GFS2: Extend umount wait coverage to full glock lifetime
[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/slow-work.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 static const u32 gfs2_old_fs_formats[] = {
43         0
44 };
45
46 static const u32 gfs2_old_multihost_formats[] = {
47         0
48 };
49
50 /**
51  * gfs2_tune_init - Fill a gfs2_tune structure with default values
52  * @gt: tune
53  *
54  */
55
56 static void gfs2_tune_init(struct gfs2_tune *gt)
57 {
58         spin_lock_init(&gt->gt_spin);
59
60         gt->gt_incore_log_blocks = 1024;
61         gt->gt_logd_secs = 1;
62         gt->gt_quota_simul_sync = 64;
63         gt->gt_quota_warn_period = 10;
64         gt->gt_quota_scale_num = 1;
65         gt->gt_quota_scale_den = 1;
66         gt->gt_new_files_jdata = 0;
67         gt->gt_max_readahead = 1 << 18;
68         gt->gt_stall_secs = 600;
69         gt->gt_complain_secs = 10;
70 }
71
72 static struct gfs2_sbd *init_sbd(struct super_block *sb)
73 {
74         struct gfs2_sbd *sdp;
75
76         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
77         if (!sdp)
78                 return NULL;
79
80         sb->s_fs_info = sdp;
81         sdp->sd_vfs = sb;
82
83         gfs2_tune_init(&sdp->sd_tune);
84
85         init_waitqueue_head(&sdp->sd_glock_wait);
86         atomic_set(&sdp->sd_glock_disposal, 0);
87         spin_lock_init(&sdp->sd_statfs_spin);
88
89         spin_lock_init(&sdp->sd_rindex_spin);
90         mutex_init(&sdp->sd_rindex_mutex);
91         INIT_LIST_HEAD(&sdp->sd_rindex_list);
92         INIT_LIST_HEAD(&sdp->sd_rindex_mru_list);
93
94         INIT_LIST_HEAD(&sdp->sd_jindex_list);
95         spin_lock_init(&sdp->sd_jindex_spin);
96         mutex_init(&sdp->sd_jindex_mutex);
97
98         INIT_LIST_HEAD(&sdp->sd_quota_list);
99         mutex_init(&sdp->sd_quota_mutex);
100         init_waitqueue_head(&sdp->sd_quota_wait);
101         INIT_LIST_HEAD(&sdp->sd_trunc_list);
102         spin_lock_init(&sdp->sd_trunc_lock);
103
104         spin_lock_init(&sdp->sd_log_lock);
105
106         INIT_LIST_HEAD(&sdp->sd_log_le_buf);
107         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
108         INIT_LIST_HEAD(&sdp->sd_log_le_rg);
109         INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
110         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
111
112         mutex_init(&sdp->sd_log_reserve_mutex);
113         INIT_LIST_HEAD(&sdp->sd_ail1_list);
114         INIT_LIST_HEAD(&sdp->sd_ail2_list);
115
116         init_rwsem(&sdp->sd_log_flush_lock);
117         atomic_set(&sdp->sd_log_in_flight, 0);
118         init_waitqueue_head(&sdp->sd_log_flush_wait);
119
120         INIT_LIST_HEAD(&sdp->sd_revoke_list);
121
122         mutex_init(&sdp->sd_freeze_lock);
123
124         return sdp;
125 }
126
127
128 /**
129  * gfs2_check_sb - Check superblock
130  * @sdp: the filesystem
131  * @sb: The superblock
132  * @silent: Don't print a message if the check fails
133  *
134  * Checks the version code of the FS is one that we understand how to
135  * read and that the sizes of the various on-disk structures have not
136  * changed.
137  */
138
139 static int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
140 {
141         unsigned int x;
142
143         if (sb->sb_magic != GFS2_MAGIC ||
144             sb->sb_type != GFS2_METATYPE_SB) {
145                 if (!silent)
146                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
147                 return -EINVAL;
148         }
149
150         /*  If format numbers match exactly, we're done.  */
151
152         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
153             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
154                 return 0;
155
156         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
157                 for (x = 0; gfs2_old_fs_formats[x]; x++)
158                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
159                                 break;
160
161                 if (!gfs2_old_fs_formats[x]) {
162                         printk(KERN_WARNING
163                                "GFS2: code version (%u, %u) is incompatible "
164                                "with ondisk format (%u, %u)\n",
165                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
166                                sb->sb_fs_format, sb->sb_multihost_format);
167                         printk(KERN_WARNING
168                                "GFS2: I don't know how to upgrade this FS\n");
169                         return -EINVAL;
170                 }
171         }
172
173         if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
174                 for (x = 0; gfs2_old_multihost_formats[x]; x++)
175                         if (gfs2_old_multihost_formats[x] ==
176                             sb->sb_multihost_format)
177                                 break;
178
179                 if (!gfs2_old_multihost_formats[x]) {
180                         printk(KERN_WARNING
181                                "GFS2: code version (%u, %u) is incompatible "
182                                "with ondisk format (%u, %u)\n",
183                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
184                                sb->sb_fs_format, sb->sb_multihost_format);
185                         printk(KERN_WARNING
186                                "GFS2: I don't know how to upgrade this FS\n");
187                         return -EINVAL;
188                 }
189         }
190
191         if (!sdp->sd_args.ar_upgrade) {
192                 printk(KERN_WARNING
193                        "GFS2: code version (%u, %u) is incompatible "
194                        "with ondisk format (%u, %u)\n",
195                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
196                        sb->sb_fs_format, sb->sb_multihost_format);
197                 printk(KERN_INFO
198                        "GFS2: Use the \"upgrade\" mount option to upgrade "
199                        "the FS\n");
200                 printk(KERN_INFO "GFS2: See the manual for more details\n");
201                 return -EINVAL;
202         }
203
204         return 0;
205 }
206
207 static void end_bio_io_page(struct bio *bio, int error)
208 {
209         struct page *page = bio->bi_private;
210
211         if (!error)
212                 SetPageUptodate(page);
213         else
214                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
215         unlock_page(page);
216 }
217
218 static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
219 {
220         const struct gfs2_sb *str = buf;
221
222         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
223         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
224         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
225         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
226         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
227         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
228         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
229         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
230         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
231         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
232         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
233
234         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
235         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
236         memcpy(sb->sb_uuid, str->sb_uuid, 16);
237 }
238
239 /**
240  * gfs2_read_super - Read the gfs2 super block from disk
241  * @sdp: The GFS2 super block
242  * @sector: The location of the super block
243  * @error: The error code to return
244  *
245  * This uses the bio functions to read the super block from disk
246  * because we want to be 100% sure that we never read cached data.
