block: separate priority boosting from REQ_META
[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         init_completion(&sdp->sd_locking_init);
76         spin_lock_init(&sdp->sd_statfs_spin);
77
78         spin_lock_init(&sdp->sd_rindex_spin);
79         mutex_init(&sdp->sd_rindex_mutex);
80         INIT_LIST_HEAD(&sdp->sd_rindex_list);
81         INIT_LIST_HEAD(&sdp->sd_rindex_mru_list);
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 | REQ_PRIO, 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_alloc_root(inode);
435         if (!dentry) {
436                 fs_err(sdp, "can't alloc %s dentry\n", name);
437                 iput(inode);
438                 return -ENOMEM;
439         }
440         *dptr = dentry;
441         return 0;
442 }
443
444 static int init_sb(struct gfs2_sbd *sdp, int silent)
445 {
446         struct super_block *sb = sdp->sd_vfs;
447         struct gfs2_holder sb_gh;
448         u64 no_addr;
449         int ret;
450
451         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
452                                 LM_ST_SHARED, 0, &sb_gh);
453         if (ret) {
454                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
455                 return ret;
456         }
457
458         ret = gfs2_read_sb(sdp, silent);
459         if (ret) {
460                 fs_err(sdp, "can't read superblock: %d\n", ret);
461                 goto out;
462         }
463
464         /* Set up the buffer cache and SB for real */
465         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
466                 ret = -EINVAL;
467                 fs_err(sdp, "FS block size (%u) is too small for device "
468                        "block size (%u)\n",
469                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
470                 goto out;
471         }
472         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
473                 ret = -EINVAL;
474                 fs_err(sdp, "FS block size (%u) is too big for machine "
475                        "page size (%u)\n",
476                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
477                 goto out;
478         }
479         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
480
481         /* Get the root inode */
482         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
483         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
484         if (ret)
485                 goto out;
486
487         /* Get the master inode */
488         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
489         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
490         if (ret) {
491                 dput(sdp->sd_root_dir);
492                 goto out;
493         }
494         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
495 out:
496         gfs2_glock_dq_uninit(&sb_gh);
497         return ret;
498 }
499
500 /**
501  * map_journal_extents - create a reusable "extent" mapping from all logical
502  * blocks to all physical blocks for the given journal.  This will save
503  * us time when writing journal blocks.  Most journals will have only one
504  * extent that maps all their logical blocks.  That's because gfs2.mkfs
505  * arranges the journal blocks sequentially to maximize performance.
506  * So the extent would map the first block for the entire file length.
507  * However, gfs2_jadd can happen while file activity is happening, so
508  * those journals may not be sequential.  Less likely is the case where
509  * the users created their own journals by mounting the metafs and
510  * laying it out.  But it's still possible.  These journals might have
511  * several extents.
512  *
513  * TODO: This should be done in bigger chunks rather than one block at a time,
514  *       but since it's only done at mount time, I'm not worried about the
515  *       time it takes.
