
1353         int error;
1354
1355         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1356         if (error) {
1357                 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1358                        dev_name, error);
1359                 return ERR_PTR(error);
1360         }
1361         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super,
1362                  path.dentry->d_inode->i_sb->s_bdev);
1363         path_put(&path);
1364         if (IS_ERR(s)) {
1365                 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1366                 return ERR_CAST(s);
1367         }
1368         if ((flags ^ s->s_flags) & MS_RDONLY) {
1369                 deactivate_locked_super(s);
1370                 return ERR_PTR(-EBUSY);
1371         }
1372         sdp = s->s_fs_info;
1373         return dget(sdp->sd_master_dir);
1374 }
1375
1376 static void gfs2_kill_sb(struct super_block *sb)
1377 {
1378         struct gfs2_sbd *sdp = sb->s_fs_info;
1379
1380         if (sdp == NULL) {
1381                 kill_block_super(sb);
1382                 return;
1383         }
1384
1385         gfs2_meta_syncfs(sdp);
1386         dput(sdp->sd_root_dir);
1387         dput(sdp->sd_master_dir);
1388         sdp->sd_root_dir = NULL;
1389         sdp->sd_master_dir = NULL;
1390         shrink_dcache_sb(sb);
1391         kill_block_super(sb);
1392         gfs2_delete_debugfs_file(sdp);
1393         kfree(sdp);
1394 }
1395
1396 struct file_system_type gfs2_fs_type = {
1397         .name = "gfs2",
1398         .fs_flags = FS_REQUIRES_DEV,
1399         .mount = gfs2_mount,
1400         .kill_sb = gfs2_kill_sb,
1401         .owner = THIS_MODULE,
1402 };
1403
1404 struct file_system_type gfs2meta_fs_type = {
1405         .name = "gfs2meta",
1406         .fs_flags = FS_REQUIRES_DEV,
1407         .mount = gfs2_mount_meta,
1408         .owner = THIS_MODULE,
1409 };
1410