247  * A super block is read twice only during each GFS2 mount and is
248  * never written to by the filesystem. The first time its read no
249  * locks are held, and the only details which are looked at are those
250  * relating to the locking protocol. Once locking is up and working,
251  * the sb is read again under the lock to establish the location of
252  * the master directory (contains pointers to journals etc) and the
253  * root directory.
254  *
255  * Returns: 0 on success or error
256  */
257
258 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
259 {
260         struct super_block *sb = sdp->sd_vfs;
261         struct gfs2_sb *p;
262         struct page *page;
263         struct bio *bio;
264
265         page = alloc_page(GFP_NOFS);
266         if (unlikely(!page))
267                 return -ENOBUFS;
268
269         ClearPageUptodate(page);
270         ClearPageDirty(page);
271         lock_page(page);
272
273         bio = bio_alloc(GFP_NOFS, 1);
274         bio->bi_sector = sector * (sb->s_blocksize >> 9);
275         bio->bi_bdev = sb->s_bdev;
276         bio_add_page(bio, page, PAGE_SIZE, 0);
277
278         bio->bi_end_io = end_bio_io_page;
279         bio->bi_private = page;
280         submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
281         wait_on_page_locked(page);
282         bio_put(bio);
283         if (!PageUptodate(page)) {
284                 __free_page(page);
285                 return -EIO;
286         }
287         p = kmap(page);
288         gfs2_sb_in(&sdp->sd_sb, p);
289         kunmap(page);
290         __free_page(page);
291         return 0;
292 }
293
294 /**
295  * gfs2_read_sb - Read super block
296  * @sdp: The GFS2 superblock
297  * @silent: Don't print message if mount fails
298  *
299  */
300
301 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
302 {
303         u32 hash_blocks, ind_blocks, leaf_blocks;
304         u32 tmp_blocks;
305         unsigned int x;
306         int error;
307
308         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
309         if (error) {
310                 if (!silent)
311                         fs_err(sdp, "can't read superblock\n");
312                 return error;
313         }
314
315         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
316         if (error)
317                 return error;
318
319         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
320                                GFS2_BASIC_BLOCK_SHIFT;
321         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
322         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
323                           sizeof(struct gfs2_dinode)) / sizeof(u64);
324         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
325                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
326         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
327         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
328         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
329         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
330         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
331                                 sizeof(struct gfs2_meta_header)) /
332                                 sizeof(struct gfs2_quota_change);
333
334         /* Compute maximum reservation required to add a entry to a directory */
335
336         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
337                              sdp->sd_jbsize);
338
339         ind_blocks = 0;
340         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
341                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
342                 ind_blocks += tmp_blocks;
343         }
344
345         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
346
347         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
348
349         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
350                                 sizeof(struct gfs2_dinode);
351         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
352         for (x = 2;; x++) {
353                 u64 space, d;
354                 u32 m;
355
356                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
357                 d = space;
358                 m = do_div(d, sdp->sd_inptrs);
359
360                 if (d != sdp->sd_heightsize[x - 1] || m)
361                         break;
362                 sdp->sd_heightsize[x] = space;
363         }
364         sdp->sd_max_height = x;
365         sdp->sd_heightsize[x] = ~0;
366         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
367
368         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
369                                  sizeof(struct gfs2_dinode);
370         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
371         for (x = 2;; x++) {
372                 u64 space, d;
373                 u32 m;
374
375                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
376                 d = space;
377                 m = do_div(d, sdp->sd_inptrs);
378
379                 if (d != sdp->sd_jheightsize[x - 1] || m)
380                         break;
381                 sdp->sd_jheightsize[x] = space;
382         }
383         sdp->sd_max_jheight = x;
384         sdp->sd_jheightsize[x] = ~0;
385         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
386
387         return 0;
388 }
389
390 static int init_names(struct gfs2_sbd *sdp, int silent)
391 {
392         char *proto, *table;
393         int error = 0;
394
395         proto = sdp->sd_args.ar_lockproto;
396         table = sdp->sd_args.ar_locktable;
397
398         /*  Try to autodetect  */
399
400         if (!proto[0] || !table[0]) {
401                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
402                 if (error)
403                         return error;
404
405                 error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
406                 if (error)
407                         goto out;
408
409                 if (!proto[0])
410                         proto = sdp->sd_sb.sb_lockproto;
411                 if (!table[0])
412                         table = sdp->sd_sb.sb_locktable;
