switch open-coded instances of d_make_root() to new helper
[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/export.h>
18 #include <linux/namei.h>
19 #include <linux/mount.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/quotaops.h>
22
23 #include "gfs2.h"
24 #include "incore.h"
25 #include "bmap.h"
26 #include "glock.h"
27 #include "glops.h"
28 #include "inode.h"
29 #include "recovery.h"
30 #include "rgrp.h"
31 #include "super.h"
32 #include "sys.h"
33 #include "util.h"
34 #include "log.h"
35 #include "quota.h"
36 #include "dir.h"
37 #include "trace_gfs2.h"
38
39 #define DO 0
40 #define UNDO 1
41
42 /**
43  * gfs2_tune_init - Fill a gfs2_tune structure with default values
44  * @gt: tune
45  *
46  */
47
48 static void gfs2_tune_init(struct gfs2_tune *gt)
49 {
50         spin_lock_init(&gt->gt_spin);
51
52         gt->gt_quota_simul_sync = 64;
53         gt->gt_quota_warn_period = 10;
54         gt->gt_quota_scale_num = 1;
55         gt->gt_quota_scale_den = 1;
56         gt->gt_new_files_jdata = 0;
57         gt->gt_max_readahead = 1 << 18;
58         gt->gt_complain_secs = 10;
59 }
60
61 static struct gfs2_sbd *init_sbd(struct super_block *sb)
62 {
63         struct gfs2_sbd *sdp;
64
65         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
66         if (!sdp)
67                 return NULL;
68
69         sb->s_fs_info = sdp;
70         sdp->sd_vfs = sb;
71         set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
72         gfs2_tune_init(&sdp->sd_tune);
73
74         init_waitqueue_head(&sdp->sd_glock_wait);
75         atomic_set(&sdp->sd_glock_disposal, 0);
76         init_completion(&sdp->sd_locking_init);
77         spin_lock_init(&sdp->sd_statfs_spin);
78
79         spin_lock_init(&sdp->sd_rindex_spin);
80         mutex_init(&sdp->sd_rindex_mutex);
81         sdp->sd_rindex_tree.rb_node = NULL;
82
83         INIT_LIST_HEAD(&sdp->sd_jindex_list);
84         spin_lock_init(&sdp->sd_jindex_spin);
85         mutex_init(&sdp->sd_jindex_mutex);
86
87         INIT_LIST_HEAD(&sdp->sd_quota_list);
88         mutex_init(&sdp->sd_quota_mutex);
89         init_waitqueue_head(&sdp->sd_quota_wait);
90         INIT_LIST_HEAD(&sdp->sd_trunc_list);
91         spin_lock_init(&sdp->sd_trunc_lock);
92
93         spin_lock_init(&sdp->sd_log_lock);
94         atomic_set(&sdp->sd_log_pinned, 0);
95         INIT_LIST_HEAD(&sdp->sd_log_le_buf);
96         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
97         INIT_LIST_HEAD(&sdp->sd_log_le_rg);
98         INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
99         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
100
101         init_waitqueue_head(&sdp->sd_log_waitq);
102         init_waitqueue_head(&sdp->sd_logd_waitq);
103         spin_lock_init(&sdp->sd_ail_lock);
104         INIT_LIST_HEAD(&sdp->sd_ail1_list);
105         INIT_LIST_HEAD(&sdp->sd_ail2_list);
106
107         init_rwsem(&sdp->sd_log_flush_lock);
108         atomic_set(&sdp->sd_log_in_flight, 0);
109         init_waitqueue_head(&sdp->sd_log_flush_wait);
110
111         INIT_LIST_HEAD(&sdp->sd_revoke_list);
112
113         mutex_init(&sdp->sd_freeze_lock);
114
115         return sdp;
116 }
117
118
119 /**
120  * gfs2_check_sb - Check superblock
121  * @sdp: the filesystem
122  * @sb: The superblock
123  * @silent: Don't print a message if the check fails
124  *
125  * Checks the version code of the FS is one that we understand how to
126  * read and that the sizes of the various on-disk structures have not
127  * changed.
128  */
129
130 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
131 {
132         struct gfs2_sb_host *sb = &sdp->sd_sb;
133
134         if (sb->sb_magic != GFS2_MAGIC ||
135             sb->sb_type != GFS2_METATYPE_SB) {
136                 if (!silent)
137                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
138                 return -EINVAL;
139         }
140
141         /*  If format numbers match exactly, we're done.  */
142
143         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
144             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
145                 return 0;
146
147         fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
148
149         return -EINVAL;
150 }
151
152 static void end_bio_io_page(struct bio *bio, int error)
153 {
154         struct page *page = bio->bi_private;
155
156         if (!error)
157                 SetPageUptodate(page);
158         else
159                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
160         unlock_page(page);
161 }
162
163 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
164 {
165         struct gfs2_sb_host *sb = &sdp->sd_sb;
166         struct super_block *s = sdp->sd_vfs;
167         const struct gfs2_sb *str = buf;
168
169         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
170         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
171         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
172         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
173         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
174         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
175         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
176         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
177         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
178         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
179         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
180
181         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
182         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
183         memcpy(s->s_uuid, str->sb_uuid, 16);
184 }
185
186 /**
187  * gfs2_read_super - Read the gfs2 super block from disk
188  * @sdp: The GFS2 super block
189  * @sector: The location of the super block
190  * @error: The error code to return
191  *
192  * This uses the bio functions to read the super block from disk
193  * because we want to be 100% sure that we never read cached data.