516  */
517 static int map_journal_extents(struct gfs2_sbd *sdp)
518 {
519         struct gfs2_jdesc *jd = sdp->sd_jdesc;
520         unsigned int lb;
521         u64 db, prev_db; /* logical block, disk block, prev disk block */
522         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
523         struct gfs2_journal_extent *jext = NULL;
524         struct buffer_head bh;
525         int rc = 0;
526
527         prev_db = 0;
528
529         for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
530                 bh.b_state = 0;
531                 bh.b_blocknr = 0;
532                 bh.b_size = 1 << ip->i_inode.i_blkbits;
533                 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
534                 db = bh.b_blocknr;
535                 if (rc || !db) {
536                         printk(KERN_INFO "GFS2 journal mapping error %d: lb="
537                                "%u db=%llu\n", rc, lb, (unsigned long long)db);
538                         break;
539                 }
540                 if (!prev_db || db != prev_db + 1) {
541                         jext = kzalloc(sizeof(struct gfs2_journal_extent),
542                                        GFP_KERNEL);
543                         if (!jext) {
544                                 printk(KERN_INFO "GFS2 error: out of memory "
545                                        "mapping journal extents.\n");
546                                 rc = -ENOMEM;
547                                 break;
548                         }
549                         jext->dblock = db;
550                         jext->lblock = lb;
551                         jext->blocks = 1;
552                         list_add_tail(&jext->extent_list, &jd->extent_list);
553                 } else {
554                         jext->blocks++;
555                 }
556                 prev_db = db;
557         }
558         return rc;
559 }
560
561 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
562 {
563         char *message = "FIRSTMOUNT=Done";
564         char *envp[] = { message, NULL };
565         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
566         ls->ls_first_done = 1;
567         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
568 }
569
570 /**
571  * gfs2_jindex_hold - Grab a lock on the jindex
572  * @sdp: The GFS2 superblock
573  * @ji_gh: the holder for the jindex glock
574  *
575  * Returns: errno
576  */
577
578 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
579 {
580         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
581         struct qstr name;
582         char buf[20];
583         struct gfs2_jdesc *jd;
584         int error;
585
586         name.name = buf;
587
588         mutex_lock(&sdp->sd_jindex_mutex);
589
590         for (;;) {
591                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
592                 if (error)
593                         break;
594
595                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
596                 name.hash = gfs2_disk_hash(name.name, name.len);
597
598                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
599                 if (error == -ENOENT) {
600                         error = 0;
601                         break;
602                 }
603
604                 gfs2_glock_dq_uninit(ji_gh);
605
606                 if (error)
607                         break;
608
609                 error = -ENOMEM;
610                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
611                 if (!jd)
612                         break;
613
614                 INIT_LIST_HEAD(&jd->extent_list);
615                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
616                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
617                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
618                         if (!jd->jd_inode)
619                                 error = -ENOENT;
620                         else
621                                 error = PTR_ERR(jd->jd_inode);
622                         kfree(jd);
623                         break;
624                 }
625
626                 spin_lock(&sdp->sd_jindex_spin);
627                 jd->jd_jid = sdp->sd_journals++;
628                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
629                 spin_unlock(&sdp->sd_jindex_spin);
630         }
631
632         mutex_unlock(&sdp->sd_jindex_mutex);
633
634         return error;
635 }
636
637 static int init_journal(struct gfs2_sbd *sdp, int undo)
638 {
639         struct inode *master = sdp->sd_master_dir->d_inode;
640         struct gfs2_holder ji_gh;
641         struct gfs2_inode *ip;
642         int jindex = 1;
643         int error = 0;
644
645         if (undo) {
646                 jindex = 0;
647                 goto fail_jinode_gh;
648         }
649
650         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
651         if (IS_ERR(sdp->sd_jindex)) {
652                 fs_err(sdp, "can't lookup journal index: %d\n", error);
653                 return PTR_ERR(sdp->sd_jindex);
654         }
655         ip = GFS2_I(sdp->sd_jindex);
656
657         /* Load in the journal index special file */
658