413         }
414
415         if (!table[0])
416                 table = sdp->sd_vfs->s_id;
417
418         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
419         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
420
421         table = sdp->sd_table_name;
422         while ((table = strchr(table, '/')))
423                 *table = '_';
424
425 out:
426         return error;
427 }
428
429 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
430                         int undo)
431 {
432         int error = 0;
433
434         if (undo)
435                 goto fail_trans;
436
437         error = gfs2_glock_nq_num(sdp,
438                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
439                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
440                                   mount_gh);
441         if (error) {
442                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
443                 goto fail;
444         }
445
446         error = gfs2_glock_nq_num(sdp,
447                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
448                                   LM_ST_SHARED,
449                                   LM_FLAG_NOEXP | GL_EXACT,
450                                   &sdp->sd_live_gh);
451         if (error) {
452                 fs_err(sdp, "can't acquire live glock: %d\n", error);
453                 goto fail_mount;
454         }
455
456         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
457                                CREATE, &sdp->sd_rename_gl);
458         if (error) {
459                 fs_err(sdp, "can't create rename glock: %d\n", error);
460                 goto fail_live;
461         }
462
463         error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
464                                CREATE, &sdp->sd_trans_gl);
465         if (error) {
466                 fs_err(sdp, "can't create transaction glock: %d\n", error);
467                 goto fail_rename;
468         }
469
470         return 0;
471
472 fail_trans:
473         gfs2_glock_put(sdp->sd_trans_gl);
474 fail_rename:
475         gfs2_glock_put(sdp->sd_rename_gl);
476 fail_live:
477         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
478 fail_mount:
479         gfs2_glock_dq_uninit(mount_gh);
480 fail:
481         return error;
482 }
483
484 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
485                             u64 no_addr, const char *name)
486 {
487         struct gfs2_sbd *sdp = sb->s_fs_info;
488         struct dentry *dentry;
489         struct inode *inode;
490
491         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
492         if (IS_ERR(inode)) {
493                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
494                 return PTR_ERR(inode);
495         }
496         dentry = d_alloc_root(inode);
497         if (!dentry) {
498                 fs_err(sdp, "can't alloc %s dentry\n", name);
499                 iput(inode);
500                 return -ENOMEM;
501         }
502         dentry->d_op = &gfs2_dops;
503         *dptr = dentry;
504         return 0;
505 }
506
507 static int init_sb(struct gfs2_sbd *sdp, int silent)
508 {
509         struct super_block *sb = sdp->sd_vfs;
510         struct gfs2_holder sb_gh;
511         u64 no_addr;
512         int ret;
513
514         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
515                                 LM_ST_SHARED, 0, &sb_gh);
516         if (ret) {
517                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
518                 return ret;
519         }
520
521         ret = gfs2_read_sb(sdp, silent);
522         if (ret) {
523                 fs_err(sdp, "can't read superblock: %d\n", ret);
524                 goto out;
525         }
526
527         /* Set up the buffer cache and SB for real */
528         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
529                 ret = -EINVAL;
530                 fs_err(sdp, "FS block size (%u) is too small for device "
531                        "block size (%u)\n",
532                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
533                 goto out;
534         }
535         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
536                 ret = -EINVAL;
537                 fs_err(sdp, "FS block size (%u) is too big for machine "
538                        "page size (%u)\n",
539                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
540                 goto out;
541         }
542         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
543
544         /* Get the root inode */
545         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
546         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
547         if (ret)
548                 goto out;
549
550         /* Get the master inode */
551         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
552         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
553         if (ret) {
554                 dput(sdp->sd_root_dir);
555                 goto out;
556         }
557         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
558 out:
559         gfs2_glock_dq_uninit(&sb_gh);
560         return ret;
561 }
562
563 /**
564  * map_journal_extents - create a reusable "extent" mapping from all logical
565  * blocks to all physical blocks for the given journal.  This will save
566  * us time when writing journal blocks.  Most journals will have only one
567  * extent that maps all their logical blocks.  That's because gfs2.mkfs
568  * arranges the journal blocks sequentially to maximize performance.
569  * So the extent would map the first block for the entire file length.
570  * However, gfs2_jadd can happen while file activity is happening, so
571  * those journals may not be sequential.  Less likely is the case where
572  * the users created their own journals by mounting the metafs and
573  * laying it out.  But it's still possible.  These journals might have
574  * several extents.
575  *
576  * TODO: This should be done in bigger chunks rather than one block at a time,
577  *       but since it's only done at mount time, I'm not worried about the
578  *       time it takes.