194  * A super block is read twice only during each GFS2 mount and is
195  * never written to by the filesystem. The first time its read no
196  * locks are held, and the only details which are looked at are those
197  * relating to the locking protocol. Once locking is up and working,
198  * the sb is read again under the lock to establish the location of
199  * the master directory (contains pointers to journals etc) and the
200  * root directory.
201  *
202  * Returns: 0 on success or error
203  */
204
205 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
206 {
207         struct super_block *sb = sdp->sd_vfs;
208         struct gfs2_sb *p;
209         struct page *page;
210         struct bio *bio;
211
212         page = alloc_page(GFP_NOFS);
213         if (unlikely(!page))
214                 return -ENOBUFS;
215
216         ClearPageUptodate(page);
217         ClearPageDirty(page);
218         lock_page(page);
219
220         bio = bio_alloc(GFP_NOFS, 1);
221         bio->bi_sector = sector * (sb->s_blocksize >> 9);
222         bio->bi_bdev = sb->s_bdev;
223         bio_add_page(bio, page, PAGE_SIZE, 0);
224
225         bio->bi_end_io = end_bio_io_page;
226         bio->bi_private = page;
227         submit_bio(READ_SYNC | REQ_META, bio);
228         wait_on_page_locked(page);
229         bio_put(bio);
230         if (!PageUptodate(page)) {
231                 __free_page(page);
232                 return -EIO;
233         }
234         p = kmap(page);
235         gfs2_sb_in(sdp, p);
236         kunmap(page);
237         __free_page(page);
238         return gfs2_check_sb(sdp, silent);
239 }
240
241 /**
242  * gfs2_read_sb - Read super block
243  * @sdp: The GFS2 superblock
244  * @silent: Don't print message if mount fails
245  *
246  */
247
248 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
249 {
250         u32 hash_blocks, ind_blocks, leaf_blocks;
251         u32 tmp_blocks;
252         unsigned int x;
253         int error;
254
255         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
256         if (error) {
257                 if (!silent)
258                         fs_err(sdp, "can't read superblock\n");
259                 return error;
260         }
261
262         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
263                                GFS2_BASIC_BLOCK_SHIFT;
264         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
265         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
266                           sizeof(struct gfs2_dinode)) / sizeof(u64);
267         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
268                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
269         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
270         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
271         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
272         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
273         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
274                                 sizeof(struct gfs2_meta_header)) /
275                                 sizeof(struct gfs2_quota_change);
276
277         /* Compute maximum reservation required to add a entry to a directory */
278
279         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
280                              sdp->sd_jbsize);
281
282         ind_blocks = 0;
283         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
284                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
285                 ind_blocks += tmp_blocks;
286         }
287
288         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
289
290         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
291
292         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
293                                 sizeof(struct gfs2_dinode);
294         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
295         for (x = 2;; x++) {
296                 u64 space, d;
297                 u32 m;
298
299                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
300                 d = space;
301                 m = do_div(d, sdp->sd_inptrs);
302
303                 if (d != sdp->sd_heightsize[x - 1] || m)
304                         break;
305                 sdp->sd_heightsize[x] = space;
306         }
307         sdp->sd_max_height = x;
308         sdp->sd_heightsize[x] = ~0;
309         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
310
311         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
312                                  sizeof(struct gfs2_dinode);
313         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
314         for (x = 2;; x++) {
315                 u64 space, d;
316                 u32 m;
317
318                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
319                 d = space;
320                 m = do_div(d, sdp->sd_inptrs);
321
322                 if (d != sdp->sd_jheightsize[x - 1] || m)
323                         break;
324                 sdp->sd_jheightsize[x] = space;
325         }
326         sdp->sd_max_jheight = x;
327         sdp->sd_jheightsize[x] = ~0;
328         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
329
330         return 0;
331 }
332
333 static int init_names(struct gfs2_sbd *sdp, int silent)
334 {
335         char *proto, *table;
336         int error = 0;
337
338         proto = sdp->sd_args.ar_lockproto;
339         table = sdp->sd_args.ar_locktable;
340
341         /*  Try to autodetect  */
342
343         if (!proto[0] || !table[0]) {
344                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
345                 if (error)
346                         return error;
347
348                 if (!proto[0])
349                         proto = sdp->sd_sb.sb_lockproto;
350                 if (!table[0])
351                         table = sdp->sd_sb.sb_locktable;
352         }
353
354         if (!table[0])
355                 table = sdp->sd_vfs->s_id;
356