659         error = gfs2_jindex_hold(sdp, &ji_gh);
660         if (error) {
661                 fs_err(sdp, "can't read journal index: %d\n", error);
662                 goto fail;
663         }
664
665         error = -EUSERS;
666         if (!gfs2_jindex_size(sdp)) {
667                 fs_err(sdp, "no journals!\n");
668                 goto fail_jindex;
669         }
670
671         if (sdp->sd_args.ar_spectator) {
672                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
673                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
674                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
675                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
676         } else {
677                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
678                         fs_err(sdp, "can't mount journal #%u\n",
679                                sdp->sd_lockstruct.ls_jid);
680                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
681                                gfs2_jindex_size(sdp),
682                                gfs2_jindex_size(sdp) - 1);
683                         goto fail_jindex;
684                 }
685                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
686
687                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
688                                           &gfs2_journal_glops,
689                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
690                                           &sdp->sd_journal_gh);
691                 if (error) {
692                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
693                         goto fail_jindex;
694                 }
695
696                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
697                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
698                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
699                                            &sdp->sd_jinode_gh);
700                 if (error) {
701                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
702                                error);
703                         goto fail_journal_gh;
704                 }
705
706                 error = gfs2_jdesc_check(sdp->sd_jdesc);
707                 if (error) {
708                         fs_err(sdp, "my journal (%u) is bad: %d\n",
709                                sdp->sd_jdesc->jd_jid, error);
710                         goto fail_jinode_gh;
711                 }
712                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
713                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
714                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
715
716                 /* Map the extents for this journal's blocks */
717                 map_journal_extents(sdp);
718         }
719         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
720
721         if (sdp->sd_lockstruct.ls_first) {
722                 unsigned int x;
723                 for (x = 0; x < sdp->sd_journals; x++) {
724                         error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x),
725                                                      true);
726                         if (error) {
727                                 fs_err(sdp, "error recovering journal %u: %d\n",
728                                        x, error);
729                                 goto fail_jinode_gh;
730                         }
731                 }
732
733                 gfs2_others_may_mount(sdp);
734         } else if (!sdp->sd_args.ar_spectator) {
735                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
736                 if (error) {
737                         fs_err(sdp, "error recovering my journal: %d\n", error);
738                         goto fail_jinode_gh;
739                 }
740         }
741
742         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
743         gfs2_glock_dq_uninit(&ji_gh);
744         jindex = 0;
745
746         return 0;
747
748 fail_jinode_gh:
749         if (!sdp->sd_args.ar_spectator)
750                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
751 fail_journal_gh:
752         if (!sdp->sd_args.ar_spectator)
753                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
754 fail_jindex:
755         gfs2_jindex_free(sdp);
756         if (jindex)
757                 gfs2_glock_dq_uninit(&ji_gh);
758 fail:
759         iput(sdp->sd_jindex);
760         return error;
761 }
762
763
764 static int init_inodes(struct gfs2_sbd *sdp, int undo)
765 {
766         int error = 0;
767         struct gfs2_inode *ip;
768         struct inode *master = sdp->sd_master_dir->d_inode;
769
770         if (undo)
771                 goto fail_qinode;
772
773         error = init_journal(sdp, undo);
774         if (error)
775                 goto fail;
776
777         /* Read in the master statfs inode */
778         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
779         if (IS_ERR(sdp->sd_statfs_inode)) {
780                 error = PTR_ERR(sdp->sd_statfs_inode);
781                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
782                 goto fail_journal;