579  */
580 static int map_journal_extents(struct gfs2_sbd *sdp)
581 {
582         struct gfs2_jdesc *jd = sdp->sd_jdesc;
583         unsigned int lb;
584         u64 db, prev_db; /* logical block, disk block, prev disk block */
585         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
586         struct gfs2_journal_extent *jext = NULL;
587         struct buffer_head bh;
588         int rc = 0;
589
590         prev_db = 0;
591
592         for (lb = 0; lb < ip->i_disksize >> sdp->sd_sb.sb_bsize_shift; lb++) {
593                 bh.b_state = 0;
594                 bh.b_blocknr = 0;
595                 bh.b_size = 1 << ip->i_inode.i_blkbits;
596                 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
597                 db = bh.b_blocknr;
598                 if (rc || !db) {
599                         printk(KERN_INFO "GFS2 journal mapping error %d: lb="
600                                "%u db=%llu\n", rc, lb, (unsigned long long)db);
601                         break;
602                 }
603                 if (!prev_db || db != prev_db + 1) {
604                         jext = kzalloc(sizeof(struct gfs2_journal_extent),
605                                        GFP_KERNEL);
606                         if (!jext) {
607                                 printk(KERN_INFO "GFS2 error: out of memory "
608                                        "mapping journal extents.\n");
609                                 rc = -ENOMEM;
610                                 break;
611                         }
612                         jext->dblock = db;
613                         jext->lblock = lb;
614                         jext->blocks = 1;
615                         list_add_tail(&jext->extent_list, &jd->extent_list);
616                 } else {
617                         jext->blocks++;
618                 }
619                 prev_db = db;
620         }
621         return rc;
622 }
623
624 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
625 {
626         char *message = "FIRSTMOUNT=Done";
627         char *envp[] = { message, NULL };
628         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
629         ls->ls_first_done = 1;
630         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
631 }
632
633 /**
634  * gfs2_jindex_hold - Grab a lock on the jindex
635  * @sdp: The GFS2 superblock
636  * @ji_gh: the holder for the jindex glock
637  *
638  * Returns: errno
639  */
640
641 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
642 {
643         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
644         struct qstr name;
645         char buf[20];
646         struct gfs2_jdesc *jd;
647         int error;
648
649         name.name = buf;
650
651         mutex_lock(&sdp->sd_jindex_mutex);
652
653         for (;;) {
654                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
655                 if (error)
656                         break;
657
658                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
659                 name.hash = gfs2_disk_hash(name.name, name.len);
660
661                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
662                 if (error == -ENOENT) {
663                         error = 0;
664                         break;
665                 }
666
667                 gfs2_glock_dq_uninit(ji_gh);
668
669                 if (error)
670                         break;
671
672                 error = -ENOMEM;
673                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
674                 if (!jd)
675                         break;
676
677                 INIT_LIST_HEAD(&jd->extent_list);
678                 slow_work_init(&jd->jd_work, &gfs2_recover_ops);
679                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
680                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
681                         if (!jd->jd_inode)
682                                 error = -ENOENT;
683                         else
684                                 error = PTR_ERR(jd->jd_inode);
685                         kfree(jd);
686                         break;
687                 }
688
689                 spin_lock(&sdp->sd_jindex_spin);
690                 jd->jd_jid = sdp->sd_journals++;
691                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
692                 spin_unlock(&sdp->sd_jindex_spin);
693         }
694
695         mutex_unlock(&sdp->sd_jindex_mutex);
696
697         return error;
698 }
699
700 static int init_journal(struct gfs2_sbd *sdp, int undo)
701 {
702         struct inode *master = sdp->sd_master_dir->d_inode;
703         struct gfs2_holder ji_gh;
704         struct gfs2_inode *ip;
705         int jindex = 1;
706         int error = 0;
707
708         if (undo) {
709                 jindex = 0;
710                 goto fail_jinode_gh;
711         }
712
713         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
714         if (IS_ERR(sdp->sd_jindex)) {
715                 fs_err(sdp, "can't lookup journal index: %d\n", error);
716                 return PTR_ERR(sdp->sd_jindex);
717         }
718         ip = GFS2_I(sdp->sd_jindex);
719
720         /* Load in the journal index special file */
721
722         error = gfs2_jindex_hold(sdp, &ji_gh);
723         if (error) {
724                 fs_err(sdp, "can't read journal index: %d\n", error);
725                 goto fail;
726         }
727
728         error = -EINVAL;
729         if (!gfs2_jindex_size(sdp)) {
730                 fs_err(sdp, "no journals!\n");
731                 goto fail_jindex;
732         }
733
734         if (sdp->sd_args.ar_spectator) {
735                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
736                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
737         } else {
738                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
739                         fs_err(sdp, "can't mount journal #%u\n",
740                                sdp->sd_lockstruct.ls_jid);
741                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
742                                gfs2_jindex_size(sdp),
743                                gfs2_jindex_size(sdp) - 1);
744                         goto fail_jindex;
745                 }
746                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