357         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
358         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
359
360         table = sdp->sd_table_name;
361         while ((table = strchr(table, '/')))
362                 *table = '_';
363
364         return error;
365 }
366
367 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
368                         int undo)
369 {
370         int error = 0;
371
372         if (undo)
373                 goto fail_trans;
374
375         error = gfs2_glock_nq_num(sdp,
376                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
377                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
378                                   mount_gh);
379         if (error) {
380                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
381                 goto fail;
382         }
383
384         error = gfs2_glock_nq_num(sdp,
385                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
386                                   LM_ST_SHARED,
387                                   LM_FLAG_NOEXP | GL_EXACT,
388                                   &sdp->sd_live_gh);
389         if (error) {
390                 fs_err(sdp, "can't acquire live glock: %d\n", error);
391                 goto fail_mount;
392         }
393
394         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
395                                CREATE, &sdp->sd_rename_gl);
396         if (error) {
397                 fs_err(sdp, "can't create rename glock: %d\n", error);
398                 goto fail_live;
399         }
400
401         error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
402                                CREATE, &sdp->sd_trans_gl);
403         if (error) {
404                 fs_err(sdp, "can't create transaction glock: %d\n", error);
405                 goto fail_rename;
406         }
407
408         return 0;
409
410 fail_trans:
411         gfs2_glock_put(sdp->sd_trans_gl);
412 fail_rename:
413         gfs2_glock_put(sdp->sd_rename_gl);
414 fail_live:
415         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
416 fail_mount:
417         gfs2_glock_dq_uninit(mount_gh);
418 fail:
419         return error;
420 }
421
422 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
423                             u64 no_addr, const char *name)
424 {
425         struct gfs2_sbd *sdp = sb->s_fs_info;
426         struct dentry *dentry;
427         struct inode *inode;
428
429         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
430         if (IS_ERR(inode)) {
431                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
432                 return PTR_ERR(inode);
433         }
434         dentry = d_make_root(inode);
435         if (!dentry) {
436                 fs_err(sdp, "can't alloc %s dentry\n", name);
437                 return -ENOMEM;
438         }
439         *dptr = dentry;
440         return 0;
441 }
442
443 static int init_sb(struct gfs2_sbd *sdp, int silent)
444 {
445         struct super_block *sb = sdp->sd_vfs;
446         struct gfs2_holder sb_gh;
447         u64 no_addr;
448         int ret;
449
450         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
451                                 LM_ST_SHARED, 0, &sb_gh);
452         if (ret) {
453                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
454                 return ret;
455         }
456
457         ret = gfs2_read_sb(sdp, silent);
458         if (ret) {
459                 fs_err(sdp, "can't read superblock: %d\n", ret);
460                 goto out;
461         }
462
463         /* Set up the buffer cache and SB for real */
464         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
465                 ret = -EINVAL;
466                 fs_err(sdp, "FS block size (%u) is too small for device "
467                        "block size (%u)\n",
468                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
469                 goto out;
470         }
471         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
472                 ret = -EINVAL;
473                 fs_err(sdp, "FS block size (%u) is too big for machine "
474                        "page size (%u)\n",
475                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
476                 goto out;
477         }
478         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
479
480         /* Get the root inode */
481         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
482         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
483         if (ret)
484                 goto out;
485
486         /* Get the master inode */
487         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
488         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
489         if (ret) {
490                 dput(sdp->sd_root_dir);
491                 goto out;
492         }
493         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
494 out:
495         gfs2_glock_dq_uninit(&sb_gh);
496         return ret;
497 }
498
499 /**
500  * map_journal_extents - create a reusable "extent" mapping from all logical
501  * blocks to all physical blocks for the given journal.  This will save
502  * us time when writing journal blocks.  Most journals will have only one
503  * extent that maps all their logical blocks.  That's because gfs2.mkfs
504  * arranges the journal blocks sequentially to maximize performance.
505  * So the extent would map the first block for the entire file length.
506  * However, gfs2_jadd can happen while file activity is happening, so
507  * those journals may not be sequential.  Less likely is the case where
508  * the users created their own journals by mounting the metafs and
509  * laying it out.  But it's still possible.  These journals might have
510  * several extents.
511  *
512  * TODO: This should be done in bigger chunks rather than one block at a time,
513  *       but since it's only done at mount time, I'm not worried about the
514  *       time it takes.