783         }
784
785         /* Read in the resource index inode */
786         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
787         if (IS_ERR(sdp->sd_rindex)) {
788                 error = PTR_ERR(sdp->sd_rindex);
789                 fs_err(sdp, "can't get resource index inode: %d\n", error);
790                 goto fail_statfs;
791         }
792         ip = GFS2_I(sdp->sd_rindex);
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         return 0;
803
804 fail_qinode:
805         iput(sdp->sd_quota_inode);
806 fail_rindex:
807         gfs2_clear_rgrpd(sdp);
808         iput(sdp->sd_rindex);
809 fail_statfs:
810         iput(sdp->sd_statfs_inode);
811 fail_journal:
812         init_journal(sdp, UNDO);
813 fail:
814         return error;
815 }
816
817 static int init_per_node(struct gfs2_sbd *sdp, int undo)
818 {
819         struct inode *pn = NULL;
820         char buf[30];
821         int error = 0;
822         struct gfs2_inode *ip;
823         struct inode *master = sdp->sd_master_dir->d_inode;
824
825         if (sdp->sd_args.ar_spectator)
826                 return 0;
827
828         if (undo)
829                 goto fail_qc_gh;
830
831         pn = gfs2_lookup_simple(master, "per_node");
832         if (IS_ERR(pn)) {
833                 error = PTR_ERR(pn);
834                 fs_err(sdp, "can't find per_node directory: %d\n", error);
835                 return error;
836         }
837
838         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
839         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
840         if (IS_ERR(sdp->sd_sc_inode)) {
841                 error = PTR_ERR(sdp->sd_sc_inode);
842                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
843                 goto fail;
844         }
845
846         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
847         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
848         if (IS_ERR(sdp->sd_qc_inode)) {
849                 error = PTR_ERR(sdp->sd_qc_inode);
850                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
851                 goto fail_ut_i;
852         }
853
854         iput(pn);
855         pn = NULL;
856
857         ip = GFS2_I(sdp->sd_sc_inode);
858         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
859                                    &sdp->sd_sc_gh);
860         if (error) {
861                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
862                 goto fail_qc_i;
863         }
864
865         ip = GFS2_I(sdp->sd_qc_inode);
866         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
867                                    &sdp->sd_qc_gh);
868         if (error) {
869                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
870                 goto fail_ut_gh;
871         }
872
873         return 0;
874
875 fail_qc_gh:
876         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
877 fail_ut_gh:
878         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
879 fail_qc_i:
880         iput(sdp->sd_qc_inode);
881 fail_ut_i:
882         iput(sdp->sd_sc_inode);
883 fail:
884         if (pn)
885                 iput(pn);
886         return error;
887 }
888
889 static int init_threads(struct gfs2_sbd *sdp, int undo)
890 {
891         struct task_struct *p;
892         int error = 0;
893
894         if (undo)
895                 goto fail_quotad;
896
897         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
898         error = IS_ERR(p);
899         if (error) {
900                 fs_err(sdp, "can't start logd thread: %d\n", error);
901                 return error;
902         }
903         sdp->sd_logd_process = p;
904
905         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
906         error = IS_ERR(p);
907         if (error) {
908                 fs_err(sdp, "can't start quotad thread: %d\n", error);
909                 goto fail;
910         }
911         sdp->sd_quotad_process = p;
912
913         return 0;
914
915
916 fail_quotad:
917         kthread_stop(sdp->sd_quotad_process);
918 fail:
919         kthread_stop(sdp->sd_logd_process);
920         return error;
921 }
922
923 static const match_table_t nolock_tokens = {
924         { Opt_jid, "jid=%d\n", },
925         { Opt_err, NULL },
926 };
927
928 static const struct lm_lockops nolock_ops = {
929         .lm_proto_name = "lock_nolock",
930         .lm_put_lock = gfs2_glock_free,
931         .lm_tokens = &nolock_tokens,
932 };
933
934 /**
935  * gfs2_lm_mount - mount a locking protocol
936  * @sdp: the filesystem
937  * @args: mount arguments
938  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
939  *
940  * Returns: errno
941  */
942
943 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
944 {
945         const struct lm_lockops *lm;
946         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
947         struct gfs2_args *args = &sdp->sd_args;
948         const char *proto = sdp->sd_proto_name;
949         const char *table = sdp->sd_table_name;
950         const char *fsname;
951         char *o, *options;
952         int ret;
953
954         if (!strcmp("lock_nolock", proto)) {
955                 lm = &nolock_ops;
956                 sdp->sd_args.ar_localflocks = 1;