747
748                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
749                                           &gfs2_journal_glops,
750                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
751                                           &sdp->sd_journal_gh);
752                 if (error) {
753                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
754                         goto fail_jindex;
755                 }
756
757                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
758                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
759                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
760                                            &sdp->sd_jinode_gh);
761                 if (error) {
762                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
763                                error);
764                         goto fail_journal_gh;
765                 }
766
767                 error = gfs2_jdesc_check(sdp->sd_jdesc);
768                 if (error) {
769                         fs_err(sdp, "my journal (%u) is bad: %d\n",
770                                sdp->sd_jdesc->jd_jid, error);
771                         goto fail_jinode_gh;
772                 }
773                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
774
775                 /* Map the extents for this journal's blocks */
776                 map_journal_extents(sdp);
777         }
778         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
779
780         if (sdp->sd_lockstruct.ls_first) {
781                 unsigned int x;
782                 for (x = 0; x < sdp->sd_journals; x++) {
783                         error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x));
784                         if (error) {
785                                 fs_err(sdp, "error recovering journal %u: %d\n",
786                                        x, error);
787                                 goto fail_jinode_gh;
788                         }
789                 }
790
791                 gfs2_others_may_mount(sdp);
792         } else if (!sdp->sd_args.ar_spectator) {
793                 error = gfs2_recover_journal(sdp->sd_jdesc);
794                 if (error) {
795                         fs_err(sdp, "error recovering my journal: %d\n", error);
796                         goto fail_jinode_gh;
797                 }
798         }
799
800         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
801         gfs2_glock_dq_uninit(&ji_gh);
802         jindex = 0;
803
804         return 0;
805
806 fail_jinode_gh:
807         if (!sdp->sd_args.ar_spectator)
808                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
809 fail_journal_gh:
810         if (!sdp->sd_args.ar_spectator)
811                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
812 fail_jindex:
813         gfs2_jindex_free(sdp);
814         if (jindex)
815                 gfs2_glock_dq_uninit(&ji_gh);
816 fail:
817         iput(sdp->sd_jindex);
818         return error;
819 }
820
821
822 static int init_inodes(struct gfs2_sbd *sdp, int undo)
823 {
824         int error = 0;
825         struct gfs2_inode *ip;
826         struct inode *master = sdp->sd_master_dir->d_inode;
827
828         if (undo)
829                 goto fail_qinode;
830
831         error = init_journal(sdp, undo);
832         if (error)
833                 goto fail;
834
835         /* Read in the master statfs inode */
836         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
837         if (IS_ERR(sdp->sd_statfs_inode)) {
838                 error = PTR_ERR(sdp->sd_statfs_inode);
839                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
840                 goto fail_journal;
841         }
842
843         /* Read in the resource index inode */
844         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
845         if (IS_ERR(sdp->sd_rindex)) {
846                 error = PTR_ERR(sdp->sd_rindex);
847                 fs_err(sdp, "can't get resource index inode: %d\n", error);
848                 goto fail_statfs;
849         }
850         ip = GFS2_I(sdp->sd_rindex);
851         sdp->sd_rindex_uptodate = 0;
852
853         /* Read in the quota inode */
854         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
855         if (IS_ERR(sdp->sd_quota_inode)) {
856                 error = PTR_ERR(sdp->sd_quota_inode);
857                 fs_err(sdp, "can't get quota file inode: %d\n", error);
858                 goto fail_rindex;
859         }
860         return 0;
861
862 fail_qinode:
863         iput(sdp->sd_quota_inode);
864 fail_rindex:
865         gfs2_clear_rgrpd(sdp);
866         iput(sdp->sd_rindex);
867 fail_statfs:
868         iput(sdp->sd_statfs_inode);
869 fail_journal:
870         init_journal(sdp, UNDO);
871 fail:
872         return error;
873 }
874
875 static int init_per_node(struct gfs2_sbd *sdp, int undo)
876 {
877         struct inode *pn = NULL;
878         char buf[30];
879         int error = 0;
880         struct gfs2_inode *ip;
881         struct inode *master = sdp->sd_master_dir->d_inode;
882
883         if (sdp->sd_args.ar_spectator)
884                 return 0;
885
886         if (undo)
887                 goto fail_qc_gh;
888
889         pn = gfs2_lookup_simple(master, "per_node");
890         if (IS_ERR(pn)) {
891                 error = PTR_ERR(pn);
892                 fs_err(sdp, "can't find per_node directory: %d\n", error);
893                 return error;
894         }
895
896         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
897         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
898         if (IS_ERR(sdp->sd_sc_inode)) {
899                 error = PTR_ERR(sdp->sd_sc_inode);
900                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
901                 goto fail;
902         }
903
904         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
905         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
906         if (IS_ERR(sdp->sd_qc_inode)) {
907                 error = PTR_ERR(sdp->sd_qc_inode);
908                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
909                 goto fail_ut_i;
910         }
911
912         iput(pn);
913         pn = NULL;
914