515  */
516 static int map_journal_extents(struct gfs2_sbd *sdp)
517 {
518         struct gfs2_jdesc *jd = sdp->sd_jdesc;
519         unsigned int lb;
520         u64 db, prev_db; /* logical block, disk block, prev disk block */
521         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
522         struct gfs2_journal_extent *jext = NULL;
523         struct buffer_head bh;
524         int rc = 0;
525
526         prev_db = 0;
527
528         for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
529                 bh.b_state = 0;
530                 bh.b_blocknr = 0;
531                 bh.b_size = 1 << ip->i_inode.i_blkbits;
532                 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
533                 db = bh.b_blocknr;
534                 if (rc || !db) {
535                         printk(KERN_INFO "GFS2 journal mapping error %d: lb="
536                                "%u db=%llu\n", rc, lb, (unsigned long long)db);
537                         break;
538                 }
539                 if (!prev_db || db != prev_db + 1) {
540                         jext = kzalloc(sizeof(struct gfs2_journal_extent),
541                                        GFP_KERNEL);
542                         if (!jext) {
543                                 printk(KERN_INFO "GFS2 error: out of memory "
544                                        "mapping journal extents.\n");
545                                 rc = -ENOMEM;
546                                 break;
547                         }
548                         jext->dblock = db;
549                         jext->lblock = lb;
550                         jext->blocks = 1;
551                         list_add_tail(&jext->extent_list, &jd->extent_list);
552                 } else {
553                         jext->blocks++;
554                 }
555                 prev_db = db;
556         }
557         return rc;
558 }
559
560 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
561 {
562         char *message = "FIRSTMOUNT=Done";
563         char *envp[] = { message, NULL };
564
565         fs_info(sdp, "first mount done, others may mount\n");
566
567         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
568                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
569
570         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
571 }
572
573 /**
574  * gfs2_jindex_hold - Grab a lock on the jindex
575  * @sdp: The GFS2 superblock
576  * @ji_gh: the holder for the jindex glock
577  *
578  * Returns: errno
579  */
580
581 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
582 {
583         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
584         struct qstr name;
585         char buf[20];
586         struct gfs2_jdesc *jd;
587         int error;
588
589         name.name = buf;
590
591         mutex_lock(&sdp->sd_jindex_mutex);
592
593         for (;;) {
594                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
595                 if (error)
596                         break;
597
598                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
599                 name.hash = gfs2_disk_hash(name.name, name.len);
600
601                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
602                 if (error == -ENOENT) {
603                         error = 0;
604                         break;
605                 }
606
607                 gfs2_glock_dq_uninit(ji_gh);
608
609                 if (error)
610                         break;
611
612                 error = -ENOMEM;
613                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
614                 if (!jd)
615                         break;
616
617                 INIT_LIST_HEAD(&jd->extent_list);
618                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
619                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
620                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
621                         if (!jd->jd_inode)
622                                 error = -ENOENT;
623                         else
624                                 error = PTR_ERR(jd->jd_inode);
625                         kfree(jd);
626                         break;
627                 }
628
629                 spin_lock(&sdp->sd_jindex_spin);
630                 jd->jd_jid = sdp->sd_journals++;
631                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
632                 spin_unlock(&sdp->sd_jindex_spin);
633         }
634
635         mutex_unlock(&sdp->sd_jindex_mutex);
636
637         return error;
638 }
639
640 static int init_journal(struct gfs2_sbd *sdp, int undo)
641 {
642         struct inode *master = sdp->sd_master_dir->d_inode;
643         struct gfs2_holder ji_gh;
644         struct gfs2_inode *ip;
645         int jindex = 1;
646         int error = 0;
647
648         if (undo) {
649                 jindex = 0;
650                 goto fail_jinode_gh;
651         }
652
653         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
654         if (IS_ERR(sdp->sd_jindex)) {
655                 fs_err(sdp, "can't lookup journal index: %d\n", error);
656                 return PTR_ERR(sdp->sd_jindex);
657         }
658
659         /* Load in the journal index special file */
660
661         error = gfs2_jindex_hold(sdp, &ji_gh);
662         if (error) {
663                 fs_err(sdp, "can't read journal index: %d\n", error);
664                 goto fail;
665         }
666
667         error = -EUSERS;
668         if (!gfs2_jindex_size(sdp)) {
669                 fs_err(sdp, "no journals!\n");
670                 goto fail_jindex;
671         }
672
673         if (sdp->sd_args.ar_spectator) {
674                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
675                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
676                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
677                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
678         } else {
679                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