957 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
958         } else if (!strcmp("lock_dlm", proto)) {
959                 lm = &gfs2_dlm_ops;
960 #endif
961         } else {
962                 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
963                 return -ENOENT;
964         }
965
966         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
967
968         ls->ls_ops = lm;
969         ls->ls_first = 1;
970
971         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
972                 substring_t tmp[MAX_OPT_ARGS];
973                 int token, option;
974
975                 if (!o || !*o)
976                         continue;
977
978                 token = match_token(o, *lm->lm_tokens, tmp);
979                 switch (token) {
980                 case Opt_jid:
981                         ret = match_int(&tmp[0], &option);
982                         if (ret || option < 0) 
983                                 goto hostdata_error;
984                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
985                                 ls->ls_jid = option;
986                         break;
987                 case Opt_id:
988                         /* Obsolete, but left for backward compat purposes */
989                         break;
990                 case Opt_first:
991                         ret = match_int(&tmp[0], &option);
992                         if (ret || (option != 0 && option != 1))
993                                 goto hostdata_error;
994                         ls->ls_first = option;
995                         break;
996                 case Opt_nodir:
997                         ret = match_int(&tmp[0], &option);
998                         if (ret || (option != 0 && option != 1))
999                                 goto hostdata_error;
1000                         ls->ls_nodir = option;
1001                         break;
1002                 case Opt_err:
1003                 default:
1004 hostdata_error:
1005                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1006                         return -EINVAL;
1007                 }
1008         }
1009
1010         if (sdp->sd_args.ar_spectator)
1011                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s", table);
1012         else
1013                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u", table,
1014                          sdp->sd_lockstruct.ls_jid);
1015
1016         fsname = strchr(table, ':');
1017         if (fsname)
1018                 fsname++;
1019         if (lm->lm_mount == NULL) {
1020                 fs_info(sdp, "Now mounting FS...\n");
1021                 complete_all(&sdp->sd_locking_init);
1022                 return 0;
1023         }
1024         ret = lm->lm_mount(sdp, fsname);
1025         if (ret == 0)
1026                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1027         complete_all(&sdp->sd_locking_init);
1028         return ret;
1029 }
1030
1031 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1032 {
1033         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1034         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1035             lm->lm_unmount)
1036                 lm->lm_unmount(sdp);
1037 }
1038
1039 static int gfs2_journalid_wait(void *word)
1040 {
1041         if (signal_pending(current))
1042                 return -EINTR;
1043         schedule();
1044         return 0;
1045 }
1046
1047 static int wait_on_journal(struct gfs2_sbd *sdp)
1048 {
1049         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1050                 return 0;
1051
1052         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1053 }
1054
1055 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1056 {
1057         struct super_block *sb = sdp->sd_vfs;
1058         char ro[20];
1059         char spectator[20];
1060         char *envp[] = { ro, spectator, NULL };
1061         sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1062         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1063         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1064 }
1065
1066 /**
1067  * fill_super - Read in superblock
1068  * @sb: The VFS superblock
1069  * @data: Mount options
1070  * @silent: Don't complain if it's not a GFS2 filesystem
1071  *
1072  * Returns: errno
1073  */
1074
1075 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1076 {
1077         struct gfs2_sbd *sdp;
1078         struct gfs2_holder mount_gh;
1079         int error;
1080
1081         sdp = init_sbd(sb);
1082         if (!sdp) {
1083                 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1084                 return -ENOMEM;
1085         }
1086         sdp->sd_args = *args;
1087
1088         if (sdp->sd_args.ar_spectator) {
1089                 sb->s_flags |= MS_RDONLY;
1090                 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
1091         }
1092         if (sdp->sd_args.ar_posix_acl)
1093                 sb->s_flags |= MS_POSIXACL;
1094         if (sdp->sd_args.ar_nobarrier)
1095                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1096
1097         sb->s_flags |= MS_NOSEC;
1098         sb->s_magic = GFS2_MAGIC;
1099         sb->s_op = &gfs2_super_ops;