915         ip = GFS2_I(sdp->sd_sc_inode);
916         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
917                                    &sdp->sd_sc_gh);
918         if (error) {
919                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
920                 goto fail_qc_i;
921         }
922
923         ip = GFS2_I(sdp->sd_qc_inode);
924         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
925                                    &sdp->sd_qc_gh);
926         if (error) {
927                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
928                 goto fail_ut_gh;
929         }
930
931         return 0;
932
933 fail_qc_gh:
934         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
935 fail_ut_gh:
936         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
937 fail_qc_i:
938         iput(sdp->sd_qc_inode);
939 fail_ut_i:
940         iput(sdp->sd_sc_inode);
941 fail:
942         if (pn)
943                 iput(pn);
944         return error;
945 }
946
947 static int init_threads(struct gfs2_sbd *sdp, int undo)
948 {
949         struct task_struct *p;
950         int error = 0;
951
952         if (undo)
953                 goto fail_quotad;
954
955         sdp->sd_log_flush_time = jiffies;
956
957         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
958         error = IS_ERR(p);
959         if (error) {
960                 fs_err(sdp, "can't start logd thread: %d\n", error);
961                 return error;
962         }
963         sdp->sd_logd_process = p;
964
965         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
966         error = IS_ERR(p);
967         if (error) {
968                 fs_err(sdp, "can't start quotad thread: %d\n", error);
969                 goto fail;
970         }
971         sdp->sd_quotad_process = p;
972
973         return 0;
974
975
976 fail_quotad:
977         kthread_stop(sdp->sd_quotad_process);
978 fail:
979         kthread_stop(sdp->sd_logd_process);
980         return error;
981 }
982
983 static const match_table_t nolock_tokens = {
984         { Opt_jid, "jid=%d\n", },
985         { Opt_err, NULL },
986 };
987
988 static void nolock_put_lock(struct kmem_cache *cachep, struct gfs2_glock *gl)
989 {
990         struct gfs2_sbd *sdp = gl->gl_sbd;
991         kmem_cache_free(cachep, gl);
992         if (atomic_dec_and_test(&sdp->sd_glock_disposal))
993                 wake_up(&sdp->sd_glock_wait);
994 }
995
996 static const struct lm_lockops nolock_ops = {
997         .lm_proto_name = "lock_nolock",
998         .lm_put_lock = nolock_put_lock,
999         .lm_tokens = &nolock_tokens,
1000 };
1001
1002 /**
1003  * gfs2_lm_mount - mount a locking protocol
1004  * @sdp: the filesystem
1005  * @args: mount arguements
1006  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
1007  *
1008  * Returns: errno
1009  */
1010
1011 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
1012 {
1013         const struct lm_lockops *lm;
1014         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1015         struct gfs2_args *args = &sdp->sd_args;
1016         const char *proto = sdp->sd_proto_name;
1017         const char *table = sdp->sd_table_name;
1018         const char *fsname;
1019         char *o, *options;
1020         int ret;
1021
1022         if (!strcmp("lock_nolock", proto)) {
1023                 lm = &nolock_ops;
1024                 sdp->sd_args.ar_localflocks = 1;
1025                 sdp->sd_args.ar_localcaching = 1;
1026 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1027         } else if (!strcmp("lock_dlm", proto)) {
1028                 lm = &gfs2_dlm_ops;
1029 #endif
1030         } else {
1031                 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
1032                 return -ENOENT;
1033         }
1034
1035         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
1036
1037         ls->ls_ops = lm;
1038         ls->ls_first = 1;
1039
1040         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
1041                 substring_t tmp[MAX_OPT_ARGS];
1042                 int token, option;
1043
1044                 if (!o || !*o)
1045                         continue;
1046
1047                 token = match_token(o, *lm->lm_tokens, tmp);
1048                 switch (token) {
1049                 case Opt_jid:
1050                         ret = match_int(&tmp[0], &option);
1051                         if (ret || option < 0) 
1052                                 goto hostdata_error;
1053                         ls->ls_jid = option;
1054                         break;
1055                 case Opt_id:
1056                         /* Obsolete, but left for backward compat purposes */
1057                         break;
1058                 case Opt_first:
1059                         ret = match_int(&tmp[0], &option);
1060                         if (ret || (option != 0 && option != 1))
1061                                 goto hostdata_error;
1062                         ls->ls_first = option;
1063                         break;
1064                 case Opt_nodir:
1065                         ret = match_int(&tmp[0], &option);
1066                         if (ret || (option != 0 && option != 1))
1067                                 goto hostdata_error;
1068                         ls->ls_nodir = option;
1069                         break;
1070                 case Opt_err:
1071                 default:
1072 hostdata_error:
1073                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1074                         return -EINVAL;
1075                 }
1076         }
1077
1078         if (sdp->sd_args.ar_spectator)
1079                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s", table);
1080         else
1081                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u", table,
1082                          sdp->sd_lockstruct.ls_jid);
1083
1084         fsname = strchr(table, ':');
1085         if (fsname)
1086                 fsname++;
1087         if (lm->lm_mount == NULL) {
1088                 fs_info(sdp, "Now mounting FS...\n");
1089                 return 0;
1090         }
1091         ret = lm->lm_mount(sdp, fsname);
1092         if (ret == 0)
1093                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1094         return ret;
1095 }
1096