680                         fs_err(sdp, "can't mount journal #%u\n",
681                                sdp->sd_lockstruct.ls_jid);
682                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
683                                gfs2_jindex_size(sdp),
684                                gfs2_jindex_size(sdp) - 1);
685                         goto fail_jindex;
686                 }
687                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
688
689                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
690                                           &gfs2_journal_glops,
691                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
692                                           &sdp->sd_journal_gh);
693                 if (error) {
694                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
695                         goto fail_jindex;
696                 }
697
698                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
699                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
700                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
701                                            &sdp->sd_jinode_gh);
702                 if (error) {
703                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
704                                error);
705                         goto fail_journal_gh;
706                 }
707
708                 error = gfs2_jdesc_check(sdp->sd_jdesc);
709                 if (error) {
710                         fs_err(sdp, "my journal (%u) is bad: %d\n",
711                                sdp->sd_jdesc->jd_jid, error);
712                         goto fail_jinode_gh;
713                 }
714                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
715                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
716                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
717
718                 /* Map the extents for this journal's blocks */
719                 map_journal_extents(sdp);
720         }
721         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
722
723         if (sdp->sd_lockstruct.ls_first) {
724                 unsigned int x;
725                 for (x = 0; x < sdp->sd_journals; x++) {
726                         error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x),
727                                                      true);
728                         if (error) {
729                                 fs_err(sdp, "error recovering journal %u: %d\n",
730                                        x, error);
731                                 goto fail_jinode_gh;
732                         }
733                 }
734
735                 gfs2_others_may_mount(sdp);
736         } else if (!sdp->sd_args.ar_spectator) {
737                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
738                 if (error) {
739                         fs_err(sdp, "error recovering my journal: %d\n", error);
740                         goto fail_jinode_gh;
741                 }
742         }
743
744         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
745         gfs2_glock_dq_uninit(&ji_gh);
746         jindex = 0;
747
748         return 0;
749
750 fail_jinode_gh:
751         if (!sdp->sd_args.ar_spectator)
752                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
753 fail_journal_gh:
754         if (!sdp->sd_args.ar_spectator)
755                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
756 fail_jindex:
757         gfs2_jindex_free(sdp);
758         if (jindex)
759                 gfs2_glock_dq_uninit(&ji_gh);
760 fail:
761         iput(sdp->sd_jindex);
762         return error;
763 }
764
765
766 static int init_inodes(struct gfs2_sbd *sdp, int undo)
767 {
768         int error = 0;
769         struct inode *master = sdp->sd_master_dir->d_inode;
770
771         if (undo)
772                 goto fail_qinode;
773
774         error = init_journal(sdp, undo);
775         if (error)
776                 goto fail;
777
778         /* Read in the master statfs inode */
779         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
780         if (IS_ERR(sdp->sd_statfs_inode)) {
781                 error = PTR_ERR(sdp->sd_statfs_inode);
782                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
783                 goto fail_journal;
784         }
785
786         /* Read in the resource index inode */
787         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
788         if (IS_ERR(sdp->sd_rindex)) {
789                 error = PTR_ERR(sdp->sd_rindex);
790                 fs_err(sdp, "can't get resource index inode: %d\n", error);
791                 goto fail_statfs;
792         }
793         sdp->sd_rindex_uptodate = 0;
794
795         /* Read in the quota inode */
796         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
797         if (IS_ERR(sdp->sd_quota_inode)) {
798                 error = PTR_ERR(sdp->sd_quota_inode);
799                 fs_err(sdp, "can't get quota file inode: %d\n", error);
800                 goto fail_rindex;
801         }
802
803         error = gfs2_rindex_update(sdp);
804         if (error)
805                 goto fail_qinode;
806
807         return 0;
808
809 fail_qinode:
810         iput(sdp->sd_quota_inode);
811 fail_rindex:
812         gfs2_clear_rgrpd(sdp);
813         iput(sdp->sd_rindex);
814 fail_statfs:
815         iput(sdp->sd_statfs_inode);
816 fail_journal:
817         init_journal(sdp, UNDO);
818 fail:
819         return error;
820 }
821
822 static int init_per_node(struct gfs2_sbd *sdp, int undo)
823 {
824         struct inode *pn = NULL;
825         char buf[30];
826         int error = 0;
827         struct gfs2_inode *ip;
828         struct inode *master = sdp->sd_master_dir->d_inode;
829
830         if (sdp->sd_args.ar_spectator)
831                 return 0;
832
833         if (undo)
834                 goto fail_qc_gh;
835
836         pn = gfs2_lookup_simple(master, "per_node");