1100         sb->s_d_op = &gfs2_dops;
1101         sb->s_export_op = &gfs2_export_ops;
1102         sb->s_xattr = gfs2_xattr_handlers;
1103         sb->s_qcop = &gfs2_quotactl_ops;
1104         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1105         sb->s_time_gran = 1;
1106         sb->s_maxbytes = MAX_LFS_FILESIZE;
1107
1108         /* Set up the buffer cache and fill in some fake block size values
1109            to allow us to read-in the on-disk superblock. */
1110         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1111         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1112         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1113                                GFS2_BASIC_BLOCK_SHIFT;
1114         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1115
1116         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1117         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1118         if (sdp->sd_args.ar_statfs_quantum) {
1119                 sdp->sd_tune.gt_statfs_slow = 0;
1120                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1121         } else {
1122                 sdp->sd_tune.gt_statfs_slow = 1;
1123                 sdp->sd_tune.gt_statfs_quantum = 30;
1124         }
1125
1126         error = init_names(sdp, silent);
1127         if (error)
1128                 goto fail;
1129
1130         gfs2_create_debugfs_file(sdp);
1131
1132         error = gfs2_sys_fs_add(sdp);
1133         if (error)
1134                 goto fail;
1135
1136         error = gfs2_lm_mount(sdp, silent);
1137         if (error)
1138                 goto fail_sys;
1139
1140         error = init_locking(sdp, &mount_gh, DO);
1141         if (error)
1142                 goto fail_lm;
1143
1144         error = init_sb(sdp, silent);
1145         if (error)
1146                 goto fail_locking;
1147
1148         error = wait_on_journal(sdp);
1149         if (error)
1150                 goto fail_sb;
1151
1152         /*
1153          * If user space has failed to join the cluster or some similar
1154          * failure has occurred, then the journal id will contain a
1155          * negative (error) number. This will then be returned to the
1156          * caller (of the mount syscall). We do this even for spectator
1157          * mounts (which just write a jid of 0 to indicate "ok" even though
1158          * the jid is unused in the spectator case)
1159          */
1160         if (sdp->sd_lockstruct.ls_jid < 0) {
1161                 error = sdp->sd_lockstruct.ls_jid;
1162                 sdp->sd_lockstruct.ls_jid = 0;
1163                 goto fail_sb;
1164         }
1165
1166         error = init_inodes(sdp, DO);
1167         if (error)
1168                 goto fail_sb;
1169
1170         error = init_per_node(sdp, DO);
1171         if (error)
1172                 goto fail_inodes;
1173
1174         error = gfs2_statfs_init(sdp);
1175         if (error) {
1176                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1177                 goto fail_per_node;
1178         }
1179
1180         error = init_threads(sdp, DO);
1181         if (error)
1182                 goto fail_per_node;
1183
1184         if (!(sb->s_flags & MS_RDONLY)) {
1185                 error = gfs2_make_fs_rw(sdp);
1186                 if (error) {
1187                         fs_err(sdp, "can't make FS RW: %d\n", error);
1188                         goto fail_threads;
1189                 }
1190         }
1191
1192         gfs2_glock_dq_uninit(&mount_gh);
1193         gfs2_online_uevent(sdp);
1194         return 0;
1195
1196 fail_threads:
1197         init_threads(sdp, UNDO);
1198 fail_per_node:
1199         init_per_node(sdp, UNDO);
1200 fail_inodes:
1201         init_inodes(sdp, UNDO);
1202 fail_sb:
1203         if (sdp->sd_root_dir)
1204                 dput(sdp->sd_root_dir);
1205         if (sdp->sd_master_dir)
1206                 dput(sdp->sd_master_dir);
1207         if (sb->s_root)
1208                 dput(sb->s_root);
1209         sb->s_root = NULL;
1210 fail_locking:
1211         init_locking(sdp, &mount_gh, UNDO);
1212 fail_lm:
1213         gfs2_gl_hash_clear(sdp);
1214         gfs2_lm_unmount(sdp);
1215 fail_sys:
1216         gfs2_sys_fs_del(sdp);
1217 fail:
1218         gfs2_delete_debugfs_file(sdp);
1219         kfree(sdp);
1220         sb->s_fs_info = NULL;
1221         return error;
1222 }
1223
1224 static int set_gfs2_super(struct super_block *s, void *data)
1225 {
1226         s->s_bdev = data;
1227         s->s_dev = s->s_bdev->bd_dev;
1228
1229         /*
1230          * We set the bdi here to the queue backing, file systems can
1231          * overwrite this in ->fill_super()
1232          */
1233         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1234         return 0;
1235 }
1236
1237 static int test_gfs2_super(struct super_block *s, void *ptr)
1238 {
1239         struct block_device *bdev = ptr;
1240         return (bdev == s->s_bdev);
1241 }
1242
1243 /**
1244  * gfs2_mount - Get the GFS2 superblock
1245  * @fs_type: The GFS2 filesystem type
1246  * @flags: Mount flags
1247  * @dev_name: The name of the device
1248  * @data: The mount arguments
1249  *
1250  * Q. Why not use get_sb_bdev() ?