1097 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1098 {
1099         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1100         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1101             lm->lm_unmount)
1102                 lm->lm_unmount(sdp);
1103 }
1104
1105 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1106 {
1107         struct super_block *sb = sdp->sd_vfs;
1108         char ro[20];
1109         char spectator[20];
1110         char *envp[] = { ro, spectator, NULL };
1111         sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1112         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1113         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1114 }
1115
1116 /**
1117  * fill_super - Read in superblock
1118  * @sb: The VFS superblock
1119  * @data: Mount options
1120  * @silent: Don't complain if it's not a GFS2 filesystem
1121  *
1122  * Returns: errno
1123  */
1124
1125 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1126 {
1127         struct gfs2_sbd *sdp;
1128         struct gfs2_holder mount_gh;
1129         int error;
1130
1131         sdp = init_sbd(sb);
1132         if (!sdp) {
1133                 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1134                 return -ENOMEM;
1135         }
1136         sdp->sd_args = *args;
1137
1138         if (sdp->sd_args.ar_spectator) {
1139                 sb->s_flags |= MS_RDONLY;
1140                 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
1141         }
1142         if (sdp->sd_args.ar_posix_acl)
1143                 sb->s_flags |= MS_POSIXACL;
1144         if (sdp->sd_args.ar_nobarrier)
1145                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1146
1147         sb->s_magic = GFS2_MAGIC;
1148         sb->s_op = &gfs2_super_ops;
1149         sb->s_export_op = &gfs2_export_ops;
1150         sb->s_xattr = gfs2_xattr_handlers;
1151         sb->s_qcop = &gfs2_quotactl_ops;
1152         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1153         sb->s_time_gran = 1;
1154         sb->s_maxbytes = MAX_LFS_FILESIZE;
1155
1156         /* Set up the buffer cache and fill in some fake block size values
1157            to allow us to read-in the on-disk superblock. */
1158         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1159         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1160         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1161                                GFS2_BASIC_BLOCK_SHIFT;
1162         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1163
1164         sdp->sd_tune.gt_log_flush_secs = sdp->sd_args.ar_commit;
1165         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1166         if (sdp->sd_args.ar_statfs_quantum) {
1167                 sdp->sd_tune.gt_statfs_slow = 0;
1168                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1169         }
1170         else {
1171                 sdp->sd_tune.gt_statfs_slow = 1;
1172                 sdp->sd_tune.gt_statfs_quantum = 30;
1173         }
1174
1175         error = init_names(sdp, silent);
1176         if (error)
1177                 goto fail;
1178
1179         gfs2_create_debugfs_file(sdp);
1180
1181         error = gfs2_sys_fs_add(sdp);
1182         if (error)
1183                 goto fail;
1184
1185         error = gfs2_lm_mount(sdp, silent);
1186         if (error)
1187                 goto fail_sys;
1188
1189         error = init_locking(sdp, &mount_gh, DO);
1190         if (error)
1191                 goto fail_lm;
1192
1193         error = init_sb(sdp, silent);
1194         if (error)
1195                 goto fail_locking;
1196
1197         error = init_inodes(sdp, DO);
1198         if (error)
1199                 goto fail_sb;
1200
1201         error = init_per_node(sdp, DO);
1202         if (error)
1203                 goto fail_inodes;
1204
1205         error = gfs2_statfs_init(sdp);
1206         if (error) {
1207                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1208                 goto fail_per_node;
1209         }
1210
1211         error = init_threads(sdp, DO);
1212         if (error)
1213                 goto fail_per_node;
1214
1215         if (!(sb->s_flags & MS_RDONLY)) {
1216                 error = gfs2_make_fs_rw(sdp);
1217                 if (error) {
1218                         fs_err(sdp, "can't make FS RW: %d\n", error);
1219                         goto fail_threads;
1220                 }
1221         }
1222
1223         gfs2_glock_dq_uninit(&mount_gh);
1224         gfs2_online_uevent(sdp);
1225         return 0;
1226
1227 fail_threads:
1228         init_threads(sdp, UNDO);
1229 fail_per_node:
1230         init_per_node(sdp, UNDO);
1231 fail_inodes:
1232         init_inodes(sdp, UNDO);
1233 fail_sb:
1234         if (sdp->sd_root_dir)
1235                 dput(sdp->sd_root_dir);
1236         if (sdp->sd_master_dir)
1237                 dput(sdp->sd_master_dir);
1238         if (sb->s_root)
1239                 dput(sb->s_root);
1240         sb->s_root = NULL;
1241 fail_locking:
1242         init_locking(sdp, &mount_gh, UNDO);
1243 fail_lm:
1244         gfs2_gl_hash_clear(sdp);
1245         gfs2_lm_unmount(sdp);
1246         while (invalidate_inodes(sb))
1247                 yield();
1248 fail_sys:
1249         gfs2_sys_fs_del(sdp);
1250 fail:
1251         gfs2_delete_debugfs_file(sdp);
1252         kfree(sdp);
1253         sb->s_fs_info = NULL;
1254         return error;
1255 }
1256
1257 static int set_gfs2_super(struct super_block *s, void *data)
1258 {
1259         s->s_bdev = data;
1260         s->s_dev = s->s_bdev->bd_dev;
1261
1262         /*
1263          * We set the bdi here to the queue backing, file systems can
1264          * overwrite this in ->fill_super()
1265          */
1266         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1267         return 0;
1268 }
1269
1270 static int test_gfs2_super(struct super_block *s, void *ptr)
1271 {
1272         struct block_device *bdev = ptr;
1273         return (bdev == s->s_bdev);
1274 }
1275
1276 /**
1277  * gfs2_get_sb - Get the GFS2 superblock
1278  * @fs_type: The GFS2 filesystem type
1279  * @flags: Mount flags
1280  * @dev_name: The name of the device
1281  * @data: The mount arguments
1282  * @mnt: The vfsmnt for this mount
1283  *
1284  * Q. Why not use get_sb_bdev() ?