837         if (IS_ERR(pn)) {
838                 error = PTR_ERR(pn);
839                 fs_err(sdp, "can't find per_node directory: %d\n", error);
840                 return error;
841         }
842
843         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
844         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
845         if (IS_ERR(sdp->sd_sc_inode)) {
846                 error = PTR_ERR(sdp->sd_sc_inode);
847                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
848                 goto fail;
849         }
850
851         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
852         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
853         if (IS_ERR(sdp->sd_qc_inode)) {
854                 error = PTR_ERR(sdp->sd_qc_inode);
855                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
856                 goto fail_ut_i;
857         }
858
859         iput(pn);
860         pn = NULL;
861
862         ip = GFS2_I(sdp->sd_sc_inode);
863         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
864                                    &sdp->sd_sc_gh);
865         if (error) {
866                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
867                 goto fail_qc_i;
868         }
869
870         ip = GFS2_I(sdp->sd_qc_inode);
871         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
872                                    &sdp->sd_qc_gh);
873         if (error) {
874                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
875                 goto fail_ut_gh;
876         }
877
878         return 0;
879
880 fail_qc_gh:
881         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
882 fail_ut_gh:
883         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
884 fail_qc_i:
885         iput(sdp->sd_qc_inode);
886 fail_ut_i:
887         iput(sdp->sd_sc_inode);
888 fail:
889         if (pn)
890                 iput(pn);
891         return error;
892 }
893
894 static int init_threads(struct gfs2_sbd *sdp, int undo)
895 {
896         struct task_struct *p;
897         int error = 0;
898
899         if (undo)
900                 goto fail_quotad;
901
902         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
903         error = IS_ERR(p);
904         if (error) {
905                 fs_err(sdp, "can't start logd thread: %d\n", error);
906                 return error;
907         }
908         sdp->sd_logd_process = p;
909
910         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
911         error = IS_ERR(p);
912         if (error) {
913                 fs_err(sdp, "can't start quotad thread: %d\n", error);
914                 goto fail;
915         }
916         sdp->sd_quotad_process = p;
917
918         return 0;
919
920
921 fail_quotad:
922         kthread_stop(sdp->sd_quotad_process);
923 fail:
924         kthread_stop(sdp->sd_logd_process);
925         return error;
926 }
927
928 static const match_table_t nolock_tokens = {
929         { Opt_jid, "jid=%d\n", },
930         { Opt_err, NULL },
931 };
932
933 static const struct lm_lockops nolock_ops = {
934         .lm_proto_name = "lock_nolock",
935         .lm_put_lock = gfs2_glock_free,
936         .lm_tokens = &nolock_tokens,
937 };
938
939 /**
940  * gfs2_lm_mount - mount a locking protocol
941  * @sdp: the filesystem
942  * @args: mount arguments
943  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
944  *
945  * Returns: errno
946  */
947
948 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
949 {
950         const struct lm_lockops *lm;
951         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
952         struct gfs2_args *args = &sdp->sd_args;
953         const char *proto = sdp->sd_proto_name;
954         const char *table = sdp->sd_table_name;
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 (lm->lm_mount == NULL) {
1015                 fs_info(sdp, "Now mounting FS...\n");
1016                 complete_all(&sdp->sd_locking_init);
1017                 return 0;
1018         }
1019         ret = lm->lm_mount(sdp, table);
1020         if (ret == 0)
1021                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1022         complete_all(&sdp->sd_locking_init);
1023         return ret;
1024 }
1025
1026 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1027 {
1028         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1029         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1030             lm->lm_unmount)
1031                 lm->lm_unmount(sdp);
1032 }
1033
1034 static int gfs2_journalid_wait(void *word)
1035 {
1036         if (signal_pending(current))
1037                 return -EINTR;
1038         schedule();
1039         return 0;
1040 }
1041
1042 static int wait_on_journal(struct gfs2_sbd *sdp)
1043 {
1044         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1045                 return 0;
1046
1047         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1048 }
1049
1050 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1051 {
1052         struct super_block *sb = sdp->sd_vfs;
1053         char ro[20];
1054         char spectator[20];
1055         char *envp[] = { ro, spectator, NULL };
1056         sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1057         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1058         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1059 }
1060
1061 /**
1062  * fill_super - Read in superblock
1063  * @sb: The VFS superblock
1064  * @data: Mount options
1065  * @silent: Don't complain if it's not a GFS2 filesystem
1066  *
1067  * Returns: errno
1068  */
1069
1070 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1071 {
1072         struct gfs2_sbd *sdp;
1073         struct gfs2_holder mount_gh;
1074         int error;
1075
1076         sdp = init_sbd(sb);
1077         if (!sdp) {
1078                 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1079                 return -ENOMEM;
1080         }
1081         sdp->sd_args = *args;
1082
1083         if (sdp->sd_args.ar_spectator) {