1251  * A. We need to select one of two root directories to mount, independent
1252  *    of whether this is the initial, or subsequent, mount of this sb
1253  *
1254  * Returns: 0 or -ve on error
1255  */
1256
1257 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1258                        const char *dev_name, void *data)
1259 {
1260         struct block_device *bdev;
1261         struct super_block *s;
1262         fmode_t mode = FMODE_READ | FMODE_EXCL;
1263         int error;
1264         struct gfs2_args args;
1265         struct gfs2_sbd *sdp;
1266
1267         if (!(flags & MS_RDONLY))
1268                 mode |= FMODE_WRITE;
1269
1270         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1271         if (IS_ERR(bdev))
1272                 return ERR_CAST(bdev);
1273
1274         /*
1275          * once the super is inserted into the list by sget, s_umount
1276          * will protect the lockfs code from trying to start a snapshot
1277          * while we are mounting
1278          */
1279         mutex_lock(&bdev->bd_fsfreeze_mutex);
1280         if (bdev->bd_fsfreeze_count > 0) {
1281                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1282                 error = -EBUSY;
1283                 goto error_bdev;
1284         }
1285         s = sget(fs_type, test_gfs2_super, set_gfs2_super, bdev);
1286         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1287         error = PTR_ERR(s);
1288         if (IS_ERR(s))
1289                 goto error_bdev;
1290
1291         if (s->s_root)
1292                 blkdev_put(bdev, mode);
1293
1294         memset(&args, 0, sizeof(args));
1295         args.ar_quota = GFS2_QUOTA_DEFAULT;
1296         args.ar_data = GFS2_DATA_DEFAULT;
1297         args.ar_commit = 30;
1298         args.ar_statfs_quantum = 30;
1299         args.ar_quota_quantum = 60;
1300         args.ar_errors = GFS2_ERRORS_DEFAULT;
1301
1302         error = gfs2_mount_args(&args, data);
1303         if (error) {
1304                 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1305                 goto error_super;
1306         }
1307
1308         if (s->s_root) {
1309                 error = -EBUSY;
1310                 if ((flags ^ s->s_flags) & MS_RDONLY)
1311                         goto error_super;
1312         } else {
1313                 char b[BDEVNAME_SIZE];
1314
1315                 s->s_flags = flags;
1316                 s->s_mode = mode;
1317                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1318                 sb_set_blocksize(s, block_size(bdev));
1319                 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1320                 if (error)
1321                         goto error_super;
1322                 s->s_flags |= MS_ACTIVE;
1323                 bdev->bd_super = s;
1324         }
1325
1326         sdp = s->s_fs_info;
1327         if (args.ar_meta)
1328                 return dget(sdp->sd_master_dir);
1329         else
1330                 return dget(sdp->sd_root_dir);
1331
1332 error_super:
1333         deactivate_locked_super(s);
1334         return ERR_PTR(error);
1335 error_bdev:
1336         blkdev_put(bdev, mode);
1337         return ERR_PTR(error);
1338 }
1339
1340 static int set_meta_super(struct super_block *s, void *ptr)
1341 {
1342         return -EINVAL;
1343 }
1344
1345 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1346                         int flags, const char *dev_name, void *data)
1347 {
1348         struct super_block *s;
1349         struct gfs2_sbd *sdp;
1350         struct path path;
1351         int error;
1352
1353         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1354         if (error) {
1355                 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1356                        dev_name, error);
1357                 return ERR_PTR(error);
1358         }
1359         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super,
1360                  path.dentry->d_inode->i_sb->s_bdev);
1361         path_put(&path);
1362         if (IS_ERR(s)) {
1363                 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1364                 return ERR_CAST(s);
1365         }
1366         if ((flags ^ s->s_flags) & MS_RDONLY) {
1367                 deactivate_locked_super(s);
1368                 return ERR_PTR(-EBUSY);
1369         }
1370         sdp = s->s_fs_info;
1371         return dget(sdp->sd_master_dir);
1372 }
1373
1374 static void gfs2_kill_sb(struct super_block *sb)
1375 {
1376         struct gfs2_sbd *sdp = sb->s_fs_info;
1377
1378         if (sdp == NULL) {
1379                 kill_block_super(sb);
1380                 return;
1381         }
1382
1383         gfs2_meta_syncfs(sdp);
1384         dput(sdp->sd_root_dir);
1385         dput(sdp->sd_master_dir);
1386         sdp->sd_root_dir = NULL;
1387         sdp->sd_master_dir = NULL;
1388         shrink_dcache_sb(sb);
1389         kill_block_super(sb);
1390         gfs2_delete_debugfs_file(sdp);
1391         kfree(sdp);
1392 }
1393
1394 struct file_system_type gfs2_fs_type = {
1395         .name = "gfs2",
1396         .fs_flags = FS_REQUIRES_DEV,
1397         .mount = gfs2_mount,
1398         .kill_sb = gfs2_kill_sb,
1399         .owner = THIS_MODULE,
1400 };
1401
1402 struct file_system_type gfs2meta_fs_type = {
1403         .name = "gfs2meta",
1404         .fs_flags = FS_REQUIRES_DEV,
1405         .mount = gfs2_mount_meta,
1406         .owner = THIS_MODULE,
1407 };
1408