1285  * A. We need to select one of two root directories to mount, independent
1286  *    of whether this is the initial, or subsequent, mount of this sb
1287  *
1288  * Returns: 0 or -ve on error
1289  */
1290
1291 static int gfs2_get_sb(struct file_system_type *fs_type, int flags,
1292                        const char *dev_name, void *data, struct vfsmount *mnt)
1293 {
1294         struct block_device *bdev;
1295         struct super_block *s;
1296         fmode_t mode = FMODE_READ;
1297         int error;
1298         struct gfs2_args args;
1299         struct gfs2_sbd *sdp;
1300
1301         if (!(flags & MS_RDONLY))
1302                 mode |= FMODE_WRITE;
1303
1304         bdev = open_bdev_exclusive(dev_name, mode, fs_type);
1305         if (IS_ERR(bdev))
1306                 return PTR_ERR(bdev);
1307
1308         /*
1309          * once the super is inserted into the list by sget, s_umount
1310          * will protect the lockfs code from trying to start a snapshot
1311          * while we are mounting
1312          */
1313         mutex_lock(&bdev->bd_fsfreeze_mutex);
1314         if (bdev->bd_fsfreeze_count > 0) {
1315                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1316                 error = -EBUSY;
1317                 goto error_bdev;
1318         }
1319         s = sget(fs_type, test_gfs2_super, set_gfs2_super, bdev);
1320         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1321         error = PTR_ERR(s);
1322         if (IS_ERR(s))
1323                 goto error_bdev;
1324
1325         memset(&args, 0, sizeof(args));
1326         args.ar_quota = GFS2_QUOTA_DEFAULT;
1327         args.ar_data = GFS2_DATA_DEFAULT;
1328         args.ar_commit = 60;
1329         args.ar_statfs_quantum = 30;
1330         args.ar_quota_quantum = 60;
1331         args.ar_errors = GFS2_ERRORS_DEFAULT;
1332
1333         error = gfs2_mount_args(&args, data);
1334         if (error) {
1335                 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1336                 if (s->s_root)
1337                         goto error_super;
1338                 deactivate_locked_super(s);
1339                 return error;
1340         }
1341
1342         if (s->s_root) {
1343                 error = -EBUSY;
1344                 if ((flags ^ s->s_flags) & MS_RDONLY)
1345                         goto error_super;
1346                 close_bdev_exclusive(bdev, mode);
1347         } else {
1348                 char b[BDEVNAME_SIZE];
1349
1350                 s->s_flags = flags;
1351                 s->s_mode = mode;
1352                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1353                 sb_set_blocksize(s, block_size(bdev));
1354                 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1355                 if (error) {
1356                         deactivate_locked_super(s);
1357                         return error;
1358                 }
1359                 s->s_flags |= MS_ACTIVE;
1360                 bdev->bd_super = s;
1361         }
1362
1363         sdp = s->s_fs_info;
1364         mnt->mnt_sb = s;
1365         if (args.ar_meta)
1366                 mnt->mnt_root = dget(sdp->sd_master_dir);
1367         else
1368                 mnt->mnt_root = dget(sdp->sd_root_dir);
1369         return 0;
1370
1371 error_super:
1372         deactivate_locked_super(s);
1373 error_bdev:
1374         close_bdev_exclusive(bdev, mode);
1375         return error;
1376 }
1377
1378 static int set_meta_super(struct super_block *s, void *ptr)
1379 {
1380         return -EINVAL;
1381 }
1382
1383 static int gfs2_get_sb_meta(struct file_system_type *fs_type, int flags,
1384                             const char *dev_name, void *data, struct vfsmount *mnt)
1385 {
1386         struct super_block *s;
1387         struct gfs2_sbd *sdp;
1388         struct path path;
1389         int error;
1390
1391         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1392         if (error) {
1393                 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1394                        dev_name, error);
1395                 return error;
1396         }
1397         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super,
1398                  path.dentry->d_inode->i_sb->s_bdev);
1399         path_put(&path);
1400         if (IS_ERR(s)) {
1401                 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1402                 return PTR_ERR(s);
1403         }
1404         if ((flags ^ s->s_flags) & MS_RDONLY) {
1405                 deactivate_locked_super(s);
1406                 return -EBUSY;
1407         }
1408         sdp = s->s_fs_info;
1409         mnt->mnt_sb = s;
1410         mnt->mnt_root = dget(sdp->sd_master_dir);
1411         return 0;
1412 }
1413
1414 static void gfs2_kill_sb(struct super_block *sb)
1415 {
1416         struct gfs2_sbd *sdp = sb->s_fs_info;
1417
1418         if (sdp == NULL) {
1419                 kill_block_super(sb);
1420                 return;
1421         }
1422
1423         gfs2_meta_syncfs(sdp);
1424         dput(sdp->sd_root_dir);
1425         dput(sdp->sd_master_dir);
1426         sdp->sd_root_dir = NULL;
1427         sdp->sd_master_dir = NULL;
1428         shrink_dcache_sb(sb);
1429         kill_block_super(sb);
1430         gfs2_delete_debugfs_file(sdp);
1431         kfree(sdp);
1432 }
1433
1434 struct file_system_type gfs2_fs_type = {
1435         .name = "gfs2",
1436         .fs_flags = FS_REQUIRES_DEV,
1437         .get_sb = gfs2_get_sb,
1438         .kill_sb = gfs2_kill_sb,
1439         .owner = THIS_MODULE,
1440 };
1441
1442 struct file_system_type gfs2meta_fs_type = {
1443         .name = "gfs2meta",
1444         .fs_flags = FS_REQUIRES_DEV,
1445         .get_sb = gfs2_get_sb_meta,
1446         .owner = THIS_MODULE,
1447 };
1448