1084                 sb->s_flags |= MS_RDONLY;
1085                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1086         }
1087         if (sdp->sd_args.ar_posix_acl)
1088                 sb->s_flags |= MS_POSIXACL;
1089         if (sdp->sd_args.ar_nobarrier)
1090                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1091
1092         sb->s_flags |= MS_NOSEC;
1093         sb->s_magic = GFS2_MAGIC;
1094         sb->s_op = &gfs2_super_ops;
1095         sb->s_d_op = &gfs2_dops;
1096         sb->s_export_op = &gfs2_export_ops;
1097         sb->s_xattr = gfs2_xattr_handlers;
1098         sb->s_qcop = &gfs2_quotactl_ops;
1099         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1100         sb->s_time_gran = 1;
1101         sb->s_maxbytes = MAX_LFS_FILESIZE;
1102
1103         /* Set up the buffer cache and fill in some fake block size values
1104            to allow us to read-in the on-disk superblock. */
1105         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1106         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1107         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1108                                GFS2_BASIC_BLOCK_SHIFT;
1109         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1110
1111         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1112         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1113         if (sdp->sd_args.ar_statfs_quantum) {
1114                 sdp->sd_tune.gt_statfs_slow = 0;
1115                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1116         } else {
1117                 sdp->sd_tune.gt_statfs_slow = 1;
1118                 sdp->sd_tune.gt_statfs_quantum = 30;
1119         }
1120
1121         error = init_names(sdp, silent);
1122         if (error)
1123                 goto fail;
1124
1125         snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1126
1127         gfs2_create_debugfs_file(sdp);
1128
1129         error = gfs2_sys_fs_add(sdp);
1130         if (error)
1131                 goto fail;
1132
1133         error = gfs2_lm_mount(sdp, silent);
1134         if (error)
1135                 goto fail_sys;
1136
1137         error = init_locking(sdp, &mount_gh, DO);
1138         if (error)
1139                 goto fail_lm;
1140
1141         error = init_sb(sdp, silent);
1142         if (error)
1143                 goto fail_locking;
1144
1145         error = wait_on_journal(sdp);
1146         if (error)
1147                 goto fail_sb;
1148
1149         /*
1150          * If user space has failed to join the cluster or some similar
1151          * failure has occurred, then the journal id will contain a
1152          * negative (error) number. This will then be returned to the
1153          * caller (of the mount syscall). We do this even for spectator
1154          * mounts (which just write a jid of 0 to indicate "ok" even though
1155          * the jid is unused in the spectator case)
1156          */
1157         if (sdp->sd_lockstruct.ls_jid < 0) {
1158                 error = sdp->sd_lockstruct.ls_jid;
1159                 sdp->sd_lockstruct.ls_jid = 0;
1160                 goto fail_sb;
1161         }
1162
1163         if (sdp->sd_args.ar_spectator)
1164                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1165                          sdp->sd_table_name);
1166         else
1167                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1168                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1169
1170         error = init_inodes(sdp, DO);
1171         if (error)
1172                 goto fail_sb;
1173
1174         error = init_per_node(sdp, DO);
1175         if (error)
1176                 goto fail_inodes;
1177
1178         error = gfs2_statfs_init(sdp);
1179         if (error) {
1180                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1181                 goto fail_per_node;
1182         }
1183
1184         error = init_threads(sdp, DO);
1185         if (error)
1186                 goto fail_per_node;
1187
1188         if (!(sb->s_flags & MS_RDONLY)) {
1189                 error = gfs2_make_fs_rw(sdp);
1190                 if (error) {
1191                         fs_err(sdp, "can't make FS RW: %d\n", error);
1192                         goto fail_threads;
1193                 }
1194         }
1195
1196         gfs2_glock_dq_uninit(&mount_gh);
1197         gfs2_online_uevent(sdp);
1198         return 0;
1199
1200 fail_threads:
1201         init_threads(sdp, UNDO);
1202 fail_per_node:
1203         init_per_node(sdp, UNDO);
1204 fail_inodes:
1205         init_inodes(sdp, UNDO);
1206 fail_sb:
1207         if (sdp->sd_root_dir)
1208                 dput(sdp->sd_root_dir);
1209         if (sdp->sd_master_dir)
1210                 dput(sdp->sd_master_dir);
1211         if (sb->s_root)
1212                 dput(sb->s_root);
1213         sb->s_root = NULL;
1214 fail_locking:
1215         init_locking(sdp, &mount_gh, UNDO);
1216 fail_lm:
1217         gfs2_gl_hash_clear(sdp);
1218         gfs2_lm_unmount(sdp);
1219 fail_sys:
1220         gfs2_sys_fs_del(sdp);
1221 fail:
1222         gfs2_delete_debugfs_file(sdp);
1223         kfree(sdp);
1224         sb->s_fs_info = NULL;
1225         return error;
1226 }
1227
1228 static int set_gfs2_super(struct super_block *s, void *data)
1229 {
1230         s->s_bdev = data;
1231         s->s_dev = s->s_bdev->bd_dev;
1232
1233         /*
1234          * We set the bdi here to the queue backing, file systems can
1235          * overwrite this in ->fill_super()
1236          */
1237         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1238         return 0;
1239 }
1240
1241 static int test_gfs2_super(struct super_block *s, void *ptr)
1242 {
1243         struct block_device *bdev = ptr;
1244         return (bdev == s->s_bdev);
1245 }
1246
1247 /**
1248  * gfs2_mount - Get the GFS2 superblock
1249  * @fs_type: The GFS2 filesystem type
1250  * @flags: Mount flags
1251  * @dev_name: The name of the device
1252  * @data: The mount arguments
1253  *
1254  * Q. Why not use get_sb_bdev() ?
1255  * A. We need to select one of two root directories to mount, independent
1256  *    of whether this is the initial, or subsequent, mount of this sb
1257  *
1258  * Returns: 0 or -ve on error
1259  */
1260
1261 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1262                        const char *dev_name, void *data)
1263 {
1264         struct block_device *bdev;
1265         struct super_block *s;
1266         fmode_t mode = FMODE_READ | FMODE_EXCL;
1267         int error;
1268         struct gfs2_args args;
1269         struct gfs2_sbd *sdp;
1270
1271         if (!(flags & MS_RDONLY))
1272                 mode |= FMODE_WRITE;
1273
1274         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1275         if (IS_ERR(bdev))
1276                 return ERR_CAST(bdev);
1277
1278         /*
1279          * once the super is inserted into the list by sget, s_umount
1280          * will protect the lockfs code from trying to start a snapshot
1281          * while we are mounting
1282          */
1283         mutex_lock(&bdev->bd_fsfreeze_mutex);
1284         if (bdev->bd_fsfreeze_count > 0) {
1285                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1286                 error = -EBUSY;
1287                 goto error_bdev;
1288         }
1289         s = sget(fs_type, test_gfs2_super, set_gfs2_super, bdev);
1290         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1291         error = PTR_ERR(s);
1292         if (IS_ERR(s))
1293                 goto error_bdev;
1294
1295         if (s->s_root)
1296                 blkdev_put(bdev, mode);
1297
1298         memset(&args, 0, sizeof(args));
1299         args.ar_quota = GFS2_QUOTA_DEFAULT;
1300         args.ar_data = GFS2_DATA_DEFAULT;
1301         args.ar_commit = 30;
1302         args.ar_statfs_quantum = 30;
1303         args.ar_quota_quantum = 60;
1304         args.ar_errors = GFS2_ERRORS_DEFAULT;
1305
1306         error = gfs2_mount_args(&args, data);
1307         if (error) {
1308                 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1309                 goto error_super;
1310         }
1311
1312         if (s->s_root) {
1313                 error = -EBUSY;
1314                 if ((flags ^ s->s_flags) & MS_RDONLY)
1315                         goto error_super;
1316         } else {
1317                 char b[BDEVNAME_SIZE];
1318
1319                 s->s_flags = flags;
1320                 s->s_mode = mode;
1321                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1322                 sb_set_blocksize(s, block_size(bdev));
1323                 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1324                 if (error)
1325                         goto error_super;
1326                 s->s_flags |= MS_ACTIVE;
1327                 bdev->bd_super = s;
1328         }
1329
1330         sdp = s->s_fs_info;
1331         if (args.ar_meta)
1332                 return dget(sdp->sd_master_dir);
1333         else
1334                 return dget(sdp->sd_root_dir);
1335
1336 error_super:
1337         deactivate_locked_super(s);
1338         return ERR_PTR(error);
1339 error_bdev:
1340         blkdev_put(bdev, mode);
1341         return ERR_PTR(error);
1342 }
1343
1344 static int set_meta_super(struct super_block *s, void *ptr)
1345 {
1346         return -EINVAL;
1347 }
1348
1349 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1350                         int flags, const char *dev_name, void *data)
1351 {
1352         struct super_block *s;
1353         struct gfs2_sbd *sdp;
1354         struct path path;
1355         int error;
1356
1357         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1358         if (error) {
1359                 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1360                        dev_name, error);
1361                 return ERR_PTR(error);
1362         }
1363         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super,
1364                  path.dentry->d_inode->i_sb->s_bdev);
1365         path_put(&path);
1366         if (IS_ERR(s)) {
1367                 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1368                 return ERR_CAST(s);
1369         }
1370         if ((flags ^ s->s_flags) & MS_RDONLY) {
1371                 deactivate_locked_super(s);
1372                 return ERR_PTR(-EBUSY);
1373         }
1374         sdp = s->s_fs_info;
1375         return dget(sdp->sd_master_dir);
1376 }
1377
1378 static void gfs2_kill_sb(struct super_block *sb)
1379 {
1380         struct gfs2_sbd *sdp = sb->s_fs_info;
1381
1382         if (sdp == NULL) {
1383                 kill_block_super(sb);
1384                 return;
1385         }
1386
1387         gfs2_meta_syncfs(sdp);
1388         dput(sdp->sd_root_dir);
1389         dput(sdp->sd_master_dir);
1390         sdp->sd_root_dir = NULL;
1391         sdp->sd_master_dir = NULL;
1392         shrink_dcache_sb(sb);
1393         kill_block_super(sb);
1394         gfs2_delete_debugfs_file(sdp);
1395         kfree(sdp);
1396 }
1397
1398 struct file_system_type gfs2_fs_type = {
1399         .name = "gfs2",
1400         .fs_flags = FS_REQUIRES_DEV,
1401         .mount = gfs2_mount,
1402         .kill_sb = gfs2_kill_sb,
1403         .owner = THIS_MODULE,
1404 };
1405
1406 struct file_system_type gfs2meta_fs_type = {
1407         .name = "gfs2meta",
1408         .fs_flags = FS_REQUIRES_DEV,
1409         .mount = gfs2_mount_meta,
1410         .owner = THIS_MODULE,
1411 };
1412